Coordinated Traffic Signal Control Via Bayesian Optimization For Hierarchical Conditional Spaces

We study the problem of coordination control of multiple traffic signals to mitigate traffic congestion. The parameters we optimize are the coordination pattern and offsets. A coordination pattern indicates which traffic signals are coordinated. Offsets show how traffic signals can be coordinated. We aim at finding the optimal combination of the coordination pattern and offsets. In this paper, we treat it as an optimization problem whose search space is a conditional space with hierarchical relationships; the coordination pattern determines the controllability of the offsets. Then, we tackle the problem by proposing a novel method built upon Bayesian Optimization, called BACH. Experiments demonstrate that BACH successfully optimizes coordination control of traffic signals and BACH outperforms various state-of-the-art approaches in the literature of traffic signal control and Bayesian Optimization in terms of best parameters found by these methods with a fixed budget.