Power equipment multi-modal fusion anomaly identification method, system, device and medium

By adopting a multimodal fusion anomaly identification method, the shortcomings of single-modal monitoring in power equipment anomaly identification are solved. It realizes unified access and unified alarm of multimodal data, improves the accuracy and stability of anomaly identification, and adapts to complex application scenarios.

CN122365006APending Publication Date: 2026-07-10CHANGZHOU YIZHI TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
CHANGZHOU YIZHI TECHNOLOGY CO LTD
Filing Date
2026-04-16
Publication Date
2026-07-10

AI Technical Summary

Technical Problem

Existing methods for identifying anomalies in power equipment rely on single-modal monitoring, which is easily affected by noise interference and environmental changes, resulting in high false alarm rates and high false alarm rates. Furthermore, fixed thresholds are difficult to adapt to different equipment and scenarios, leading to threshold failure or frequent false alarms.

Method used

A multimodal fusion anomaly identification method is adopted. By acquiring electrical quantity time series data and auxiliary modal data (such as visible light images, video frames, audio data, and infrared thermal imaging data), and after standardization processing, the method utilizes modal existence masking, attention mechanism, and adaptive threshold to achieve multimodal feature fusion and anomaly determination.

Benefits of technology

It enables unified access, unified modeling, and unified alarm for multimodal data of power equipment, adapting to modal missingness, data heterogeneity, and threshold drift in complex application scenarios, improving the accuracy and stability of anomaly identification, and reducing false alarms and missed alarms.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122365006A_ABST
    Figure CN122365006A_ABST
Patent Text Reader

Abstract

This invention relates to the field of anomaly detection technology, and in particular to a method, system, device, and medium for multimodal fusion anomaly identification of power equipment. The method includes: acquiring monitoring data of the target power equipment and associating the monitoring data with equipment identifiers and timestamps; the monitoring data includes at least electrical quantity time series data and at least one auxiliary modal data; standardizing the monitoring data to construct standardized sample data and writing the standardized sample data into the GoLD dataset, while simultaneously establishing an index file for associating sample identifiers, modal paths, labels, and timestamps; constructing a modal existence mask based on the existence of each modal data, the modal existence mask being used to characterize whether the electrical quantity modality and each auxiliary modality are valid; inputting the electrical quantity time series data into an electrical quantity encoder, inputting the existing auxiliary modal data into the corresponding modal encoders respectively, fusing the output features of each modality to obtain fused features; outputting anomaly probability based on the fused features, and / or outputting anomaly score based on reconstruction error, and determining whether the current monitoring data is abnormal based on a threshold.
Need to check novelty before this filing date? Find Prior Art