Abstract: To identify the operational state of key components in the wind turbine, we propose a denoising autoencoder with time-dimension information for fusing the multiple sensor signals of these key components. Firstly, the relationship between the signals of each sensor in the SCADA data is learned and the change trend of the operational state in the time dimension is obtained to establish a normal model of the key components. Then, real-time data collected by the wind farm is input into the normal model, and the operational state of the key components are identified based on the distribution of the output residual of the normal model. Finally, we use actual fault data from the generator and gearbox in the wind farm for verification. The results show that obtaining the state change trend in the time dimension improves the ability to accurately describe the operational state of the key components, and the proposed method effectively improves the accuracy of the identification of the operational state of these key components.