Optimizing AoI in Mobility-Based Vehicular Fog Networks: A Dueling-DDQN Approach

Vehicular fog computing (VFC) is a growing approach for delivering low-latency services to IoT devices and intelligent traffic systems. In VFC, vehicles and roadside units (RSUs) collaborate to form a fog infrastructure, processing and storing real-time traffic data at the network edges. However, vehicle mobility creates challenges in maintaining stable communication and network connectivity. Differences in distance, location, speed, and direction between vehicles and infrastructure can disrupt vehicle-to-infrastructure (V2I) communication, impacting reliability. To address these challenges, we explored dynamic fog formation using mobile vehicles and RSUs in city scenarios. We focus on real-time RSU association and fog formation to prevent outdated information from causing failures in V2I communication. We propose an approach utilizing deep reinforcement learning (DRL), particularly the dueling double deep Q-network (Dueling DDQN) algorithm. Through simulations using the Constant Speed Mobility (CSM) model, we compared its performance with DQN, DDQN, and Dueling DQN. Results show that Dueling DDQN outperforms the other methods, effectively improving the freshness of real-time system information.