A Short-term Photovoltaic Power Prediction Method Considering the Reuse of Historical Information

To reduce the impact of photovoltaic output volatility on the power system, it is necessary to continuously improve the accuracy of photovoltaic power prediction (PVPP). This paper proposes a short-term power prediction method considering the reuse of photovoltaic historical information. Firstly, the historical similarity and spatial similarity of photovoltaic power are measured by using the Pearson correlation coefficient, providing more important decision features for PVPP. Then, the mutual information algorithm is used to evaluate the importance of all input features for the prediction model′s decision-making, and the top 75% feature combinations are selected as the final input feature set. Finally, a short-term photovoltaic power prediction method based on bidirectional recurrent residual network is constructed. Simulation results show that the root mean square error of the proposed short-term photovoltaic power prediction model at the 4th hour is 0.0572, which has a significant improvement in prediction accuracy compared with other models, demonstrating certain effectiveness and applicability.