Research On Health Status Assessment And Prediction Of Wind Turbines Based On Long Short Term Memory Neural Network

As a main form of clean energy power generation technology,wind power technology has developed rapidly in the context of China:s sustained high-speed economic development,gradual shortage of resources and increasingly serious environmental problems in recent years.However,due to the wind turbine is operating in the harsh environment,which is resulting in frequent wind turbine failures,reducing the utilization of wind energy and increasing the cost of operation and maintenance.Therefore,the operation condition monitoring of wind turbines has received more and more attention,and it has become more important to predict the faults of wind turbines in advance and accurately.At present,the bearing fault characteristic signal of the variable working condition monitored by the wind turbine Condition Monitoring System(CMS)is very weak.The traditional diagnostic signal processing approaches can not clearly extract the bearing vibration signal characteristics,so it is difficult to realize the early state monitoring of the bearing.The SCADA data monitored by the wind turbine Supervisory Control And Data Acquisition(SCADA)system varies widely and has strong randomness,which makes it difficult to extract information.Moreover,there is a lack of effective theory and method to extract useful information,which results in these datas can not accurately monitor the state of the equipment,and the method can not prevent the deterioration of equipment failure and have not been effectively utilized.In order to effectively judge the operation status of wind turbines,this paper takes the SCADA data provided by the wind farm and the simulated wind farm bearing vibration data provided by the test bench as the research object,and puts forward a health state assessment analysis method based on Long Short-Term Memory(LSTM)neural network.The LSTM neural network is a method for processing time series data,by learning historical data,the relationship between time series can be found and the inherent rules can also be found,and the current operation state of the wind turbine is predicted early.In this paper,the following aspects are studied in view of some problems existing in the early operation monitoring of wind turbines:(1)Aiming at the problem that the wind turbine bearing fault characteristic signal is weak and the fault characteristics are difficult to extract,this paper proposes a LSTM neural network based on wind fami health assessment method.The method is to directly input the simulated wind farm bearing vertical vibration datas provided by the test bench into the established LSTM network,and then smooth the LSTM prediction datas by Exponentially Weighted Moving Average(EWMA).The result shows that the method has a certain evaluation effect on the running condition of the bearing.(2)Aiming at the problem that the wind turbines have large SCADA data,many parameters and difficult information extractionrthis paper proposes a LSTM neural network based on wind farm health assessment method.Firstly,the SCADA data needs to be preprocessed.Secondly,the preprocessed data is extracted by monotonicity and correlation metrics,then the extracted data is input into the constructed LSTM network.Finally,the LSTM prediction data is smoothed by EWMA.The feasibility and validity of this method are proved by comparing the final output results with the results of the on-site inspection of wind turbine.