Research On Wind Turbine Blade Icing Prediction Based On Data Driven

With the increase of energy demand for economic development,the wind power industry develops rapidly,and a growing number of wind turbines are put into operation.Due to the particularity of wind power acquisition,wind farms are built in areas with cold and damp air.Harsh conditions lead to frequent failures of wind turbines,the icing rate of wind turbine blades is relatively high,the wind farm benefits are seriously affected,and even safety issues may be caused.To solve the problem of wind turbine blade icing,the key is to accurately predict the wind turbine blade early icing situation,timely icing prevention and early deicing.This paper takes wind turbine blades of a wind power airport as the research object,selects two months of wind turbine blade working condition data and environmental data in SCADA system for analysis,and at the same time fully learns the mechanism of blade icing,and uses XGBoost algorithm to predict wind turbine blade icing.The main research work includes the following aspects:First of all,the data in this paper comes from the SCADA system of the wind farm.The icing problem is analyzed,the data variables used are understood,the working condition information and data are explained,and the operation principle of the wind turbine is mastered.Then,the data is preprocessed,which mainly includes missing value analysis,data standardization,and extraction of valid data.And feature selection of variables.Firstly,through the physical principle analysis and mathematical statistical analysis of the icing phenomenon,the wind speed,the power,and the temperature difference between the ambient temperature and the cabin temperature are three effective variables that can reflect the icing of the wind turbine blade.Through data visualization,the variables related to the three types of blades were analyzed:blade angle,blade speed,and pitch motor temperature.Further analysis of the relationship between them revealed that the three blades in these three aspects under normal operating conditions and freezing conditions The trends are consistent,and pitch_angle_mean,pitch_speed_mean and moto_tmp_mean are selected as new variables for analysis.Secondly,for the remaining variables and the three new variables,feature extraction is performed using the Relief algorithm,and the weights of each variable are calculated.Finally,seven feature variables are selected for building a classification model,which lays the foundation for the effective performance of the model.Finally,for the imbalance between icing data and normal data in this paper,set the icing data as positive samples and the normal running data as negative samples,and use the Smooth algorithm to perform data equalization processing.In order to determine whether the wind turbine blades are in the icing state,the supervised learning XGBoost extreme gradient lifting model is used for classification and prediction,and the samples are identified as icing data or normal data.The training set data is used for model training and learning,and the model is verified with the validation set.Finally,the model performance is evaluated through multiple evaluation indicators.The evaluation results prove that the method can effectively identify the working conditions,and the operation and maintenance strategy of the blade is proposed based on the icing prediction results.