A Probabilistic Routing Algorithm Based on CNN and Q-Learning for Vehicular Edge Network

Abstract

Vehicular edge networks represent a novel architecture that utilizes vehicles as mobile edge nodes, characterized by high-speed dynamic changes. To effectively transmit data in vehicular edge networks, opportunistic routing methods can be employed, selecting suitable relay nodes based on encounter opportunities between nodes. Although existing opportunistic routing algorithms primarily select the optimal transmission path based on the encounter characteristics between nodes, the dynamism of network topology, the uncertainty of node mobility, and the heterogeneity between nodes still pose significant challenges to the implementation of opportunistic routing. In response to this, we propose a probabilistic routing algorithm based on Convolutional Neural Networks (CNN) and Q-learning, named PRCQ. This algorithm predicts node state transition probabilities using decomposed latent node features and dynamically adjusts optimal routing strategies using Q-learning. Extensive simulations were conducted on the NS-2.35 simulator based on two different city scenarios to evaluate the performance of the PRCQ algorithm compared to other existing algorithms. The results indicate that, compared with other existing opportunistic routing algorithms, PRCQ exhibits superior performance in terms of average transmission delay and packet delivery ratio.