Critical scenarios identification in power system simulations using graph measures and machine learning

Abstract

It is essential that electrical power systems are constructed with a reliable and resilient infrastructure. The evaluation of convergence scenarios of the load flow is a technique widely used to study the reliability of energy systems. This paper considers the classification of convergence scenarios under different loading and power generation conditions. Scenarios where the solution is not converging are evaluated using machine learning algorithms. A data set is built from power system topological representation and the simulation of load flows. Algorithms including Support Vector Machine, K-Nearest-Neighbor, and Decision Trees are evaluated and compared. The trained models can be used as a step in the contingency analysis process to be able reduce the computational time and effort in the execution of load flow calculations.