选择性递推LSSVR及其在过程建模中的应用Selective Recursive LSSVR and its Applications in Process Modeling
刘毅,陈坤,王海清,李平
摘要(Abstract)：
提出一种改进的递推最小二乘支持向量机方法,用于非线性MIMO(multi-input multi-out)过程的可在线更新辨识和软测量建模。该算法在向前递推时只引入预报误差较大的样本更新模型,以提高模型的稀疏性和泛化能力。在向后删减时推导了基于快速留一法的模型修剪准则,以克服随意删减样本的缺点。通过有选择性的向前、向后递推,模型的推广能力和计算性能均得以保证,且更加适应过程的时变性。在连续搅拌釜式反应器过程的辨识和重油催化裂化的软测量建模研究,均表明所提出算法的有效性和优越性。
关键词(KeyWords)： 在线最小二乘支持向量机;选择性递推;过程辨识;软测量建模
基金项目(Foundation): 国家自然科学基金(20776128);; 浙江省自然科学基金资助项目(Y107032)
作者(Author): 刘毅,陈坤,王海清,李平
参考文献(References)：
- [1]Sjberg J,Zhang Q H,Benveniste A et al.Nonlinear black-box modeling in system identification:a unified overview[J].Automatic,1995,31(12):1691-1724.
- [2]Wang L X.Adaptive Fuzzy Systems and Control:Design and Stability Analysis[M].New Jersey:Prentice-Hall,1994.
- [3]Suykens J A K,van Gestel T,de Brabanter J et al.Least Squares Support Vector Machines[M].Singapore:World Scientific,2002.
- [4]PENG Zhen-rui(彭珍瑞),WANG Bao-liang(王保良),HUANG Zhi-yao(黄志尧)et al.Improved LS-SVM based voidage measurement of oil-gas two-phase flow(基于改进的LS-SVM测量油气两相流空隙率)[J].J Chem Eng of Chinese Univ(高校化学工程学报),2008,22(1):18-22.
- [5]Tang H S,Xue S T,Chen R et al.Online weighted LS-SVM for hysteretic structural system identification[J].Engineering Structure.2006,28(12):1728-1735.
- [6]ZHANG Hao-ran(张浩然),WANG Xiao-dong(汪晓东).Incremental and online learning algorithm for regression least squares support vector machine(回归最小二乘支持向量机的增量和在线式学习算法)[J].Chinese J Comput(计算机学报),2006,29(3):400-406.
- [7]FENG Rui(冯瑞).Study on Support Vector Machines and its Application in Soft Sensor of Industrial Processes(支持向量机及其在工业过程软测量中的应用研究)[D].Shanghai(上海):Shanghai Jiaotong University(上海交通大学),2003.
- [8]Wang H Q,Li P,Gao F R et al.Kernel classifier with adaptive structure and fixed memory for process diagnosis[J].AIChE J,2006,52(10):3515-3531.
- [9]Cawley G C,Talbot N L C.Preventing over-fitting during model selection via Bayesian regularisation of the hyper-parameters[J].J Mach Learn Res,2007,8(4):841-861.
- [10]Golub G H,van Loan C F.Matrix Computations.3rd ed[M].Baltimore:The John Hopkins University Press,1996.