Research On The Optimization Of Edge Collaborative Resource Allocation In Intelligent Railway

With the development of railway Internet of Things and big data,as well as the improvement of equipment intelligence level,augmented reality,video or voice processing generate massive data,which brings new requirements for network delay and wireless bandwidth.The cloud computing model cannot meet the quality requirements of time-critical services,but the emerging mobile edge computing(MEC)can solve the above problems.In addition,because of the limited computing resources of vehicular terminals of railway and edge servers,and considering the complexity of jointly optimizing computing offloading and resource allocation,the optimization of edge collaborative resource allocation in intelligent railway has important significance.Based on this,the main research contents of this thesis are as follows:Firstly,becasue of the resource limitation of edge servers,a system architecture of edge coordination in intelligent railway is proposed,including computing offloading model,handover model,and delay model,which transforms the MEC network throughput maximization problem into an integer linear programming problem.In the MEC networks,a heuristic-based throughput maximization algorithm(TMEC)is proposed to simultaneously find the shortest path with the maximum delay and the processing edge server,and two benchmark algorithms,namely Min-Load and Min-Link algorithm for comparison.The simulation performance analysis of the task completion rate and execution time of the three algorithms is given.The research shows that when the resources of the edge server are limited,the number of service tasks supported by TMEC is much larger than that of the other two algorithms.Secondly,considering delay-insensitive services such as public broadcast,because of the high complexity of the heuristic algorithm and the continuous optimization goal of resource allocation,an edge cooperative resource allocation algorithm is proposed based on the deep deterministic policy gradient(DECRA),which can achieve adaptive computing offload and resource allocation.To exploit the correlation of input states,a convolutional neural network based edge collaboration resource allocation(CECRA)scheme is proposed.The research shows that both DECRA and CECRA algorithms converge and the cumulative task completion rates reach 94% and 97%,respectively.Finally,considering the delay-sensitive services such as emergency communication,because of the problem of high switching delay of edge servers in private networks,a public-private fusion system model is proposed,and the optimization goal is to maximize the weighted sum computing rate.Since the optimization problem is a non-convex mixed integer programming problem,the problem is decomposed into two sub-problems of offloading decision-making and resource allocation,which are solved by binary search and greedy algorithm respectively.Considering the high complexity of joint optimization of N-order binary variables,we combine deep reinforcement learning(DRL)with knearest neighbor algorithm(KNN)and order-preserving optimization algorithm(OP),named DRL-KNN and DRL-OP.The research shows that both DRL-KNN and DRL-OP can converge and the normalized calculation rate can reach 99% and 95%,respectively.