Search

A systematic procedure is presented for calibrating and validating a microscopic model of safety performance. The context in the model application is the potential for rear-end crashes at signalized intersections. VISSIM v.4.3 provides the simulation platform for estimating the safety performance for individual vehicles and has been calibrated and validated using separate samples of observed vehicle tracking data extracted from the FHWA/NGSIM program. The calibration exercise involves four sequential steps: (1) heuristic selection of initial model inputs, (2) statistical screening using a Plackett-Burnman design, (3) fractional factorial analysis relating inputs to safety performance, and (4) genetic algorithm procedure for obtaining best estimate input values. Three measures of safety performance were considered: crash potential index, number of vehicles in conflict and total conflict duration per vehicle. Model consistency was assessed by comparing simulated and observed safety performance based on a separate validation sample of vehicle tracking data. The suggested procedure was found to effectively estimate model input parameters that closely matched safety performance measures in the observed validation data. This procedure yields an objective and efficient means for simulation model calibration applied for estimating safety performance at signalized intersections.

Frequently transportation engineers are required to make difficult safety investment decisions in the face of uncertainty concerning the cost-effectiveness of different countermeasures. For certain types of highway-railway grade crossings, this problem is further aggravated due to the lack of observed before and after collision data that reflects the impact of specific countermeasures. This study proposes a Bayesian data fusion method as an attempt to overcome these challenges. In this framework, we make use of previous research findings on the effectiveness of a given countermeasure, which could vary by jurisdictions and operating conditions to obtain a priori inference on its expected effects. We then use locally calibrated models, which are valid for a specific jurisdiction, to develop the current best estimates regarding the countermeasure effects. By using a Bayesian framework, these two sources are integrated to obtain the posterior distribution of the countermeasure effectiveness. As a result, the outputs provide information not only of the expected collision response to a specific countermeasure but also its variance and corresponding probability distribution for a range of likely values. Examples from Canadian highway-railway grade crossing data are used to illustrate the proposed methodology and the specific effects of prior knowledge and data likelihood on the combined estimates of countermeasure effects.