基于优化的风电与光伏聚合容量估计及馈线功率预测方法

In 2050, clean deep electrification will require the connection of innumerable distributed energy resources (DERs) to power distribution utilities. The current utility practice of requiring complete DER information for operational applications, such as short-term feeder power prediction (FPP), will not be feasible in the future considering innumerable DERs. Unlike existing studies on DER hosting capacity estimation—which focus on determining the maximum DER capacity a system can integrate without grid reinforcements—this study addresses the future challenge of estimating the aggregate capacity (EAC) of connected DERs. Additionally, it aims to forecast the feeder power at the medium-voltage (MV) level. The proposed approach integrates advanced optimization techniques with deep neural network (DNN) models. An optimization method is introduced, where the first stage involves training basic solar and wind DNN models. In the next stages, EAC values for DERs are determined, and a load model is developed. Once the models for DERs and loads are trained and their aggregate capacities are determined, the framework enables real-time short-term FPP by utilizing weather and chronological data as inputs. The proposed model is tested on real utility data, demonstrating an average accuracy of 97.45% for EAC and 97.29% for FPP. A comparison with a direct FPP shows the superiority of the proposed sequential approach.