Online prediction of DGA results for intelligent condition monitoring of power transformers

Nandagopan, G and Beevi, K Sabeena and Lekshmi, A S Kunju and Vishnupriya, K J and Kumar, Deepa S and Abhijith, S and Rishika, K K; (2022) Online prediction of DGA results for intelligent condition monitoring of power transformers. In: 2022 IEEE International Conference on Power Electronics, Smart Grid, and Renewable Energy (PESGRE). IEEE, Piscataway, NJ, pp. 1-6. ISBN 9781665448376 (

[thumbnail of Nandagopan-etal-PESGRE-2022-Online-prediction-of-DGA-results-for-intelligent-condition-monitoring]
Text. Filename: Nandagopan_etal_PESGRE_2022_Online_prediction_of_DGA_results_for_intelligent_condition_monitoring.pdf
Accepted Author Manuscript
License: Strathprints license 1.0

Download (1MB)| Preview


Transformers form a major part of a power system in transmission as well as distribution of power. Considering the criticality, finance, and time involved in repair, periodic condition monitoring and maintenance of transformers are the key to ensure electrical safety as well as stable operation of the large interconnected power system. Dissolved Gas Analysis (DGA) is an established tool used to determine the incipient faults within the transformer by analyzing the concentration of different gases in the transformer oil and giving early warnings and diagnoses. Currently, transformers worldwide utilise online sensors to monitor dissolved gases and moisture content in oil. The online DGA sensor uses a small amount of oil from transformer to perform real-time DGA analysis and gives the ppm content of dissolved gases for further course of action. Considering the large quantity of assets and the huge amount of data produced, it is imperative to develop a tool to aid the operators in assimilating the available data for diagnosis and proactive decision making. The present study improvises AI techniques to predict future dissolved gas concentrations using real time DGA data collected from the transmission utility of the country. The prediction helps to forecast the trend of development of incipient faults in the transformer. The complete project scope is to develop a highly reliable diagnostic tool to emulate the decision-making ability of a human expert in transformer DGA analysis to enhance transformer life. In the present paper, models based on Auto-regressive Integrated Moving Average (ARIMA), Long Short-Term Memory (LSTM), and Vector Auto Regression (VAR) are implemented to predict DGA data of three in-service transformers. DGA data is forecasted for up to 8 monthly samples in the future, and the accuracy of results is compared with each other. The LSTM-VAR combined model is seen to provide the best results among them.