Multi-Scale Fusion Fuzzy Method for Accurate and Interpretable Prediction of Photovoltaic Power Generation

Photovoltaic (PV) power systems can provide clean, carbon-free energy and have become an important pathway for achieving carbon neutrality goals. Accurate prediction of PV power output is crucial for the rational scheduling and allocation of solar power resources. To further enhance the prediction accuracy at some local regions, especially around the peak or bottom points, this paper proposes a novel fusion fuzzy model for PV power forecasting. This model begins with the development of a distributed function-weighted fuzzy method (DFWFM), which effectively simplifies the structure of traditional fuzzy systems and significantly reduces the number of fuzzy rules. And an iterative regularized least squares method is developed for tuning the parameters of the DFWFM. Furthermore, a data-driven multi-scale fusion model (DMFM) is proposed to alleviate local prediction errors by training additional local models with data from high-error regions. The construction processes for such a multi-scale fusion model are also provided. Finally, detailed experiments are conducted, comparing the DFWFM-DMFM with five other prediction models. Experimental and comparative results demonstrate that this strategy substantially enhances the prediction accuracy in specific local areas, thereby improving the overall performance of the forecasting model.