The Research Of Sleep-scheduling Strategy In Wireless Sensor Networks Based On Reinforcement Learning

With the rapid development of micro-electromechanical systems,wireless sensor nodes are provided with data sensing,wireless communications,collaboration and other functions.It can obtain the physical information of the sensing target within the network coverage area by immersion easily.Generally,wireless sensor networks deploy a great number of sensor nodes by means of broadcasting in untouchable regions.Due to the miniaturization of sensor nodes,most of them are equipped with a limited and non-replaceable power supply.Therefore,it is not only impossible to replace or recharge the energy for the node(the deployment environment is unreachable),but also unrealistic(the number of nodes is too much).So,how to utilize the node energy effectively and maximize the WSN's lifetime are the key factor to evaluate the network quality of service.Firstly,because of the deployment characteristics of sensor nodes in the target area which is large scale and high density,it may lead the overlapping sensing regions of the nodes.Therefore,the overlapping regions will cause high data redundancy for node collection and transmission,not to mention the data collected by part of nodes are not different from the data collected by all the nodes.Secondly,that all the nodes work at the same time will lead to singnal channel interference and network congestion,reduce network throughput,therefore,the data of nodes is also easy to be lost.In the condition of guaranteeing network performance,scheduling part of nodes into sleeping state properly and timely,which can effectively improve the energy consumption efficiency and prolong the network lifetime.The main content of this paper is on the basis of understanding and mastering the characteristics of wireless sensor networks.By studying the characteristics of reinforcement learning algorithm and using the advantages of reinforcement learning distributed and environment-free model,this paper studies the sleep-wake scheduling problem in WSN network.The main work of the paper is as follows:(l)By the point of view of the network,we propose a Q-learning based sleep-wake scheduling algorithm(QSA)for wireless sensor network.Firstly,the wireless sensor network is regarded as a multi-agent system.The Markov Decision Process(MDP)is used to build the model of problem,then system's state set,action set and reword function are defined following.And then we use the QSA algorithm to maximize the value function by trial-and-error process,and finally it will get an optimal scheduling strategy.The experimental results show that QSA algorithm has higher energy efficiency as well as ensure an acceptable perception efficiency.(2)To improve the node energy utilization,a Q-learning based MAC layer scheduling approach(QMSA)for wireless sensor network is presented.QMSA deals with the problem of sleep-wake scheduling of the individual sensors for wireless sensor networks in MAC layer.By directly interacting with the surrounding environment in a distributed fashion,the multi-agent system is able to learn an efficient policy to scheduling the sensor nodes.Simulation results reveal that the proposed algorithm is able to fulfill the energy consumption of the network to maximize the network's lifetime as well as ensure an acceptable perception efficiency.