基于Block-RPLS模型自适应更新的质量预测方法
|
摘要
工业过程包含动态、时变等过程特性.传统的基于PLS方法的质量预测采用的是固定模型,难以实时修正和学习新的过程信息,从而导致建模效率和精度降低,针对该问题提出一种自适应的块式递推偏最小二乘法(Block-RPLS)模型质量预测方法,用于在线调整PLS模型的结构和参数.采用滑动窗方法确定更新的数据块,利用矩阵相似性理论分析窗内数据的结构特性,得到该滑动窗的特征矩阵.同时,引入局部离群因子(LOF)检测滑动窗内离散偏离程度较大的更新数据,通过交叉验证方法修正PLS模型参数自适应学习过程的时变信息.最后,通过数值仿真和青霉素发酵过程的质量预测实验验证所提出方法的有效性.
The industrial process includes the dynamic and time-varying characteristics. The quality prediction method based on the tradition PLS model fails to learn and follow the process information on real time by using the fixed training data set, which often causes the problems of poor model effectiveness and accuracy on quality prediction performance. Thus, a new quality prediction method based on the adaptive block recursive partial least squares(Block_RPLS)is proposed to modify the construction and the parameter of PLS model online. The sliding window method is used to determine the updated data blocks and the structural characteristics of data blocks in this sliding window is analyzed by using the matrix similarity theory. At the same time, the local outlier factor(LOF) is introduced to detect the outlier data that has large discrete deviation in the sliding window, then, the time-varying information of the outlier data is adapted by the cross validation method to learn the PLS model parameters adaptively. Finally, the effectiveness of the proposed method is demonstrated by the numerical simulation and the experiment of the penicillin fermentation process.
The industrial process includes the dynamic and time-varying characteristics. The quality prediction method based on the tradition PLS model fails to learn and follow the process information on real time by using the fixed training data set, which often causes the problems of poor model effectiveness and accuracy on quality prediction performance. Thus, a new quality prediction method based on the adaptive block recursive partial least squares(Block_RPLS)is proposed to modify the construction and the parameter of PLS model online. The sliding window method is used to determine the updated data blocks and the structural characteristics of data blocks in this sliding window is analyzed by using the matrix similarity theory. At the same time, the local outlier factor(LOF) is introduced to detect the outlier data that has large discrete deviation in the sliding window, then, the time-varying information of the outlier data is adapted by the cross validation method to learn the PLS model parameters adaptively. Finally, the effectiveness of the proposed method is demonstrated by the numerical simulation and the experiment of the penicillin fermentation process.
引文
[1]Benaicha A,Mourot G,Benothman K,et al.Determination of principal component analysis models for sensor fault detection and isolation[J].Int J of Control,Automation and Systems,2013,11(2):296-305.
[2]Mnassri B,Adel E M E,Ouladsine M.Reconstruction-based contribution approaches for improved fault diagnosis using principal component analysis[J].J of Process Control,2015,33:60-76.
[3]Pozo F,Vidal Y.Wind turbine fault detection through principal component analysis and statistical hypothesis testing[J].Advances in Science&Technology,2016,101(1):45-54.
[4]Rato T J,Reis M S.Fault detection in the Tennessee Eastman benchmark process using dynamic principal components analysis based on decorrelated residuals(DPCA-DR)[J].Chemometrics Intelligent Laboratory Systems,2013,125(7):101-108.
[5]Hulland J.Use of partial least squares(PLS)in strategic management research:A review of four recent studies[J].Strategic Management J,2015,20(2):195-204.
[6]Hair J F,Hult G T M,Ringle C M,et al.A primer on partial least squares structural equation modeling(PLS-SEM)[M].2nd ed.Thousand Oaks:Sage Publications,2016:170-280.
[7]Denham M C.Prediction intervals in partial least squares[J].J of Chemometrics,2015,11(1):39-52.
[8]Jessen F,Lametsch R,Bendixen E,et al.Extractin information from twodimensional electrophoresis gels by partial least squares regression[J].Proteomics,2015,2(1):32-35.
[9]Helland K,Berntsen H,Borgen O S,et al.Recursive algorithm for partial least squares regression[J].Chemometrics Intelligent Laboratory Systems,1992,14(1):129-137.
[10]Qin S J.Recursive PLS algorithms for adaptive data modeling[J].Computers Chemical Engineering,1998,22(4/5):503-514.
[11]Qin S J.Partial least squares regression for recursive system identification[C].IEEE Conf on Decision&Control.New York:IEEE Press,1994:2617-2622.
[12]汪小勇,梁军,刘育明,等.基于递推PLS的自适应软测量模型及其应用[J].浙江大学学报:工学版,2005,39(5):676-680.(Wang X Y,Liang J M,Liu Y M,et al.Recursive PLS based adaptive soft-sensor model and its application[J].J of Zhejiang University:Engineering Science,2005,39(5):676-680.)
[13]Jalalvand A R,Gholivand M B,Goicoechea H C,et al.Advanced and tailored applications of an efficient electrochemical approach assisted by As LSSR–COW–r PLS and finding ways to cope with challenges arising from the nature of voltammetric data[J].Chemometrics Intelligent Laboratory Systems,2015,146(8):437-446.
[14]Matias T,Souza F,Rui A,et al.On-line sequential extreme learning machine based on recursive partial least squares[J].J of Process Control,2015,27:15-21.
[15]Asmund Rinnan,Andersson M,Ridder C,et al.Recursive weighted partial least squares(r PLS):An efficient variable selection method using PLS[J].J of Chemometrics,2014,28(5):439–447.
[16]Arablouei R,Doan?ay K,Werner S,et al.Adaptive distributed estimation based on recursive least-squares and partial diffusion[J].IEEE Trans on Signal Processing,2014,62(14):3510-3522.
[17]周建英,吴小培,张超,等.基于滑动窗的混合高斯模型运动目标检测方法[J].电子与信息学报,2013,35(7):1650-1656.(Zhou J Y,Wu X P,Zhang C,et al.A moving object detection method based on sliding window gaussian mixture model[J].J of Electronics Information,2013,35(7):1650-1656.)
[18]Cason T P,Absil P A,Dooren P V.Iterative methods for low rank approximation of graph similarity matrices[J].Linear Algebra Its Applications,2013,438(4):1863-1882.
[19]Han M,Zhang Z K.Fault detection method based on weighted kernel independent component analysis[J],Control and Decision,2016,31(2):242-248.
[20]Ma H H,Hu Y,Shi H B.Fault detection of complex chemical processes using Mahalanobis distance-based local outlier factor[J].CIESC J.2013,64(5):1674-1682.
[21]Zhao Y,Balboni F,Arnaud T,et al.Fault experiments in a commercial-scale PV laboratory and fault detection using local outlier factor[C].IEEE,Photovoltaic Specialists Conference.New York:IEEE Press,2014:3398-3403.
[22]Bing S,Shi H,Ma Y,et al.Multisubspace principal component analysis with local outlier factor for multimode process monitoring[J].Industrial Engineering Chemistry Research,2014,53(42):16453-16464.
[2]Mnassri B,Adel E M E,Ouladsine M.Reconstruction-based contribution approaches for improved fault diagnosis using principal component analysis[J].J of Process Control,2015,33:60-76.
[3]Pozo F,Vidal Y.Wind turbine fault detection through principal component analysis and statistical hypothesis testing[J].Advances in Science&Technology,2016,101(1):45-54.
[4]Rato T J,Reis M S.Fault detection in the Tennessee Eastman benchmark process using dynamic principal components analysis based on decorrelated residuals(DPCA-DR)[J].Chemometrics Intelligent Laboratory Systems,2013,125(7):101-108.
[5]Hulland J.Use of partial least squares(PLS)in strategic management research:A review of four recent studies[J].Strategic Management J,2015,20(2):195-204.
[6]Hair J F,Hult G T M,Ringle C M,et al.A primer on partial least squares structural equation modeling(PLS-SEM)[M].2nd ed.Thousand Oaks:Sage Publications,2016:170-280.
[7]Denham M C.Prediction intervals in partial least squares[J].J of Chemometrics,2015,11(1):39-52.
[8]Jessen F,Lametsch R,Bendixen E,et al.Extractin information from twodimensional electrophoresis gels by partial least squares regression[J].Proteomics,2015,2(1):32-35.
[9]Helland K,Berntsen H,Borgen O S,et al.Recursive algorithm for partial least squares regression[J].Chemometrics Intelligent Laboratory Systems,1992,14(1):129-137.
[10]Qin S J.Recursive PLS algorithms for adaptive data modeling[J].Computers Chemical Engineering,1998,22(4/5):503-514.
[11]Qin S J.Partial least squares regression for recursive system identification[C].IEEE Conf on Decision&Control.New York:IEEE Press,1994:2617-2622.
[12]汪小勇,梁军,刘育明,等.基于递推PLS的自适应软测量模型及其应用[J].浙江大学学报:工学版,2005,39(5):676-680.(Wang X Y,Liang J M,Liu Y M,et al.Recursive PLS based adaptive soft-sensor model and its application[J].J of Zhejiang University:Engineering Science,2005,39(5):676-680.)
[13]Jalalvand A R,Gholivand M B,Goicoechea H C,et al.Advanced and tailored applications of an efficient electrochemical approach assisted by As LSSR–COW–r PLS and finding ways to cope with challenges arising from the nature of voltammetric data[J].Chemometrics Intelligent Laboratory Systems,2015,146(8):437-446.
[14]Matias T,Souza F,Rui A,et al.On-line sequential extreme learning machine based on recursive partial least squares[J].J of Process Control,2015,27:15-21.
[15]Asmund Rinnan,Andersson M,Ridder C,et al.Recursive weighted partial least squares(r PLS):An efficient variable selection method using PLS[J].J of Chemometrics,2014,28(5):439–447.
[16]Arablouei R,Doan?ay K,Werner S,et al.Adaptive distributed estimation based on recursive least-squares and partial diffusion[J].IEEE Trans on Signal Processing,2014,62(14):3510-3522.
[17]周建英,吴小培,张超,等.基于滑动窗的混合高斯模型运动目标检测方法[J].电子与信息学报,2013,35(7):1650-1656.(Zhou J Y,Wu X P,Zhang C,et al.A moving object detection method based on sliding window gaussian mixture model[J].J of Electronics Information,2013,35(7):1650-1656.)
[18]Cason T P,Absil P A,Dooren P V.Iterative methods for low rank approximation of graph similarity matrices[J].Linear Algebra Its Applications,2013,438(4):1863-1882.
[19]Han M,Zhang Z K.Fault detection method based on weighted kernel independent component analysis[J],Control and Decision,2016,31(2):242-248.
[20]Ma H H,Hu Y,Shi H B.Fault detection of complex chemical processes using Mahalanobis distance-based local outlier factor[J].CIESC J.2013,64(5):1674-1682.
[21]Zhao Y,Balboni F,Arnaud T,et al.Fault experiments in a commercial-scale PV laboratory and fault detection using local outlier factor[C].IEEE,Photovoltaic Specialists Conference.New York:IEEE Press,2014:3398-3403.
[22]Bing S,Shi H,Ma Y,et al.Multisubspace principal component analysis with local outlier factor for multimode process monitoring[J].Industrial Engineering Chemistry Research,2014,53(42):16453-16464.