摘要
支持向量机是统计学习领域新兴的一种机器学习方法，已经在模式识别等领域得到有效的应用，也成为非线性系统辨识的有力工具。文献中将支持向量机方法与径向基函数神经网络相结合，所提出的支持向量神经网络算法被证明对非线性系统的建模是无偏的。但到目前为止，将支持向量机方法用于解决工业过程控制问题还不多见。
ABS树脂聚合过程具有高度非线性，时变性的特点。其聚合过程对温度及温度变化反应灵敏，给系统的建模和控制提出了较高的要求。
本文应用支持向量神经网络算法，对ABS树脂聚合过程进行建模，并与自适应逆控制方法相结合，提出了基于支持向量神经网络的非线性系统自适应逆控制方法。该方法利用支持向量神经网络，根据现场采集的数据训练网络，建立ABS树脂聚合过程逆系统模型，将此逆模型作为控制器，其输出作用于被控对象。同时根据期望输出与系统实际输出之差，通过PID算法，调整逆模型控制器的输入，实现对逆模型控制器的在线修正，形成了基于支持向量神经网络的逆模型自适应控制算法。对非线性系统进行的仿真结果表明，支持向量神经网络的建模精度高于普通神经网络，且具有训练时间短，泛化能力强等优点；基于支持向量神经网络的自适应逆控制方法具有简捷、可靠、有效、鲁棒性强的特点。与一般基于神经网络的控制器相比，它有能根据误差直接调节控制器参数以适应模型及环境变化的自适应能力，具有很强的鲁棒性；与传统自适应控制系统相比，它不需要精确的数学模型，具有广泛的适用性。
本文工作表明：支持向量神经网络作为一种高精度建模工具，不仅可以成功应用于解决如ABS树脂聚合温度建模等复杂工业过程的建模问题，还可以为复杂工业过程的控制提供有力的工具。
Support Vector Machine is a new machine learning method in the field of statistics. It has been used in the field of pattern identify effectively. It is also a powerful tool in nonlinear system discrimination. The support vector machine method combining with radial basis function neural network (SVNN) has been proved no-biased on the modeling of nonlinear system, but SVNN method has seldom been used to solve the problem in industry procedure control by now.
The polymerization process of ABS resin is nonlinear and time-varying. It is sensitive to the temperature and the change of temperature, which has higher requirement on system modeling and controller designing.
In this paper, SVNN method is used model the polymerization process of ABS resin. Furthermore, combing SVNN method with adaptive inverse control method, this paper proposes a new adaptive inverse control method for nonlinear system. By this method, an inverse model of polymerization process of ABS is proposed; the SVNN is trained using data from the production field. This inverse model is used as a controller and its output is used to the controlled object. Meantime, the output error is used to adjust the input of the inverse model controller by the PID method, and the correction of the inverse model controller is realized at the same time. The results of the simulation show that the precision of this method higher than normal neural network, furthermore, the SVNN method shortens the training time and improves the generalization ability. The adaptive inverse control arithmetic based on the support vector neural network has characters of simplicity, credibility, effectivity and robustness. Compared with the normal neural networks controller, the proposed method can adapt the change of model by adjusting the parameters of controller directly, and shows stronger robustness; compared with the conventional adaptive control system, the method proposed in this paper needs no precise mathematical model, so it has extensive application.
This paper's work shows that support vector neural network can not only be used in solving the modeling problem such as temperature control of the ABS resin polymerization process successfully, but also provide a powerful tool in the complicated industry process control.
引文
[1] 钟义信．智能理论与技术．北京：人民邮电出版社，1992
[2] Abhijit S．Pandya，Robert B．Macy．神经网络模式识别及其实现．北京：电子工业出版社，1999
[3] 匡启和，刘建业，姜长生．基于支持向量机的惯导初始对准系，航天控制．2001,3：42-47
[4] Vapnik. The nature of statistical learning theory. New York USA: Spring Verlag,1995(张学工译．统计学习理论的本质．北京：清华大学出版社，2000)
[5] 李凯，郭子雪．基于SVM的函数模拟，河北大学学报(自然科学版)．2001，21(1)：7-8
[6] 王国胜，钟义信．支持向量机的若干新进展，电子学报．2001，29(10)：1397-1400
[7] P.Bartlett, J. Shawe-Taylor. Generalization performance of support vector machine and other pattern classifiers. Cambridge: Schlkopf, C.J.C.Burges, A.J.Smola, editors, 1999, Advance in Kernel Methods----Support vector learning, 43-54
[8] K.P.Muller, A.J.Smola. Predicting time series with support vector machines. Proceedings of ICANN'97, Springer LNCS 1327,1997,999-1004
[9] Smola, A.J., B.Scholkopf. A tutorial on support vector regression, Neurocolt technical report. Royal Holloway College, University of London, 1998
[10] B．Widrow，E．Walach．刘树棠，韩崇昭译．自适应逆控制(第一版)．西安：西安交通大学出版社，2000
[11] 李春文，冯元琨．多变量非线性控制的逆系统方法(第一版)．北京：清华大学出版社，1991
[12] W.C.Chan, C.W. Chan, K.C. Cheung, C.J. Harris. On the modeling of nonlinear dynamic systems using support vector neural networks. Engineering application of artificial intelligence, 2001, 14:105-113
[13] 黄立本，张立基．合成树脂及应用丛书ABS树脂及其应用．北京：化学工业出版社，2001
[14] 吴健华，吴今培．基于支持向量机的模式识别方法，五邑大学学报(自然科学版)．2002，14(1)：6-10
[15] 张学工．关于统计学习理论与支持向量机，自动化学报．2000，26(1)：32-42
[16] 阎辉，张学工，马云潜，李衍达．基于变异函数的径向基核函数参数估计，自动化学报．2002，28(3)：450-455
[17] Cherkassky V, Mulier F. Learning from Data: Concepts, Theory and Methods. NY: John Viley&Sons. 1997
[18] 阎辉，张学工，李衍达．支持向量机与最小二乘法的关系研究，清华大学学报(自然科学版)，2001，41(9)：77-80
[19] 张燕昆，杜平，刘重庆．基于主元分析与支持向量机的人脸识别方法，上海交通大学学报．2002，36(6)：884-886
[20] 卢增祥，李衍达．交互支持向量机学习算法及其应用．清华大学学报(自然科学版)．1999，39(7)：93-97
[21] HuangJ, et al. Face Pose Discrimination Using Support Vector Machines (SVM), Proc. Of 14th International Conference on Pattern Recognition (ICPR'98),1998
[22] 田盛丰，黄厚宽．基于支持向量机的数据库学习算法，计算机研究与发展．2000，37(1)：17-22
[23] Bradley P. Mathematical Programming Approaches to Machine Learning and Data Mining. PhD thesis,
University of Wisconsin, Computer Sciences Department, Madison, WI, USA, TR-98-11, 1998
[24] Suykens JAK, et al. Optimal control by least squares support vector machines, Neural Networks. 2001,14(1):23-25
[25] 刘江华，程君实，陈佳品．支持向量机训练算法综述，信息与控制．2002，31(1)：45-50
[26] Osuna E, Freund R, Girosi F. Training support vector machines: An application to faced election. In Proceedings of CVPR'97,PuertoRico, 1997
[27] John CP. Fast Training of Support Vector Machines using Sequential Minimal Optimization. In ScholkopfB. Et al (ed.), Advances in Kernel Methods-Support Vector Learning, Cam-bridge, MA, MIT Press, 1999,185-208
[28] Keerthi SS, et al. Improvements to Platt's SMO Algorithm for SVM Classier Design, TRCD-99-14, Dept. of Mecha. And Prod. Engin. National Uni. Of Singapore, 1999
[29] Chih WH, et al. A Simple Decomposition Method for Support Vector Machines. Technical report, National Taiwan University, 1999
[30] Friess TT, et al. The kernel adapt on algorithm: A fast and simple learning procedure for support vector machines, ICML98. 1998:188-196
[31] Zhang X. Using class-center vectors to build support vector machines. In Proceedings of NNSP'99,1999
[32] Drezet, P.M.L., Harrison, R.F. Support vector machines for system identification. UKACC International Conference on CONTROL'98, 1-4 September 1998, 688-692
[33] 李爱军，王新民，刘惠英吕．非线性系统的神经网络自适应逆控制，测控技术．2002，21(11)：61-63
[34] 陈小红，侯春海，钱积新．逆系统方法的径向基函数网络实现，控制与决策．1998，3(2)：146-150
[35] 陈小红，钱积新．RBFN逆控制系统综合设计方法，浙江大学学报(工学版)．2000，34(2)：126-129
[36] 沈毅力，王宏贵，李天石．电液伺服加载系统的神经网络逆控制，系统仿真学报．2002，14(12)：1663-1665
[37] 郝晓弘，邵辉，胡冷石，杨新华．基于动态递归网的无刷直流电动机自适应逆控制，计算机仿真．2002，19(1)：101-103
[38] 常东来，沈传文，苏彦民．神经元逆控制在异步机调速系统中的应用，电力电子技术．1998，1：7-13
[39] 李春文，张平，冯元琨．一种基于逆系统方法的化学反应器改进控制方案，控制与决策．1998，13(5)：577-616
[40] 黄田根．PVC聚合釜温串级分程调节系统，化工自动化及仪表．1998，25(6)：17-18
[41] 王玲．PVC聚合釜自动控制系统，氯碱工业．2000，4：33-34
[42] 冯斌，须文波，刘飞，反应釜系统中的模糊串级控制策略，无锡轻工大学学报．1999，18(4)：72-75
[43] 张乃尧，阎平凡．神经网络与模糊控制．北京：清华大学出版社，1998
[44] J. Zhang, A. J. Morris, E. B. Martin, C. Kiparissides, Prediction of polymer quality in batch polymerisation reactors using robust neural networks,Chemical Engineering Journal.1998(69): 135-143
[45] I.M.Galvan ,J.M. Zaldivar, Application of recurrent neural networks in batch reactors Part Ⅱ: Onlinear inverse and predictive control of the heat transfer fluid temperature,Chemical Engineering and Processing, 1998,37:149-161
[46] 吴伟林，基于神经元网络的间歇反应釜智能控制器的设计，微计算机信息．1997，13(5)：11-13
[47] 赵斌，张如辉，齐占庆．基于U-滤波LMS算法的自适应逆控制系统，自动化与仪器仪表．2002，6：7-10
[48] Jackson IRH. Convergence properties of radial basis functions, Constr Approx. 1988, (4):243-264
[49] MS博格著，余庆余译．非线性与泛函分析．北京：科学出版社，1989
[50] 谭永红．基于BP网络的自适应控制，控制理论与应用．1994，11(1)：84-87