Research On Travel Time Prediction And Reliability Evaluation Of Urban Road Based On Online Car-hailing Trajectory Data

In recent years,the disposable income of Chinese residents has gradually increased,and the traffic demand of residents has been increasing.Although the construction of urban roads is still being strengthened,the traffic congestion problem has always caused troubles to residents.In this context,intelligent transportation system as a new means of rapid development and has been widely used in improving the efficiency of road traffic,traffic congestion and other aspects play a pivotal role.Travel time is one of the most important travel indicators for residents.Timely and accurate prediction results of travel time can not only provide travel decision-making information for travelers,but also contribute to the management and development of intelligent transportation system.Based on the spatial-temporal correlation analysis results of trip time data,this paper builds a prediction model of urban road trip time based on deep learning combination method,and evaluates the reliability of predicted trip time values based on historical data sets.First of all,this paper draws a conclusion about the research status of map matching,travel time prediction and travel time reliability assessment,and ensures the research path of this paper.According to the characteristics of the original data set used and the data needs of this study,the pre-processing process of trajectory data in this paper is given.The temporal and spatial scope of the case is determined by analyzing the order quantity and the starting and ending position of the order.Secondly,on the basis of data preprocessing,this paper expresses the idea of travel time and provides the computing way of travel time data.Pearson correlation coefficient expresses the correlation of travel time data from the perspectives of spatial-temporal things,and the analysis conclusion can provide theoretical and data support for the prediction model and input matrix construction in the following paper.Then,a prediction model(AC-BILSTM-GRU)combining convolutional neural network,bidirectional long and short-term memory network,gated cyclic unit and attention mechanism is proposed.In this combined model,CNN is used to obtain the spatial characteristics of travel time data,Bi LSTM and GRU are used to obtain the time characteristics of travel time data,and the attention mechanism is used to improve the feature capture ability of the model.A case study on the trajectory data of xi ’an online car hailing shows that the accuracy and efficiency of this model are better than other comparison models.Finally,the reliability of travel time is evaluated based on parametric method.Normal distribution,lognormal distribution,Weibull distribution,gamma distribution,three-parameter Bohr distribution and multivariable Gaussian mixture model were used to fit the travel time sample data of different dates and different periods.The model that passed the K-S test and had the lowest AIC value was the best fitting model for each case.The reliability of the predicted trip time of the three cases was analyzed by using five reliability evaluation indexes: average trip time,historical arrival rate,buffer time,buffer index and planned time index,so as to provide travelers with auxiliary decision-making information.There are 54 pictures,24 tables and 78 references.