全球能源互联网2025，Vol.8Issue(3)：368-377,10.DOI:10.19705/j.cnki.issn2096-5125.2025.03.012
面向数字孪生配电网的低时延业务决策方法
Low Delay Service Decision Method for Digital Twin Distribution Network
摘要
Abstract
Digital twin(DT)can significantly improve the real-time decision-making ability of the distribution network by learning optimal service decisions through model parameter iterations.On the other hand,edge computing can reduce the communication and computational burden of data-driven DT models by sinking model training tasks to the edge side.However,the distribution of resources and model training requirements are incompatible,resulting in a lot of resource waste.This paper proposes a low delay service decision method for digital twin distribution network.First,a low-latency service decision-making framework is constructed,and edge collaboration is used to improve the accuracy and efficiency of distribution network DT model training.Establish the optimization problem of minimizing the weighted sum of model training delay and loss.Second,an edge cooperative decision algorithm based on bidirectional greedy evolution is proposed.Q learning is utilized to jointly optimize the timeliness and accuracy of model training under the condition of unknown global information,and solves the problem of cooperative server selection conflict based on greedy strategy.Finally,the simulation results show that compared to reinforcement learning-based service decision(RSD)algorithm and greedy-based service decision(GSD)algorithm.The proposed algorithm reduces the global loss function by 43.32%and 71.33%,and reduces the total delay of global model aggregation by 14.87%and 56.42%.关键词
配电网/数字孪生/边缘协作/低时延业务决策/双向贪婪演进Key words
distribution network/digital twin/edge collaboration/low latency service decision/bidirectional greedy evolution分类
动力与电气工程引用本文复制引用
彭琳钰,刘晴,刘旭,汤玮,郑智浩,刘康,廖畅..面向数字孪生配电网的低时延业务决策方法[J].全球能源互联网,2025,8(3):368-377,10.基金项目
贵州电网有限责任公司科技项目(066500GS62200017). Science and Technology Project of Guizhou Power Grid Co.,Ltd.(066500GS62200017). （066500GS62200017）
