A neural network-based scheme for fault diagnosis of power transformers

https://doi.org/10.1016/j.epsr.2004.11.013Get rights and content

Abstract

This research presents an artificial neural network (ANN)-based scheme for fault diagnosis of power transformers. The scheme is designed to detect the fault, estimate the faulted side, classify the fault type and identify the faulted phase.

The proposed fault diagnosis scheme (FDS) consists of three hierarchical levels. In the first level, a pre-processing of input data is performed. In the second level, there is an ANN which is designed to detect the fault and determine the faulted side if any. In the third level, there are two sides diagnosis systems. Each system is dedicated to one side and consists of one ANN in series with four paralleled ANNs (for fault type classification).

The proposed FDS is trained and tested using local measurements of three-phase primary voltage and primary and secondary currents. These samples are generated using EMTP simulation of the High Dam 15.75/500 kV transformer substation in Upper Egypt. All the possible fault types were simulated. The fault locations and fault incipience time were varied within each fault type. Testing results proved that the performance of the proposed ANN-based FDS is satisfactory.

Introduction

Large power transformers are considered important and precious equipment of the power system. If a power transformer experiences a fault, it is necessary to take the transformer out of service as soon as possible so that the damage is minimized. The costs associated with repairing a damaged transformer may be very high. An unplanned outage of a power transformer can cost electric utilities millions of dollars. Consequently, it is of great importance to minimize the frequency and duration of unwanted outages of power transformers. Accordingly, high demands are imposed on power transformer protective relays. The protection degree of a power transformer is assessed according to its importance and rating [1].

Fault diagnosis (FD) of power systems plays a crucial role in power system monitoring and control that ensures a stable electrical power supply to consumers. FD of power systems involves identifying the location and nature of faults occurring in the power system due to different disturbances [2]. This FD function is then the most basic fault handling, function of power system supervisory control and data acquisition (SCADA) systems. Fig. 1 shows the modules of a protective relay in which the fault diagnosis module is the most important module.

Fault type classification is an essential protective relaying feature due to its significant effect on the enhancement of relaying scheme operation. Correct operation of major protective relays may depend on the fault classification [3]. Faulted phase selection is as important as fault detection. It would lead to increase the system stability and system availability by allowing single-pole tripping. Single-pole tripping has many benefits like improving the transient stability and reliability of the power system, reducing the switching over-voltages and shaft torsional oscillations of large thermal units [4].

The conventional analytical based approaches are expected to be affected by system operating conditions and complete faulted phase selection cannot be achieved through these analytical approaches. These approaches are also time-consuming [5], [6].

The artificial neural networks (ANNs) provide a very interesting and valuable alternative because they can deal with the most complex situations which are not defined sufficiently for deterministic algorithms to execute. ANN can also handle nonlinear tasks. They are parallel data processing tools capable of learning functional dependencies of data. They are robust with respect to incorrect or missing data. Protective relaying based on ANN is not affected by change in the system operating conditions. ANNs also have high computation rates, large input error tolerance and adaptive capability [7].

Many literatures concerned with the application of the ANN-based algorithm for transformer protection have been reported in [8], [9], [10], [11]. Most of these literatures are relying on laboratory experimental transformer model. However, small size transformer usually behaves differently compared with the large power transformer during inrush and fault periods. Other literatures have formulated the training file considering R-L transients which are apparently unjustified. A contribution to the ANN-based transformer protection has been reported; however, the developed relay is directed to serve in single-phase transformers and does not extend to cover the three-phase power transformers. In all the aforementioned ANN relay efforts, there are not enough details on the rules applied for selecting the training classes, how the training patterns constitute a full coverage of the class domain and different ANN architectures. Also, the evaluation of the ANN relay sensitivity and stability boundaries has not been addressed.

In [12], a transformer model during energization and fault periods is developed using the electromagnetic transient program (EMTP). The training sets for the ANN are formulated considering various conditions including different fault classes. Different ANN structures are tested for various training strategies. Then, the most suitable ANN construction is selected for the proposed ANN relay. The response of the proposed ANN relay is measured and compared with the differential relay. It is found that the proposed relay is more efficient regarding the speed of detection, sensitivity and stability boundaries.

In this paper, an ANN-based fault diagnosis scheme (FDS) for power transformer external faults has been developed. The functions of the FDS under study are to detect the fault, estimate the faulted side and classify the fault type as well as determine the faulted phase. The required specifications of the FDS should include high reliability and fast response. Arranging the ANNs to be working in parallel make the FDS faster.

Section snippets

System under study

The EMTP [13] is used in simulating the transients of power system elements including transformers. EMTP transformer model cannot directly give winding internal fault and inrush current cases but it gives the ability to adapt the model for the transformers equivalent circuit.

The power system considered for this study is the Upper Egypt Power System (UEPS). It consists of generating stations, substations, transformers, power lines and loads. The power line from High Dam 500 kV power station

Architecture of the proposed FDS

The important module of a protective relay is the fault diagnosis module. The functions required to be performed by the proposed FDS are: fault detection, faulted side estimation, fault type classification and faulted phase selection. Since this is a multi-task problem, it is preferable to decompose it to individual sub-problems where each sub-problem is dedicated to one task. Using an ANN for only one task makes it more powerful and increase the learnability of the ANN. So, the proposed FDS

Faulted side detection (ANN #1)

ANN #1 is designed to detect the occurrence of fault and locate its side, primary or secondary fault. This network is trained based on the data included in the training set (40 cases) and tested using the data included in the test set (20 cases).

Fig. 7 shows a sample of the output testing results of ANN #1 (only eight cases are included). Each case study includes pre-fault, during fault and post-fault periods. Pre-fault as well as post-fault conditions are indicated as normal (0.1 level), while

Conclusions

The design details of an ANN-based FDS for transformer external faults have been presented. The investigation of the performance of the proposed FDS under various fault conditions leads to the following conclusions:

  • the architecture of the proposed FDS has the advantage of assigning one task to each ANN;

  • the adequate length of the data window is one-fourth cycle (four samples of three-phase voltages and primary and secondary currents);

  • the time response of the proposed FDS is fairly fast due to

References (15)

  • B. Kasztenny et al.

    Digital relays improve protection of large transformers

    IEEE Comput. Appl. Power

    (1998)
  • Y. Sekine et al.

    Fault diagnosis of power systems

    Proc. IEEE

    (1992)
  • Applied Protective Relaying, Westinghouse Electric Corporation, Relay Investment-Division, Coral Springs, FL, USA,...
  • Single phase tripping and autoreclosing of transmission lines

    IEEE Trans. Power Deliv.

    (1992)
  • T. Dalsten et al.

    Multi-neural network based fault area estimation for high speed protective relaying

    IEEE Trans. Power Deliv.

    (1996)
  • M. Kezunovic et al.

    Detect and classify faults using neural nets

    IEEE Trans. Comput. Appl. Power

    (1996)
  • CIGRE TF 38.06.06, Artificial neural networks for power systems, Electra, No. 159, April 1995, pp....
There are more references available in the full text version of this article.

Cited by (85)

  • <inf>Fault</inf> detection <inf>and diagnosis in power transformers: a comprehensive review and classification of publications and methods</inf>

    2022, Electric Power Systems Research
    Citation Excerpt :

    Between 1991 and 2000, the majority of research focused on mathematical models of the system's physical mechanism and structure [7–63]. Between 2001 and 2010, most studies have concentrated on accuracy and computational efficiency [64–170]. In this decade, some new data driven approaches have been presented due to the monitoring technologies, measured data, availability of physical models, and communication infrastructure.

  • A novel model based on wavelet LS-SVM integrated improved PSO algorithm for forecasting of dissolved gas contents in power transformers

    2018, Electric Power Systems Research
    Citation Excerpt :

    These techniques were standardized in 1978 by IEC and later revised in 2008 [4]. In the last two decades, artificial intelligent (AI) techniques have been proposed for transformer fault diagnosis, such as fuzzy logic inference system [5–17], artificial neural network (ANN) method [18–26], expert systems [27–29], grey clustering analysis method [30,31], and rough set theory method [32,33], among others. Additional AI techniques have been used by researchers, such as self-organizing polynomial networks [34], organizing-map algorithm [35], data mining approach [36], extension theory [37], Bayesian network [38], and kernel-based possibilistic theory [39], etc.

  • Dissolved Gas Analysis of Transformer: An Approach Based on ML and MCDM

    2023, IEEE Transactions on Dielectrics and Electrical Insulation
  • Radial Power Distribution System Fault Classification Model Based on ANFIS

    2023, International Journal on Recent and Innovation Trends in Computing and Communication
View all citing articles on Scopus
View full text