Research On Anomaly Detection Of Wind Turbines State Parameters And Condition Assessment Of Wind Turbines

In recent years,wind power has developed rapidly in the world and has become the third largest source of power generation in China.As the most critical equipment of wind power plants,wind power units often work in the natural conditions with changeable climate and harsh environment.The operation and maintenance costs as high as 20%-35% seriously affect the economic benefits of wind farms and restrict the development of wind power industry.SCADA(Supervisory control and data acquisition)system is a commonly used operation status monitoring system for wind turbines.However,SCADA data are susceptible to natural conditions and operation conditions,so a large amount of status information contained in SCADA is difficult to be directly utilized.On the basis of studying the structure and principle of wind turbine units,this paper fully excavates the SCADA data of wind turbine units,and studies the abnormal detection of wind turbine state parameters and the whole machine state evaluation method based on data drive and machine learning algorithm.The main contents of this paper are as follows:Firstly,the operation principle and SCADA system of wind turbine are studied.Fully grasp the wind power unit composition,operation principle,each component fault statistics information.The SCADA system and its monitoring parameters of wind turbine are studied.The state parameters of wind turbine are defined and classified.Combined with the fan operating conditions and modeling requirements,the sample data selection and normalization method of the prediction model were determined,which provided the data basis for the subsequent research.Secondly,in view of the traditional input parameter selection method is lack of scientific rationality,fails to consider the question of redundancy between parameters,considering from the perspective of correlation analysis and prediction model,are established based on maximum information coefficient(Maximal information coefficient,MIC),and based on Back propagation neural network(Back propagation neural network,the BPNN)forecasting model of input parameter selection model.The comparison shows that the parameter selection method based on BPNN can further reduce the redundancy of input data and the complexity of the model on the premise of ensuring higher accuracy of the prediction model.Thirdly,the abnormal detection model of wind turbine state parameters is studied.The Least squares support vector machines(LSSVM)and Genetic algorithm(GA)were used to optimize the prediction model of BPNN state parameters,and the accuracy of each modelwas compared and analyzed.Finally,the residual error,mean square error(MSE)of the predicted parameters were calculated by the sliding window statistical method,and the threshold value was set based on the "3 principle",and a dual indicator anomaly detection model based on the residual error and MSE was established.Example analysis shows that the sliding window statistical method can eliminate the influence of random factors and reduce the misjudgment rate of the abnormal detection model.Finally,a wind turbine state evaluation model based on Extreme learning machine(ELM)and degradation index is constructed.Considering the influence of fan operating conditions on different evaluation indexes,the evaluation indexes are divided into two categories,and the first category is combined by function fitting to obtain its dynamic threshold.Then,the relative deterioration degree is introduced,and the deterioration degree of the two indexes is taken as the input of the state evaluation model.Secondly,a new neural network ELM is adopted to construct a state evaluation model of wind turbine,which is compared with the state evaluation model based on Random forest(RF)and BPNN.The advantages of the data processing method and the state evaluation model are verified by an example.