Classification method for multiple power quality disturbances via label distribution enhancement and multi‐granular feature optimization

Large‐scale integration of distributed generation and widespread use of power electronic equipment make power quality disturbances (PQDs) more complicated. There are still unsolved problems for the classification of multiple power quality disturbances (MPQDs) that consist of various kinds of single disturbances: (1) since the difference between the contribution degrees of every contained single disturbance to the composite disturbance is ignored, the logical labels cannot completely describe the composite disturbance; (2) the optimal feature in one granular space may not be optimal in another. These drawbacks lead to the degradation of MPQD classification, which should be considered a multi‐label model rather than single‐label model used in existing methods. Therefore, this paper proposes a novel method to improve the classification performance of MPQDs. The label distribution, representing the contribution degree of a single disturbance to the composite disturbance, is introduced. In addition, the high‐dimensional feature space is reduced by multigranular optimization, where the fuzziness and redundancy are removed by the modified rough‐set method. To improve the performance of the classifier, the ensemble classification model based on homogeneous classifier integration is proposed to integrate the base classifiers constructed by the feature vectors from different granularity spaces. A large number of field recordings are applied to validate the proposed method. The results show that the proposed method performs better than traditional methods, especially under noisy environments. This paper proposes a new power quality classification method via label distribution enhancement (LDE) and multi‐granular feature optimization. Thousands of samples according to field recordings are applied to validate the effectiveness of the proposed method.