PV Prediction Using Hierarchical BiLSTM-RFR Model Considering Meteorological Factors

Prediction to the photovoltaic (PV) power is critical to maintain power balance and secure operation of power system. To improve the prediction accuracy, this paper newly proposes a hierarchical prediction model with the complementary empirical mode decomposition (CEEMD) by combining bidirectional long short-term memory neural network (BiLSTM) and the random forest regression (RFR). At first, the PV power is decomposed to several intrinsic modes using the CEEMD. Then by analyzing the correlation between intrinsic modes and meteorological factors with the Spearman correlation coefficient, the prediction model based on the BiLSTM and RFR is proposed for the strongly and weakly correlated components. Finally, the resultant components are combined to obtain the predicted PV power. The simulation results validate the prediction accuracy with the proposed model.