Key points are not available for this paper at this time.
Abstract Traffic microsimulation models are crucial for intelligent transportation systems evaluation, but careful parameter calibration is required for credible pre‐ and post‐ITS comparisons. However, the back‐box and stochastic nature of the system make calibration challenging. Sensitivity analysis (SA) helps to identify influential parameters, but scenario dependency limits its generalisability. Metrics such as root mean squared relative error (RMSRE) can oversimplify the stochasticity in traffic data, compromising calibration quality. Furthermore, calibration for specific key performance indicators (KPIs) does not ensure the reliability of other KPIs. The authors propose the simultaneous calibration of driving behaviour parameters without prior sensitivity information. They demonstrate the effect of the error metric and objective function facets on the calibration efficiency and parameter convergence consistency. Results indicate that employing SA to identify influential parameters is unnecessary. Each parameter converges to a stable point by responding directly to the information within the objective function or due to the interactions with other parameters. Therefore, simultaneous calibration of multiple KPIs and maintaining the stochasticity structure in the data—enhanced calibration efficiency and parameter convergence consistency. Additionally, using probabilistic dissimilarity metrics that consider the entire distribution, such as the Wasserstein distance, outperform the K–S distance and RMSRE.
Abuamer et al. (Fri,) studied this question.