Hybrid multi-scale feature extraction and parameter optimization for day-ahead photovoltaic power forecasting

As a clean and sustainable energy source, photovoltaic (PV) power generation plays an increasingly important role in modern power systems. However, the inherent variability and uncertainty of PV output, driven by meteorological conditions, present significant challenges for accurate forecasting. To address this issue, this study proposes a multi-level PV power forecasting framework that integrates historical power data, numerical weather prediction (NWP) information, and spatial meteorological features. An optimized XGBoost model enhanced by a genetic algorithm is developed to improve forecasting performance based on historical power data. Furthermore, informative meteorological variables are selected and incorporated to construct a hybrid forecasting model. The impact of higher-resolution spatial meteorological information is also investigated. Experimental results demonstrate that the proposed framework consistently improves forecasting accuracy, with the optimized model significantly reducing prediction errors compared with conventional machine learning approaches. The findings highlight the value of combining historical, meteorological, and spatial information for reliable PV power forecasting, providing support for renewable energy integration and power system operation.