A Surrogate Function in Cellular GA for the Traffic Light Scheduling Problem

The traffic light scheduling problem is undoubtedly one of the most critical problems in a modern traffic management system. Appropriate traffic light planning can improve traffic flows, reduce vehicles’ emissions, and provide benefits for the whole city. Metaheuristics, notably the Cellular Genetic Algorithm (cGA), offer an alternative way of solving this optimization problem by providing “good solutions” to adjust the traffic lights to mitigate traffic congestion. However, one of the unresolved issues is these methods use very time-consuming operations. Specifically, the evaluation is a complex process since a simulator should be executed to get the quality of the solutions. In this work, we focus on this topic and propose using an artificial neural network (as a surrogate system) to tackle this problem. Our experiments show very promising results since our proposal can significantly reduce the execution time while maintaining (and even, in some scenarios, improving) the quality of the solutions.