PV Residual Current Classification and Identification Based on Support Vector Machine

In order to solve the problem that the residual current threshold discriminating mechanism of grid-connected dc systems lacks the ability of fault type identification, a photovoltaic (PV) residual current identification method based on support vector machine (SVM) classification is proposed based on the difference between PV residual current and dc residual current in time–frequency domain characteristics. First, the difference between the two residual current characteristics is analyzed by simulation and experiment. Then, the residual current signal is decomposed and reconstructed by wavelet packet transform to enhance the characteristic difference. Use the random forest algorithm to compare the distinguishing ability of different features to the PV residual current so that the original residual current waveform is transformed into four feature quantities for the subsequent algorithm input: information entropy, margin factor, pulse factor, and kurtosis. Finally, a classification and identification algorithm based on Bayesian optimized SVM is proposed, which achieved 100% identification accuracy of PV residual current. A PV residual current detection (RCD) and recognition prototype is developed to verify the reliability of the proposed algorithm. The work provides an effective solution for the safety protection of residual current and fault line selection in a dc distribution system.