Health Status Evaluation And Fault Early Warning Of Wind Turbine Based On SCADA Data

The instability of operating conditions has brought numerous failures to wind turbine generators,and the abnormal state of the units has not been detected in time before the unit is shut down,resulting in unit shutdown,greatly reducing the power generation efficiency of the units.In order to detect abnormal conditions and fault trends of wind turbines in advance and reduce the economic losses associated with abnormal shutdown,it is necessary to evaluate the health status of wind turbines.Then,fault early warning is carried out for some units,so as to provide reference for operation and maintenance strategies for operation managers and reduce the loss of failure shutdown.The main research content of this article includes:(1)In this paper,a more reasonable and feasible wind turbine health status evaluation system based on fuzzy comprehensive evaluation is proposed.Firstly,relevant parameters of the gear box,generator,engine room,Control system And paddle changing system are extracted on account of the SCADA(Supervisory Control And Data Acquisition)data recorded during the motion of the unit with sampling time of 10 minutes.At the same time,the temperature parameters are modified to eliminate the influence of power and environmental factors.Then,according to the converted wind speed under the standard conditions,the working conditions are divided to determine the best fluctuation range of each index parameter under each working condition.And using respectively the methods of subjective and objective weighting method to comprehensively ascertain the weights of all levels of evaluation.A vague synthesis appraisal system with twolevel was built to evaluate the health status of wind turbines by using 22 parameters extracted from four main components.(2 Establish a combined fault warning model based on multiple linear regression and BP(Back Propagation)neural network.The representative parameters of each component are selected as the output of the prediction model and the parameters with strong pertinence are chose as the input parameters.Then the forecast models on account of multiple linear regression,BP neural network forecast model and optimal weighted forecast model are established.The residual of these three models were compared by training data,and the sliding window theory was introduced to improve the accuracy of early warning.Through example analysis,in contrast with to the previous methods,the health status assessment method presented in this article is more rational and reliable,and the results of assessment are more realistic.At the same time,the combined early warning model based on optimal weighting can detect component failures in advance and reduce false positives under normal operating conditions.