CRLM: A cooperative model based on reinforcement learning and metaheuristic algorithms of routing protocols in wireless sensor networks

• Published in: Computer networks (Amsterdam, Netherlands : 1999) Vol. 236; p. 110019
• Main Authors: Wang, Zhendong, Shao, Liwei, Yang, Shuxin, Wang, Junling, Li, Dahai
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.11.2023
• Subjects: Cluster rules; Cluster-based protocol; Hierarchical routing; Network lifetime; Nonuniform cluster; Wireless sensor networks; Wireless sensor networks; Cluster-based protocol; Nonuniform cluster; Network lifetime; Cluster rules; Hierarchical routing
• ISSN: 1389-1286; 1872-7069
• DOI: 10.1016/j.comnet.2023.110019

In wireless sensor networks, reasonable clustering and routing are keys to efficient energy utilization. However, the selection of cluster heads and routes is NP-hard. Most of the existing routing protocols use heuristic or metaheuristic optimization algorithms to solve this problem. Most protocols regard the selection of the cluster head and routing as two independent problems. However, the selection of cluster heads will affect the selection of routes, and there is a certain relationship between the two stages. Therefore, considering these two problems independently, the solution obtained is not necessarily the optimal solution in the network. In addition, most of the existing routing protocols are still subject to conventional clustering and conventional multi-hop communication in the network, which is extremely unfavorable for reducing the energy consumption of nodes. In this paper, we propose a cooperative model based on reinforcement learning and metaheuristic algorithms called CRLM, in which we use reinforcement learning to enhance the merit-seeking capability of the metaheuristic algorithm and use the algorithm to solve network communication schemes (clustering and routing are considered as one phase). The communication scheme also achieves load balancing of clusters within the network through pruning and employs a novel multi-hop model to reduce network energy waste. Compared to E-ALWO, ChOA-HGS, GATERP, GWO, IPSO-GWO, and LEACH, CRLM has 56%, 95%, 34.5%, 85.7%, 116.7%, and 140.7% improvements in network lifetime. [Display omitted] •Control the updating of meta-heuristic algorithms using reinforcement learning.•The Integrated Cluster-Routing Solution is used to logically pair cluster and route.•A pruning-based cluster pruning rule is used to balance the load on cluster heads.•Spanning multi-hop is developed to reduce the energy consumption of cluster heads.