Predicting operational failures in low-voltage distribution networks using machine learning-based classification models

The advancement of smart grids has allowed distributed generators and energy storage systems to be connected to low-voltage networks, making the planning and operation of distribution systems more complex. With consumers and prosumers becoming critical elements, data analytics and artificial intelligence technologies are necessary to understand load behaviour and operational factors causing failures. This work proposes the use of machine learning-based classification models to predict critical distribution transformers, based on power flow processed with 60-minutes interval and historical weather data. The aim is to assist professionals in allocating investments for maintenance and construction in low-voltage networks to improve systems reliability. Using real grid network data from a Brazilian Distributed System Operator, the Random Forest was the best model for the problem solution, achieved 99.17% accuracy and 70% precision in identifying critical distribution transformers.