Design And Optimization Of Online Task Allocation Mechanism Based On Lyapunov Optimization Theory

With 5G communications,new sensing technologies based on portable mobile devices and digital signals have emerged,and technologies such as the Internet of Things,cloud computing,and big data have rapidly evolved to enable larger,more portable,decentralized technologies to collect and process data.Under the large-scale and portable distributed data collection and processing method,an effective task allocation mechanism has become a powerful means to control resource allocation and save costs.However,the fixed allocation method and the incompleteness of participant information hinder the design of the task allocation mechanism.The fundamental reason is that the system lacks dynamic adjustment for random participants.The offline task allocation mechanism is often too ideal and practical.There are very few cases in which the participants have complete information.Therefore,it is necessary to design a dynamic task allocation mechanism to effectively motivate participants to participate in tasks,and to dynamically adjust the allocation method to maximize the system goals.For dynamic task allocation in mobile crowdsensing(MCS),sweep coverage and stability control are two key issues that need to be solved in MCS task allocation.A fair online task allocation mechanism is proposed,which is composed of an online rating protocol and a stability control strategy.Among them,the online rating protocol integrates task quality,rating update,collusion identification,and payment determination,which can simultaneously deal with participants’ free-riding and collusion behaviors.Stability control strategies aim at scenarios where the future information of tasks and participants cannot be predicted,and decisions are made only rely on the current information in order to maximize social welfare and balance network stability under proportional fairness constraints.Finally,through rigorous theoretical analysis and experimental comparison with two benchmark experiments,the correctness and effectiveness of the online perception task allocation mechanism proposed in this paper are jointly proved.For online task allocation in edge computing based on federated learning,two Lyapunov-based online task allocation control strategies,FedOCA and LDRL-Fed,are proposed to intelligently select edge nodes to participate in model training tasks in each round according to the state of the node data set.Among them,FedOCA is an online task allocation control strategy.Aiming at the problem that the future information of the edge node data cannot be predicted,based on the Lyapunov optimization theory,only the current information is used to formulate an independent online control strategy,which can effectively assign the model to the edge node and dynamically adjust the assignment strategy in each round.Aiming at the convergence accuracy of Non-IID data for federated learning,a task allocation strategy based on deep reinforcement learning,LDRL-Fed,is proposed to improve the accuracy of the convergence rate for Non-IID data.We rigorously prove the feasibility of the strategy FedOCA from the theoretical aspect.Based on the two data sets of MNIST and Fashion MNIST,we compare the strategy LDRL-Fed and FedOCA with Fedavg and verify the correctness and effectiveness of the online task allocation mechanism in terms of the degree of data Non-IID,training rounds and batches,and the number of nodes.