Customized convolutional network-based road traffic congestion detection in VANET

Abstract

Vehicular Ad-Hoc Networks (VANETs) play a crucial role in modern transportation systems by facilitating real-time communication among vehicles and infrastructure. One of the primary challenges in VANETs is road traffic congestion, which can significantly impact mobility and safety. This paper presents an advanced framework for managing congestion in Vehicular Ad Hoc Networks (VANETs) by combining a Customized Convolutional Neural Network (CCNN) with a Self-Improved Coati Optimization Algorithm (SICOA). The framework encompasses a comprehensive network model for vehicle and infrastructure communication, a Customized CNN-based detection system that estimates congestion probabilities using mobility speed, bandwidth occupancy, link quality, and delay, and a congestion control mechanism optimized by SICOA. The CNN model, designed with multiple convolutional layers, batch normalization, dropout regularization, and a modified loss function, accurately predicts congestion levels and categorizes nodes based on their congestion probability (Cp). The SICOA then optimizes data transmission paths by evaluating fitness criteria such as Received Signal Strength Indicator (RSSI), throughput, energy consumption, and Packet Delivery Ratio (PDR). This integration of machine learning and optimization techniques enhances network performance, increases throughput, and reduces latency, providing a significant improvement over traditional methods and offering a promising solution for efficient VANET congestion management. With the CCNN-SICOA scheme, the results showed a bandwidth of 0.198, a delay of 1.741 s, a link quality rating of 92.319, and a mobility factor of 0.127.