A Hybrid Physics-Informed Neural Network
Michael Ajao-Olarinoye, Vasile Palade, Fei He, Petra A. Wark, Zindoga Mukandavire, Seyed Mousavi · 2025
COVID-19 poses a complex and unprecedented challenge to public health and society, requiring scientific efforts to understand, model, diagnose, and control the disease. The integration of physics-informed neural networks (PINNs) with epidemiological models offers a powerful tool for understanding the dynamics of infectious diseases and developing effective strategies to control and mitigate their impact. In this study, we develop a novel hybrid approach to modelling and forecasting the dynamics and the demand for healthcare resources. We take advantage of the concept of PINNs to capture the dynamics of infectious diseases, estimate key parameters, and unobserved states in the ability of the model, using the neural network to integrate physical laws directly into the learning process of a modified compartmental model. We also make use of the strength of recurrent neural networks (RNNs) to combine data generated by a PINN and lagged covariates of target variables to forecast demand from the intensive care unit demand using different variations of RNN models. The results indicate that our proposed hybrid framework is highly effective in capturing the dynamics of the pandemic and providing insight into the dynamics of the disease.