Method for recognizing fixed-point deformation abnormal signal by fusing multi-scale entropy and attention weight

By integrating multi-scale entropy and attention weights, high-precision identification and adaptive learning of fixed-point deformation anomaly signals were achieved, solving the problem of low identification accuracy in existing technologies and improving the adaptability and interpretability of the model.

CN122413243APending Publication Date: 2026-07-17HUBEI EARTHQUAKE ADMINISTRATION (SEISMOLOGY RES INST OF CHINA EARTHQUAKE ADMINISTRATION)
View PDF 1 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
HUBEI EARTHQUAKE ADMINISTRATION (SEISMOLOGY RES INST OF CHINA EARTHQUAKE ADMINISTRATION)
Filing Date
2026-05-19
Publication Date
2026-07-17

AI Technical Summary

Technical Problem

Existing technologies lack adaptive learning capabilities in the identification of fixed-point deformation anomaly signals, resulting in low accuracy in deformation data identification, especially with limited generalization ability under different geological structures, noise environments, and different types of deformation data.

Method used

This paper employs a method that integrates multi-scale entropy and attention weights. Through multi-scale coarsening, amplitude-weighted permutation entropy calculation, feature attention filtering, and ensemble model training, it achieves adaptive learning and high-precision recognition. Specific steps include: multi-scale coarsening, amplitude-weighted permutation entropy integration, feature attention filtering, and ensemble model training. Key features are selected using attention weights and combined into a decision tree model.

Benefits of technology

It improves the recognition accuracy and generalization ability of fixed-point deformation anomaly signals, can automatically identify key features in different types of deformation data, suppress noise, adapt to multiple deformation modes, improve the adaptability and interpretability of the model, and reduce computational complexity.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122413243A_ABST
    Figure CN122413243A_ABST
Patent Text Reader

Abstract

融合多尺度熵和注意力权重的定点形变异常信号识别方法，先依据多个定点形变异常信号数据得到多个幅值加权排列熵向量，再将所有向量整合为标准化特征矩阵；然后将编码标签定义为输出，再依据标准化矩阵通过注意力权重机制筛选得到有效特征集，有效特征集定义为输入，然后设置集成模型的结构与参数，在每轮迭代中，依据有效特征集训练单棵决策树，最终由多颗单棵决策树组合得到集成模型；将待识别的定点形变异常信号数据处理为待识别的有效特征集，然后输入集成模型，再得到定点形变异常的种类；注意力权重机制筛选得到有效特征，以使模型具备自适应学习能力，提升分类精度。因此，本设计具备自适应学习能力，且对于形变数据的识别精度较高。
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Earthquake prediction method and device, electronic equipment and storage medium

    CN117741744A