A Hybrid Data- and Physics-Driven Method for Forecasting Photovoltaic Power Throughout the Entire Snowfall Process

The entire snowfall process can be divided into two stages: snowfall onset and snow covering. During this process, the snow coverage on the surface of photovoltaic panels exhibits a complex time-series variation pattern, and the numerical weather forecast struggles to to reflect the actual solar irradiance received by photovoltaic panels, increasing the difficulty of power generation forecasting. However, there is currently a lack of research on photovoltaic power forecasting throughout the entire snowfall process. Therefore, this study proposes a hybrid data- and physics-driven method for photovoltaic power forecasting throughout the entire snowfall process. Firstly, a power forecasting model based on BiGRU is developed to obtain the initial prediction results, and the differential characteristics of power prediction errors during the snowfall onset and snow cover periods are analyzed. Secondly, a data-driven correction model based on XGBoost and a physics-driven correction model based on the improved Marion model are constructed to determine the correction terms of the forecasting results for the entire snowfall process by leveraging the strengths of the two models. Finally, a reverse difference weighted strategy is employed to dynamically combine different correction results, yielding the final prediction results. The effectiveness of the proposed method is verified using the data from a photovoltaic power station in North Hebei.