A Novel Classification and Identification Framework for Power Quality Disturbances Via Constraint and Prior Knowledge Guided Multi-Label Learning

Abstract The accurate detection and classification of power quality disturbances (PQDs) is a crucial step to ensuring the stable operation of power systems. Unfortunately, the current PQDs identification and classification methods suffer from high model complexity and large parameter counts, which significantly limits its applicability on embedded devices and other mobile terminals. To address this issue, a novel classification and identification framework for power quality disturbances via constraint and prior knowledge guided multi-label learning (CPGML) is proposed in this paper. This method leverages constraint networks and prior knowledge to reduce the difficulty of recognizing disturbance types for the main network, which improves recognition accuracy with only a slight increase in model complexity. In this method, PQDs feature information is first extracted in the main network using a one-dimensional convolutional and group-convolutional network (1D-CGCN). Then, a SimAM (a simple, parameter-free attention mechanism)-optimized Gated Recurrent Unit (SimGRU) is utilized to capture the temporal features of PQDs. Additionally, a constraint network is designed to identify the number of disturbances within PQDs, and prior knowledge is incorporated to assist recognition. Finally, the predictions of the constraint network and the outputs of the main network are combined using a label threshold function based on the constraint network to obtain the final classification results. The simulation and practical testing results demonstrate that the proposed CPGML-based method significantly reduces model complexity while achieving high accuracy and lightweight recognition of complex disturbances.