A power quality detection and classification algorithm based on FDST and hyper-parameter tuned light-GBM using memetic firefly algorithm. (January 2022)

• Record Type: Journal Article
• Authors: Panigrahi, Rasmi Ranjan; Mishra, Manohar; Nayak, Janmenjoy; Shanmuganathan, Vimal; Naik, Bighnaraj; Jung, Young-Ae
• Is Part Of: Measurement. Volume 187(2022)
• Publication Date: 2022-01
• Subjects: Power quality -- Feature extraction -- Classification -- Fast discrete s-trasform -- PSO -- Light GBM -- IEEE-1159 standard; EPQ Electric power quality -- PQ Power quality -- PQD Power quality disturbance -- FE Feature extraction -- STFT Short-time Fourier transform -- WT Wavelet transform -- ST S-transform -- LGBM Light-gradient boost machine -- MFA Memetic firefly algorithm -- FDST Fast discrete s-transform -- EMD Empirical mode decomposition -- HHT Hilbert Huang transform -- ANN Artificial neural networks -- SVM Support vector machine -- DT Decision tree -- KNN K-nearest neighbor -- ELM Extreme machine leaning -- MAT maximum amplitude versus time -- MAF maximum amplitude versus frequency -- XGBM Extreme Gradient Boosting Machine -- GOSS Gradient-based One Side Sampling -- EFB Exclusive Feature Bundling -- η General parameter -- Λ regularization parametr -- l(·) loss function -- y^j predicted class -- yj original class -- fi fitness; Weights and measures -- Periodicals; Measurement -- Periodicals; Measurement; Weights and measures; Periodicals; 530.8
• DOI: 10.1016/j.measurement.2021.110260
• Languages: English
• ISSNs: 0263-2241

Abstract

Highlights: PQ disturbances (PQDs) detection and classification using FDST and LightGBM (LGBM) The hyper-parameters of LGBM are optimized through memetic firefly algorithm. Performance is validated on 25 types of PQDs (both single and multiple events) The overall detection accuracy is found to be 99.714% with noiseless condition. The detection accuracy is also verified in different level of noisy conditions. Abstract: Presently, the issue of power quality (PQ) disturbances in electrical power system has been greater than before owing to increased use of power electronics based nonlinear loads. This work has proposed a hybrid PQ detection and classification algorithm that uses fast-discrete-S-transform (FDST) as feature extraction (FE) technique and memetic firefly algorithm (MFA) based Light-gradient-boost-machine (LGBM) as a classifier. In general, 25 types of PQ signals, comprising both single and multiple disturbances, are studied considering the IEEE-1159 standard. A 3.2 kHz sampling frequency is used on ten cycles of distorted waveforms for the FE. The experimental results clearly proves the effectiveness of the proposed approach with high detection accuracy (99.714% with synthetic data and 99.66% with simulated data), less computational complexity and immune to noisy environments. To end, this work has performed a comparative study with other contemporary FE techniques and classifiers, and in addition with other previously published work.