Photovoltaic Power Forecasting Model Based on Wavelet Packet Transform and Similar Day Theory

This paper proposes a hybrid GRA-WPD-GRU deep learning model for improving photovoltaic power output prediction accuracy. The model combines Grey Relational Analysis (GRA) for similar day selection, Wavelet Packet Decomposition (WPD) for sequence processing, and Gated Recurrent Unit (GRU) networks for prediction. Using data from a Chinese photovoltaic power station, our model demonstrates superior prediction accuracy and stability across various seasonal and weather conditions compared to traditional machine learning approaches, as evaluated by MAE, RMSE, and $\mathbf{R}^{\mathbf{2}}$ metrics. This improved accuracy addresses the challenges of integrating intermittent photovoltaic power into existing power grids.