Short-term Photovoltaic Power Prediction based on ICEEMDAN and Optimized Deep Hybrid Kernel Extreme Learning Machine

Aiming at the problem of low prediction accuracy for photovoltaic power generation due to the strong randomness and volatility, a prediction model based on Improved Complete Ensemble Empirical Mode Decomposition With Adaptive Noise (ICEEMDAN) and Beluga Whale Optimization Deep Hybrid Kernel Extreme Learning Machine (BWO-DHKELM) is proposed. Firstly, the historical data are analyzed by Pearson Correlation Coefficient, and the meteorological data with high correlation are obtained as the input features of the prediction model. Secondly, the PV power is decomposed by ICEEMDAN to reduce its volatility. Then, DHKELM is constructed for each subsequence and several parameters of the model are optimized by BWO. Finally, the predicted values of each subsequence are summed to obtain the final prediction results. The effectiveness and superiority of the proposed model are verified by using real data from a PV plant in Jiangsu, China as an example.