Identification And Prediction Of Short-Time Traffic Conditions On Urban Roads Based On Trajectory Data Of Ride-Hailing

There is an increasingly serious congestion nowadays,with the promotion of big data and smart transportation,accurate perception and prediction of traffic congestion will largely help travelers to obtain timely congestion information and make corresponding strategies.As a mutual product of fast-developing smart terminal technology and growing traffic demand,RideHailing service provides great convenience to urban residents.The massive data brought by equipped positioning systems can also help to explore the trip characteristics of residents and traffic flow operation conditions,which can help to alleviate urban traffic congestion.Therefore,based on mining the spatial and temporal characteristics of Ride-Hailing trip,this study aims to construct an evaluation and perception model for automatic identification of urban road traffic conditions,and to capture and predict urban traffic short-time fluctuation conditions with neural networks,so as to achieve real-time perception and early warning of urban traffic conditions.In this study,the GPS data of Ride-Hailing in Chengdu city for a period of 28 days in 2018 are employed to extract the travel trajectories of Ride-Hailing and to analyze the spatial and temporal characteristics of Ride-Hailing travel on this basis.For temporal features,the travel frequency at macro level and the average speed of roads at micro level are dominant;for spatial features,the spatial correlation among adjacent roads is profiled by focusing on travel hotspot areas of road network,and the spatial occupancy of selected roads is further analyzed.Based on the spatial and temporal dependence of traffic conditions,this study constructs a multidimensional urban road traffic operation evaluation index.The speed performance gradient index(SPG)is proposed creatively,and combines with delay time ratio(DTR)and road section saturation(V/C),which describes the traffic conditions of urban roads from the perspective of speed,time and space,respectively.In addition,this study improves the DBSCAN algorithm(Density-Based Spatial Clustering of Applications with Noise)and then establishes the DBSCAN algorithm based on partition adaptive learning for clustering the multidimensional traffic operation evaluation indexes,in order to update the traffic operation evaluation criteria in a more scientific and reasonable way.Finally,to address the volatility of traffic flow in the short term,the ATT-TCN(Attention Mechanism-Temporal Convolutional Network)is used to predict the traffic evaluation indexes with spatial and temporal information in the short term.The prediction results are compared with the ATT-TCN without spatial awareness information.The prediction results are compared with the ATT-TCN neural network without incorporating spatial perception information,the long-and short-term memory neural network,and the gated recurrent unit neural network to verify the accuracy and sensitivity of this model to data fluctuation.Based on inputting the prediction values of evaluation indexes into the traffic condition evaluation criteria for rank determination,a complete closed loop of traffic condition identification and prediction is formed.It is found that the trajectory characteristics of Ride-Hailing exhibit obvious short-time volatility and spatial correlation,which puts forward more serious requirements for the subsequent traffic flow prediction.The DBSCAN algorithm based on partitioned adaptive learning established in this study achieves automated sensing and evaluation of urban road traffic conditions,and divides urban road traffic conditions into six classes,which means that the improved DBSCAN algorithm improves the contour coefficient by 0.07 compared with the K-means++ algorithm.When using ATT-TCN model for short-time prediction of urban road traffic condition,the average absolute error of prediction of the model with spatial information for the three evaluation indexes is reduced by 0.4,0.034,and 0.01,respectively,compared with this model without spatial information.The comparison test of the current mainstream neural network models also proves that the average absolute error of the prediction of SPG and V/C of ATT-TCN model is reduced by 0.05 and 0.034,respectively,compared with the long and short term memory neural network model,and by 0.046 and 0.038,respectively,compared with the gated circulation unit neural network model.Due to the sensitivity of the ATT-TCN model to data fluctuations,the average absolute error of the prediction of DTR by the ATT-TCN algorithm in the peak period is 0.078 lower than that of the neural network model of the long and short term memory.The prediction accuracy of traffic condition prediction and judgment model proposed in this study reaches 91.18%.