GRAPH ATTENTION-BASED MULTI-SCALE WAVELET INTELLIGENCE FRAMEWORK FOR HYBRID POWER QUALITY DISTURBANCE CLASSIFICATION
Keywords:
Power Quality Disturbances, Graph Attention Network, Discrete Wavelet Transform, Multi-Resolution Analysis, Hybrid Disturbances, Smart Grid Monitoring, Explainable Artificial Intelligence.Abstract
The increasing penetration of renewable energy resources, power electronic converters, electric vehicle infrastructure, and nonlinear industrial loads has significantly increased the occurrence of complex and hybrid power quality disturbances (PQDs) in modern smart grids. Conventional wavelet–ANN frameworks rely on static feature vectors and treat wavelet sub-bands independently, limiting their capability to capture cross-frequency interactions inherent in hybrid disturbances.
This paper proposes a Graph Attention-Based Multi-Scale Wavelet Intelligence (GAMWI) framework for accurate and interpretable classification of IEEE Std. 1159-compliant single and hybrid power quality disturbances. In the proposed approach, multi-resolution wavelet energy features are transformed into structured graph representations that model inter-scale dependencies among frequency bands. A lightweight Graph Attention Network (GAT) dynamically assigns adaptive importance weights to relational frequency interactions, thereby improving disturbance separability and enhancing interpretability of the classification process.
Simulation results demonstrate that the proposed framework achieves a classification accuracy of 99.21%, outperforming conventional DWT-ANN and CNN-based classifiers while maintaining lower computational complexity. The proposed method also exhibits strong robustness under noisy operating conditions, making it suitable for real-time smart grid power quality monitoring applications.
