Research On Computing Task Offloading Of Mobile Edge Computing Based On Trust Evaluation

With the continuous development of unmanned aerial vehicle(UAV)technology,UAV edge computing has emerged as a new computing paradigm with wide application prospects and development potential.However,UAV edge computing currently faces the problems of high costs and insufficient security evaluation indicators.To address these issues,this thesis proposes a trust evaluation and task caching-unloading scheme for UAV edge computing.Firstly,three indicators,including privacy degree(PD),monitoring effectiveness against malicious hijacking(MH),and the ability to defend against black hole attacks(BH),are established to measure the privacy,security,and anti-attack capability of the UAV edge computing system.Based on these three indicators,a trust evaluation model for UAV edge computing is established,and reinforcement learning algorithm is adopted to optimize the trust value of the UAV edge computing system.Secondly,from the perspective of UAV edge computing nodes,a multi-user edge computing model is established,and the waiting probability indicator is introduced to evaluate the computing efficiency of the mobile edge computing system.The cost function of the mobile edge computing system is constructed to transform the multi-objective optimization problem into a single-objective nonlinear programming problem.For solving the problem,Monte Carlo tree search and decision tree pruning algorithms are used to obtain the optimal unloading strategy from the perspective of the UAV edge node.Finally,based on the cost function and trust evaluation model,a cost model for Io T devices in UAV edge computing is established,with the optimization objective of minimizing delay and energy consumption,and the trust value as a constraint.To address the issue of poor convergence and local optimal solutions in existing reinforcement learning algorithms,this thesis proposes the Monte Carlo Q-learning algorithm.The experimental results show that the proposed Monte Carlo Q-learning algorithm has good convergence and ability to escape from local optimal solutions,with a maximum improvement of up to 18.4% compared to other algorithms.