Scholarlink

<p>Wind Turbines (WT) may be continuously monitored thanks to Supervisory Control and Data Acquisition (SCADA) systems, which aid in predictive maintenance and performance optimization. Power output, temperature, humidity, wind speed, rotor and generator speed, and other operational and environmental factors are all included in the SCADA dataset used in this study. Wind speed and the theoretical power curve were found to be the most significant predictors by correlation analysis following preprocessing, which included handling missing data, eliminating redundant features, and encoding time-based variables. With the help of Grid Search Cross-Validation (GSCV), the hyperparameters of six Machine Learning (ML) models were adjusted after 75% of the data had been used for training and 25% for testing. The results show that GSCV-KNN outperforms more intricate ensemble and deep learning models, with the highest R2 of 0.976 and the lowest error metrics. Predictive performance is driven by wind speed, theoretical power curves, and temporal variables, as confirmed by feature importance and SHAP analysis. The study provides useful, understandable, and affordable insights for predictive maintenance in wind energy systems by showcasing a lightweight yet incredibly accurate modeling framework. Lightweight optimized models for SCADA-based wind turbine monitoring are still understudied despite the popularity of sophisticated ensemble and deep learning approaches. This study systematically aimed to benchmark six ML models and shows that Grid Search-optimized KNN achieves superior accuracy and robustness while preserving interpretability and computational efficiency.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Supervisory Control and Data Acquisition System for Monitoring Wind Turbine Operational Performance Using Machine Learning

  • 期刊论文

  • 发表日期：

    2026年6月16日

查看全文
  • Lanqin Yang,
  • Xiaoling Zhang,
  • Xin Xu

摘要

Wind Turbines (WT) may be continuously monitored thanks to Supervisory Control and Data Acquisition (SCADA) systems, which aid in predictive maintenance and performance optimization. Power output, temperature, humidity, wind speed, rotor and generator speed, and other operational and environmental factors are all included in the SCADA dataset used in this study. Wind speed and the theoretical power curve were found to be the most significant predictors by correlation analysis following preprocessing, which included handling missing data, eliminating redundant features, and encoding time-based variables. With the help of Grid Search Cross-Validation (GSCV), the hyperparameters of six Machine Learning (ML) models were adjusted after 75% of the data had been used for training and 25% for testing. The results show that GSCV-KNN outperforms more intricate ensemble and deep learning models, with the highest R2 of 0.976 and the lowest error metrics. Predictive performance is driven by wind speed, theoretical power curves, and temporal variables, as confirmed by feature importance and SHAP analysis. The study provides useful, understandable, and affordable insights for predictive maintenance in wind energy systems by showcasing a lightweight yet incredibly accurate modeling framework. Lightweight optimized models for SCADA-based wind turbine monitoring are still understudied despite the popularity of sophisticated ensemble and deep learning approaches. This study systematically aimed to benchmark six ML models and shows that Grid Search-optimized KNN achieves superior accuracy and robustness while preserving interpretability and computational efficiency.