Performance evaluation of the savonius hydrokinetic turbine using soft computing techniques

Abstract

Savonius hydrokinetic turbine is a vertical axis turbine suitable for generating electrical power used in rivers and canals. The performance of the turbine is described in terms of power coefficient. The power coefficient (CP) depends on input variables such as aspect ratio, overlap ratio, blockage ratio, number of blades, blade arc angle, blade shape factor, twist angle, Reynolds number, and Tip Speed Ratio (TSR). In this study, different soft computing methods, namely CatBoost, Artificial Neural Network (ANN), Random Forests (RF), Multivariate Adaptive Regression Splines (MARS), Adaptive Neuro-Fuzzy Inference System (ANFIS), ANFIS-Genetic algorithm (ANFIS-GA), ANFIS-Slime Mold Algorithm (ANFIS-SMA), ANFIS-Marine Predators Algorithm (ANFIS-MPA) along with Linear Regression (LR) were used to predict the power coefficient of Savonius hydrokinetic turbines for the first time. In order to obtain appropriate data, experimental tests were conducted in an open water channel for different design configurations of Savonius turbines. Further, more data were collected from different references. These data were used for the training and testing process of the models. The precision of the methods was evaluated using multiple statistical indices. Three different training-testing scenarios were prepared to provide a reliable predictive model, and in all scenarios, the CatBoost method was superior. Sensitive analysis showed that aspect ratio is the most important design parameter among all effective parameters. As a result, the CatBoost is recommended to predict a Savonius hydrokinetic turbine's performance considering multi-input variables.