gLaSDI: Parametric physics-informed greedy latent space dynamics identification

Xiaolong He, Youngsoo Choi, William D. Fries, Jonathan L. Belof, Jiun-Shyan Chen

Abstract: Highlights•Non-intrusive reduced-order modeling based on physics-informed greedy latent-space dynamics identification is proposed.•An autoencoder and parametric models are trained interactively to identify intrinsic and simple latent-space dynamics.•A physics-informed adaptive greedy sampling algorithm is introduced to search for optimal training samples on the fly.•A kNN convex interpolation scheme is applied to exploit local latent-space dynamics for enhanced generalization.•The proposed method achieves 17-2,658x speed-up and 1-5% relative errors for various nonlinear dynamical problems.