A Methodology to Diagnose Transformer Faults Based on the Multi-Objective Optimization and the Case-Based Reasoning

There exists a complicated mapping relationship between the types of transformer faults and the compositions and concentrations of dissolved gases in transformer oil. A methodology based on multi-objective optimization and case-based reasoning (CBR) is proposed to diagnose transformer faults by analyzing the dissolved gas analysis data. Considering the characteristics of the features used in the transformer diagnosis model, the targeted improvements are made in the preprocessing phase of the features to protect the diversities of samples. The quality of the feature subset is quantified by the internal cross-validation based on the CBR, and the feature subset is optimized by NSGA-II, considering both the classification performance and the size of the feature subset. The effect of the algorithm parameters on the performance is investigated, and the results validate the robustness of the proposed methodology. The proposed methodology is applied to diagnose cases from the IEC TC 10 database, achieving a 100% fault type recognition accuracy and a 92.6% severity evaluation accuracy.