Research on Cross-Domain Power Data Fusion and In-Depth Analysis Based on Multimodal Model

View at publisher Summarise with VIRAT
Authors: Zhanjie Liu, Xiaonan Gao, Jiaqi Yuan
Cited by: 0
References: 9

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

A fault classification method based on CNN-LSTM-Attention hybrid neural network is proposed in this paper to meet the challenge of multi-modal data fusion and parsing of power grid equipment. By integrating multi-source data such as time series, image and text, the model realizes the collaborative diagnosis of four kinds of faults, namely, sudden change of electrical quantity, deterioration of insulation medium, abnormal mechanical and thermodynamic, and coupling of information physics. The experimental results show that the overall accuracy of the fusion framework is 0.942, and the macro average$F 1$score is 0.93, which is significantly better than single modality and the early fusion baseline. In particular, the F1 score of 0.91 is achieved in the cyber-physical coupled fault diagnosis, which verifies the effectiveness of cross-modal dynamic feature alignment and collaborative learning. This study provides a reliable technical path for intelligent diagnosis of complex power grid faults, and further promotes the precision and automation of operation and maintenance decision-making.

Fields of Study & Subjects

Power Systems Fault Detection Machine Fault Diagnosis Techniques Power Transformer Diagnostics and Insulation

Publication details

Journal / source
International Conference on Electrical Automation and Artificial Intelligence (ICEAAI)
Pages
648–652
Publication date
16 January 2026
Language
EN