Predicting the insulating paper state of the power transformer based on XGBoost/LightGBM models

Power transformer plays a crucial role in the power networks. Most of the transformer malfunctions was due to failure in the insulating systems. The utilities keen on the contentious operation of the power network, so early detection of the transformer faults avoiding the undesirable outage of the transformer from the service. The insulating paper state is an indication to the transformer health and aging of it may lead to failure of the transformer, so some periodic and routine test must be performed on the insulting oil to get information about the insulating paper condition. The value of the degree of polymerization (DP) is a key of the insulating paper state. Various recommended tests such as the dissolved gases (DGA), breakdown voltage (BDV), oil interfacial tension (IF), oil acidity (ACI), moisture content (MC), oil color (OC), dielectric loss (Tan δ), and furans concentration specifically (2-furfuraldehyde (2-FAL)) were carried out to correlate between these variables and the DP and then to the insulating oil state. The collected data from these tests were used to supply XGBoost/LightGBM to build artificial intelligence model to predict the insulating paper state. The results indicated that the great ability of the proposed model to predict the insulating state with high accuracy. Of these various configurations for these two classification models, one achieved perfect prediction accuracy (100%) of the insulating paper state prediction accuracy. The other configurations showed accuracy values from 1.0 down to 0.955.