Research On Data-driven Monitoring And Warning Technologies For Critical Compoents In Wind Turbines

Along with the rapid development of the global wind power industry,the single-unit capacity of wind turbines has gradually increased and the total installed capacity has increased.The problems such as high unit failure rate,low operating efficiency,short life and poor reliability have become increasingly prominent.Therefore,it is of great academic research significance and engineering application value to carry out research on wind turbine condition monitoring and early fault warning technology for reducing the failure rate and operation and maintenance cost,ensuring the safe operation of the unit and maximizing the economic benefits of the wind farm.The thesis addresses the application requirements for condition monitoring and fault warning technology in wind farms,based on the monitoring data of the Supervisory Control and Data Acquisition(SCADA),specifically focusing on several critical components,including gearbox,generator and converters.Based on the latest theoretical techniques in the field of machine learning and data mining,this thesis aims to study data-driven condition monitoring and fault warning methods for wind turbine critical components,which can provide technical supports to achieve effective monitoring,accurate assessment and fault warning for the critical components.The main work can be summarized as follows:Firstly,taking the gearbox as the research object,in order to solve the problem that traditional static monitoring methods are difficult to effectively monitor the operating status of wind turbine gearbox,a health monitoring framework based on the echo state network and the extreme theory is studied.On the one hand,to better capture the complex non-linear relationship between different SCADA monitoring variables,a gearbox SCADA vibration prediction model based on the echo state network is proposed.On the other hand,aiming at the problem of low detection accuracy and poor adaptability of traditional monitoring methods,a dynamic threshold monitoring method based on the extreme theory is proposed to determine the dynamic control limit.The effectiveness of the proposed monitoring method is verified by the real SCADA data collected from wind farms.Secondly,taking the generator as the research object,a multi-parameter fusion monitoring approach is proposed based on multimodal deep denoising autoencoders to deal with the complex multivariate SCADA data with the high-dimensionality,nonlinear correlated characteristics and redundant information.The relevant variables related to the operating status of the generator are first selected to build multi-parameter fusion monitoring model.Then a deep denoising autoencoder network based normal behavior model is built using the normal data collected from wind turbines,and accordingly,the detection thresholds are determined by using the kernel density estimation method.When a fault occurs,the relations of different generator monitoring variables will be disturbed and the build health monitoring model cannot well reconstruct the faulty input data and produce a larger reconstruction error.Once it exceeds the pre-defined threshold,the fault alarms will be triggered.The proposed approach was evaluated using the measured SCADA data and the comparative results with traditional multivariate statistical monitoring method demonstrate its effectiveness and superiority.Thirdly,taking the doubly-fed converter as the research object,the mathematical model of the doubly-fed converter is established,and then to address the high-frequency resonance detection and suppression from the grid-side converter,a virtual damping method based on the current of the grid side converter as a state variable is proposed.With the proposed method,the effect of the series inductance of the filter inductor is realized by software control,and the high frequency resonance suppression is realized without increasing the hardware cost.Furthermore,an intelligent monitoring system for the wind turbine converters are designed and developed to measure the corresponding the alternative current and voltage data,the input and output status of the signal feedback switch and to record and save data waveforms.Also,the system can have the capability of the remote transmission of the recorded data via the Ethernet network,which can provide accurate and reliable data support and assisted analysis for the converter fault diagnosis for technical professionals.The current and voltage waveforms under different scenarios of the normal grid connection,the normal offline grid and the faulty offline grid,are recorded and then are used to test the performance of the developed intelligent monitoring systems for converters.Lastly,to meet the requirements for the practical application of the wind farm,to improve the monitoring and diagnosis ability of the existing SCADA monitoring systems,a health monitoring and fault warning system is developed to integrate the statistical analysis functions of the original SCADA monitoring system.Specifically,an abnormal monitoring and performance evaluation module is designed to extend the monitoring capability of the existing SCADA system.The developed system has been tested in a wind turbine from a real wind farm.