A Novel Power Quality Disturbance Detection Using Circular Coordinate Transition and Optimized Convolutional Neural Networks

Ensuring high power quality is a critical objective for electricity distributors, particularly in modern power systems with an increasing penetration of distributed generation. The complexity of power quality disturbances (PQDs), often involving multiple simultaneous events, poses significant challenges for effective detection and classification. This paper presents a novel advanced detection method for complex PQDs. The approach integrates a circular coordinate transition (CCT) algorithm, which transforms one-dimensional PQD signals into two-dimensional images with distinctive, classifiable features. Additionally, an optimized convolutional neural network (Opt-ConvNet) is introduced, incorporating depthwise separable convolutions (DSC), low and high-level feature fusion (LHFF), and adjustable tuning parameters (ATP) to enhance feature extraction and classification performance. The proposed model exhibits robustness to noise and demonstrates strong applicability to real-world datasets. Experimental results validate the effectiveness of the method, achieving 99.2% classification accuracy for 55 synthetic PQD classes at a signal-to-noise ratio (SNR) of 20 dB and exceeding 97.5% accuracy across three different experimental datasets at the same SNR level. Beyond PQD classification, the proposed methodology has the potential for broader applications in power system signal processing.