An AI-based Approach for Improved Power Quality Disturbances Detection and Classification

Abstract

Power quality disturbances (PQDs) have become more common in contemporary electrical systems due to the increasing implementation of non-linear loads and renewable energy technologies that use power electronics. Traditional machine learning techniques mostly depend on manually engineered features to detect and classify PQDs. Those restricts their ability to generalize across various PQD categories. In our research, we are going to present an enhanced deep learning methodology for accurately identifying and categorizing multiple power quality disturbances using AI techniques. We used the SEED Power Quality Disturbance Dataset for training and testing our models. We developed and evaluated four artificial intelligence approaches: one-dimensional Convolutional Neural Network (CNN- 1D), conventional CNN, Long Short-Term Memory (LSTM), and a combined CNN-LSTM architecture. The CNN-LSTM hybrid model outperformed others in classification accuracy. This research highlights the superiority of deep learning models in automatically learning temporal and spatial features, thus eliminating the need for manual feature extraction.