Optimizing Lane-Change Decisions in VANETs: A Communication-Driven Approach

Abstract

Vehicular ad hoc networks (VANETs) are pivotal in intelligent transportation systems (ITSs), enabling swift vehicle communication. The study focuses on optimizing highway traffic flow, particularly discretionary lane changing (DLC) via vehicle-to-vehicle (V2V) communication. To address the dynamic highway environment, an intelligent DLC decision-making model integrating deep learning techniques is proposed. The model employs the enhanced competitive swarm optimization (ECSO) algorithm for traffic density and the improved locust search (ILS) algorithm for vehicle mobility prediction. The nonlinear autoregressive dynamic neural network (NAR-DNN) serves as the decision-making framework, offering choices such as free lane change, forced lane change, and no lane change. SUMO and NS2 simulations evaluate the model, demonstrating its efficacy in establishing efficient communication links. Results show significant improvements over traditional frameworks, with the NAR-DNN achieving superior performance in packet delivery rates (12%–18%), connectivity probability (10%–15%), message delay (15%–20%), and average lane-change duration (8%–12%), respectively. These findings highlight the NAR-DNN's effectiveness in enhancing traffic management and safety within VANETs, offering promising insights for future ITS advancements.