Abstract: Against the backdrop of the large-scale development and cost reduction of wind turbines, coupled with the application of lightweight blades made of carbon fiber composite materials, blade fracture accidents occur frequently. Currently, ensuring the trouble-free operation of wind farms is the primary concern of customers. This paper establishes a self-coding neural network model based on Long Short-Term Memory(LSTM) and a multivariate control chart model based on the Bootstrap threshold respectively. The operation data of wind turbine blades are analyzed, and the abnormal states are monitored. Both methods can effectively suppress noise and identify the abnormal states of the blades.