Scholarlink

<p>This study develops a three-stage probabilistic framework to propagate wind forecast uncertainty into available transfer capability (ATC) assessment. As a case study, the framework is applied to a 21 MW wind farm on Jeju Island, South Korea. In stage 1, a random forest model corrects numerical weather prediction wind-speed forecasts, reducing the mean absolute error from 1.78 to 1.25 m/s. In stage 2, gradient boosted quantile regression generates 19 wind power quantiles (0.05–0.95), where the corrected wind-speed input decreases the continuous ranked probability score from 1.43 to 1.13 and the average absolute coverage deviation from 10.22 to 3.79 percentage points. In stage 3, the probabilistic wind power trajectories are translated into ATC profiles, yielding a 90 % prediction-interval bandwidth ranging from 0.00 to 7.63 MW. The proposed framework supports the selection of conservative, median, or high-confidence transfer margins and provides a practical basis for risk-informed operation under variable wind conditions.</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

Probabilistic wind power forecasting using gradient boosted quantile regression for power system uncertainty analysis

  • 期刊论文

  • 发表日期：

    2026年7月10日

查看全文
  • Yoojeong Koh,
  • Jin Hur

摘要

This study develops a three-stage probabilistic framework to propagate wind forecast uncertainty into available transfer capability (ATC) assessment. As a case study, the framework is applied to a 21 MW wind farm on Jeju Island, South Korea. In stage 1, a random forest model corrects numerical weather prediction wind-speed forecasts, reducing the mean absolute error from 1.78 to 1.25 m/s. In stage 2, gradient boosted quantile regression generates 19 wind power quantiles (0.05–0.95), where the corrected wind-speed input decreases the continuous ranked probability score from 1.43 to 1.13 and the average absolute coverage deviation from 10.22 to 3.79 percentage points. In stage 3, the probabilistic wind power trajectories are translated into ATC profiles, yielding a 90 % prediction-interval bandwidth ranging from 0.00 to 7.63 MW. The proposed framework supports the selection of conservative, median, or high-confidence transfer margins and provides a practical basis for risk-informed operation under variable wind conditions.