| Using the road traffic supply and demand relationship to achieve refined urban traffic management is the main content of intelligent transportation,and the dynamic OD estimation of urban road network is an important support on the traffic demand side.Existing theories and methods of dynamic traffic assignment model is difficult to accurately describe the real road network vehicle travel path selection law,and the OD estimation model tuning problem is time-consuming and laborious.A wide range of vehicle detectors are provided for intelligent traffic to provide a variety of traffic big data.The vehicle trajectory data represented by the license plate data collected by the HD intersection vehicle detector can provide the vehicle's travel path,and provide the data for estimating road network dynamic OD.Artificial intelligence technology represented by deep learning provides a new solution to traffic science problems.This study is based on the vehicle travel chain obtained from the vehicle license plate data,and uses the supervised learning method in deep learning as the means.The main contents include three aspects: vehicle travel chain division,vehicle trajectory reconstruction,and road network dynamic OD estimation.The specific research contents are summarized as follows.Obtaining a travel route for a single trip of the vehicle is a basic step of subsequent dynamic OD estimation.It is easy to extract the travel chains of all vehicles from the license plate data on the same day,but the vehicle travel chain usually includes multiple trips,so the vehicle travel chain needs to be divided.Considering the time-varying characteristics of the travel time of the road segment,based on the microwave data and the length attribute of the road segment,using the search idea of dijkstra algorithm,a short-circuit solving method for the traffic community suitable for urban road network is proposed.The traffic obtained by this method is obtained.The shortest travel time of the specific time period of the cell is compared with the actual travel time recorded by the license plate data,thereby completing the division of the vehicle travel chain.The purpose of incomplete vehicle trajectory reconstruction in this paper is to provide more accurate and complete road observable traffic flow for dynamic OD estimation.Firstly,according to the integrity of the vehicle trajectory,the vehicle travel trajectory obtained by the vehicle travel chain is divided into three categories: complete vehicle trajectory,incomplete vehicle trajectory and single recorded travel trajectory.Considering the possible missing situation of the vehicle trajectory,design for lstm neural networks and neural network input and output data frame,capture trajectory characteristics of the vehicles in the road network;the accuracy of the trained model is evaluated by the test set,and the evaluation results show that the reconstruction method proposed by the paper meets the actual needs;based on the LSTM neural network vehicle trajectory model,this paper proposes a reconstruction method of incomplete vehicle trajectory,which uses this method to complete the incomplete vehicle trajectory obtained from the license plate data.Combining CNN and LSTM neural networks in supervised learning,construct a dynamic OD allocation model with both spatiotemporal characteristics,and quickly estimate dynamic OD with trained distribution model based on evolutionary strategy.Firstly,construct the CNN+LSTM model framework for dynamic OD allocation problem,capture the dynamic OD of the road network and the observable inlet flow,and observe the spatio-temporal relationship of the traffic at the intersection.Secondly,a dynamic OD estimation method of the evolutionary strategy thoughts based on the trained distribution model comes up;finally,the traffic distribution performance and dynamic OD estimation performance of the model are evaluated respectively.The results show that the proposed method can estimate the dynamic OD of urban road network quickly and accurately. |
PDF Full Text Request