Regression models for predicting the effect of trash rack on flow properties at power intakes

Abstract

Vortex flow characteristics in a reservoir and horizontal water intake have been predicted by using regression models in this numerical research. In this paper, three standalone machine learning models – Random Forest (RF), K-nearest neighbours (KNN), Gradient Boosting (GB) – and a proposed hybrid model based on Lévy Jaya Algorithm (LJA) and GB (LJA-GB) are employed to estimate the effect of trash racks on flow properties at power intakes. The experimental data which are prepared for the proposed study in this paper were obtained through a rectangular laboratory tank 8.3 m3 with various submergence depths and Froude numbers on nine trash racks with 63.7%–84.1% opening, made out of 2, 2.5, 4, and 6 mm thick copper wire. The outcomes revealed that the proposed LJA-GB model shows the best overall performance among the four models used for estimation. Thus, the LJA-GB model has the lowest mean absolute error (MAE) (0.3344), mean squared error (MSE) (0.1784), and root mean squared error (RMSE) (0.4223) values and highest R-squared ([Formula: see text]) (0.9899) and Willmott’s index (WI) values (0.9508) in the testing stage metrics for [Formula: see text] estimation and MAE (0.0061), MSE (0.0001), RMSE (0.0073), [Formula: see text] (0.9971), WI (0.9727) for [Formula: see text] estimation. Whereas the RF and KNN models exhibited poor performance in both stages of estimation.

Keywords

• Hydro power plant
• experimental modelling
• vortex strength
• regression models