Transformer Intelligent Diagnosis Method Based on Multi-source Information Fusion

The state-of-the-art fault diagnosis methods for transformers have some disadvantages, e.g., single information source and low diagnosis accuracy, which result in unreliable judgments. Based on transformer information fusion technology, this paper proposes a fault diagnosis method combining probability extreme learning machine (PELM) and improved Dempster-Shafer (D-S) evidence theory. First, three types of data, including dissolved gas in oil, electrical test, and oil chemical experiment, are selected as the input characteristics for fault diagnosis. Furthermore, the method of relative deterioration degree is adopted for preprocessing. Then, the improved extreme learning machine is used to convert the preliminary diagnosis results into a probability distribution matrix. Afterwards, the improved D-S evidence theory algorithm is applied to fuse multiple evidence bodies to achieve comprehensive diagnosis of transformers, which makes the diagnosis results more accurate. Experimental results show that the fault diagnosis accuracy rate of the method proposed is 98.27%. The real-life fault data provided by a certain maintenance company have been tested to verify the effectiveness of the proposed method and the performance of the proposed method is analyzed.