Research On OD Estimation Method Of Urban Road Network Based On Encoder-Decoder Framework

Online and fast estimation of dynamic flow OD for highway networks is critical to active traffic management that uses the relationships between highway supplies and traffic demands.For urban road networks,existing theories and methods are still constrained in that the dynamic traffic assignment models are incapable to accurately describe the rules of vehicle route choices for real road networks,the optimization process of finding OD estimation model solutions is time-consuming and grueling,and that the spatiotemporal evolution rules of highway network traffic demands are still superficially studied.Under the new circumstances with the rapid development of AI technologies represented by deep learning,this study utilizes the new identifiable vehicle location data.Starting with the accurate trajectories reconstruction for identifiable vehicles,a dynamic traffic assignment model is constructed,in which the incremental updating mechanism is considered,so that the mapping relationships between flow ODs and the station volumes are accurately depicted.Then,an online OD estimation model is constructed using the mapping relationships,and a heuristic genetic search optimization algorithm is used to fast find the model solutions.Finally,the spatiotemporal evolution rules of road network traffic demands are explored based on the results of dynamic traffic assignment and OD estimation,through the analysis of uncertainty levels.Firstly,the paper adopt Encoder-Decoder framework to carry on dynamic traffic assignment baesd on identifiable vehicle location data.The research mainly includes three parts: building Encoder and Decoder framework;determining the parameters of the neural network;determining the form of input and output.After the DTA,the paper choose GEH to evaluate the neural network,and the result indicate that most of the road fulfill the precision requirements.Then,the heuristic algorithm baesd OD estimation method is put forward based on the DTA machine which was researched above.Fist of all,the paper pick up the historic OD pattern and OD fluctuation feature as the initial condition and Iteration scope of the heuristic algorithm.Next,the OD estmation method is modelled by “Multigroup parallel Evolutionary Strategies+(μ+λ)＿ES” algorithm and MAPE is used to evaluate the model.Then,the paper optimize the model from two aspects: historic OD pattern optimization and Variation intensity optimization.As a result,the precision of the model is promoted to a certain degree.Furthermore,the paper design a increment renew strategy to relieve the “Catastrophic Forgetting” of DTA model.Finally,the paper employ Gaussian Mixture Model to modelling the OD travel time and OD demand based on the OD data which is output by OD estimation model。And OD travel time reliability is analysed by “Mean” and “Buffer Index”.The analysis of traffic uncertainty levels is a valuable attempt for researching the spatiotemporal evolution rules of road network traffic demands.