Development and implementation of calibration method for macrosimulation traffic models

Macroscopic traffic flow models are of paramount importance for traffic surveillance, control, and infrastructure planning, providing critical insights into aggregate traffic behavior. After developing a traffic flow model for a specific freeway segment or network, it must be calibrated against real-world data from the site to determine the optimal model parameters. Model calibration and validation assess two key qualities of a robust mathematical model: its ability to replicate real-world behavior and its applicability to different scenarios or locations. Model calibration aims to determine an optimal set of parameter values that minimize the discrepancy between real-world data and the model output, reflecting reality as accurately as possible. Meanwhile, model validation tests the ability of the calibrated model to generalize by applying it to data from different time periods or days, and potentially to forecast future traffic conditions. Several issues need to be considered in model calibration. Among the most important are complex traffic phenomena such as capacity drops, disruptions caused by entering and exiting ramp flows, and stop-and-go waves. Additionally, the persistent excitation of the traffic system is crucial to ensure that all traffic regimes and not only free-flow, dense, and congested regimes are sufficiently represented. Due to these issues, traffic model calibration has remained a complex challenge, requiring not only extensive data but also a robust methodology to integrate and interpret this data effectively. In previous studies, model calibration is typically formulated as a system identification problem of nonlinear dynamic systems and is solved using various optimization algorithms and the source of real data is limited to an average of a few days or only a few hours during the morning/evening, with limited model validation. This study introduces an aggregate calibration method for transport system models that leverages about one year of mobility data alongside other data sources within an optimization framework which not only considers the standard traffic conditions but also explicitly accounts for congestion and anomalies, such as accidents and special events, ensuring its applicability under diverse traffic conditions. The study delivers a benchmarking framework for mobility data analysis, addressing practical challenges and opportunities offered by telematics data, enabling the effective integration of mobility counts data with control system performance outcomes. The study explores the effectiveness of machine learning techniques in traffic pattern recognition, offering new perspectives on data-driven approaches in traffic management. Furthermore, comprehensive results are presented for widely used optimization algorithms which had not been compared before for calibrating METANET model including Particle Swarm Optimization (PSO), Nelder-Mead (N-M), Simultaneous Perturbation Stochastic Approximation (SPSA), and the Firefly Algorithm, highlighting their performance in model calibration. The proposed method is validated across multiple traffic scenarios, including both standard and abnormal conditions, demonstrating its robustness and adaptability. By addressing these challenges and providing innovative tools, this work advances the accuracy, reliability, and applicability of traffic models, paving the way for more efficient and sustainable transportation systems.