Identification And Prediction Of Safety Risk States For Urban Roads Based On Ride-hailing Trajectory Data

Road traffic accidents seriously affect the efficiency of traffic operation and the safety of life and property,and there is an urgent need to carry out research on active prevention and control of road safety risks in the context of the severe traffic safety situation.The identification and prediction of safety risk states of urban roads is the focus of traffic safety research due to their high utilization rate and complex and variable traffic operation conditions,and the real-time dynamic traffic operation characteristics are closely related to the safety risk states,while the ride-hailing trajectory data,as one of the floating vehicle GPS data,can accurately reflect the real traffic operation characteristics of the road and is often used to characterize the traffic states of the road.The research on the safety risk states of each road section at each time from the overall road level is important for controlling the overall safety risk states of the road network and taking related measures.Therefore,from the perspective of active safety prevention,the safety risk states of urban roads are taken as the research object,and the traffic operation parameters related to the safety risk states are extracted from them with the support of the trajectory data of ride-hailing,and a real-time and effective safety risk states identification and prediction method is established to carry out relevant research on the safety risk states of urban road sections at each time.The main research contents are as follows:(1)The traffic operation parameters related to the safety risk states are extracted from the trajectory data of net-approved vehicles,and a missing value filling model is established to fill in the missing values in the time-series data of the parameters.A pre-processing method considering the study object and the characteristics of ride-hailing trajectory data is proposed,and the spatio-temporal units are divided to discretize the study area and the trajectory data in time and space;six traffic operation parameters,namely,average velocity,average acceleration,standard deviation of velocity,design speed consistency,average stops and imbalance rate of velocity,are extracted from each spatio-temporal unit in the mid-flow direction as safety risk evaluation indexes and calculated;an End-to-End Generative Adversarial Network(E~2GAN)model is constructed to learn the distribution of the indicator temporal data using adversarial generative training to fill in the missing values.(2)A method that can identify the safety risk states of urban roads based on high-dimensional index time-series data is proposed to identify the safety risk state of each spatio-temporal unit.An unsupervised Safety Risk State Deep Embedded Clustering Network(SRSDEC)is established to identify the safety risk states of urban roads based on high-dimensional spatio-temporal units through Deep Auto-encoder Neural Network(DAE).The feature extraction and clustering processes jointly optimize the safety risk state clustering effect through information interaction,solving the mismatch problem due to the separation of feature extraction and clustering processes,and improving the clustering effect;a safety risk states discrimination model based on Light Gradient Boosting Machine(Light GBM)algorithm is developed to discriminate the safety risk states by supervised learning.(3)An indirect prediction method of safety risk states based on traffic operation parameters prediction is proposed to predict the safety risk states in the short term future.A Bidirectional Long Short-Term Memory Network(Bi-LSTM)model is constructed to extract more comprehensive time-varying information and distribution features for short-term prediction of traffic operation parameters by simultaneous forward and backward learning;combined with the established feature extraction and safety risk states discriminant models,the corresponding safety risk states are predicted based on the predicted values of the short-term traffic operation parameters.In summary,the method is designed from three aspects:index extraction and missing data filling of traffic operation parameters related to safety risk states,safety risk states identification and safety risk states prediction,and the research on safety risk states of urban roads is carried out to obtain safety risk states of different road sections and time periods of urban roads in a timely and accurate manner.The research results can provide a basis for traffic management departments to formulate control and guidance measures and driver warning,which is important to reduce the safety risk level and accident probability.