Fault Prediction And Diagnosis Of Wind Turbine Drivetrain Based On LSSVM

Because of energy shortage and serious environmental pollution,in recent decade s wind energy source becomes a rising concern in the world.A large number of win d power plants have been built at home and abroad,the wind power installed capacit y and power generation have increased rapidly.With the rapid development of wind power industry,because of remote location,poor working environment and inefficient operation and maintenance of wind power plants,wind turbine drivetrain sometimes b reak down.It results in long downtime and reduces power generation and economic benefits.It is necessary to research the fault prediction and diagnosis of wind turbine drivetrain.Current fault prediction is mostly aimed at a single component and a system or a whole machine can't be considered comprehensively.In this text drivetrain is regar ded as a whole and on the basis of the fault mechanism of the drivetrain,the tempe rature monitoring amounts of 4 key positions of the drivetrain are selected as the fau lt indices.The LSSVM regression model is built,its inputs are wind speed,ambient temperature,active power and 4 related previous temperature monitoring amounts of drivetrain and outputs are 4 related temperature monitoring amounts.The fitting effec t of LSSVM regression model is compared with NSET regression model.Deviation o f predictive value and actual value of the model is defined as 'deviation value' of th e drivetrain.The GMM can fit probability distribution of deviation values of normal drivetrain and logarithm likelihood probability(LLP)of drivetrain can be calculated t hrough the GMM,which is fault quantification index of system.The potential faults can be detected by comparison of indices and threshold.The fault prediction method is proved to be sensitive and accurate with actual SCADA data.Rolling bearing is an important part of drivetrain and easy to damage.In the te xt,identification of bearing fault location is researched based on vibration signals of simulated faulted bearings in laboratory.There are 5 fault locations of inner racew ay,ball,outer centered raceway,etc.Firstly,the root mean square value,skewness,k urtosis,singular spectrum entropy,power spectrum entropy and wavelet packet energy entropy are extracted as the eigenvalues of faults identification from the original vib ration time-domain signals.But a sort of eigenvalue is not good for the direct identif ication of 6 fault locations.So the eigenvalues of 56 vibration samples and classification labels are computed and validated in the fault identification model based on PS O-LSSVM.The model compares with PSO-SVM,the experimental results indicate th at the LSSVM algorithm has higher fault identification accuracy,faster operation spee d and less time-consuming.