Enhanced Heterogeneous Vehicular Networks With Intelligent Congestion Avoidance Mechanism via Regularized Q‐Value‐Based Graph Generalized Neural Network Transformer

ABSTRACT The rapid development of heterogeneous vehicular networks (HetVNETs) has transformed the transportation industry by enabling vehicle‐to‐vehicle and vehicle‐to‐infrastructure communication. However, as traffic load increases, these networks face severe congestion, leading to unreliable and insecure communication. Congestion control in HetVNETs is challenging due to dynamic topologies, multiple communication protocols, and varying traffic intensities. Traditional congestion control techniques struggle to address these issues, necessitating an intelligent mechanism to detect and prevent data congestion in advance. This research introduces the enhanced heterogeneous vehicular networks with intelligent congestion avoidance mechanism via regularized Q‐value‐based graph generalized neural network transformer (RQ‐GGNN‐ArJ). The proposed hybrid framework integrates a graph‐based generalized neural network for modeling dynamic network topologies, a regularized Q‐value transformer for adaptive dedicated short‐range communications (DSRC) transmission power control to ensure real‐time congestion mitigation, and artificial jelly‐driven adaptive optimization (ArJ‐AO) for fine‐tuning weight parameters and loss functions. These components collectively form a highly efficient congestion avoidance mechanism with real‐time decision‐making capabilities. The proposed system achieves remarkable performance, with 99.8% prediction accuracy in identifying congestion patterns, a 99.6% reduction in packet loss, a 99.7% improvement in communication reliability, and 99.3% resource utilization. Therefore, the RQ‐GGNN‐ArJ framework establishes a new benchmark for intelligent congestion management in HetVNETs.