Research On Machine Learning Based Power Controlling In Security Transmission Systems With Energy Harvesting

It is an environment-friendly and high cost-effectiveness power supply solution that collecting energy from the environment by using energy harvesting technology to power the nodes in wireless communication systems and networks.However,the randomness and uncertainty of the energy harvesting process also bring new challenges.The energy management and power control at the energy harvesting communication nodes is one of the important problems that need to be solved.Reinforcement learning is an adaptive online learning method that does not rely on the prior knowledge of the environment.The agent learns the strategy in the interaction process with the environment.This thesis solves the power control problem in security communication systems with energy harvesting based on reinforcement learning.The contents of the thesis are as follows.1.The transmission power control in a wireless security communication system consisting of an energy harvesting source node,a destination node and an eavesdropping node is studied.Under the condition that the environmental state information is unknown in advance,the online power control algorithm based on reinforcement learning is presented,which relies only on the current system state.The power control process is modeled as a Markov decision process,and two power control schemes are given.The first method discretizes the continuous state first,and then uses the Q learning algorithm to obtain the transmission power.The second method uses neural network to approximate the continuous Q value function,and then use deep Q network to obtain the transmission power.2.The problem of transmission node scheduling and transmission power control in a wireless security communication system composed of multiple energy harvesting source nodes,a destination node and an eavesdropping node is studied.The environmental state information in the system is unknown in advance.Firstly,the transmission node is selected form the perspective of the security performance and energy efficiency.Then the power control problem is map to a Markov decision process.The environmental information is dealt as the state,the power is dealt as the action,and secrecy rate is dealt as the reward.Finally,an algorithm based on deep deterministic policy gradient is used to maximize the average secrecy rate.