Research On Urban Traffic Flow Forecasting Based On LS-WSVM

The real-time and accurate urban traffic flow forecasting is one of the key technology of intelligent transportation systems (ITS), which provide favorable data support for the urban traffic guidance system, and also provide the important technical support for the traffic accident detection and other subsystems of the ITS.The analysis of the inherent characteristics and predictability of urban traffic flow, studying the dynamics system of traffic flow, the discrimination of chaotic characteristic the according to the largest Lyapunov index, but also to use this index to identify the time series predictability. To the one-dimensional time series reconstruction to multi-dimensional space, which will hide the internal factors in the multidimensional space revealed, the study for the time series of phase space reconstruction, and calculate the reconstruction parameters by C-C method. The simulated experiments of the actual traffic flow time series, verified the predictability of the traffic flow time series by the calculation results.The thesis studied the principle and applicability of the standard LS-SVM, and analyzed the characteristics of several commonly used kernel functions. According to the shortcomings, the article introduced the wavelet theory, constructed the Morlet wavelet kernel function which meets the Mercer theorem, build the traffic flow forecasting model based on the least squares wavelet support vector machine (LS-WSVM), and optimized the model parameters by the grid search and the improved grid search. According to the simulated experiments of the actual traffic flow time series, the article analyzed the advantages and disadvantages, and further improved and constructed LS-WSVM model based on hybrid wavelet kernel function.For comparison, the article constructed the standard LS-SVM prediction model and BP neural network model, did the simulated experiments of the actual traffic flow time series, and analyzed predictors, the results showed that the LS-WSVM model of the mixed kernel function effectively improved the accuracy of the traffic flow forecasting and its generalization ability, was more suitable for the real-time online traffic flow forecasting.