Research On Icing Diagnosis Of Wind Turbine Blades Based On SCADA Data

Wind energy is a new type of sustainable and green energy.With the global energy shortage and environmental pollution becoming more and more serious,wind power generation technology has become a research hotspot in the energy field.In the process of wind power generation,blades are the main components of wind turbine to obtain wind energy,and the aerodynamic characteristics and structural performance of blades affect the running status of wind turbine.However,when the wind turbine operates in cold climate,it is easy to have a large amount of icing on the blades,which will destroy the aerodynamic characteristics and structural performance of the blades,lead to a decline in power generation and shorten the service life of the unit.In addition,maintenance or shutdown caused by severe icing of blades will bring huge losses to the operation of the wind farm,and even large ice cubes falling off the blades will pose threats and hazards to the surrounding environment and personnel.Due to the large number of sensors distributed,the performance changes of wind turbines caused by external environment or internal hardware will be reflected in the operation data.Therefore,the icing of wind turbine blades can be diagnosed by statistics and analysis of a large number of data in wind farms.To solve the problem of blade icing of large horizontal axis wind turbine in cold climate,this paper studies the diagnosis model of blade icing of wind turbine based on the operation data in SCADA(Supervisory Control and Data Acquisition)of wind farm,related knowledge of wind turbine field and data analysis,and provides theoretical basis for related engineering applications.The research contents of this paper mainly include:firstly,because the ratio of blade icing data and normal data in the data set is seriously unbalanced,this paper adopts SMOTE combined with Edited Nearest Neighbours sampling to solve the problem of data imbalance;Then,chi-square test is used to screen the characteristics of the running data in SCADA,and the characteristics that can reflect the icing of blades are screened out to solve the problem of large amount of data and redundant data.Then,the dimensions of the selected features are reduced by using the relevant knowledge of wind turbine field,and a feature dimension reduction method based on Pearson correlation coefficient and PCA(Principal Components Analysis)is proposed to realize feature dimension reduction.Then compare and analyze several single classifiers in machine learning,select three single classifiers CART(Classification and Regression Tree),RF(Random Forest)and MLP(Multilayer Perceptron)to build a Stacking model,and compare and analyze the classification results of single classifier and combination model;Finally,the performance index of the model is tested by using the competition data set of the wind turbine blade icing prediction contest and the actual engineering data of a wind farm in Yunnan Province.The results show that the Pearson correlation coefficient combined with PCA feature dimension reduction method proposed in this paper is effective,and it also shows that the Stacking model constructed in this paper can improve the accuracy of blade icing diagnosis compared with a single classifier.