Multi-source Data-driven Real-time Risk Prediction For Expressway Accidents

The urban expressway is an important part of the urban transportation system.It is characterized by large traffic capacity and saving travel time,which provides a lot of convenience for the long-distance express transportation in the city.With the increase of the number of motor vehicles in the city,the traffic load of the urban expressway is also increasing,which makes the traffic accidents occur frequently.And these traffic accidents will lead to traffic congestion,and a lot of property damage and casualties.This not only affects people’s life and property safety,but also causes serious environmental pollution and energy consumption.Therefore,the traffic safety level of urban expressways needs to be improved urgently.The real-time traffic flow status is the most direct factor leading to the occurrence of traffic accidents,and with the development of intelligent transportation systems,massive and multi-source high-precision traffic flow data has been generated,which provides a good data foundation for the real-time risk prediction of urban expressways.In order to improve the traffic safety of urban expressways,this paper constructs an accident risk prediction model for urban expressways based on the historical traffic accident data of urban expressways and real-time traffic flow data obtained by multi-source detectors.The specific research content of this paper includes the following aspects.The first is the extraction and fusion of real-time traffic flow parameters based on multisource data.By preprocessing the data of the motion detector and the cross-section detector,the matching real-time traffic flow data within the time and space of the accident is extracted,and the corresponding traffic flow parameters are calculated.However,different types of detectors have their own advantages and disadvantages.The cross-section detector can obtain the traffic flow information near the detector position and close to the full sample,but cannot accurately capture the running status of the entire road section.The detector can track traffic flow information for the entire road segment,but it cannot guarantee a sufficient sample of vehicles.In order to comprehensively utilize two different data sources,this paper adopts the DS evidence theory method to integrate the traffic flow characteristics of the two data sources to obtain accurate traffic flow information and provide a data basis for subsequent accident risk modeling research.The second is construction of a real-time prediction model for urban expressway traffic accident risk.Traffic flow data in three spatial ranges are extracted based on fusion data and cross-sectional detector data.Considering multiple real-time traffic flow characteristics comprehensively,a real-time prediction model of urban expressway accident risk is constructed by using Balanced Bagging sample resampling technology and XGBoost integrated machine learning algorithm.The study found that compared with a single data source,the fusion data,the urban expressway accident risk model has higher prediction accuracy,and can better fit the nonlinear complex relationship between real-time traffic flow characteristics and traffic accident risk.Therefore,the data fusion method can be applied to the field of traffic safety to improve the accuracy of the accident prediction model.At the same time,this study found that the model of the traffic flow parameters in the spatial range of the four upstream and downstream sections performed the best.The experimental conclusions can provide a theoretical reference for establishing an accident risk prediction model based on real-time traffic flow characteristics in the future.Finally,analysis of the correlation between real-time traffic flow characteristics and urban expressway accident risk.Based on the optimal real-time prediction model of urban expressway accident risk,feature importance analysis method and partial dependence graph are used to analyze the importance of features to the model and the correlation between real-time traffic flow characteristics and accident risk.The study found that the four characteristics of upstream speed dispersion,occupancy difference,speed dispersion difference,and downstream speed have a greater impact on accident risk,and the influence of some real-time traffic flow characteristics on accident risk is given.The research conclusions are of great significance for the management and control of urban expressways and improving the safety level of expressways.This paper can help traffic management departments to give early warning of traffic accident risks in a targeted manner and make timely traffic management scheduling.At the same time,it deepens the understanding of the correlation between real-time traffic flow characteristics and accident risk,and provides an effective basis for traffic safety management and prevention,and improves the level of urban road traffic safety.