On the Application of Machine Learning in Savonius Wind Turbine Technology: An Estimation of Turbine Performance Using Artificial Neural Network and Genetic Expression Programming

This article addresses the application of artificial neural network (ANN) and genetic expression programming (GEP), the popular artificial intelligence, and machine learning methods to estimate the Savonius wind rotor's performance based on different independent design variables. Savonius wind rotor is one of the competent members of the vertical-axis wind turbines (VAWTs) due to its advantageous qualities such as direction independency, design simplicity, ability to perform at low wind speeds, and potent standalone system. The available experimental data on Savonius wind rotor have been used to train the ANN and GEP using matlab r2020b and genexprotools 5.0 software, respectively. The input variables used in ANN and GEP architecture include newly proposed design shape factors, number of blades and stages, gap and overlap lengths, height and diameter of the rotor, freestream velocity, end plate diameter, and tip speed ratio besides the cross-sectional area of the wind tunnel test section. Based on this, the unknown governing function constituted by the aforementioned input variables is established using ANN and GEP to approximate/forecast the rotor performance as an output. The governing equation formulated by ANN is in the form of weights and biases, while GEP provides it in the form of traditional mathematical functions. The trained ANN and GEP are capable to estimate the rotor performance with R-2 approximate to 0.97 and R-2 approximate to 0.65, respectively, in correlation with the reported experimental rotor performance.