Численное моделирование задачи ассимиляции граничных данных при восстановлении старшего коэффициента модели диффузии–конвекции–реакции
Рубрика: Математика
Статья в выпуске: 4 т.18, 2026 года.
Бесплатный доступ
Рассматривается задача восстановления коэффициента температуропроводности в стационарной модели диффузии-конвекции-реакции по нормальной производной на части границы области функционирования модели. Задача сведена к минимизации функционала невязки, которая осуществляется методом проекции градиента. Градиент находится аналитически из системы оптимальности. Численное моделирование показало работоспособность метода и его чувствительность к выбору начального приближения, шага спуска и приближения градиента.
Короткий адрес: https://sciup.org/147255553
IDS: 147255553 | УДК: 517.9+550.331 | DOI: 10.14529/mmph260404
Numerical Simulation of the Boundary Data Assimilation Problem in Recovering the Leading Value of a Diffusion-Convection-Reaction Model
The paper considers an inverse problem for a stationary hydrodynamic diffusion–convection–reaction model. The goal is to recover the spatially distributed thermal diffusivity coefficient from observations of the normal derivative of the temperature on a part of the boundary. The forward problem is formulated as a boundary value problem with a homogeneous Dirichlet condition and prescribed velocity, reaction coefficient, and heat source. The inverse problem is treated as a control problem for the diffusivity coefficient and reduced to the minimization of a data-misfit functional over a convex set of admissible coefficients. An explicit expression for the Frechet gradient of the misfit functional is derived by using an adjoint equation and the optimality system. Approximations of the gradient based on a finite set of basis functions and on a simplified functional are proposed. The projected gradient method with a line search for the step size is applied. Numerical experiments are carried out in OpenFOAM using the finite volume method on a grid of 2500 cells. For several initial approximations the method converges to a satisfactory reconstruction: the relative error decreases to 2,6 %–9,1 % within 47–56 iterations. The results demonstrate the effectiveness of the proposed approach and its strong sensitivity to the initial guess, the step size, and the gradient approximation. The developed procedure is applicable to parameter identification in heat and mass transfer and control problems.