Research on transformer health condition evaluation method based on clustering analysis and dynamic feature extraction

In order to realize the dynamic assessment and early warning of the transformer state, and gradually improve the intelligent level of equipment fault diagnosis, this paper proposes a transformer health condition evaluation method based on cluster analysis and dynamic feature extraction. Based on the data of key state quantity of dissolved gas in transformer oil, the health condition of transformer is divided into 3 kinds of characteristics: “health”, “sub-health” and “abnormality”. In this method, The Gaussian mixture model is used to analyze the fault case data of the transformer static feature extraction, by introducing time series parameters, the hidden Markov model is used to convert the extracted static features into dynamic features to complete the construction of the fault case database, and the dynamic assessment and short-term early warning of the transformer health status are realized through feature value matching. This method was used to verify the status assessment of transformer overheating faults, and the equipment health status assessment was carried out on 65 transformers in the medium and high temperature overheating case database and health equipment case database. The analysis results show that, under multiple cross-validation, the average prediction accuracy of the transformer's health status is above 95%, which has high engineering application value.