Citation: Mao, W.; Wei, B.; Xu, X.;
Chen, L.; Wu, T.; Peng, Z.; Ren, C.
Fault Diagnosis for Power
Transformers through
Semi-Supervised Transfer Learning.
Sensors 2022, 22, 4470. https://
doi.org/10.3390/s22124470
Academic Editors: Kelvin K. L. Wong,
Dhanjoo N. Ghista, Andrew W. H. Ip
and Wenjun (Chris) Zhang
Received: 27 April 2022
Accepted: 9 June 2022
Published: 13 June 2022
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
iations.
Copyright: © 2022 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
Article
Fault Diagnosis for Power Transformers through
Semi-Supervised Transfer Learning
Weiyun Mao
1
, Bengang Wei
2,
* , Xiangyi Xu
1
, Lu Chen
1
, Tianyi Wu
1
, Zhengrui Peng
1
and Chen Ren
1
1
State Grid Shanghai Electric Power Research Institute, Shanghai 200437, China;
maoweiyun1991@163.com (W.M.); xuxiangyi_1986@126.com (X.X.); chenlu8766@163.com (L.C.);
wtyfenghua@163.com (T.W.); skanson@163.com (Z.P.); 18901658588@163.com (C.R.)
2
State Grid Shanghai Municipal Electric Power Company, Shanghai 200122, China
* Correspondence: wbgsj@126.com
Abstract:
The fault diagnosis of power transformers is a challenging problem. The massive multi-
source fault is heterogeneous, the type of fault is undetermined sometimes, and one device has only
met a few kinds of faults in the past. We propose a fault diagnosis method based on deep neural
networks and a semi-supervised transfer learning framework called adaptive reinforcement (AR)
to solve the above limitations. The innovation of this framework consists of its enhancement of the
consistency regularization algorithm. The experiments were conducted on real-world 110 kV power
transformers’ three-phase fault grounding currents of the iron cores from various devices with four
types of faults: Phases A, B, C and ABC to ground. We trained the model on the source domain and
then transferred the model to the target domain, which included the unbalanced and undefined fault
datasets. The results show that our proposed model reaches over 95% accuracy in classifying the
type of fault and outperforms other popular networks. Our AR framework fits target devices’ fault
data with fewer dozen epochs than other novel semi-supervised techniques. Combining the deep
neural network and the AR framework helps diagnose the power transformers, which lack diagnosis
knowledge, with much less training time and reliable accuracy.
Keywords:
semi-supervised transfer learning; fault type diagnosis of power transformers; three-
phase grounding current of the iron core; deep neural network
1. Introduction
This section illustrates the research’s motivation, target and framework, followed by a rel-
evant literature review of the recent research about the fault diagnosis of power transformers.
After that, we state the research gap, our contributions and this paper’s organisation.
1.1. Motivation, Target and Framework
Power transformers are important equipment in electrical power systems, and trans-
former failures will negatively impact whole systems. An early-stage diagnosis of trans-
former fault type can save the high cost of repairing and downtime. The three-phase
grounding currents of the iron core generated during the turning on of transformers are
important indexes to assess the states of power transformers. The value of the normal
grounding current of the iron core is quite small, while the abnormal grounding current
value will increase substantially. The warning threshold for the grounding current of the
iron core is 100 mA for the 110 kV power transformers and the faults are caused by Phases
A, B, C and ABC to ground fault. This research aims to construct a reliable model that
can analyse the fault type after 10 min from fault occurrence based on the abnormal three-
phase grounding current data of the iron core, and the model can be quickly transferred to
different devices with quick retraining to obtain good performance.
Benefiting from the evolution of machine learning, more and more supervised or
unsupervised models have been adopted to pinpoint the linear or nonlinear relationships
Sensors 2022, 22, 4470. https://doi.org/10.3390/s22124470 https://www.mdpi.com/journal/sensors
