Data-based Condition Prediction Of Wind Turbine Generator System And Outlier Identification Of Pitch System

Wind power generation develops rapidly, so that the capacity of wind turbinegenerator is larger and larger and the control system becomes more complex. Thelarge-scale wind farms are also developing from onshore to offshore, even to abyssalregions. However, with the construction and operation of large-scale grid-connectedwind farms, the high failure rate and operation maintenance costs have become an issue.So it is important to improve the operational reliability and utilization of Wind TurbineGenerator System (WTGS). Based on mastering the work and control principle of agrid-connected wind turbine, by using the operation data and fault information，condition prediction of WTGS and outlier identification of pitch system were analyzedin this thesis. The following research works were to be carried:①Based on in-depth understanding of the operating principle and basic structureof a grid-connected WTGS, the control system of WTGS and itsmonitoring of operating parameters are presented. The failure mode andfailure mechanism of the subsystem with high failure rate are also analyzed.②Considering the distinction of the indicating trends and the indicatingvariable rules of the historical data based on different time intervals, by extracting themulti-group unequal interval time sequence from the monitored data of a WTGS, byusing the average weakening buffer algorithm, the non-equidistant grey predictionmodels GM(1,1) of the selected data time sequences are established, respectively.Secondly, by introducing the concept of the relational degree, the prediction result withthe maximal relational degree was selected by comparing with the actual measuredvalues. In addition, the running conditional parameter of the rotor speed and thetemperatures of main components of the practical850kW variable speed constantfrequency WTGS were forecasted by using the proposed non-equidistant greyprediction model. Finally, the rotor speed of a2MW WTGS was used to forecast, andthe prediction results were compared with that of BP neural network and support vectormachine (SVM) methods.③In order to assess and predict the failure of WTGS accurately,it needs to get thecharacteristics operating parameters of the faults firstly. The pitch control system istaken as an example to select the characteristics operating parameters of the faults.According to the operation data of the pitch control system and its fault information, the normal operation data set and fault data set of the pitch control system are constructed.The characteristics operating parameters of the pitch control system faults are selectedby the Relief Algorithm. The effectiveness of the selected characteristics operatingparameters is verified with BP neural network classifier.④By using characteristic parameters of pitch systems failures of wind turbine,outlier identification of pitch system based on the distance are studied. Firstly, theobservation vector of the pitch system under normal operation is constructed. Secondly,the regression model of observation vector, which inputs wind speed and outputsobservation vector, is established by using Support Vector Regression (SVR)theory.Then, calculate the distance between the observation vector and measured vectorand identify outlier state of the pitch system. Finally, the normal and abnormal operatingdata of2MW wind turbine pitch system are used to validate the proposed outlieridentification algorithm.The research results can be applied to the wind turbine control system and windfarm SCADA system to assess the status and predict the faultsof the wind turbineaccurately.