Research On Large Wind Farms Intelligent Maintenance System

By the end of 2015,the wind power installed capacity is 129 million kilowatts in China and 6.69 million kilowatts in Shanxi Province.At the end of 13th Five-Year Plan,the number is respectively expected to 240 and 18 million kilowatts.With the increasing of wind power installed capacity,the safe,efficient and economical operation of wind power becomes more and more important.Due to the long-term operation of the wind turbine in harsh environment,the failure occurred frequently,and the maintenance cost is high.Therefore,it is of great value to study on intelligent maintenance system of wind farm to improve the operation reliability,reduce the maintenance cost and optimize the maintenance decision.This article is carried out data acquisition,fault feature extraction,fault prediction,intelligent maintenance decision and system framework design.The main works are summarized as follows:Apart from the vibration data acquisition based on condition monitoring and fault diagnosis system,an increase of current,temperature,video,audio,smoke,grid parameters,which provides the basis of designing and developing intelligent maintenance system.The common failure modes including gear box,generator and other key components of wind turbine are analyzed,and the characteristic parameters of the vibration signal in time domain and frequency domain are given.In view of the existing feature extraction method is difficult to quantitatively describe the degree of equipment failure,the concepts of complexity is introduced.Due to the complexity metrics can be used to comprehensive measure the signal trends,a new method that the combination of wavelet packet,envelopment analysis and sample entropy fast algorithm to extract fault characteristic value is put forward.Through the analysis of wind turbine fault simulation experiment and field data,it turns out to be correct and effective method.On the basis of the feature extraction,fault prediction model and method are studied.Due to the traditional grey forecasting model has the defects such as accumulated data sequence requirements with exponential nature and low accuracy of multi-step prediction,grey model is improved by the method of Markov and equal dimension.By combining the improved grey model and Elman neural network,it is put forward to the prediction method of improved grey Elman neural network wind turbine vibration characteristics.After analyzing the cases,the method has higher prediction accuracy than the gray model or Elman neural network model alone and meets the requirements of engineering application.Then,the single variable model is further studied,and multivariable fault prediction model is proposed due to that the traditional single variable model has the disadvantages of single feature dependence,lack of information and low accuracy of prediction.Starting from the view of multivariable characteristics,the multivariate least squares support vector machine and parameter optimization are studied.In study of components degraded condition assessment of failure threshold setting method and fault prediction theory,the introduction of correlation analysis feature selection method,the method is proposed to wind turbine multivariable fault prediction based on PSO-MLSSVM.The feasibility of the method is proved by simulation and practical application.Based on the above research results,we have designed the intelligent maintenance system from the actual demand of wind farm,and established the working flow chart of the system.Besides,we have divided and introduced its function of each module in detail,and designed the convenient interface of maintenance system to meet the practical wind farm.