Advanced power curve modeling for wind turbines: A multivariable approach with SGBRT and grey wolf optimization

Abstract

Accurate power curve modeling is crucial for improving the operational efficiency and performance of grid-connected wind turbines (WTs). To enhance the modeling quality and eliminate input variable interactions, this paper proposes a novel multivariable power curve prediction approach that integrates advanced machine learning techniques, namely stochastic gradient boosting regression tree (SGBRT) and grey wolf optimization (GWO), with innovative data preprocessing and feature selection methods. The specific works and novelties are as follows. 1) The raw data is cleaned in a two-dimensional Copula space, using wind wheel speed as an auxiliary criterion and a probabilistic description, to handle data uncertainties and nonlinear dependencies. 2) A partial mutual information (PMI) method is presented for data characteristics analysis, based on which eight significant parameters are selected as modeling input variables, reducing computational complexity while enhancing prediction accuracy. 3) A power curve prediction model considering multiple input variables is established using SGBRT, and its hyperparameters are optimized through a GWO algorithm, guided by a fitness function combining the indicators of root mean square error (RMSE), mean absolute error (MAE) and R squared (R²). 4) Validated with real SCADA data from WTs in service, the proposed model achieves superior performance, with the smallest standardized residuals (6.56 %), RMSE (around 27 kW), MAE (19.27 kW), and superior average R² (98.61 %) for all speed regions. Comparative studies indicate that the proposed approach outperforms existing methods, offering significant improvements in accuracy, efficiency, robustness and adaptability for WT power curve modeling.