Research On Prognostic And Health Management Of Wind Turbine Based On Machine Learning

In recent years,with the rapid growth of wind power industry,people have higher requirements for the safety,stability and reliability of unit operation.In order to make up for the problems of "excessive maintenance" and "insufficient maintenance" existing in the operation and maintenance mode of wind turbines,fault prediction and health management(PHM)is introduced into the field of wind power,PHM is expected to find the early symptoms or minor faults as soon as possible,and make the corresponding maintenance plan in advance,which is of great significance to reduce the sudden outage probability of wind turbines,improve the safe operation of wind turbines and the production efficiency of the whole wind farm.Therefore,considering the operation safety of power grid and units,this paper carried out the research of wind turbine fault prediction and health management method based on machine learning,mainly including three aspects of fault diagnosis and fault prediction.(1)In the aspect of fault diagnosis,in order to solve the problem of sample imbalance caused by the lack of fault samples in fault diagnosis,this paper uses Conditional Generative Adversarial Networks(CGAN)to expand the fault samples,and then introduces Convolutional Neural Networks(CNN)model and sets a small convolution kernel to extract features more carefully,comprehensively and accurately,so as to avoid the impact of subjective screening input on the results and improve the efficiency The overall accuracy of classification diagnosis.(2)In the aspect of fault prediction,this paper establishes a Deep Belief Network(DBN)fault prediction model based on the monitoring data of SCADA system,and uses sliding window to process the data to realize online prediction.Because the internal rules of data will be destroyed in fault state,this paper selects the threshold based on adaptive exponentially weighted moving average(EWMA)to detect the change trend of reconstruction error,and uses the failure rate as the decision criterion to determine the fault of wind turbine,so as to reduce the false alarm caused by the extreme value of wind speed,the cause of fault is preliminarily determined according to the residual variation of input and output.The experimental results prove the feasibility and sensitivity of this method.