On-line supervised learning for dynamic security classification using probabilistic neural networks

Abstract

This paper addresses the problem of dynamic security classification of electric power systems using multiclass pattern recognition. In particular, on-line supervised learning using Probabilistic Neural Networks is applied. The various patterns are recognized by calculating probabilities of belonging to each class. These probabilities are used in a subsequent decision-making stage to achieve classification. The learning of each class can be performed in parallel. Results regarding performance of the proposed pattern recognition tested on the dynamic security of an actual island power system are presented and discussed.

Review of Data-Driven Techniques for On-Line Static and Dynamic Security Assessment of Modern Power Systems

“…However, it is also possible to consider the rate of change of frequency (RoCoF), i.e., the time derivative of the power system frequency (df/dt), often in combination with the frequency minimum/maximum values [119]- [122].…”

“…The GD/FD (49) was used and returned an error of 8-9% when applied to events simulated using the Taipower system. In the same year, frequency nadir and RoCoF were combined in a DT binary classification problem in [122].…”

Application of Machine Learning on Power System Dynamic Security Assessment

“…An online supervised learning method for multiclass classification of power system frequency stability, tested with on-line data from an actual power system, is presented in [10]. The method is general and can be applied to all types of stability problems, by simply changing the security classification criteria.…”