基于PatchTST-STL模型的光伏发电功率日前预测OA

Accurate day-ahead forecasting of photovoltaic(PV)power is crucial for energy management in electrical grids.Addressing the challenges of high randomness and volatility in PV power,which often lead to low prediction accuracy,this paper proposes a novel day-ahead photovoltaic power forecasting method based on the PatchTST-STL model.The method enhances the input and architecture of the Transformer by introducing the concept of Patches and applying Seasonal and Trend decomposition using Loess(STL).Each time step is redefined as a single token,preserving local dependencies within each token,thereby enhancing the model's ability to capture local patterns while leveraging the Transformer's multi-head self-attention mechanism to extract long-term dependencies from the sequence.Considering the complexity of PV sequence patterns,the STL process is employed to handle multivariate PV sequences,separating them into trend,seasonal,and residual components,which are then inputted as independent channels.The Grey Wolf Optimizer(GWO)is utilized to optimize the Patch size,a critical hyperparameter,ensuring rapid convergence of the algorithm.Experimental results on the DKASC dataset demonstrate that,compared to the Informer model,our proposed method achieves an average reduction of 36.7％in Mean Absolute Error(MAE)and 11.7％in Root Mean Square Error(RMSE).Furthermore,when compared to the traditional Transformer,the MAE and RMSE are reduced by an average of 57.6％and 30.9％,respectively,showcasing the superior performance of our approach in PV power forecasting.