Fault Detection Of Double-fed Wind Turbine Gearbox System Based On LightGBM

Wind turbines are a typical nonlinear complex system with frequent failures of components,including wind turbines,gearboxes,generators,converters,yaw systems,pitch systems,and hydraulic systems.Regular system fault detection of the gearbox of the wind turbine can reduce wind farm power generation accidents and reduce the economic loss and power loss caused by the long downtime of the wind turbine.This paper proposes a method for fault diagnosis of wind turbine gearboxes based on the LightGBM algorithm.(1)Aiming at the problem that there are many features in the SCADA data of the wind turbine gearbox system and it is difficult to obtain the optimal feature set,a feature selection method of the wind turbine gearbox system based on Spearman and the maximum information coefficient is proposed.Analyze the failure mechanism of the gear box system of the wind turbine,select the fault characteristic parameters of the gear box system from the massive wind turbine SCADA data,remove the invalid and redundant features,and select the characteristics that can effectively reflect the failure of the gear box system.The experimental results show that the Pearson feature selection method and the maximum information coefficient method can screen the massive wind turbine gearbox data to a great extent,improve the lack of artificial selection condition parameters,and extract the gearbox data from the SCADA database while reducing the dimensionality.(2)In view of the large amount of data calculation in the traditional optimization algorithm and the blind search for the optimal combination of parameters,a wind turbine fault diagnosis method based on the improved LightGBM algorithm is proposed,combined with Bayesian hyperparameter optimization,and a method based on the improved LightGBM is established.The fault diagnosis model for wind turbines can efficiently improve the accuracy of fault diagnosis for wind turbines.Compared with the original algorithm,the improved LightGBM algorithm is more accurate.The real-time data experiment results of a wind farm show that this method has a good effect on the fault diagnosis of wind turbine gearboxes.(3)Aiming at the problem of unbalanced samples of the fault and normal types of wind turbines,a strategy of LightGBM improved with cost-sensitive functions is proposed,which improves the accuracy of fault recognition of wind turbines under unbalanced sample conditions,and reduces the rate of false negatives.False alarm rate.At the same time,the diagnostic effects of cost-sensitive gradient boosting decision tree algorithm and cost-sensitive AdaBoost algorithm on the same data set are compared,and the effectiveness of the method is verified.