Fault Prediction Of Wind Turbin Based On D-S Evidence Fusion

With the rapid consumption of non-renewable energy,energy problem has been the urgent need to satisfy,wind energy has get much attention among those renewable and clean energy.Wind power generator is a key component of energy conversion,and wind turbine fault diagnosis and maintenance is the primary condition to ensure the normal operation of the wind turbine.Wind turbines often installed in extreme environment or the sea level,the traditional equipment maintenance is waiting until the serious damage and then sent maintenance personnel to repair,which not only waste a lot of manpower,sometimes because of the wind operating for a long time with damaged,resulting in irreversible fault,so how can predict the fault of wind turbine as soon as possible has become a problem worthy of studying.This thesis is based on history data and repairment log of Tuoshan wind field in Dalian,aimed at establishing fault forecast model of double-fed induction generator.The fault of the wind turbin to be identified includes the stator coil short circuit,the rotor coil short circuit,the bearing damage and the rotor eccentricity,the first two belong to the electrical fault,and the latter belongs to the mechanical fault.We extract vibration and current feature vectors on frequency domain through wavelet packet,and then establish fault forecast model through the D-S evidence fusion theory.The traditional fault diagnosis is analyzing the fault that machine is already damaged,so the diagnosis model only can be used to predict the fault kind which machine is damaged.This paper build the prediction model from the data set which wind turbin is still running until one hour later,although the wind turbin is still in operation,but the vibration parameters and current parameters have been abnormal.We can stop the wind machine running once found the abnormal parameter state,which can prevent the wind turbin been irreversible damaged if continue operating,also the prediction of the fault type can be a great reference for maintenance personnel.The traditional fault diagnosis is mainly aimed at the frequency characteristics of vibration signal or current signal to construct a pattern recognition classifier such as support vector machine or neural network,this thesis has used D-S evidence fusion method to combine two space classifier from vibration and current feature vector.Because the generator is a coupled system through the electromagnetic field,so the abnormal current characteristics can be used to assist in the diagnosis of mechanical failure,abnormal vibration characteristics can also be used to assist in the diagnosis of electrical fault.D-S evidence fusion needs basic probability assignment of the recognition framework,and standard support vector machine can output prediction probability distribution after transformation,the probability distribution can be used as the basic probability assignment of BPA at the same time,the support vector machine has a prediction accuracy of each classification,the prediction accuracy is the local credibility,with local credibility as weights can modify the initial BPA,after corrected can make two evidence fusion conflict factor smaller,the experiments show that the fusion fault prediction model has higher accuracy compared with non-fusion.