Research On Forecasting The Degradation Trend Of Vulnerable Components Of Offshore Wind Turbines

Offshore wind power has the advantages of strong wind force,stable wind direction,large capacity of single machine and no land occupation,which has gradually become the future development direction of China's wind power industry.However,due to the influence of typhoon,tide,sea wave and other adverse environment,the vulnerable components of offshore wind turbine units frequently fail.Therefore,it is of great significance to carry out the prediction research on the degradation trend of vulnerable components of offshore wind turbine units to reduce the frequency of failure and the cost of operation and maintenance.This article extracts reasonable degradation indicators and establishes an effective prediction model to reveal the degradation trend of vulnerable components and realize early warning of component failures.The main research contents are as follows:(1)Considering the accuracy of the prediction of the degradation trend of the vulnerable components,the complex working conditions of the offshore wind turbine are first classified,and then the prediction research is carried out for a specific working condition.A differential evolution algorithm combined with Fisher criterion(FDE)is proposed to optimize the kernel principal component analysis(KPCA)to realize the multi-parameter operating condition feature extraction.Based on this,the fuzzy C-means clustering algorithm(FCM)is used to establish the working condition division model,and the actual operating condition data of the offshore wind farm is used for verification.(2)A new prediction method based on preferred wavelet and Mahalanobis distance(MD)using GRU neural network is proposed.Firstly,the wavelet packet decomposition is used to extract the energy characteristics of the vulnerable component data,and the principle of optimal decomposition layers is established,and the optimal wavelet basis function is selected according to the energy fluctuation change rate.Then,the MD between the eigenvectors is calculated as the index of the performance degradation of vulnerable components.The index is used as the input of GRU network to construct the prediction model of degradation trend,and the algorithm is verified by the measured bearing life data.(3)After predicting the early failure of vulnerable components,a method of fault diagnosis for vulnerable components based on kernel fuzzy C-means clustering(KFCM)is proposed.By improving grey wolf optimization(GWO),the clustering centers and kernel parameters under the best classification results are obtained.According to the similarity between the test data of vulnerable parts and the samples of each clustering center in the kernel space,firstly judge whether the samples belong to the known fault,then diagnose the fault category,and take the measured gear fault data as an example to verify the algorithm.