Neural network based online traffic signal controller design with reinforcement training

Traffic congestion leads to problems like delays, decreasing flow rate, and higher fuel consumption. Consequently, keeping traffic moving as efficiently as possible is not only important to economy but also important to environment. Traffic system is a large complex nonlinear stochastic system. Traditional mathematical methods have some limitations when they are applied in traffic control. Thus, computational intelligence (CI) technologies gain more and more attentions. Neural Networks (NNs) is a well developed CI technology with lots of promising applications in traffic signal control (TSC). In this paper, a neural network (NN) based signal controller is designed to control the traffic lights in an urban traffic road network. Scenarios of simulation are conducted under a microscopic traffic simulation software. Several criterions are collected. Results demonstrate that through online reinforcement training the controllers obtain better control effects than the widely used pre-time and actuated methods under various traffic conditions.