Fault Early Warning For Key Components Of Wind Power Transmission System Based On Long Short-term Memory Network

With the rapid development of the global economy,the world 's demand for energy is also increasing.Traditional energy sources such as oil and coal have been gradually diminishing,subsequent reserves are limited,and the negative impact on the environment is great.In recent years,countries around the world are actively seeking new energy sources.Among them,wind energy has been the focus of attention for its green,environmental protection and renewable advantages.In addition,the installed capacity of wind turbines has increased year by year,and has gradually developed from land to offshore with richer wind energy.However,with the rapid development of the wind power industry,various problems have become more prominent.For example,as the service life of a unit increases,its utilization of wind energy is low,reliability is poor,faults occur frequently,and maintenance costs are high.Therefore,it is of great significance for the steady development of the wind power industry to carry out operation and maintenance work such as condition monitoring,fault early warning,and timely maintenance of related components when the unit is running.In the current era of big data,informationization and intelligence,how to make better use of the massive data generated during the operation of wind turbines,mine the state information of the data,and tailor the appropriate operation and maintenance model for each unit to become the field of wind power Research hotspots and difficulties of condition monitoring technology.This topic starts from the needs and problems of fault diagnosis of wind power equipment,through a more in-depth analysis of the unit's operating status and data characteristics,based on machine learning and deep learning methods,researches on early warning of key component failures,aiming to achieve the unit's status Monitoring and fault early warning.The main work of the paper is as follows:(1)Analyze the operating principle of key components of the unit drive system,the types and causes of failures,determine the data source using SCADA data as a model,and summarize their characteristics to lay the foundation for subsequent research methods.(2)Taking the oil temperature failure of the gearbox as the research object,the method of early warning of time series data failure based on long-term and short-term memory networks is studied.First,a long-term and short-term memory network model is constructed for the time-series correlation of SCADA data;then,the health data filtered through training is used to reconstruct the gearbox oil temperature and establish an unsupervised model;finally,the gearbox data is monitored online and rebuilt When the error exceeds the threshold,a fault warning is performed.The significance of the model on the one hand attaches importance to the correlation of data in the time dimension,and on the other hand builds the model through unsupervised means,which solves the problem of data imbalance due to lack of faulty data.(3)Further analyze the characteristics of SCADA data,mine the features of SCADA data,and study the early warning method of blade cracking faults based on long-term and short-term memory networks.First of all,on the one hand,as mentioned above,the SCADA data has a temporal correlation;on the other hand,the unit changes at any time during the operating conditions,and the weight of the final result is different for each dimension feature in different operating conditions.Aiming at the characteristics of the two parties,a network model is designed that can dynamically change the feature weights of each dimension according to the different working conditions,and can obtain the temporal correlation of data.By using the labeled blade cracking data training model,the fault early warning of blade cracking is realized.Finally,the validity of the model is verified by using actual data and different methods.(4)Finally,a fault early warning module is constructed for the wind turbine drive system,and several functions of data pre-processing,model training,and model testing are implemented by means of a GUI interface.On the one hand,the method proposed in the article is further verified;on the other hand,the GUI interface interactive method is more intuitive and convenient for secondary training and testing of the model.