Research On Intelligent Fault Warning Of Key Components Of Wind Turbine Based On Machine Learning

The national policy goals of "peaking carbon,neutralizing carbon","accelerating planning and building a new energy system,and promoting the formation of a green and low carbon production and lifestyle" are put forward successively,which brings great opportunities for our wind power market.In recent years,the installed capacity of wind power generation is growing rapidly,the contribution rate of wind power to the economic development of our country continues to climb,but a large number of fans have been out of the quality guarantee period,leading to the increase of pressure on wind power enterprise operation and maintenance,how to use advanced big data and artificial intelligence technology to effectively improve the traditional post-operation and maintenance mode,reducing cost and increasing efficiency for enterprises and enabling intelligent construction becomes an important problem to be solved urgently.This paper focuses on the three key components of wind turbine with high failure frequency--gearbox oil temperature,gearbox output shaft and generator,and studies the fault warning system of wind turbine.Integrated machine algorithm is used to establish a fault warning model,and the minimum residual mean is adopted as the warning threshold to realize fault early warning.Firstly,the gearbox oil temperature fault warning is studied using Light GBM algorithm.After cleaning 1,521,163 pieces of operation and maintenance data,three data sets with different feature dimensions were generated by feature screening combined variance selection method,Pearson correlation coefficient and recursive feature elimination method.We used grid search to optimize Light GBM parameters,and comprehensively compared the fitting effect and training time of different data sets.It was found that Light GBM algorithm performed better in data set 2(dimension 32).Light GBM is compared with the other two boosting important algorithms XGBoost and Cat Boost in data set 2.The comprehensive analysis shows that the goodness of fit,mean square error and mean relative error of the gearbox oil temperature early-warning model based on Light GBM reach 0.998,0.00003 and 0.0027,and the training time of the model is 1617.78 seconds,which is only 1/2 of the data set 1(dimension 70).Gearbox oil temperature warning model based on Light GBM has the best performance.Finally,5.2℃,the minimum mean of the residual prediction of the two historical faults of the gearbox oil temperature,was taken as the warning threshold.Experimentally,the gearbox oil temperature fault could be warned at least 8 hours(h)in advance,and the feature(i.e.potential fault factors)importance order of the gearbox oil temperature warning model could be obtained by using SHAP.Provide technical support for subsequent wind turbine fault diagnosis and location diagnosis.Secondly,the gearbox output shaft fault early warning system is established.The Pearson correlation coefficient was combined with the importance ranking of machine learning model features to screen 10 important features,and three mainstream machine learning algorithms such as XGBoost,Light GBM and Cat Boost were used for modeling.The results show that the early-warning model of gearbox output shaft based on XGBoost has good predictive ability,the early-warning threshold is 3.4℃,and the early-warning time is 9h.Finally,the early warning of generator failure of wind turbine is studied.Taking the temperature of generator bearing B as the modeling object,a similar research process of gearbox fault warning was adopted to establish an early warning model of generator bearing B based on XGBoost,with an early warning threshold of 4.7℃ and an early warning time of 14 h.The above key component model has been measured in a wind field in Weihai,Shandong province,and the results show that the component temperature prediction accuracy is more than 95%.In this paper,the study of wind turbine fault warning system can greatly reduce the operation and maintenance costs of wind farms,reduce the consumption of fossil energy such as coal,and make contributions to the transformation of old and new kinetic energy.