Predictive modeling for savonius hydrokinetic turbine performance: a machine learning investigation

This paper investigates the use of artificial intelligence (AI) utilizing machine learning (ML) techniques to predict the performance of Savonius water turbines. Four ML models, including CatBoost (Categorial Boosting), random forest, gradient boosting, and support vector regression (SVR), were used to forecast turbine performance. Independent input variables were selected based on principles of Fluid Mechanics. This selection included independent variables representing the turbine's geometrical properties, water flow characteristics, operational conditions, and test conditions. This investigation reveals that the CatBoost and random forest models exhibit superior predictive capabilities. They are characterized by high Coefficient of Determination values, demonstrating their robustness in handling the complex dynamics of water turbine performance. CatBoost achieves the best results on the test dataset with mean absolute error (MAE), mean squared error (MSE), and Coefficient of Determination values of 0.0147, 0.0004, and 0.9643. The comprehensive evaluation approach (employing MAE, MSE, and Coefficient of Determination) helps in understanding each model's predictive accuracy and error behaviors, thereby guiding the selection of the most suitable model for this particular engineering application. This study introduces implemented AI frameworks for predicting turbine performance, addressing challenges caused by the absence of specific mathematical formulas for turbine performance and limited data availability in this field.