Research On Handover Mechanism Of Visible Light Heterogeneous Network Based On Reinforcement Learning

The rapid development of information technology,the endless emergence of intelligent services and the diversification of network development have made the spectrum resource bottleneck and security defect of radio frequency communication increasingly prominent.Visible light communication(VLC)has abundant spectrum resources,high security,and strong anti-electromagnetic interference capability.It is a powerful means to solve current communication challenges.However,the characteristics of VLC’s limited coverage and not easy to penetrate the obstruction are the principle of its communication security,and also the reason why the signal is easily interrupted when it communicates alone.Therefore,integrating other wireless communication methods to construct a VLC heterogeneous network is a reliable means to achieve stable and high-speed communication.Vertical handover is a key technology to support user mobility in heterogeneous networks.It is very important to ensure communication continuity and user service quality.Therefore,the research on the handover mechanism of VLC heterogeneous networks is of great significance.The main innovations of this paper are as follows:(1)This paper proposes a VLC and infrared communication hybrid network model for scenes with complex electromagnetic background noise or high security requirements such as ship cabins,aircraft cabins,smart logistics workshops and smart manufacturing plants to provide users with strong connectivity,safe and high-speed communication without electromagnetic interference.(2)This paper designs a VLC heterogeneous network multi-attribute decision vertical handover algorithm based on Q-learning,Sarsa and Sarsa(λ)reinforcement learning algorithm,and establishes the reinforcement learning model of handover problem.In this model,user sensed the environment autonomously,selected the action and optimized the handover strategy according to the feedback of the environment,and realized automatic feature extraction and information modeling.And comprehensively considers three network attributes of received signal strength,data transmission rate and signal-to-interference-to-noise ratio,combined the handover cost to construct a reward function to realize the quantification of environmental feedback,in which the weight factor of each attribute is determined by analytic hierarchy process.With the help of distance attenuation factor λ and E matrix,Sarsa(λ)algorithm strategically updates all steps that promote the occurrence of reward,and optimizes the global environment perception ability of the handover algorithm.Simulation show that compared with Q-learning,Sarsa and the existing handover algorithm,Sarsa(λ)algorithm has faster policy convergence speed,less average handover times and higher system throughput at different user movement speeds.(3)On the basis of traditional reinforcement learning,this paper introduces backward propagation neural network to approximate the Qvalue function,and proposes a multi-attribute decision adaptive vertical handover algorithm for VLC heterogeneous network based on deep Qlearning network.And the empirical playback mechanism is used to ensure the convergence stability of the neural network and the alternative updating method of the target neural network is integrated to realize the time-difference idea.The algorithm optimizes the adaptability of the handover strategy evolution,further improves the convergence speed of the strategy,reduces the average handover times and improves the system throughput.