Turbulent Reynolds Stresses Prediction using Stochastic Gradient Boosting Regression

Abstract

Predicting turbulent Reynolds stresses (TRS) accurately is crucial for the advancement of fluid dynamics and engineering applications. This study presents an application of stochastic gradient boosting regression (GBR) to predict TRS within a turbulent vertical axisymmetric jet. Leveraging a comprehensive dataset encompassing flow rate, spatial development ratios, velocity profiles, and root mean square of velocity fluctuations, we engaged in a thorough analysis to capture the intricate dynamics governing TRS. Our approach involved the application of GBR, a machine learning algorithm renowned for its precision and flexibility. GBR’s capability to optimize any differentiable loss function was utilized to minimize predictive errors iteratively. The model’s performance was scrutinized using metrics such as mean squared error (MSE), root mean squared error (RMSE), mean absolute error (MAE), and the coefficient of determination. Intriguingly, our findings revealed that feature scaling, a common preprocessing technique, did not enhance the model’s performance. The unscaled model exhibited superior accuracy, challenging the notion that feature scaling is universally beneficial. The study’s findings underscore the importance of empirical validation of preprocessing techniques and spotlight the effectiveness of GBR in modeling TRS. The insights from this research could have significant implications for turbulence modeling practices and machine learning methodologies within fluid dynamics and other related fields.