Basic Problems In "Mechanism Model+Identification Model" Strategy In Traffic Flow Forecasting

Traffic flow forecasting with high accuracy is the premises and key in traffic control and traffic abduction in Intelligent Transportation System (ITS). Some recent studies just use the data to be forecasted as"only"data, which take forecasting as a pure"data to data"mathematical process. However, the chief way to improve the accuracy of forecasting must be the study on strategies of forecasting.As an improvement of combined forecasting strategy, the basic problems of the"mechanism model + identification model"forecasting strategy which can comprehensively consider the structure information and historical behavior of complex systems applying in short-term traffic flow forecasting were studied in this dissertation.The main contents and results are:(1) Applied"mechanism model + identification model"to short-term traffic flow forecasting, with the discussions of the basic structure of"mechanism model + identification model". Proved that Lyapunov exponent of differential system of real nonlinear dynamical system keeps the same.(2) Considering the influence on forecasting accuracy by the length of the time series to be forecasted from the viewpoint of statistics, used the methods of ANN, AR etc. to do numerical test. The theory and results of the experiments showed that as the length of the series grows, the accuracy of forecasting grows as that the standard deviation estimation of confidence interval grows with the increase of number of the samples according toχ2 distribution. (3) Transplanted the ensemble method in numerical weather forecasting into short-term traffic flow forecasting, and tested it by ANN and AR respectively to get the result. The paper also attempted to make experiment and further discuss on the influence of noise intensity and times of ensemble on forecasting's accuracy. The numerical test suggests that the ensemble method improve the accuracy of short-term traffic flow forecasting.The research suggests that using"mechanism model + identification model"strategy and the strategy of behavior's forecasting after multi-models estimations advocated by DDDAS to forecast short-term traffic flow is a beneficial new discussion of forecasting strategy. These methods can obviously improve the credibility and the accuracy of short-term traffic flow forecasting.