Owing that the monitoring variables of the wind turbine supervisory control and data acquisition(SCADA) system are highly correlated and coupled, a wind turbine fault detection method is proposed based on multi-parameter correlation coupling and mutual information auto-encoder. The correlation matrix of multidimensional time series is established to describe the coupled relationship of SCADA data. The mutual information based variational auto-encoder is proposed for the encoding-decoding reconstruction of the correlation matrix. The wind turbine health evaluation index is constructed based on the reconstruction error of the correlation matrix, and then the early fault threshold is set through the update iteration of the exponential weighted moving average model. The proposed method is verified using the SCADA monitoring data of wind turbines in two wind farms. The results show that this method can effectively mine the internal correlation coupling information of SCADA multivariable time series, which can effectively improve the accuracy of wind turbine anomaly detection and the robustness to environmental interferences.