Aiming to the problem that high cost and error of traditional fault monitoring method， a fault monitoring method is proposed based on the temperature model for fault monitoring of the wind turbine main bearings. The multiple linear regression model， grey model， support vector machine regression model and their combination forecasting model of main bearing temperature under normal operation are established respectively. Based on the combined prediction model， the sliding window method is introduced to analyse the change of temperature residual mean and standard deviation in the faulty condition of the main bearing. The failure of the main bearing is judged by comparing the confidence interval of the mean value or standard deviation of temperature residual with the set critical value. The results indicate that， the determination coefficient of the main bearing temperature combination forecasting model value is 0.049 3， 0.002 7 and 0.000 2 higher than that of the multivariate linear regression model， the grey prediction model and the support vector machine regression model， the combination forecasting model by introducing a sliding window method can reflect the abnormal situation of main bearing temperature in time， which provides high-precision monitoring of the main bearing fault state and formulating scientific and healthy maintenance strategy .