Multi-Step Short-Term Forecasting of Photovoltaic Power Utilizing Autoformer with Prophet

The prediction of photovoltaic (PV) power generation faces certain challenges, primarily due to the high uncertainty of solar irradiance. The accuracy of PV power prediction is critical for the stability and reliability of power grids. However, existing models often perform poorly in long-sequence, multi-step prediction tasks, and there is still room for improvement in feature extraction from historical data. Therefore, this study proposes a novel forecasting method based on Autoformer and Prophet, combining the advantages of Autoformer in long-term sequence prediction with the strengths of Prophet in feature extraction to enhance the accuracy of PV power generation forecasting. First, the Autoformer encoder extracts seasonal components from complex time series data, while the decoder continuously utilizes the past seasonal components provided by the encoder for optimization. Then, Prophet extracts trend-cycle and seasonal components from the time series data input into the decoder. Finally, Autoformer predicts photovoltaic power generation based on the extracted features. The feasibility and superiority of the hybrid model are verified by comparing it with other models. The results show that the proposed method performs well across various performance evaluation metrics in the short-term PV prediction tasks, significantly outperforming other approaches.