Intelligent service migration for the internet of vehicles in edge computing: A mobility-aware deep reinforcement learning framework

Abstract

The restricted coverage of edge servers in the Internet of Vehicles (IoV) results in service migration as vehicles traverse various regions, potentially escalating operational costs and diminishing service quality. However, existing service migration schemes inadequately address the dynamic attributes of high-speed mobile vehicles and the temporal variability of the network. To overcome this issue, we propose a mobility-aware deep reinforcement learning framework based on vehicle behavior prediction for service migration. Firstly, taking the service processing latency, migration latency, and energy consumption as metrics, a constrained model is established to minimize long-term costs. Given the considerable uncertainty in the associational behaviors between high-speed mobile vehicles and edge servers, a vehicle behavior prediction method utilizing the Hidden Markov Model (HMM) is then proposed. On this basis, we design a mobility-aware reinforcement learning service migration algorithm based on a Double Dueling Deep Q-Network (D3RLSM) incorporating a prioritized experience replay mechanism to extract vehicular state features accurately and optimize the training process. Compared with several baseline methods, D3RLSM shows its effectiveness in reducing service latency and energy consumption.