To accurately predict the operation status of wind turbines

realize early warning of wind turbine failures

and reduce the maintenance and operation costs

this article proposes an QPSO-CNN-Bi-LSTM-Attention based wind turbine condition prediction method.Firstly

KPCA algorithm is used to extract principal components of wind turbine SCADA monitoring data

and SPE is calculated to construct training set and testing set of wind turbine condition Secondly

the hyperparameters of the QPSO-CNN-Bi-LSTM-Attention model are optimized based on the training set data of the normal and abnormal conditions of the wind turbine

and the SPE of the testing set data is calculated to realize early warning of abnormal state by contrast SPE with its threshold.Finally

the effectiveness and accuracy of the proposed method are verified by an example analysis of the SCADA monitoring data of 1.5 WM wind turbine in a wind farm in Inner Mongolia.Case studies show that the proposed method can better extract the principal components hidden in nonlinear data

and has better prediction performance and stability compared with CNN

LSTM and CNN-LSTM methods in condition prediction of wind turbines.