Chaos Theory And Data Fusion Based Short-term Prediction For Traffic Flow

Nowadays, traffic problems, i.e., traffic congestion, traffic safety, and traffic emission, have drawn much attention from both academic and industrial circles. To address this issue, there is a research need to study the real-time and accurate prediction of traffic flow as it can leverage the traffic guidance and control. However, the accuracy of existing methods for traffic flow prediction is restricted due to the following two aspects: on one hand, as many traffic parameters(such as speed, occupancy and flow) can characterize the traffic system, hence, single parameter based prediction method cannot fully capture the state evolution of traffic flow; On the other hand, due to the nonlinearity and complexity of traffic flow, linear based prediction method cannot guarantee the prediction error to converge to an expected range. Hence, considering the above-mentioned issues, it motivates us to study the prediction of short-time traffic flow based on the chaos theory and multi-source data fusion.The focus of this thesis is on the traffic flow prediction based on the chaos theory and multi-source data fusion. In particular, the chaotic characteristics of traffic system is firstly identified using the the maximum Lyapunov exponent. Then, phase space of multiple parametric time series is reconstructed based on the phase space reconstruction theory. In addition, multiple parametric time series in the reconstructed higher dimension space are fused according to the Bayesian theory. Also, the prediction of short-time traffic flow is performed using support vector regression(SVR) and radial basis function neural networks(RBFNN), respectively. Finally, field test data are used to verify the effectiveness of the proposed methods.The main work includes:1. Chaotic characteristics identification of traffic flow based on the maximum Lyapunov exponentConsidering the chaotic characteristics of traffic system, it enables to capture the characteristics of traffic flow using the chaos theory. To this end, the maximum Lyapunov exponents of multiple measures, such as average speed, average occupancy and average traffic flow, are first calculated using the small-data method so as to identify the chaotic characteristics. Results demonstrate that the three groups of 5 min statistical measure of traffic data, including average speed, average occupancy and average traffic flow, are chaotic to a certain extent.2. Phase space reconstruction of multi-parameter time series based on the phase space reconstruction theorySince traffic flow is chaotic in Section 1, the phase space reconstruction theory is used to capture the strange attractors in a high-dimension reconstruction space. Consequently, the change of the corresponding time series will eventually tend a specific trajectory in that high-dimentional phase space. To this end, the values of the parameters including the embedding delay ? and dimension m are calculated using the C-C algorithm and the G-P algorithm, respectively. Subsequently, the phase space of multi-parameter time series is reconstructed based on the phase space reconstruction theory. Simulation result shows that the chaotic attractors of different time series have the characteristics of both similarity and uniqueness in that high-dimentional phase space.3. Phase point fusion of multi-parameter time series in high-dimentional phase space based on the Bayesian estimation theoryAccording to Takens' embedding theorem, it is feasible in theory to reconstruct the phase space of the original system based on a single parameter time series. However, as the chaotic attractors of different time series of the same system show different features in a high-dimentional space, and the information of multi-parameter time series is richer than that of a single parameter time series. Therefore, the phase points are fused in terms of the redundancy and complementarity of multi-parameter time series using the Bayesian estimation theory. Simulation results show that the fusion can produce a new fusion phase space that includes all the main features of multiple measures within it.4. The short-term traffic flow prediction based on multi-source data fusionBased on the Sections 1, 2, and 3, a multi-parameter fusion based model is proposed to predict the short-term traffic flow. In particular,(1) For single parameter, support vector regression(SVR) and radial basis function neural network(RBFNN) based prediction models are analyzed, respectively.(2) For multiple parameters, multi-parameter fusion based SVR and RBF neural network prediction models are proposed using the phase space reconstruction theory and Bayesian estimation theory. After theoretic analyses, field traffic data is used to verify the effectiveness of the proposed models.(3) Considering the performance of the proposed predicition models, the multiple parameters based short-term predicition for traffic flow is compared to the single parameter based prediction, and the multi-parameter fusion based SVR prediction model is compared to the multi-parameter fusion based RBF predicition model. Simulation results show that the performance of the proposed multi-parameter based prediction model is improved in terms of the accuracy compared with the single parameter based prediction model, and the performance of the multi-parameter fusion based RBF prediction model is superior to the multi-parameter fusion based SVR prediction model.