前向神经网络控制理论研究及其应用
|
摘要
非线性系统广泛存在于自然界。由于经典和现代控制方法存在一个共同的局限性：就是要求预先知道被控对象的数学模型，但实际上许多对象具有复杂的不确定性和时变性；此外还具有复杂的非线性。虽然在控制理论中有系统辨识的手段，但是对于非线性时变系统尚无成熟的和系统的辨识理论与方法，要实行有效的实时控制就很难了。人工神经元网络有表示任意非线性关系和自学习等能力，给解决这些问题提供了新思想和新方法。
本课题主要针对前向神经网络——BP网络理论与控制器设计进行研究。首先重点对BP网络的结构和学习算法进行了深入研究，揭示了动量因子与网络收敛速度、收敛精度之间的关系，并提出了一种改进的算法。然后研究了采用自适应学习率BP网络的辨识方法，仿真说明其可以自适应地跟踪辨识被控对象。在此基础上，基于补偿控制思想，利用神经PID对传统PID进行补偿，设计了一种混合PID控制器；神经网络与自校正控制的结合，使得自校正方法能对非线性系统实现比较理想的控制效果；利用模糊控制与神经网络控制各自的优点，提出了一种基于快速BP算法的神经网络自适应模糊控制器，能够对系统进行在线控制。最后将混合PID控制应用于H型钢连轧机张力系统中，实现微张力控制，仿真结果说明其较传统PID具有更好的性能，同时也为其它类似系统的控制提供一些参考。
There are lots of non-linear systems in the nature.Because classical and morden control methods have common limitation:mathematical model of the plant has to be known in advance.But in fact many plants are intricate uncentaintied and time-varying.In addition,they also are non-linear.Though there are methods of system identification in control theories,the identification theories and ways of non-linear and time-varying system are not mature and systematic.It is very difficult to carry out effective real-time control.The neural networks can approximate random non-linear relations and study by itself ,and it provides new thoughts and new ways for solving these problems.
This paper focuses on the theories and controller designs of forward neural netwoks—BP network.At first, the structure and algorithms of BP network are deeply researched,the ralations between momentum factor and convergence speed、convergence accuracy are revealed and a kind of improved BP algorithm is presented.Then the identification method based on BP network with adaptive learning rate is studied and the simulaton indicates it can adaptively track the plant. Based on compensation control thought,through using neural network PID controller as compensation tache of traditional PID controller,a kind of hybrid PID controller is designed.Combined with neural network,self-tuning control can realized better control over non-linear system.Utilizing the advantages of fuzzy control and neural network control,a kind of neural network adaptive fuzzy controller based on fast BP algorithm is presented and it may control the plant on line.At last,the hybrid PID controller is applied in tension system of H-section continuous rolling mills and tension-free control is realized.The simulation results show that it has better performance than traditional PID controller.At the same time,it provides some reference for the control of other similar systems.
本课题主要针对前向神经网络——BP网络理论与控制器设计进行研究。首先重点对BP网络的结构和学习算法进行了深入研究，揭示了动量因子与网络收敛速度、收敛精度之间的关系，并提出了一种改进的算法。然后研究了采用自适应学习率BP网络的辨识方法，仿真说明其可以自适应地跟踪辨识被控对象。在此基础上，基于补偿控制思想，利用神经PID对传统PID进行补偿，设计了一种混合PID控制器；神经网络与自校正控制的结合，使得自校正方法能对非线性系统实现比较理想的控制效果；利用模糊控制与神经网络控制各自的优点，提出了一种基于快速BP算法的神经网络自适应模糊控制器，能够对系统进行在线控制。最后将混合PID控制应用于H型钢连轧机张力系统中，实现微张力控制，仿真结果说明其较传统PID具有更好的性能，同时也为其它类似系统的控制提供一些参考。
There are lots of non-linear systems in the nature.Because classical and morden control methods have common limitation:mathematical model of the plant has to be known in advance.But in fact many plants are intricate uncentaintied and time-varying.In addition,they also are non-linear.Though there are methods of system identification in control theories,the identification theories and ways of non-linear and time-varying system are not mature and systematic.It is very difficult to carry out effective real-time control.The neural networks can approximate random non-linear relations and study by itself ,and it provides new thoughts and new ways for solving these problems.
This paper focuses on the theories and controller designs of forward neural netwoks—BP network.At first, the structure and algorithms of BP network are deeply researched,the ralations between momentum factor and convergence speed、convergence accuracy are revealed and a kind of improved BP algorithm is presented.Then the identification method based on BP network with adaptive learning rate is studied and the simulaton indicates it can adaptively track the plant. Based on compensation control thought,through using neural network PID controller as compensation tache of traditional PID controller,a kind of hybrid PID controller is designed.Combined with neural network,self-tuning control can realized better control over non-linear system.Utilizing the advantages of fuzzy control and neural network control,a kind of neural network adaptive fuzzy controller based on fast BP algorithm is presented and it may control the plant on line.At last,the hybrid PID controller is applied in tension system of H-section continuous rolling mills and tension-free control is realized.The simulation results show that it has better performance than traditional PID controller.At the same time,it provides some reference for the control of other similar systems.
引文
1 周其节, 徐建闵.神经网络控制系统的研究与展望.控制理论与应用,1992,9(6):569-577
2 冯纯伯, 刘延平.神经网络控制的现状及问题.控制理论与应用,1994,11(1):103-106
3 刘永红.神经网络理论的发展与前沿问题.信息与控制,1999,28(1):31-46
4 王永骥, 涂健.神经元网络控制.北京:机械工业出版社,1998.2:1-11
5 焦李成.神经网络系统理论.西安:西安电子科技大学出版社,1996:6-15
6 Brooks R A. Intelligence without Representation. Artificial Intelligence,1990,47:139-159
7 袁曾任.智能控制研究的新进展.信息与控制.1994,(5):257-261
8 Fukuda Toshio, Ghibata Takanori. Theory and Applications of Neural Networks for Industrial Control Systems.IEEE Transactions on Industrial Electronics,December 1992,39(6):472-489
9 S. Abid, F. Fnaiech,and M. Najim. A Fast Feedforward Training Algorithm Using a Modified Form of the Standard Backpropagation Algorithm .IEEE Trans. Neural Networks,2001,12(2):424-430
10 G. J, Wang and C. C. Chen. A Fast Multilayer Neural Network Training Algorithm Based on the Layer-by-layer Optimization Procedures.IEEE Trans. Neural Networks, 1996,7:768-775
11 V. R. Sotino and L. C. K. Hui. Use of a Quasi-Newton Method in Feedforward Neural Networks Construction Algorithm.IEEE Trans. Neural Networks,1995,6:272-277
12 M. T. Hugan and M. B. Monhaj. Traning Feedforward Networks with the Marquardt Algorithm.IEEE Trans. Neural Networks,1994,5:989-993
13 姜天戟, 袁曾任.新激活函数下前馈型神经网络及其在天气预报中的应用.信息与控制,1995,.24(1):47-51
14 袁曾任.人工神经元网络极其应用.北京:清华大学出版社,1999:66-74
15 丛爽.神经网络、模糊系统及其在运动控制中的应用.合肥:中国科学技术大学出版社,2001.5:20-31
16 高雪鹏, 丛爽.BP网络改进算法的性能对比研究.控制与决策,2001,16(2):167-171.
17 张立明.人工神经网络的模型及其应用.上海:复旦大学出版社,1992:32-47
18 王旭, 王宏, 王文辉.人工神经元网络原理与应用.沈阳:东北大学出版社,2000.12:51-56
19 Tian Y. RLS Training Algorithm for Multilayer Feedforward Neural Networks and Its Application to System Identification.Control Theory and Application,1994,11(5):594-599
20 Narebdra K. S. and Parthasarathy K. Identification and Control for Dynamic Systems Using Neural Networks.IEEE Trans. Neural Networks,1990,1(1):4-27
21 G.Cybenco. Approximation by Superpositions of a Sigmoidan Function.Math.Contr. Signal Sys,1989,2(4):303-314
22 K.I. Funahashi. On the Approximate Realization of Continuous Mapping by Neural Networks.Intel.Conf.NN,1989
23 K.Hornik. Approximation Capabilities of Multilayer Feedforward Network.Neural Networks,1991,2(4):251-257
24 T. Chen&H. Chen. Approximation to Nonlinear Continuous Functions by Neural Networks with Application to Dynamic System.IEEE Transaction on Neural Networks, 1993,3(4):910-918
25 K. Hornik, M. Stinchcombe & H. White. Universal Approximation of an Unknown Mapping and Its Derivatives Using Multilayer Feedforward Networks.Neural Networks, 1990,3:551-560
26 W.D. Hu & Z.Z. Wang. The Approximation of Arbitrary Functions with Multilayer BP Neural Networks.Proc.IJCNN'92,Beijing,China,1992
27 何玉彬, 李新忠.神经网络控制技术及其应用.北京:科学出版社,2000:44-54
28 焦李成.神经网络的应用与实现.西安:西安电子科技大学出版社,1996:319-333
29 鲍晓红, 贾英民.用含动态隐层的前馈网辨识非线性系统.自动化学报,1997,23(5):689-693
30 徐耀玲, 戴汝为.人工神经网络在系统辨识中的应用.自动化学报,1991,17(1):91-94
31 阎平凡, 张长水.人工神经网络与模拟进化计算.北京:清华大学出版社,2000:23-27
32 TAN Yonghong and DANG Xuanju,Achiel Van Cauwenberghe. PID Gradient Algorithm for Neural Network Based Generalized Non-linear PID Controller.Control Theory and Application,2000,17(6):861-867
33 王耀南, 童调生, 蔡自兴.基于神经元网络的智能PID控制及应用.信息与控制,1994, 23(3):185-189
扈宏杰, 陈敬泉, 刘强等.基于神经网络补偿的PID控制方法的研究.电气传
34 动,2000,5:3-5
35 谭永红.神经网络自适应PID控制及其应用.模式识别与人工智能,1993,6(1):81-85
36 闻新, 周露, 王丹力等.MATLAB神经网络应用设计.北京:科学出版社,2000:212-243
37 程卫国, 冯峰, 姚东等.MATLAB5.3应用指南.北京:人民邮电出版社, 1999:178-189
38 施阳, 李俊, 王惠刚等.MATLAB语言工具箱——TOOLBOX实用指南.西安:西北工业大学出版社,1999:149-156
39 罗小青, 孙优贤.基于神经网络的前馈学习控制器.信息与控制,1994,23(4):311-314
40 陈卫田, 施颂椒.基于神经网络的非线性自适应控制.控制理论与应用,1996,13(5):545-552
41 Fu-chuang Chen. Back-Propagation Networks for Neural Self-Tuning Adaptive Control.IEEE Conrol Systems Magazine ,1990,4(1):44-48
42 Yoshiaki I et al. Neural Network Application for Direct Feedback Controller.IEEE Trans Neural Networks,1992,3(2):224-231
43 K. S. Narendra and S.Mukhopadhyay. Adaptive Control of Nonlinear Multivariable System Using Neural Networks.Neural Networks,1994,7(5):737-752
44 Ahmed M. S. Neural Net Based MRAC for a Class of Non-linear Plants. IEEE Trans. Neural Networks,2000,13(1):111-124
45 F. Sun,Z. Sun and P.-Y. Woo. Stable Neural-Network-based Adaptive Control for Sampled Data Systems. IEEE Trans. Neural Networks,1998,9:956-968
46 黄战, 戴冠中.神经网络控制器的研究.信息与控制,1991,20(1):27-31
47 谭永红.基于BP神经网络的自适应控制.控制理论与应用,1994,11(1):84-87
48 谭永红.采用神经网络的自校正控制.人工智能与模式识别,1991,4(3):75-79
49 沈瑛, 张翠芳.基于BP神经网络的模型参考自适应控制.西南交通大学学报,2001,36(5):553-556
50 徐丽娜.神经网络控制.哈尔滨:哈尔滨工业大学出版社,1999.5:67-77
51 杨子厚.神经网络技术在钢铁工业中的应用——神经网络在控制系统中的应用(上).冶金自动化,1996,21(2):48-52
52 杨子厚.神经网络技术在钢铁工业中的应用——神经网络在控制系统中的应用(下).冶金自动化,1997,21(3):48-55
53 王耀南.神经网络自适应模糊控制在温度控制系统中的应用.信息与控制,1996,25(4):245-251
54 李士勇.模糊控制·神经网络和智能控制.哈尔滨:哈尔滨工业大学出版社,1998:408-418
55 应行仁, 曾南.采用BP神经网络记忆模糊规则的控制.自动化学报,1991,17(1):63-67
56 刘曙光, 魏俊民, 竺志超.模糊控制技术.北京:中国纺织出版社,2001.6:202-212
57 彭小奇, 梅炽, 周孑民等.多变量系统的模糊神经网络控制模型及其应用.控制理论与应用,1995,12(3):351-359
58 陈非, 敬忠良, 姚晓东.一种模糊神经网络的快速参数学习算法.控制理论与应用,2002,19(4):583-587
59 Wang L X,Mendel J M. Fuzzy Basis Function,Universal Approximation,and Orthogonal Least Squares Learning.IEEE Trans.Neural Networks,1992,3(5):807-814
60 张乃尧, 阎平凡.神经网络与模糊控制.北京:清华大学出版社,1998:242-251
61 孙霆.连轧张力控制及其仿真研究.燕山大学硕士学位论文,1988:55-72
62 姜平, 宋文忠.神经元自适应PID控制器的研究.电气自动化,1998,(1):11-13
63 周箴, 赵金, 万淑芸等.模糊单神经元混合控制在交流传动系统中的应用.电气传动,1998,(1):21-23
64 陈伯时.电力拖动自动控制系统.北京:机械工业出版社,1992.5:76-90
65 丁科, 候朝桢, 梁文林.神经网络模糊控制在交流调速系统中的应用研究.控制与决策,2001,16(4):500-503
66 胡建元, 吴隆安, 黄心汉等.单个神经元的自适应控制.电气自动化,1993,15(1):19-20
67 王晓东, 陈伯时, 夏承光.基于S型函数单神经控制元的直流传动系统.电气自动化,1996,(5):4-7
68 王京, 陈辉.对角回归神经网络在直流双闭环调速系统中的应用.电气传动,1999，(4):25-28
2 冯纯伯, 刘延平.神经网络控制的现状及问题.控制理论与应用,1994,11(1):103-106
3 刘永红.神经网络理论的发展与前沿问题.信息与控制,1999,28(1):31-46
4 王永骥, 涂健.神经元网络控制.北京:机械工业出版社,1998.2:1-11
5 焦李成.神经网络系统理论.西安:西安电子科技大学出版社,1996:6-15
6 Brooks R A. Intelligence without Representation. Artificial Intelligence,1990,47:139-159
7 袁曾任.智能控制研究的新进展.信息与控制.1994,(5):257-261
8 Fukuda Toshio, Ghibata Takanori. Theory and Applications of Neural Networks for Industrial Control Systems.IEEE Transactions on Industrial Electronics,December 1992,39(6):472-489
9 S. Abid, F. Fnaiech,and M. Najim. A Fast Feedforward Training Algorithm Using a Modified Form of the Standard Backpropagation Algorithm .IEEE Trans. Neural Networks,2001,12(2):424-430
10 G. J, Wang and C. C. Chen. A Fast Multilayer Neural Network Training Algorithm Based on the Layer-by-layer Optimization Procedures.IEEE Trans. Neural Networks, 1996,7:768-775
11 V. R. Sotino and L. C. K. Hui. Use of a Quasi-Newton Method in Feedforward Neural Networks Construction Algorithm.IEEE Trans. Neural Networks,1995,6:272-277
12 M. T. Hugan and M. B. Monhaj. Traning Feedforward Networks with the Marquardt Algorithm.IEEE Trans. Neural Networks,1994,5:989-993
13 姜天戟, 袁曾任.新激活函数下前馈型神经网络及其在天气预报中的应用.信息与控制,1995,.24(1):47-51
14 袁曾任.人工神经元网络极其应用.北京:清华大学出版社,1999:66-74
15 丛爽.神经网络、模糊系统及其在运动控制中的应用.合肥:中国科学技术大学出版社,2001.5:20-31
16 高雪鹏, 丛爽.BP网络改进算法的性能对比研究.控制与决策,2001,16(2):167-171.
17 张立明.人工神经网络的模型及其应用.上海:复旦大学出版社,1992:32-47
18 王旭, 王宏, 王文辉.人工神经元网络原理与应用.沈阳:东北大学出版社,2000.12:51-56
19 Tian Y. RLS Training Algorithm for Multilayer Feedforward Neural Networks and Its Application to System Identification.Control Theory and Application,1994,11(5):594-599
20 Narebdra K. S. and Parthasarathy K. Identification and Control for Dynamic Systems Using Neural Networks.IEEE Trans. Neural Networks,1990,1(1):4-27
21 G.Cybenco. Approximation by Superpositions of a Sigmoidan Function.Math.Contr. Signal Sys,1989,2(4):303-314
22 K.I. Funahashi. On the Approximate Realization of Continuous Mapping by Neural Networks.Intel.Conf.NN,1989
23 K.Hornik. Approximation Capabilities of Multilayer Feedforward Network.Neural Networks,1991,2(4):251-257
24 T. Chen&H. Chen. Approximation to Nonlinear Continuous Functions by Neural Networks with Application to Dynamic System.IEEE Transaction on Neural Networks, 1993,3(4):910-918
25 K. Hornik, M. Stinchcombe & H. White. Universal Approximation of an Unknown Mapping and Its Derivatives Using Multilayer Feedforward Networks.Neural Networks, 1990,3:551-560
26 W.D. Hu & Z.Z. Wang. The Approximation of Arbitrary Functions with Multilayer BP Neural Networks.Proc.IJCNN'92,Beijing,China,1992
27 何玉彬, 李新忠.神经网络控制技术及其应用.北京:科学出版社,2000:44-54
28 焦李成.神经网络的应用与实现.西安:西安电子科技大学出版社,1996:319-333
29 鲍晓红, 贾英民.用含动态隐层的前馈网辨识非线性系统.自动化学报,1997,23(5):689-693
30 徐耀玲, 戴汝为.人工神经网络在系统辨识中的应用.自动化学报,1991,17(1):91-94
31 阎平凡, 张长水.人工神经网络与模拟进化计算.北京:清华大学出版社,2000:23-27
32 TAN Yonghong and DANG Xuanju,Achiel Van Cauwenberghe. PID Gradient Algorithm for Neural Network Based Generalized Non-linear PID Controller.Control Theory and Application,2000,17(6):861-867
33 王耀南, 童调生, 蔡自兴.基于神经元网络的智能PID控制及应用.信息与控制,1994, 23(3):185-189
扈宏杰, 陈敬泉, 刘强等.基于神经网络补偿的PID控制方法的研究.电气传
34 动,2000,5:3-5
35 谭永红.神经网络自适应PID控制及其应用.模式识别与人工智能,1993,6(1):81-85
36 闻新, 周露, 王丹力等.MATLAB神经网络应用设计.北京:科学出版社,2000:212-243
37 程卫国, 冯峰, 姚东等.MATLAB5.3应用指南.北京:人民邮电出版社, 1999:178-189
38 施阳, 李俊, 王惠刚等.MATLAB语言工具箱——TOOLBOX实用指南.西安:西北工业大学出版社,1999:149-156
39 罗小青, 孙优贤.基于神经网络的前馈学习控制器.信息与控制,1994,23(4):311-314
40 陈卫田, 施颂椒.基于神经网络的非线性自适应控制.控制理论与应用,1996,13(5):545-552
41 Fu-chuang Chen. Back-Propagation Networks for Neural Self-Tuning Adaptive Control.IEEE Conrol Systems Magazine ,1990,4(1):44-48
42 Yoshiaki I et al. Neural Network Application for Direct Feedback Controller.IEEE Trans Neural Networks,1992,3(2):224-231
43 K. S. Narendra and S.Mukhopadhyay. Adaptive Control of Nonlinear Multivariable System Using Neural Networks.Neural Networks,1994,7(5):737-752
44 Ahmed M. S. Neural Net Based MRAC for a Class of Non-linear Plants. IEEE Trans. Neural Networks,2000,13(1):111-124
45 F. Sun,Z. Sun and P.-Y. Woo. Stable Neural-Network-based Adaptive Control for Sampled Data Systems. IEEE Trans. Neural Networks,1998,9:956-968
46 黄战, 戴冠中.神经网络控制器的研究.信息与控制,1991,20(1):27-31
47 谭永红.基于BP神经网络的自适应控制.控制理论与应用,1994,11(1):84-87
48 谭永红.采用神经网络的自校正控制.人工智能与模式识别,1991,4(3):75-79
49 沈瑛, 张翠芳.基于BP神经网络的模型参考自适应控制.西南交通大学学报,2001,36(5):553-556
50 徐丽娜.神经网络控制.哈尔滨:哈尔滨工业大学出版社,1999.5:67-77
51 杨子厚.神经网络技术在钢铁工业中的应用——神经网络在控制系统中的应用(上).冶金自动化,1996,21(2):48-52
52 杨子厚.神经网络技术在钢铁工业中的应用——神经网络在控制系统中的应用(下).冶金自动化,1997,21(3):48-55
53 王耀南.神经网络自适应模糊控制在温度控制系统中的应用.信息与控制,1996,25(4):245-251
54 李士勇.模糊控制·神经网络和智能控制.哈尔滨:哈尔滨工业大学出版社,1998:408-418
55 应行仁, 曾南.采用BP神经网络记忆模糊规则的控制.自动化学报,1991,17(1):63-67
56 刘曙光, 魏俊民, 竺志超.模糊控制技术.北京:中国纺织出版社,2001.6:202-212
57 彭小奇, 梅炽, 周孑民等.多变量系统的模糊神经网络控制模型及其应用.控制理论与应用,1995,12(3):351-359
58 陈非, 敬忠良, 姚晓东.一种模糊神经网络的快速参数学习算法.控制理论与应用,2002,19(4):583-587
59 Wang L X,Mendel J M. Fuzzy Basis Function,Universal Approximation,and Orthogonal Least Squares Learning.IEEE Trans.Neural Networks,1992,3(5):807-814
60 张乃尧, 阎平凡.神经网络与模糊控制.北京:清华大学出版社,1998:242-251
61 孙霆.连轧张力控制及其仿真研究.燕山大学硕士学位论文,1988:55-72
62 姜平, 宋文忠.神经元自适应PID控制器的研究.电气自动化,1998,(1):11-13
63 周箴, 赵金, 万淑芸等.模糊单神经元混合控制在交流传动系统中的应用.电气传动,1998,(1):21-23
64 陈伯时.电力拖动自动控制系统.北京:机械工业出版社,1992.5:76-90
65 丁科, 候朝桢, 梁文林.神经网络模糊控制在交流调速系统中的应用研究.控制与决策,2001,16(4):500-503
66 胡建元, 吴隆安, 黄心汉等.单个神经元的自适应控制.电气自动化,1993,15(1):19-20
67 王晓东, 陈伯时, 夏承光.基于S型函数单神经控制元的直流传动系统.电气自动化,1996,(5):4-7
68 王京, 陈辉.对角回归神经网络在直流双闭环调速系统中的应用.电气传动,1999，(4):25-28