Backpropagation and gradient descent for an optimized dynamic mode decomposition

TL;DR

This paper introduces a robust optimization method for dynamic mode decomposition that leverages deep learning techniques like automatic differentiation and stochastic gradient descent to analyze complex, noisy data.

Contribution

It presents a novel approach combining deep learning tools with DMD, enabling simultaneous computation of eigenvalues, modes, and amplitudes with regularization and physical constraints.

Findings

Effective on complex, noisy datasets

Successfully applied to transonic shock buffet data

Enhanced flexibility and robustness in DMD analysis

Abstract

We present a robust and flexible optimization approach for dynamic mode decomposition analysis of data with complex dynamics and low signal-to-noise ratios. The approach borrows techniques and insights from the field of deep learning. Specifically, we employ automatic differentiation and stochastic gradient descent to compute eigenvalues, modes, and mode amplitudes simultaneously. The method allows embedding regularization or physical constraints into the operator definition. The optimization approach is applied to three examples of increasing complexity, the most challenging of which is an experimental dataset of transonic shock buffet on a swept at realistic flight conditions.