Scholarlink

<p>Wind energy is a key pillar of the global transition toward clean and low-carbon power, and improving wind turbine performance using existing infrastructure has become increasingly important. With the widespread deployment of SCADA systems and the availability of weather data, wind farms now generate large volumes of operational information that can be used for data-driven optimization. However, most studies up to 2025 focus mainly on improving wind power prediction accuracy using machine learning models that behave as black boxes and rely on correlation rather than true cause-effect relationships. This limits their usefulness for real operational decision-making, as they cannot reliably identify which turbine parameters genuinely drive power improvements. To address this gap, this study proposes a hybrid explainable causal machine learning framework for wind turbine power maximization. The framework combines random forest and XGBoost models to achieve accurate power prediction, while LIME is used to explain model behavior and identify influential operational and environmental features. Causal machine learning techniques are then applied to remove confounding effects and estimate true causal impacts using average treatment effect (ATE) and conditional average treatment effect (CATE). The suggested method performs consistently under actual operating circumstances, as shown by validation over various wind speed ranges. The hybrid model has a high prediction accuracy of 93.20%, which shows that it is good at picking out the important things that happen in the way turbines work. More significantly, causal optimization produces a mean power increase of 14.937 kW, which is an 11.41% increase over pre-treatment levels. This clearly shows that combining explainability and causal reasoning with good prediction not only builds model trust but also provides useful, practical, actionable changes in wind turbine efficiency. The proposed framework offers a transparent and practical decision-support tool for wind farm operators, delivering measurable energy gains without additional hardware investment and supporting long-term renewable energy and sustainability goals.</p> Graphical abstract <p></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

Optimizing wind energy production using explainable AI and causal machine learning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月22日

查看全文
  • Priyadharshini Rengasamy,
  • R. Rajesh

摘要

Wind energy is a key pillar of the global transition toward clean and low-carbon power, and improving wind turbine performance using existing infrastructure has become increasingly important. With the widespread deployment of SCADA systems and the availability of weather data, wind farms now generate large volumes of operational information that can be used for data-driven optimization. However, most studies up to 2025 focus mainly on improving wind power prediction accuracy using machine learning models that behave as black boxes and rely on correlation rather than true cause-effect relationships. This limits their usefulness for real operational decision-making, as they cannot reliably identify which turbine parameters genuinely drive power improvements. To address this gap, this study proposes a hybrid explainable causal machine learning framework for wind turbine power maximization. The framework combines random forest and XGBoost models to achieve accurate power prediction, while LIME is used to explain model behavior and identify influential operational and environmental features. Causal machine learning techniques are then applied to remove confounding effects and estimate true causal impacts using average treatment effect (ATE) and conditional average treatment effect (CATE). The suggested method performs consistently under actual operating circumstances, as shown by validation over various wind speed ranges. The hybrid model has a high prediction accuracy of 93.20%, which shows that it is good at picking out the important things that happen in the way turbines work. More significantly, causal optimization produces a mean power increase of 14.937 kW, which is an 11.41% increase over pre-treatment levels. This clearly shows that combining explainability and causal reasoning with good prediction not only builds model trust but also provides useful, practical, actionable changes in wind turbine efficiency. The proposed framework offers a transparent and practical decision-support tool for wind farm operators, delivering measurable energy gains without additional hardware investment and supporting long-term renewable energy and sustainability goals.

Graphical abstract