Research On Forecasting Method Of Road Traffic Flow Under Emergency

With the rapid development of economy, traffic demand is also increasing. Traffic congestion, environment pollution and safety problem caused by unexpected events have achieved wide attention of traffic management department and related researchers. Traffic under sudden events has the characteristics of sudden change, episodic and nonlinear. The complexity and correlation between the internal mechanism and the functions have not gotten enough attention.It is not only helpful for traffic management, but also useful for people’s travel routes and time to forecast traffic flow in sudden event section exactly. There are many forecasting methods, just like ARIMA prediction based on time series data, Kalman filter prediction method based on space state data and support vector machine forecasting algorithm based on temporal data. All these methods are so complex and need model parameters setting for several times. Besides, the impacts of events on traffic prediction are not considered.So this paper analyzed the correlation between temporal and spatial of the detected traffic data on adjacent road sections under the condition of traffic events based on the characteristics of temporal and spatial. This paper proposed a kind of random forest prediction model considering event factors and effectiveness of this method is verified by the practical data. The research results of this paper are summarized as follows:(1) A traffic incident acquisition method based on 122 alarm information is proposed.This paper established spatial location information database of Beijing place name. Based on this database, this paper extracted corresponding layers and the corresponding regions as a place name dictionary by "Hierarchical sub block". In the basis of location information of traffic events, the traffic flow data collected by detectors is identified and the specific state of the data is known.(2) The collected traffic data is divided into the traffic data under traffic events and nonevent traffic data by location information retrieval technology. After analyzing traffic speed changing data, adjacent detector data under emergency condition is found to have a strong temporal and spatial correlation.(3) Traffic predictions considering emergency factors have been done through the random forest method, ARIMA and Kalman filter algorithm.(4) This paper proposed a spatial-temporal data fusion prediction method in order to improve the prediction accuracy. The prediction results of time series data prediction method and spatial sequence prediction method are fused by the least square method in this algorithm and a new result is shown. At last, the proposed method is evaluated by the mean absolute error percentage (MAP), the mean absolute error (MA) and the mean square error (MS).