Bayesian Optimization of Road Pricing using Agent-based Mobility Simulation

Road pricing policies are frequently debated but not widely adopted. Tools for designing near practice-ready policies are still missing, especially considering the complex dynamics between the different levels of traveller decision-making and the networks’ performance. We couple an agentand activity-driven mobility simulator with a Bayesian Optimization (BO) framework for designing optimal road pricing policy in a daily mobility and transportation network system. We extend the literature with a BO-framework application to distance-based road pricing under a departuretime and route-choice sensitive demand model combined with a detailed mesoscopic network. We then tested a general BO and a recently proposed contextual BO algorithm for SimMobility and computational performance. Both identified a similar optimum distance-based pricing, with the second being more computationally efficient. Nonetheless, iterations number, increasing search space and dimensionality could limit their performance. Lastly, the effects of the identified policy were analyzed by leveraging the outcome capabilities of SimMobility.