Model Validation Against an Actual System

Ultimately we would want to calibrate our simulation against an actual transit system modeled by it. Due to the discrete timestep nature of our schedule optimization model, the simulation would only be capable of providing an approximation of a live system's performance. However, if the live system uses the same schedule optimization algorithm used in our simulation, we wouldn't expect simulated versus live performance to differ appreciably unless vehicles run late and passengers miss connections. The simulation currently does not model these types of unexpected events, but adding such probabilistic failures to the simulation shouldn't pose much of a challenge. The challenge lies in calibrating those probabilities against those that might occur in the live system due to factors discussed in section .