With the rapid development of wind power industry, a large number of wind turbines are put into use. In order to make wind turbine operate efficiently and reduce the high maintenance cost due to faulty downtime, wind turbine fault prediction methods were researched. Based on the supervisory control and data acquisition(SCADA) data of a 2 000 kW doubly fed wind turbine for 14 months, a convolutional neural network-bidirectional long short-term memory(CNN-BiLSTM) algorithm model was established. Firstly, a convolutional neural network was used to reject the abnormal noise data in the input data, extract the key local features of the input data, and simplify the complexity of the input data. Then the processed input data was put into a bi-directional long and short-term memory network to perform active power prediction, and the active power prediction was performed in a bi-directional long and short-term memory network. Finally, the processed input data was put into a two-way long-short time memory network for active power prediction, and the residual difference between the real and predicted values of active power was analyzed to complete the prediction of wind turbine faults. The results show that the constructed algorithm model has the stability of fault prediction, and can eliminate the misprediction caused by many factors, and make the fault prediction 4 days earlier than the SCADA system of the wind turbine, which provides a guarantee for avoiding the sudden shutdown due to the deterioration of the fault.