Method for Error State Assessment of Current Transformers Based on Multimodal Global-Local Fusion Network

作者：Shaojun Wu, Banghai Yu, Songhui Zhang, Haoran Sui, Changxiao Zhang, Chunhui Wang, Yekui Tan · 年份：2026 · DOI：10.1109/iceaai68945.2026.11442599 · 研究领域：Machine Fault Diagnosis Techniques、Power Systems Fault Detection、Electrical Fault Detection and Protection

In power system safe operation, error state assessment of current transformers (CTs) is crucial for accurate measurement and reliable protection. Addressing limitations in existing research-foreign single-modal analysis fails to capture error correlation characteristics, while domestic fusion methods lack synchronous extraction of global and local features, leading to insufficient assessment accuracy under complex conditions-this paper proposes an error state assessment method based on a Multimodal Global-Local Fusion Network (MG-LFNet). First, one-dimensional current time-series data is converted into multimodal two-dimensional image data via GAF, MTF, and RPM algorithms. Then, global operation rules and local defect features are synchronously extracted, followed by accurate error state assessment through multimodal feature fusion and an SVM classifier. Experimental results show the method achieves 99.67 % accuracy and 0.998 AUC value; even under 5 dB high-noise environment, accuracy remains 95.2 % with a standard deviation of only 0.21 %. Its comprehensive performance outperforms mainstream methods, meeting real-time online assessment needs of CT error states under complex working conditions.