Multiple-Time-Scale energy management strategy for virtual power plants considering dynamic weights and a Data-Model driven prediction model

As the installed capacity of wind and solar power keeps rising and electric vehicle charging loads are integrated on a large scale, the uncertainty on both the source and load sides of virtual power plants has notably grown. This makes it arduous to guarantee stable and optimal dispatching. A multiple-time-scale energy management strategy for virtual power plants is proposed to mitigate these uncertainties. On the generation side, a hybrid neural network method for wind-solar output prediction with dynamic weights is proposed, which can greatly eliminate the errors caused by the increase of prediction time scale. On the load side, a data-model driven electric vehicle charging load prediction method is proposed, which combines neural network prediction with road network models. The proposed method improves the accuracy of the basic parameters of the road network model. In addition, improvements are made on the basis of the traditional Multi server Markovian Arrival and Exponential Service Time Queueing Model to accurately describe the queueing and charging behaviours of electric vehicles at charging stations. Case studies validate that the proposed energy management method achieves higher accuracy than the method that directly predicts the electric vehicle charging load using neural networks and the traditional rolling optimization method without dynamic weights, leading to reduced overall costs for virtual power plants. Notably, the proposed method reduces the dispatch cost by 14.83%.