Soft Computing Approaches for Wind Turbine Power Prediction: A Study using Gradient Boosting Trees and LightGBM

This research explores wind turbine power prediction using advanced machine learning algorithms, specifically Gradient Boosting Trees (GBT) Regressor and LightGBM. The dataset, comprising 50,530 records, includes wind speed, wind direction, month, and hour as input features and compares the predicted power output with theoretical and actual power measurements. Statistical analysis of the dataset revealed key insights into the distribution and relationships of features, which were visualized through Kernel Density Estimate (KDE) plots and a correlation heatmap. Wind speed showed a strong positive correlation (0.93) with both theoretical and actual power outputs, confirming its critical role in turbine performance. Model evaluation demonstrated that LightGBM outperformed the GBT Regressor in predictive accuracy. LightGBM achieved an R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> score of 0.983 and a lower Mean Absolute Error (MAE) of 71.23 kW compared to GBT’s R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> score of 0.981 and MAE of 83.68 kW. LightGBM’s predictions closely aligned with the theoretical power curve, especially at higher wind speeds, with minimal deviations. The results highlight LightGBM’s superior performance and its potential to optimize wind turbine operations by accurately predicting power output across various conditions.