Signal Processing-Free Intelligent Model for Power Quality Disturbances Identification

Integrating different types of renewable energy sources in the power system substantially challenges the power quality (PQ), directly affecting the system's stability and service life span. The rise of power quality disturbances (PQD) generates irregularities in voltage and current waveforms, harming smart grid networks and linked devices. Traditional methods for PQD classification use complicated feature extraction techniques, which can be computationally expensive and lack scalability. This research proposes applying basic convolutional neural network (CNN) models for automated PQD detection and categorization as a prospective solution to these issues. By directly examining PQD images generated from signal data, these models reduce the requirement for human-crafted features. The study analyzes alternative CNN setups, training datasets, and disturbance types to measure model performance. The results demonstrate that these simple CNN models maintain stable accuracy values in normal and noisy environments, even with increasing classes and noise, the models managed to maintain a high-performance level with up to 99.39% accuracy for 17 classes when the Adam optimizer was used instead of RMSprop. The models could deal with noise-related disturbances, still achieving accuracy as high as 96.42% when trained by just 50% of the dataset under 30dB SNR (Signal to Noise Ratio) conditions. Moreover, comparing the two frequencies on 50Hz and 60Hz performance does not show the equivalent models' robustness over different operating levels. This study highlights the potential of CNNs in boosting power quality disturbance categorization and presents paths for further inquiry in model refining and optimization. The study focuses on CNN-based models applied in power quality disturbance detection and classification research.