Research On Fault Detection Of Wind Turbine Based On Subspace Identification And K-L Divergence Analysis

Along with the increasing of the volume of wind turbines,the structure is more complex,the installation location is becoming more and more remote and so on various aspects factor,the environment faced by wind turbines more complex,especially offshore wind turbines,which will result in higher fault rate of wind turbines,as a result,workers have to regular maintenance,this will cause wind turbines downtime has greatly increased,at the same time the corresponding operation maintenance cost also increase greatly.In addition,if the fault of the wind turbine cannot be found and solved in time,it will cause greater damage to the wind turbine and cause unnecessary accidents.Therefore,it is necessary to predict the faults of wind turbines at an early stage.The residual often contains the fault information of the system,so the generation of the residual is an important part of the fault detection process.However,the wind turbine is faced with various uncertainties,noise interference and other factors in the operation process,which will have a certain impact on the generation of residual.In addition,the switching of wind turbine operating conditions will also make the residuals under different operating conditions have different statistical characteristics,which will affect the accuracy of residual evaluation.Finally,false alarm and missed detection frequently occur in the process of fault detection.To solve the above problems,the thesis proposes a fault detection method for wind turbine based on subspace identification and Kullback-Leibler(K-L)divergence analysis.Firstly,for various uncertainties and noise interference problems,the method of residual generation based on subspace identification is adopted,and the robustness index is introduced to make the obtained residual data sensitive to faults and robust to unknown interference and other factors.Then,in order to avoid the influence of wind turbine operating condition switching on the generated residual,the operating conditions were divided according to the control strategy of wind turbine to obtain the residual data under different operating conditions,so that the residual can be effectively evaluated in the next step.Finally,in order to further reduce the false alarm rate and missed detection rate of fault detection,residual evaluation method based on K-L divergence analysis is adopted to analyze the difference of residual probability density function,and gradient descent algorithm is used to optimize the threshold,and K-L divergence and threshold are compared to realize the fault detection of wind turbine.The proposed method is applied to the NREL-5MW offshore wind turbine benchmark model,at the same time,10 kinds of sensor and actuator faults generated in the simulation of 3 groups of different average wind speeds were diagnosed,and compared with other methods,the statistical results show that the accuracy of this method can reach above 90%,which is better than other methods.