Short-Term Photovoltaic Power Combination Forecasting Method with Time-Varying Weights

Abstract

This letter presents a variable weight short-term photovoltaic (PV) power combination forecasting method based on similar numerical weather prediction (NWP)data. This method firstly identifies and reconstructs the abnormal data, and divides the photovoltaic power data after data processing into the training set, intermediate set, and test set together with NWP data. Secondly, the parameters of single prediction models are optimized by training set data. Thirdly, in the intermediate set, based on the prediction results of single models and the actual photovoltaic power data, the optimal weight coefficient at each time is obtained by constructing the objective function and constraint conditions. Lastly, by matching the NWP data of the intermediate set closest to the NWP data to be predicted, the corresponding weight coefficient set is extracted to determine the time to be predicted. Several experimental results are provided, demonstrating the superiority of the proposed variable weight combination model over both fixed weight combination models and single prediction models.