Solar irradiance feature extraction and support vector machines based weather status pattern recognition model for short-term photovoltaic power forecasting

Photovoltaic power forecasting (PVPF) can help energy management system and power grid to improve the proportion of solar energy in total energy consumption. Classification modeling according to different weather types is an effective means to improve the accuracy of PVPF under various weather statuses. However, the weather type of historical data (WTHD) is missing in some cases, which will cause great difficulties to classification modeling because the data without WTHD cannot be used for the model training. To identify the missing WTHD, a solar irradiance feature extraction and support vector machines (SVM) based weather statuses pattern recognition (WSPR) model for short-term PVPF (ST-PVPF) is presented. To ensure the feasibility and reduce the workload of classification modeling, four generalized weather classes (GWC) covering all weather types are constituted, and GWC based classification modeling approach for ST-PVPF is proposed subsequently. The SVM model for WSPR is built with input features extracted from solar irradiance data. Through a case study, the effectiveness and performance of the WSPR model are verified and evaluated. The influences of different input dimensions and feature combinations are also analyzed and discussed. The results indicate that the missing WTHD can be effectively recovered as GWC by the proposed model.