A hybrid method enabling predictive digital twin during the design phase of wind turbines

One of the main challenges in the design phase of wind turbines is the estimation of power production in the real environment. Although there are some computer-aided simulations, wind tunnel experimental data, and empirical formulas, most of them lack accurate forecasts. In this study, a novel hybrid model, which is a part of a pre-deployment predictive digital twin, is proposed using statistical techniques and machine learning models for accurate and fast forecasting in the design phase of a wind turbine. The statistical model captures the primary dynamics of the wind turbine power curve, while the machine learning model eliminates errors in the model. A logistic function is used as the statistical model for the regression problem, and an artificial neural network with two hidden layers is used for the error model. The model is trained with the dataset available online. The trained model is executed within the digital twin platform, which is built on Unity 3D, enabling the acquisition of both visual and analytical results. The proposed model exhibits 8 percent less root mean square error and 17 percent less mean absolute error compared to the pure statistical technique approach.