Assessment of power quality improvement in microgrid using ReliefF-based feature selection and LSTM recurrent neural network

The nonlinear loads, distributed generation and intermittency of renewable energy (RE) are serious issues in modern microgrids and result in a degradation of power quality. This study presents a data-driven framework, ReliefF–LSTM, for power-quality disturbance classification and power-quality improvement assessment, which is a MATLAB/Simulink-simulated IEEE 33-bus microgrid. ReliefF is used to select the power-quality features that are relevant and the LSTM is used to classify the temporal patterns into seven power quality disturbances. This forecasted disturbance class is passed to a subsequent compensation section which is tested with total harmonic distortion (THD), power factor and voltage deviation. Electric measurements such as voltage, current, frequency, power factor, active power and harmonic components are then measured and processed, followed by signal conditioning and multi-domain feature extraction. The algorithm ReliefF selects the most relevant and non-redundant features of the power quality indicators, and the selected features are grouped in temporal sequences. The power-quality disturbance conditions are then specifically learned and numerically predicted using a Long Short-Term Memory (LSTM) recurrent neural network, which is not directly used to generate the control signals. The ReliefF–LSTM model is formulated exclusively as a sequence-to-label classifier that identifies seven power-quality operating conditions. It does not predict continuous electrical values or directly generate control signals. The predicted disturbance class is subsequently provided to a separate power-electronic compensation stage, whose effectiveness is evaluated using THD, power factor, and voltage-regulation measurements. The experimental comparative analysis is made with eight benchmark techniques such as Artificial Neural Network (ANN), RNN, GRU, LSTM, CNN-LSTM, BiLSTM, TLBO-ANN, and GJO-SAPF. Based on the predicted power-quality conditions, following disturbance classification, the separately implemented DSTATCOM reduced THD from 14.7% to 3.1%, corresponding to a 78.91% reduction, improved the power factor from 0.78 to 0.94, and reduced the maximum bus-voltage deviation from 0.082 to 0.032 p.u., corresponding to a 60.98% reduction. The outcomes of these results have confirmed the accuracy of the proposed framework in terms of temporal disturbance prediction, and the power quality enhancement and intelligent microgrid management following that prediction.

Authors

Publication Details

Journal
Scientific Reports
Published
2026-11-14
DOI
https://doi.org/10.1038/s41598-026-80308-4
Primary Topic
Power Quality and Harmonics
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Assessment of power quality improvement in microgrid using ReliefF-based feature selection and LSTM recurrent neural network

P. Divya Banu, Sundar Govindasamy
Scientific Reports
Power Quality and Harmonics
article

Assessment of power quality improvement in microgrid using ReliefF-based feature selection and LSTM recurrent neural network

P. Divya Banu, Sundar Govindasamy
article en

Abstract

The nonlinear loads, distributed generation and intermittency of renewable energy (RE) are serious issues in modern microgrids and result in a degradation of power quality. This study presents a data-driven framework, ReliefF–LSTM, for power-quality disturbance classification and power-quality improvement assessment, which is a MATLAB/Simulink-simulated IEEE 33-bus microgrid. ReliefF is used to select the power-quality features that are relevant and the LSTM is used to classify the temporal patterns into seven power quality disturbances. This forecasted disturbance class is passed to a subsequent compensation section which is tested with total harmonic distortion (THD), power factor and voltage deviation. Electric measurements such as voltage, current, frequency, power factor, active power and harmonic components are then measured and processed, followed by signal conditioning and multi-domain feature extraction. The algorithm ReliefF selects the most relevant and non-redundant features of the power quality indicators, and the selected features are grouped in temporal sequences. The power-quality disturbance conditions are then specifically learned and numerically predicted using a Long Short-Term Memory (LSTM) recurrent neural network, which is not directly used to generate the control signals. The ReliefF–LSTM model is formulated exclusively as a sequence-to-label classifier that identifies seven power-quality operating conditions. It does not predict continuous electrical values or directly generate control signals. The predicted disturbance class is subsequently provided to a separate power-electronic compensation stage, whose effectiveness is evaluated using THD, power factor, and voltage-regulation measurements. The experimental comparative analysis is made with eight benchmark techniques such as Artificial Neural Network (ANN), RNN, GRU, LSTM, CNN-LSTM, BiLSTM, TLBO-ANN, and GJO-SAPF. Based on the predicted power-quality conditions, following disturbance classification, the separately implemented DSTATCOM reduced THD from 14.7% to 3.1%, corresponding to a 78.91% reduction, improved the power factor from 0.78 to 0.94, and reduced the maximum bus-voltage deviation from 0.082 to 0.032 p.u., corresponding to a 60.98% reduction. The outcomes of these results have confirmed the accuracy of the proposed framework in terms of temporal disturbance prediction, and the power quality enhancement and intelligent microgrid management following that prediction.

Scientific Reports
Affordable and clean energy, Industry, innovation and infrastructure
Power Quality and Harmonics
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon