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

In recent years,wind energy as a clean energy has developed rapidly.However,most of the wind turbines are installed in high latitude,high altitude and other cold places,which makes the wind turbines often suffer from the problem of blade icing.If we can diagnose the ice fault of the wind turbine blade timely and accurately,and deicing the wind turbine timely,it is of great significance to ensure the safe and stable operation of the wind turbine.At present,the most direct and effective method to diagnose the ice fault of wind turbine blades is the artificial observation method,which is extremely time-consuming and labor-consuming.Therefore,it is an urgent problem to find a simple and efficient method for ice fault diagnosis of wind turbine blades.Based on the SCADA data of wind turbine,this paper puts forward the characteristic quantity which can represent the icing of wind turbine blades,and then uses machine learning algorithm to build the ice fault diagnosis model of wind turbine blades.The model takes the characteristic data as the input and outputs the status tag of whether the wind turbine blade is frozen or not,which realizes the diagnosis of the wind turbine blade icing fault.The main research work of this paper is as follows:(1)A set of standardized training set data preprocessing method is proposed.Firstly,the invalid samples and outliers in the training set data are eliminated.Then,for large sample data,we use strong rule filtering method,similarity function method and fixed length sampling method to undersample.For small sample data,smote algorithm is used to over sample,and finally achieve the class balance of training set data to get standard training set data.(2)The optimal eigenvalue group of the ice fault diagnosis model input of the wind turbine blade is extracted.On the one hand,the filter method is used to select the important features with strong correlation with the icing of wind turbine blades.On the other hand,the physical model analysis method,data visualization analysis method and sliding window analysis method are used to construct the features.Finally,a total of 7 transient features and 8 statistical features are extracted to form the optimal feature group of model input.(3)A fault diagnosis model of wind turbine blade icing is built and verified by an example.Firstly,the support vector machine(SVM)of the traditional machine learning algorithm is selected,and the particle swarm optimization algorithm is used to optimize SVM,and the ice fault diagnosis model of wind turbine blades based on SVM is constructed.Then,the long-term and short-term memory network(LSTM)in deep learning is selected,and the adaptive time estimation method is used to train LSTM.Finally,the model evaluation index based on confusion matrix is constructed,and the model is validated by an example.The results show that the model based on SVM and the model based on LSTM can diagnose the ice fault of wind turbine blades better,and the classification performance of the model based on LSTM is slightly better than that of the model based on SVM.In addition,the performance of the model constructed by using different eigenvalues is compared,and the results show that the classification performance of the model constructed by using the optimal eigenvalue group is the best,which is better than the model constructed by using the original SCADA data or transient eigenvalues,which proves the effectiveness of the research method in this paper.