Artificial Neural Networks Application for Top Oil Temperature and Loss of Life Prediction in Power Transformers

The development of precise tools for power transformers temperature prediction allows a better use of equipment's nominal capacity, extending its useful life and possibility of strategic planning based on possible future operating scenarios. Proposition of temperature prediction models is of great interest to those responsible for power transformers. This article presents the development of Artificial Neural Networks (ANNs) as a tool for top oil temperature prediction in power transformers and justifies the use of Non-linear AutoRegressive with eXogenous inputs (NARX) model and input parameters according to the thermal behavior of the transformers under study. Also, a comparison from the perspective of the loss of life between the ANNs response and monitoring data is made, as the prediction for fictitious future operating scenarios is presented for method validation. The results obtained demonstrate that the developed ANNs replicate in a very satisfactory way the thermal behavior of the transformers under study. Error remains small during most of the prediction horizon, approximately 2 degrees C in absolute values (about 4% nominal). Allowing operators to assess dispatch among their equipment, extending the useful life and avoiding unexpected situations, featuring a very useful tool in power plants and substations and opening paths for new studies.