Research On Dynamic Origin-destination Matrix Estimation Of Urban Network Using Simulation Method

Advanced traffic management system (ATMS) is the important research part of intelligent transport system whose core is to release pre-trip traffic network information and route guidance information on the way through traffic management center(TMC for short). Besides, DTMS will also genetate various time-dependent traffic controls plan so that travel behaviour guidance and traffic congestion can be eased. The implement of DTAS must regard time-varying OD matrix as basic input. Therefore the estimation and prediction of time-varying OD matrix become the first basic and important work to be done in the procedure of DTMS. Time-dependent OD matrix is often described as traffic distribution flow in consecutive time intervals with short span (15~20min on average). However, it is tuff and expensive task to obtain these data through direct traffic survey, common method is to estimate it through partial observed link flow data. Because of convenience to detect and real-time property of the observed flow data, it is possible to make such estimation and prediction.Time-dependent OD matrix problem generally include off-line and real-time estimation. The former tries to estimate time-dependent OD matrix in the case of known link flow data whose purpose is to acquire traffic distribution information, while the later is first to make an estimation of current OD matrix and then predict the OD flow in future time intervals, then an iterate procedure is started whose purpose is to make preparation for traffic management and control system. In recent decades, researchers in different field have made a more in-depth analysis from their own viewpoint on dynamic OD matrix estimation and prediction, and a lot of theoretical framework and mathematical model have been proposed. This thesis firstly racall the research progress of this field, study the advantages and drawbacks in each model, make clear the key issues, and then study the effective estimation method on dynamic assignment matrix, multi-path choice problem, the influence of error in assignment matrix on the estimated OD matrix, and also kalman filter prediction model, etc. it will lay the foundation for further research, main work of this thesis follows:1. Study review of OD matrix estimation and prediction problem is made. This review is done from static to dynamic part and a classification of various methods is made as well as the advantage and disadvantage of different ideas and models. Finally the comparation of static and dynamic estimation method is carried out.2. For the lack of valid estimation method on dynamic traffic assignment matrix, present a simulation-based dynamic assignment matrix estimation method and corresponding calculation model. Dynamic assignment matrix describes the mapping between the current observed link flows and the unobserved OD flows in prior periods. It is the key parameter in both off-line and real-time OD estimation problem. Simulation method can give a more realistic and vivid description of dynamic interaction between traffic demand side and supply side, and more important, needn't to meet first-in-first-out (FIFO) rule. New method introduces“restricted road network traversal algorithm”to set paths in simulation project and Path-Size logit model to determine driver's route choice so that otherwise IIA shortcoming within multiple logit model may be avoided. Calculation model of assignment matrix is able to overcome the unrealistic ideal“equal-headway among vehicles during trip”assumption that made by analysis model. Empirical studies show that new method can produce better assignment result.3. For the lack of consideration of traffic assignment matrix estimation error on the OD estination result, present the improved offline dynamic OD estimation models which take the random error of assignment matrix into account. We firstly analyze the various errors in the procedure of assignment matrix estimation, then the endogenous effects on OD matrix be estimated are analyzed and then two kinds of improved models, the improved simultaneous and sequential estimation procedure, are proposed. The former one trys to calculate the entire OD matrix in all intervals and corresponding assignment matrix one time, while the later trys to make estimation for each OD matrix. In addition, a sharp increase in optimization variables and the function form transformation make us to solve the new models using genetic algorithm which has glabal convergence. Empirical studies have shown that the new model can greatly improve the estimation accuracy of OD matrix.4. Build a state-space model which take the intersection detection flow into account and propose corresponding one-step kalman filter algorithm to estimate and predict dynamic OD matrix. The detection flow in interaction can reflect the driver's driving state under the control of signal plan and some non-motorized transport interference(such as bicycle and pedestrian) and its discontinuous character is an effective supplement of link flow which may be regard as continuous.Our idea is firstly to determine the state vector, construct state transition and measurement equation relate to ODME and then to use kalman filter algorithm to estimate current OD matrix and predict the next one. In addition, we propose the method to estimate the measurement error variance, the variance matrix of state transition error, assignmtnt matrix as well as initial state of dynamic system.Empirical studies indicate that adding this information to measurement equation can greatly improve the prediction accuracy.