Research On Target Tracking Technology Based On Deep Reinforcement Learning In The Internet Of Vehicles

The Internet of vehicles(Io V)is the application of the Internet of things in the field of intelligent transportation,and it is also an important part of the intelligent transportation system.In essence,the Io V is a huge wireless sensor networks(WSNs),in which wireless communication between roadside nodes and vehicles can be carried out,so as to sense the vehicle position in real time and interact with the vehicle data.In order to ensure the real-time communication between the vehicle and the network,the vehicle network needs to have the ability of high-precision target tracking.However,the current main tracking methods rely on traditional technologies such as GPS,which do not have enough tracking accuracy,and are vulnerable to weather and construction factors.Based on the target tracking scenario in the Io V,this paper designs three dynamic scheduling algorithms for sensor nodes based on reinforcement learning(RL),and achieves the following results.(1)In the target tracking environment of WSNs,based on the characteristics of the nodes in the network,the system needs to schedule sensor nodes dynamically to reduce the sensor node operation energy consumption under the condition of ensuring the tracking accuracy.To solve the dynamic scheduling problem of WSNs nodes under time synchronization,a value iterative scheduling algorithm based on Kalman filter(KF)tracking is proposed to make sequential decision for sensor nodes.The algorithm can dynamically adjust the active range of sensor nodes based on the target position predicted by KF.The simulation results show that the proposed Q-learning node scheduling algorithm can consume less energy while maintaining the same tracking accuracy compared with the sarsa node scheduling algorithm.(2)Further considering the problem of data transmission time asynchrony between sensor nodes in the above model,a target tracking model based on extended Kalman filter(EKF)is constructed.Because the EKF will produce a higher dimension state space in the model,a double-layer depth value iterative scheduling algorithm based on the EKF is proposed.The algorithm is based on the EKF to predict the target position,and the result is used as the center to dynamically adjust the activation range of sensor nodes.The simulation results show that under the same tracking accuracy,the proposed node scheduling algorithm based on doublelayer depth Q network can save more sensor node operation and communication energy than the value iterative scheduling algorithm based on KF tracking.(3)In the process of EKF,the tracking results will produce errors due to omitting the higher-order terms in the state transition function,while particle filter(PF)can predict the target through Monte Carlo(MC)sampling,which has better accuracy than EKF,so the nonlinear target tracking model is constructed based on PF.However,the complexity of PF algorithm will increase exponentially with the increase of sampling points.In order to improve the convergence speed of the algorithm,a strategy iterative scheduling algorithm based on PF is designed.Based on the prediction results of PF,the algorithm dynamically adjusts the activation range of sensor nodes.The simulation results show that the proposed scheduling algorithm based on depth deterministic strategy has faster convergence speed than the scheduling algorithm based on value iteration.At the same time,compared with the EKF model,the scheduling algorithm based on PF model can save more sensor node operation and communication energy in the same accuracy.