Data-based Health Assessment Of Wind Turbine

Wind turbine is the main equipment of transforming wind power into electricity,which is a very complicated nonlinear dynamic system.There are many key parts and components in one wind turbine,and evaluating the running performance of these components is very important for the wind turbine maintenance.This paper mainly studies the health assessment of the key components bearing and the wind turbine.The health index is put forward to describe the degradation process of bearing and wind turbine performance.Example experiments show that the health index can not only accurately track the degradation process of bearing and wind turbine but also realize the early anomaly identification.This paper also studies the methods for the prediction of degradation performance and proposes an improved extreme machine learning methods.The main contents of this paper are as follows:Firstly,the degradation performance evaluation of the bearing is studied in this paper.In order to accurately evaluate the degradation process,this paper firstly extracts as many features as possible from the vibration signals of the bearing,and selects features that are more sensitive to the degradation process of bearing.Based on the selected features,the self-organizing map network and the minimum quantification error are used to construct the evaluation model and health index.Finally,six bearing's vibration data are used to validate the method and index.Then,this paper further explores the wind turbine health assessment problem under the complex and varied operating conditions.In order to realize the accurate evaluation of the wind turbine health assessment,this paper firstly divides the operating conditions of wind turbine.Then for each condition,the paper constructs a multi-feature fusion model based on GMM as the evaluation background model and puts forward the health index based on the Mahalanobis distance.The results show that the proposed method can not only accurately evaluate health-state of the wind turbine,but also identify the abnormal behavior of the wind turbine in advance.Finally,this paper studies the method of health-state trend prediction,and proposes an improved Extreme Learning Machine.This improved methods firstly uses the PSO algorithm to obtain the optimal input weights and biases,which overcomes the shortcomings of the network stability caused by arbitrary assignment.Then the mean value of the two excitation functions is used as the hidden layer's output to improve the accuracy of the prediction.The validation results show that compared with the BP neural network,least squares support vector machine and the original extreme learning machine,the improved ELM's prediction accuracy is higher.