Fault diagnosis of permanent magnet direct-drive wind turbine converter based on lifting wavelet and random forest

Abstract

The wind power converter is very prone to failure in the permanent magnet direct-drive wind power system. For the healthy operation of permanent magnet direct-drive wind power system, a wind power converter fault diagnosis method based on improved wavelet and random forest is proposed. In Simulink platform, a model of permanent magnet direct-drive wind power system is built to simulate the normal and fault states of wind power converter to obtain converter fault signals. The lifting scheme is used to construct wavelet to obtain lifting wavelet, and the genetic algorithm is used to optimize the lifting wavelet to construct the optimal lifting wavelet. Then the relative wavelet energy, waveform index, margin index and peak index are calculated by the detail coefficients of the lifting wavelet decomposition. The converter fault feature vector is obtained, and finally the feature vector is input to the random forest for converter fault diagnosis. It is shown that the random forest model has good classification performance for wind power converter fault diagnosis in comparative analysis with extreme learning machine and BP neural network.