Scholarlink

<p>Wind power has emerged as an essential element of worldwide renewable energy systems; yet, the substantial aberrant data within wind turbine SCADA systems present considerable hurdles to grid stability and operational efficiency, adversely impacting power forecasting precision and grid scheduling dependability. This research presents a multi-variable wind power anomaly correction technique utilizing a physics-aware transformer. The method develops an adaptive temporal encoding mechanism to identify multi-scale periodic patterns and formulates a physics-informed composite loss function that incorporates reconstruction accuracy, temporal smoothness, and capacity constraints to ensure that correction outcomes adhere to the physical laws governing wind power systems. Validation using actual SCADA data from a 280&#xa0;MW offshore wind farm in Fujian indicates that the suggested method realizes a 3.04% enhancement in MAE and a 0.68% enhancement in RMSE relative to the optimal baseline method, XGBoost, in anomaly correction tasks. LSTM prediction models trained with corrected data provide a 10.2% improvement in MAE and a 20.8% enhancement in R2 for 24-step long-term forecasting compared to the deletion technique, thereby greatly enhancing the accuracy and stability of wind power forecasting. The research findings offer a viable technical method for managing wind power data quality and improving forecasting accuracy.</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

Multi-variable wind power anomaly correction using Physics-Aware Transformer with adaptive temporal encoding

  • 期刊论文

  • 发表日期：

    2026年2月28日

查看全文
  • Yanjuan Tang,
  • Zuyun Luo

摘要

Wind power has emerged as an essential element of worldwide renewable energy systems; yet, the substantial aberrant data within wind turbine SCADA systems present considerable hurdles to grid stability and operational efficiency, adversely impacting power forecasting precision and grid scheduling dependability. This research presents a multi-variable wind power anomaly correction technique utilizing a physics-aware transformer. The method develops an adaptive temporal encoding mechanism to identify multi-scale periodic patterns and formulates a physics-informed composite loss function that incorporates reconstruction accuracy, temporal smoothness, and capacity constraints to ensure that correction outcomes adhere to the physical laws governing wind power systems. Validation using actual SCADA data from a 280 MW offshore wind farm in Fujian indicates that the suggested method realizes a 3.04% enhancement in MAE and a 0.68% enhancement in RMSE relative to the optimal baseline method, XGBoost, in anomaly correction tasks. LSTM prediction models trained with corrected data provide a 10.2% improvement in MAE and a 20.8% enhancement in R2 for 24-step long-term forecasting compared to the deletion technique, thereby greatly enhancing the accuracy and stability of wind power forecasting. The research findings offer a viable technical method for managing wind power data quality and improving forecasting accuracy.