Prediction of electrical power disturbances using machine learning techniques

Electrical power disturbances have negative social, economic, and political impacts. They can lead to catastrophic results that may end with blackouts. Increased understanding, analysis, and prediction of electrical disturbances can help to avoid the occurrence of major disturbances or at least to limit their consequences. This paper develops a system that predicts the type of electrical disturbances using machine learning techniques (MLTs). The proposed system is used for features selection and classification of an open source electrical disturbances dataset available online. Ant colony optimization is used for the features selection and 5 MLTs are adopted for classification; k-nearest neighbor, artificial neural networks, decision tree, logistic regression, and naïve bayes. The findings and results showed that the proposed system has the ability to efficiently classify electrical disturbances with a classification accuracy ranging from 74.57 to 86.11% depending on the classifier used.