Short-term PV Power Prediction Based on Feature Fusion and Multi-strategy HPO

Aiming at the problem of having too many feature factors in PV power prediction, the complex mapping relationship between key features and power is difficult to mine effectively, and the prediction accuracy is not high. A combined prediction model based on the fusion of features and multi-strategy optimization of the Hunter-Prey Optimization Algorithm (HPO) optimized GRU neural network is proposed. Firstly, the Pearson correlation coefficient method is used for feature screening, and then the PV power data are decomposed by Successive Variational Mode Decomposition (SVMD). Then, the HPO algorithm is improved by introducing the nonlinear strategy, the dynamic decay strategy, and the elite retention strategy, resulting in the IHPO algorithm with better performance. The hyperparameters of the GRU are then optimized using the IHPO algorithm. Finally, an SVMD-IHPO-GRU short-term PV power prediction model is developed. Through simulation experiments on data from a PV power station in Xinjiang and comparisons with many different combinations of models and decomposition methods, the prediction model is proven to have better robustness and accuracy.