Innovative transformer neural network for wind density function estimation at different hub heights of turbine.

Accurate estimation of wind power potential is important for resource assessment to install wind turbine. Weibull distribution functions (WDF) have been widely used and it is a function of wind speed (WS). With Turbine hub height WS get changes and it form complex nonlinear equations with WDF. To compute this paper introduces an innovative Transformer Neural Network (TNN) model for WDE estimation leverage self attention mechanism to capture complex pattern. For this wind power potential (WPP) of a site located in the Northeastern India is selected. The novelty is that WPP is performed up to 80 m height and not up to a 150 m height for North-Eastern India. Cubic Factor (CF) method is used for the evaluation of Weibull parameters, i.e., scale 'c' and shape 'k'. CF and k are independent of height and are found to be 1.96 and 1.95, respectively. The scale varies from 2.65 to 3.90 from height 10 m to 150 m. TNN performance is evaluated by Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), R-squared (R²), Mean Bias Error (MBE), and Mean Absolute Percentage Error (MAPE). For the Cumulative Distribution Function (CDF), the model achieved an MSE of 0.0012, RMSE of 0.0357, MAE of 0.0238, R² of 0.9170, MBE of -0.0071, and MAPE of 12.3158%. In comparison, the Weibull density functions (WDF) estimation yielded an MSE of 0.0003, RMSE of 0.0178, MAE of 0.0130, R² of 0.9039, MBE of -0.00018, and MAPE of 11.67%. The results demonstrate the Transformer model's high accuracy and robustness in estimating WDF, making it a reliable tool for assessing wind energy potential at different turbine hub heights.