An Optimized Resource Allocation for Task Offloading in MEC-Enabled 5G-V2I Network Congestion

Abstract

This paper proposes an adaptive congestion-aware task offloading framework for 5G vehicular-to-infrastructure (V2I) networks with Mobile Edge Computing (MEC). We develop a deterministic fluid-limit approximation of the underlying data offloading queue to capture congestion dynamics arising from varying vehicular densities, and integrate it with an Incremental Double Deep Q-Network (I-DDQN) algorithm that adapts its policy across unpredictable traffic scenarios. Unlike conventional DDQN baselines, the proposed method incorporates incremental state-value updates, recent-experience replay, and queue-aware reward shaping, enabling fast behavioural adjustment without model retraining. Comprehensive evaluations using a custom SUMO–Mininet co-simulation of Dublin’s urban topology demonstrate that I-DDQN reduces average latency by 22%, transmission energy by 27%, and improves bandwidth utilisation by 19% compared to DDQN and DQN under congestion levels ranging from 100–450 vehicles. These results validate the effectiveness of the deterministic approximation and the incremental RL design in maintaining low-latency, energy-efficient offloading in highly dynamic V2I environments. The findings also indicate that I-DDQN continues to deliver improvements in latency, energy consumption, and bandwidth utilisation, even when the initial assumptions about traffic and buffer conditions are relaxed.