Enhancing Power Quality Disturbance Classification Through Ensemble Learning and Statistical Techniques

In an era driven by technological advancements, the reliable and uninterrupted delivery of electrical power is paramount for the seamless operation of various critical infrastructures. This paper addresses the imperative need for accurate and efficient classification of power quality disturbances, introducing a novel approach that is a simple yet highly effective feature extraction method based on statistical features. Our investigation centers on the analysis of time-series data encompassing dynamic current and voltage values collected from diverse power distribution systems. The statistical features extracted from this time-series data serve to represent the signal in the feature space. By relying on statistical features, the combination of computational efficiency and enhanced interpretability distinguish our approach as exceptionally accessible. The gradient boosting algorithm is employed to select the key features that significantly contribute to classification. Ensemble learning methods such as Random Forest and XGBoost algorithms were used to identify disturbances in the power grid. The effectiveness and robustness of this framework are assessed across multiple datasets of power quality disturbances, demonstrating its capability for accurate identification and classification surpassing the established benchmarks. The positive outcomes achieved through the proposed methodology suggest that our approach is a noteworthy achievement in modern power networks.