Several Issues In Short-term Traffic Flow Forecasting

Intelligent transportation systems(ITS) is a total definition of new road traffic system which includes people, road and vehicle. The system uses the most advanced electronical and corresponding technology. Traffic control and management system is the most important research area in ITS, and the traffic flow forecasting problem is the core issue of the traffic control and management. Therefore, how to predict realtime and accurate traffic flow information is the key point which can decide the results of traffic control and management. This paper mainly discusses two kinds of short-term traffic flow forecasting methods which are Kalman filter method based on the model and neural network method based on the data. The basic ideas of the papers are as follows: based on the historical data, building a integrated forecasting model under the detection data missing to improve forecasting accuracy; using the neural network nonlinear mapping ability, and using historical data to train neural network, then get parameters needed in Kalman filter to expand the scope of application of Kalman filter.The main works and contributions in the dissertation are as follows.(1) Discusses the traffic issues and short-term traffic flow forecasting theory, and compares the current forecasting methods.(2) Studies on the series of methods of Kalman filter which can be used in short- term traffic flow forecasting, including Kalman filter, Kalman smoother, extended Kalman filter, short-term traffic flow forecasting based on Kalman filter and the macro traffic flow models applied to the extended Kalman filter. Afterwards proposes a integrated forecasting model which can solve the detection data missing problem. The model has two parts, one is rebuilding section, and another is forecasting section. In rebuilding section, we use the historical data to create the historical trend and then use the trend to fix missing data; in forecasting section, we retrofit Kalman smoother to improve the real-time forecasting ability, and the experience shows the model enhances the accuracy of forecasting.(3) Studies on the short-term traffic flow forecasting method based on the neural network, discusses the structures and models of neural network. After that, we propose a new short-term traffic flow forecasting model that combines generalized regression neural network(GRNN) and Kalman filter, the model can make use of GRNN's good abilities of non-linear mapping, and the parameters can be obtained through the neural network's training. The model does not need a accurate mathematic model which is always not easy to be obtained. Then, the experience shows that the new method expands the scope of application of Kalman filter and also has good results.(4) According to above theory researches, the algorithms presented in this article are used in Shenzhen intelligent traffic control and manement system. We design and implement the traffic information processing subsystem which includes short- term traffic flow forecasting module, that enhances the effectiveness of traffic control and management.