Task Offloading and Resource Optimization Method Based on PER-MATD3
By setting up edge servers at the edge of the network,Mobile Edge Computing(MEC)provides sufficient computing and storage resources for end-user devices nearby and meets the delay and energy consumption requirements of emerging applications.We study the joint optimization of task offloading strategy,computing capacity,and power allocation for multi-users in mobile edge computing,a mixed integer nonlinear programming problem.We formulate it as a Multi-agent Markov Decision Process(MAMDP)to minimize the system cost.To solve this problem,we propose a multi-agent twin delay deep deterministic policy gradient algorithm combined with prioritized experience replay(PER-MATD3),which alleviates the overestimation problem of the value function and increases the learning efficiency.In addition,an offloading action generation algorithm was designed to transform the output continuous action reconstruction into discrete offloading action,and the effectiveness of the offloading action was guaranteed.Simulation experimental results show that the PER-MATD3 algorithm is superior to other benchmark algorithms in the convergence speed and the reward value of convergence in the training process.The proposed method can effectively reduce the total cost and delay in the task off-loading and resource allocation problem of mobile edge computing while maintaining a very low task failure rate.
mobile edge computingtask offloadingresource allocationdeep reinforcement learningprioritized experience replay
万方数据
维普
