Graph Machine Learning Edge Classification Approaches in Unbalanced Datasets: Transmission Line Contingency Screening in Electrical Power Systems

In order to deal with relational data, classical machine learning techniques are being improved to consider graph data strucutres, which comprises the graph machine learning field. Many networks can be modeled using graph structures, such as social networks, communication networks and power system networks. When dealing with security assessment of power system networks, the identification of the most severe contingency scenarios is called the contingency screening step, which can be frame as a graph editing problem in the case of equipment failures. In this work, the use of graph machine learning approaches for solving the transmission line contingency problem is evaluated. The problem is formulated as a binary classification task on the edges of a graph and two learning approaches are proposed and compared.