Scholarlink

<p>Impact of data scaling methods on regression performance of machine learning algorithms is analyzed for predicting wind turbine output power. The analysis encompasses ten distinct data scaling techniques employed during the data preprocessing phase: Standard Scaler (SS), Min Max Scaler (MMS), Robust Scaler (RS), Max Absolute Scaler (MAS), Normalizer (N), Quantile Transformer (QT), Power Transformer (PT), Pareto Scaling (PS), Mean Centered (MC), and Decimal Scaling Normalization (DSN). Following preprocessing, seven regression algorithms including Linear Regression (LR), Support Vector Regression (SVR), K-Nearest Neighbors (KNN), Elastic Net (EN), Decision Tree (DT), Adaptive Boosting (AdaBoost), and Extreme Gradient Boosting (XGBoost) are implemented. These regression models are combined with data scaling methods to forecast the output power of wind turbine. To ensure the generalizability and robustness of the findings, three diverse datasets including the Texas Wind Turbine Dataset, the Wind Turbine SCADA Dataset, and the Wind Power Forecasting Dataset are utilized. This is the first study to systematically compare ten scaling methods across three diverse wind datasets, which highlights the novelty and comprehensive scope of the analysis compared to previous works that typically focused on fewer scaling methods or single datasets. The XGBoost model with DSN scaling achieves the best performance on the Texas Wind Turbine Dataset (R² improved from 0.9996 to 0.999964, RMSE reduced by 99.98%), while the XGBoost model with N scaling delivers the best results on both the Wind Turbine SCADA Dataset (R² improved from 0.991633 to 0.994031, RMSE reduced by 99.35%) and the Wind Power Forecasting Dataset (R² improved from 0.999 to 0.999428, RMSE reduced by 99.97%). The results reveal that data scaling significantly influences regression accuracy, and its effect varies across datasets and models. The findings highlight the comprehensive evaluation of 10 scaling techniques, 7 regression models, and 3 datasets, offering new insights into the interaction between preprocessing methods and machine learning performance. The results underscore the critical role of data scaling in optimizing machine learning models for wind energy prediction.</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

Comparative Evaluation of Data Scaling Techniques for Wind Turbine Power Prediction Using Machine Learning

  • 期刊论文

  • 发表日期：

    2025年12月18日

查看全文
  • Kadir Ileri

摘要

Impact of data scaling methods on regression performance of machine learning algorithms is analyzed for predicting wind turbine output power. The analysis encompasses ten distinct data scaling techniques employed during the data preprocessing phase: Standard Scaler (SS), Min Max Scaler (MMS), Robust Scaler (RS), Max Absolute Scaler (MAS), Normalizer (N), Quantile Transformer (QT), Power Transformer (PT), Pareto Scaling (PS), Mean Centered (MC), and Decimal Scaling Normalization (DSN). Following preprocessing, seven regression algorithms including Linear Regression (LR), Support Vector Regression (SVR), K-Nearest Neighbors (KNN), Elastic Net (EN), Decision Tree (DT), Adaptive Boosting (AdaBoost), and Extreme Gradient Boosting (XGBoost) are implemented. These regression models are combined with data scaling methods to forecast the output power of wind turbine. To ensure the generalizability and robustness of the findings, three diverse datasets including the Texas Wind Turbine Dataset, the Wind Turbine SCADA Dataset, and the Wind Power Forecasting Dataset are utilized. This is the first study to systematically compare ten scaling methods across three diverse wind datasets, which highlights the novelty and comprehensive scope of the analysis compared to previous works that typically focused on fewer scaling methods or single datasets. The XGBoost model with DSN scaling achieves the best performance on the Texas Wind Turbine Dataset (R² improved from 0.9996 to 0.999964, RMSE reduced by 99.98%), while the XGBoost model with N scaling delivers the best results on both the Wind Turbine SCADA Dataset (R² improved from 0.991633 to 0.994031, RMSE reduced by 99.35%) and the Wind Power Forecasting Dataset (R² improved from 0.999 to 0.999428, RMSE reduced by 99.97%). The results reveal that data scaling significantly influences regression accuracy, and its effect varies across datasets and models. The findings highlight the comprehensive evaluation of 10 scaling techniques, 7 regression models, and 3 datasets, offering new insights into the interaction between preprocessing methods and machine learning performance. The results underscore the critical role of data scaling in optimizing machine learning models for wind energy prediction.