Temperature Prediction of Disconnecting Switch Based on Memory Regression Metric Learning

In the power system, most of the faults of disconnecting switch are eventually expressed as the form of heat. Therefore, we can detect the faults by observing its temperature. Most of the existing models employ conventional machine learning algorithms to learn the mapping function between temperature-related features and device temperature. These models do not make full use of the history information of the device to predict the current temperature value. However, in fact, the temperature variation of the disconnecting switch is continuous and sequential. In this paper, we propose a model based on Memory Regression Metric Learning (MRML) to predict the temperature of disconnecting switch. This model employs the historical features of the disconnecting switch together with a new feature for the temperature prediction, and uses metric learning to eliminate the impact of the data dimension. Experiments show that our model has better performance in temperature prediction than others.