Disturbance Classification Method for Microgrids Based on EEMD-Transformer-SVM

Abstract

Accurate real-time power system disturbance classification is beneficial in avoiding system faults. However, in the process of disturbance detection, the quality of data obtained from the synchronous phase measurement unit (PMU) can be problematic, seriously affecting its application in disturbance classification. Moreover, existing methods are unable to accurately classify data with excessive noise. To address this problem, a disturbance classification method based on ensemble empirical mode decomposition, Transformer neural network, and support vector machine (EEMD-Transformer-SVM) was proposed. First, considering the nonlinear and non-stationary characteristics of microgrid disturbance data, using ensemble empirical mode decomposition to extract data features could effectively reduce the difficulty of fitting nonlinear fluctuation patterns in machine learning models, while avoiding interference between local features. Moreover, to capture and amplify the effective information in the data, a Transformer with a multilayer self-attention encoder network was proposed, which could further transform the data features after EEMD. Finally, the features were classified using a support vector machine. Based on the Consortium for Electrical Reliability Technology Solutions (CERTS) microgrid system, the proposed method was tested under different disturbance data to verify its accuracy and efficiency. By testing the data classification performance in different scenarios, the method demonstrated a high level of generalization.