Unknown Power Quality Disturbances Classification Using Transfer Learning and Deep Learning Architectures

Effective classification of power quality disturbances (PQDs) is crucial for maintaining the operational stability and efficiency of power systems. However, existing classification methods often fail to adapt to unknown disturbances, requiring extensive retraining with large datasets, which limits their usage in real-time applications. To overcome this challenge, a novel approach utilizing transfer learning is proposed. This technique allows the PQD classification model to be efficiently updated with a small dataset, significantly reducing the computational burden. The method leverages a hybrid CNN-GRU architecture, which integrates both spatial and temporal features of PQD signals, enhancing its ability to classify disturbances with high accuracy. This approach demonstrates strong adaptability and computational efficiency, making it suitable for real-time power system monitoring.