Congestion forecast framework based on probabilistic power flow and machine learning for smart distribution grids

Abstract

The increase in renewable energy sources and new technologies such as electric vehicles and storage can generate uncertainties in distribution grid operations, increasing the likelihood of congestions in power lines. Distribution system operators (DSOs) face several challenges while operating their grids in such conditions. These congestions deteriorate the electrical equipment in the long term, reducing its life span. This work proposes a framework to predict grid asset congestions on a daily basis. A congestion forecast framework is proposed by combining probabilistic power flows and machine learning algorithms to support DSOs in their daily decision-making. The framework is tested on a modified IEEE-33 bus system and CINELDI MV Reference system with hourly synthetic data. The results showed that the framework is able to closely predict the congestions on the lines. Computational capabilities are reported and discussed. The study indicates that the framework is a suitable tool for day-to-day congestion predictions in smart distribution grids yielding low error in expected values.