Wind turbine generator gearbox bearing state prediction method

The invention relates to a wind turbine generator gearbox bearing state prediction method, which comprises the following steps of: screening variables which are strongly correlated with a gearbox bearing state as input variables of a stack noise reduction self-encoding network structure through a mutual information method by taking unit operation data acquired by a wind turbine monitoring and data acquisition system as basic data; the method comprises the following steps: carrying out stack noise reduction on a high-speed shaft bearing temperature, carrying out dimension reduction processing on the high-speed shaft bearing temperature, carrying out modeling on variables processed by a stack noise reduction self-encoding network through Gaussian process regression, establishing a gear box bearing state prediction model, and monitoring a prediction residual error change trend of the high-speed shaft bearing temperature by using a sequential probability ratio test method. Therefore, the temperature abnormity of