Traffic congestion problem has been boring many cities worldwide since the new century. Somehow the conventional bus system fails to act as the major role in meeting the increasing urban travel demand, because of its deficiency, uncomfortableness and great latency. TSP (Transit Signal Priority) has been recognized to be an effective way to improve the level of service of public transit. However applying TSP to convetioanl bus system is unpractical because of the great number of bus routes and criss-corsses. At the same time, BRT (Bus Rapid Transit), as a new public transit mode, is attracting more and more interests of researchers and engineers. The BRT routes are simple and of high level of informatization. This dissertation focuses on the problem of providing TSP for BRT buses. Firstly the dissertation gives a brief introduction to BRT systems and reviews the PTP (Public Transit Priority) research. PTP methods can be roughly classified into three categories: passive priority, active priority and real-time priority. Characteristics, implementation frameworks and development histories of the three categories of methods are summarized. Then the dissertation proposes a passive PTP method based on signal coordination. This method requires that all signals have the same cycle length. Signal coordination is achieved by estimating the travel times between consecutive signals and calculating proper signal offset for each signal. This method is simple and easy to implement and can result in great travel time saving for buses under proper conditions. Next the dissertation proposes a continuous travel time prediction model towards active TSP system based on a deep network. The deep network is composed of a stacked autoencoder and a logistic regression predictor. The vehicle speed, traffic flow, traffic density and signal time when the vehicle passes the detector are taken as network input and the output gives the predicted continuous travel time to the intersection. Big data (simulated) are applied to train the model. On the testing data set it produces state-of-the-art prediction accuracy with mean absolute error of 4.13 seconds. Based on the research above, it’s ready for the dynamic active TSP model. Within this model, three signal priority action are combinedly used: green extension, red truncation and phase insertion. Action window for each priority action is located within the signal cycle according to the temporal width of predicted bus arrival window. ...
修改评论