Task Offloading Management in the Internet of Vehicles Based on an Improved TD3 Algorithm

Abstract To reduce latency and energy consumption of vehicles in the Internet of Vehicles (IoV) environment and achieve reasonable task offloading, this paper proposes a task offloading algorithm named SCSO-GRU-TD3 (SGT) based on a three-dimensional environment. The SGT algorithm aims to minimize the weighted cost of latency and energy consumption, modeling the task offloading problem as a Markov Decision Process. Building on the Twin Delayed Deep Deterministic Policy Gradient (TD3), it employs a GRU network optimized by a Grouped Auxiliary Memory(GAM) to learn environmental information and guide TD3 in making decisions. Furthermore, it introduces a Sand Cat Swarm Optimization algorithm that combines Whale Optimization, Elite Guidance, and Cauchy Mutation to explore the solution space and enhance sample quality, thereby improving model stability and convergence performance. Simulation results demonstrate that, compared with the TD3, MGU-TD3, GRU-TD3, and BiLSTM-DDPG algorithms, the proposed SGT algorithm reduces the total cost (weighted sum of latency and energy consumption) by 69%, 70%, 9%, and 5%, respectively.