Reduced Order Modeling of 2D Turbulent Electro-Thermo-Convection

Turbulent buoyancy-driven convection (known as Rayleigh-Bénard convection) is hard to predict. Computational fluid dynamics (CFD) simulations are often used to model the dynamics of the fluid. When solving for turbulent or transient dynamics, running CFD simulations can be extremely time consuming. In its place, machine learning can be used to forecast these complicated dynamics within the fluid. Two machine learning models that can be used with this time series data are the Long Short-Term Memory (LSTM) and Echo State Network (ESN) models. This presentation will discuss how these machine learning models, when trained on CFD output data, can predict the dynamics of Rayleigh-Bénard convection.