Coefficient identification in a SID modelling of epidemic propagation for COVID-19 by using the Physics-Informed Neural Network approach

This work introduces a new approach utilizing Physics-Informed Neural Networks (PINNs) and a novel Susceptible-Infected-Death (SID) mechanistic model to dynamically estimate both the state variables and transmission rate of an epidemic model. PINNs leverage both available data and the underlying system equations, enabling them to address ill-posed inverse problems typically encountered in epidemiological modeling. We present a split-weight strategy for implementing PINNs that significantly improves accuracy and computational efficiency compared to traditional joint approaches. Our method is rigorously evaluated using real-world data from the COVID-19 pandemic in Shanghai. The split-weight PINN framework and SID model demonstrates superior performance, achieving up to a 12% and 16% increase in accuracy. This research offers a powerful tool for enhancing the reliability and adaptability of epidemic models, enabling more accurate predictions and informed public health interventions.