Multi‐task learning method for classification of multiple power quality disturbances

In this study, the authors propose a multi‐task learning with deconvolution network (MTL‐DN) method for the multi‐label classification of multiple power quality disturbances (MPQDs). First, the labels of MPQDs are assigned to three groups corresponding to three learning tasks and the label correlations among various PQDs are utilised in the joint learning of interrelated tasks. A weighted joint loss function is adopted to balance multiple tasks and to ensure that all of the tasks achieve the global optimum. Second, considering the effect of the pooling operation on transient disturbances, a deconvolution network is employed to reconstruct the erased feature and to merge them into high‐level feature for the final classification. Finally, the authors employed two sets of evaluation metrics to verify the validity of the MTL‐DN method and compared it with three state‐of‐the‐art multi‐label classification methods. Extensive experiments based on simulated and real‐world datasets demonstrated that their method performed better and it greatly improved the accuracy rate for identifying MPQDs under different signal‐to‐noise ratio conditions.