A Transformer Warning Method Based on Pattern Recognition and Statistical Analysis

Temperature is one of the most important characteristics reflecting internal faults in transformers. By monitoring the unconventional operating behavior of temperature, potential fault risks that may exist in transformers can be timely grasped. This study takes the transformer’s multiple parameters as the analysis object, and applies a combination of pattern recognition and statistical analysis to achieve early warning of potential transformer hazards. In the proposed method, firstly, the OPTICS+K-Means dual clustering algorithm is used to construct a transformer temperature early warning state model from multi-parameter historical data; secondly, the state model is used to match a set of most similar data for the observed data, and the residuals between the two are used to measure the degree of equipment abnormality; then, the Mann-Kendall test method is used to test the trend characteristics of the residuals, so as to determine whether the transformer has abnormal operating behavior. The experimental results show that the proposed method has the ability to detect transformer temperature anomalies in advance.