An adaptive thresholding method for sample learning

By employing an adaptive thresholding method based on sample learning, utilizing kernel density estimation and weighted incremental learning, combined with sliding window and isolated forest models, the threshold is dynamically updated, solving the problem of sample distribution drift in industrial equipment and improving the accuracy and stability of fault diagnosis.

CN121743718BActive Publication Date: 2026-11-10MCC5 GROUP SHANGHAI CORPORATION LIMITED
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202511774268.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2025-11-28
Publication Date
2026-11-10
Estimated Expiration
2045-11-28

AI Technical Summary

Technical Problem

Existing technologies struggle to capture real-time sample distribution drift in industrial equipment, leading to decreased fault diagnosis accuracy and potentially causing production accidents.

Method used

An adaptive thresholding method based on sample learning is adopted. Distribution drift is detected by kernel density estimation. The threshold is dynamically updated by combining weighted incremental learning and an isolated forest model. The sliding window and exponential weight decay mechanism are integrated to achieve real-time adaptation to sample distribution and filtering of abnormal samples.

Benefits of technology

It achieves accurate detection and dynamic response to sample distribution drift, improves the long-term reliability and anti-interference capability of fault diagnosis, and meets the real-time requirements of industrial sites.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121743718B_ABST
    Figure CN121743718B_ABST
Patent Text Reader

Abstract

This invention discloses an adaptive thresholding method based on sample learning, comprising five steps: sample feature extraction and distribution representation, sample distribution drift detection, threshold correction based on weighted incremental learning, anomaly sample filtering mechanism, and threshold update triggering and execution. It collects equipment operation data in real time and extracts features, uses kernel density estimation combined with JS distance to detect sample distribution drift, implements sample weighting using a sliding window and exponential weight decay mechanism, calculates a dynamic threshold based on the confidence interval of a weighted regression model, and filters out anomaly samples using an isolated forest. This invention can adapt to the dynamic changes in sample distribution in industrial scenarios in real time, improving the long-term accuracy and robustness of fault diagnosis. It has strong engineering feasibility and meets the real-time requirements of industrial sites.
Need to check novelty before this filing date? Find Prior Art