Abstract
Solar energy is one of the most influenceable renewable resources. Photovoltaic(PV) system is widely used in converting solar energy into electricity. Therefore, monitoring the working state of PV systems in real-time to make sure the PV systems working in reasonable condition is a crucial task. The key work in this task is forecasting PV power generation in real-time. Since the PV power generation is greatly depend on the weather conditions which include many variables. This problem becomes a big data issue. Techniques which are relevant with big data involving data mining, machine leaning and deep learning are adopted in solving this problem. A solution is given by utilizing these techniques. Various methods of data mining are adopted in analyzing the big data, and machine learning and deep learning algorithm are utilized in implementing five forecasting models for PV power generation.The five forecasting model are : near regression model, Lasso regression model and Ridge regression model, SVR model, and MLP model. In order to have good view of the forecasting results, visualization corresponding with the five models are given. The five forecasting model are tested and evaluated by means of many different measurement including explained variance score (EVS) mean square error(MSE), and $R^2$ score(R2) and processing time. The evaluation result of the five forecasting model is reasonable which can be taken into utilization in PV system real-time monitoring.