By establishing a data-driven fault prediction model,the fault state can be separated from the normal state,and the accurate diagnosis of wind turbine fault can be realized.Therefore,a wind turbine fault diagnosis strategy based on Markov clustering and feedforward neural network was proposed.Data clustering was realized through Markov distance evaluation,and normal data and abnormal data were separated.Then,based on feedforward neural network,three prediction models of wind turbine,gearbox and generator were constructed according to engineering experience.Finally,the proposed fault prediction strategy was verified by the experimental prototype data.The experimental results show that the proposed wind turbine fault prediction strategy can effectively identify wind turbine output power anomalies,gearbox temperature anomalies and generator temperature anomalies,which is conducive to reasonable maintenance scheduling.