The Predication Evaluation Of The Wind Turbine Gearbox Health Trend Based On PCA-NAR Neural Network

Several factors,such as bad weather and environment,lead to the problems of component damage and state deterioration of wind turbines.Gearbox,as a key component of wind turbines,is prone to damage,which brings challenges to the safe and stable operation of power system.The preventive maintenance of gearbox is the key to solve this problem.The traditional "regular maintenance" method has the problems of "insufficient maintenance" and "excessive maintenance",which is not conducive to reducing maintenance costs and improving reliability.In order to ensure the safe and stable operation of wind turbine gearbox,this paper takes the gear box data and correlation analysis in SCADA monitoring system as the starting point,establishes the relative deterioration model of monitoring indicators to avoid the interference of abnormal data,and further establishes the evaluation and prediction model of gear box deterioration,so as to realize the recognition of abnormal state of gear box and the prediction of change trend.The details are as follows:In this paper,the wind turbine gearbox is taken as the research object.Considering the influence of the external environment on various internal parameters,the external environmental indicators(such as wind speed,power,ring temperature)and the internal monitoring indicators of the SCADA monitoring system are calculated by Spearman coefficient and mutual information theory.Correlation coefficients(temperature,speed,pressure,etc.),linear correlation and nonlinear correlation are combined to select certain types of environmental variables with high correlation using K-means clustering(K-means).The monitoring indicators are divided into working conditions,and the relative deterioration degree model is established according to the working conditions and the box pattern analysis method.On the basis of the above-mentioned establishment of the relative deterioration degree model,the indicators of the same component are reduced by the principal component analysis method.Finally,the deterioration trend of the unit is obtained.In order to further predict the state after the gearbox to provide operational reference basis,the NAR neural network is used to predict the degradation trend,and the degradation trend model is obtained.The results show that the deterioration model based on box diagram can reduce the disturbance of outliers and has a certain robustness.At the same time,PCA is used to reduce the dimension of various data,which can provide a better overall evaluation of gearbox,and also reduce the structure of prediction model and prediction time.For time series data,NAR neural network model has certain advantages.Compared with the widely used BP network and traditional time series model,its accuracy is higher,and its operation speed is also greatly improved.It has good practicability for early detection of the deterioration value of wind turbines and reminding staff.In addition,the PCA evaluation method is used to integrate the parameters and make a prediction,which reduces the loss time caused by the prediction of multiple parameters.