結合動態交通指派之旅次起迄推估與預測之研究

DTA-BASED DYNAMIC ORIGIN-DESTINATION DEMANDS ESTIMATION AND PREDICTION MODEL

本研究主要利用卡門濾波方法結合動態交通指派模式DynaTAIWAN，建構旅次起迄推估模型，以歷史旅次起迄（Origin-Destination, O-D）流量與路段流量資料，建立模式之參數矩陣。本研究主要貢獻包括：（1）系統狀態參數以O-D流量差值（deviation）取代O-D流量值，使推估之O-D流量更精確且常態化；（2）結合動態交通指派模式，以程式擷取指派矩陣；（3）建立MySQL儲存與管理資料庫，將動態O-D推估與預測之程序分解成七大步驟，未來可逐步改善過程之精確度與效率。並透過車流模擬軟體DynaTAIWAN於三實驗路網上進行模式之驗證，並建構相關敏感度分析，包括小路網、市區號誌化路網與50節點高速公路與市區道路混合路網，並經由評估指標均方根誤差（RMSE）及卡方檢定檢測推估之結果。經由結果顯示，三路網上之RMSE值分別小於2、4、17，卡方檢定顯示推估與實際O-D流量值無差異，本研究所提出之動態O-D推估模式具有精確且合理之推估結果。

This research aims at integrating dynamic traffic assignment model DynaTAIWAN with the Kalman Filtering (KF) approach to construct the dynamic Origin-Destination (O-D) estimation and prediction model; the dynamic parameters based on the historical and real time data are generated to meet the dynamic traffic conditions. The contributions include: 1. the model takes the deviations of O-D flows from historical averages instead of O-D flows as the state vectors to increase the accuracy and normalization of estimations; 2. the time-dependent assignment matrix is gained in advance via C++, the historical O-D flows are assigned into DTA and the vehicle trajectory data accounts for calculating the assignment parameters; 3. the procedure of O-D estimation and prediction is decomposed into 7 steps, and the efficiency and accuracy can be improved step by step. Numerical experiments to illustrate the proposed model are conducted in three networks: a small test network, a signalized urban network and a 50-node mixed network, and several sensitivity analyses are performed. The measurement criteria includes RMSE and the chi-square test which are utilized to examine the results, the results show that the RMSE values are less than 2, 4, and 17 on the three networks respectively, and the chi-square tests reveal there are no differences between the estimated and real O-D flows. The numerical results indicate that estimated O-D values from the proposed model are reasonable and accurate.