Task Classification Based Resource Allocation in Parked Vehicle Assisted Vehicle Edge Computing: A Deep Reinforcement Learning Approach

With the development of the Internet of Vehicles (IoV), Vehicular Edge Computing (VEC) is proposed and widely studied as a solution to solve the problem that in-vehicle applications need a lot of computing resources to maintain operation. However, the computing resources of edge servers in VEC are limited compared to cloud servers. To improve computational efficiency, more accurate computational strategies need to be provided for different types of tasks, and available computing resources should also be fully explored. In this paper, we first construct a multi-vehicle VEC offloading model that takes into account the computing resources of parked vehicles. Considering different execution time and energy consumption constraints of tasks, we then classify tasks into different types, to make more appropriate decisions. After that, an optimization problem is formulated to minimize the average weighted cost of time slots. We leverage the Markov Decision Process (MDP) to model the optimization problem. To solve this problem, we propose a value decomposition network based task offloading and resource allocation algorithm (VDNBA), so that each vehicle can independently make its partial offloading decision and computing resources allocation strategy. Eventually, extensive simulations with comparative benchmarks are conducted to evaluate the effectiveness of our proposed algorithm.