Prediction of power supply interruption incidents using correlation matrices and SVM

Abstract

The quality of the electric power distribution service directly impacts the quality of life of people. Incidents that lead to power supply interruptions can occur, and it is important that a quick response is given by the company responsible for restoring the service. This work proposes a model for predicting these incidents, aiming to assist energy distribution companies in improving the quality and agility of their service and, consequently, avoiding the application of sanctions by the regulatory agencies. The forecast is based on the application of support vector machines for regression. It considers historical data of incidents occurred in four regions of the state of Rio de Janeiro, Brazil (Cabo Frio, Macaé, Niteroí and Resende), as well as climatic data from the same regions, both collected in the period between 2017 and 2018. To improve the performance of the model, an automatic feature selection technique was applied, based on correlation matrices. Values of mean absolute percentage error were around 15% for three of the four regions studied, sufficient for the model to assist in the strategic decision-making process by the company. Also, all results obtained by the complete model are 30% better than those obtained without feature selection, which highlights the importance of this work.