The invention discloses a rotor turn-to-turn
short circuit diagnosis method based on an exciting current-
shaft voltage harmonic nonlinear correlation coefficient. The method comprises the following steps: synchronously collecting parameters, and preprocessing the collected parameters through
wavelet threshold denoising and temperature normalization; extracting
harmonic analysis according to the
shaft voltage harmonic characteristics; identifying magnetic saturation, temperature and load nonlinear factors and calculating corresponding coefficients; performing dynamic correction on the Pearson's
correlation coefficient through the nonlinear correction coefficient, and combining with
fuzzy logic weighting to obtain
fuzzy logic weighting; defining a working condition
feature vector, and constructing a working condition adaptive dynamic model comprising a
linear regression basic model and an LSTM neural network; and calculating a dynamic threshold value based on the basic threshold value and the nonlinear factor, and judging the turn-to-turn
short circuit trend of the rotor according to the comparison of the
correlation coefficient and the dynamic threshold value and a prediction result. According to the invention, through combination of the nonlinear correction
correlation coefficient and the LSTM neural network, dynamic compensation of nonlinear factors is realized, and the accuracy and reliability of rotor turn-to-turn
short circuit fault diagnosis are improved.