Scholarlink

<p>Due to environmental and equipment-related factors, wind farm data acquisition systems often contain abnormal and missing data, which adversely affects the accuracy of wind power prediction. To address this issue, a wind farm anomaly data processing method based on Daubechies wavelet decomposition, DBSCAN, and VAE is proposed. By analyzing the distribution characteristics of wind speed-power data, anomalies are categorized into clustered and dispersed types. Wavelet decomposition and DBSCAN are then employed to identify and clean the anomalous data. During the reconstruction phase, VAE is utilized to generate smooth and complete power data, filling in gaps created by erroneous deletions and missing values. Case study results demonstrate that, in anomaly detection, the accuracy rate reaches 89.81%, with a misjudgment rate of 2.41%; in power reconstruction, the reconstruction error is 1.15%, with an accuracy rate of 98.25%, outperforming other comparative algorithms. Additionally, wind power prediction using a BiLSTM model on data processed by this method shows an accuracy improvement of 2.28% over raw data.</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

Anomalous Data Processing for Wind Farms Based on Daubechies Wavelet, DBSCAN, and VAE

  • 期刊论文

  • 发表日期：

    2025年8月1日

查看全文
  • Xunwen Su,
  • Hangyuan Wei,
  • Jiahao Geng,
  • Yu Han,
  • Zijing Wu,
  • Siqi Zhang

摘要

Due to environmental and equipment-related factors, wind farm data acquisition systems often contain abnormal and missing data, which adversely affects the accuracy of wind power prediction. To address this issue, a wind farm anomaly data processing method based on Daubechies wavelet decomposition, DBSCAN, and VAE is proposed. By analyzing the distribution characteristics of wind speed-power data, anomalies are categorized into clustered and dispersed types. Wavelet decomposition and DBSCAN are then employed to identify and clean the anomalous data. During the reconstruction phase, VAE is utilized to generate smooth and complete power data, filling in gaps created by erroneous deletions and missing values. Case study results demonstrate that, in anomaly detection, the accuracy rate reaches 89.81%, with a misjudgment rate of 2.41%; in power reconstruction, the reconstruction error is 1.15%, with an accuracy rate of 98.25%, outperforming other comparative algorithms. Additionally, wind power prediction using a BiLSTM model on data processed by this method shows an accuracy improvement of 2.28% over raw data.