Evolution strategies-tuned support vector machine-based classification of inter-area oscillations

Tools for real-time monitoring of inter-area oscillations are now commercially available. These tools have been validated in many power systems with different characteristics and are in operation in some control rooms. Yet missing, however, are tools that can assist an operator to identify the root cause of poorly damped oscillations and propose appropriate countermeasures. As a step towards this direction, this paper describes the construction of a support vector machine model trained to classify potential operating points according to their corresponding oscillation damping ratios. Evolution strategies are used to tune the SVM hyperparameters, including the selection of its kernel function, such that the accuracy of the resulting model is as high as possible.