The Research On Condition Assessment Of Wind Turbine And Early Warning Of Gearbox Based On Roughset&Aritificial Neural Network

With global wind power industry developing fast and installed wind capacityincreasing significantly, complicated structure and operating environment makeswind turbine face an unprecedented challenge on operation and maintenance.Real-time condition assessment and early warning of abnormal operating conditionoffer a safe and efficient operating of wind turbine, having a great significance tooptimizing maintenance strategy. This dissertation presents a method for real-timecondition assessment of wind turbine and early warning of gearbox on the premise ofno extra sensors, that based on a large number of monitoring data offered by SCADAand considering the relationship among the monitoring projects.The characteristic parameters of the model are selected by the attributesreduction algorithm of the rough set theory. The attributes reduction, which makesuse of attributes’ importance as heuristic information, is implemented again on thebasis of ordinary reduction, combining with discretization method of naive algorithm,semi-naive algorithm and equal frequency binning and attributes reduction algorithmsof genetic algorithm and Johnson’s algorithm. The characteristic parameters of themodels are selected based on the reduction result of each method. The characteristicparameters of the real-time condition assessment model are reduced with windturbine active power as decision attribute and other SCADA monitoring variables ascondition attributes. While the characteristic parameters of gearbox’s early warningmodel are reduced with the temperature of gearbox’s input shaft and output shaft andgearbox oil as decision attribute respectively and other SCADA monitoring variablesas condition attributes.The proposed real-time condition assessment model combines level-limit alarmsystem of SCADA and active power difference value between monitoring value andpredicted value of forecasting model. According to data verification, the proposedmodel based on rough set and TS fuzzy neural network is superior to the BP neuralnetwork and SVM model in terms of the prediction accuracy and it can accurately identify the abnormal operation of the wind turbine. The early warning model ofgearbox is based on the prediction models of the temperature of gearbox’s input shaftand output shaft and gearbox oil. Other trend prediction models ofgear box temperature only considered the temperature monitoring values fora period of time before the current value. While the proposed model takes intoaccount other monitoring variables’ influence on the gearbox’s temperature toestablish TS fuzzy neural network prediction model. The model can alarm that thegearbox is in abnormal conditions when the difference value between monitoringvalue and predicted value of forecasting model about gearbox’ temperature is overlimit.This dissertation presents models for real-time condition assessment of windturbine and early warning of gearbox based on the real data getting from SCADAsystem of wind turbine with good maneuverability and generalization.