Machine Learning Based Post-Earthquake Performance Rapid Assessment of Power Transformer in Substation

Abstract To assess the performance status of power transformers in substations after earthquakes, this study proposes an intermediate mechanical index (IMI) directly related to electrical function through finite element partition. A simulation model of the equipment is created to gather seismic response data of IMI and facilitate the prediction of easily-monitored mechanical indices (EMMI). Subsequently, a composite map from the EMMI to electrical function, can be constructed by integrating the previously indicated conversion and simulation model. Using the response data as training samples, the map would be represented by a proxy model embedded individual or ensemble machine learning algorithms. After collecting the actual seismic responses via monitoring during earthquakes, a post-earthquake evaluation will be performed based on the input-output operation of the map proxy model. This method is used for assessing the electrical contact status of a grid-side bushing affixed to a high-voltage converter transformer. The error analysis confirmed the precision of this proposed method. This approach could be similarly used for post-earthquake assessments of other equipment and electrical functions, even though this work provides validation for just one instance.