Research On Routing Algorithm Of Underwater Wireless Sensor Network Based On Reinforcement Learning

As a product of the third revolution of the information technology industry,the Internet of Things(Io T)technology makes it possible to connect all things.Internet of Underwater Things(Io UT)as an extension of the Internet of Things technology makes it possible to monitor largescale waters.With the development of wireless communication technology,the Underwater Wireless Sensor Network(UWSN)composed of sensor nodes has become the carrier of this technology.The collection of water information usually requires multiple nodes in the network to work together.However,due to the characteristics of the underwater environment,the communication between nodes has the disadvantages of low bandwidth,high delay,high transmission loss,and network topology changes.How to design and implement efficient and stable routing algorithms for UWSN is a challenging content.The traditional UWSN routing algorithm mainly considers the current state of sensor nodes when making decisions,and lacks flexible control and global control of the entire UWSN.With the improvement of sensor performance and the improvement of intelligent algorithm related theories,the application of intelligent algorithms to UWSN routing design is a promising content.According to the characteristics of UWSN and the design requirements of communication routing algorithms,this paper proposes a Fuzzy logic Combined with Qlearning Opportunistic Routing Protocol(FCQOR)based on the combination of fuzzy logic and reinforcement learning.The research content of this article mainly includes the following aspects:(1)In this paper,a method to deal with the influence of node mobility on network topology is proposed,and defines the concept of the applicability of node packet forwarding based on node energy and time delay factors,and designs a prediction model of the applicability of node packet forwarding based on multi-factor fusion;(2)According to the suitability of node forwarding packet,the adaptability learning of node forwarding packet is carried out based on reinforcement learning,and the state-action value,immediate return and update process of the learning process are designed,so as to accelerate the learning convergence speed and improve the adaptability of the algorithm to the network;(3)From the perspective of accelerating the speed of algorithm convergence,an update strategy based on the destination node is proposed in the initial stage of route establishment,and an update strategy based on dynamic threshold is proposed during the routing operation;(4)An opportunistic attribute-based packet forwarding strategy is proposed to increase the reliability of the designed routing algorithm.In this strategy,a hold time mechanism is used to suppress redundant packets,define the expected link quality of a hop node,and use it to limit the number of packet forwarding candidate nodes.Finally,the performance of FCQOR algorithm is tested on the network simulation platform NS-3,and the performance of the algorithm is further analyzed and compared with the existing routing algorithm.Simulation results show that FCQOR algorithm has good applicability to the topology changes brought by node mobility.Under the same network size,FCQOR algorithm has better network performance than other algorithms.