Intelligent identification method for pipe string leakage of ultra-deep gas well based on time series data and time-frequency fusion BiLSTM model

CN121997035BActive Publication Date: 2026-06-19CHINA UNIV OF PETROLEUM (EAST CHINA)
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
CHINA UNIV OF PETROLEUM (EAST CHINA)
Filing Date
2026-04-10
Publication Date
2026-06-19

AI Technical Summary

Technical Problem

Existing technologies struggle to effectively identify the location and size of leaks in ultra-deep gas well production tubing, especially small leaks. Furthermore, existing models cannot fully capture transient and full-process leak characteristics, leading to difficulties in quantitative identification.

Method used

A composite deep learning model is constructed by using a BiLSTM model based on time-series data and time-frequency fusion, combined with multi-scale CNN, fast Fourier transform (FFT), bidirectional long short-term memory network (BiLSTM), adaptive multi-head attention mechanism (AMA) and gated information fusion architecture (GRN), and multi-module fusion to improve the accuracy of leak identification.

Benefits of technology

It achieves high-precision identification of the location and size of leaks in ultra-deep gas well tubing, especially sub-millimeter-level detection accuracy for small and medium-sized leaks and location prediction accuracy of about 30m, and has robustness and generalization ability in complex environments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121997035B_ABST
    Figure CN121997035B_ABST
Patent Text Reader

Abstract

This invention relates to an intelligent identification method for ultra-deep gas well tubing leaks based on a time-series data and time-frequency fusion BiLSTM model, belonging to the field of petroleum engineering technology. The method includes: Step 1: acquiring the physical features of the model input and constructing an experimental dataset; Step 2: constructing and training a multi-module fusion ensemble deep learning model; Step 3: performing intelligent identification of ultra-deep gas well tubing leaks using the trained multi-module fusion ensemble deep learning model. This invention's method can efficiently learn the input physical features and achieve high-precision regression prediction for leak location and size tasks.
Need to check novelty before this filing date? Find Prior Art