Real-time Risk Prediction And Spatiotemporal Impact Analysis For Freeway Accident

Pro-active traffic safety management is a hot research topic in the field of traffic safety.It is of great significance to effectively prevent the occurrence of traffic accidents and effectively reduce the cost caused by accidents.Real-time risk prediction and spatiotemporal impact analysis of freeway accident are two important components in the active traffic safety control system,providing a straightforward and effective decision-making process for determining strategies of active traffic safety management,and gaining a lot of attention from world-wide traffic researchers and experts.So far,there have been plenty of related studies in this field.It is worth to mention that traffic accident is a comprehensive result influenced and interacted by human,vehicle,road as well as environment,leading to a great challenge in modeling and analyzing it effectively and accurately.Considering the differences of continuous traffic flow at home and abroad and the ease of data acquisition,in this paper,the accident data,the traffic flow data as well as the corresponding weather data,was collected from the 1-5 freeway in UA,and analyzed and used to build the corresponding models.The most important variables were considered and selected by utilizing a data-driven method,and the real-time accident risk prediction theory and method were mainly studied.On this basis,an effective accident impact analysis model was proposed to quantify the effect caused by accidents in time and space dimension.The main contributions of this paper are summarized as follows:First,a real-time accident risk prediction method based on Random Forest algorithm combined with a data-driven threshold selection strategy was proposed.Through training a validation model on a separate validation data set,a proper threshold was determined by comprehensively considering accident recognition rate,non-accident recognition rate,overall recognition rate and accident precision rate.Experimental results show that the proposed method based on random forest algorithm is superior to the typical logistic regression method while the comprehensive performance was improved by employing the automatic threshold selection strategy.Second,an accident impact analysis model was put forward to overcome the disadvantage of existing methods,i.e.,some hypothesises are often needed in the process of modeling which are not necessarily true in reality.The speed variation coefficient was selected as the analysis indictor,which was employed to construct a spatiotemporal contour plot using a linear interpolation technique.On this basis,the local polynomial fitting algorithm was used to extract the contour of the accident impact.Then four quantitative indictors are calculated,including the maximum impact time range,the maximum impact space range,the impact intensity and the shock wave propagation speed.Results show the proposed method can effectively conduct impact analysis and the quantitative indicators can be quantified accurately and timely.