Linking damping of electromechanical oscillations to system operating conditions using neural networks

This paper presents the application of Neural Networks to link the damping of electromechanical oscillations in the Nordic power system to the measured operating conditions. Different neural network topologies have already been presented in the literature for this application, but using exclusively data from simulations. The primary objective of the paper is to analyze how these topologies behave with data from a real power system. The damping of the 0.35 Hz electromechanical oscillation has been first estimated from a large amount of Phasor Measurements Units (PMU) measurements for a two years period. Three neural network models are trained with power system variables as generation, load and power flows over cross-border lines measured during year 2010, used as input, and the estimated damping from PMU measurements during the same year, used as target. The neural network models are then tested with the data from 2011 with the aim of estimating the damping. The results indicate that neural networks can correctly predict more than 80% of the operating conditions resulting in low damping during the entire year 2011. The presented method is purely measurement-based and it can be used in conjunction with other traditional model-based planning methods to predict oscillatory stability limits.