Traffic variable estimation and traffic signal control based on soft computation

This paper studies traffic variable estimation, and presents a method of estimation for the number of vehicle waiting for queue (NVWQ) based on neuro-fuzzy at urban intersection. We present results of training the neural network for a detectorized intersection in Changsha City. The accuracy of NVWQ estimation using the fuzzy neural networks approaches is more than 90%. The fuzzy neural networks have the advantages of both fuzzy expert systems (knowledge representation) and artificial neural networks (learning). The fuzzy neural networks can be trained successfully to estimate NVWQ for different traffic flow patterns and different conditions of intersection. This greatly reduces a lot of effort of extracting a traffic expert's knowledge into fuzzy if-then rules. All we have to do is to present the training data to the network which can figure out its own rules through internal representation. In the traffic signal control system, detection of traffic variables at intersection, such as NVWQ is very important and is the basic input data to determine signal timing. We also discuss traffic signal control based on fuzzy system and genetic algorithms (GA). The fuzzy controller has the ability to adjust its signal timing in response to the changing traffic conditions on a real-time basis. Our proposed controller produces lower vehicle delays and percentage of stopped vehicles than the traffic-actuated controller.