Generator condition monitoring method based on SAE and multi-source data fusion

Abstract

With the increasing number of units involved in power system regulation and the increasing proportion of industrial load, a single data source has been unable to meet the accuracy requirements of online monitoring of unit conditions in the new power system. Based on the stacked autoencoder (SAE) network, combined with multi-source data fusion technology and adaptive threshold, a generator condition monitoring method is proposed. First, a SCADA–PMU data fusion method based on the weighted D–S evidence theory is proposed. Second, the auto-coding technology is introduced to build a stacked self-coding deep learning network model, extract the deep features of the training dataset, and build a generator fault detection model. Finally, by smoothing the reconstruction error and combining it with the trend change in the state monitoring quantity detected by the adaptive threshold, the fault judgment is realized. The simulation results show that, compared with the traditional method based on a single data source, the proposed method has higher robustness and accuracy, thus effectively improving the refinement level of generator condition monitoring.