基于强化学习的电力通信网路由配置优化方法  

Reinforcement Learning Based Optimization Method for Power Communication Network Routing Configuration

作　　者：赵若涵 温树峰 王甜甜 陈泱吟 ZHAO Ruohan;WEN Shufeng;WANG Tiantian;CHEN YangYin(Experimental and Verification Center of State Grid Electric Power Research Institute Co.,Ltd.,Nanjing 210000,China;Bestlink Technologies Co.,Ltd.,Nanjing 210012,China)

机构地区：[1]国网电力科学研究院有限公司实验验证中心,江苏南京210000 [2]嘉环科技股份有限公司,江苏南京210012

出　　处：《通信电源技术》2023年第23期97-99,共3页Telecom Power Technology

摘　　要：常规电力通信网路由的配置优化方法费时费力且效果不佳,因此提出基于强化学习的电力通信网路由配置优化方法。设计中先根据虚拟路由器冗余协议来提升专线利用率,以实现出口流量的负载均衡。在此之后进行基于强化学习动态权重优化策略的制定,将业务量大的流量端所经过的链路权重变小,根据动作离散化提升神经网络策略的更新效率,实现电力通信网的路由配置优化。通过实验证明,提出方法在流量强度分别为50%和100%的网络负载条件下,平均网络延时最大值为0.55 s,配置优化效果较好。The configuration optimization method for conventional power communication network routing is time-consuming,laborious,and ineffective.A reinforcement learning based optimization method for power communication network routing configuration is proposed.In the design,the first step is to improve the utilization of dedicated lines based on the virtual router redundancy protocol to achieve load balancing of export traffic.Afterwards,a dynamic weight optimization strategy based on reinforcement learning will be developed to reduce the weight of the links passing through the high-volume traffic end,and improve the efficiency of neural network strategy update based on action discretization,achieving optimization of power communication network routing configuration.Through experiments,it has been proven that the proposed method achieves a maximum average network delay of 0.55 s under two network load conditions of 50%and 100%traffic intensity,respectively.The configuration optimization effect is good.

关 键 词：强化学习 电力通信网 配置优化 

分 类 号：TM73[电气工程—电力系统及自动化]