Reduced Order Model
A modeling approach that approximates the high-dimensional dynamics of an original system with a small number of degrees of freedom. Common implementations include proper orthogonal decomposition, Krylov subspace methods and deep neural network surrogate models, used to accelerate multiphysics simulation.
Timeline (last 90 days)
Reduced-order models (ROM) approximate the high-dimensional dynamics of the original system through a small number of degrees of freedom; common approaches include proper orthogonal decomposition (POD), Krylov subspace methods, and deep neural network surrogate models.
A single transient solve of a complete three-dimensional electrochemical-thermal-electrical coupled finite element model takes hours or even days; after a reduced-order model is trained, a single inference is usually on the order of milliseconds.
The accuracy of ROM is limited by the coverage of the training data, and the error increases significantly when extrapolating to operating conditions outside the training set.
All Facts (3)
Reduced-order models (ROM) approximate the high-dimensional dynamics of the original system through a small number of degrees of freedom; common approaches include proper orthogonal decomposition (POD), Krylov subspace methods, and deep neural network surrogate models.
50%UnverifiedA single transient solve of a complete three-dimensional electrochemical-thermal-electrical coupled finite element model takes hours or even days; after a reduced-order model is trained, a single inference is usually on the order of milliseconds.
50%UnverifiedThe accuracy of ROM is limited by the coverage of the training data, and the error increases significantly when extrapolating to operating conditions outside the training set.
50%