Transformer fault diagnosis based on Bayes active learning

Dissolved Gas Analysis (DGA) is the most commonly used method for on-site power transformer fault diagnosis. Artificial Intelligence (AI) techniques enhance fault diagnosis but rely on large amounts of labeled data. Active learning optimizes this by selectively querying uncertain, informative samples for manual labeling, which reduces costs and improves diagnostic accuracy with minimal number of labeled data. This paper introduces an innovative Bayes Active Learning Method based Entropy Sampling Strategy(BALM-ESS) that employs entropy-based sampling strategy in active learning process to select the most valuable samples for training. The method is validated using the IEC TC10 dataset and is compared with Bayes Random sampling (BRS) which randomly selecting some samples from the unlabeled sample. The findings demonstrate the effectiveness of the proposed algorithm and its robustness in tackling classification problems by spending as few labeled samples as possible.