Chaotic Analysis And Prediction For Traffic Flow Time Series

It is now recognized that the traffic flow inducible system is the best way to improve traffic efficiency and mobility. The key and prerequisite for achieving traffic flow inducible system is the prediction of chaotic time series, therefore, accurate and real-time traffic flow time series prediction has become one of the hot research to Intelligent Transportation System. Because of the highly nonlinear,complexity and indeterminacy of the traffic flow, means that the traditional traffic flow time series pretreatment technology can not be achieved satisfactory results. The chaos theory researched for the non-linear characteristics has known as the 20th century's third revolution to natural sciences, so it provides a scientific basis for chaotic characteristics of the extraction to traffic flow time series.For the traffic flow time series prediction, the classic predictive technology does not have the capacity of adaptive and self-learning, but artificial neural networks are highly nonlinear, self-organizing, adaptive, fault-tolerant and real-time features that non-linear problems an be satisfactorily resolved by artificial neural networks in the field of TransportationIn the thesis , with the support of the Chongqing Municipal Science and Technology Commission NSFC key projects "Urban Traffic Congestion Dynamic Network of Early Warning Technology Research and Ease the Decision-making" (Ref: CSTC 2006BA6016),based on the analysis phase space reconstruction theory and the characteristics of the time series, it was chaotic feature extraction pretreatment for traffic flow time series. On this basis, the model building and predictive research of traffic flow chaotic time series is realized by neural networks, the main research results include:Based on the analysis phase space reconstruction theory and the characteristics of the time series, the chaotic features was researched and pointed out that the city's traffic flow sequence is a chaotic system, by qualitative analysis and quantitative terms.It was research that prediction of traffic flow chaotic time series based on chaotic algorithm. First, the traffic flow time series chaotic feature is extracted by chaos theory. pretreatment for traffic flow time series, and the wavelet neural networks model was build by this. Second, the chaotic mechanism and the chaotic probability is described. Based on chaotic learning algorithm, and the wavelet neural networks fast learning algorithm of traffic flow time series is designed based on chaotic algorithm. Last, a single-step and multi-step prediction of traffic flow chaotic time series is researched by BP neural networks, wavelet neural networks and wavelet neural networks based on chaotic algorithm. The results showed that the wavelet neural networks predictive performance is better than the BP networks and the wavelet neural networks by the simulation results and root-mean-square value.Based on the consistency of Volterra functional model and ANN model,the VNNTF model is proposed to make prediction of traffic flow series. With the traffic flow time series limited memory performance and phase space reconstruction mathematical significance, the method of truncation order to the traffic flow chaotic time series Volterra functional model is proposed. At the same time, the conclusion is proposed that the number of truncation terms of traffic flow Volterra model is equivalent to the length of the maximum memory of traffic flow signal and is also equivalent to the minimum phase space embedding dimension. At last, the VNNTF networks learning algorithm is designed, and a single-step and multi-step prediction of traffic flow chaotic times series is researched by VNNTF networks, the VNNTF networks predictive performance is better than the BP networks and Volterra functional model by the simulation results and root-mean-square value.By researching the modeling basis on traffic flow chaotic time series and FIR networks, the FIRTF model is proposed to make prediction of traffic flow series. It is proposed that the gray correlation analysis method to determining the number of hidden layer neurons, in the circumstances to vector signals to the FIRIF neural networks hidden layer neuron signals. Based on FIR networks learning algorithm, the FIRTF networks adaptive chaos mechanism algorithm is designed. At last, by the simulation results and root-mean-square value, the traffic FIRTF neural networks learning and prediction owned on a better advantages, because that the traffic flow FIRTF and the adaptive algorithm combination integrated gda-BP networks, gdm-BP networks, gda-feedback-BP networks and gdm-feedback-BP networks of the four characteristics.In the finally part, the summary to the research works of the full text is given, and the direction of further study is pointed out about traffic flow time series chaotic Characteristics and prediction.