基于多智能体深度强化学习的车联网任务安全卸载OA

With the development of intelligent transportation,the integration of mobile edge computing has alleviated the re-source limitations of vehicles in the Internet of Vehicles(IoV).However,the lack of efficient task offloading strategies still af-fects the Quality of Experience(QoE)of users.To address the security issues during task offloading,this paper proposes a task security offloading strategy based on Multi-Agent Deep Reinforcement Learning(MADRL).Firstly,a task model,acommunica-tion model,and a computation model are established based on the dynamic environment of vehicles.The task offloading problem is then transformed into a Markov Decision Process(MDP)according to the system model.Secondly,the Multi-Agent Deep De-terministic Policy Gradient(MADDPG)algorithm is utilized to handle the convergence issues of a single agent in a dynamic envi-ronment,ensuring the efficient offloading of tasks.Finally,differential privacy and weighted federated learning are incorporated into the algorithm training process to mitigate the impact of heterogeneous edge node resources and Non-IID data on the training results,while also preventing attackers from inferring the model through parameter reverse engineering,thus protecting user pri-vacy.Simulation results show that the proposed algorithm can effectively improve system execution efficiency while ensuring user privacy during the training process.