A method for the behavioral analysis of partial discharges in hydrogenerators by generalized linear models

This study proposes a method for the analysis of partial discharges in hydrogenerators through generalized linear models from data collected by capacitative couplers associated to phases A, B and V for the measurement of partial discharges in the stator windings of a hydrogenerator located at a hydroelectric plant. The dependent variable was considered the number of partial discharge events, explained by the variables of amplitude and phase angle. The applied models used to represent the data were the Poisson and Negative Binomial models. The latter showed the best fit. With the estimates of the model, it was possible to observe that the probability of partial discharge occurrence increased for low amplitude intervals and that these were distributed throughout all angle classes.