Automatic Power Quality Disturbance Feature Extraction Using Fast Iterative Filtering

Abstract

Power quality (PQ) events have become increasingly complex due to the growing integration of renewable energy resources into electric grids. Hence, there is a pressing need for expressive latent representations of the sensorized grid components that can aid the timely detection of PQ disturbances (PQDs). Nevertheless, traditional feature extraction methods, including those based on frequency or decomposed time domains, are generally appropriate for specific subsets of possible PQDs and heavily rely on theoretical or experiential intuition. To address these challenges, this study introduces fast iterative filtering (FIF)—an entirely data-driven approach for uncovering hidden features within disturbed signals. FIF utilizes iterative convolution with a filter constructed from the signal itself to decompose non-stationary signals into intrinsic mode functions, eliminating the need for prior knowledge. Furthermore, fast Fourier transform (FFT) is embedded to improve computational efficiency, enabling real-time recognition of PQDs. Evaluated using synthetic signals per IEEE 1159-2019 standard, FIF shows excellent extraction performance, being capable of representing the predictive content of feature tensors, including interpretative patterns of PQDs.