Short-Term Photovoltaic Power-Forecasting based on Machine Learning

Photovoltaic (PV) power short-term forecasting is of great significance to power system control and operations. To improve the forecasting accuracy, this paper proposes a machine-learning-based short-term forecasting approach for PV power. The proposed approach is built on multiple sub-models which contribute to high prediction accuracy. Firstly, a training set is used to build sub-models with better performance. Secondly, each sub-model is fused by an equal weight fusion method or a least square method, and the integrated model generates the prediction results. Finally, the effectiveness of the method is verified by an example. The simulation results verify that the proposed forecasting approach has a better prediction performance than the equal weight synthesis methods and an improved prediction accuracy than the traditional method.