摘要
在新能源广泛接入背景下,配电网业务数据采集终端接入规模与信号采集频次激增,对网络的业务承载能力提出更高要求。该文首先通过整合云边端算力资源,构建以最大化云边端协同处理数据量为目标的优化问题,并引入李雅普诺夫优化理论将其转换为仅依赖当前时隙信息的在线优化问题,实现排队时延与长时平均采集数据量的协同保障;然后,提出基于改进深度Q网络(DQN)的配电网云边端协同处理算法,通过引入基于贪婪策略的Q值排序机制与双重经验回放机制,解决多终端处理决策耦合导致的资源竞争冲突,在保障样本多样性处理能力的同时提升算法的收敛性;最后,仿真结果证明所提算法能够有效适配配电网采集业务高密度、高频次的发展趋势。
With the widespread access of renewable energy,the access scale of distribution network service data acquisition devices and data acquisition frequency have surged.The distribution network acquisition services are rapidly developing towards high-frequency,massive,and computationally intensive directions.It is significant to fully utilize the potential of cloud-edge-end collaboration to enhance the service carrying capacity of the network.Recently,service processing methods based on cloud-edge-end collaboration have been proposed.However,these methods still face several challenges.First,the coupling of long-term constraint guarantees and short-term processing decision optimization makes it difficult for single-slot short-term decisions to achieve long-term constraint coordination.Second,the differentiated performance requirements of services and limited network resources lead to interdependence among multi-device processing decisions.Existing methods lack a collaborative processing mechanism,making it challenging to resolve decision conflicts caused by competition.Finally,most current methods adopt random sampling mechanisms,overlooking the differences among samples in the action experience pool,resulting in poor convergence and optimization performance in resolving competition conflicts under resource-constrained scenarios.To address these challenges,this paper proposes a cloud-edge-end collaborative service processing mechanism for high-frequency data acquisition in distribution network.Firstly,a cloud-edge-end multi-level collaborative service processing framework for high-frequency acquisition in the distribution network is designed.It constructs differentiated models for local computing,edge processing,and cloud processing to meet the varied computing requirements of data acquisition services.Further,under the premise of ensuring queuing delay and long-term average data collection constraints,the objective of maximizing the amount of cloud-edge-end collaborative processed data is set,which ensures sufficient underlying data support for the normal operation of new power services while reducing queuing delay.Subsequently,the concept of virtual queues from Lyapunov optimization theory is introduced to transform the original problem into an online optimization problem that only depends on current slot information.It plays an important role in achieving the coordinated guarantee of delay and throughput.Then,an improved deep Q-network based cloud-edge-end collaborative processing algorithm for distribution network is proposed,which includes five stages of initialization,action selection,conflict resolution,learning,and updating.Specifically,in the action selection and conflict resolution stages,a greedy strategy-based Q-value sorting mechanism is introduced.It selects the action with the highest Q-value as the processing decision of the device for the current slot,and resolves wireless channel and edge server resource selection conflicts caused by multi-device processing decision coupling through edge-end collaboration.In the learning stage,considering the importance of different device services and the confidence of action samples,a dual replay experience pool is designed to ensure sample diversity,effectively avoiding data loss potentially caused by aggressive strategies.This greatly improves the convergence of the algorithm.The proposed algorithm ensures the orderly operation of cloud-edge-end services in distribution networks.Finally,the effectiveness and rationality of the proposed algorithm are verified through simulation examples.The simulation results show that the proposed algorithm can increase the amount of cloud-edge-end collaborative processed data by 11.71%and 14.86%,reduce queuing delay by 24.68%and 26.09%.It can also increase the average data acquisition volume by 8.87%and 7.44%.At the same time,it significantly reduces the backlog of device layer queue backlog and greatly improves the convergence speed of the algorithm.The author team will further consider information synchronization and security issues during data transmission and processing.
作者
于子淇
刘健阳
陈亚鹏
周振宇
孙中伟
Yu Ziqi;Liu Jianyang;Chen Yapeng;Zhou Zhenyu;Sun Zhongwei(School of Electrical and Electronic Engineering,North China Electric Power University,Beijing 102206,China;State Grid Beijing Haidian Electric Power Supply Company,Beijing 100195,China)
出处
《电工技术学报》
北大核心
2025年第11期3502-3513,共12页
Transactions of China Electrotechnical Society
基金
国家电网公司总部科技项目资助(No.52094021N010(5400-202199534A-0-5-ZN))。
关键词
配电网
高频采集
云边端协同
业务处理
深度强化学习
Distribution network
high frequency acquisition
cloud-edge-end collaboration
service processing
deep reinforcement learning
