Relay Selection Algorithm Based On Deep Reinforcement Learning In M2M Communication

I In recent years,with the development of smart terminal multimedia services,more intelligent wireless devices are gradually integrated into life.The Internet of Things(Io T)has occupied an extremely important position in future communications services.It realizes the information sharing and communication between any person and thing at any time and any place.M2 M communication is a very important part of the Internet of Things.Without human intervention,Machine Type Communication Device(MTCD)can autonomously complete data communication with each other.However,because the nodes of the M2 M network have certain rules according to the service characteristics of nearby devices.The traditional relay selection method often considers the channel conditions,and can reduce the transmission power of the device and the node during the transmission on the premise of achieving a certain communication quality,thereby reducing energy consumption.However,if some of the MTCDs communicate more frequently,the number of times that individual relay nodes are selected far exceeds that of other nodes,causing the energy consumption speed to greatly exceed other nodes,resulting in uneven energy consumption.The lifetime of individual nodes will be much smaller than that of other nodes,which degrades the overall performance of the network.If only the residual energy is taken into account,the energy of nodes is equalized,but the energy consumption of single communication will be increased due to the distance or the bad channel condition,and the overall performance of the network will also be reduced.It can be seen that in both reducing energy consumption and balancing energy consumption,an equilibrium point needs to be found to maximize the lifetime of the network.To solve this problem,this paper adopts Qo E index as the evaluation standard of service quality,which is used to calculate the minimum energy transmitted by relay nodes.With the DQN method,the system autonomously selects suitable relay nodes,and finds a balance point in reducing energy consumption and energy balance,thereby extending the life of the entire system.Aiming at the problems of too large randomness and slow convergence speed,it is proposed to add a prior rule before performing the action selection strategy to exclude some relay node selections that cannot be realized in actual situations,which will not affect the original action selection strategy.Through the setting of rules,unnecessary exploration is reduced,and the speed of model training is accelerated.Finally,based on the scenario modeling and the above ideas,a complete process of relay node selection algorithm based on deep reinforcement learning is proposed.In addition,the paper analyzes the algorithm by simulation.First,the parameters of the learning and discount factors in the algorithm and the weight parameters in the reward function were optimized by simulation.Then use the optimized parameter model to compare and analyze with the traditional relay selection algorithm.The simulation results show that the DQN-based relay node selection algorithm comprehensively considers energy consumption and energy balance,and has significant advantages over traditional relay selection algorithms in terms of system life,and can adjust the weight parameters according to subjective requirements.