Complex Power Quality Disturbances Classification Based on Multi-label Active Learning

Power quality (PQ) disturbances generated during the power grid operation are complicated and volatile in real life. If a large number of complex PQ disturbances (CPQDs) from power grid monitoring devices are all labelled artificially, then it may consume a lot of human resources. To effectively utilize these unlabeled data collected by the monitors to improve the accuracy of the learning model, this study proposed an approach for the recognition of CPQDs using a multi-label active learning strategy. First, the study presented a novel active learning strategy based on label exclusiveness and ranking score (LERS) by analyzing the label relation among different PQ disturbances. Second, the strategy was incorporated into the multi-label extreme learning machine classifier to train and identify CPQDs. Finally, extensive experiments in the study validate the effectiveness of the proposed method. Results indicate that LERS improves the performance of the classification model by adding the most informative sample. As the number of labelled samples increases from 1000 to 8000, the evaluation metric MicroF1 reaches more than 0.7. The corresponding labelling cost of the proposed strategy is reduced by 40% compared with other strategies when obtaining certain accuracy. This study provides a specific reference for recognizing CPQDs and has a bright application prospect. © 2022. School of Science, IHU. All rights reserved.