Research On Feature Data Extraction Method Of Wind Turbines Based On Data Mining

In recent years,wind power generation has developed rapidly,and a large number of wind farms have been completed and put into use.However,due to the uncertainty of the wind speed,the wind turbine is in a working environment with alternating load for a long time,so the probability of failure of the wind turbine is high.For most of the wind farms in the remote suburbs and up to hundreds of meters,the maintenance cost is difficult and the maintenance time is long,which has a great impact on the economic benefits of the wind farm.Therefore,it is very necessary to develop wind turbines related research on fault analysis and fault prediction.Based on the understanding of the mechanical structure of the wind turbine and the principle of related faults,this paper analyzes the historical data of the SCADA system and uses the relevant methods of data mining to predict the fault of the wind turbine,and validates the model through the pitch system in the unit.The data used in the relevant experiments in this paper are all derived from the SCADA data of a wind farm in Beijing and the unit fault data in the field.First analyze the relevant parameters recorded in the fault data,and select the appropriate state parameters for later fault diagnosis.In order to ensure the reliability of SCADA data,data cleaning methods such as mean method,regression method,and nuclear density estimation method are used to clean the original data.The ReliefF algorithm is used to select the variables that have a great influence on the fault state parameters of the pitch system,and the characteristics of the wind turbine operating data are selected to pave the way for the prediction of the fault state parameters.By comparing the BP neural network and the wavelet BP neural network,the selected features are used as model inputs to predict the fault state parameters.Aiming at the shortcomings of wavelet neural network,such as long calculation time and easy to fall into local minimum points,this paper uses genetic algorithm to optimize wavelet neural network to improve the accuracy of prediction model.The experimental simulation is carried out by Matlab,and the final prediction error is analyzed.The fault prediction method adopted in this paper has high accuracy.