Reinforcement Learning-Based Edge Server Placement in the Intelligent Internet of Vehicles Environment
Zhou Zhou, Jemal Abawajy
Abstract
Generative AI-enabled Intelligent Transportation Systems (ITS) are revolutionizing modern transportation by improving safety, efficiency, and adaptability, fostering the development of intelligent Internet of Vehicles (IoV) ecosystems. A critical component of such systems is the deployment of edge servers, which provide localized, low-latency computing, storage, and processing capabilities near data sources. However, determining the optimal placement of edge servers to maximize safety, efficiency, and adaptability in IoV systems remains a significant challenge. Current solutions often optimize one or two metrics, such as energy efficiency or latency, but fail to account for holistic performance improvements. This paper proposes an edge server placement algorithm, ISC-QL, which combines Improved Spectral Clustering (ISC) and Q-Learning to address this gap. In the first stage, the ISC algorithm clusters base stations based on their geographic distribution and the number of connected users, identifying preliminary deployment locations for edge servers. In the second stage, a Q-Learning approach is employed to dynamically refine these initial placements, ultimately determining the optimal locations for edge servers. Simulation experiments using real-world data from Shanghai Telecom validate the effectiveness of the ISC-QL algorithm. Compared to benchmark approaches, ISC-QL demonstrates significant performance improvements, including a 50% increase in load balancing, a 22% reduction in average energy consumption, and a 16% decrease in average delay.