When the wind turbine fails, the sensor data implies the fault features. To mine multiple features from sensor data, a fault diagnosis model of wind turbine based on auto-encoder and one-dimension convolutional neural network (CNN) is proposed in this paper. A one-dimensional convolutional layer is constructed to identify multiple features of time series data, and the features extraction of time-sequence data of all sensors can be realized. Under the action of the full connection layer, faults can be identified accurately by adjusting the network structure and parameters. For sensor data containing noise in complex operation environment, a de-noising method based on auto-encoder is proposed. The de-noising effect of the auto-encoder reconstructs the noise signal into the original signal, which improves the recognition effect of the fault in the noise environment. The simulation results show that the proposed method has obvious advantages in accuracy and robustness compared with the model-based method and other data-driven methods.