Missing Data Imputation in Power Transformer Monitoring Based on a Spatiotemporal Recurrent Neural Network

Power transformer is one of the core of power systems. It is crucial to accurately and proficiently estimate the missing operational monitoring data of power transformers for precisely assessing the operating condition of transformers and the operation of power systems. In contrast to other monitoring data, the operational monitoring data of power transformers exhibits strong temporal and feature correlations. However, most imputation methods independently study one of these correlations without fully integrating these types of correlations, leading to insufficient imputation accuracy. Therefore, this paper proposes an imputation method based on Machine Learning (ML) that combines temporal correlation and feature correlation. It can learn the development trend of changes in a single feature’s temporal dimension within transformer data, supplemented by the variation characteristics of other dimensional features in the same environment, taking into account both temporal and spatial complex features simultaneously, thereby improving the imputation accuracy.