Research On SCADA Data Cleaning Method For Wind Turbine

With the continuous development and progress of information technology,the analysis and utilization of large data become the key to the development of various industries,wind power is no exception.Through the analysis of the operation data of the wind turbine,it can judge and forecast the running state of the fan,and find out the fault in time,which is of great significance to the safety and economy of the unit operation.The data analysis of the wind turbine is based on the data collected by the SCADA system,but the collected raw data often contains a lot of "dirty data"-that is,missing data,duplicate data,abnormal data and so on.If the analysis with the raw data containing"dirty data" results in a large deviation in the analysis results,the operation of the state to determine a great impact,therefore,the need for raw SCADA data for data cleaning.In this paper,SCADA operation data of a wind field in Anhui Province are taken as an example to study the method of data cleaning of wind turbine.The main contents of the project are as follows:Firstly,Due to the random variation of wind speed,it is difficult to analyze operating data based on single unit.Therefore,the concept of adjacent unit is proposed and the data is cleaned by establishing the data model of the correlation between adjacent units.However,there is no clear conclusion on how to determine the ratio of adjacent units.Based on Pearson's correlation coefficient,this paper gives a more accurate minimum length of correlation coefficient by analyzing a large amount of historical data,and analyzes the influence of different wind resources on the correlation coefficient.Secondly,the data anomaly detection method research.The least squares support vector machine(SVM)method is used to establish the neighborhood model which reflects the similarity between wind turbines.The data of the same variables of multiple adjacent units are analyzed by horizontal comparison,and the residuals are analyzed by prediction.Thirdly,the data interpolation method research.Two kinds of interpolation methods are proposed:one is the time series interpolation method based on ARMA model,and the prediction accuracy of the model is improved by the combination of forward and reverse.The other is based on the random forest method of the adjacent unit interpolation method,respectively,for a single adjacent units and multiple adjacent units of the interpolation were analyzed.