Research On Mobile Charging Path Planning Algorithm For Wireless Sensor Network Based On DQN

In wireless sensor networks(WSN),energy is one of the most important challenges in the application of wireless sensor networks.Wireless charging technology is undoubtedly a viable solution to the energy problem.One or more wireless mobile chargers(MCs)with large-capacity batteries are deployed in WSNs,which are called wireless rechargeable sensor networks(WRSNs).In the practical application of wireless rechargeable sensor network(WRSN),the energy consumption rate of sensor nodes changes dynamically due to environmental reasons,the amount of data collected and many other uncertain factors.However,existing wireless charging schemes still suffer from high node mortality or low charging efficiency.To this end,this paper proposes a charging scheme based on Deep Reinforcement Learning(DRL): Deep Q Network(DQN)charging algorithm.In the DQN network model,we introduce a neural network to build a bridge between the environmental state and the value of charging behavior.After the training of the DQN network model is completed,the value of actions corresponding to various states can be obtained.MC only needs to read the real-time residual energy of the sensor nodes in the WRSN to make charging choices.This mode is compared with the traditional on-demand charging scheme.Such a mode has great advantages in fairness and adapting to changes in energy consumption of sensor nodes compared to the traditional on-demand charging scheme.The main work of this paper is as follows:(1)We propose a charging scheme based on deep reinforcement learning: DQN(Deep Q Network)charging algorithm.The algorithm uses two identical neural networks to work together,in which the Q network is responsible for training the MC in real time and updating the network weights,the target network is responsible for fitting the predicted value and the target value,and the Q network is updated every fixed time step.The weight of the network is copied to itself until the loss function is completely converged,and the final weight value of the network is calculated.After that,the MC will make the charging selection decision with the highest value according to the state it is in.(2)We build a simulated wireless rechargeable sensor network scenario and train the DQN charging algorithm from different angles.In order to maximize the charging efficiency of the MC,we set the reward value of reinforcement learning as the ratio of the single charging amount of the MC to the sensor node to the single moving distance.(3)We compare the proposed DQN charging algorithm with other algorithms:NJNP(Nearest-Job-Next with Preemption),random strategy and greedy algorithm in many aspects.A large number of simulations show that the DQN charging scheme proposed in this paper is effective in energy utilization and the number of hungry nodes,better than other charging schemes.