详情

原文传递 EVALUATION OF INCIDENT DETECTION METHODOLOGIES

题名：	EVALUATION OF INCIDENT DETECTION METHODOLOGIES
作者：	Hani S. Mahinassani, Carl Haas, Josh Peterman, and Sam Zhou
关键词：	Incident detection methodologies, freeway Congestion, traffic management strategies
摘要：	The detection of freeway incidents is an essential element of an area’s traffic management system. Incidents need to be detected and handled as promptly as possible in order to minimize traffic delays. Various algorithms and detection technologies are examined to determine which combinations offer optimized detection performance. This study represents an effort to compile, compare, and tank available incident detection strategies. Based on an extensive literature review, as well as on interviews with traffic management personnel, die California algorithm No. 8, McMaster algorithm, Minnesota (DELOS), and Texas algorithms were selected for testing. The performance of these algorithms was assessed using extensive incident and traffic data from San Antonio, Texas. For training purposes, die data were separated into subsets for calibration and testing. During calibration, algorithm parameters were optimized via a Monte Carlo estimation process. Trained algorithms were then tested and evaluated according to traffic data aggregation (smoothing) and incident type. Results verify the validity of the calibration process, though algorithm performance varied slightly between calibration and testing phases. Each algorithm performed differently under different situations. Based on this perception, a holistic data (algorithm) fusion and information fusion model was developed to exploit the advantages of different algorithms and incident detection resources. The fusion approaches were explored, and fusion results on the calibration data set were analyzed. Finally, recommendations were proposed and future work was identified.
报告类型：	科技报告