This PDF file contains the front matter associated with SPIE Proceedings Volume 12329, including the Title Page, Copyright information, Table of Contents, and Conference Committee listings.

In the modeling of hybrid vehicle, the neural network model in this paper can optimize the actual measurement and data parameter analysis of the engine by using GT software on the PC side. In the modeling process of hybrid vehicle engine, the basic model is established first, and the output torque and running power of the basic model are optimized after testing, so as to reduce the error of true value, and the dynamic performance of hybrid locomotive is calculated by using the neural network model. Starting from the era background of the continuous improvement of the performance of hybrid vehicles, this paper analyzes and studies the modeling and model verification of L hybrid vehicles, builds the test data results of the hybrid vehicle model, and studies its control strategy by analyzing the hybrid vehicle power system.

With the continuous expansion of the application fields of deep learning, human life is becoming more and more wonderful. This paper mainly uses the application of deep learning in the field of image recognition and uses deep learning algorithm to recognition the problem of flowers. Due to the huge number of flowering species, How to automatically recognition flowers through the system gives people more understanding of flowers while appreciating them, which is a topic with practical application value. This paper constructs a deep learning multi-classification model based on ResNet, and conducts training based on PyTorch framework on the data set of more than 3000 pictures of five kinds of flowers. After 50 rounds of iterative training, it finally achieves 91% average accuracy on the test dataset.Finally, a recognition system based on this model is realized by visualization technology, so as to facilitate users to effectively recognize the kind of flowers through this system.

The purpose of this paper is to examine the current state of research on the application of virtual reality technology in autism spectrum disorders in China. A precise search for virtual reality and autism was accessed through CNKI's advanced search topic, and there were 21 results. By carefully reading and studying the research results of others, this paper analyzes the latest applications of virtual reality technology in the treatment of autism spectrum disorders. Virtual reality technology can effectively reduce unreasonable fear of specific situations and objects and significantly enhance social-emotional recognition and processing for people with ASD. By compensating for language barriers through nonverbal expression, more people with ASD can transfer the skills learned in virtual reality to real social life.

Cross-modality person re-identification, also known as RGB-infrared person re-identification, is a person recognition and retrieval technology between visible and infrared images. As an essential technologies of video intelligent surveillance system, the technology is urgently needed in security monitoring and suspect tracking. The modality discrepancy and intra-class variations in cross-modality person re-identification make it difficult to extract effective discriminable features of person identity. There are significantly fewer effective discriminable features in the shared feature space caused by shared network parameters. In order to solve this problem, an attention-guided cross-modality person re-identification is designed in this paper. The attention module uses the attention mechanism to focus on the salient features, so as to guide the network to put more parameters on the learning of salient features, so that the network can learn more discriminable, features from salient features. A comparative experiment on two standard data sets of cross-modality person re-identification verifies the effectiveness and advancement of the proposed method. For example, the proposed method improves rank-1 by 17.79% on SYSU-MM01 and 16.55% on mAP.

The sequential recommendation aims to recommend items that may be of interest to users based on user behavior sequence information. However, most of the current sequential recommendation tasks cannot adequately model the longterm preferences and short-term intentions of users when mining user preferences. In order to model and integrate the user's long-term and short-term preferences effectively, a new sequential recommendation method, which integrates item relationship and user preference, named IRUP, is proposed. Firstly, Graph Convolution Network (GCN) and an itemrelation level attention mechanism are utilized to model long-term and short-term preferences of users respectively; secondly, a co-attention mechanism is used to learn the cross-correlation information between long-term and short-term preferences, which enhances the user's preference representation; finally, the long-term preference and the short-term preference are fused, and the inner product is introduced to calculate the recommendation list. The experimental results on two public datasets Beauty and Home show that the proposed method IRUP can effectively improve the recommendation performance. Compared with the best baseline method KDA, the two evaluation metrics HR@5 and NDCG@5 have an average improvement of 8.28% and 7.03%, respectively.

According to the harsh condition of the conditional independence of feature attributes in the Naive Bayesian algorithm, an improved Naive Bayesian classification method based on feature correlation weighting is proposed. The degree of correlation between feature attributes, class attributes and feature attributes is judged by symmetric uncertainty, irrelevant feature attributes and redundant attributes are eliminated, and then the correlation between feature attributes and class variables is high and the correlation degree between feature attributes is low. The basic principle of the weight assignment of feature attributes, and finally experimental verification on the public data set. According to the experimental results, the algorithm can effectively reduce the influence of the correlation between the features and effectively improve the classification accuracy.

Key performance indicators are important tools for enterprises to evaluate smart manufacturing. However, equipment data on the workshop is Multi-Source Heterogeneous Data that is difficult to interconnect, and equipment state is affected by multiple mutually coupled factors, resulting in the inability to accurately calculate KPIs. A standardized information model is the key to solve the problem. Therefore, this paper addresses the problem that intelligent KPI evaluation data is difficult to integrate and multi-source, analyzes the connection between KPI calculation parameters and different production states of equipment, and decouples the multi-source factors that distinguish equipment states from multiple dimensions, while clarifies the information model architecture of industrial robots and establishes a semantically consistent and standardized information model of industrial robots according to OPC UA specification.

With the advent of the era of big data, a large number of sensors are scattered on traffic roads, generating traffic flow data, posing huge challenges to traffic flow prediction. Therefore, capturing the spatial features of traffic flow data is greatly significant to improve the accuracy of traffic flow predicted models. Analysis of spatial and temporal characteristics of traffic flow data, this paper proposes a Spatial-Temporal Attention (MA-STLGAT) model to predict the traffic speed in the traffic network. The Graph Attention Network (GAT) assigns different weights to neighbor nodes to capture the dynamic topology structure. Learning the temporal correlation of traffic flow data through Long ShortTerm Memory network (LSTM), and Multi-Head Attention is introduced to learn the contribution of nodes, improve the convergence speed and save computing costs. The experimental results on real data set PeMSD7 show that the predict performance of MA-STLGAT is better than other baseline models.

In the Reid method which is based on deep learning, the combination of local features and global features can provide a robust representation for person retrieval. Human pose information can provide the location of the human skeleton, thus effectively directing the network to focus more on these critical regions, while also helping to reduce noise interference from background or occlusion. However, previous pose based approaches may not take full advantage of the pose information and rarely consider the different contributions of independent local features. In this paper, we design a pose guided multi-branch graph convolutional attention network (PG-MBGCAN) for the problem of re-identification of occluded pedestrians, which exploits the pose relationship to improve the robustness of pedestrian features and considers the contributions of different local features. Experimental results show that the proposed method in this paper has good performance in different datasets.

The society has entered the era of big data, computer network technology has improved people's living standards to a new level, and continues to progress toward the direction of artificial intelligence. Artificial intelligence is one of the three major scientific and technological achievements of the 20th century, and mathematics is its key theoretical foundation. Taking artificial intelligence in the era of big data as the research object, this paper analyzes the application of traditional mathematical operations in artificial intelligence in the field of computer science and technology in the future from the perspective of mathematical thinking. This paper mainly introduces the typical mathematical basis of artificial intelligence, including linear algebra, probability theory, mathematical statistics, discrete mathematics and mathematical modeling, and prospects the development trend of mathematical basis of artificial intelligence, hoping to provide some reference for the application of artificial intelligence in computer network technology.

The electric vehicle industry has greatly developed in this years, which has led to significant growth in the number of fast-charging stations for electric vehicles. This paper takes an electric bus DC fast charging station as the research object, and proposes a smart charging station system based on the Internet of Things. It not only realizes the interconnection and intercommunication of data between the equipment and facilities of the charging station, the local control platform, and the cloud platform, but also realizes comprehensive intelligent management combining local control and remote control. According to the historical load data collected by the smart charging station, this paper also studies the performance of LSTM in the load prediction of the charging station. Build a load prediction model in MATLAB, determine the optimal network structure through simulation experiments, and select the appropriate input and output to train the model. The simulation outcomes show that the prediction accuracy of LSTM for charging station load is higher than that of traditional BP neural network, which lays a foundation for the research on energy control strategies of smart charging stations.

Photovoltaic systems are commonly used in daily life as an important device for collecting solar energy. It is important to model PV systems, mainly to simulate the I-V response of solar cells under various conditions. In order to accurately estimate all unknown parameters of different PV models, a new hybrid algorithm, called LSHADE-TSO, is proposed by hybridizing LSHADE and the tuna swarm optimization (TSO). The spiral foraging search and parabolic foraging search of TSO are introduced into the mutation strategy in LSHADE to improve the exploration ability and population diversity. In addition, this paper adds the crossover factor (CR) ranking, top α r1 selection, strategies to improve the convergence efficiency. LSHADE-TSO is applied in solving photovoltaic parameter identification by comparing with well-known algorithms on photovoltaic parameter identification problems in recent years. The data results show that the LSHADETSO algorithm has better convergence and better search capability than other algorithms.

Acronym Identification means to recognizing acronyms and their definitions in a given paragraph by an algorithm. Existing studies have shown that the pre-training model has significant performance advantages in acronyms recognition. However, the vanilla Bert measures randomly masking tokens in the pre-training, which may also abbreviate the token of acronyms in sentence. The continuous masking method will be more helpful to the recognition and interpretation of acronyms. In this paper, we propose to use the SpanBERT model for acronym recognition and obtaining the transfer relationship between indicators with the help of a conditional random field (CRF). Through experimental comparison, the results demonstrate that our method increases the F1 value compared with the Bert and BERT-CRF models.

In cloud manufacturing (CMfg), various services with different functions are aggregated into an optimal service composition to handle complex manufacturing tasks, while maximizing service quality. However, the collaborative effect between services has a significant impact on the service composition, and the conflict between tasks may also affect the operation of the service composition in a real manufacturing environment. To handle these problems, this paper proposes an adaptive model of manufacturing service composition based on conflict detection. First, the preference information of both service parties and the collaborative performance between the services are considered to obtain a more realistic and satisfactory service composition. Subsequently, resource conflicts are detected in real time, and resolved according to their types, so as to realize the self-adaptation of the service composition. Finally, the illustrative case and experimental results are conducted to verify the feasibility and effectiveness of the proposed model.

In order to make full use of operation and maintenance records for in-depth mining to support management, improve operation and maintenance work management level and execution efficiency, this paper analyzes and preprocesses operation and maintenance log files by studying artificial intelligence technology. A user behavior expression model is constructed from each dimension and log analysis experiments are carried out. Through the research on LSTM and BiLSTM methods, a data-driven abnormal behavior intelligent analysis method based on Bi-LSTM is proposed. The experimental results show that the Bi-LSTM method in this paper has an accuracy of 86% and strong operability in detecting abnormal operation and maintenance operations, and has good performance in detecting abnormal operation and maintenance operations of users. The command line is used as an analysis of abnormal operation and maintenance behavior of audit data.

Stock price prediction has sparked the interest of financial investors and scholars, and it is also a research topic for academics. Because of the non-linearity and fluctuation of stock prices, classic statistical methods for stock price prediction are less than ideal. When it comes to analyzing time series data, the deep learning model provides a lot of advantages. In this paper, we crawl the historical stock price data of GOOGLE and provide a stock prediction system based on a feedforward neural network to further optimize the prediction model. We use RMSE, MAE and MAPE to verify the model’s prediction accuracy. The empirical results indicate that the feedforward neural network provides good effectiveness and feasibility.

Artificial intelligence (AI) technologies are rapidly proliferating around the world. In this paper, we first present the current state of AI applications around the world, analyzing the current use of iBorderCtrl in border control and the problems that exist. The current state of domestic security against terrorist attacks and mass migration problems highlights the need for a better and more effective border management system. At the heart of the AI-based iBorderCtrl project is the ability to perform automated "deception detection" and "risk assessment" in border crossing security. iBorderCtrl's technology is built on the controversial discipline of "emotion recognition science". Systems that rely on machine learning, especially those involving facial recognition technology, have so far been problematic and biased.

Aiming at the problem that the classical evidence combination formula will fail in case of serious conflict between evidences, an improved method of Dempster-Shafer (D-S) evidence theory based on reward and punishment factors is proposed. Firstly, a conflict allocation formula based on the basic probability assignments is constructed, and then a calculation method of reward and punishment factors based on credibility is proposed. The reward and punishment factors are substituted into the conflict allocation formula to correct the proportion of conflict allocation. Finally, the modified conflict allocation formula is used for evidence fusion. Simulation results show that the proposed algorithm can effectively solve the conflict between pieces of evidence and has better convergence and stability compared to the classical combination formula and other improved methods when multiple pieces of evidence are fused. In the process of practical application, the performance of the algorithm can be changed by adjusting the size of specific parameters, which improves the flexibility of the application of the algorithm.

The status of port logistics is becoming more and more important, and the construction of port informatization has become a new trend of logistics development. However, there are some problems in the management and retrieval of relevant knowledge resources in port logistics, such as information islands and lack of semantic interaction, which lead to the differences in the management of relevant knowledge resources and the low quality of retrieval services. Through the IDEA design platform, this paper uses the Jena reasoning package and related toolkits to design a relatively simple autonomous question-answering system, which realizes the reasoning and query of the port logistics container knowledge resource ontology knowledge base and demonstrates the reliability of the research content based on the accuracy of the query results. Experimental tests show that the accuracy of the system is high, and it has practical use and promotion value.

Aiming at the problem of shelter of identification and concentration analysis such as wearing masks in offline learning under the background of epidemic situation, the identification method of action energy map based on change of eye and the concentration analysis method based on eye movement offset were proposed. The eye contour maps in a complete blink cycle were extracted from the self-built blink database videos to synthesize the action energy map. Wavelet packet decomposition was used to extract features, and kernel principal component analysis (KPCA) was used to reduce the data dimension, then support vector machine (SVM) was selected for classification training. The accuracy can reach 95%. The gradient center algorithm was used to get the eyeball center and collect the videos in the states of focus and non focus in the online class, then the Manhattan distance between the eyeball center and the eye contour center was used to measure the eye movement offset, finally the threshold to divide focus and non focus was defined.to optimize the interactive effect of online education.

Aiming at the nonlinear characteristics of PMSM vector control system, such as slow response speed, poor steady-state performance and external disturbance, a method based on fuzzy PID controller is proposed to control PMSM. And by improving the grey predictive control, the feedback data of the controller is predicted in advance, and the fuzzy control system is further improved. The simulation results show that the fuzzy PID controller based on the improved grey prediction improves the control accuracy and response speed of the PMSM control system, reduces the overshoot of the system, and improves the robustness of the vector control system.

Aiming at the following problems encountered by Multiple Supervised Fully Convolutional Networks (MSFCN ) in liver tumor segmentation: rich blood vessels around the liver, complex organs, and inconspicuous boundaries, which have a great impact on accurate tumor segmentation, plus The tumor boundary is not clear and the segmentation is difficult. In this paper, we propose a method to automatically segment liver and tumor in CT abdominal images using Cascaded Fully Convolutional Networks (CFCN) and 3D Conditional Random Fields (CRF). We train and cascade two fully convolutional neural networks for combined segmentation of the liver and its tumors. In the first step, our trained FCN segmented the liver from abdominal CT images as the input of the second FCN, which removed the influence of the environment around the liver. The second FCN segmented the tumor region only based on the liver obtained in the first step. Finally, 3D conditional random field is used to optimize the segmented image obtained by cascaded FCN, and the edge information of liver tumor image is extracted to solve the problem of unclear boundary of liver tumor and complete the segmentation.

Text classification is the most common application of Natural Language Processing(NLP), and Transformer models have dominated the field in recent years. Currently, pre-training modeling of text through deep learning methods is a common way of text classification. This paper firstly proposes an improved XLNet text classification model based on the problems of long-term dependence and insufficient contextual semantic expression in previous pre-trained language models, and uses XLNet pre-trained language modeling to represent text as low-dimensional word vectors to obtain sequences. Secondly, the generated word vector sequence is passed into the LSTM network, and the two-way features of the sentence are extracted by the memory unit of the LSTM. On the basis of effectively extracting text features, the Multi-head attention model is used to calculate the multi-angle attention probability, to focus on the important words. Finally, text classification is achieved through a fully connected network. The experimental results show that the accuracy of the model is 0.9457 and the loss rate is 0.3133. Compared with other related models, it has achieved better results.

The temperature and pressure in the transition ladle are two important factors that affect the thickness of ultra-thin amorphous alloy, and there is a complex nonlinear relationship between the two factors and the thickness. To solve the problem that the control model cannot be determined in the process of thickness control of ultra-thin amorphous alloy, a regression model of ultra-thin amorphous alloy thickness based on improved PSO-GRNN is proposed. And the regression model based on GRNN takes temperature and pressure as input and thickness as output, which is a model that describes the nonlinear relationship between them. Based on the standard PSO algorithm, this paper introduces crossover and mutation operations, and the inertia weight of the PSO algorithm is improved. The smoothing parameter of GRNN is optimized by the improved PSO algorithm. Combined with the production field data, the model is simulated and analyzed. Comparing the method in this paper with BP, RBF, GRNN, and standard PSO-GRNN models, the results show that the method in this paper can effectively improve the accuracy of the model and can be used as the control model of the ultra-thin amorphous alloy thickness control system.

To complete the simulation model design of grassland land use change based on Repast Simphony platform, it is necessary to simulate and predict the state feedback and specific change process of grassland land under different influencing factors such as complexity, nonlinearity and interactivity. In this process of model design, the grassland land use change in a certain area is taken as the research object, the present situation of grassland land use and the significance of model establishment are deeply discussed, and the model establishment objectives, model environment and so on are analyzed, and finally the simulation experiment is completed. Through simulation, it can be determined that the transfer of animal husbandry labor force, industrial utilization mode and government policy have certain influence on the sustainable development of grassland ecology. At this stage, the problems of oversimplification and low simulation degree existing in the application of multi-agent system are perfected and supplemented, which has high practical value. Therefore, the multi-agent simulation model of grassland land use change can provide some support for the government to carry out grassland management and herdsmen to make rational use of grassland.

In view of the problems that the existing subjective evaluation of vibration comfort is easily influenced by people's psychological state, leading to the large deviation of the score value, and the existing subjective score has no quantifiable scientific index, a new idea of calibrating the subjective score system by using the modern physiological and psychological model is put forward. The specific research contents are as follows: a human vibration test platform which can realize vertical vibration is established, and the applied vibration excitation signal is set according to the standard; 15 volunteers were recruited to carry out systematic and scientific pre-test training and experiments. A large number of experimental data verified the applicability of Stevens' laws of physical psychology to Chinese people. Finally, the conclusion that the subjective evaluation of vibration comfort can be completed by setting up a questionnaire on this basis is obtained.

To solve the problem of water-coal ratio imbalance under the frequent changes of coal quality in thermal power supercritical units, this paper proposes a vector autoregressive (VAR) model to achieve self-optimization of water-coal ratio of the unit under steady state. This method takes a supercritical 350MW unit as an example. By screening the time series data of key parameters of the unit under steady-state operating conditions, the optimal order of VAR model is determined. Using the VAR model, we can find the optimal value of water-coal ratio under the corresponding operating conditions. And this paper further adopts error analysis, uncertainty analysis, and impulse response analysis to ensure the accuracy and safety of the optimal value of water-coal ratio. In actual operation, the unit can use the VAR model to achieve self-optimization, that is, to find the optimal water-coal ratio under different coal qualities, and to form the water-coal ratio curve under the current coal quality. Practice has proved that this method can effectively solve the problems of large deviation of the main steam temperature and main steam pressure caused by the unsuitable boiler feed water strategy due to frequent changes in coal quality of the unit.

In order to shorten the identification time of vehicle to obstacles and realize the real-time control of vehicle obstacle avoidance, a new method of vehicle obstacle avoidance control based on big data and real-time modeling was proposed. Using real-time modeling technology, a database containing a large number of driving data under different conditions is regarded as the mathematical model of the system, and the gain and control time of the control input are calculated by extracting the data, so as to avoid obstacles. The numerical analysis shows that the method eliminates the complicated mathematical model and only needs simple arithmetic operation and sorting to realize the real-time control, which has the advantage of short computation time. Compared with the previous control method, the new control method can avoid various obstacles more effectively.

To improve the prediction accuracy of the system load, this paper couples k-means clustering algorithm and neural network to predict the system load by taking into account the component characteristics of user load. Firstly, the typical curve of each 10kV distribution line is calculated. Secondly, the distribution lines are clustered into different categories through kmeans clustering according to the typical curves so that the user-side load can be effectively decomposed. Finally, the system load is predicted by using neural network with the load forecast result of different components as features. The experimental results show that the system load prediction considering the user-side load component analysis is beneficial to improve the prediction accuracy of the system load. This research has certain engineering application value.

It is possible for the distribution network to absorb large-scale distributed wind power resources. Through gas turbines and P2G, the uncertainty brought by wind power integration can be effectively solved, which can alleviate the gas pressure of the natural gas network. Wind power resources are referred to in this paper, and their error distribution is fitted. On this basis, the distributed optimization model of distribution network and gas distribution network is established, in which the uncertainty of wind power output is constrained by chance. The P2G and gas turbine are used as flexible and adjustable resources to deal with the uncertainty. Specifically, the nonlinear term in the model is linearized and solved by the CPLEX solver. A comparative analysis of specific cases is carried out. It is proved that the research content of this paper can realize more flexible and diversified means of system regulation, enhance the ability of the system to withstand uncertainty, and is conducive to the operation of the system under the influence of uncertainty.

This paper analyzes the limitations of the existing work, and studies the channel access strategy for these limitations. Queuing mechanism and secondary sensing mechanism are used to improve channel utilization, reduce the number of node access requests, avoid packet conflicts, and prioritize according to the types of nodes to reduce the number of nodes competing in the channel. The QMACP protocol is proposed, and finally the protocol is analyzed by simulation.

In people re-identification tasks, the most intuitive method for selecting people representation features is to directly extract a global feature map of the people. However, relying on global features alone often fails to accurately identify people in the presence of occlusion, misalignment and background interference. In addition, due to interference from factors such as low camera resolution and illumination, some key local features (e.g. carried objects, body parts such as the face or limbs) are not clearly observed. To this end, we are inspired by fine-grained image classification tasks and propose an attentionbased fine-grained feature network model (AFGF) for people re-identification tasks, which will fully consider the relationship between global and local features and incorporate attention mechanisms to effectively extract fine features of people and improve their discrimination ability. The effectiveness of our model is validated on the Market-1501 and DukeMTMC-reID datasets, and the experimental results show that the method improves significantly on both the supervised baseline and unsupervised baselines.

Cybersecurity event extraction aims to extract threat intelligence information from unstructured text and provide a data basis for correlation analysis of cybersecurity events. However, in the field of cybersecurity, the complexity of terminology in event data as well as the problem of polysemy and crossover between Chinese and English in Chinese texts pose great challenges to event extraction. Existing methods usually perform event extraction in a pipelined manner, ignoring the dependencies between event elements and event triggers. Therefore, a joint event extraction model is proposed in this paper. The model captures the contextual features of sentences in the encoding layer by an improved temporal convolutional network (TCN) and then enhances the dependency features between triggers and arguments using a multihead attention mechanism to achieve sentence-level joint extraction of Chinese cybersecurity events. We conducted comparison experiments on real cybersecurity news data to evaluate the event extraction performance of the model. Compared with a baseline model LSTM, the method improves the F1 values by 13.2% and 33.2% in two subtasks of trigger extraction and argument extraction, respectively.

Since the applied-field magnetoplasmadynamic thruster (AF-MPDT) has the unique characteristics of high plasma density, strong magnetic flux density and non-equilibrium state, an innovative simulation model, namely the multi-component magnetic hydrodynamic (MHD) model, is proposed in this work. The conventional MHD model includes 2 sets of equations, which are the fluid describing the motion of plasma and the Maxwell equations for the electromagnetic description. In the proposed multi-component MHD model, the Maxwell equations are maintained, but the fluid part is improved by considering the two components of the plasma, electrons and ions, separately. In order to describe the relationship between the electrons and ions, the high-frequency collision model of the two components and the strong electromagnetic acceleration model of the plasma are further coupled with the fluid part. Various significant parameters such as the plasma density distribution and the ion velocity distribution of the AF-MPDT can be obtained by solving the model with a high spatial-temporal resolution Total Variation Diminishing (TVD) scheme. The acquired simulation results show that the density and axial velocity of both electrons and ions are low when the magnetic flux density is absent or low. Interestingly, with the increase of the magnetic flux density, electrons and ions gradually gather near the anode and cathode, leading to an increase of both the density and velocity, and therefore the performance of the thruster is greatly improved.

Aiming at the problems that the existing gesture recognition technology needs video devices or sensors, is easily affected by ambient light and has high deployment cost, this paper proposes a gesture recognition method based on WIFI signal and lightweight depth network. The amplitude and phase of channel state information are extracted from WIFI signal, the amplitude is filtered, and the phase is unwrapped and linearly transformed. Use normalization and interpolation to transform CSI signals into RGB pictures. A lightweight depth network model based on MobileNet_V2 is used to classify RGB pictures with combined amplitude and phase, and the width factor α=0.5 is changed to reduce the parameters of the model. The channel and spatial attention module are embedded in the inverted residual structure to improve the classification accuracy. Experimental results on public data sets show that the average accuracy rate of this method for recognizing six common push-pull gestures reaches 96.4%. The amount of calculation is 30% of the basic model, and the classification accuracy is improved by 1.3%.

The aim of this paper is to examine the use of virtual reality technology in psychological counselling and therapy. Enter the precise search for virtual reality and psychotherapy through CNKI's advanced search topic with 171 results. Read carefully and thoroughly research and analyse the findings of others, VR technology is a leading-edge human-computer interactive user interface that allows users to fully immerse themselves in a natural environment created by a computer, resulting in an increasingly sophisticated realistic simulation effect. VR technology was once known for its use in the prevention of childhood phobias and PTSD, but in recent years it has been expanded to include applications for autism, intellectual disability and brain damage recovery. Although there are still some shortcomings in virtual reality technology, its use in mental illness has accelerated significantly with the development of science and technology. Virtual reality technology can be combined with counselling therapy to improve the patient's psychiatric symptoms, improve the coordination and motivation of the patient's limbs; allowing the patient to achieve more psychological satisfaction and pleasure in life, thus improving the quality of life.

Social progress has driven industrial transformation and upgrading, and numerous people from physical labor to mental labor so that the proportion of sedentary people is increasing year by year. Poor sitting posture and sedentary are important causes of sub-health today. The premise of correcting poor sitting posture is to identify human sitting posture using realtime monitoring devices such as smart chair. Therefore, it is of great significance to study a sitting posture recognition method. In this article, we propose a data pre-processing method and an improved ResNet for the recognition of sitting posture. Using Cross-mean Filtering and Extreme Value Processing Functions to preprocess data can effectively filter out data noise and reduce the probability of algorithm misclassification. Experiments show that it is easy to analyze the user's sitting posture using this method, and the probability of physiological left-right tilt being misclassified as poor sitting posture is reduced, which means that the improved method proposed in this article has obvious advantages in recognizing sitting posture.

Traditional single-model document-level summary extraction methods lack a multi-level representation of features and cannot extract feature information from documents well. To this end, the text uses the BigBird model and Transformer model to extract the sentence features and document features of the document respectively, to obtain a more comprehensive multi-level feature representation of the document and extract a more accurate document summary. In addition, the BigBird model for the sentence encoding layer, which increases the length of the input text sequence through a sparsity attention mechanism, solves the problem that traditional single-model document-level summary extraction methods cannot focus on more comprehensive contextual information and reduces the model time and space complexity. Comparative experiments are conducted on a generic dataset, and the experimental results demonstrate the effectiveness of the proposed method.

In order to meet the needs of complex equipment group enterprises to carry out model-based definition (MBD) and modelbased system engineering (mbse) innovative development modes, break through key technologies such as design resource management, adaptation and sharing in product collaborative development process, on the basis of integrated management of MBD resources and mbse resources, integrate enterprise mbse design environment, collaborative design system, product data management system and other design environments, build a collaborative development environment throughout the product life cycle, open up the task flow, data flow and knowledge flow in the process of product development, take aviation products as the application verification object, realize aircraft collaborative development based on process and resource integration and sharing, and improve aircraft development efficiency and innovation ability.

This paper takes sentiment analysis as the research direction and collects the data of users' comments on hotel rooms. The research methods used include text corpus preprocessing, Word2vec model, support vector machine, AUC value calculation, feature classification based on supervised learning method, emotion analysis and ordered logistic regression model, etc. Word2vec is used to map words into k-dimensional vector space, and PCA algorithm is used to reduce the dimensionality of the results. According to Word2vec, the comment corpus can be divided into 7 categories of feature dimensions, and then feature recognition is carried out, and emotion scores are calculated by emotion analysis technology. Finally, classify and summarize the emotional scores of each comment in different feature categories. In this paper, through the establishment of Word2vec model, the analysis of consumers' emotional tendency to the hotel, to provide the direction of improving the service level for the hotel, is conducive to the management and development of the hotel.

In the context of the deepening of computer science and technology and digitalization, the innovative application of virtual reality technology in the teaching environment and methods of colleges and universities highlights its value. This paper firstly compares the recent research on the application of virtual reality technology in experimental and practical teaching; next, it elaborates its application status in detail and specifically through three subject areas: political science, medicine, and psychology. The immersive teaching environment built by virtual reality technology provides students with multichannel perception, body movement participation, etc. to help teach more efficiently. Finally, the challenges of this technology being widely used in higher education are presented and an attempt is made to look at its future.

Simultaneous localization and mapping (SLAM) based on vision is a research hot spot in the field of robotics. This paper first introduces the classic framework of visual SLAM, and analyzes the role of each module in the entire framework. Moreover, several classic visual SLAM algorithms are introduced, the principle of each algorithm is explained, the corresponding optimization scheme of each algorithm is given, and the advantages and disadvantages of monocular visual SLAM, stereo visual SLAM, and RGB-D visual SLAM are summarized. Additionally, the development of visual SLAM is prospected in this paper.

As one of the most exciting events of the Winter Olympics, short track speed skating has received widespread attention. The performance of short track speed skating competition is often related to the indicators of athletes' physical state and sports level. How to dig out the key factors affecting performance from numerous indicators has become the most concerned issue for coaches and athletes. Traditional methods mainly use numerical statistics and correlation modeling to analyze the relationship between different indicators and grades. These methods cannot accurately and effectively analyze and evaluate indicators due to rough modeling granularity and incomplete assumptions. In view of this, we propose a comprehensive analysis method for short track speed skating state performance. Specifically, we design a comprehensive importance analysis model based on decision tree, which can be used to model the importance of different indicators to grades. Then, we build a comprehensive evaluation model to evaluate the competitive state and sports performance to help coaches intuitively judge the pros and cons of athletes' state performance. We conduct comprehensive experiments based on simulation dataset. The experiment analyzes the key factors that affect the comparison performance, and evaluates the state performance.

Increasingly complicated environment of intelligent decision-making system and the inherent unexplainable algorithm black-box problem in the decision-making process can easily result in a man-computer trust crisis and affect users' decision-making behavior. Therefore, investigating assessment method of trustworthiness of intelligent decision-making system is extremely urgent under sophisticated application scenarios and high-risk tasks. Targeting the difficulties of trustworthiness assessment of intelligent decision-making systems, an assessment method of the trustworthiness of intelligent decision-making system is proposed in this paper. The existing concept and connotation of artificial intelligence trustworthiness is analyzed to standardize the attribute requirements of trustworthiness of decision-making system combined with complex application scenarios. The index system of trustworthiness assessment of the intelligent decision-making system is constructed based on the analysis of the attribute requirements. According to assessment indices, the trustworthiness assessment method of the intelligent decision-making system is proposed. It provides an important theoretical and technical reference for realizing the cognition of trustworthiness degree of the intelligent decision-making system from a perceptual perspective to a rational perspective.

Chinese relation extraction is designed to extract the relation between a given entity pair from unstructured text. Neither existing character input-based nor word-input methods can take full advantage of the semantic information in the data. Although studies have explored this, there are still problems with the lack of semantic information and the low speed of reasoning. To address the issues, we propose FCW-BERT: Fuse Characters and Words-BERT for Chinese RE, which add word information in a flat structure as input, and enhance the interaction between characters and words with a relative position encoding. Furthermore, we utilize TextCNN for further local feature extraction, experiments on two real-world datasets in distinct domains show consistent and significant superiority and robustness of our model, as compared with other baselines.

Through quantitative and qualitative research methods, this paper investigates the application of online learning platform for English reading among Chinese college students, and puts forward the construction mode of self-learning platform for English reading based on artificial intelligence due to application problems and students’ needs. The survey shows that both English reading websites and smart-phone apps are students’ favorite reading platforms. The former are more comfortable and convenient to read with rich reading resources and the latter are easy to carry and can act as learning assistants. The main problems of the current online English reading platform include: the classification of articles on the website is based on topic but not the difficulty; it has few related background information or extended learning material; it is lack of synchronous exercises covering listening, speaking, reading, writing and translating skills; the knowledge learned cannot be effectively recorded or stored in personal learning corpus. Based on these problems, the AI-based learning platform for English reading lays emphasis on the following aspects: classifying online articles due to their difficulty, supplying reading articles with extended reading material and supporting English exercises, recording key words and accumulating knowledge in personalized self-learning corpus.

With the application and development of advanced technologies such as the Internet, big data and artificial intelligence in the field of transportation, automobiles are undergoing revolutionary changes. From the perspective of design, automobiles have developed from a vehicle to a design object containing personal space, public space and social space. Therefore, the development of automobiles in the direction of intelligent Internet has become a boom and trend of the automobile industry. As the basis of automobile development and testing, the application scenarios of intelligent automobiles are of great importance. This paper will discuss the application scenarios of intelligent automobiles.

In the smart grid environment, the participation of electric vehicles in demand side response will affect the operation of the grid.How to effectively use electric vehicles to make power grid operation more stable will be the future development of smart grid to face. In the light of the uncertainty created by the participation electricity users in the demand side response, this paper establishes electricity users' demand side response model based .By studying the load characteristics of electric vehicles and developing an effective trading mechanism, both power grid companies and users can reach a deal.this paper establishes electricity users' demand side response model based on two kinds of price mechanism (i.e. real-time pricing and time-of-use pricing) to analyse changes in consumer consumption patterns and load transfers. Testing the improved the participation of users in the demand side response model, this paper makes use of numerical stipulations under different restrictions through MATLAB.

Currently, the policy of energy system transformation represents a national strategy. For the development of the renewable energy industry, utility subsidies are required, and the forecast subsidy allocation amount continues to be an extremely important issue. Under this guide, we use a multiple linear regression algorithm to simulate and calculate the undetermined coefficients for wind power, photovoltaic power, capacity of biomass power projects, amounts of on-grid electricity, subsidy period, and the amount allocated by the governments as independent variables, respectively; and the amount to be allocated in the next year as the result. For example, the undetermined coefficients for wind power we calculated are -0.7579, 0.0747, 0.9664, 0.9134, and 47.863, respectively. We then put these undetermined coefficients into multiple linear regression, and obtain a new model for calculating energy subsidies of various types. The results indicate that multiple linear regression plays a significant role in the application of subsidy prediction, and provides a more reliable method for enterprises to estimate the number of subsidies allocated.

Catenary is the core component of high-speed railway traction power supply system, its operation and maintenance safety are directly related to the safe operation of high-speed railway. In order to solve the problem of various parts and defects but rare actual defect samples on catenary support device, a few-shot defect detection method based on manifold smoothing and graph neural network is proposed. Firstly, the catenary parts are extracted from coarse to fine in two stages through the Yolo v5 target detection network. Then, the wide residual network is used to capture the feature representation of the input image, and the graph neural network is used to construct the undirected graph to transmit the label information from the labeled support samples to the unlabeled query samples. Thus, the prediction of the input image is obtained. Moreover, embedding propagation is used to smooth the low dimensional manifold constructed by graph neural network. It updates the nodes in the graph according to the surrounding node states, so as to alleviate the distribution deviation generated in the process of few-shot learning, improve the classification accuracy of defective samples and enhance the generalization performance of the model. Through the defect detection of three parts of the catenary, the average comprehensive defect detection accuracy is 93.9%. The results show that in the case of a variety of catenary parts and a small number of samples, the proposed few-shot defect detection method can accurately locate the parts in the catenary and effectively detect the defect samples.

Object detection is an essential subject in the field of computer vision. It is widely used in various scenes and greatly shares the work of human beings. In view of the lack of real-time performance of pedestrian detection in the field of object detection and the excessive load of data returned by the central server in the cloud computing environment, this paper proposes an improved SSD model under the edge computing architecture. The intelligent mobile terminal completes the real-time data collection, and the edge server undertakes the preliminary calculation, The cloud server is responsible for complex computing and model training. The SSD algorithm is lightweight improved to better adapt to the performance of the edge server. The idea of feature pyramid is introduced to fuse the semantic information carried by the deep feature map with the shallow feature map to improve the detection accuracy of small objects. Through experiments, the model has achieved about 0.72 map and about 0.034s single frame processing speed, which can be applied to real life.

Partial label learning s a type of weakly supervised learning, which learns to predict one label as the correct answer from a given candidate label set. It is a great challenge to improve the accuracy of recognition because of the combination of the two difficult learning conditions, partial supervised learning and partial supervised learning. However, these methods all use similar matrices for tag propagation. We introduced the attention mechanism, proposed a attention-based semisupervised partial label learning (ASPLL) method to address the label contamination issue in PLL through reliable label propagation. Using attention mechanism instead of similarity matrix can significantly improve the accuracy of the algorithm. We evaluate the performance of our method on real-world web image datasets. The experimental results on the web image dataset show that in the semi supervised partial labeling web image classification, ASSPL algorithm is significantly better than the mainstream semi-supervised partial labeling learning algorithms, such as SSPL, PARM.

In order to solve the flow shop problem efficiently, an immune algorithm is proposed to solve the flow shop scheduling problem. The algorithm is designed according to the immune system mechanism of human or other higher animals, takes the scheduling objectives and constraints as the antigen, takes the solution of the problem as the antibody, encodes the antibody with natural numbers according to the workpiece processing order, and takes the reciprocal of the maximum process time as the fitness function. Through the introduction of isolation niche and other technologies, the adaptability of the immune algorithm is improved, the diversity of the population is ensured, and the premature convergence is overcome, The convergence speed is improved. The benchmark test of flow shop problem shows that the algorithm not only has good scalability in the scale of solving the problem, but also is lower than tabu search algorithm and simulated annealing algorithm in operation time, which verifies the effectiveness of the algorithm.

The objects of antacid bacilli microscopic images are small in size and complex in background, traditional detection algorithms have low accuracy in identifying antacid bacilli, while deep learning-based object detection studies lack corresponding datasets and are difficult to detect small objects. To address this two problems, a dataset TB-data for antacid bacilli recognition is constructed by microscopic vision platform, and an adaptation improvement algorithm for antacid bacilli is proposed based on Cascade R-CNN: using a sliding window cutout strategy to cut the original map into multiple submaps and change the relative size of the object to the original map; using Multi-scale Feature Map Stack method to repeatedly fuse different resolution information for multi multi-scale information.The experiments show that the AP50 of the improved model can reach 86.5% and has a frame rate of 16.1fps, which is 7.8% higher than the AP50 of Faster R-CNN, with real-time, accurate and lightweight features, providing a new solution for the recognition of antacid bacilli microscopic images.

With the increasing popularity of the Internet, network security issues are also increasing and there are more and more fraudulent web pages, which also brings great obstacles to the governance of network security. However, for the detection of fraudulent web pages, most of the previous detection methods are based on web page characteristics. Multiple information on web page content, such as web page text, HTML, images, and other content is also an important feature for detecting and discovering fraudulent web pages. In this paper, the text dataset of fraudulent web pages obtained by the crawler technique is used to extract the features of the text content sentences using the bidirectional LSTM model, and the feature weights of the text dataset of fraudulent web pages are enhanced by introducing an attention mechanism to obtain a feature vector representation of the text data sentences and then the classification detection is carried out. The experiments show that the accuracy and F1 values of the bidirectional LSTM model are higher than those of other models. The bidirectional LSTM model on the addition of the attention mechanism performed better, yielding an accuracy result of 90.85%, which is a 2.7% increase in accuracy and 3.09% increase in F1 value compared to the bidirectional LSTM model. Therefore, the method is effective in detecting fraudulent web pages and has some practicality in cyber security governance.

A new adaptive image fog removal model is designed, which combines multi-scale guided filter algorithm with hue model, in order to solve the problems that traditional methods can easily cause the result image to appear contrast loss, halo obvious, color distortion and color bias serious. The paper proposed a new transmittance estimation method which combines multi-scale guided filter and dark channel prior theory, aiming at preserve the color information belongs to the images better. In addition, a more efficient structure preservation filter is introduced to optimize the coarse transmittance distribution and improve the overall effect of the defrosting result image. As we can see during the result, when compared to the classic algorithm, the proposed method to this paper can not only achieves better fog removal effect, but also effectively suppresses halo and dispersion.

Aiming at the problems of low model detection speed and low detection accuracy caused by pedestrian occlusion and changing pedestrian poses in dense pedestrian detection, a dense pedestrian detection algorithm with Hierarchical Vision Transformer Prediction Head (STPH)-YOLOv5 is proposed. Based on YOLOv5, the original YOLOv5 prediction head is replaced with a Hierarchical Vision Transformer Prediction Head (STPH) to explore the prediction potential with selfattention mechanism. In addition, a coordinate attention mechanism is integrated to accurately capture the location and orientation information of pedestrians in the image. In order to avoid the loss of feature information caused by the deepening of the network, the fused multi-scale feature network (BiFPN) is used in the neck part of the improved algorithm to replace the PANet in the original YOLOv5, which makes the prediction network more sensitive to targets of different sizes and improves the overall model. The detection ability reduces the missed detection rate and false detection rate of the model.

Grayscale image colorization is a process of adding reasonable color information to an image, and converting grayscale images into color images is an important and difficult image processing task. The process of colorization is to predict the color information corresponding to the grayscale image by the colorization model. In this paper, the proposed multi-scale input adversarial generative network coloring model with multi-scale input is used in the generator, and the input condition map is fused with the network feature map using the input fusion module so that the network can better utilize the information on each scale of grayscale image for color prediction and improve the effect of colorization. The effectiveness of the generator in this paper is verified experimentally. And it is compared with several existing methods to prove the improvement of the colorization effect of this paper's method.

Based on the characteristics of power load and considering various meteorological factors, this paper improved BiLSTM model for forecasting. On the improved Bi-LSTM model, the prediction effect of historical power load data at the peak is significantly better than the original model; Then, considering the meteorological factors mined by KNN algorithm, different meteorological factors and historical load data are used as the input end of the prediction model to predict, and the corresponding evaluation criteria are obtained. Simulation results show that the prediction accuracy is improved after considering the influence of multiple meteorological factors. Compared with the previous methods, the proposed method has higher prediction accuracy.

The domestic urban traffic pressure and road congestion are serious. The vigorous development of public transport is undoubtedly an effective measure to solve urban traffic problems without increasing or expanding roads. Among them, the bus signal priority control system is one of the main ways to improve the service level and operation efficiency of public transport. However, due to the complex urban traffic road environment in China and a large number of non-motor vehicles and pedestrians, it is difficult to achieve the precise positioning and detection of public transport vehicles. Therefore, the research on the bus signal priority control system in China is difficult to be applied to practical projects. Due to the complex urban traffic road environment in China and a large number of non-motor vehicles and pedestrians, it is difficult to achieve the precise positioning and detection of public transport vehicles. Therefore, this paper puts forward the concept of a precise bus signal priority detection system and improves it from three aspects: the acquisition of a precise lane-level bus signal priority control area, the high-precision positioning of public transport vehicles, and the optimal detection location setting of bus signal priority control area. In order to achieve the high-precision positioning of public transport vehicles in the bus signal priority control area, the optimal detection position of the bus signal priority control area is set to achieve the precise control of bus signal priority. Based on Beidou high-precision positioning and UAV aerial photography, this paper proposes a method to obtain an accurate lane level bus signal priority control area. The experimental results show that the method can obtain an accurate lane level area with an error of less than 5 cm.

With the rapid development of new energy vehicles, problems such as “charging difficulty” are becoming increasingly prominent. The charging space is occupied by fuel vehicles, resulting in the inability of new energy vehicles to charge after they arrive at the charging station. This paper proposes a license plate recognition algorithm model which combines the three steps of vehicle detection, license plate detection, and license plate recognition, which improves the robustness of license plate recognition in complex scenes. In view of the lack of Chinese license plate data, the data enhancement method is adopted to improve the recognition accuracy of Chinese provincial abbreviations and common color license plates such as blue, green, and yellow. According to the characteristics of a small license plate target and easy rotation in the license plate recognition task, the WPOD-NET network is selected for license plate detection. Finally, the license plates of new energy vehicles and fuel vehicles are distinguished by recognizing the color and law of license plates, so as to provide supporting information for the generation of fuel vehicle occupancy alarm. In this paper, the data collected by the cameras of some charging stations in a city are used for verification. The verification results show that the accuracy of oil vehicle occupancy recognition is as high as 97%.

This paper study the path planning problem of autonomous underwater vehicles. We first give a discrete potential field method, which indicates that the global optimal path can be found by searching along the direction of the decreasing potential energy. Then we propose an improved reinforcement learning path planning algorithm, which uses discrete potential energy as the reward function to speed up the convergence of the algorithm and realizes the path planning in the flow field by modifying the reward function. Finally, we demonstrate the effectiveness of the proposed algorithm by simulation experiments.

The traditional A* algorithm has many turns, long paths, and slow speed in the global path planning of unmanned vehicles. For these problems, this paper proposes an improvement scheme. Add a reverse search from the end point to the starting point and generate a straight-line path between two nodes when judging that there is no obstacle between the target nodes in both directions. In addition, redundant nodes are eliminated, and the path is smoothed based on cubic B spline curves. Simulation results show that compared with the traditional A* algorithm, the path planned by the improved A* algorithm has shorter length, fewer and smoother turns, and the planning time is reduced by 14%-75%, which achieves the optimization of the path planning of the A* algorithm.

Aiming at the problem that the vehicle cannot accurately determine whether there is a malicious lane change of the preceding vehicle during high-speed cruise, a method based on recurrent neural network (RNN) is proposed to automatically determine the malicious lane change of the vehicle during high-speed cruise. Use HOG, Hear-like, SURF, LBP features to be fused into the SVM algorithm to detect the front sight distance environment, simulate different malicious lane changing scenarios as input features, preprocess the data and import it into a recurrent neural network to build an RNN model. Learning and predictability determine the tendency of vehicles to change direction maliciously and use the grey prediction method (GM model) to predict the behavior of vehicles maliciously changing lanes, so as to maintain a minimum safe distance between vehicles, thereby improving the safety of high-speed cruising and reducing the loss of traffic efficiency in the application process of variable speed limit control. Through the simulation experiment, the system can correctly judge the vehicle malicious lane change with an accuracy of 99.2%.

As one of the most important energy, power quality is very important to the safe and stable operation of distribution network. According to national standards and evaluation requirements, a power quality evaluation method is proposed based on Minibatch K-Means algorithm and random forest algorithm. At first, abnormal data are processed. Processed data are clustered analysis by Minibatch K-Means algorithm. Imbalance sample is processed for clustering data by SMOTE. At last, labeled data are extracted features by random forest algorithm. The training model is saved. The simulation results show that power quality is evaluated by Minibatch K-Means algorithm and random forest algorithm, which solves the problem that labelless power quality data can't be classified by integrating multiple indexes. It achieves a comprehensive, efficient and rapid evaluation of power quality in distribution network.

Recommendations based on similarity search are widely used. However, when the number of users and items is large, it may become an efficiency bottleneck. In this paper, we propose a high-precision, modular, and efficient recommendation component: the binarization encoder. We use Siamese neural networks to jointly learn binary codes of users and items. The Siamese networks are composed of two autoencoders, whose encoder and decoder are sharing weights further. Their latent factors are discretized as efficient representations of users and items. In order to improve the learning ability of this model, a comprehensive objective function is proposed, which considers three objectives: improving the reconstruction quality of input vectors, reducing the information redundancy between dimensions, and finetuning the predictions with observed ratings. We use collaborative filtering based on matrix factorization on real-world datasets as basic systems and evaluate the proposed method. The experimental results prove the effectiveness of our method.

The introduction of Industry 4.0 brings opportunities for the transformation of production modes. As an important link in the production process, the production logistics process often cannot be accurately evaluated. This paper focuses on the production logistics process in the workshop and analyzes the logistics performance indicators evaluation system from different dimensions. A performance indicators evaluation system oriented to Industry 4.0 is designed and proposed, which uses original data from the production site to calculate or evaluate the indicators, to provide reliable data for the production system optimization.

Target detection in autonomous driving scenes refers to the output of environmental target information by relevant algorithms after the vehicle collects all-round image, radar, GPS and other information through sensors. Target detection actually includes two tasks : positioning and classification. The positioning task outputs the location information of the target relative to the vehicle, and the classification task needs to output the specific categories of the target. This paper mainly studies 3D target detection based on monocular RGB image. This method only needs monocular camera, which has the lowest cost and the fastest running speed.

Dynamic data flow is a data model with the characteristics of high speed, high dimensionality, change, and potential infinity, and its data distribution evolves over time. Broad Learning System (BLS) is a learning model with fast training speed, simple structure and high accuracy. It has the characteristics of online learning and rapid adjustment of the model structure. The traditional online BLS algorithm cannot capture the changing trend of the dynamic data flow with time in time, so it is not suitable for dealing with the dynamic data flow problem. This paper proposes a novel online broad learning algorithm (MDO-BLS) that combines sliding window strategy and forgetting factor method. This BLS model has a forgetting mechanism, that is, the model has the ability to learn new samples online while forgetting the knowledge of old samples, so that it can quickly capture the trend of data changes. To verify the performance of the proposed BLS model with forgetting mechanism, experiments are performed using a dynamic regression dataset. The experimental results show that the BLS model with forgetting mechanism is superior to the traditional BLS and some other online algorithm in both prediction accuracy and time overhead, so it is more suitable for application in dynamic data flow environment.

Aiming at the complex optimization problem of power electronic distribution network planning, a multi-objective genetic algorithm based on Multi Strategy improvement (genetic algorithm for short) is proposed. Combining genetic algorithm with distribution network planning effectively, the chromosome group coding mode of genetic algorithm in planning scheme is studied; The multi strategy improvement of genetic algorithm involves the improvement of population selection, crossover and mutation operators and adaptive genetic operators; Through the heuristic algorithm, the ability of searching the optimal solution of the population can be improved. The performance of the improved genetic algorithm based on multi strategy is tested by Schaffer function and Griewank function, and its composition, search characteristics and search optimization process are analyzed and discussed respectively. The results show that the improved genetic algorithm based on multi strategy has great advantages in search accuracy and calculation efficiency, and is of great value for distribution network planning and optimization.

Improving the resilience of urban integrated energy systems is of great significance for dealing with extreme events. For this reason, a high-resilience urban integrated energy system is discussed. Firstly, the architecture of the high-resilience urban regional energy interconnection system is proposed, and the content of the architecture is expounded. Secondly, the response process of the system under extreme events is analyzed. Finally, a series of indexes related to resilience assessment are proposed to measure the resilience of the system.

In order to solve the problem of optimal operation and revenue distribution of multi-energy power stations (data centers, electric vehicle charging stations, distributed photovoltaics, wind power, and energy storage stations), this paper proposes a multi-station integration collaborative optimization strategy. First, exploring the energy supply mode for the data center under the multi-station integration scenario, and the subjects in the multi-station integraion mode are divided into three subjects: the power grid company, the equipment investor and the data center. Second, various deveices involved in multistation integration are modeled ,and the modeling of energy storage station is combined with special requirements for the reliability of the data center. Practical demand is taken into account in the modeling of charging station as the electric car is divided into three kinds of charging methods. Considering the randomness of electric vehicles, the interconnection time and initial SOC are randomly generated, and subject to the charging pile number and power limit. Later, optimal dispatch is carried out with the maximum benefit of the equipment investor as the objective function. Finally, the analysis of the results shows that the multi-station integration mode can bring benefits to the above three subjects, and it is a win-win mode, which has a development prospect.

At present, the training of UHVDC transmission system mainly uses simulation software to carry out operation skills training and technical skills improvement training for the operation and maintenance personnel of the converter station. However, the UHVDC transmission system has plenty of electrical equipment, the operation steps are rigorous, and the operation logic is complex. In the actual operation and maintenance work, for the same operation task during the UHVDC system switching operation, due to the different order of equipment operation, it has multiple operation procedures. If only one evaluation logic is adopted, the evaluation of simulation operation will be inaccurate. In order to solve this problem, an evaluation system based on the state logic algorithm of UHVDC transmission system is designed in this paper. Firstly, the position state database of system equipment is established, secondly, the logical sequence relationship of the action logic of the UHVDC transmission system is established. Finally, pre-set evaluation criteria for switching operation are constructed. According to the operation state change logic in the simulation operation record, the evaluation result is obtained by comparing with the pre-set evaluation standard. After the algorithm is implemented, the simulation operations that meet the operation expectations and the state logic of the UHVDC system are all correct operations. The evaluation system provides fast, efficient and accurate evaluation for the digitally simulated UHVDC system switching operation.

Undetected elderly fall will result in serious consequences, therefore, in the informatization of elderly care services, fall detection and intelligent warning for the elderly at home is increasingly significant, which has become a hot spot of the current research. A fall detection scheme for the elderly in embedded devices by using the video stream analysis method based on the target detection model and combined with the change of human posture to determine the fall state is proposed in this paper. The clustering algorithm is used to analyze the centralized value of the aspect ratio of the dataset according to the characteristics of the single-type target detection scene in the application and the aspect ratio of the anchor in the selected SSD algorithm is reset. The method of self-adaptive judgment of positive and negative samples in China improves the identification efficiency of the elderly by using the characteristics of the data set itself. Theoretical analysis and experimental results show that the accuracy of elderly identification is increased by 5.7%, and the accuracy of fall event judgment is increased by 6.3%.

Based on the public-oriented complex environment and complex tasks, this paper designs an autonomous navigation robot platform for task scheduling, and studies its key technologies. Aiming at the problem of simultaneous localization and mapping (SLAM) of autonomous navigation robot platform in complex environment and multitasking, a hybrid particle filter algorithm (M-PF Mixture Particle Filter) is proposed to solve the particle attenuation problem encountered in traditional particle filter. A global path planning algorithm is proposed on the path planning problem. Finally, the human skeleton point recognition is used to achieve the specific functions. In this paper, TSRP- HCI robot platform is used to test various methods proposed in this paper for specific indoor cruise scenarios. The test results show that the task scheduling oriented autonomous navigation robot platform proposed in this paper has advantages in positioning accuracy, navigation safety, algorithm efficiency and human-computer interaction mode, and has considerable application value. It can interact with human beings in a dynamic environment and carry out task scheduling at the same time to complete the navigation task from the target point to the termination point independently.

Under the vigorous promotion of global energy Internet, China's AC/DC power grid construction has been developed rapidly and both artificial intelligence technology and machine learning technology have been developed by leaps and bounds. Computer-based image pattern recognition technology is an important part of computer application technology. With the rapid development of digital image processing technology and pattern recognition technology, image pattern recognition technology has been more comprehensively applied in various aspects such as industry, agriculture, national defense, scientific research and medicine. In this paper, we introduce the image detection technology and image detection algorithm for substation inspection robots and acquire image data files through image acquisition terminal sensors to realize the categorization of feature amount. So as to realize the annotation of pre-set features based on the training model. An optimized classifier and optimized image features are used to build an efficient and accurate image classification model. Artificial intelligence image recognition technology for power equipment image recognition is important to further improve the automation and intelligence level of substations.

Fake news has always been a problem for online communities. It can build mistrust, create boundaries between people, and lower productivity in many government departments. Distinguishing fake news by human is slow and expensive, therefore, machines were proposed by many scholars to do the job. There are mature machine learning algorithms available to use for distinguishing fake news and real news. Despite the high accuracy of most well designed machine learning models, they can still make mistakes on some news. In this paper, we compared three machine learning models for news classification and evaluate biases in these models. The first one is Embedding Bag, second one is CNN network, and then the Naive Bayes Classifier. After tidying and vectorizing the data, they were feed to these classifiers to get the model. As a result, all of the models achieved very high accuracy. All of them had an accuracy of higher than 90 percent. The biases for all of the approaches are primarily sampling biases. The less represented words have higher chances to be mis classified than those more represented words in the database.

Text-based cross-modal person re-identification refers to the retrieval of the corresponding identity image in the pedestrian image set through the linguistic description of the pedestrian. Aiming at the problem that there is no interaction between the image-text feature extraction branch, we propose a text-to-image person re-identification method based on mutual teaching. The proposed method adds interactive operation in the image text feature extraction part to make the image and text information respectively guide the feature extraction of another modality, which realizes the information transfer between image and text modalities to close the distance between two modalities. The method is experimentally verified on the CUHK-PEDES dataset. The experimental results show the effectiveness and superiority of the proposed method.

Sleep scoring is manually marked by professional neurophysicians, which is a time-consuming and subjective error prone work. An automatic classification algorithm dealing with this task is of great significance to auxiliary medical diagnose. In this paper, data set are collected from the clinical physiological signals of 7 patients from Xianga Hospital. Signal preprocessing and feature extraction algorithm combines the advantages of principal component analysis and wavelet decomposition in only single EEG channel. In result, the five fold cross validation on the data set of all 7 patients has achieved an average accuracy of 77.18% and the highest accuracy of 83.24%. In addition, it is also found that the stable and optimal EEG channel is C3-M2, and the extracted 41 wavelet statistical features are not sensitive to signal noise, and it is found that the random forest model has better classification performance.

The torque pulsation generated by brushless DC motors during motion can lead to reduced system stability and reliability, as well as mechanical vibration, resonance and noise problems. And they require the installation of position sensors to provide commutation signals, which in turn increases the complexity and cost of the motor structure. To address these problems, a research strategy combining traditional dual closed-loop control with the extended Kalman filtering algorithm is proposed based on the study of traditional motor control. On the basis of the traditional double closed-loop control system, a feedback signal processed by extended Kalman filter algorithm is added.The simulation results show that this strategy can get accurate estimation of speed and rotor position, and the system with the extended Kalman filtering algorithm is more stable and better controlled when the motor speed is given and changed.

Based on the foundation design of an onshore wind power project in China, the finite element analysis by Midas GTS NX is used to study the stress distribution and the overall displacement of the wind turbine foundation anchored by prestressed rock bolt. This work focuses on the position of the most unfavorable stress of foundation under load condition, and the influence of different bolt diameters on the overall displacement of foundation. The results show that the stress at the junction of foundation and platform column, and the anchor end of rock bolt are significantly larger than other positions, which should be paid attention to in design. The foundation displacement decreases with the increase of bolt diameter, but when the bolt diameter is greater than 60mm, the change trend decreases. Safe and economical bolt diameter should be selected in the design.

The operating range of airborne infrared warning systems is significantly reduced at low altitude, this will increase the difficulty in detecting the target of low-altitude aircraft, so it is very important to study the operating range of airborne infrared warning systems at low altitude. Firstly, an infrared detection model including target, atmospheric transmission, and detector is established. Secondly, the cloud, rain, snow, and ground disturbances at low altitudes are analyzed, and the operating range models are established. Finally, the operating range of the airborne infrared warning system at low altitude is simulated, simulation results show that the mid-infrared band meets the airborne infrared warning system requirements at low altitude, the models can play an important role in the quantitative study of the detection performance of the airborne infrared warning system at low altitude.

With the increasing demand for product customization, Single and Small-batch (SS) discrete manufacturing has emerged, but there is lack of scientific quality control methods for this manufacturing mode. This paper analyzes the mechanism of quality problems in the discrete manufacturing process and compares the differences between SS production and massproduction. In order to cope with the environmental changes in SS production, a domain adaptation-based model is proposed to minimize the discrepancy between source domain and target domain, which enhance the generalization ability of the prediction model in SS production.

It is conducive to fault treatment and protection to effectively identify the type of single-phase high resistance grounding fault in distribution network. A single-phase grounding fault type identification method, based on improved residual network, is proposed. Firstly, the direct features of waveform data are automatically obtained by using the feature extraction module in the residual network.The time-frequency characteristics of the original fault electrical quantity are analyzed by Fourier transform, and the energy domain characteristics are added on this basis.Secondly, the manually extracted features are replaced by the original waveform data, and the single-phase grounding fault classification model of distribution network based on improved residual network is established. Finally, through the verification of field recording data, compare the original fault data with that after extracting multi-domain features. It is proved that this method can quickly and effectively complete the high resistance grounding fault identification, and has no requirements for the actual waveform length and sampling rate, so it is more practical.

Since the existing methods for the simulation and deformation analysis for ultra-thin plate hull segment hoisting are poorly specialized, and relying on experience is prone to deviations, it is difficult and inefficient to directly use finite element analysis. In response to the above problems, a 3D visualization simulation analysis method for deformation of ultra-thin plate hull segment hoisting is proposed, and special hoisting tools are designed to optimize the hoisting process. Based on the dynamic virtual simulation system specially designed for the hoisting of hull segments, the tensile force during the hoisting process of ultra-thin plate segments is obtained, and then combined with ANSYS software, the finite element analysis of the deformation during the hoisting process is carried out. The experimental results show that compared with the traditional hoisting process, the improved hoisting process in this paper can effectively control the stress and deformation of hull segments, realize the accurate simulation of the hoisting of the ultra-thin plate hull segments, and have a great reference significance for the actual hoisting process planning.

The connector for extravehicular operation by astronauts is a key component for repairing devices in the space station. The paper studies mechanism of the vacuum cold welding phenomenon of the connector, analyzes its threats to the connector, provides appropriate measures for preventing vacuum cold welding and verifies these measures through experiment. The results indicate that the connector does not occur vacuum cold welding phenomenon during the experiment and the above measures are effective. Therefore, the study is important and significant for ensuring the long-term and reliable operation of the connector in the space station.

Aiming at the shortcomings of existing photovoltaic monitoring systems, a photovoltaic monitoring system based on Cat.1 medium-speed IoT network was designed. The system used the combiner box as the monitoring node, STM32 as the development platform, the communication module adopted USR-G771 DTU, and the cloud platform was the Tlink IoT platform. After uploading to the cloud platform, people can log in to the cloud platform using WeChat or a browser to view the operating status of the photovoltaic system. Experiments show that the system can monitor the photovoltaic system for a long time, with good real-time data, high signal strength, low cost, easy delivery, and low power consumption, which effectively ensures the stable operation of the photovoltaic system and saves the investment of manpower and material resources.

Based on the application requirements of the complex border defense and coastal defense, an intelligent photoelectric detection and identification system is proposed. This paper introduced its main components and working strategy. The system adopts dual band detection of visible and infrared light to realize all-weather detection of specific regions, and uses the intelligent identification technology of image features to implement the accurate identification of specific targets. The integrated communication mode of ground-based 4G communication and low earth orbit satellite communication is selected to quickly upload the compressed detection information, to ensure the stability and continuity of communication in various environments. The system is expected to exert an important effect on target detection and identification in the field of border and coastal defense in the future.

The cable patrol patrol robot can distinguish the fire door by image recognition technology and realize the function of crossing the fire door autonomously. However, the effect of image recognition is greatly affected by light intensity. In some cable tunnels, the dim environment makes the effect of image recognition fire door poor, which seriously affects the passage of robots. In order to improve the identification accuracy of fire door in dim environment, a method of tunnel fire door identification by patrol robot in dim environment was proposed. Firstly, the original image was preprocessed, and the image enhancement technology was mainly used to enhance the details of the collected image through guided filtering. The saliency detection algorithm based on feature clustering and multi-scale fusion can effectively reduce the interference of background information and false target, and obtain the feature image of target image. A target recognition network based on improved YOLO-V4 was adopted, which added one auxiliary network, which can effectively improve the performance of the whole feature extraction network. The attention mechanism was adopted to fuse the feature information of the auxiliary network and the backbone network, which enhanced the effective information channel, suppressed the invalid information channel, and improved the efficiency of network recognition. The experimental results show that this method can improve the accuracy of image recognition in dim environment and effectively solve the problem that patrol robot cannot accurately identify the fire door due to insufficient light.

The barrel finishing is an important part of finished machining in advanced manufacturing technology, and the abrasive media is the key factor affecting the machining effect and efficiency. Rules construction is the key to the expert system being applied to the optimization of the abrasive media in the barrel finishing. Currently, when the type of parts has changed, the abrasive media optimization accuracy of the original expert reasoning model decreased. Therefore, the paper proposes an automatic construction method of interval value rules for the abrasive media based on hierarchical classification. Firstly, the existing case base is optimized by the improved fuzzy C-means (FCM) algorithm of the subtraction cluster to obtain the training set. Secondly, each eigenvalue grade and the affiliated interval are determined, and the improved hierarchical classification generating representation is used to establish interval value rules. Finally, simulation results show that the comprehensive accuracy of the improved expert reasoning model is 93.76%, whichis3.29% higher than the original expert reasoning model, and the decision-making guidance for the development of new processes is provided.

Deep learning technology is widely used in side-channel attacks, improving the ability of profiling side-channel attacks. A model based on Bagging ensemble strategy is proposed to improve the classification performance of neural networks in Deep Learning based Side Channel Attacks based. In this model, the weighted voting method based on class probability is used to combine each base model, and the difference of each classifier is fully combined. The weighted coefficient is calculated by reciprocal method to improve the classification accuracy. Finally, the maximum likelihood estimation is used to recover the key to further improve the attack efficiency. Experimental results show that when attacking open data sets, the side channel attack method based on Bagging integrated strategy improved model has a higher success rate, and the number of attacks required for successful attack can be reduced by more than 16.3%.

For power grids, reactive power optimization is a key research problem. Increasing reactive power affects the security. The purpose of this work is to study the optimization of swarm particle basis. Optimal energy system regression. The role of progressive group optimization in solving the problem of reactive energy enhancement is being studied in depth, how to improve the problem to be solved when the PSO algorithm is applied to energy optimization and upgrading is closely explored. The coupled equation is the most iterative and uses Newton's method to calculate the power flow. In the two grid systems of IEEE14 interface and IEEE30 interface, the difference, advantages and disadvantages. The experimental results show that the network loss of the advanced particle optimization swarm is 4.86 MW less than that of the standard particle optimization.

Accessibility of special vehicles on urban roads is the fundamental guarantee for dealing with emergencies. On the road in China, especially in the high-speed gate area, special vehicles such as ambulances and fire engines are often blocked. Given the situation in which a particular carcass through the highway, this paper aims to prioritize special vehicles and promote the active concession of social vehicles for special vehicles in space. This paper proposes a method for determining the transfer of special vehicles on highways. For social vehicles in high-speed door areas, the violation of compaction lines caused by avoiding special vehicles can be automatically eliminated, eliminating the concerns of social vehicle drivers, improving the initiative of transferring unique cards, and realizing the priority of special vehicles.

Ultra-redundant Manipular is a sort of highly flexible bionic robot. It has small cross-sectional profile and is highly adaptable to multi-obstacle and non-structural environments. Different tools can be installed on the end effector to accomplish certain complex operational tasks. Ultra-redundant manipulator can operate movement in confined spaces such as pipes, aero engines and post disaster debris. Reasonable path planning method is required to avoid collisions between the ultra-redundant manipulator and the environment during task execution. An improved head-steering path planning method is proposed in this paper. A nonlinear programming approach is used to solve the optimization problem, and the minimum value of the nonlinear objective function is solved to ensure that the end of the ultra-redundant manipulator follows the target path with minimum error, and to realize the "collision-free" path of the ultra-redundant manipulator in the non-structural environment. In this paper, firstly, the kinematic model of the ultra-redundancy manipulator is established, then the objective function of the path planning method is formulated. Simulation experiments are conducted for aircraft engine inspection and the approach of target points in a narrow space. The experimental results show that the head-steering path planning method proposed in this paper is effective and feasible.

Due to the irregularity of taking the elevator every year, there are endless cases of accidents. Therefore, this paper designs and implements an intelligent elevator monitoring system. In order to realize the drive of the USB camera, the system adopts the Android bottom V4L2 protocol video image acquisition, and adopts the Android Studio software realizes the development of intelligent monitoring APP based on Tensorflow framework, which can realize real-time detection and monitoring of functions such as illegal parking of self-propelled (electric) vehicles, illegal smoking, and no masks.

With the gradual increase of ev penetration, a large number of charging piles are putting into operation. As a power electronic device, the power quality problem of charging piles is prominent, which will affect the power grid and nearby equipments. Focosing on the problem of difficult field detection, this paper studied the overall architecture of plug and play test system and completes the design of detection system device, communication system and platform application. In this paper, the actual test of the test system was carried out to check its function. And the operation and maintenance strategy of charging pile based on power quality was formulated by referring to the existing standards. It has been proved that the research results of this paper can provide technical support for the intelligent operation and maintenance of charging piles.

Many large bridges are being built across large rivers and the height of ships is also increasing. Due to the super high hull, the bridge structure (cross-sea bridge) across the water route has frequent ship collisions, which not only threatens people's safety, but also brings immeasurable economic losses. In addition, the resulting bridge damage and collapse hazards are huge; the risk of the ship hitting the bridge cannot be ignored in the process of bridge structure planning, design and operation throughout its life. Therefore, in view of the phenomenon of ultra-high ships in clear sea areas, a detection algorithm for ultra-high ships based on lidar is proposed. First, the single-line lidar based on ROS is used to obtain environmental information, and the lidar data is processed by the median filtering algorithm for the clear sea area, so as to remove the erroneous point cloud data caused by noise and avoid affecting the subsequent analysis. The point cloud data of some sectors will be deleted through the laser_filters plug-in that comes with ROS, and then the filtered point cloud will be clustered using the adaptive density clustering algorithm DBSCAN, and then the Graham scanning method will be used to establish an external convex hull for the point set. Mark the superelevated ships. The results show that this study can quickly detect ultra-high ships in the clear sea area, and can track objects in real time. Make further contributions to the detection and research of ultra-high ships in clear waters.

Predictive control is based on a certain model, which uses the past input and output to predict the output at some time in the future. It can obtain the optimal control strategy for the current and future sampling periods, by calculating with control constraints and prediction error. In this paper, a predictive control algorithm of the position loop of the antenna control system is proposed. According to the required maximum speed and maximum acceleration, the speed is reasonably predicted, the motion of the antenna is planned, and the fast and stable tracking of the target is realized. The algorithm has simple parameters and strong adaptability.

Wind turbine failure data is only a small part of the huge amount of operational data. Classification of a small amount of fault data often leads to overfitting of the model. To address the above problem, a few-shot learning method for wind turbine fault classification is proposed by using a prototypical network with Mahalanobis distance. This is done by preprocessing a small amount of high-dimensional fault data using PCA and LSTM methods to obtain training samples for a prototypical network of Mahalanobis distance. The distance function that fits best to the wind power fault data is found by training as the distance function of the prototypical Mahalanobis distance network. The experimental results show that The accuracy of few-shot learning for wind turbine fault classification using the Mahalanobis distance prototypical network is much higher than that of conventional machine learning classification.

The computer vision-based automatic aiming system is proposed for the problem that the accuracy of firearms and other devices decreases due to psychological and environmental factors during the confrontation between police and criminals. The scheme first performs processing operations such as semantic segmentation and edge detection on 8,571 images and then obtains a training model by the YOLOv5 algorithm. Finally, the trained model is used for image verification and real-time detection to effectively track targets and mark the killing range. The experiments show that the system has the effect of improving shooting accuracy and can provide theoretical support for the subsequent development of assisted shooting hardware.

The wear of the brake pad will lead to the increase of the braking gap, thus prolonging the braking time and threatening the driving safety of the driver. In this paper, the working principle of each functional component of the EMB actuator is analyzed, and the simulation model of each component is established. At the same time, a three closed loop PI controller suitable for the actuator is designed, and the control strategy considering the braking gap is added. The expected clamping force is applied at the control end to compare the changes of the three parameters of EMB, namely, clamping force, motor speed and screw displacement, with and without brake clearance control. The simulation results show that this strategy can eliminate the braking gap as soon as possible and improve the response speed of the actuator. Moreover, after the braking, the screw can quickly return to the initial position to prepare for the next braking, which effectively avoids the adverse braking conditions that may occur due to the change in the size of the braking gap.

A heavy-duty industrial robot with a load of 3.5kn and a positioning accuracy of 0.05mm was combined with the end effector of friction stir welding and refitted into a robot friction stir welding system. In the friction stir welding machining test, it was found that although the structural strength of the robot met the use requirements, its stiffness under some working conditions didn’t meet the requirements, resulting in low positioning accuracy and unsatisfactory machining accuracy of the friction stir welding robot. To solve this problem, a closed-loop controlled heavy-duty friction stir welding robot system based on ultrasonic positioning technology was designed. The system was composed of dual ultrasonic transmitters, multiple ultrasonic receivers, positioning processor. Two ultrasonic transmitters with different transmission frequencies were arranged at two different positions of the end effector, and then two groups of receivers that could distinguish the ultrasonic frequency were used to receive the ultrasonic signal. The real-time coordinates of the two ultrasonic transmitters in the positioning space were calculated by the positioning system to determine the position and posture of the end effector. The end effector was controlled to track the predetermined trajectory in real time to realize the closed-loop control of the robot. Based on the three-point positioning principle, the relevant calculation and analysis were carried out, which proved the rationality and feasibility of the system in theory. The experimental results show that the positioning accuracy of the system is within 1mm and the posture error is within 0.01˚.

With the rapid development of mobile service robot industry, its application scenarios have become more complex and diverse. SLAM (Simultaneous Localization and Mapping) with a single sensor is difficult to meet the application requirements. Multisensor fusion for accurate positioning and mapping navigation has become one of the research hotspots in this field. In order to solve the problem of single sensor simultaneous localization and map construction in indoor small environment, the contour is fuzzy and the effect is not ideal. Based on ORB-SLAM2 fusion and improved Gmapping algorithm, this paper proposes a multi-sensor real-time localization mapping scheme based on laser radar and vision fusion. This algorithm combines the visual information provided by the depth camera and the scanning frame provided by the laser radar to make up for the shortcomings of a single sensor and form a real-time positioning and map building system with higher robustness. The experimental results show that the integrated SLAM scheme can obtain a more comprehensive and accurate map, which meets the expected design requirements.

Under the background of vehicle lightweight and electrification, distributed drive electric vehicle has gradually become a hot direction of vehicle research. This paper introduces the characteristics of distributed drive electric vehicle, the layout configuration of distributed drive system and the technical difficulties faced by four-wheel hub motor, analyzes the research direction and development status of dynamic control technology of distributed drive vehicle, points out that improving the effectiveness and robustness of control algorithm is the main direction of future research, and looks forward to the development prospect of distributed drive electric vehicle, It can be used as a reference for scholars engaged in research in this field.

In recent years, the probability of large-scale power outage in distribution network caused by extreme events continues to increase, which poses a serious threat to people's life, market economy and social stability. With the continuous development of renewable energy, more and more distributed power sources are connected to the distribution network, which significantly improves the power supply reliability of the distribution network and also provides a way to ensure the continuous power supply of key facilities in extreme events. Based on the analysis of the severe damage caused by extreme events to distribution network, this paper summarizes the current power supply recovery ideas, power supply recovery model and power supply recovery algorithm of distribution network under extreme events, and finally prospects the development trend of power supply recovery problem of distribution network under extreme events.

In the context of the dual carbon target, China has continued to adjust the share of renewable energy. In this paper, wind power generation and photovoltaic power generation systems are coupled to the CCHP system, and the excess power of the system is stored through hydrogen production. In this paper, the physical and mathematical models of the RE-CCHPH2 system are established by using mixed integer programming model. The objective function of optimization is the lowest system cost. The system's cooling, heating and electrical load requirements can be met by optimizing equipment capacity. Finally, the system is evaluated with primary energy saving rate, carbon dioxide emission reduction rate and cost saving rate. The model verifies the feasibility of the new coupling system and solves the problems of wind and light abandoning to a certain extent. New system reduces waste of renewable energy and improves energy efficiency. And it provides a solution for the future urban energy system.

This paper analyzes and researches the intelligent information construction of the governor control system of hydropower station. Based on IEC-61850 communication protocol, this paper puts forward the composition framework and function strategy of the governor control system, which is suitable for digitalization, communication networking, integration standardization and decision-making intelligence of intelligent hydropower stations. It provides effective solutions for the safety and stability, intelligence, fault early warning diagnosis, and status trend analysis of the hydropower station's governor control system, which is beneficial to promote the construction of intelligent hydropower stations for reference.

In the “Whole County Photovoltaic” background, the large-scale access of distributed photovoltaic(DPV) to the distribution network poses a challenge to the hosting capacity of the distribution network. Scene generation and scene reduction method was used to describe the uncertainty of photovoltaic(PV) power, Monte Carlo simulation method was used to sample the uncertainty of PV configuration, and the hosting capacity analysis method of distribution network considering the uncertainty of PV power and configuration mode was proposed. The forward and back power flow method is adopted, and a standard example of IEEE33 node is used to illustrate adaptability and effectiveness of the proposed method.

At present, China is dominated by the slowdown in energy consumption growth and accelerated structural substitution, and formed a new pattern of energy development. The National Energy Administration recently issued a notice on promoting the source network load-storage integration and multi-energy complementary development, emphasized the importance of energy storage in the power network, and energy storage will not be a "furnishing". As the core support for the development of new energy, the energy storage technology covers many aspects such as the power supply side, the grid side and the user side. This article studies energy storage technology from the user side. First, it introduces the current types of energy storage technology and the advantages and disadvantages. Secondly, it analyzes the economic benefits of electrochemical energy storage technology on the industrial and commercial users under the peak-to-valley arbitrage mode. Finally, the problems of energy storage and future energy storage trends are summarized. Through the research on energy storage technology for industrial and commercial users, it can be known that among the three types of batteries, lead-acid batteries, lithium-ion batteries and all-vanadium flow batteries, lithium-ion batteries are currently the best choice for user-side energy storage.

The proposal of the "double carbon" goal puts forward higher requirements for the energy transformation of the power industry. With the access of multiple loads, more and more complex diversity and random uncertainty problems are brought to the power grid. Therefore, a strategy of reducing loss of distribution network under the interaction of source network load and storage is proposed. Firstly, the multi load output characteristics of clean energy, electric vehicles and energy storage systems are studied. Secondly, based on the multi load output characteristics, the strategies and model of reducing loss of distribution network under the interaction of source network load and storage is constructed. Finally, an example is analyzed. The results show that the proposed the strategies of reducing loss of distribution network under the interaction of source, network, load and storage can better coordinate power grid resources and improve energy efficiency.

Bionic waterweeds scouring protection facility is a new flexible protective measure and an important means to protect submarine transmission cables. In this paper, a real cable in Zhoushan, Zhejiang Province is taken as the research object, and a mathematical model is established to preliminarily evaluate the erosion protection effect of bionic waterweeds. The results show that larger resistance coefficient and wider protection range will achieve better siltation promotion effect.

In this paper, the power performance, economical performance, handling stability and driving safety are regarded as major control targets of Distributed Drive Electric Bus (DDEB). According to vehicle driving information, the weight coefficients of multiple control targets are calculated by neural network. Furthermore, multi-objective optimization algorithm is carried out to determine the driving torque of each wheel with Multi-Objective Evolutionary Algorithm based on Decomposition (MOEA/D). The slalom test results demonstrate that the torque allocation control strategy can satisfy the demand of longitudinal driving force and direct yaw moment as far as possible, and improve the handling stability of test bus effectively.

In order to improve the maneuverability and stability of a certain type of multi-axle special vehicle under low adhesion coefficient and high-speed conditions, an active steering controller is designed with this type of multi-axle special vehicle as the research object. Firstly, the dynamic model of this type of special vehicle is built based on Trucksim, and the accuracy of the model is verified by real vehicle experiments, which lays a foundation for the research on active steering control. Then, an active steering controller is designed based on the Ackerman steering principle with the zero centroid slip angle as the control objective. Finally, the model of the controller is built through simulink, and the cosimulation with the vehicle dynamics model of Trucksim is carried out to verify the designed active steering controller. The simulation results show that the designed active steering controller has an ideal control effect on the roll stability and handling stability of the vehicle under high speed conditions and low adhesion road coefficient.

Through the analysis of venue lighting characteristics and market research, the existing intelligent venue lighting energy control system still has some problems, such as complex layout lines, short service life of lamps, poor comfort and so on. In view of the above problems, this paper proposes a venue lighting energy control system, the overall design idea to the venue intelligent construction concept as the starting point, with the Arduino microcontroller as the core control device, the use of light sensors and infrared sensors to detect the light intensity and number of people in the museum, according to the light intensity and number of people in the museum, control the opening or closing of the lighting lights. This system can carry out automatic control of sub-regions and time-sharing according to the lighting characteristics of different application occasions, thereby saving electricity consumption, and can also monitor the status of electrical equipment in real time, providing the optimal power-saving status and lighting effect, significantly improving the efficiency of lighting electricity, and improving the energy-saving space of the venue lighting system.

A deep investigation was carried out on the application of distributed wind power generation technology in Jiangyin Port, a typical inland port in China. The detail demand- supply balance of wind-generated electricity (i.e. electricity generation, consumption and exportation grid) was analyzed based on the real application. Furthermore, the CO₂ mitigation effect was estimated. The results indicated that there were 40.133 million kWh in 2021 generated by wind generation power technology, of which 56.46% was consumed by Jiangyin Port and 43.28% was supplied to the grid . Application of wind power generation technology in Jiangyin Port in 2021 can acquired obvious carbon mitigation effect, reducing 2.8×10⁴ tons CO₂ emissions. Wind power generation technology is a effective and feasible measure for promoting the task of carbon peak and carbon neutralization.

Battery Energy Storage System (BESS) has been rapidly developed and widely used in power systems at home and abroad, but Bess has not deeply understood the impact of AC Short-Circuit Current in power system. And the existing short-circuit current calculation standards do not involve the short-circuit current analysis method considering the influence of Bess in case of AC system failure. Based on Bess electromagnetic transient model, this paper analyzes the influence mechanism of energy storage output current on AC short-circuit current under AC short-circuit fault. The research results provide a reference for reasonably evaluating the impact of energy storage on AC system short-circuit current and optimizing the setting of Bess control and protection parameters.

This paper designs a multi-point wireless temperature monitoring system based on NRF24L01. The system consists of a master node and three slave nodes. The slave node uses the STM32 CPU to control the temperature sensor DS18B20 to collect temperature, and wirelessly sends the collected temperature data to the master node through the wireless transceiver module NRF24L01, and displays the temperature in real time on the TFT LCD display of the node. The master node uses STM32 CPU to control the wireless transceiver module NRF24L01 to receive the temperature data sent from multiple slave nodes, and at the same time displays the temperature data of multiple slave nodes on the TFT LCD display of this node in real time, if it exceeds the range of the temperature setting value , the monitoring system alarms. The advantage of this design is the use of wireless data transmission, multi-point data collection, avoiding wiring difficulties during wired transmission, and saving resources, and through test verification, the wireless temperature data transmission distance between the master node and the slave node can reach 100 meters , the temperature measurement accuracy is 0.1 °C.

YOLO (You Only Look Once) is a high-precision real-time object detection algorithm widely adopted in the visual perception of self-driving cars. However, its performance will be dramatically blocked in raining and foggy conditions. In order to address the detection of vehicles and pedestrian targets of self-driving vehicles during their driving on the road, this paper puts forward an improved YOLO model based on Multiscale Retinex with Colour Restoration defogging algorithm. In the proposed approach, different YOLO models (including YOLOv2, YOLOv3, YOLOv4 models) are fabricated to obtain a eligible network for vehicle and pedestrian detection in foggy environments. The experimental results indicate that the aforementioned three YOLO models have better performance based on the hypothesized approach, validating that the approach is effective in raining and foggy conditions. In addition, the YOLOv4 network model trained by the MSRCR algorithm enhances the accuracy of detection. Therefore, it can be estimated that the proposed approach is eligible for the application of real-time object detection in self-driving cars.

As one of the most crucial electrical equipment in the power system, the reliable and stable operation of distribution transformer plays an important role. For the condition detection of the transformers, all the existing research methods are based on sensors to collect electrical data. In this paper, a new detection method of abnormal power loss identification, basing on BP Neural Network is proposed. Using this new method, the operating load, environmental parameters and the loss values of distribution transformer are collected to establish the BP neural network algorithm, which can simulate the complex relationship between load, environmental parameters and transformer loss. The trained neural network model can input characteristic parameters, output the predicted loss value and determine whether the measured loss is abnormal through hypothesis test, which is used to predict the operation fault of distribution transformer. Finally, a practical example is used to analyze the effectiveness of this method.

For the failure data of a certain type of explosion-proof indicator lights, after the cumulative distribution function of twoparameter Weibull distribution was linearized, and the least square method was used to perform linear regression analysis on the failure data, shape parameters could be obtained, which show the fault mode of the indicator lights was poor heat dissipation; Through fault tree analysis to determine the bottom events one by one, it was found the failure was caused by poor heat dissipation; Then the failure indicator lights were dissembled to analyze the cause behind the failure deeply. Lastly, specific improvement measures were put forward in the paper.

The low-speed characteristics of vehicle shock absorber are mainly determined by the structural parameters of the constant through-hole on the throttle valve. By establishing the simulation model of the internal flow field of the shock absorber and using the method of computational fluid dynamics, the damping characteristics of the shock absorber at low-speed under different through-hole structures were obtained. The results show that when the volume of the constant through-hole of the throttle valve is equal, the damping force generated by the constant through-hole becomes larger and the throttling effect becomes stronger by increasing the number of steps or channels.

Chlorinated end group is a common design strategy to achieve high-performance non-fullerene acceptors (NFAs). Here, we designed and synthesized a monochlorinated small-molecule acceptor Y-BO-Cl-p with well-defined structure and studied its optical properties. The strong electronegativity of the chlorine (Cl) atom and the large polarity of the C-Cl bond will lead to an enhanced intramolecular charge transfer (ICT) effect from the electron-donating core to the chlorinated end group. Therefore, Y-BO-Cl-p has a narrow bandgap of 1.39 eV, and both its chloroform solution and solid thin film exhibit broad absorption spectra. In chloroform solution, the absorption range of Y-BO-Cl-p is mainly in 450-800 nm, and its maximum absorption peak is at 730 nm. In contrast, the absorption range of the film is mainly in 500-900 nm, and its maximum absorption peak is at 803nm. The large red shift of the absorption spectrum from solution to film is attributed to the tighter molecular packing in the Y-BO-Cl-p film. In addition, the larger Cl-substituted dipole moment also modulates the intermolecular interactions, which is beneficial to the formation of more ordered intermolecular π-π stacking. In conclusion, the chlorination modification of end group is an efficient approach to realize high-performance small molecule acceptors.

In the substation maintenance operation, the indoor equipment often faces the situation of replacing the conductive aluminum bar or copper bar in the installation cabinet.When the outdoor equipment is technically upgraded or overhauled, it will also encounter the situation of replacing and installing new wires and clips, and the precise positioning of the hole spacing in the process of making equipment such as conductive bars and wiring clips is particularly important. However, in the past, the determination of the hole distance mainly relied on the manual marking of the operator, and there was no special tool for fast and accurate positioning of the hole distance.This paper proposes an adjustable fast hole spacing positioning device suitable for substation maintenance operations, which can be positioned in both horizontal and vertical directions at the same time. It greatly shortens the operation time, provides favorable conditions for quantitative positioning work, and effectively guarantees the safe and stable operation of the power grid.

Current computer control technology is widely used in industrial automation production line, can reduce the workload, but also improve the production efficiency of the system, and the current control technology visualization, real-time monitoring is not good enough. In order to improve the efficiency of automated production lines and improve the above problems, the C sharp language was used to design the computer system of the automated production line based on the linear motor, first introduced the hardware composition and function of the automated production line, and then developed the human-machine interface and control system of the host computer, the user operated and monitored the production line through the interface. Finally, experiments were carried out on the production line prototype platform to achieve stable and high-speed operation, with the maximum speed up to 1200mm/s. Real-time human-machine interaction information could be realized, so as to improve the production line efficiency.

In view of the increasing demand for ultra-low speed and large torque direct drive motors in oil fields and military occasions, a new type of tandem direct drive compound motor was proposed, which can realize the double-side output of ultra-low speed and large torque. In this paper, the structural design of the tandem double-side output compound motor is carried out, and the finite element modeling and simulation analysis of the permanent magnet motor part and the permanent magnet gear are carried out by using Ansoft software, and the data of magnetic density distribution, no-load speed, torque curve, speed ratio, internal and external rotor speed and output torque of the permanent magnet motor are obtained. It can be seen from the simulation results of the rotation speed and torque fluctuation curves of the compound motor at no-load start and rated load start that the tandem double-side output compound motor has good stable operation and dynamic performance, and can realize the stable output of ultra-low speed and large torque.

In order to improve the survivability in complex electromagnetic environment, polarization modulation technology is widely used in modern radar. This greatly improves the low probability of interception performance of the radar and increases the interception difficulty of the radar counter reconnaissance receiver. In order to improve the interception probability of radar countermeasure reconnaissance receiver, increase the reconnaissance distance and improve the parameter measurement accuracy, a high-performance radar countermeasure reconnaissance receiver based on full polarization technology is designed in this paper. The performances of single polarization reconnaissance and full polarization reconnaissance are compared and analyzed theoretically. The simulation results of different polarization modulation signals show that the full polarization reconnaissance can effectively improve the reconnaissance performance.

In this paper, a speed measuring device for escalator handrail is designed, which can be quickly installed on the dado of escalator. When the wall panel is FRP, it can be installed directly by vacuum adsorption. When the protective wall panel cannot be installed by vacuum adsorption, it can be clamped and fixed by auxiliary support. When the mounting plate is fixed to the protective wall plate of the escalator, the speed measuring roller can be contacted or separated from the upper surface of the handrail by adjusting the vertical height of the sliding plate. The vacuum suction cup is flattened under the suction state, with good adsorption effect, safety and reliability. When it needs to be released, it only needs to loosen the sealing cover.

The welding quality of the GIS shell is critical to preventing SF6 gas leakage in the Gas Insulated Switchgear（GIS） equipment of the substation, and phased array ultrasonic testing（PAUT） technology can intuitively and efficiently inspect the butt welds of the GIS shell. Since the GIS shell is mostly non-magnetic aluminum alloy material, the scanning process can not use the magnetic wheel to allow the scanning device to adsorb on the detection surface.in order to solve the problem of coupling between the phased array probe and the surface of the GIS shell during the inspection process, automatic scanning device for PAUT of butt welds of GIS shell was developed. The scanning device uses the main frame, straps, probe lifting assembly and waterway tap assembly to effectively fix the phased array probe and the curvature wedge on the GIS shell, on the one hand, to ensure the close fit between the probe and the detection surface in the inspection, on the other hand, to ensure that the data can be automatically collected along the established trajectory during the inspection process.The scanning device ensures the quality of the collected data while improving the detection efficiency, and the feasibility of the scanning device is verified by experiments.

As a key measure of efficiency in the transportation sector, passenger flow collection in the intelligent scheduling system of public traffic vehicles is an important basis for planning, scheduling and forecasting in transportation services. In order to improve the accuracy of public transport passenger flow collection, an automatic passenger flow collection method using infrared sensors is presented. The method is highly accurate and low cost, and it uses neural network algorithms for pattern recognition, which results in an accuracy rate of 97% according to experimental statistics.

With the development of electronic technology, the study of signal distortion is particularly important. For example, if harmonic distortion is produced, it will make the multimedia audio lose its original timbre and even appear harsh sound. This design is a signal distortion detection device with MSP430F149 MCU as the core, which can complete signal processing, signal acquisition, and thus calculate the signal distortion. The overall design circuit consists of filter circuit, voltage lifting module, voltage following module, signal acquisition module, MCU and signal display module. MSP430F149 MCU controls the signal acquisition and storage, and the LCD1602 display shows the signal distortion measurement results according to the calculation formula written in the program code. In addition, the design utilizes the Fast Fourier Transform (FFT) algorithm, which not only ensures the measurement accuracy, but also greatly improves the speed of spectrum analysis.

The methods based on convolutional neural network (CNN) have greatly boosted the performance of salient object detection. However, detecting salient objects of multiple scales remains a challenge. To identify salient objects with different scales, we propose a novel multi-scale structure aware network, which can effectively integrate the semantic and detail information of multi-level features to enhance the complementarity between the features at different scales. Specifically, we first propose a global context enhancement module (GCEM) to enhance the relationship between highlevel features in CNN through the atrous spatial pyramid pooling module. Then, we utilize the channel and spatial attention mechanisms to suppress the background noises in above enhanced features. Next, we design feature integration module (FIM) to selectively fuse the low-level detail features, high-level semantic features, and global context information generated by GCEM. Finally, the proposed GCEM and FIM module are progressively cascaded three times to generate final saliency maps. Experimental results on three benchmark datasets demonstrate that our approach achieves superior performance over state-of-the-art methods.

A rodless cylinder motion actuator aimed at requirement of small space taken and high reliability motion of a certain project was designed. The design process of scheme, selection of core components, mechanical structure and control circuit were described in detail. A prototype actuator was built and motion tests were carried out. The test result shows that the actuator is stable and reliable, and the motion index meets the design requirements.

This paper analyzed the detent torque and holding torque of the Micro claw pole stepper motor by using FEM software based on the DOE(design of experiment) analysis method .The effects of air gap, magnet thickness, claw pole thickness and claw pole height on detent torque and holding torque were compared. Through optimization analysis, the optimal level of each factor of increasing the holding torque while decreasing the detent torque was obtained, which makes the peak-to-peak value of the detent torque decrease by 16.41% and the value of holding torque increase by 6.47%. Finally, the accuracy of the result was verified by comparing the test value and the simulation value of the motor. Through the analysis of the paper, it is found that air gap and permanent magnet thickness have the greatest influence on the detent torque of micro claw pole permanent magnet stepper motor. The influence of claw pole thickness on detent torque is second only to the magnet thickness, and the claw pole height has the least influence on detent torque.

How to conduct FOGs (fiber optic gyros) performance evaluation under small samples condition has been the focus of FOGs research. In this paper, an improved performance evaluation method of FOGs based on modified Bayes is proposed, which fuses the random weighting method to determine the priori distribution, and combines the likelihood function established by small sample test data to evaluate the performance of FOGs. The simulation results show that the proposed method substantially reduces the length of the confidence interval without reducing the confidence, improves the accuracy of the estimation results, and has important theoretical significance and reference value for solving the FOG performance evaluation problem under small samples.

No researches on measurement of micro angle using giant magnetoimpedance sensors have been reported. In this work, a sensitive angle measuring instrument - giant magnetoimpedance (GMI) angle sensor is reported to measure the angle of a stepper motor shaft at a long distance. The operating principle of the developed micro angle sensor is based on giant magnetoimpedance effect. By analyzing the experimental results under different conditions, the performance of GMI micro angle measurement system is discussed. The experimental results show that the GMI angle sensor can distinguish the micro angle change of 0.056° at a distance of 7 cm between sensor and coil, and still output a high voltage response of 5 V at 4 cm. With high sensitivity and resolution, GMI angle sensor shows a great ability of detecting micro angle, displaying a tremendous application potential in micro angle measurement.

The hydropower plant has been operating for more than 20 years since it was put into operation. With the commissioning of the upstream and downstream cascade hydropower plants, the operating water head of the power plant has been greatly changed compared with the initial design, and problems such as poor operating conditions and frequent accidents occur. Therefore, the technical transformation of the hydropower unit is imminent. The purpose of this paper is to solve the problem of changing the operating water head and improve the output and stability of the unit. Through research with the design institute and other units, it was decided to replace only the more efficient runner on the basis of not changing the diameter and flow path of the runner. After the technical transformation, the operation effect is good, and the stability and economic benefits have been greatly improved. This transformation can provide a reference for the transformation of the same type of units.

In this paper, the physical dimensions of UHV oil-air bushing and UHV Oil-SF6 bushing are analyzed in detail. UHV Oil-SF6 bushing adopts epoxy impregnated paper solid insulation structure, UHV oil-air bushing adopts outer porcelain sleeve structure and built-in oil impregnated paper composite insulation structure. UHV bushing prototypes are developed. At the same time, the UHV bushing and its aging test platform are comprehensively considered, and the three-dimensional model is further established. The model also considers the structural size of the bushing body and the pressure equalizing device at the tail of the bushing. The electric field on the surface of the bushing voltage equalizing cover is simulated and calculated, and the maximum electric field intensity on surface is 11kV/mm. Taking the electric field distribution at the tail of UHV bushing as the optimization object, the structural size is optimized in four schemes, and finally the optimal structural design method is obtained. Finally, according to the planning and requirements of 1000kV AC UHV project, the design parameters of UHV aging test platform are: the maximum nominal rated current of the equipment is 6.3kA, and the current can reach 7.56kA when considering the expansion factor of 1.2. The design parameters of the equipment are based on 8kA, and the maximum working current can reach 8kA. On the other hand, at present, the rated current of UHV AC bushing is mainly 2500A and 3150A. Selecting 8kA current can meet current required for accelerated aging test of bushing insulation.

Oil gas bushing is the sealed insulating part connecting two media (OIL and SF6). 1000kV oil and gas bushing is developed for the first time, and there is no relevant operation experience. It is very necessary to conduct long-term performance assessment and research under all working conditions (full voltage and full current) before it can be put into practical engineering application. The aging platform can meet the following functions at the same time: 1) it can effectively and fixedly install 1100kV UHV capacitive oil-gas bushing and oil-air bushing. 2) After the UHV bushing is installed to the aging platform for full voltage and full current operation, there will be no electric field distortion and partial discharge in the bushing capacitor core and the aging platform, so as to ensure its reasonable electric field structure design. Although the 1000kV ordinary oil air bushing independently developed has passed type test items specified in the standard, because its test can not truly reflect the reliability of its actual operation, it is also very necessary to carry out the long-term performance assessment research under all working conditions. Only in this way can the performance of 1000kV AC bushing (Oil-SF6, oil air) be comprehensively assessed, provide the strong support for the engineering application of 1000kV AC bushing, and provide basis for the real localization of 1000kV AC bushing.

The submersible mixer is used as the research object to study the influence of its installation position on the mixing effect. Based on CFD calculation software, the standard k-ε equation is used to simulate the internal flow field of the submersible mixer under different installation position conditions. The results show that the installation position is better in the diagonal than the installation position is in the same horizontal line, and the increase of viscosity of the working medium will increase the power consumed by the mixing device.

In order to explore the applicability of biodiesel on marine diesel engines, this paper is based on the bench test of a sixcylinder medium-speed diesel engine, under the load conditions of 25%, 50%, 75%, 90% ,100%, respectively. Fuel 0# diesel and three kinds of biodiesel B5, B10, B15, analyze the vibration signals measured at the cylinder head, body, fuselage, The study found that the higher the content of biodiesel, the smaller the vibration response intensity of the cylinder head, and the piston side thrust generated by burning biodiesel is also significantly smaller than that generated by burning 0# diesel. After comparing the vibration intensity, it was found that the higher the biodiesel content, the smaller the vibration intensity of the cylinder head and the fuselage, and the greater the vibration intensity of the machine foot position. The vibration intensity of diesel engine is small under low load conditions of 25%, 50% and 75%, and large under high load conditions of 90% and 100%.

Traffic is undoubtedly an indispensable part of our daily life, and the presence of traffic lights makes the movement of vehicles orderly. However, at the same intersection with traffic signals, the traffic congestion level can vary greatly on different days or at different times of the day. In this paper, a base model and an optimization model are developed to simulate the number of vehicles arriving at an intersection over a period of time, mainly using Poisson distribution probabilities. The base model solves the problem for traffic light cycles that are multiples of 10, while the optimization model solves the problem for traffic light cycles that are arbitrary at any point in time, based on the conditions given in the topic. This study aims to reasonably reduce the waiting time of vehicles and pedestrians from a planning perspective, which is of great practical importance.

At present, the traditional vehicle lead-acid battery charger has a very simple structure and a relatively simple charging method. It cannot effectively adjust the charging method according to the actual state of charge of the battery, and cannot effectively monitor, protect and repair the charging process. This paper conducts research on the above content, analyzes the structure and charging method of the existing vehicle lead-acid battery intelligent charging and maintenance system, summarizes relevant experience, and aims to provide scientific development suggestions for workers in the same field.

In the future advanced combat aircraft, the requirements for engine thrust-to-weight ratio is getting higher and higher, with current thrust-to-weight ratio approaching 10, the turbine inlet temperature will reaches 1700~1900K, which will far exceed the melting point temperature of high-temperature alloy material and eventually lead to a more severe working environment for high-pressure turbine vane. In this paper, three simplified models of turbine vane with similar structure are established, and finite element simulation of turbine vane was carried out by setting boundary conditions, the vane, temperature distribution and stress distribution and other performance data are obtained. the analysis results can be used as reference for the structure optimization of turbine vane.

In order to effectively solve the problem of rail-car’s large ramp traction in coal mine, the paper studies the operating environment conditions of underground rail-car, and obtains the design technical requirements of rubber wheel tractor. And through the vehicle transmission system matching and offset compensation traction device research, the overall layout scheme of the tractor was designed. The tractor is mainly composed of vehicle frame, bidirectional driving cab, explosion-proof diesel engine, offset compensation traction device, hydraulic transmission system, four-wheel steering axle, loading and unloading crane, hydraulic and electrical control system. Finally, the first tractor was manufactured and industrial tested in the coal mine. The results show that this scheme can effectively solve the problem of rail-car’s traction.

The natural circulation capability of the primary circuit directly affects the safe operation of the reactor under accidents such as loss-of-flow accident. In this paper, we model a megawatt-scale unit using the system analysis program RELAP5 to consider the effect of the steam generator U-tube arrangement on the natural circulation characteristics of the primary circuit. It is shown that: the flow rate of the primary circuit coolant at steady state varies according to different U-tube arrangements. During the long-term process of natural circulation, all three U-tube arrangements can establish an effective natural circulation. The horizontal-U type has the best cooling effect on the core and the highest natural circulation flow rate for the same design basis conditions. The inverted-U type steam generator showed a phenomenon in which the pressure in the outlet chamber was greater than that in the inlet chamber.