Abstract: To mitigate the annotation cost of Complex Power Quality Disturbances (CPQDs) data, a hybrid strategy active learning approach was employed along with Laplacian Extreme Learning Machine to identify CPQDs in power distribution networks. It combines different categories of active learning sampling strategies to select the most informative and representative CPQDs data for labeling. In the active learning process, the weights of different strategies are adaptively adjusted using a logarithmic function. To enhance classifier performance, a Laplacian manifold regularization was embeded into Extreme Learning Machine in both supervised and unsupervised frameworks. The proposed architecture is compared with state-of-the-art active learning algorithms on datasets generated through code synthesis and hardware, demonstrating superior performance of the proposed method.