Detecting changes in freeway traffic states using the CUSUM algorithm

Hualiang Teng, Purdue University

Abstract

The process of detecting a change in freeway traffic states should fully utilize the available information about the traffic processes before and after a state change as well as the state change occurrence distribution. The literature review revealed that only parts of this information are used by existing state change detection algorithms. In this thesis, state change detection algorithms were developed by applying the CUSUM algorithm where varying levels of a priori information on changes in traffic states can be incorporated. To apply the CUSUM algorithm, the properties of traffic processes before and after state changes and the distributions of state change occurrence were first investigated. The results indicated that it may not be realistic to apply the original CUSUM algorithm, in which full information about changes is assumed to be known. Thus, four algorithms were designed by varying the amount of prior information integrated in the formulation. These algorithms and some relevant state-of-the-art algorithms were evaluated under different simulation conditions. A practical evaluation with field data was also performed. The four versions of the CUSUM algorithm are promising for use in practice due to their desirable performance. These algorithms can be applied to other transportation problems such as determining uniform pavement sections.

Degree

Ph.D.

Advisors

Madanat, Purdue University.

Subject Area

Civil engineering|Transportation

COinS