With the continuous increase of units involving in the regulation in power system and the gradual increase in the proportion of industrial load

a sole data source has no way to meet the accuracy requirement of on-line monitoring of unit status in modern power systems. Therefore

in this paper a kind of condition monitoring method of generator is proposed in combination with a stacked denoising autoencoder network and multi-source data fusion technology. Firstly

a SCADA-PMU data fusion method based on weighted D-S evidence theory is proposed. Then

the auto-encoder technology is introduced to construct a staked denoising autoencoder deep learning network model

extract deep features from the training dataset to construct the fault detection model of generator. Finally

the fault determination is achieved by smoothing reconstruction errors and detecting trend variation in monitored parameters using an adaptive threshold. The simulation results show that the method proposed in this paper

compared to traditional method based on a sole data source

has superior robustness and accuracy

thereby enhancing the precision of fault diagnosis and condition monitoring of generator.