Research And Application Of Self-selection Wireless Network Protocol Based On Reinforcement Learning

A reinforcement learning based network optimization scheme running on a control node is designed and implemented for industrial wireless network environments where multiple wireless network protocols coexist,terminal nodes move randomly,and the number of terminals near the access point is constantly changing.Due to the different service types of each terminal node in the industrial network,data exchange between networks in the scenario of converged networks is usually not possible,and network access schemes cannot be selected intelligently,but a control node can be set to control the network access methods in the current area.With the development of machine learning technology,it becomes a feasible solution to apply the new technology to the actual wireless network access selection by machine to judge the network environment and select the end node network access method autonomously.Firstly,this paper designs a network quality data collection model,proposes a way to determine network quality parameters by analyzing the characteristics of different network protocols and synthesizing the indicators common to each protocol,and establishes an environment agent module,which is not only used for the collection of network quality data,but also can provide feedback to the decisions made by the decision-making network part and execute the instructions issued by the decisionmaking network.Secondly,this paper proposes a switching control architecture for DDQN-based network protocols to solve the overestimation problem that can arise in DQN networks,improve the stability of selecting wireless network protocols in dynamic network environments,reduce the ping-pong effect,and lower the average energy consumption.The architecture sets the input data,action space,and reward function of the According to the different services undertaken by nodes,different service nodes are classified,weights are obtained using hierarchical analysis,different service type division is defined,and node conflicts are simulated by random movement of nodes,which enables the algorithm to feedback reliable results of network protocol selection according to the dynamic network environment.Finally,this paper completes the environment model design and simulation of the above algorithm on NS3 network simulation platform and performs the algorithm selection result by feeding back the environment network quality through NS3 platform.Through simulation experiments,it can be seen that the node throughput can be stabilized at the maximum value under the DDQN algorithm prediction,and the optimization scheme is more suitable compared with other reinforcement learning algorithms,and the mean value of node energy consumption is reduced.