Day-Ahead Photovoltaic Power Forecasting Using an NWP-Driven Random Forest with Spatial Downscaling

Abstract Due to the volatility of power generation of photovoltaic (PV) power plants under different weather conditions and the complexity of atmospheric processes, traditional models have certain limitations in prediction accuracy. In this paper, a prediction model based on numerical weather prediction (NWP) is constructed, which utilizes the Random Forest algorithm to predict the generation power of photovoltaic power plants for day-ahead horizons under multiple scenarios, and at the same time leverages its strengths in classification/regression and robust feature ranking. On the basis of the NWP prediction model, this paper further combines grid search and spatial interpolation to propose a spatial downscaling prediction model based on NWP, which improves the prediction accuracy of PV power by increasing the resolution of meteorological data. The example results show that compared with the traditional prediction model, the model proposed in this paper outperforms baselines, which can significantly improve the prediction accuracy of PV power generation and contribute to the further promotion of the cooperative scheduling between the grid and PV in the future.