A Deep Learning Approach to the Transformer Life Prediction Considering Diverse Aging Factors

Abstract

The reliability of a high-capacity power transformer is fundamental to the stable operation of power systems. However, characterization of the transformer aging process is a difficult task, considering the diverse aging factors in its life cycle. This prevents effective management of such equipment. In the work, we study the aging phenomenon of power system transformers, whose representative degeneration variables are extracted from real transformer operational data. Combining with the average life of the equipment, the extracted features are used as indicators for the transformer reliability evaluations. We developed a deep learning–based approach using a convolutional neural network for effective equipment life prediction. The performance of the transformer life prediction model is verified using field-test data, which demonstrates the superior accuracy of the presented approach.