A novel solar irradiance forecast model using complex network analysis and classification modeling

The output power of solar photovoltaic (PV) plant is mainly determined by the received solar irradiance of the solar PV panels. Therefore, for step wise PV power forecast methods, the accuracy of solar irradiance forecast is very important. However, the diversity of irradiance fluctuation patterns brings great challenges to PV power predictions and limits the application of existing forecast models. In this paper, a novel solar irradiance forecast model of ultra-short-term time scale is proposed, which using complex network analysis to identify the irradiance and classification modeling to realize classified forecast of irradiance. Firstly, 4 irradiance fluctuation patterns are defined according to the local historical irradiance data, then 4 BP neural networks (BPNN) are built and trained respectively using these data of different fluctuation patterns. Secondly, each four-hour irradiance series is transformed into complex network based on weighted horizontal visibility algorithm (WHVG), then three complex network features are extracted to analysis not only the time series characteristics but also the network topology of irradiance data sequence. Thirdly, the support vector machine (SVM) is used to identify the irradiance fluctuation pattern with complex network features, and the corresponding BPNN is applied to forecast the future irradiance. The accuracy of irradiance pattern identification and the subsequent classified ultra-short-term irradiance forecast are verified by simulation with two-years actual irradiance data in Mississippi, US.