Calibration, validation & verification

Abstract

This chapter explains the processes of verification, calibration, and validation in pedestrian modelling. These are essential processes in the design and use of pedestrian models that together ensure accurate simulations of pedestrian behavior. Verification confirms that the model's implementation aligns with its conceptual design, calibration adjusts model parameters to improve accuracy, and validation assesses how well the model represents real-world pedestrian movements. Verification involves a structured process of testing whether the implemented model accurately reflects the conceptual model. This is done through a series of verification test cases, which compare the simulated outcomes to what is expected from the conceptual model. Calibration and validation are interrelated but serve different purposes. Calibration is an iterative process that fine-tunes model parameters to minimize errors between simulation results and reference data. Validation, on the other hand, assesses how accurate a pedestrian model replicates pedestrian behavior and dynamics. The state-of-the-art approach involves multi-objective calibration and validation, where multiple scenarios and metrics (i.e. objectives) are used to calibrate and validate the model. The choice of objectives has a major impact on the calibration and validation results. Key is that the scenarios and metrics are chosen such that they cover and capture all the relevant behaviors and dynamics. Which behaviors and dynamics are relevant depends on the intended use of the model and the type of modelled behavior. As most pedestrian models are stochastic or use stochastic parameters it is essential that during calibration and validation replications, repeating the simulation multiple time using the same inputs, are run to deal with this. Lastly, a sensitivity analysis of the model is also important to determine which parameters the model is most sensitive to. This guides the calibration process and can ensure that the calibration is as efficient as possible. All these processes are explained in detail in this chapter. This includes descriptions of how to apply them in the context of pedestrian behavior modelling and what are important factors to consider. This chapter therefore provides guidance for both model developers in creating valid models and model users is assessing the quality of their model for the intended application.