Incremental Fuzzy Machine Learning for Power Systems Fault Classification

Abstract

The concept of power quality is related to a set of changes that may occur in the electrical system. These result in flaws or bad consumer equipment operation. Such changes (disturbances/faults) can occur in many parts of a power system, including the consumer installation and the supplying system, which may yield financial losses to both. Real-time automatic detection and classification of disturbances are therefore of fundamental importance. In this study, evolving intelligent models, that is, models supplied with incremental online learning algorithms capable of changing their parameters and structure according to new information that emerges from a data stream, are considered for pattern recognition and classification. In particular, an evolving fuzzy set-based model (FBeM) is taken into consideration. A Hodrick-Prescott filter combined with a Fast Fourier Transform technique and root mean square voltages are considered for pre-processing measured data and extracting features that indicate the presence of disturbances. The model developed in this study has reached classification performance comparable to that of state-of-the-art models in the field of power quality. Detection and classification of disturbances such as voltage swell, sub-harmonic, oscillatory transient, spikes, and notching, occurring simultaneously, were reached with an accuracy of about 99%.