On Deep Learning for Condition Assessment of Power Transformers

Accurately assessing power transformers' health condition and calculating their probability of failure (PoF) can facilitate decision-making with respect to the operation and maintenance of transformers. Among various data and information used for transformer condition assessment and PoF calculation, the Dissolved Gas Analysis (DGA) plays a key role in diagnosing transformer incipient faults. In this paper, we have developed a deep learning-based model to improves the effectiveness of DGA interpretation for transformer condition assessment. Then, the refined DGA results are used alongside other transformer measurement data to support the PoF calculation and risk cost analysis. The method developed in this paper has been validated using the dataset collected from numerous in-service power transformers.