Machine Learning-Based Fault Detection in Electrical Equipment

Abstract

Effective fault detection and diagnosis are essential for ensuring safety, effectiveness, and stability of the power supply system. The delayed detection of faults results in further breakdowns of the equipment and the power supply system. In the current paper, an effective pipeline of fault detection through machine learning and the use of the electrical fault dataset is designed. Although the selection of the appropriate machine learning algorithm is usually regarded as the primary problem in such research, the current paper focuses on the design of the effective data pipeline including the upload and processing of the dataset, removal of the redundant attributes, as well as the handling of missing values and normalization of other attributes. The class distributions and correlations are used to analyze the existence of faults in the dataset before dividing it into training and test datasets.