Reinforcement Learning-Driven Edge Server Deployment for Multi-Objective Optimization in the Intelligent Internet of Vehicles
Abstract
The deployment of edge servers (ESs) in the autonomous Intelligent Internet of Vehicles (IIoV) environment is essential for ensuring low-latency data processing, efficient resource management, and real-time decision-making in vehicular networks. This paper proposes a two-stage multi-objective optimization algorithm, termed EOBS-MFRL, which combines the Elite Opposition-Based Hunger Games Search (EOBS) algorithm with Mean Field Reinforcement Learning (MFRL). In the first stage, the EOBS algorithm is employed to determine the initial deployment positions of ESs, and in the second stage, MFRL is used to refine these positions to achieve optimal deployment. Unlike previous algorithms that focus on a limited set of factors, such as energy consumption and time delay, the EOBS-MFRL approach simultaneously optimizes load balancing, latency, energy consumption, and network reliability. Experimental results using the Shanghai Telecom dataset demonstrate that, compared to benchmark algorithms, the EOBS-MFRL algorithm improves load balancing by 42%, reduces average latency by 12%, decreases energy consumption by 3%, and enhances network reliability by 39%. The results underscore the effectiveness of the EOBS-MFRL algorithm in optimizing ESs deployment within IIoV systems.