Daily Prediction of Photovoltaic Power Generation Based on XGBoost Machine Learning

Photovoltaic (PV) power generation is an important renewable energy generation, but its variability to environmental factors, power generation fluctuates greatly, bringing certain challenges to the grid. In this study, a short-term prediction model for PV power generation using both physical modeling and machine learning techniques. Firstly, based on the geographic location of the PV power plant, combined with the solar irradiation to calculate the theoretical generating power, to obtain the power characteristics of the PV power plant. Then, a PV power plant power generation prediction model based on historical power data and considering NWP is established. After K-means clustering, the model achieves a prediction accuracy approaching 98% under low-temperature weather conditions. Finally, this paper proposes a spatial information reconstruction method integrating Kriging interpolation models with the XGBoost algorithm to address the issue of insufficiently precise weather data in short-term PV power forecasting. Prediction results demonstrate a significant improvement in accuracy compared to pre-reconstruction, with an additional nearly 10% increase in forecasting precision specifically for sunny weather conditions.