Research On WSN Node Task Scheduling Algorithm Based On Distributed Q-learning

As a self-organizing network,Wireless Sensor Network(WSN)is formed by a number of sensor nodes that receive and process information independently.It has becomes a hotspot in wireless sensor networks research that to use the limited battery power,computing,communication and storage resources reasonably in a time-varing environment.Scheduling tasks reasonably is an effective way to solve the above problems,according to the sensor node detection,processing of environmental information.As the type of small sample machine learning,reinforcement learning can make the agent make task scheduling decisions with limited processing and communication capabilities intelligently and can also make policy changes constantly according to the dynamic environment and application requirements.In our work,based on the traditional Q learning algorithm,a distributed independent Q-learning WSN node task scheduling algorithm based on improved SVM and a distributed cooperative Q-learning WSN node task scheduling algorithm based on global value function are proposed.The thesis uses a distributed independent Q-learning WSN node's task scheduling algorithm based on improved Support Vector Machine(ISVM-Q)for complex WSN scenarios,where nodes only consider their own information and have higher requirements for energy consunption performance.By using a value function as an approximates,the algorithm can effectively reducing the state-action space explosion;the parameterization of SVM model enhances the interpretation ability;the use of a sliding window of the sample is to alleviate the problems caused by too much information;the greedy strategy is combined with simulated annealing,which makes the node effectively explore the action space effectively in the early stage and avoids fell in the local optimal problem.Experimental results show that the proposed algorithm can improve the performance of data collection applications(the success rate of acquisition and reception tasks)while reducing energy consumption.In addition,aiming to solve the problem of the tocal optimal,the thesis changes the delayed rewards dynamically according to the energy and performance,and a distributed cooperative Q-learning WSN node task scheduling algorithm based on global value function(GQ)is designed.The reward information of the nodes is transmitted to the neighbor nodes as valid information,so that the optimal diffusion of the nodes to the optimal performance of the entire network is achieved.In addition,the reward function,which,is the combination of energy consumption and application performance,makes the node not only consider the application performance,but also ensure that the energy consumption is not too large.Secondly,the dynamic exploration strategy design in Q learning avoids the node in the learning process get into the local optimum.The experimental results show that the algorithm can make the node improve the application performance(moving target tracking application)and ensure the network energy consumption decrease slowly by scheduling tasks reasonably.