Research For Short-term Traffic Flow Forecasting Methoed Based On Chaos Theory

The short-term traffic flow prediction often used in the dynamic traffic guidance, the advanced traffic management and the traffic control and safety, plays an important role in the research area of traffic engineering, as well as is one of the core research issues in Intelligent Transportation Systems (ITS).A traffic system by instinct is a complex system emerging from the interactions among people, vehicles and roadway segments, is an open, far-from-equilibrium, dynamical system in which the nonlinearity of interactions and the irreversibility of processes can be observed. To deal with this kind of systems in which the time sequences of a system state only can be obtained by a certain observer, the technology is needed, named chaotic-time-series reconstruction which reconstructs global system behaviors by one-dimensional mappings of them and analyzes essential characteristics of the system.By analyzing elementary properties of traffic flow time sequences, this thesis focuses on the research of the State Space Reconstruction of chaotic time series dynamic systems, and determines parameters of the reconstructed state space. Aimed at the shortcoming of the local-region method when confirming neighbor Phase Points, the improved method is bring forward, and the error correction method is put forward to make full use of the one-step forecasting result. Based on these study, the Improved Adding-weighted One-rank Local-region Prediction model (IAOL) is established. Moreover, Combined with Neural Network which have parallel processing and strong nonlinear mapping capability, the Chaos Local Region-RBF Neural Network compositional prediction model (CLR-RBFNN) is established. At last, considering Multiresolution analysis characteristic of Wavele, the Wave-Chaos Local Region -RBF Neural Network compositional prediction model (W-CLR-RBFNN) is set up. The W-CLR-RBFNN model aims at the analysis of elementary characteristics of traffic flow from the perspective of time-frequency on different scales by considering the chaotic character of short term traffic flow, thus it is more applicable to forecasting of time series which have non-linear, nonstationary and chaotic characteristics. The results of the experiments on actual traffic flow data from the Beijing traffic system show that the W-CLR-RBFNN model is more effective than the other models.