Early Fault Identification Of Wind Turbine Based On Deep Convolutional Network

The operating conditions of wind turbines are complex and changeable,and the probability of failure is high.Early detection of unit faults and timely treatment can effectively reduce downtime,avoid major accidents and reduce operation and maintenance costs.SCADA(Supervisory Control and Data Acquisition)system of wind field can provide a large number of unit related operating variables,such as power,temperature,wind speed,etc.How to mine effective information from these data and realize early fault identification of wind turbine is a hot research topic in the field of wind turbine fault prediction.Based on the deep learning algorithm,this paper extracted the time and space features of SCADA data of wind turbine to realize the early fault identification of wind turbine.The specific research contents are as follows:(1)Based on the time and space characteristics of SCADA data,an early fault identification method of wind turbine gearbox was proposed based on the combination of Atrous Convolutional Neural Network(ACNN)and Bi-directional Long Short-Term Memory Network(Bi-LSTM).ACNN with large receptive field was firstly used to extract the spatial characteristics of monitoring variables,and then Bi-LSTM,which could extract time series information from both positive and negative directions,was used to perceive the changes of spatial characteristics in time series and monitor the health status of units.By analyzing the SCADA data of wind turbine,it is proved that the proposed method can effectively identify the gearbox faults of wind turbine in early stage.(2)ACNN+ Bi-LSTM model can analyze SCADA from the perspective of space and time,but ACNN extracts spatial features and directly inputs Bi-LSTM,without considering the influence of spatial features on the prediction results of the model.The Self-attention Mechanism(SM)is added between ACNN and Bi-LSTM,and SM,which can capture global information,is used to adjust the weight of spatial features extracted from ACNN to improve the influence of effective spatial features on the model.In order to better show the fluctuation trend of the predicted results,the SG algorithm is used to smooth the predicted results.By analyzing the SCADA data of wind turbine unit,it is proved that this method can identify the gearbox fault and electrical fault of the unit earlier.