Nonintrusive reduced order modeling framework for quasigeostrophic turbulence.

Area of Science:

• Computational fluid dynamics
• Machine learning applications in science
• Nonlinear dynamical systems

Background:

• Large-scale quasistationary systems pose significant computational challenges.
• Conventional projection-based reduced order modeling (ROM) often requires numerous modes for stability and struggles with intermittent dynamics.

Purpose of the Study:

• To develop a nonintrusive reduced order modeling (ROM) framework for large-scale quasistationary systems.
• To leverage Long Short-Term Memory (LSTM) networks for time series prediction within the ROM framework.

Main Methods:

• Utilized Proper Orthogonal Decomposition (POD) to extract modal coefficients from high-resolution data.
• Trained an LSTM model on these modal coefficients for recursive time series prediction.
• Reconstructed flow fields and time series using inverse POD transform.

Main Results:

• The proposed nonintrusive ROM framework with LSTM (ROM-LSTM) achieves stable solutions with fewer POD modes compared to ROM-Galerkin projection (ROM-GP).
• ROM-LSTM accurately captures quasiperiodic intermittent bursts and provides stable, accurate mean flow dynamics.
• Demonstrated significantly higher accuracy and faster performance with larger time step sizes than ROM-GP.

Conclusions:

• The nonintrusive ROM-LSTM framework is a robust and highly efficient method for predicting chaotic nonlinear fluid flows.
• This approach bypasses the need for prior knowledge of governing equations, offering a versatile modeling tool.