Controlled islanding of power networks using machine learning algorithm

Wide-area blackouts have occurred the last decades due to severe disturbances and cascading failures. Controlled islanding of a power system is a measure for preventing blackout and limiting the effects of disturbances. Intention islanding aims to divide the network into electromechanically stable islands, which should satisfy a number of constraints such as generator coherency, generation-demand balance, system limits, and network dynamic constraints. The controlled islanding addresses the problem of finding which lines should be disconnected to create the islands. In this paper a modified algorithm of SelfOrganizing Map (SOM) in combination with a Particle Swarm Optimization (PSO) algorithm is proposed in order to determine which lines should be disconnected. To evaluate the algorithm performance the method is applied to test networks.