Photovoltaic Power Prediction Based on Cluster Analysis and Fitting Residual Random Forest Ensemble Model

The photovoltaic power prediction is studied from three aspects in this paper, data correction, cluster analysis and prediction modelling. Firstly, the historical data of photovoltaic power station is used to explore the mapping relationship between the predicted irradiance value and the measured irradiance value. The deviation of the predicted irradiance value is corrected according to this mapping relationship. Secondly, the mutual information entropy (MIE) is introduced to revise the clustering result of k-means. Finally, a residual random forest ensemble model is proposed, and day-ahead photovoltaic power is forecasted based on solar power curve clustering. Experiments show the proposed method is effective in improving the accuracy of solar power prediction.