Scholarlink

<p>Ensuring the reliable operation of Sucker Rod Pump Systems (SRPS) is of paramount importance in the petroleum industry, demanding effective methods for the early detection of slow-developing secondary faults. The conventional methodology focuses on multi-fault classification and feature extraction based on the mechanistic model, which highly depends on the labelled dateset and sufficient mechanistic information. This paper proposes an unsupervised end-to-end learning algorithm designed for SRPS anomaly detection, denoted Anomaly Detection with Domain-specific Shapelet Learning algorithm (AD-DSL). The AD-DSL utilizes a mechanistic information matrix and introduces a sparsity-promoting objective function, enabling Shapelet-based features to learn from motor power time-series data interpretably. With a dynamic threshold and defined anomaly scores, AD-DSL monitors the variation trend of the SRPS for anomaly detection. The proposed method provides early warnings of potential issues for decision-makers. The robustness and effectiveness of the proposed method are demonstrated through quantitative comparison with baseline methods, where AD-DSL outperforms in accuracy and delivers competitive F1 scores.</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

Anomaly detection with domain specific shapelet learning for sucker rod pump system

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年7月18日

查看全文
  • Xiangyu Li,
  • Zhupei Liao,
  • Chunhua Yuan

摘要

Ensuring the reliable operation of Sucker Rod Pump Systems (SRPS) is of paramount importance in the petroleum industry, demanding effective methods for the early detection of slow-developing secondary faults. The conventional methodology focuses on multi-fault classification and feature extraction based on the mechanistic model, which highly depends on the labelled dateset and sufficient mechanistic information. This paper proposes an unsupervised end-to-end learning algorithm designed for SRPS anomaly detection, denoted Anomaly Detection with Domain-specific Shapelet Learning algorithm (AD-DSL). The AD-DSL utilizes a mechanistic information matrix and introduces a sparsity-promoting objective function, enabling Shapelet-based features to learn from motor power time-series data interpretably. With a dynamic threshold and defined anomaly scores, AD-DSL monitors the variation trend of the SRPS for anomaly detection. The proposed method provides early warnings of potential issues for decision-makers. The robustness and effectiveness of the proposed method are demonstrated through quantitative comparison with baseline methods, where AD-DSL outperforms in accuracy and delivers competitive F1 scores.