Dynamic analysis and data-driven inference of a fractional-order SEIHDR epidemic model with variable parameters

To analyze and predict the evolution of contagion dynamics, fractional derivative modeling has emerged as an important technique. However, inferring the dynamical structure of fractional-order models with high degrees of freedom poses a challenge. In this paper, to elucidate the spreading mechanism and non-local properties of disease evolution, we propose a novel fractional-order SEIHDR epidemiological model with variable parameters, incorporating fractional derivatives in the Caputo sense. We compute the basic reproduction number by the next-generation matrix and establish local and global stability conditions based on this reproduction number. By using the fractional Adams–Bashforth method, we validate dynamical behaviors at different equilibrium points in both autonomous and non-autonomous scenarios, while qualitatively analyze the effects of fractional order on the dynamics. To effectively address the inverse problem of the proposed fractional SEIHDR model, we construct a fractional Physics-Informed Neural Network framework to simultaneously infer time-dependent parameters, fractional orders, and state components. Graphical results based on the COVID-19 pandemic data from Canada demonstrate the effectiveness of the proposed framework.