基于深度强化学习的可信分簇路由协议  

Trusted Clustering Routing Protocol Based on Deep Reinforcement Learning

作　　者：段辉 石琼[1] 师智斌[1] DUAN Hui;SHI Qiong;SHI Zhibin(School of Computer Science and Technology,North University of China)

机构地区：[1]中北大学计算机科学与技术学院

出　　处：《仪表技术与传感器》2024年第2期117-126,共10页Instrument Technique and Sensor

基　　金：山西省自然科学基金项目(20210302123075)。

摘　　要：针对分簇路由协议中恶意节点充当簇头的安全性问题以及基于深度强化学习的路由协议存在收敛慢、波动大的难题,提出了一种基于信任机制和深度强化学习算法soft actor-critic(SAC)的分簇路由协议。该协议首先运用改进的标签传播算法对网络进行分簇。然后采用基于信任的簇头选举机制从簇内选出可信簇头,并采取主-从簇头机制防止簇头“叛变”成为恶意节点。最后利用SAC算法,将簇头作为智能体,实现动态路由决策。实验结果表明:该协议相较于RTRPT、SCR-TBE以及基于DQN、D3QN、PPO的路由协议,具有更优的性能和更好的收敛性。其丢包率、平均时延和网络吞吐量指标均为最优。在多个测试场景下,相较于PPO方案性能最小提升3.97%,最大提升22.39%。Addressing the security issues caused by malicious nodes acting as cluster heads in clustering routing protocols,as well as the challenges of slow convergence and substantial volatility encountered in deep reinforcement learning-based routing protocols,a clustering routing protocol was proposed based on trust mechanism and deep reinforcement learning algorithm Soft Actor-Critic(SAC)was proposed.This protocol integrates a trust mechanism and leverages the advanced deep reinforcement learning algorithm,Soft Actor-Critic(SAC).The protocol employed an enhanced label propagation algorithm to efficiently cluster the network.Then,a trust-based cluster-head election mechanism was utilized to carefully elect trustworthy cluster heads from within the cluster,and a master-slave cluster-head mechanism was adopted,effectively safeguarding against cluster heads transforming into malicious nodes.At last,the SAC algorithm was leveraged to make dynamic routing decisions,with the elected cluster heads acting as agents.Experimental results demonstrate that the protocol has better performance and convergence than RTRPT,SCR-TBE,DQN,D3QN and PPO routing protocols.Its packet loss rate,average delay and network throughput are the best.In multiple test scenarios,the performance of the protocol was improved by 3.97%and 22.39%compared with the PPO scheme.

关 键 词：无线传感器网络 分簇路由协议 网络安全 深度强化学习 信任机制 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]