Research On SCADA Data-driven Wind Turbine Modeling And Failure Early Warning Methods

As the country vigorously supports the development of new energy sources,wind energy has been widely used as a new energy source with a large amount of storage,and the wind power industry has also developed rapidly.But with the operating environment of wind turbine is more complicated and the operating time increase,the frequency of wind turbine failures will also increase.At present,on-site operators are unable to deal with the fault in time,therefore the normal operation of wind turbine has greatly affected.It is of great necessity to detect the abnormal condition with high efficiency,aiming at improving the stability of the wind turbine,which can also avoid the serious accidents.Through the modeling and analysis of wind turbine,real-time monitoring of the system’s operating status and making early warning information of abnormal conditions are important research directions.This article first introduces the basic structure and operation control principles of wind turbine.On this basis,it focuses on the analysis of common failures and causes of the wind turbine.Then,it introduces the network structure of Elman neural network,NARX dynamic neural network and GRNN neural network,and the learning algorithm of each layer of the network.Finally,combined with the data collected by the SCADA system,using comprehensive correlation index to determine input and output parameters of neural network model more reasonably.Combining three kinds of neural networks to establish models under normal operating conditions for the wind wheel system,gearbox system and generator system of 1.5MW doubly-fed wind turbine.By adjusting the network parameters,all the models are optimized to be more accurate,and the prediction effect of the models is also satisfying.Based on the normal condition model,the sliding window method and the statistical process control method can determine the corresponding system operation index thresholds,respectively.Through these thresholds,online status monitoring of the system can be achieved,so as to achieve the effect of fault warning.The results show that the neural network method mentioned in this article can achieve the highest accuracy of the wind turbine model.It lays a good foundation for the subsequent establishment of a reasonable fault early warning monitoring model.According to actual cases,the early warning can be reached before the fault occurs,which effectively improves the reliability of the unit.