Research On Fault Early Warning Technology Of Wind Turbine Gearbox Based On CNN And Long Short-term Memory Network

With the acceleration of the world’s energy transformation process,wind energy has become one of the general directions of global energy transformation due to its low-carbon,clean,and electrified characteristics.China’s wind energy resources are rich in reserves and have great development potential.Under the background of the current large-scale growth of the wind power market,it has strong resource competitiveness.In recent years,the explosive growth of the newly installed capacity of onshore wind power and offshore wind power has put forward higher requirements for the development of my country’s wind power industry and has also contributed to the realization of China’s "dual carbon" goals.As a large part of the wind turbine construction cost,the gearbox is the focus of wind turbine assembly and post-maintenance.The gearbox of wind turbine works in complex working conditions and extreme environment for a long time,and the probability of failure is higher than other components of the unit.Gearbox failure is one of the main reasons for the failure of the unit.Timeliness and accuracy are the keys to gearbox fault early warning technology research.Starting from the point of improving the timeliness and accuracy of early warning of unit gearbox faults,this paper focuses on the monitoring data of the supervisory control and data acquisition system and conducts in-depth analysis and mining of the collected data information.The main research contents are as follows.The density space unsupervised learning clustering outlier detection method based on loop parameter iteration can detect and mark outliers in SCADA data of wind turbines.Compared with the outlier detection method based on the isolated forest algorithm,more free abnormal values in the SCADA data can be detected,and the edge values under normal distribution can be screened out to obtain more core data information.Based on the generative adversarial networks model to realize the interpolation of the vacancy value of the fan SCADA data,and compared with the mean substitution interpolation method,while ensuring the integrity of the SCADA data time series,the actual situation of the unit operation is restored to the greatest extent and the time series prediction ability of long short-term memory network.Based on the multi-dimensional analysis,outlier detection and marking,and vacancy interpolation on the original data of the unit,a convolutional neural network and long short-term memory network gearbox fault early warning model based on high-quality SCADA data is constructed.The model was tested using the monitoring data of the same type of faulty unit.The experimental results show that the wind turbine fault early warning model,which has been processed by abnormal value detection marking and vacancy value interpolation,can predict the abnormal fluctuation of gear oil temperature in the early stage of the failure of the unit,and has a high fault early warning accuracy,and achieves the purpose of improving the timeliness and accuracy of wind turbine gearbox fault early warning technology.