温差驱动下基于多任务学习及可解释机器学习的综合能源系统多元负荷预测OA

With the advancement of the"dual carbon"goals and the rapid development of integrated energy systems,achieving the safe and economical operation of integrated energy systems has placed higher demands on the accuracy of multiple load forecasting.To address the shortcomings of existing research in handling load volatility and coupling,this paper proposes a short-term multiple load forecasting model for integrated energy systems(IES)based on variational mode decomposition(VMD),long short-term memory(LSTM),support vector machine(SVM),and Transformer.First,the influencing factors of multiple loads are analyzed,and the"12-hour temperature difference"is innovatively introduced as a key meteorological feature to capture abnormal load fluctuations caused by drastic temperature changes.Calendar information is quantified using one-hot encoding.Next,the original load series are decomposed using VMD.Then,the sparrow search algorithm(SSA)is employed to optimize SVM parameters and predict each IMF component,which serves as high-level feature inputs.Finally,a multi-task learning(MTL)framework is constructed,where the shared layer uses LSTM to explore the coupling relationships among multiple loads,and the task-specific layers employ Transformer encoders to capture the temporal dependencies of each load for predictions.Additionally,SHAP analysis is used to evaluate the impact of feature variables on the forecasting results.Using actual data from an industrial park in Lanzhou,China,as a case study,the experimental results demonstrate that the proposed model effectively reduces the mean absolute percentage error of load forecasting and achieves satisfactory prediction accuracy.