A methodology for data-driven signal timing optimization

The overall goals of this dissertation will contribute to the Intelligent Transportation System (ITS) field by developing a methodology to effectively utilize traffic data and to optimize traffic signal control. The statement of this dissertation is that the methodology, integrating data mining techniques, principal variables extraction methodologies, state-based modeling algorithms, and traffic simulations, can be developed to improve the performances of traffic signal control systems.; ITS systems include large numbers of traffic sensors that collect enormous quantities of data. The historical traffic data is in fact capable of proving abundant amount of information that can aid in the development of improved current control strategies. One data reorganization procedures were developed to refine the raw traffic data. Data mining tools are the keys to explore the in the traffic data. These algorithms help to find the principal variables and to classify traffic data by scientific techniques under scientific logic, not guesstimates or experiences.; This dissertation describes the research investigating the application of data mining tools to aid in the optimization of traffic signal timing control. Specifically, the principal variables extraction studies make the analysis of large traffic corridor areas are possible. One six-intersection corridor case study was conducted to illustrate that the use of data mining tools. The results of a six-intersection corridor show the techniques of principal variables extraction can effectively reduce the data space and maintain system performances.; One adaptive signal control model based on the classification of historical traffic data was also introduced, and specified by the results of six-intersection case. Using CART analysis every new surveillance observation can be assigned to one traffic state and one corresponsive signal timing plan is applied to solve this traffic demand. Two artificial traffic conditions were developed to evaluate the adaptive signal control model.; The results of this dissertation indicated that advanced data mining techniques hold high potential to assist traffic engineers in signal control system design, development and operations. This methodology has good performances in the adaptive signal control model, and the traffic state transition matrices investigation, for the development of automatically, real-time signal control systems.