Multi-Objective Stochastic Optimization Algorithms to Calibrate Microsimulation Models

Abstract

The calibration process for microscopic models can be automatically undertaken using optimization algorithms. Because of the random nature of this problem, the corresponding objectives are not simple concave functions. Accordingly, such problems cannot easily be solved unless a stochastic optimization algorithm is used. In this study, two different objectives are proposed such that the simulation model reproduces real-world traffic more accurately, both in relation to longitudinal and lateral movements. When several objectives are defined for an optimization problem, one solution method may aggregate the objectives into a single-objective function by assigning weighting coefficients to each objective before running the algorithm (also known as an a priori method). However, this method does not capture the information exchange among the solutions during the calibration process, and may fail to minimize all the objectives at the same time. To address this limitation, an a posteriori method (multi-objective particle swarm optimization, MOPSO) is employed to calibrate a microscopic simulation model in one single step while minimizing the objectives functions simultaneously. A set of traffic data collected by video surveillance is used to simulate a real-world highway in VISSIM. The performance of the a posteriori-based MOPSO in the calibration process is compared with a priori-based optimization methods such as particle swarm optimization, genetic algorithm, and whale optimization algorithm. The optimization methodologies are implemented in MATLAB and connected to VISSIM using its COM interface. Based on the validation results, the a posteriori-based MOPSO leads to the most accurate solutions among the tested algorithms with respect to both objectives.

A Multi-Objective Approach for the Calibration of Microscopic Traffic Flow Simulation Models

“…However, other multi-objective metaheuristics were not considered, and the hyperparameters of the algorithms were not properly tuned. Karimi, et al [23] achieved great results using a multi-objective stochastic optimization algorithm for the calibration of a microsimulation traffic flow model. However, they compared the performance of their proposed algorithm with the results obtained using a mono-objective approach.…”