Accurate forecast of solar irradiance is significant for related domains. Because accurate forecasts can help relevant researchers plan the management and application of solar energy that can be used in nuclear power stations and power plant. In this paper, an approach of solar irradiance forecasting based on artificial neural network (ANN) is adopted. The dataset from April 1st, 2017 to May 31st, 2018 was measured by the meteorological station in Yunnan Normal University. Multilayer perceptron model (MLP) and the variables, such as daily solar irradiance, air humidity, and relevant time parameter are employed to forecast solar irradiance in future 24 h. Moreover, the method of cross-validation is used to guarantee the robustness of experimental results. The results show the normalized root means square error (nRMSE) between the measured data and forecasted data is about 1.8~20.07% (1.8~10.6% for the sunny day, 11.6~20.07% for the cloudy day). Compared with ANN model, the nRMSE on the model of K-Nearest Neighbor (KNN), Linear Regression (LR), Ridge Regression (RR), Lasso Regression, Auto-Regressive and Moving Average (ARMA) and Decision Tree Regression (DTR), are 35%, 31%, 30%, 26%, 23% and 11% (unstable) respectively. It means that the performance of our model satisfies related applications.