Understanding and predicting interruptions index of medium voltage customers using fully connected networks

Abstract

The losses caused by the lack of electricity typically exceed the cost of the electricity itself. Improving power quality is a way to reduce or avoid loss of production in the industry, prevent fires or explosions, and minimize damages to industrial equipment. Therefore, finding customers that probably will have interruptions in advance will generate value for both the company and customers. The purpose of this study is to analyze data from units that consume electricity using neural networks and decision trees, such as self-organizing maps, CHAID and CART, and using fully connected neural networks to predict the interruption index for the next year. The results reveal an important space for improvements such as the connection between non-compliance of established indicators over time and specific points of electrical network with problems. That way supports the concessionaries to manage their infrastructure to get a better quality of the electric power network.