Research On Short-term Traffic Flow Forecasting Based On Kalman Filter And Wavelet Neural Network

With the development of urbanization and the improvement of people's living standards, the amount of motor vehicles is growing day by day, which has led to the enormous challenges to urban transportation. Under the existing road conditions, it is one of effective methods to solve the traffic problems by improvement of traffic control management, reasonable use of existing transportation facilities and giving full play to their ability. One critical problem of the intelligent transportation system (ITS) is whether or not to predict traffic flow of urban roads accurately. Therefore, short-term traffic flow prediction theory has important practical significance.The short-term traffic flow prediction theory is the way to solve the prominent contradiction of traffic problem under the present conditions, and the important theory method to control traffic problem effectively. The traditional methods are the sum of autoregressive moving average model, nonparametric regression models and so on. At present, the new research methods mainly are support vector machines, chaos prediction models, artificial neural network model and the combination forecasting based on integration of variety model features etc. However, many different theoretical models have their own features and deficiencies. For example, non-parametric regression method's structure is relatively simple, but the method is limited by the practical application case in the event of a large amount of data, because of basing on the complexity of their neighbors'search. Traditional neural network training is slow, and sometimes the situation does not converge. So, it is need to find a more reasonable prediction model to adapt to the growing emergence of a new road traffic conditions change.Kalman filter theory and the combination of wavelet neural network theory are applied in prediction of short-term traffic flow, filtering to data streams through theoretical model of Kalman, to get rid of white noise exists in data information, predicting by wavelet neural network, the effective integration of the two traffic flow Prediction theory is a useful attempt. Theoretical model of Kalman filtering effect on the data stream and wavelet neural network prediction of time-frequency characteristics make the system to achieve a good prediction. Therefore, the paper has taken Kalman filter theory and neural network theory into the field of short-term prediction for traffic flow.The main work of this paper is as follows:Firstly, the short-term traffic flow of urban roads has been filtered by Kalman filter theory to form good training samples. Secondly, the function option of mother wavelet, parameter setting, modeling, algorithm design and so on in prediction model of traffic flow based on wavelet neural network have been studied in detail. Meanwhile, improving and optimizing algorithm in wavelet neural network model which is prone to fall into local minima or convergence problems, to improve the performance of the model prediction, and predicting simulation results by training.Finally, after practical measure on status of traffic flow in the peak hours at the crossing of the urban roads, the training and simulating on data samples of traffic flow have been done by combining Kalman filter model and wavelet neural network model. According to the simulating analysis and comparison, the predictive performance of the neural network model based on Kalman filter and wavelet, BP neural network model and RBF neural network model has been compared all around. The results show that the models introduced in the paper have accuracy, feasibility and high precision in the traffic flow forecasting.