Safety-Integrated Calibration and Validation of Simulation Models for Complex Innovative Intersections Using Bivariate Extreme Value Theory

This paper presents a safety-integrated calibration and validation framework for microscopic traffic simulation models applied to two unconventional intersection designs: Diverging Diamond Interchanges (DDI) and Continuous Flow Intersection (CFI). Sites from Utah, United States, were modeled in PTV VISSIM and calibrated against three key Measures of Effectiveness (MOEs): Traffic Volume, Average Speed, and Crash Frequency. Unlike traditional calibration approaches, the proposed framework incorporates surrogate safety analysis directly into the calibration process by embedding bivariate Peak Over Threshold (POT) modeling from Extreme Value Theory (EVT). This method jointly captures conflict severity using Time to Collision (TTC) and Maximum Deceleration (MaxD), allowing crash frequency estimation from simulated conflicts. The calibration results showed strong agreement with field data, achieving GEH values below 5 for all movements and minimal speed errors. Simulation-based crash estimates differed from observed crash frequencies by only 0.07 crashes per year for the CFI. Validation using 2019 data further confirmed the predictive capability and generalizability of the models. This study contributes a robust and transferable framework for evaluating both operational efficiency and safety performance in innovative intersection design.