Model-based Constrained Bayesian Optimization of IEEE 802.11 VANET Safety Messaging

Vehicle safety remains a critical concern as road accidents occur daily. Wireless communication technologies are increasingly recognized for their potential to enhance the safety of both human-driven and autonomous vehicles. In particular, IEEE 802.11-based communication systems have been proposed and evaluated to improve road safety. Achieving low transmission latency and high reliability is essential for the development of Vehicular Ad Hoc Networks (VANETs). Given the dynamic nature of vehicular environments and stringent quality of service (QoS) requirements, adaptive network configurations are necessary. This paper presents two Bayesian model-based approaches for optimizing adaptive real-time communication parameters to achieve optimal network configurations while satisfying QoS constraints. The first approach integrates a stochastic QoS model with constrained Bayesian optimization algorithms, addressing the complexity and computational demands of traditional network simulations. The second approach incorporates a Deep Learning Neural Network (DLNN) into the Bayesian optimization framework, significantly accelerating the iterative optimization process and enabling adaptation to dynamic communication environments. The optimization algorithms are carefully designed and calibrated to ensure precision and robustness. Experimental results using Python demonstrate the efficiency and accuracy of our methods in rapidly converging to optimal parameters for IEEE 802.11-based VANETs. Compared to our previous models and existing approaches in the literature, the proposed optimization scheme offers substantial improvements in computation time, accuracy, and reliability, supporting real-time optimization of communication parameters.