Multimodal feature fusion method for identifying excitation inrush current in distribution transformer

Authors

• Fan Yan, State Key Laboratory of Disaster Prevention and Reduction for Power Grid ， Changsha University of Science & Technology ， Changsha 410114 ， China
• Luxin Zhan, Jiangmen Power Supply Bureau ， Guangdong Power Grid Co .， Ltd.， Jiangmen 529100 ， China
• Chun Chen, State Key Laboratory of Disaster Prevention and Reduction for Power Grid ， Changsha University of Science & Technology ， Changsha 410114 ， ChinaFollow
• Yi An, Electric Power Research Institute of State Grid Jiangxi Electric Power Co .， Ltd.， Nanchang 330096 ， China
• Yijia Cao, State Key Laboratory of Disaster Prevention and Reduction for Power Grid ， Changsha University of Science & Technology ， Changsha 410114 ， China

Keywords

DG, multimodal feature fusion, GAF, excitation inrush current identification, resnet 18, MLP

Abstract

With the widespread integration of high-penetration distributed generation （DG） and power electronic devices，the harmonic components of fault currents have become increasingly prominent，preventing traditional relay protection methods from effectively distinguishing between excitation inrush currents and fault currents.Moreover，existing time-frequency analysis methods rely on manually selected evaluation metrics，which often lead to misjudgments.To address these issues，this paper proposes a deep learning-based identification method that fuses image- and statistical-domain features，in which the neural network adaptively learns features for identification.Specifically，the Gramian angular field （GAF ） method is employed to transform time-series data into images，from which image features are extracted using ResNet 18.Then，statistical-domain features of the fault current are extracted，and the two types of features are fused through low-rank multimodal fusion （LMF ）.Furthermore，a multilayer perceptron （MLP ） is used to train the fused multimodal features，thereby efficiently identify fault currents and excitation inrush currents.Finally，simulation experiments are conducted to verify the feasibility and effectiveness of the proposed method.The results indicate that the proposed method achieves an identification accuracy of up to 99.72% and can be further applied to classify fault currents.

DOI

10.19781/j.issn.1673-9140.2026.04.028

First Page

340

Last Page

352

Recommended Citation

Yan, Fan; Zhan, Luxin; Chen, Chun; An, Yi; and Cao, Yijia (2026) "Multimodal feature fusion method for identifying excitation inrush current in distribution transformer," Journal of Electric Power Science and Technology: Vol. 41: Iss. 4, Article 28.
DOI: 10.19781/j.issn.1673-9140.2026.04.028
Available at: https://jepst.researchcommons.org/journal/vol41/iss4/28