Multi-Criteria Optimization for Photovoltaic Power Forecasting based on CNN-BiGRU

Accurate and stable prediction of PV generation has the potential to significantly enhance the stability of power systems and mitigate the reliance on conventional environmental resources. Nevertheless, many prediction models lack data processing and do not consider the model's generalisation capacity, resulting in poor robustness of the model. To solve this gap, a hybrid deep learning framework based on multi-criteria optimization and successive variational modal decomposition is presented. Firstly, the Pearson correlation analysis thermodynamic diagram is utilised to determine the input characteristics of the model, then the PV data undergoes a process of consecutive variational modal decomposition to mitigate the presence of high-frequency noise. Furthermore, a Bidirectional Gated Recurrent Unit (BiGRU) network is employed to further investigate the temporal dependencies within each subsequence. Meantime, multi-criteria particle swarm optimization is utilized to optimize the important parameter of BiGRU model. The experimental findings demonstrate that the prediction model has superior prediction accuracy compared to alternative models.