Research On Prediction Of Traffic Flow Using Fuzzy Neural Networks

The intelligent transport system (ITS) is an integrated field, which is established for use of changing traffic conditions, reducing traffic congestion, decreasing traffic accidents, so as to provide better services for economic and social development. As one of the ITS crucial problem, how to improve methods of traffic flow prediction and its accuracy become hot spots in the research of traffic flow prediction.This dissertation research on traffic flow prediction based on fuzzy neural network based on fuzzy neural networks and chaotic theory. On the foundation of phase space reconstruction of chaotic theory, the adaptive neuro-fuzzy inference system (ANFIS) and the T-S (Takagi-Sugeno) fuzzy neural network ensemble system are studied, then the two methods are applied to traffic flow prediction. Compared with existing prediction methods using neural networks, experimental results show that the method of fuzzy neural network is effective.The main work of this dissertation includes three aspects as follows:Firstly, the fuzzy neural network structure is studied, subsequently the structure and algorithms of ANFIS and the T-S fuzzy system based on neural network ensemble are studied in detail.Secondly, the chaotic theory is introduced into the traffic flow prediction, and then the chaotic characteristics of traffic flow time series and the selection of embedding dimension and delay time parameter are analyzed. According to the results of the recurrence plot, the predictability of traffic flow series is discussed.Finally, in accordance with experiments based on practical traffic flow series, prediction models established by ANFIS and the T-S fuzzy system based on neural network ensemble are applied to traffic flow prediction of two practical traffic flow series provided by a freeway monitoring station in Beijing and the website authorized by the British department of transportation. After different prediction models based on different embedding dimensions and delay-times is studied, and the predicted performance of conventional neural networks is discussed. The experiment shows that under conditions of optimal embedding dimension and delay-time, methods of the ANFIS and the T-S fuzzy system based on neural network ensemble obtained better predictive ability and have a higher accuracy than other conventional methods of neural networks. In the field of traffic flow time series prediction, fuzzy neural network method is much more effective than conventional methods.