STFT-ENGB: A Hybrid Time Frequency and Gradient Boosting Approach for Power Quality Disturbance Detection

Abstract

Power signal processing is a specialized domain within signal processing that focuses on the analysis, interpretation, and manipulation of signals in electrical power systems. In modern smart grids, Power Quality Disturbances (PQDs) can result in considerable operational disruptions and financial losses for energy stakeholders. This research introduces a Short-Time Fourier Transform fused Efficient Natural Gradient Boosting (STFT-ENGB) model for robust recognition of power quality disturbances with energy grid applications. A comprehensive framework used for PQD identification by leveraging advanced power signal processing techniques and time-frequency-based feature extraction. The system collects electrical measurements from the power system includes voltage and current. The Z-score normalization is a preprocessing technique for reducing noise. The STFT is utilized to extract discriminative, time-localized features from the power signals. These extracted features are then combined using a late fusion strategy to form a unified representation. The proposed method was implemented using Python 3.10.1. Extensive experiments demonstrate that the proposed STFT-ENGB approach performs better than multimodal baseline architectures, achieving superior results, with accuracy, F1-score, recall, and precision ranging from 95% to 99%. These findings offer a promising solution for real-time power signal monitoring in smart grid environments, facilitating intelligent fault diagnosis and improving the overall resilience and responsiveness of modern electrical infrastructure.