A dissolved Gases Analysis Method for Power Transformer Faults Diagnosis Based on the Observation of Subsets of Labelled Fault Data

Power transformers play a critical role in electrical power transmission and distribution, necessitating the swift and precise identification of internal faults to ensure the grid's safe and stable functionality. This paper introduces an innovative method for diagnosing power transformer faults using dissolved gas analysis (DGA). The proposed technique employs a two-step diagnostic approach involving the formation and analysis of subsets. In the initial step, subsets are formed by regrouping samples with the same ranking sequence in descending order of concentrations of the different fault-related gases. The subsequent step involves developing a traditional sub-model for each subset, utilizing gas ratios to differentiate between various faults associated with that subset. The final diagnosis model is obtained by combining the different sub-models. The proposed method has been implemented with 680 DGA samples and tested on 169 DGA samples. The performance of the proposed method was assessed against existing traditional, intelligent, and hybrid methods utilizing a part of the International Electrotechnical Commission TC10 database. The proposed technique yielded a diagnosis accuracy of 98.27% on the validation dataset, outperforming other approaches, such as the three ratios technique and the Hyosun Corporation gas ratios method, which achieved accuracies of 93.10%.