In order to research the significance of vibration noise characteristics of wind turbine on the diagnosis of operation conditions and fault of the unit，taking a permanent magnet synchronous wind turbine as an example，a multi － source data fusion noise prediction model based on genetic algorithm － based support vector regression ( GA － SVＲ) is established． The vibration noise of the unit under different operation conditions such as no － load，load and variation of wind speed are simulated， and the sample data after processing based on information entropy theory are set up． The radial and axial of generator shaft and the high － speed shaft and low － speed shaft of gearbox are selected as input variables as well as the noise data as output variables，so a GA － SVＲ feature － level fusion prediction model is established，and it is verified with the measured data． The results show that the prediction model can obtain more accurate noise fluctuation trend and predictive values in the noise prediction application，which has a practical application feasibility．