Abstract for: Deep Learning for System Dynamics Parameter Calibration

Model parameter calibration is a demanding and time-consuming task that must be completed before the system dynamics (SD) model can be used to understand and improve a system's dynamic behavior. This paper employs deep learning as a parameter calibration tool for SD models. We test our proposed deep learning calibrator on the SEIRb epidemic model. Specifically, we train our calibrator on the data generated by SEIRb, mapping the system's infection and death rate dynamics to a set of model parameters. Then, we use the trained deep learning model as a calibrator to estimate the model parameters of a previously unseen infection and death rate data. Our results show that the deep learning calibrator accurately predicts the model parameters, accelerates the calibration process, and offers more robustness to the process and measurement noise.