Generator fault detection method based on mixed attention mechanism

The invention discloses a generator fault detection method based on a mixed attention mechanism, and the method comprises the steps: firstly collecting an SCADA system monitoring signal during the historical normal operation of a generator of a wind turbine generator, and obtaining a training data set; secondly, establishing a long-short-term memory self-encoder based on a mixed attention mechanism, introducing a space attention mechanism into the encoder, introducing a time attention mechanism into a decoder, training a whole self-encoding network model by taking a minimum reconstruction error as a target, and extracting depth features of a training data set; constructing a generator fault detection model, firstly calculating an average value of each depth feature, then calculating a state index of each sample by adopting a mahalanobis distance, performing smoothing processing on a state index sequence of a training set, and calculating a health threshold by adopting a kernel density estimation method; and f