Traffic Simulation Failure Detection and Analysis

Abstract

Microscopic, stochastic traffic simulation may yield simulation failures under multiple replications. The failed runs are not valid in the estimation of traffic performance and should be excluded from the final simulation output analysis. On the other hand, these failure runs provide important clues to perform a simulation flaw diagnosis. An unconventional failure detection and analysis methodology was proposed to comprise three layers: time series inspection, spatial analysis, and causal analysis. The process of time series inspection traces the variation of indicator variables over the time domain for the purpose of detection of simulation failures. The spatial analysis identifies failure occurrence patterns, and the subsequent causal analysis judge contributing factors to simulation failures using a tabular method in combination with other tools.

A widely-used traffic simulator, CORSIM, is used as the test-bed simulator. Three real-world traffic networks were simulated as the case studies for the proposed method. The study results indicated that the proposed failure detection and analysis method is valid and effective to improve traffic simulation from multiple perspectives. Its application in the evaluation of networks testified its utility in multiple aspects. The proposed procedure helped uncover the existing deficiencies in the current simulation models, and, therefore, provide important guidance for the organized model improvement efforts. On the other hand, the procedure was also applied in the analysis of a projected traffic scenario to testify its value in the identification of critical sites on the network from the traffic engineering perspective.