Statistical machine learning and dissolved gas analysis: A review

Piotr Mirowski, Yann LeCun

Research output: Contribution to journalArticle

Abstract

Dissolved gas analysis (DGA) of the insulation oil of power transformers is an investigative tool to monitor their health and to detect impending failures by recognizing anomalous patterns of DGA concentrations. We handle the failure prediction problem as a simple data-mining task on DGA samples, optionally exploiting the transformer's age, nominal power and voltage, and consider two approaches: 1) binary classification and 2) regression of the time to failure. We propose a simple logarithmic transform to preprocess DGA data in order to deal with long-tail distributions of concentrations. We have reviewed and evaluated 15 standard statistical machine-learning algorithms on that task, and reported quantitative results on a small but published set of power transformers and on proprietary data from thousands of network transformers of a utility company. Our results confirm that nonlinear decision functions, such as neural networks, support vector machines with Gaussian kernels, or local linear regression can theoretically provide slightly better performance than linear classifiers or regressors. Software and part of the data are available at http://www.mirowski.info/pub/dga.

Original languageEnglish (US)
Article number6301810
Pages (from-to)1791-1799
Number of pages9
JournalIEEE Transactions on Power Delivery
Volume27
Issue number4
DOIs
StatePublished - 2012

Fingerprint

Gas fuel analysis
Learning systems
Power transformers
Linear regression
Learning algorithms
Support vector machines
Data mining
Insulation
Classifiers
Health
Neural networks
Electric potential
Industry

Keywords

  • Artificial intelligence
  • neural networks
  • power transformer insulation
  • prediction methods
  • statistics
  • transformers

ASJC Scopus subject areas

  • Electrical and Electronic Engineering
  • Energy Engineering and Power Technology

Cite this

Statistical machine learning and dissolved gas analysis : A review. / Mirowski, Piotr; LeCun, Yann.

In: IEEE Transactions on Power Delivery, Vol. 27, No. 4, 6301810, 2012, p. 1791-1799.

Research output: Contribution to journalArticle

@article{e6fba13a1b1749e4bdb1f7b2830e1a26,
title = "Statistical machine learning and dissolved gas analysis: A review",
abstract = "Dissolved gas analysis (DGA) of the insulation oil of power transformers is an investigative tool to monitor their health and to detect impending failures by recognizing anomalous patterns of DGA concentrations. We handle the failure prediction problem as a simple data-mining task on DGA samples, optionally exploiting the transformer's age, nominal power and voltage, and consider two approaches: 1) binary classification and 2) regression of the time to failure. We propose a simple logarithmic transform to preprocess DGA data in order to deal with long-tail distributions of concentrations. We have reviewed and evaluated 15 standard statistical machine-learning algorithms on that task, and reported quantitative results on a small but published set of power transformers and on proprietary data from thousands of network transformers of a utility company. Our results confirm that nonlinear decision functions, such as neural networks, support vector machines with Gaussian kernels, or local linear regression can theoretically provide slightly better performance than linear classifiers or regressors. Software and part of the data are available at http://www.mirowski.info/pub/dga.",
keywords = "Artificial intelligence, neural networks, power transformer insulation, prediction methods, statistics, transformers",
author = "Piotr Mirowski and Yann LeCun",
year = "2012",
doi = "10.1109/TPWRD.2012.2197868",
language = "English (US)",
volume = "27",
pages = "1791--1799",
journal = "IEEE Transactions on Power Delivery",
issn = "0885-8977",
publisher = "Institute of Electrical and Electronics Engineers Inc.",
number = "4",

}

TY - JOUR

T1 - Statistical machine learning and dissolved gas analysis

T2 - A review

AU - Mirowski, Piotr

AU - LeCun, Yann

PY - 2012

Y1 - 2012

N2 - Dissolved gas analysis (DGA) of the insulation oil of power transformers is an investigative tool to monitor their health and to detect impending failures by recognizing anomalous patterns of DGA concentrations. We handle the failure prediction problem as a simple data-mining task on DGA samples, optionally exploiting the transformer's age, nominal power and voltage, and consider two approaches: 1) binary classification and 2) regression of the time to failure. We propose a simple logarithmic transform to preprocess DGA data in order to deal with long-tail distributions of concentrations. We have reviewed and evaluated 15 standard statistical machine-learning algorithms on that task, and reported quantitative results on a small but published set of power transformers and on proprietary data from thousands of network transformers of a utility company. Our results confirm that nonlinear decision functions, such as neural networks, support vector machines with Gaussian kernels, or local linear regression can theoretically provide slightly better performance than linear classifiers or regressors. Software and part of the data are available at http://www.mirowski.info/pub/dga.

AB - Dissolved gas analysis (DGA) of the insulation oil of power transformers is an investigative tool to monitor their health and to detect impending failures by recognizing anomalous patterns of DGA concentrations. We handle the failure prediction problem as a simple data-mining task on DGA samples, optionally exploiting the transformer's age, nominal power and voltage, and consider two approaches: 1) binary classification and 2) regression of the time to failure. We propose a simple logarithmic transform to preprocess DGA data in order to deal with long-tail distributions of concentrations. We have reviewed and evaluated 15 standard statistical machine-learning algorithms on that task, and reported quantitative results on a small but published set of power transformers and on proprietary data from thousands of network transformers of a utility company. Our results confirm that nonlinear decision functions, such as neural networks, support vector machines with Gaussian kernels, or local linear regression can theoretically provide slightly better performance than linear classifiers or regressors. Software and part of the data are available at http://www.mirowski.info/pub/dga.

KW - Artificial intelligence

KW - neural networks

KW - power transformer insulation

KW - prediction methods

KW - statistics

KW - transformers

UR - http://www.scopus.com/inward/record.url?scp=84866706797&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84866706797&partnerID=8YFLogxK

U2 - 10.1109/TPWRD.2012.2197868

DO - 10.1109/TPWRD.2012.2197868

M3 - Article

AN - SCOPUS:84866706797

VL - 27

SP - 1791

EP - 1799

JO - IEEE Transactions on Power Delivery

JF - IEEE Transactions on Power Delivery

SN - 0885-8977

IS - 4

M1 - 6301810

ER -