Bayesian Learning of Optimal Utility Functions for Message Routing in Vehicular Ad Hoc Networks

This paper explores the application of machine learning techniques for acquiring utility functions in message routing within Vehicular Ad Hoc Networks (VANETs), aiming to enhance the performance of message routing. Message routing in VANETs employs a store-carry-and-forward protocol, where vehicles (nodes) holding messages traverse suitable trajectories and opportunistically forward messages upon entering the communication range of other nodes. The core of the protocols involves context-dependent decision-making on whether to forward messages to an encountered node and, if so, determining which message to transmit. To enhance the message arrival to the destination with short delay, we introduce a method building upon Wu et al.’s prior work on utility function acquisition through the structure learning of Bayesian networks. Our proposed method incorporates two key innovations: (1) integrating destination information of forwarded messages as an attribute within the Bayesian network and (2) extending the crossover operation from genetic algorithms to optimize the attribute order in the structure learning. Through extensive experiments conducted on the ONE simulator, our method demonstrates superior utility compared to the existing baselines, showcasing its effectiveness in VANET message routing improvement.