Research And System Development Of Early Warning Method For Wind Turbines Based On Deep Learning

With the vigorous development of the wind power industry,the high operation and maintenance costs of wind turbines cannot be ignored.In order to detect early failure signs of wind turbines and reduce the cost of operation and maintenance of wind turbines,this paper proposes an early warning method for wind turbines based on deep learning.The monitoring method includes SCADA(Supervisory Control and Data Acquisition)data early warning method.And wind turbine blade surface damage detection method,and finally designed a wind turbine early warning system based on actual needs.The main research contents are as follows:(1)The SCADA data of wind turbines were analyzed,and a temperature-power abnormal data identification method based on working condition division and segmented quartile method and a wind speed and power abnormal data identification method based on the optimized Kmeans-LOF algorithm were established.The results show that these two methods have better results in identifying and cleaning SCADA data anomalies than other traditional methods.(2)An early warning method for wind turbines based on LSTM-SDAE network is proposed.Use a sliding window to smooth the reconstruction error to reduce false alarms.The kernel density estimation method is used to obtain the probability distribution of the reconstruction error,and the alarm threshold is obtained.The results of the actual wind turbine operating data verification show that the LSTM-SDAE network can detect early failure symptoms before components fail,and has lower mean square error than other models.(3)A method for surface damage detection of wind turbine blades based on Mobile NetYOLOv4 network is established.Use image preprocessing methods such as image enhancement,image noise reduction,and transfer learning ideas to enhance the detection accuracy of the model.The results show that the m AP of the Mobile Net-YOLOv4 network is about 2% lower than that of the YOLOv4 network,but the number of network parameters is 1/5 of that of the YOLOv4 network,and the network training speed is 5 times that of the YOLOv4 network.(4)Combined with the above-mentioned early warning methods and blade surface damage detection methods,through the analysis of the system requirements,system architecture selection and database design,an early warning system for wind turbines is finally developed.