Abstract: Early fault warning through wind turbine condition monitoring can prevent further development of faults and reduce wind farm operation and maintenance costs. To fully explore the time sequence information of parameters of wind turbine supervisory control and data acquisition（SCADA）and the nonlinear relationship between them,A wind turbine condition monitoring and fault warning method based on deep convolutional auto-encoder（DCAE）which combines the auto-encoder（AE）and convolutional neural network（CNN）is proposed. Firstly,based on the historical SCADA offline data,the DCAE for wind turbine condition monitoring is established.Then the reconstruction error is analyzed to determine the alarm threshold. Finally,the EMWA control chart is used to monitor the status of a wind turbine in real-time. Taking the failure of a 2 MW doubly-fed wind turbine blade as an example,the proposed DCAE method is verified. The results show that the DCAE method proposed in this paper is effective in early warning of the wind turbine failure,and is superior to the existing deep learning-based wind turbine condition monitoring methods. The proposed method significantly improve reconstruction accuracy,reduce model parameters and training time.