Urban Road Short-term Traffic Flow Forecasting Based On Spatio-temporal Characteristic

Traffic Signal Control System and Traffic Flow Guidance System is a core subject of Intelligent Transportation System(ITS), the accurate, real-time traffic flow forecasting information is essential to the function of Traffic Signal Control System and Traffic Flow Guidance System. Traffic flow forecasting information can be sent to Advanced Travellers Information System(ATIS) and Advanced Traffic control system(ATMS) directly, to provide effective real-time travel information for travelers and improve the level of urban traffic control.Traffic flow forecasting has been all long the hot topic in various academic fields. Existing short-term traffic flow prediction mainly aim at traffic flow time series data of single road cross section based on statistical model, nonlinear theory, intelligent algorithm, or combination forecasting model with reference to several kinds of theory. It exists the problem of low prediction accuracy and difficult to adapt to the nonlinear and uncertainty that short-term traffic flow change showed. In allusion to these problems, this paper use urban road traffic flow spatio-temporal characteristic as the starting point, deeply analyzes the temporal characteristic and spatial characteristic of urban road traffic flow, take the influence of adjacent road traffic flow into consideration when short-term traffic flow forecast, forecasting method based on time series data has good stability characteristics and prediction method based on estimation of spatial-correlation traffic flow data has a good adaptability to nonlinearity, combine with two superiority, a new method that consider research urban road section real-time, historical traffic flow data and spatial-correlation road section traffic flow data at the same time is proposed, and the proposed method improves the short-term traffic flow prediction accuracy.Firstly, analyzed the temporal characteristic of urban traffic flow, dynamic analysis method was applied to short-term traffic flow temporal characteristic, concretely analyzed traffic flow characteristic and short-term predictability discriminated method. Refer to three types of short-term traffic flow prediction method based on different theories, gives the method of mathematical model and algorithm steps, respectively forecast to 2min, 6min and 12 min three statistical cycle traffic flow through the experiment case, compare forecast results evaluation index, analyzed the applicability of each forecast model.Secondly, analyzes the characteristics of urban road traffic flow spatial distribution and mutual-correlation, by using no detector road traffic parameter estimation, analyzes the feasibility of target road traffic flow forecasting by use space adjacent detection data. A prediction method was proposed using multivariate step linear regression to estimate the target road traffic flow. Considering the time-varying of traffic system, combining with optimal estimation ideas of Kalman filtering theory, a spatial prediction model of short-time traffic flow is presented based on Kalman filtering theory. Experiment simulation proved the effectiveness of the proposed method.Finally, in the urban road traffic flow analysis based on the spatial and temporal correlation properties, proposes two at the same time considering the road traffic flow forecast time series data and spatial adjacent and the prediction method of road traffic flow data, is a kind of dynamic weighting fusion algorithm using least squares, will be based on time series data output forecasting method and prediction method based on spatial correlation traffic data fusion output, get the final result; Another kind is the Kalman filter to optimize K-neighbor algorithm is applied to short-term traffic flow prediction, initial with the multiple linear regression to the definition of state variable weights, using the Kalman filter recursive optimal estimation principle, according to the section of data real-time acquisition of state variable weights updated online, so as to adapt to the change of traffic status. Experimental cases, compared to a single based on forecasting method of time series data and spatial data, demonstrate the superiority of the proposed method.