基于改进Informer的多时间尺度电动汽车可调度容量时空分布预测OA

High-precision and fine-grained multi-timescale prediction of electric vehicle(EV)schedulable capacity is of great significance for large-scale EV participating in power grid ancillary services.However,existing EV schedulable capacity prediction algorithms suffer from a significant reduction in prediction accuracy and efficiency when the prediction sequences are long.To address the requirements of multiple dispatching scenarios in the power system,this paper proposes an improved Informer algorithm to construct a multi-timescale,aggregated,schedulable capacity prediction model for electric vehicle aggregators(EVAs)in terms of day-ahead,ultra-short-term,and real-time scales.First,the historical schedulable capacity data of EV As obtained from processing large-scale individual EV charging records is used as the data set of the improved Informer prediction algorithm.Then,the convolutional sparse attention mechanism is used to replace the attention mechanism in the traditional Informer algorithm,and an improved Informer algorithm is constructed to enhance its performance in capturing trend information of historical time sequences,thereby improving the prediction accuracy of long sequences.Finally,more than half a million actual historical charging records of EVs in a city are used as samples to validate the proposed model at multiple timescales and spatial dimensions.The results show that the proposed method significantly improves the prediction accuracy on different timescales as well as reduces the prediction time of long sequences.