Calibrating supply parameters of large-scale DTA models with surrogate-based optimisation

This study is among the early attempts to employ a surrogate-based optimisation (SBO) approach to solve the large-scale dynamic traffic assignment (DTA) calibration problem that is characterised by an expensive-to-evaluate and non-closed-form objective function. This paper formulates the calibration of the large-scale DTA model as a bi-level optimisation problem with a non-closed objective function such that it can only be evaluated through simulation. The Kriging surrogate model is adopted to construct the response surface between the objective value and the decision variables. The SBO approach first evaluates a number of initial samples, then fits the response surface and searches for the optima via an infill process. It reduces the number of large-scale DTA runs for evaluating the objective values and saves much computational time. For demonstrative purposes, a real-world large-scale DTA model in the state of MD is calibrated with the proposed SBO approach. After 400 initial points and 100 infill points, the SBO approach reduces the calibration matching gap from 29.68 to 21.90%. It is also presented that the proposed SBO is significantly faster than the genetic algorithm in searching for better solutions. The results demonstrate the feasibility and capability of SBO in DTA calibration problems.