Modeling the Pressure-Strain Correlation in a Differential Reynolds Stress Model using Feature Engineering for an Evolutionary Algorithm

Kurzfassung

This work focuses on determining functional forms for the coefficients of the SSG Reynolds Stress model using a machine learning approach referred to as Genetic Evolution Programming (GEP). This method uses evolutionary principles to evolve a set of possible equation candidates for the coefficients over multiple iterations. While it produces interpretable and easy to implement equations, its convergence is less predictable and controllable than other machine learning techniques and shows a high sensitivity to its initialization, inputs and training parameters.