基于联邦学习的能源聚合服务商负荷预测OA北大核心

Energy aggregation service providers in the electricity market are highly dependent on the load-forecasting accuracy to maximize their market interests.However,the required forecasting data are typically scattered among different energy suppliers,and owing to data security and privacy concerns,these suppliers are often unwilling to share their data.Inaccurate load forecasting reduces the operational efficiency of energy aggregation service providers and limits the in-depth understanding of supplier operating models.To address this issue,this study proposes a load-forecasting model based on federated learning,which achieves an effective improvement in load-forecasting accuracy while ensuring data security and privacy.First,an artificial neural network with multidimensional environmental feature selection is established according to the task requirements.Subsequently,a weighted-average strategy was used to jointly train the artificial neural network,bridging the gap between individual energy entities and environmental features.Finally,the model features trained separately are sent to the server of the energy-aggregation service provider for the fitting process.The results of a case analysis using actual data from a large energy aggregation service provider in a southern city in China demonstrate that the proposed model significantly improves operational efficiency and forecasting accuracy.