Improvement of Power Transformer Faults Diagnosis Using Random Search with Classification and Regression Tree

Abstract

Dissolved gas in transformer oil is an important parameter to analyze the operating condition of transformers. Dissolved gas analysis (DGA) is also a commonly used method for transformer fault diagnosis. Compared to the traditional IEC Ratios, Rogers, Duval Triangle and Pentagon methods, the artificial intelligence algorithms improve the efficiency of transformer fault diagnosis, and also reduce the requirements and reliance on application experience. In order to explore more feature information in the DGA data and the accuracy of diagnosis results, a transformer fault diagnosis model based on random search and classification regression tree is built in the paper. Based on the interrelationship of the dissolved gases, the paper expands the number of features of DGA. In addition, the random search algorithm is used to realize the parameter optimization of CART model so as to improve the accuracy of fault diagnosis results. Based on the collected DGA sample dataset in the paper, the improvement effect of the RS algorithm on the CART model is verified and discussed. It is found that the median accuracy rate exceeds 92.3% for the power transformer diagnosis, demonstrating the effectiveness of the proposed technique.