This PDF file contains the front matter associated with SPIE Proceedings Volume 12260, including the Title Page, Copyright information, Table of Contents, and Committee Page.

Knowledge representation learning (KRL) aims to obtain the embedding of entities and relations from the information of knowledge graph (KG). Most existing methods can only model the entities in the training data, while failing to generalize to out-of-knowledge-base (OOKB) entities which only appear in the testing. To solve this issue, one common approach is to train an aggregator by leveraging the auxiliary knowledge such as neighbor information and entity descriptions. In this work, we propose a novel aggregation model called neighborhood transformer (Neighbor-T) to enhance the representations of OOKB entities. Compared with previous methods, Neighbor-T shows effectiveness on neighbor information aggregation because of self-attention mechanism. Experiments demonstrate that our enhanced representation outperforms the state-of-the-art on two knowledge graph completion tasks under OOKB setting: triple classification and entity prediction.

Crop pest control is one of the important tasks for crop yield. However, multi-class pests and high similarity in appearance bring challenges to precision recognition of pests. In recent years, deep-learning based algorithms in object detection have achieved an excellent result, such as the YOLO detector, which can balance accuracy and speed. YOLO performs well in detecting normal size objects, but has low precision in detecting small objects. The accuracy decreases notably when dealing with pest dataset, which have large-scale changes and multi-class. To solve the detection problem of multi-scale pest, we propose a detector named YOLO-pest based on YOLOv4 to improve the performance of pest detection. Our approach includes using lite but efficient backbone mobileNetv3 and lite fusion feature pyramid network. The improved detector significantly increased accuracy while remaining fast detection speed. Experiments on the constructed Croppest12 dataset show that our improved algorithm outperforms other compared methods.

Agriculture pest disaster is one of the most important reasons that reduce grain yield. Accurate recognition and detection are the core of integrated pest management (IPM). Existing deep learning-based methods improve the capacity of feature extraction, but ignore the imbalance of object number and size distribution result in insufficient performance. Therefore, we design a joint balance-distribution oriented composition (JBDOC) to detect multi-class pests with the synthetic dataset. Object bounding boxes and white background boards are used to construct the balanced synthetic dataset for training the convolutional neural network (CNN). Our JBDOC solves the distribution imbalance without methods restriction and improves the test performance without extra time consumption. We combine the JBDOC with current popular detection methods to verify the validity. Experimental results show that the JBDOC greatly improves the performance of deep learning-based detectors in the pest field.

The segmentation of the offshore farm area in the high-resolution SAR image is of great significance for the statistics of the farming area and the analysis of the rationality of the farming layout. However, the SAR images have the characteristics of a lot of noise and inconspicuous features. It is difficult to achieve precise segmentation by directly using non-learning image segmentation methods. Therefore, we propose a precise segmentation scheme for offshore farms in high-resolution SAR images based on improved UNet++. We first adopt a simulated annealing strategy for the update of the learning rate during the network training. By initializing the learning rate multiple times, we avoid the network from falling into a local optimum. Secondly, for the dataset studied, we verify that the segmentation performance of resizing the image to 256×256 pixels is better than that of 512×512 pixels. Finally, we propose an improved UNet++, which uses SE-Net as the feature extraction network to enhance the feature learning ability. Extensive experimental results show that, compared to some state-of-the-art methods, the proposed scheme achieves superior performance with a frequency weighted intersection over union (FWIoU) of 0.9853 on the high-resolution SAR offshore farm dataset.

Spectrum sensing in cognitive radio is a key technology to improve spectrum utilization. However, the current spectrum sensing algorithms have limited accuracy and are not flexible enough in complex communication environment. In this paper, a spectrum sensing algorithm based on adaptive regional model is proposed. The spatial information of the regional model is used to improve the performance of spectrum sensing. Through the adaptive adjustment, the cognitive nodes at different distances in the regional model ensures the sensing performance of the whole region. Results show that the regional model can improve the spectrum efficiency and deal with the complex environment of multi primary user base stations. The spectrum utilization of single region model is improved by 9% and the average difference between actual model accuracy and prediction accuracy is 0.48%. Through this regional model, a spectrum sensing algorithm with adaptive adjustment and high frequency spectrum utilization is realized.

In recent years, with the rapid development of mobile terminal devices and the Internet, spatial crowdsourcing has received widespread attention. The spatial crowdsourcing problem is characterized by the location information contained in the attributes of workers and tasks, the crowdsourcing platform can assign reasonable spatial tasks to workers based on their current location, and the execution of tasks will be accompanied by dynamic changes in their physical location of the workers, so task assignment is an important research content of spatial crowdsourcing problem, and the quality of task assignment methods can affect the development of its crowdsourcing platform. For the spatial crowdsourcing problem that requires a group of workers with relevant professional skills to work together to complete special application scenarios (e.g., performance-type tasks), a cost-based greedy approach is proposed to minimize platform costs by matching a suitable team of workers for spatial tasks under the constraints of workers and tasks. Extensive experiments have been conducted on synthetic datasets to demonstrate the effectiveness and efficiency of the proposed approach.

With the development of Internet technology, various network attacks have emerged one after another, seriously affecting the security of many key infrastructures such as finance, energy, and transportation. Therefore, the importance of network asset management is self-evident. How to judge the security of assets and detect lost assets has become an important research topic. This paper proposes a method for detecting lost assets based on feature optimization and active-passive detection. Firstly, it achieves the classification of abnormal traffic by extracting important features of the traffic data. And then, it detects the network assets using the combined active and passive detection method. The experiments show that this method can effectively detect the lost assets in the network and effectively provide an analysis basis for threat analysis and emergency response.

This paper focuses on WiFi indoor positioning based on received signal strength, and weighed K-nearest neighbor (WKNN) algorithm is the most classic position estimation strategy. However, locating across the entire fingerprint database takes a lot of time. In this paper, a clustering algorithm based on Self-Organization Map (SOM) is proposed to shorten the positioning time. Meanwhile, an improved WKNN algorithm is proposed to further increase the positioning accuracy. The experiment results show that the positioning time is effectively cut down after clustering and the average positioning error of the proposed algorithm is 1.18 m, which can achieve high accuracy in indoor environment.

The traceability of data leakage remains a foundational challenge faced by big data. Traditional data tracing technology is mainly based on digital fingerprint to embed lengthy code into digital works such as video, while the structured data with limited embedding space has not been given adequate consideration. In this paper, we propose a chameleon short signature by improving Khalili’s chameleon hash function and combining Boneh’s signature algorithm to achieve a one-to-many signature with a shorter message length under the same security premise. Then, we construct a traitor tracing model based on the proposed signature and design a cascade chain to complete credible data sharing and undeniable leaker detection. Security and simulation analysis show that the traitor tracing model achieves trusted data sharing and efficient traitor tracing for structured data.

Credit scoring is a significant research domain for both finance and computer science researchers. Customers’ loan application behavior is highly related to their future default performance, which requires more studies. This study applies a modified approach of sequential-pattern-mining-based classification to mine important and discriminant behavioral patterns of customers. And then, we use the patterns to predict customers’ likelihood to default on a future loan. Our approach specifically addresses the problem of data imbalance. Evaluation based on real business data shows that our approach outperforms a series of time series classification methods, including deep learning models. In addition, such pattern-based classification approaches have the merit of explainability in features. Specific patterns of customer loan applications are found to be related to their future default behavior. Therefore, our method enhances the business understanding, provides managerial insights, and thus is more likely to be accepted by the industry.

This paper proposes a method to fill in the missing traffic data by using multi-source data. Due to the regularity and specificity of traffic data, Gru network is used to capture missing patterns. The processed missing data, mask data and time interval data are input into Gru network for more in-depth information capture. The results of road speed matching for the floating vehicle data on the road in the corresponding period are further studied by Gru network, and the two results are fused to obtain the filling value of missing value.

Simultaneous localization and mapping (SLAM) based on 2D lidar is the vital technology for indoor mobile robot mapping and navigation, and graph optimization has become a common method to solve this problem in the recent years. In graph-based SLAM, loop detection is a key step to obtain global pose constraints, and the real-time performance of this process ensures that the back-end optimization of the current frame can be completed smoothly before the arrival of the data for the next moment. However, due to the limitation of mobile robot's computing resources, when the global map reaches a certain scale, the success rate of loop detection which has a positive impact on the mapping accuracy will decrease with the number of loop constraints is directly proportional to the number of all poses. Therefore, we propose a self-adaptive matching method based on genetic algorithm (GA) to calculate the loop closure constraints between the current scan and each local map, so as to speed up the loop detection process. The experimental result shows that our method is superior to the traditional graph-based SLAM solutions in large scale map construction.

In our work, we implemented one-shot action recognition that using the skeleton data. In terms of data preprocessing, we used the form of mapping skeleton sequence coordinates into signal images. In the feature extraction module, we used feature extraction based on resnet18. In the few-shot learning part, we adopted the metric neural network model based on graph neural network. Finally, the leading accuracy is realized on the ntu-rgbd120 one-shot dataset.

Traffic forecasting is one of the most important problems in the areas of intelligent transportation system, and it is the key link. It plays a major role in transportation service and navigation. However, urban traffic has its own characteristics, and the complex traffic system is highly nonlinear and stochastic, which makes traffic forecasting a very difficult problem. Although many previous methods can make the high performance for predicting in traffic forecasting, the existing research has not fully utilized the influence of spatial and temporal characteristics on prediction. In this article, we put forward a new model called Spatio-temporal multi-attention graph network. Taking into account the similar features of traffic flow every day and the interaction between road network structures, the model takes advantages of the internal dependence between the dynamic spatial network and the time dimension information to improve accuracy of forecasting. Experimental results show that our model is nicer over the others, which has good performance and gain more precision prediction accuracy.

As more and more industries and enterprises begin to develop the online/offline business model, the demand for efficient sentiment analysis of online service user comments is increasing day by day. In view of the problem that the aspect emotion analysis model is difficult to combine the semantic information of the text effectively, and the accuracy of aspect emotion classification of the texts need to be improved, this paper uses GRU model to improve the ATAE-LSTM model and improves AlBERT model through the attention mechanism based on aspect level. The ATAE-AlBERT-BiGRU model is proposed. In this paper, comparative experiments are carried out on the SemEval 2014 dataset, and the results show that the accuracy of the proposed model is significantly improved compared with the comparison model in the aspect level emotion analysis task, and it performs well in reducing the computational load of the model.

COVID-19 plays role in every part of the world; especially, it does harm to lives of people. Thus, COVID-19 sounds the alarm that is very important to build an effective mechanism to help prevent pandemic disease. In this work, dynamic network based on status value is built, which aims to help simulate the added danger level by the addition of infected people or close contacts. First, each node of this network is labelled with different kinds of status which has special value to show its danger degree. Then, the weight of the network represents the relationship of nodes; with the value of each node, average length and average spread of danger level is calculated based on the accumulation of dynamic weight. Thus, epidemic speed and scope of the infectious disease can be simulated. Moreover, the experiments compared to other networks have verified the effectiveness of our model.

Path planning acts a significant role in the motion and exploration of mobile robots. As artificial intelligence develops, path planning is also moving towards intelligent direction. Deep Q network (DQN) has low computational complexity and high flexibility, and is widely used in mobile robot path planning. DQN algorithm mainly obtains training sample data through uniform random sampling, which is easy to generate redundancy and reduce the precision of the training model. So as to reduce the redundancy of selected samples, an improved DQN method for path planning is proposed in this paper. By establishing sample similarity screening matrix, the proposed algorithm can eliminate samples with high similarity, improve model training effect, and further enhance the precision of path planning. Simulation results show that the algorithm this paper proposed has a great improvement in the convergence speed of DQN model training and the robustness of path planning.

For time-delay sensitive applications in moving edge computing scenarios, this paper proposes an offloading strategy with caching mechanism based on genetic-particle swarm optimization algorithm. The strategy combines the above two algorithms to obtain the optimal offloading ratio and caching decision in edge computing offloading. Caching completed and multiple requested tasks and their associated data to the edge cloud minimizes task unload delays. The simulation experiment results demonstrate that this strategy can significantly reduce the delay of mobile edge computing.

With the increasing number of vehicles and the shortage of spectrum resources, the development of the Internet of vehicles is severely limited. Based on game theory, this paper proposes a dynamic allocation scheme of spectrum resources in the congestion scenario. Different from the traditional spectrum resource allocation scheme, the scheme gives users cognitive ability to realize dynamic and intelligent efficient utilization of spectrum resources without relying on the central node and the mutual transmission of channel information between users. In a dense scenario where the number of wireless resources is less than the number of users, the scheme can automatically screen out the users with more advantageous communication conditions to enhance the overall communication capacity of the system. Simulation results show that the algorithm can automatically filter out the channels with relatively better conditions, and can effectively avoid the interference between different users, which is highly efficient.

The prevalence of cardiovascular diseases in China is still on the rise, and it is estimated that 330 million people are suffering from cardiovascular diseases. In terms of physical health and mental health, a heart rate detection technology that is portable can be measured at any time, safe, comfortable, simple to operate, and low-cost is essential. Given the inevitable jitter of handheld mobile phones, based on the accuracy of remote photoplethysmography (rPPG) detection technology, we propose a moving window timing sampling to refine the original video frame signal. The heart rate value can be extracted by processing such as region of interest (ROI) selection, high-pass filtering, blind source separation algorithm, Fourier transforms, and peak detection. Compared with the heart rate value obtained without using the moving window timing sampling, we found that the effect of the moving window timing is about 10 seconds is the best. The root means as the square error (RMSE), mean absolute error (MAE), and standard deviation (STD) are the lowest, 6.6929, 5.1365, 6.6165 respectively. The errors compared to the sampling without moving piecewise function are 13.53, 10.79, 14.09, The errors were reduced by 50.5%, 52.4%, and 53.04% respectively.

Person re-identification is to query person across cameras and occlusion is one of the difficulties. Previous works have proved that local feature extraction and alignment are critical for occluded person re-identification. However, directly horizontal partition causes mis-alignment and extra-cue methods highly depend on the quality of extra-cues. In this work, we propose a novel architecture including a weakly supervised mask generator without introducing extra-cues to create fine-grained semantic masks for local feature extraction and alignment, and a weight-shared fully connection to control the balance of local and global features. We also propose a general form of weighted pooling to improve gradient transfer, which gets rid of the probability explanation with softmax. Moreover, we unravel that there is a conflict between local branches and global branch, and a buffer convolution layer helps to fix this conflict. Extensive experiments show the effectiveness of our proposed method on occluded and holistic ReID tasks. Specially, we achieve 62.5% Rank-1 and 52.6% mAP (mean average precision) scores on the occluded-duke dataset.

Collaborative filtering-based algorithms are widely used to make recommendations without analyzing the contents. Time effect can be seen everywhere in our daily life. User interests will change over time, so we use the time-decay function to integrate the user-item rating matrix and adjust it by different time-decay factors to optimize the model. And we conducted experiments using the improved algorithm on the movie evaluation dataset movielens-1 m. The results show that the algorithm is able to improve the accuracy and coverage of recommendations under specific time factors, and also can partly improve the recommendation efficiency.

In this paper, a monitoring system based on thermopile array sensors is designed for real-time refreshing of thermograms in a web interface. To address the problem of inconspicuous heat map of the main target caused by the interference of the detection environment and the small temperature difference between the target to be measured and the background, a dynamic color mapping processing scheme is proposed to make the heat map of the main target displayed more clearly by continuously adjusting the contrast between the main target and the background color. The experimental results show that the method can achieve dynamic refreshing of the thermogram through a multi-device browser, the correlation of measurement data is greater than 85% compared to handheld thermometers, and the effective transmission distance is about 30 m in open range, which can effectively enhance the portability and safety of staff during COVID-19 temperature screening.

This paper aims to study the power control method for fairness energy efficiency (EE) improvement in cognitive radio networks (CRN) with interference channels, among one primary user (PU) shares the spectrum to multiple secondary users (SUs). The objective is to control the transmit power to maximize the minimum EE among all users subject to the quality of service (QoS) constraints. An extremely challenging non-convex max-min fraction optimization issue given due consideration. This work aims at developing an adaptive solving method based on deep learning (DL) techniques for the max-min EE optimization problem. To achieve such an objective, we construct a deep neural network (DNN), with the channel state information (CSI) being the input of DNN and the transmit power being the output of DNN. However, this faces two challenges. On the one hand, it is difficult to obtain label data. On the other hand, when DNNs are applied, it is very important to consider that QoS constraints should be met. These difficulties are circumvented in our work by designing an unsupervised learning strategy, in which a loss function is devised by combining the max-min EE objective and the QoS constraints via the barrier function method. The effectiveness of our proposed algorithm is ultimately demonstrated by the simulation results.

A fault diagnosis method based on RSS-ERT was proposed to deal with the mechanical wearing fault of aero-engine through combining with oil analysis. After pre-processing the element concentration data of aero-engine oil sample analysis, in order to train more models and have more randomness, Return Sampling Strategy (RSS) was adopted in the build of Extreme Random Tree (ERT) model. The coefficient of determination (R 2 ) was used to evaluate performance of the model. This model and benchmark model were used to forecast engine health parameters. The result showed that the fault diagnosis method based on RSS-ERT was accurate and superior.

This paper takes Shengwan-Shigu area, which is located at the junction of Southwest Henan Province and Hubei Province, as the research area. Eleven evaluation factors are selected, and the multiple regression analysis function of SPSS software is used to analyze the sensitivity of each index. The sensitivity analysis results of the evaluation index show that the geological environment quality in the study area is the most sensitive to the changes of soil and water loss and desertification. Then establish the geological environment quality evaluation index system. In this paper, the Tensorflow deep learning library based on Python language is used to construct the neural network model to evaluate the geological environment quality of the study area; by using Tensorboard visualization tool, the complex neural network training process is visualized, and the neural network model is debugged and optimized. Finally, the evaluation results are visualized by the map editing function of MAPGIS software. In this paper, the geological environment quality is divided into three levels: good area, middle area and poor area. The evaluation results show that the overall conditions of the study area are good. The results of geological environment zoning and geological hazard survey points superposition in the study area show that the evaluation process and results are reasonable and feasible. The research conclusion of this paper can provide scientific basis for regional geological environment management in areas with serious soil and water loss and rocky desertification, and has important practical significance.

Coal is a major part of the world’s energy. The low calorific value of coal is the key index to measure the calorific value of coal. Therefore, real-time detection of low calorific value of coal mine plays an indispensable guiding role in the formulation of resource mining plan. To solve the above problems, this paper proposes a DP_WTELM (D: direct orthogonal signal correction; P: principal component analysis; W: whale optimization algorithm) model based on machine learning to detect the low calorific value of coal. The model is improved based on the classical forward propagation network that is extreme learning machine (ELM). The prediction accuracy of the network is improved by using preprocessing, whale optimization algorithm, and increasing the network depth. The experiments show that in contrast to ELM models, the DP_WTELM model has better performance in predicting the low calorific value of coal and can meet the industrial requirements.

The message board text put forward by netizens on a certain issue is a suggestion or opinion, which is sparse and emotional. The traditional LDA model cannot solve the sparsity problem of the short message and ignores the emotional factor. In order to solve the above problems, a message board information extraction method based on LDA model and RNN model is proposed. First, eigenvalues are introduced to classify the text to solve the sparsity problem based on LDA model. Second, RNN model is used to realize the emotional features on the basis of message text vectorization. The experiment shows that the LDA model with eigenvalues has better topic extraction ability compared with the traditional model, and with the fusion of the RNN model, it can comprehensively display the potential information of the message text. As a result, the proposed method achieves information extraction maximize.

As we all know, price is a very important factor that affects product sales. To a certain extent, price cuts will increase sales, and price increases within the acceptable range of users can increase manufacturers’ profits. Therefore, for manufacturing companies, scientific product pricing has always been a tricky issue. The pricing strategy model of traditional manufacturing companies is generally based on traditional estimates and conduct price reduction promotions. At present, there are not many pricing strategy models with better scalability for manufacturing companies, especially there are not many quantitative models that can effectively evaluate the impact of competing products on this product. For manufacturing companies, the easiest way to affect sales is to modify product prices. Therefore, based on the relatively novel ConvLSTM neural network model, this article constructs a pricing strategy model for manufacturing companies. To build a pricing strategy model based on cross-domain and cross-brand data, the traditional LSTM model cannot capture the complex relationships between different dimensions of data. Therefore, this article introduces the improved ConvLSTM neural network model of the LSTM model into the field of pricing strategy, and first passes the relevant data through the convolutional layer before ConvLSTM to fully explore the hidden high-dimensional logical associations between the cross-manufacturer and cross-domain data. Therefore, this chapter uses the ConvLSTM model to predict sales based on cross-domain and cross-brand data. At the same time, statistical methods are used to check the confidence interval of the prediction results to enhance the reliability of the model. Finally, use the predictive model to traverse the reasonable pricing interval to obtain the simulated highest sales and optimal product pricing. This chapter finally verifies the superiority of the ConvLSTM-based pricing strategy model proposed in this chapter through design comparison experiments.

Aiming at the requirements of appearance content generation of industrial products, this paper proposes an intelligent appearance product generation network architecture based on multi-channel encoder and dual attention module. Through multiple encoders, our network can learn the semantic information at different levels of the image, and then intelligently generate the appearance product image through the learned semantic information. Our model takes the line diagram as the input and can also input the sample image of the image to be generated. The network can generate appearance products that are consistent with the contour of the line graph and the color of the sample graph. The experimental results of the algorithm are qualitatively and quantitatively evaluated to verify that the algorithm can effectively generate appearance product images with high quality.

The detection of smart contracts vulnerability is a valuable research problem because smart contracts hold a huge amount of cryptocurrency. In the past, popular detection tools were mainly based on some traditional techniques such as fuzzing and symbolic execution, which rely on fixed expert features or patterns and often miss many vulnerabilities. Recent machine learning approaches alleviate this issue but do not notice the semantic information in the source code. In this paper, we develop a system called SVChecker to classify the smart contract source code written in Solidity. To show the superiority of our system, we conduct experiments on more than 40,000 smart contracts collected from Ethereum. Empirically, our experimental results demonstrate that our system outperforms all popular detection tools.

With the rapid development of social economy, people pay more and more attention to physical health problems. Heart rate detection, blood oxygen detection and blood pressure detection are important basis for cardiac diagnosis, which can help people to carry out early medical diagnosis. Combined with Arduino, Android and Bluetooth technology, a simple and easy-to-use health monitoring system based on wireless personal area network is proposed in this paper. The system uses reflective photoelectric sensors to obtain heart rate signal, digital blood pressure modules to obtain blood pressure information and MAX30102 blood oxygen sensors to obtain blood oxygen information. The data acquired by the sensor is processed by Arduino, the heart rate and blood oxygen information are displayed on the LCD1602 in real time. Meanwhile, the smart phone APP is used to communicate with Arduino via Bluetooth to perform blood pressure measurement. The test results show that the system can obtain health parameters timely and accurately, and the overall operation of the system is stable and practical.

Autonomous operation for unmanned surface vehicle (USV) is the core competence for USV, and an important mean for unmanned and intelligent operation. USVs can be used for multi-mission, so autonomous operation for USV faces a lot of challenges, such as complex and changeable marine environmental factors, limited means of awareness in sea, low degree of automated operational decisions, difficult integration control of sailing and combat equipment. This study summarizes the development status of autonomous operation for USVs from awareness, decision and control. It addresses the difficult issues and current main technologies of autonomous operation for USVs, and points out the deficiencies of the technologies. Finally, based on the analyses of development status, it addresses the development trend of autonomous operation for USV.

In order to establish a unified user model in multiple networks, a method of user identity alignment in social networks has been proposed. Mainly focusing on the user identity alignment with homogeneous network with only one type of node and edge, the former studies has been separated into three types: (1) studies based on network topology only, (2) studies based on user behavior only, (3) studies based on both user-generated content and network topology. But the defect of the former studies is obvious that there is no real social platform with only one type of node and edge in the network. This type of network is called a heterogeneous network. This paper proposes a model that can perform user identity alignment on heterogeneous networks, named user alignment across heterogeneous networks based on meta-path attention (MGUIL). MGUIL fuses meta-path features by introducing a graph attention mechanism in two heterogeneous networks and obtains local and global information through a two-layer GAT network, finally aligning the information in both networks with a unified framework. This method not only solves the alignment problem on heterogeneous network but also considers the global information propagation as a unified framework. We compare it with the existing method in real networks and confirm that MGUIL can improve user identity alignment accuracy.

The load of user level integrated energy system changes rapidly and it is difficult to predict accurately. Therefore, a day ahead load forecasting method of integrated energy system based on multi-model combination was proposed. Firstly, the long short-term memory (LSTM) network model, convolutional neural network (CNN) model and harmony search (HS) optimized light gradient boosting machine (LightGBM) model were established. Then, the inverse root mean square error method (IRMSE) was used to combine the forecasting results of the three models to obtain the final forecasting value. The effectiveness of the proposed method was verified by the actual data of an integrated energy system. The results show that the proposed method is superior to the single prediction model and the simple average combination model, and has the best prediction accuracy for electric, cooling and heat loads.

In order to provide accurate ship-motion prediction for safe seaboard operations, this paper presents a new ship-motion prediction algorithm based on modified covariance (MCOV) method and neural networks. This algorithm firstly uses the MCOV method to analysis spectrums of the ship motion. And then, major spectrums of the ship motion are used to find the ship motion model by using the neural network (NN). Simulation results show that this method can present a confidence performs on ship-motion prediction.

Rolling bearings are one of the key components in rotating machinery equipment. It is of great significance to carry out state monitoring and residual life prediction for rolling bearings to strengthen service management of bearings and maximize the use value of bearings. In this paper, a bearing remaining life prediction model based on STFT-CNN was built. The STFT transform was performed on the original signal before the CNN model was input, and the one-dimensional time series signal was converted into the time-frequency domain. Finally, experimental verification was completed on the IEEE PHM 2012 dataset, and comparative experiments were conducted. Experimental results show that the residual life curve predicted by the STFT-CNN model is more accurate and fits the actual curve.

Photosynthetic rate (Pn) of plants is determined by environment, such as temperature, carbon dioxide (CO₂), and light. Light environment includes light intensity (LI) and light quality (LQ). It is important to build a predictive model of protected crops’ Pn where LI, LQ and other environmental factors are comprehensively considered. In this paper, cucumber was taken as experimental material, and a nested experiment was designed to measure the Pn under different temperature, CO₂ concentration ([CO₂]), LI and LQ. On the bases of these measured data, a predictive model of Pn was built by using support vector regression (SVR) algorithm. The performance of training set with coefficient of determination (DC) of 0.9990, and the root-mean-square error (RMSE) of 0.0478 µmol·m^-2·s^-1 demonstrated that the model is highly accurate after training. The validation results of predictive model showed that the fitting slope was 1.0015, and the intercept was 0.0223 between measured and predicted Pn values, which indicated that the model was accurate to calculate the Pn of plants under different environment.

The increase in surface trash salvage tasks has led to problems such as high staff workload, high labor costs, and low work efficiency. A garbage salvage system with autonomous cruising, identification, and detection is designed for this problem. This paper is based on deep learning technology to detect surface garbage and salvage garbage by robotic arm to solve the problem of low intelligence of surface garbage salvage as well as the problem caused by rising labor cost and insufficient labor, and to improve the efficiency of surface garbage salvage task. The system is equipped with a ROS robot operating system and uses lidar to acquire environmental information, realize map construction and autonomous cruise of the salvage vessel, use the YOLO v4 target detection model to detect garbage, and then apply the detected target and location information to the intelligent garbage salvage system. Five common types of garbage on the water surface are collected, and the average detection speed reaches 45 FPS and the average recognition accuracy is up to 88% through experimental verification, meeting the real-time system application.

The position service provided by global satellite navigation system (GNSS) has high-precision characteristics, it is widely used in the field of building deformation monitoring. As the baseline between the receivers is short for the above application scope, the common clock source method can be used to eliminate some error items. For common clock SDCP measurement, the uncalibrated phase delay is a major error item. The UPD feature and the usual estimation methods were analyzed and verified by experimental tests in the paper. The experimental results testify that the UPD is a random variable, which varies with the ambient temperature. The UPD can be eliminated by using ARMA filter, and the ARMA filtered SDCP position accuracy can be improved by more than 20% compared to the DDCP positioning result.

In order to solve the problem of flight delays caused by limited flight schedule resources, an inbound flight schedule resource optimization model was established. Considering the interval constraints on arrival points during the control and command process, as well as constraints on airport capacity, flight transfer connections, wake interval, and other constraints; genetic algorithms are used to find the time schedule with the least total flight delays and the least amount of delayed flights. Taking the flight time resource allocation of double runway airport as an example, the proposed algorithm is simulated and verified by AirTop software. The optimization results show that the flight schedule optimized by the proposed algorithm can run smoothly and can completely absorb nonserious flight delays.

Psoriasis is a chronic disease, which has affected over 125 million patients around the world. While the nail psoriasis is more common in psoriasis, it is time-consuming and subjective accurately assess the severity of psoriasis. With the development of deep learning and machine learning, more and more automated methods are proposed for the assessment of lesional psoriasis. However, there are few automated methods for accessing nail psoriasis. This paper proposes an automatic evaluation system for nail psoriasis based on deep learning. The system consists of a cascaded neural network, including nail detection model, nail lesion detection model and quadrant classification model, and combined with the scoring algorithm to obtain the nail psoriasis severity index (NAPSI) automatically. On the dataset we built, the mAP of the nail detection model is 0.909, and the accuracy of the quadrant classification model is 0.765. Through the detection of nail lesions with two models, it can be concluded that the mAP of the best model is 0.24. The models and algorithm have been applied and verified in the application of intelligent assessment.

When it comes to image classification at a device terminal a traditional machine learning method tends to pose a risk of privacy disclosure, while the federated learning is able to alleviate such privacy problem to a certain extent by saving device data locally to train the local model. In this paper images are classified based on horizontal federated learning and FedAvg optimization algorithm. Classification model train is carried out based upon CIFAR-10 dataset and ResNet-18. When the learning rate is appropriate, the optimization algorithm adopted has faster convergence, fewer communication rounds and better classification effect. The algorithm remains convergent and even shows faster convergence in the case of heterogeneous data.

In deep convolutional neural networks, the traditional Softmax loss function lacks the ability to distinguish similar classes. In order to solve this problem, the idea of increasing the inter-class spacing and reducing the intra-class spacing is widely recognized. The large margin angular loss (LMAL) loss function is introduced to reduce the intra-class spacing by L2-standardization of the features and weight vectors of the Softmax loss function. At the same time, LMAL also has a good ability to distinguish deeper features. Combined the LMAL loss function and the VGG-16 model, the results on three independent datasets show that the image recognition accuracy of the improved model has been significantly improved.

Differential evolution (DE) and fireworks algorithm (FWA) are two good optimization algorithms. Each of them have many advantages and have been widely used. Yet it is also inadequate. DE is easy to fall into local optimization and its parameters are difficult to set. FWA is not enough to exploit the local space. In this paper, a hybrid algorithm which combines DE and FWA (FWADE) is presented. By mixing and randomly redistributing the population of DE and FWA in the process of evolution, FWADE has the virtues of DE and FWA. The algorithm has fast solution speed to find global solution. The paper describes the algorithm process of FWADE in detail. The presented algorithm is used to identify the parameters of double diode model of PV cell and the result is compared with that of DE and FWA. The result shows that the hybrid algorithm can get a better solution in parameters identification of double diode model of PV.

As a new framework mode, the mobile edge computing (MEC) can overcome the problems of energy limitation and long delay in the operation of equipment by combining with simultaneous wireless information and power transfer (SWIPT). In this paper, we propose a SWIPT-MEC task hierarchical offloading model based on multi-objective optimization. The purpose is to offload different computing tasks to different MEC offloading levels through multi-objective optimization, so that the equipment can effectively reduce the consumption of energy and time in the face of computing task intensive scenes. Based on this model, a multi-objective MEC hierarchical optimization offloading strategy is proposed by using multi-objective optimization improved strength Pareto evolutionary algorithm 2 (SPEA2). The results of simulation experiments demonstrate that the multi-objective MEC task hierarchical offloading strategy can reduce the energy consumption rate and time consumption rate of the equipment during the offloading process.

With the development of China’s economy, the environmental quality is deteriorating, and the problem of air pollution has become particularly prominent. People’s high quality of life is closely related to air pollution. Air quality information is information that people will inevitably pay attention to every day. Therefore, research on air quality prediction methods is of very practical significance for revealing the changing laws of urban air quality, grasping air quality, and guiding people’s travel and lifestyle. This paper takes Beijing’s PM2.5 pollution as an example to study air quality prediction methods. Firstly, analyzing the correlation between air pollutant concentration and meteorological factors, establishing a GA-BP pollutant concentration prediction model with meteorological factors and historical pollutant concentration as input factors, and verifying GA-BP through a comparison experiment with the standard BP prediction model. Subsequently, based on the GA-BP pollutant concentration prediction model, a progressive prediction method was proposed, and the concentration prediction process of PM2.5 was used to predict the concentration of other five air pollutants. Based on the prediction of pollutant concentration, it refers to the calculation method of the air quality index to predict the AQI and AQI level. Comparing the predicted level with the actual level, verifying the feasibility and accuracy of the prediction method, establishing an air quality prediction system with GA-BP hybrid algorithm as the core.

Using artificial intelligence technology to realize intelligent detection and monitoring of transmission lines is one of the key technologies in the construction of new power system. This paper introduces the target detection of transmission channel based on deep learning method. The deep learning method has the advantages of little influence of super parameters on the results, strong feature extraction ability and strong anti-interference ability. This paper mainly introduces the network framework, including YOLO, SSD, R-FCN, fast R-CNN, etc., and analyses the advantages and disadvantages of various methods. At the same time, the transmission channel insulator detection is verified for various network architectures, and the experiments show the applicability of various types to transmission channel target detection.

In view of the characteristics of standard quality evaluation index diversification, multi-dimension and the result susceptible to subjective factors, a standard quality evaluation model based on T-S model of fuzzy neural network is established by introducing fuzzy system theory. This model combines the advantages of fuzzy system and neural network, such as good fuzzy knowledge expression ability and adaptive ability. It has the advantages of optimal result approximation, short training time and fast convergence speed. This model was used to evaluate the quality of 7 standards in the field of petroleum engineering. The results show that the fuzzy neural network method could solve the fuzzy data processing problem in standard quality evaluation, and the reliability of the evaluation model could also meet the requirements of standard quality evaluation.

Based on the traditional drug-target interactions, a drug-target hypernetwork evolution model was constructed using hypergraph theory. The evolutionary law of the growth of drug-target interactions was analyzed by mean-field theory, and it was found that the distribution of drug-target hypernetwork conformed to a power-law distribution, and further theoretical analysis obtained that the power exponent of the distribution was correlated with the growth rate of the target species corresponding to drug development. A larger exponent tends to explore new targets. By analyzing the drug target data collected from the drugbank in 2021, it was confirmed that the empirical results are consistent with the theoretical analysis and simulation results.

The three-dimensional visualization of coal seam structure in coal mines is the foundation for coal enterprises to implement transparent mining. Existing three-dimensional modeling systems generally have problems such as poor portability and single interpolation methods. Therefore, this paper adopts the development mode of front and rear separation, and designs and realizes a three-dimensional visualization system of no plug-in and strong portable field coal seam structure based on Springboot+Vue+Mysql. The system interpolates the seam elevation in unknown areas using the optimized Crekin interpolation method, uses the Delaunay rule to generate an irregular triangle network on the coal seam surface, and extends the triangle network along the coal seam elevation to generate a three-dimensional seam model based on generalized triprism prism; finally, with the help of Three.js three-dimensional engine, the browser side displays the three-dimensional effect of the coal seam structure of the minefield. The application of this system has certain guiding significance and reference value for coal enterprises to make reasonable preparations.

With the continuous improvement of automation technology, unmanned substations are developing rapidly throughout China. In order to more conveniently guarantee the basic operation and management of unattended substations, centralized control systems for substations are increasingly used in major real-time systems. In order to ensure the effective conduct of centralized control, State Grid Corporation of China proposed construction of a new generation of central control station equipment monitoring system. As a result, a huge amount of electricity data is generated in real time and need to be processed continuously. For instance, historical and current electricity data are analyzed differently, so that real-time analysis helps users to make quick decisions about centralized electricity regulation. This paper addresses the real-time processing and analysis of data in the centralized control system of a distributed substation from the perspective of real-time data processing. Specifically, our proposal is focused on accomplishing real-time distance processing of spatial-temporal data representations (graphs can represent many structured forms of data). To accelerate this computation, we propose a set of spatial indexing techniques, as well as the implementation of a complete KNN query system on this basis, and our approach has experimentally proven to be superior in terms of performance taken in constructing both.

CAPTCHAs are widely utilized on the Internet to partially protect against computer attacks, and text-based CAPTCHAs are commonly used. In order to make the more flexible attack, this paper provides a framework with configurable options based on k-NN, including three major parts: preprocessing the binary image, building standard library and recognizing image. The standard library is built from training dataset, where the third part can be an option to drop out some characters with a high similarity, and the library is used for testing dataset. A bit-based similarity model is proposed, where "and" and "or" bit operations are executed, and the result is the ratio of both operations. Finally, the framework is applied into four typical scenarios, MNIST handwriting database, CAPTCHAs built by the CAPTCHA generator, online CAPTCHAs of CNKI website, and CAPTCHAs within open source PHP DedeCMS, the average classification accuracy is 97.05%. As a result, the model is simple but effective, the framework can work well for text-based CAPTCHAs and handwritten numbers, which may make associated websites pay more attention to current authentication mechanism, and it offers flexibility to cover more algorithms and application scenarios by implementing different logics of preprocessing according to defined APIs.

This paper established a theory framework of the correlation between consumers’ web information search and related products’ sales. Taking the Chinese customers’ search behavior using Baidu search engine and the search data left during the decision-making process, this paper built up and filtered a search keyword thesaurus with high correlation of automobile sales and leading time difference. Then a brand automobile monthly sales prediction model was set up based on BP Neutral Network. On this basis, this paper took a specific automobile model for example and predicted its monthly sales for a month. The predicted results showed that the absolute average percentage error was 5.6%, which was 0.5% lower than the MAPE model by improved principal component analysis, and the prediction accuracy was improved. The validity and rationality of the model were verified. This paper provides a new idea for product sales forecasting of automobile enterprises, and also can be used as a reference for other industries.

In this article, a passive electronic lock system under trusted IoT is designed. The system is built on a combined public key frame structure and applies SM2/SM4 series of international data encryption algorithm (IDEA). It can meet higher security requirement of various application scenarios like in military, hospital and financial systems. In this study, exclusive identification is issued to staffs, equipment, packages and systems that involved in the electronic lock system. Lifecycle management service is provided. The operator-lock-system bidirectional identity authentication and secure data interaction are achieved. In this method, the proposed system is advanced in terminal reliability, access authentication, transmission safety, data tracking and boundary control.

Aiming at solving the problem that it is difficult for users to resist location semantic inference attacks when they consume location-based services (LBS) in a road network environment, a location privacy protection method based on the semantic similarity of road under user collaboration is proposed: For LBS consumption, in the roads provided by collaborative users, we use the semantic similarity between each road and the road where the requesting user is located to determine the most suitable road, build an anonymous road set, and pass the agent. The method is theoretically analysed, and experiments are carried out on the anonymous success rate and anonymous time based on the Brinkhoff road network data generator. Experimental results show that this method has better privacy protection and higher service quality.

With the continuous development of mobile positioning technology and smart phones, users can use smart phones to obtain and share location information of themselves and various surrounding points of interest (POI) anytime, anywhere, and share their own activity information, thus forming a location-based social network (LBSN). A large amount of user data is generated in LBSN. How to quantitative analyze the effects of context to manager’s view, how to extract the hidden user feature through the analysis of user data is of important research significance for the intelligent analysis of user characteristics. In this paper, first, a muti-dimensional user feature construction method is proposed, which extracts user feature from different influencing factors. Second, the fitness of user to a feature is analyzed. Third, a unified model is used to characterize this applicability. The method can promote the transformation from user data to user feature and help solve the problem of “explosive data but poor knowledge”. Experimental verification shows that the method is feasible and can realize the mining of muti-dimensional feature of users.

Logistics support capability (LSC) plays an important role in troop’s combat effectiveness, therefore the evaluation on LSC is the key supporting means for its improvement. This paper compared several models and their applications in the LSC evaluation area, including the fuzzy comprehensive evaluation model, grey evaluation model, network DEA (Data Envelopment Analysis) model, evaluation models based on belief rule-base (BRB) as well as evaluation models based on D-S evidence theory. Among these models, fuzzy comprehensive evaluation model can quantify some factors with unclear border or hard to quantify; grey evaluation model is suitable for the uncertain system; network DEA model effectively deals with multiple inputs and outputs on the same kind of decision-making units (DMUs); evaluation model based on BRB handle effectively the information uncertainty and evaluation index types diversity; evaluation model based on D-S evidence theory could effectively analyze the consistency of evaluation index and increase the discrimination of the evaluation. However, these evaluations mainly focus on the theoretical research and hardly can be applied in reality. In this paper, the suggestions about attention to the purpose of LSC evaluation, application of new information technology, and evaluation practicality were proposed in order to improve LCS evaluation.

Wangxin application is an instant messaging software launched by Alibaba Group that can support Windows, Android, and iOS operating systems, and its chat messages are stored in the local folders of smart terminals. In this paper, we study the encryption protocol of the Wangxin application and propose a reverse analysis scheme for the Wangxin application. Since the data encryption such as chat messages are stored in the database, we analyze the encryption mechanism of the local database in detail and give a decryption key extraction method based on dynamic binary instrumentation. Finally, through practical tests, our scheme can accurately extract the decryption key and restore the local database, providing a complete solution for the Wangxin application data forensics.

In accordance with the problems of body measurement technology and difficulty in obtaining body samples for common consumers in large-scale customization of clothing, simulation body samples and the machine learning method with positive images were proposed to be adopted to predict the body data. Firstly, simulated shooting of positive images from different angles was conducted on the basis of 116 simulation body samples. In addition, characteristic parameters were obtained through extracting the body silhouette and identifying key points and multiple linear regression and neural networks were used to conduct the machine learning of chest circumference, waistline and hipline, with the prediction model established. The experiment results showed that the error mean of two models in the sample set was less than 3cm and two models were equipped with the certain insensitivity to shooting angles, which may be promoted to clothing size customization of online large-scale customization of clothing.

It can improve the learning effect and reduce the learning burden to recommend exercises which are in accordance with the teaching objectives to students. How to accurately judge the needs of students and recommend exercises that are consistent with the teaching objectives is an urgent problem to be solved in the field of personalized education. This paper proposes a personalized exercise recommendation method for teaching objectives. This method can recommend exercises that are highly compatible with the syllabus for students according to their selected knowledge points and expected score range. According to the experimental test, the average prediction success rate of this method is 72.5%, 57.2% higher than the exercise recommendation method based on collaborative filtering and KNN, and 10.6% higher than the exercise recommendation method based on cognitive diagnosis and probability matrix decomposition.

This paper proposes a set of IoT connection management and data service platform. The IoT platform can avoid the disadvantages that hardware systems of multiple manufacturers and multiple systems need to be adapted separately. The unified access method of IoT sensing devices is studied to enable the access work of devices to be carried out in a configurable way and improve the efficiency of device access and management. The platform provides powerful data channels for users and help the dual-direction communications of terminals like multi communication between sensors, actuators, embedded devices and intelligent appliances. It also supports concurrent and massive access of devices, million-level messages concurrent processing performance and provides multiple protections of devices and realizes data extraction, storage, processing and integration from multiple IoT devices.

The theory of the six degrees of separation states that there are no more than six people between someone and any stranger, meaning that someone can meet any stranger through no more than six intermediaries. As one of the most popular social media applications in China, Weibo and its user data can be employed as the testing tool for the practical significance the theory holds. In this paper, an indirect network acquisition system containing two subsystems: Weibo user data acquisition system and user relationship analysis system has been established by Web crawler, and the system’s availability and effectiveness have been tested by system service test, which shows the possibility of the Indirect network acquisition system.

Aiming at solving the problem of power system simulation in the construction of ship maneuvering simulator, this paper modeled the power system of 7000DWT ship and completed the model simulation display in Unity3D. A set of user interface of ship handling system satisfying ergonomics is developed by modeling and simulating ship handling system. According to the final simulation results, the data of the real ship and that of the simulation are within the allowable error range, which can well observe the scene in which the navigation condition of the ship changes due to the adjustment of the bell on the Unity3D interface.

This paper presents an automatic classification algorithm of Mars surface lineament structure based on Resnet50 in DEM (digital elevation model) data. This work aims to reduce the time spent by planetary researchers on the collection of lineament structure samples, so as to concentrate on scientific research based on lineament structure data. This method avoids the problems that traditional DTA (digital terrain analysis) technology can only be used locally and the judgment threshold is difficult to set due to the large differences in Mars around the planet. The highest accuracy of crater is 98.15%, the highest accuracy of dorsum is 100%, the highest accuracy of Vallis is 94.44%, and the highest total accuracy is 87.96%.

In order to predict the psychological status of teenagers accurately and quickly and find their psychological problems in time, this paper proposes a BP neural network classification model based on LM algorithm. Compared with BP neural network of steepest descent algorithm, the LM algorithm has better convergence speed and accuracy in the inversion coefficient experiment. In the Chinese text sentiment analysis experiment, the precision, recall and F1 values of LM-BP were better than SD-BP. Based on LM-BP model, this paper presents a design scheme of teenagers mental health classification and recognition system. The scheme has good stability, availability and scalability, which lays a technical foundation for the implementation of the system.

Aiming at the problems of incomplete information and low data mining efficiency in the current student management system, a college student portrait system is established based on Hadoop big data processing technology. The system collects student data from various business platforms in colleges and universities, and uses HDFS for data storage; uses canopy and k-means based clustering algorithms for multi-dimensional analysis of student data; uses Echart tool to visualize the analysis results and generate student portraits. Experiments show that the student portrait system based on canopy and k-means can describe students' images in multiple dimensions and help schools understand students more comprehensively.

Government policy performance evaluation based on the satisfaction of the masses is a hot issue, involving the design of evaluation system, model construction, parameter simulation and so on. Based on Delphi method, research resolves the government’s policy, the types of the masses, builds satisfaction evaluation index system, Then, based on entropy value method, the satisfaction of a city is evaluated, and finally, the grey correlation measurement is used. Empirical results show that the weights of urban residents, rural residents, business workers, civil servants and migrants are respectively 0.1603, 0.2227, 0.1573, 0.3005, 0.1591, and their evaluations to the government are respectively 3.5781, 3.6414, 3.6969, 3.7734, 3.6276. The results show that the performance evaluation of local government is a comprehensive evaluation of people to the government performance, can give feedback to the appeals of the masses and get public opinion analysis in hot areas in time, is conducive to the improvement of government work. The evaluation system constructed has high discrimination and promotion ability.

Ever since the concept of artificial intelligence (AI) was first coined in 1955, the quest for sophistication and improvement of existing technologies paved the way for the continuous development of AI technologies. Nowadays, AI technologies are redefining and disrupting the way people work and live in many different domains. This paper mainly focuses on AI applications in two fields closely related to people’s life - children & elderly care and short video industries. It first introduces several prevailing AI technologies applied in children & elderly care and short video industries, and then uses two case studies from Cubo Ai and Tiktok to elaborate the applications in the corresponding fields.

Common sense knowledge is a part of knowledge reasoning. At present, common sense knowledge reasoning is also a hot and difficult point in the field of knowledge reasoning. In order to enable machines to use human language, and according to some content reasoning or inference rules, to get the results of common sense reasoning. This paper mainly explains how to obtain the common knowledge base, improve the reasoning ability of the machine to the social common knowledge, and test the common knowledge reasoning dataset by using the epoch-making pre-training model through the deep reinforcement learning method. The experiment shows that by adjusting the parameters of the data, the accuracy can be improved, which is also conducive to the realization of the downstream social common sense reasoning question and answer task.

In order to explore the impact of urban rail transit on the second-hand real estate market, based on the difference-in-difference model, this paper takes Wuhan Metro Line 7 as the research object, and makes an empirical analysis from two perspectives of different districts and different distance ranges. The results show that the operation of Wuhan Metro Line 7 has a statistical significance on the housing price in Jiangxia District, Hongshan District, Dongxihu District and Jiang’an District, and the impact coefficients are 1426.2, 786.7, 647.3 and 520.2, respectively. In terms of the different distances of Metro Line 7, the opening of the subway within 1200 meters has a significant positive influence on the second-hand housing prices along the periphery, and the degree of impact is 500m, 800m-1200m, and 500 m-800 m in descending order. Quantitative analysis of the impact of the subway on the surrounding housing prices can provide the government with a reasonable guide resident to purchase houses and provide relevant evidence for the scientific implementation of land acquisition and demolition.

An AI-based clinical thinking mining discovery system that can be used on PCs and mobile terminals (cell phones, pads, etc.). The system utilizes artificial intelligence technology to obtain massive medical case data and case discussions from the Internet for data extraction and organization, and intelligently classifies them by disease type and clinical presentation. Using natural language processing technology and intelligent mapping mechanism of medical terms, it mines the information of clinical features, tests, inspections, disposal measures and reasons (drugs, surgeries, etc.) of the extracted cases, abstracts and visualizes the diagnostic rules and clinical treatment channels of the cases. The medical case data processed on the Internet will be combined with the typical case data in the HIS system to form a clinical knowledge repository of diseases, guide junior doctors and students to conduct clinical thinking training and consolidate medical knowledge.

This research seeks to investigate the behaviour profile on social media platforms, using the case of Tik Tok and QQ. Social media platforms are used for many purposes, including the requirements and gratifications, as well as to satisfy various social media needs. As such, these factors were utilised as the independent variables of this study, while social media adoption was treated as a dependent variable. Primary data was collected by using a questionnaire survey to obtain quantitative information. A survey method including self-administered questions was used to collect the study’s data. This study used random sampling to identify individuals who would likely use media sites like as Tik Tok and QQ. Based on the findings of this study, uses and gratifications, social media needs, and social media technology have a positive effect on social media adoption. Owing to the increasing number of social media platforms as a result of technological changes, there is need for researchers to continuously investigate the key factors influencing the adoption of different social media platforms.

In recent years, cases of medical insurance fraud emerge in endlessly. We urgently need to develop an effective way to detect fraud. However, efficiently mining the heterogeneous medical text data is a complicated and tough assignment in fraud detection. Therefore, a medical insurance fraud detection model with knowledge graph and machine learning is proposed in this paper. Firstly, a knowledge graph with 53,164 nodes and 1,209,847 edges is built based on the medical insurance text data of 20,000 insured members. Secondly, representation learning and improved label propagation algorithm (LPA) are used for feature engineering based on the knowledge graph. On this basis, combined with the expense data, the medical insurance fraud detection model is constructed by using easy ensemble and XGBoost. The experimental results show that the model proposed in this paper greatly improves the effect of medical insurance fraud detection. In addition, it is proved that text data plays an important role in medical insurance fraud detection.

With the rapid development of the Internet and social networks, how to analyze the behavior trajectory, emotional changes and relationship trends of online public opinion events have become an important research topic today. This article combines human behavior dynamics and sentiment analysis methods to study online public opinion events. This article uses Python to capture hot public opinion topics and comments on social platforms, and uses human dynamics to analyze public opinion events from time interval distribution and activity. Then this paper uses the maximum likelihood estimation method to evaluate the power exponent of its distribution, and finally uses BosonNLP and sentiment intensity to analyze the sentiment of the comment objects in the public opinion event. Experimental results show that the time interval of group public opinion events obeys a power-law distribution, and its activity is positively correlated with the power exponent. The sentiment analysis method of public opinion events based on human behavior dynamics performs well, and the dominant sentiment of the comment object is distributed with a power exponential. The number of likes and the sentiment value of follow-up comments effectively improve the results of sentiment analysis, and social platforms play an important role in the communication of sentiments.

China’s economy has changed from the stage of high speed growth to the stage of high quality development and emphasized the new pattern of double circulation with large domestic circulation as the main body. As an important content to build a new development pattern, the development mode of the retail also changes from scale speed type to quality benefit type. Based on the DEA-Malmquist productivity index method, this paper analyzes the panel data of 72 retail listed enterprises in China from 2015 to 2019. The results show that the pure technology efficiency change index of all retail listed enterprises shows the trend of first decreasing and then increasing, while the scale technology efficiency change index and the technology progress index show the trend of first increasing and then decreasing. The interaction of the three makes the total factor productivity index a relatively stable trend, which indicates that the steady growth of retail sales scale is mainly due to the improvement of the performance quality of retail enterprises in the industry rather than the objective external environment such as inflation.

Since the reform and opening up, China’s manufacturing industry has experienced considerable development. This paper divides the development of China’s manufacturing industry from 1978 to 2020 into three periods: recovery period, adjustment period and intelligent development period. This paper collects 855 manufacturing policies issued by the Chinese government in these three periods, uses LDA topic model and synergy network to study the policy evolution trend in these periods, and studies policy synergy combined with the classification of policy instruments. The results show that the first period of China’s manufacturing industry mainly focuses on reform and opening up, and pays more attention to technology renewal, product manufacturing and process improvement. The second period pays more attention to product quality, technology, enterprise development, industry informatization and standardization. The third period focuses on the development of intelligence, informatization, technology and production process, enterprise management and so on. The results of policy synergy analysis show that the use of China's environmental policy is redundant. There are some deficiencies in demand policy. At the same time, policies should pay more attention to basic industries.

With the popularization and development of the Internet, microblogs have become a mainstream social network platform. The evaluation of user influence has become a research hotspot. Most of the existing researches calculate influence by improving PageRank. But these researches ignored the relationship between users’ interest theme similarity and information dissemination, and didn’t have enough analysis about the interaction behaviors among users. Aiming at these problems, we proposed a new microblog user influence algorithm—MUI-ISIDA (microblog user influence based on interest similarity and information dissemination ability) in this paper, which takes into account users’ interest theme similarity and information dissemination ability. We verified the effectiveness of the proposed algorithm on Sina microblog dataset. The experimental results show that compared with PageRank and MR-UIRank, the proposed algorithm has achieved higher accuracy in user influence ranking.

As a new form of content marketing in China, webcast shopping is in the stage of rapid development and exploration. Based on the SOR theoretical model, this study established a model of the impact of webcast information display on consumers’ purchase behaviours. The external stimulus information at the level of products, network anchors and consumer groups will help consumers to establish cognize at the level of product, relationship and emotional and further generate purchase intentions and behaviours. Through empirical analysis, the findings of this study are showed as follows. First, the information transmitted at the product level via webcast significantly and positively affects the establishment of consumer product, relationship and emotional cognition. Second, the information transmitted at the network anchor level also significantly and positively affects the establishment of consumers’ cognition of products, relationships, and emotions. Third, the information at consumer group level transmitted through webcast only significantly and positively affects the establishment of relationship cognition, and the cognition at the product and emotion level is not significantly influenced. Forth, the impact of product, relationship, and emotional cognition established by online live shopping on consumers’ purchase behaviours are significantly positive.

Promoting blended learning is the trend of teaching reform in Chinese universities. The construction of blended courses should be student-centered, in which the students’ choosing willingness is of great significance. Based on summarizing the types of blended courses in colleges and universities in China, this paper analyzes the willingness of college students on different types of blended courses and the influencing factors. According to the research, faculty have different perceptions about the selection of online resources and the weight of online and offline courses in the process of blended course construction, and now, four types of blended courses have been formed in China. Students’ willingness to choose different types of courses is affected by their perceptions of blended courses, curriculum design, and learning demands. In order to promote the role of blended courses in higher education, colleges and universities should scientifically design and reasonably adopt blended courses.

At present, there are some problems in marine data, such as low utilization rate and poor visualization effect. Based on B/S development mode, open source 3D engine Cesium and WebGL and other computer related technologies, this paper designs and develops a visualization platform integrating marine observation data and marine environmental field data, effectively manage and use the marine data of the South China Sea, provide data and information products and services, make full use of visualization methods to display the South China Sea marine data, and provide strong support for economic development and scientific research in the South China Sea, real-time monitoring of the South China Sea marine environment, disaster warning, etc.

With the development of global economy, the enterprise competition intensifies. As the main operator of an enterprise, senior manager’s operation behaviors and decisions can largely affect the future development of the enterprise, which becomes the core competitive factors of the enterprise. But during the actual production and operation process, with the outstanding agent contradiction between managers and shareholders, enterprises carry out incentive measures on senior managers one after another for alleviating this kind of contradiction and lowering the agent costs. The classic upper echelon theory (UET) helps people in understanding the internal relationship of the characteristics of senior managers and enterprises decision-making behavior and then how they affect enterprise performance. However, former researches on the relationship between senior manager team’s characteristics and enterprise performance are all based on the average level or heterogeneity of the whole manager team. There are little studies on the relationship of the incentive and performance based on the vertical dyad linkage (VDL) characteristic differences of the chairman and other senior managers. So this paper proposes a research on the effects of senior managers’ incentive and enterprise performance, and discusses the relationship of chairman-senior manager team’s VDL characteristics and the way to adjust the incentive and performance. By analyzing the sample data of public enterprises in Shanghai and Shenzhen during 2014-2018, this paper probes into the relationship of incentive and performance based on theory analysis and case studies, and the regulation on the relationship between the VDL characteristics of chairman-senior managers.

PDF (portable document format) documents are widely used in information publishing, academic exchanges and daily business. Phishing attacks with malicious PDF documents have become an important means of APT (advanced persistent threat) organizations. Researchers have found that more than 90% of malicious PDF documents launch attacks by JavaScript code. The current detection models’ generalization is not enough to detect unknown malicious samples. This paper proposes a method for detecting malicious PDF documents based on benign samples. The method uses benign PDF documents as training data, and uses features at the semantic level of JavaScript code. The JavaScript keywords and usage methods frequently used in malicious PDF documents are taken as important features to improve the robustness of the model. Then, we use One-class SVM (support vector machine) machine learning algorithm to detect malicious PDF documents containing JavaScript code. Compared with the detection model trained with malicious PDF documents, the method proposed in this paper improves the generalization performance while maintaining a higher detection rate.

This paper studies whether there is a correlation between the richness of health manpower resource and the utilization efficiency of health resources. The number of health workers per thousand people was used to measure the richness of health manpower resources, and the BCC model of data envelopment analysis was used to calculate the relative health resource utilization efficiency of all cities in Hubei in 2019.To explore the correlation between the richness of health manpower resources and the utilization efficiency of health resources clearly, the correlation test was carried out. The results showed that the utilization efficiency of health resources was negatively correlated with the richness of health human resources in 15 cities except Wuhan, the provincial capital city, and Shennongjia. Therefore, it is considered that the richness of health manpower resources has an impact on the utilization efficiency of health resources, which may form a “resource curse”, that is, the richness of resources may lead to the inefficient utilization of resources. It is suggested that the government should introduce policies to control the number of health workers in cities with low utilization efficiency and high resource abundance, encourage medical graduates to find jobs in cities which equipped with less health workers and high utilization, guarantee the health manpower in areas with less health workers, and promote the harmonious and balanced development of the whole province.

This paper uses the statistics for the school shop to order data collection and data simulations by Flexsim simulation software analysis, found that there exists a problem of the unreasonable operation site, the number of workbench through there on the spot, the analysis of the distance apart, use ECRS four principles of industrial engineering, by engineering methods adjust the job site, The labor intensity of the staff is reduced, and the meal time of the customers is reduced, and the utilization rate of the hearth is improved.

Multi-agent was introduced to study the issue of multilateral and multi-attribute fair negotiation in e-commerce. In this paper, a mathematical model is established through the Hungarian solution to solve the problem. Taking into account the multi-attribute and multilateral conditions and fairness in the negotiation, the real negotiation information is simulated in the form of randomly generated data, and the evaluation profit system and bilateral negotiation model are established to solve the problem and then the profit matrix can be obtained. Analogy 0-1 assignment problem is solved by Hungarian solution. Through numerical experiments and analysis, it indicates that this algorithm can obtain the overall maximum profit value that the system can achieve, and make the number of matching people as large as possible to achieve better negotiation results.

Tropical cyclones (TCs) have a serious impact on marine fisheries. More than 80% of marine disasters are caused by TCs. Based on the data of the best track of TCs in the Northwest Pacific Ocean, this paper analyzes the spatial distribution of TC influence duration. The analysis shows that: (1) in the Northwest Pacific, tropical depressions have the most extensive influence; the influence range of tropical storms is similar to that of severe tropical storms; typhoons gradually decrease around the southeast side of Taiwan island; From severe typhoon level, the influence of TCs is moving away from the mainland; super typhoons have obvious influence only in the east of Malaysia archipelago. (2) In terms of marine fisheries, severe tropical storm and typhoon level of wind resistant marine fishery facilities should try to avoid the waters on both sides of the line between Taiwan island and Malaysia islands; severe typhoon and super typhoon level of wind resistant marine fishery facilities are only at risk in the east of Taiwan island and Malaysia islands, and are basically safe in other waters.

At present, the total employment of college graduates is under great pressure and the structural contradictions are very prominent. The State Council of China clearly proposes to strengthen the employment service and career guidance for college graduates. It is of great significance to fully integrate the information of campus teaching management and student ability evaluation system and improve the level of campus data governance for realizing personalized employment and career planning services. This paper comprehensively uses the methods of educational and psychological measurement, multivariate statistics and web technology, and establishes a multi-dimensional student information database and all-round talent portrait through the construction of data analysis model, which lays the foundation of quantitative analysis for the realization of two-way accurate matching of talent supply and demand.