Energy Reports (Apr 2023)
Early warning of stator winding overheating fault of water-cooled turbogenerator based on SAE-LSTM and sliding window method
Abstract
Aiming at the early warning of overheating defects in the stator winding of water-cooled turbogenerators, this paper proposes a novel method based on SAE-LSTM and sliding window method by combining the Sparse Auto-Encoder (SAE) and the Long–Short Term Memory network (LSTM) with highly time dependent time series data characteristics. Firstly, the sparse auto-encoder is used to reconstruct the operation data collected by the Distributed Control System (DCS) installed on the turbogenerator to extract the data characteristics; Secondly, the LSTM prediction model optimized by attention mechanism is used to predict the outlet temperature of each slot of the stator winding of the turbogenerator under normal working condition; Then, the sliding window method is adopted to detect the stator winding overheating defect, and the alarm threshold is defined based on both the maximum mean value and maximum standard deviation of the predicted residual within the window. Finally, the proposed method is validated by using the historical DCS data of a turbo generator with stator winding overheating defect before failure,and the results show that compared with the traditional threshold warning method, the proposed method can warn the defects 85 h in advance, which provides strong support for the stable operation of the turbogenerator.
