AHNet: Implementation of novel vehicle safety measure prediction framework using adaptive hybrid deep learning network in fog computing-enabled Internet of Vehicle

Abstract

Fog computing is one of the distributed and decentralized architectures. Several emerging approaches have been proposed to generate optimal solutions for efficiently managing the latency and bandwidth, but they are not suitable for handling complex network training data. Thus, it can efficiently enhance the computational complexity and minimize the convergence speed due to this; more security-related complexities are presented in the Internet of Vehicles (IoV) devices. To improve the safety measures in vehicles, a novel approach is implemented in this work. At the initial stage, the necessary data is synthetically generated. The collected data has included vehicle sensor information, Roadside Units (RSUs) data and V2V communication data. Then, the Adaptive Hybrid Network (AHN) is employed for predicting the safety measures of vehicles. The AHN is developed by leveraging the strengths of Spatio Temporal Attention-based Fuzzy Autoencoder with a Bi-Directional Gated Recurrent Unit (STA-FA-Bi-GRU). Moreover, several parameters in STA-FA-Bi-GRU are tuned using Revised Fitness-based Humboldt Squid Optimization (RFSO) to attain more accurate vehicle safety measure predicted outcomes. Finally, the proposed model predicts the vehicle’s safety measures such as traffic congestion, accident risk and driver behavior. At last, several experiments are executed in various conditions among the existing frameworks to prove their efficacy. Thus, the proposed approach proved that it has a more reliable performance than the conventional methods to enable better fog computing in IoV.