GPU based ordinal optimization for traffic signal coordination

Abstract

Traffic signal coordination has long been a hot topic in Intelligent Transportation Systems (ITS) research. Simulation-based optimization is an important method to optimize the coordination designs as the traffic systems are not easily described by mathematical models accurately. For the traffic simulator in the method, micro-simulation is becoming more and more popular than the macroscopic or mesoscopic simulation in recent years, as the micro-simulation can describe drivers' decision-making process and their driving behaviors in the framework of ITS. However, the computing burden for the micro-simulation is usually very heavy as a large number of vehicles are modeled and simulated separately. For the optimizer in the method, iterative algorithms such as Genetic Algorithms (GA), Ant Colony Optimization (ACO), and Particle Swarm Optimization (PSO) are widely used to find the global optimal solutions. However, these algorithms usually take too much time to converge when a large-scale road network is encountered. To alleviate the computing burden and speed up the convergence, we build an agent-based traffic simulator and employ Ordinal Optimization (OO) as the optimizer. We accelerate both the simulator and the optimizer with Graphics Processing Unit (GPU), which has been applied successfully in many areas for parallel computing. Simulation results show the effectiveness of the OO method and the power of GPU for parallel computing.