Modified Multi-Feature Fusion of ITD and DNN for PV Microgrid Fault Detection

Aiming at the problem of waveform distortion and false components in the Intrinsic Time-scale Decomposition (ITD)algorithm, an Modified Intrinsic Time-scale Decomposition (MITD) algorithm is proposed. Compared with the traditional ITD, the MITD algorithm can not only smooth the signal, but also retain the characteristic data effectively. In order to solve the problem of fault detection in photovoltaic (PV) microgrids, a multi-feature fusion fault detection scheme combined with MITD-DNN is designed. Firstly, the three-phase voltage amplitude of the branch is collected, the modified ITD is used to decompose the collected data and extract the characteristic values, after feature selection, the fused feature matrix was trained in deep neural network (DNN), and fault types and fault phases were accurately identified according to the classification results of deep neural network. In order to evaluate the performance of the proposed fault detection scheme, a microgrid based on IEC-61850 standard is comprehensively evaluated and studied, and the test results prove the effectiveness of the scheme.