Calibration of urban rail simulation models: A methodology using SPSA algorithm

Rail simulation model calibration is a process of adjusting model parameters while comparing model output with observations from the real rail system. There is a lack of systematic methodology for calibrating urban rail simulation models. Based on a simulator developed for urban rail operations and control, the paper demonstrates a methodology of calibrating model parameters, and specifically, fine-tuning some of the simulation inputs. The calibration process is modeled as a multi-variate optimization problem and solved by the Simultaneous Perturbation Stochastic Approximation (SPSA) algorithm. A case study of the Massachusetts Bay Transportation Authority (MBTA) Red Line shows that the methodology improves the simulation model dramatically in terms of replicating the track block runtimes. At the same time, it upgrades the station specific dwell time parameters and enhances a-priori boarding rates at stations fairly effectively.