Methods for robust calibration of traffic simulation models

Well-calibrated traffic simulation model predictions can be highly valid if various conditions arising due to time-of-day, work zones, weather, etc. are appropriately accounted for during calibration. Calibration of traffic simulation models for various conditions requires larger datasets to capture the stochasticity. In this study we use datasets spanning large time periods to, especially, incorporate variability in traffic flow and speed. However, large datasets pose computational challenges. With the increase in number of stochastic factors, the numerical methods suffer from curse of dimensionality. We propose a novel methodology to address the computational complexity in simulation model calibration under highly stochastic traffic conditions. This methodology is based on sparse grid stochastic collocation, which treats each stochastic factor as a different dimension and uses a limited number of points where simulation is performed. A computationally-efficient interpolant is constructed to generate the full distribution of the simulated output. We use real-world examples to calibrate for different times of day and conditions and show that proposed methodology is more efficient than traditional Monte Carlo-type sampling. We validate the model using a hold-out dataset and also show the drawback of using limited data for macroscopic simulation model calibration. Modelers could often face situations with limited data in calibrating for a particular condition, often when using traffic sensor data. We augment the current data with other sources when sensor data is missing. For calibrating microscopic traffic simulation models needing customized models augmenting the default modeling, require detailed site-specific data. In such cases same generic calibration methodology may not be applicable and specialized formulations are required. We propose the use of a simulation-based optimization (SBO) framework for calibration of toll plaza models that economizes on data requirements. The novelty of the SBO framework is that parameters corresponding to unavailable data can be used as calibration parameters. Using case studies the benefits of the SBO framework are demonstrated. Furthermore, we combine the sampling and interpolation using stochastic collocation with the SBO framework. Using this hybrid framework, we perform calibration to obtain distribution of output from the toll plaza model that closely follows the observed measures at the toll plaza.Ph.D.Includes bibliographical referencesby Sandeep Mudigond