Proactive MEC Resource Allocation Using Deep Learning for Real-Time Collision Avoidance in Connected Vehicular Networks

Abstract

This paper proposes a proactive resource allocation framework for multi-access edge computing (MEC) systems in connected vehicular networks, targeting real-time collision avoidance applications. In such environments, dynamic workload variations caused by vehicle mobility and traffic density can lead to inefficient resource utilization and latency violations under conventional reactive allocation strategies. To address this challenge, we develop a deep learning-based prediction model using a Gated Recurrent Unit (GRU) to forecast short-term CPU and memory demand. The predicted workload is integrated into a latency-aware orchestration mechanism that performs proactive resource allocation under capacity constraints. Extensive simulations using real-world vehicular mobility traces demonstrate that the proposed approach significantly improves system performance. In particular, it reduces latency and SLA violations by more than 50% compared to reactive allocation strategies, while improving CPU utilization efficiency. The results highlight the effectiveness of combining deep learning-based workload prediction with proactive resource orchestration for next-generation MEC-enabled vehicular systems.