Unsupervised Disturbance Identification Using Synchronous Phasor Measurement

The task of disturbance identification is to quickly identify disturbances occurring in the power system and determine their type, while traditional disturbance identification methods usually start with supervised feature extraction, relying on learning the features of specific events to classify disturbances in the streaming data, and the labelling work required for this type of feature extraction is difficult to carry out in practice due to the high complexity and workload. To address this issue, this paper proposes a disturbance identification method using synchronous phasor measurement. The method utilizes a Generative Adversarial Networks (GAN) model for unsupervised feature extraction, and a three-window parallel framework is established by targeting the time-scale features of different disturbances. Finally, the disturbances are classified by the features extracted from the above framework. The proposed method can be used for power system disturbance identification in the case with no labelled event, and the method can identify transmission network events while providing a warning for power quality events in the distribution network where the synchronized measurement device is located. Its effectiveness is verified in field data.