Traffic Signal Control Based on Adaptive Neuro-Fuzzy Inference

Abstract

An adaptive neuro-fuzzy inference system is developed and tested for traffic signal controlling. From a given input data set, the developed adaptive neuro-fuzzy inference system can draw the membership functions and corresponding rules by its own, thus making the designing process easier and reliable compared to standard fuzzy logic controllers. Among useful inputs of fuzzy signal control systems, gap between two vehicles, delay at intersections, vehicle density, flow rate and queue length are often used. By considering the practical applicability, the average vehicle inflow rate of each lane is considered in this work as inputs to model the adaptive neuro-fuzzy signal control system. In order to define the desired objectives of reducing the waiting time of vehicles at the signal control, the combined delay of vehicles within one signal cycle is minimized using a simple mathematical optimization method The performance of the control system was tested further by developing an event driven traffic simulation program in Matlab under Windows environment. As expected, the neuro-fuzzy logic controller performed better than the fixed time controller due to its real time adaptability. The neuro-fuzzy controlling system allows more vehicles to pass the junction in congestion and less number of vehicles when the flow rate is low. In particular, the performance of the developed system was superior when there were abrupt changes in traffic flow rates.