Research On Fault Diagnosis Of Wind Turbine Pitch System Based On Deep Learning

The pitch control system of wind turbine is related to the reliable and stable operation of the whole wind turbine.The working environment of wind turbine pitch system is complex,and the wind speed changes randomly,resulting in frequent failures.At the same time,the strong coupling and nonlinearity of the variables among the pitch subsystems make it difficult to detect and locate the faults directly.In order to improve the accuracy of fault diagnosis of wind turbine pitch system,this thesis is based on the data of actual operation of the pitch system of supervisory control and data acquisition(SCADA)system,with deep learning in the frontier of artificial intelligence as the main technical and theoretical support,and pitch system as the main research object.Aiming at the above problems,a fault diagnosis method of wind turbine pitch system based on deep learning is proposed.The main tasks are as follows:First of all,due to the huge amount of data information of wind power SCADA pitch system,many characteristic parameters are collected,and it is difficult to obtain fault data.In this thesis,the normalized and synthetic minority sample oversampling technique(SMOTE)algorithm is used to preprocess the data.In order to solve the problem of gradient dispersion in feature extraction of general auto-encoder network when there are many parameters.Based on the SCADA data of wind turbine electric pitch system,the PRe LU function is selected as the activation function,the batch standardization(BN)algorithm is introduced,the Adam optimizer is selected,the neural network weight is updated iteratively based on the training data,and then the loss function is calculated.The network is trained with the minimum loss function as the objective.Ultimately,the softmax classifier is used to output the weight of each component of the pitch system and the probability of occurrence.The data set in SCADA of wind turbine pitch system is selected,and the comparison and verification are carried out through the stacked auto encoder(SAE)before and after the improvement.The verification results show that the batch standardized SAE network has more optimized network model and higher recognition accuracy,which also provides a reference for wind turbine fault diagnosis.Aiming at the problem of compound faults that often occur in pitch systems,an improved SAE pitch system fault diagnosis method based on multi-label classification is proposed,and two classification models are designed: single-label fault model and multi-label fault model.On the basis of improving the single fault of the SAE network model,the single label fault is used for classification.For SAE network feature extractor and multiple two classifiers constitute a multi label classification fault diagnosis model.The effectiveness of the algorithm is verified by the example of two composite fault types of pitch system.The composite fault diagnosis can be classified and identified by SAE network as a single category,and the performance index of multi label classification is evaluated by confusion matrix.The SAE network model can effectively diagnose multiple single faults under the condition of composite fault samples,which has good application value.