Time–Frequency Convolution Neural Network for Classification of Single and Combined Power Quality Disturbances

Detecting and classifying Power Quality Disturbances (PQDs) is crucial for maintaining the reliable and efficient operation of electrical power systems. This paper introduces a novel approach to classify single and combined PQDs using a Time–Frequency Convolutional Neural Network (PQDTFCNNet). A total dataset of 9000 signals pertaining to fifteen different PQDs, including nine single and six combined instances, were generated in accordance with the guidelines outlined in the IEEE 1159 standard. The proposed method performs 1D convolution on input raw PQD signals, followed by a Continuous Wavelet Transform (CWT) layer, to transform the data into 2D representation, leveraging the advantages of 2D convolution. The training and validation of the model have been performed with 6000 and 1500 signals, respectively. Upon testing with unseen 1500 signals, the proposed technique has achieved a remarkable 99.73% accuracy. Moreover, the proposed method demonstrated reliable performance under 20 dB noise condition achieving 98.07% accuracy. These findings underscore the potential of the PQDTFCNNet in effectively addressing the challenges associated with PQD classification.