Research On Wind Turbine Gearbox Fault Early Warning Based On Deep Belief Network

As a clean and pollution-free green energy,wind energy has been greatly developed in recent years.However,wind turbines are generally installed in remote mountainous or coastal areas,and the bad natural environment makes wind turbine failures occur frequently.The gearbox system of wind turbine is a very important part of wind turbine.Its stress condition is complex.If it fails,it will cause serious consequences such as shutdown.It is of great significance for the stable and safe operation of the fan to excavate the SCADA data of the wind turbine and carry out the fault early warning of the gearbox.In this paper,a fault early warning strategy based on deep belief network algorithm is proposed and studied in detail.The strategy mainly consists of three parts: data preprocessing and feature extraction of wind turbine SCADA,early warning model based on deep belief network,and fault early warning combined with box line diagram and deterioration index.Firstly,based on the in-depth understanding of the working principle and typical faults of the fan gearbox system,the data preprocessing and feature extraction methods of the fan SCADA are studied.The abnormal data is deleted according to the parameter threshold and the data preprocessing is completed by DBSCAN clustering algorithm.In order to reduce the complexity of modeling and the influence of irrelevant parameters,the original features are selected according to the operation mechanism and empirical knowledge of the wind turbine.The maximum mutual information(MIC)algorithm is used to extract the features of the initial selection,and the parameters with high correlation with the gearbox oil pool temperature are obtained.Secondly,an early warning model based on deep belief network is proposed.After deeply studying the principle and optimization scheme of deep belief network algorithm,a fault early warning model of wind turbine gearbox based on deep belief network algorithm(DBN)is established.(1)In order to improve the modeling accuracy,a DBN-condition partition modeling scheme combining fuzzy c-means clustering and deep belief network is proposed.The fuzzy c-means clustering algorithm(FCM)is used to divide the working conditions according to the wind speed,power and generator torque.The DBN model is used to establish the early warning model of each sub working condition.(2)Considering the long training time of deep belief network,in order to meet the real-time requirements of the model and reduce the computational complexity,a DBN-ELM gearbox fault early warning model combining deep belief network and extreme learning machine(ELM)is proposed.Extreme learning machine(ELM)is used to replace BP layer of DBN model to reduce training time.The results show that DBN-condition partition model has great advantages in model accuracy and generalization ability,and DBN-ELM model has great advantages in training time.Finally,a fault early warning scheme based on threshold analysis is proposed.Considering that the working condition of wind turbine fluctuates greatly and the output residual of model monitoring parameters fluctuates greatly,it is difficult to judge the early warning time accurately.A fault early warning scheme combining box line diagram and deterioration index is proposed.The box line diagram is used to analyze the residual sequence and determine the early warning threshold,which is used as the discrimination index of fault early warning.Based on the threshold value of box graph,the deterioration index is defined to evaluate the deterioration level of wind turbine gearbox.Using the SCADA data of a wind turbine in January2018 in Hebei Province as the fault sample,the simulation analysis of an example is carried out to verify the feasibility and effectiveness of the gearbox system fault early warning strategy based on deep belief network proposed in this paper.