The Research Of "mechanism + Identification" Strategy In Short-term Highway Traffic Flow Based On Robust Statistics

The short-term traffic flow forecasting is the one of the key technologies of traffic control system and traffic flow guidance system, which are subsystems of the intelligent transport system. Traffic flow is a typical non-linear time series, and it is required to find effective and robust nonlinear forecasting method. At present, combination forecasting model is the main strategy of the traffic flow forecasting, and the"mechanism + identification"forecasting strategy as a new forecasting strategy is a refinement and improvement of the combination forecasting strategy. The in-depth and comprehensive research of the combination forecasting model is the basis of improvement and development of the "mechanism + identification" forecasting strategy. These two points are the focus of this paper.The main contents and results of this paper are:(1) The"mechanism + identification"forecasting strategy is applied to the short-term traffic flow forecasting, the concept of the low-level non-linear combination forecasting is put forward, and its reliability and robustness are preliminarily analyzed and proved with the knowledge of time series analysis and mathematical statistics. Moreover, the idea that the signal noise ratio is enhanced with nonlinear transformations is further perfected, and four basic forms of the nonlinear transformation are given to improve the traffic flow forecasting accuracy. This paper not only gives the mathematical proof of these four non-linear transformations, but also confirms the effectiveness of these transformations to improve signal noise ratio from the perspective of the wavelet transform and power spectral analysis, finally, the results are verified in the Matlab simulation experiments.(2) The bootstrap method is introduced into the combination forecasting model including five models, and improves effectively the forecasting results; and for combination forecasting model respectively including seven and twelve forecasting models, effects of the combination forecasting are studied with different training samples, and are compared with the simple average method, the variance derivative method and the Dickinson method, moreover, changes of the forecasting results of different forecasting models and combination forecasting methods are given theoretical explanation; A direct experimental basis is provided for the improvement of Dickinson's most excellent combined weight coefficients.(3) The effect of sample size on forecasting accuracy with the general and adaptive control theory dynamic forecasting methods for the short-term traffic are studied, and they are added to the further combination forecasting model.