Scholarlink

<p>To address the challenges of identifying small leaks in buried pipelines and the low detection accuracy of infrasound leak signals, a multi-feature collaborative leakage discrimination framework is constructed signals. First, 12-dimensional leak features are extracted, and a Mutual Information optimized Random Forest (MI-RF) is introduced to perform preliminary feature screening. On this basis, a Full Subset Optimal Search—Mahalanobis Distance Adaptive Discrimination Mechanism (MDFSS) is proposed to establish a leak classification and discrimination framework. The experimental results show that for sudden leakage, the optimal feature combination is Peak-to-Peak Value and Mean Frequency, achieving an average identification Accuracy of 96.33% across different apertures; for persistent leakage, the optimal feature combination is Root Mean Square and Standard Deviation of Frequency, achieving an average identification accuracy of 94.17%. Ablation experiments verify the advantages of each stage of the MI-RF-MDFSS framework. Under the typical 2&#xa0;mm persistent leakage condition, the proposed method, relying on the optimal feature combination, achieves a leak identification accuracy of 96.5%, outperforming traditional discrimination methods whose accuracy ranges only between 92 and 95%. Future work will further validate the generalization capability and engineering applicability of this method in complex industrial noise environments.</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

Research on Infrasound-based Pipeline Leakage Signal Identification Using Random Forest

  • 期刊论文

  • 发表日期：

    2026年6月30日

查看全文
  • Wenjing Du,
  • Yongmei Hao,
  • Mudan Yang,
  • Jun Shen,
  • Fei Li

摘要

To address the challenges of identifying small leaks in buried pipelines and the low detection accuracy of infrasound leak signals, a multi-feature collaborative leakage discrimination framework is constructed signals. First, 12-dimensional leak features are extracted, and a Mutual Information optimized Random Forest (MI-RF) is introduced to perform preliminary feature screening. On this basis, a Full Subset Optimal Search—Mahalanobis Distance Adaptive Discrimination Mechanism (MDFSS) is proposed to establish a leak classification and discrimination framework. The experimental results show that for sudden leakage, the optimal feature combination is Peak-to-Peak Value and Mean Frequency, achieving an average identification Accuracy of 96.33% across different apertures; for persistent leakage, the optimal feature combination is Root Mean Square and Standard Deviation of Frequency, achieving an average identification accuracy of 94.17%. Ablation experiments verify the advantages of each stage of the MI-RF-MDFSS framework. Under the typical 2 mm persistent leakage condition, the proposed method, relying on the optimal feature combination, achieves a leak identification accuracy of 96.5%, outperforming traditional discrimination methods whose accuracy ranges only between 92 and 95%. Future work will further validate the generalization capability and engineering applicability of this method in complex industrial noise environments.