计算机与应用化学
計算機與應用化學
계산궤여응용화학
COMPUTERS AND APPLIED CHEMISTRY
2013年
7期
735-738
,共4页
EM-PCA%化工过程%缺失数据%补值
EM-PCA%化工過程%缺失數據%補值
EM-PCA%화공과정%결실수거%보치
EM-PCA%chemical process%missing data%imputation
化工过程数据具有变量多，数据量大的特点，而且在测量过程中易发生数据缺失。为减少数据缺失对数据分析及故障检测过程的影响，需要对缺失数据进行补值。本文采用EM-PCA(Expectation Maximization algorithm for Principal Component Analysis)补值算法对TE(Tennessee Eastman)化工过程数据的随机缺失进行补值。选择不同的初值设置方法，并选取不同主元数对不同缺失率下的数据进行补值，应用补值与原始数值的平均相对误差来评价补值结果。结果显示当选用的主元数增大时，补值结果趋于稳定，而且EM-PCA补值算法的误差小于使用平均值法补值及当前值补值方法的误差。补值能够为后续的过程故障检测提供完整的数据，对化工过程的监控具有重要的意义。
化工過程數據具有變量多，數據量大的特點，而且在測量過程中易髮生數據缺失。為減少數據缺失對數據分析及故障檢測過程的影響，需要對缺失數據進行補值。本文採用EM-PCA(Expectation Maximization algorithm for Principal Component Analysis)補值算法對TE(Tennessee Eastman)化工過程數據的隨機缺失進行補值。選擇不同的初值設置方法，併選取不同主元數對不同缺失率下的數據進行補值，應用補值與原始數值的平均相對誤差來評價補值結果。結果顯示噹選用的主元數增大時，補值結果趨于穩定，而且EM-PCA補值算法的誤差小于使用平均值法補值及噹前值補值方法的誤差。補值能夠為後續的過程故障檢測提供完整的數據，對化工過程的鑑控具有重要的意義。
화공과정수거구유변량다，수거량대적특점，이차재측량과정중역발생수거결실。위감소수거결실대수거분석급고장검측과정적영향，수요대결실수거진행보치。본문채용EM-PCA(Expectation Maximization algorithm for Principal Component Analysis)보치산법대TE(Tennessee Eastman)화공과정수거적수궤결실진행보치。선택불동적초치설치방법，병선취불동주원수대불동결실솔하적수거진행보치，응용보치여원시수치적평균상대오차래평개보치결과。결과현시당선용적주원수증대시，보치결과추우은정，이차EM-PCA보치산법적오차소우사용평균치법보치급당전치보치방법적오차。보치능구위후속적과정고장검측제공완정적수거，대화공과정적감공구유중요적의의。
Chemical process data has many variables and great amount of data, and sometime missing data occurs during measurement. Imputation of data is needed to improve the data quality and reduce the influence to process modeling. The EM-PCA(Expectation Maximization algorithm for Principal Component Analysis) algorithm is used for imputation of random missing data from TE(Tennessee Eastman)chemical process. Different types of initial value, different number of principal component are used for different missing rate, and the average relative error is used to evaluate the results. The results showed that when the principal component number increases, the complement value approached stability, and the relative error of imputation error of EM-PCA is less than the average value method and current value method. The imputation can provide complete data for the subsequent process fault diagnosis and improve data analysis process significantly.
