Prediction Of Fan Blade Icing Based On Machine Learning Algorithm

Wind power generation is a very important renewable and clean energy,and its importance is increasing day by day in the national economy.The main machinery of wind power generation is electric power equipment.Wind power generators are normally located in high altitude or high latitude areas with good resources.However,in the low temperature climate,the unit has the problem of surface icing.Among them,the problem of blade icing is the most serious.In order to ensure the normal and safe operation of wind turbines in ice-prone areas,the observation and removal of blade icing is very important.The phenomena of blade icing,material and inherent structure change,circuit current and voltage change caused by low temperature will seriously affect the normal operation and good performance of wind turbine.Most of the traditional methods are to study whether the wind turbine blades are icing or not,such as wind turbine blade icing diagnosis technology based on vibration detection,wind turbine blade icing monitoring method based on piezoelectric ceramic technology,etc.There are also very few literatures to consider this problem from the perspective of machine learning.This paper innovatively adopts the method of multi-model fusion to solve this problem on the basis of predecessors,and further proposes a better scheme that can be used in practice,that is,the model can be constantly updated.Based on the real data of a wind farm,this paper preprocesses one of the wind turbine data by various algorithms,including,but not limited to,missing value processing based on Lagrange difference method,sample imbalance processing based on SMOTE oversampling algorithm,clustering-based small cluster partition method and atypical outlier detection method.Variance screening method,chi-square test,mutual information method,recursive feature elimination method,feature selection method based on L1 penalty item and model-based feature selection method are used to screen features.Strong correlation features are screened according to the aggregate comparison results,and unnecessary features are removed.Then,based on the data,the multi-model fusion scheme is innovatively selected for modeling after comparison.Then the ice prediction of another wind turbine is carried out by using the established model.The final result shows that the prediction effect is good,which proves that the scheme is effective.In order to put this scheme into practical use,considering the increase of data scale in the future,this paper innovatively chooses a large data platform based on SparkMLlib.The model can be constantly updated and self-prediction ability can be improved.