正规化模糊神经网络及其在提钒炼钢智能控制模型中的应用研究
|
摘要
控制学术权威Austrom指出：模糊逻辑控制、神经网络和专家系统是三种典型的智能控制方法。由于专家系统在实际应用中有较多的问题和困难(如：知识的获取主要靠人工移植，因此成本高、效率低；采用常规的分明集，致使它的推理能力弱，容易产生匹配冲突、组合爆炸等问题)，现在智能控制的重点则集中在模糊逻辑、神经网络以及二者的结合应用上。特别是二者的结合，由于融合了各自的优点，并能在一定程度上克服二者一些固有的缺陷，因而已受到人们的广泛关注，成为智能控制研究的热点。
本文针对单纯模糊系统在传统建模方法上的缺陷：难以提取和调整模糊规则与隶属函数，并且其工作量往往随变量数目的增长呈指数级增长，在前人提出的正规化模糊神经网络(NFNN)结构基础上，改进了已有的规则合并的算法。利用这种算法，可以将由神经网络自动获取的模糊规则进行合并与调整，并通过一个函数的仿真实验验证了算法的有效性。
转炉炼钢过程是一个复杂的多元多相高温反应过程，其模型的建立则是一个典型的多元非线性的映射过程，尤其在大多中小型转炉不宜增设副枪的条件下，如何在传统的静态模型之外寻找性能更优越的控制模型，就显得比较突出和迫切。本文即是把NFNN应用于某钢厂转炉提钒炼钢静态控制模型的建立问题之中，建立了转炉炼钢的冷却剂加料子模型。论文最后的测试结果说明了这种建模方式的有效性与实用性，也再次验证了规则合并的有效性。
Professor Austrom points out that fuzzy logic control, neural networks and specialist system are three typical intelligent control methods. Specialist system faces many problems and difficulties, such as knowledge acquirement depends on manual transplantation which leads to high cost and low efficiency; applying general clarified set brings forward with feeble reasoning ability, matching collision and combination explosion. Nowadays modern intelligent control focuses on fuzzy logic and neural networks or the combination of the two. Especially the combination of fuzzy logic and neural networks has already become the hotspot in intelligent control research because it absorbs the strongpoint while overcomes the weak point of both.
This paper aims at the two main limitations of pure fuzzy system in traditional modeling. One is the difficulties in extracting and adjusting the fuzzy rules and subject functions, the other is the workload increasing exponentially along with the number of variables. So we improve the old regular combination arithmetic on the base of normal fuzzy neural networks (NFNN). Through the arithmetic, we can assemble and combine fuzzy rules gained by neural networks automatically. Furthermore we test its effectiveness by functional simulation experiment.
Converter steel-making is a complicated diverse high temperature reaction process and the foundation of the model also is a diverse non-linear mapped process. So it is emergent to find a more superior control model abandoned the tradition static model in the condition that many medium and small converter cannot use an assistant gun.
Moreover, we apply NFNN in the problems of static controlling modeling of steel-making and abstract vanadium by converter and come up with the refrigerant adding sub-model. The simulated result accounts for the usefulness and practicability of the modeling method and the regular combination once again.
本文针对单纯模糊系统在传统建模方法上的缺陷：难以提取和调整模糊规则与隶属函数，并且其工作量往往随变量数目的增长呈指数级增长，在前人提出的正规化模糊神经网络(NFNN)结构基础上，改进了已有的规则合并的算法。利用这种算法，可以将由神经网络自动获取的模糊规则进行合并与调整，并通过一个函数的仿真实验验证了算法的有效性。
转炉炼钢过程是一个复杂的多元多相高温反应过程，其模型的建立则是一个典型的多元非线性的映射过程，尤其在大多中小型转炉不宜增设副枪的条件下，如何在传统的静态模型之外寻找性能更优越的控制模型，就显得比较突出和迫切。本文即是把NFNN应用于某钢厂转炉提钒炼钢静态控制模型的建立问题之中，建立了转炉炼钢的冷却剂加料子模型。论文最后的测试结果说明了这种建模方式的有效性与实用性，也再次验证了规则合并的有效性。
Professor Austrom points out that fuzzy logic control, neural networks and specialist system are three typical intelligent control methods. Specialist system faces many problems and difficulties, such as knowledge acquirement depends on manual transplantation which leads to high cost and low efficiency; applying general clarified set brings forward with feeble reasoning ability, matching collision and combination explosion. Nowadays modern intelligent control focuses on fuzzy logic and neural networks or the combination of the two. Especially the combination of fuzzy logic and neural networks has already become the hotspot in intelligent control research because it absorbs the strongpoint while overcomes the weak point of both.
This paper aims at the two main limitations of pure fuzzy system in traditional modeling. One is the difficulties in extracting and adjusting the fuzzy rules and subject functions, the other is the workload increasing exponentially along with the number of variables. So we improve the old regular combination arithmetic on the base of normal fuzzy neural networks (NFNN). Through the arithmetic, we can assemble and combine fuzzy rules gained by neural networks automatically. Furthermore we test its effectiveness by functional simulation experiment.
Converter steel-making is a complicated diverse high temperature reaction process and the foundation of the model also is a diverse non-linear mapped process. So it is emergent to find a more superior control model abandoned the tradition static model in the condition that many medium and small converter cannot use an assistant gun.
Moreover, we apply NFNN in the problems of static controlling modeling of steel-making and abstract vanadium by converter and come up with the refrigerant adding sub-model. The simulated result accounts for the usefulness and practicability of the modeling method and the regular combination once again.
引文
[1]潘树范. 国内外氧气顶吹转炉提钒现状及对攀钢转炉提钒有关问题的探讨.钢铁钒钛.第16卷.第1期.1995
[2]Austrom K J.Directions inIntelligent Control.Plenary Session.IFAC International Symposium. ITAC91 （Intelligent Tuning and Adaptive Control .1991: 15-17
[3]Mamdani E H, Assilian S. An Experiment in Linguistic Synthesis with a Fuzzy Logic Controller. Int J Man-Machine Studies. 1975. 7:1-13
[4]Holmblad L P, Ostergaard J J. Control of a Kiln by Fuzzy Logic. In: Gupta M M, Sanchez E, Eds. Fuzzy Information and Decision Problems. Amsterdam: North-Holland.1982: 25-29
[5]谢康林, 傅劲游. 神经模糊逻辑控制系统隶属函数和推理规则的确定.上海交通大学学报.1997.8: 28-33
[6]Procyk T J，Mamdani E H. A Linguistic Self-Organizing Process Controller.Automatic.1979.15: 15-30
[7]Kosko B. Neural networks and fuzzy systems: A dynamical systems approach to intelligence. NJ: Prentice-Hall. 1992
[8]Isaka S. On Neural Approximation of Fuzzy Systems. IJCNN'92. Baltimore.1992. 1: 263—268
[9]Kawamurn A, et al. A Prototype of Neural-Fuzzy Cooperation Systems. IEEE Fuzzy'92.1992. 1:1276-1282
[10]Okada H, et al. Initializing Multilayer Neural Networks with Fuzzy Logic.IJCNN'92. 1992 :239-244
[11]Dumitrache I, et al. A Real Time Control Experiment Using a Neural Network Based Linguistical Approach. IEEE Fuzzy'92,.1992: 123-127
[12]Nagao T, et al. A Genetic Method for Optimization of Asynchronous Random Neural Networks and its Application to Action Control. IJCNN'93. 1993: 1893-1896
[13]Arabshahi P, et al. Fuzzy Control of Backpropagation, IEEE Fuzzy'92.1992: 967-972
[14]Choi Jai, et al. Fuzzy parameter adaptation in neural systems. IJCNN'92.Baltimore.1992:232-238
[15]张良杰, 李衍达. 基于模糊逻辑与组合插值技术的新型一维全局优化算法. 人工智能与模式识别专辑. 清华大学学报.1994.Sup4
[16]张良杰, 李衍达. 模糊神经网络技术的新近发展. 信息与控制. 1995. 2: 39-44
[17]阎平凡, 张长水. 人工神经网络与模拟进化计算. 2000. 清华大学出版社
[18]Rumelhart D E, Hinton G E, Williams R J. Learning Representation by Backpropagation Erros[J].Nature .1986. 323(6188): 533-536
[19]Wang L X，Mendel J M. Fuzzy basis functions. universal approximation and orthogonal least-squares learning. IEEE Trans. On Neural Networks. 1992. vol. 3. 807-814
[20]Jang J S R. Functional Equivalence Between Radial Basis Function Networks and Fuzzy Inference System. IEEE Trans. On NN. 1993 (1): 156-158
[21]Hunt K J.Extending the Functional Equivalence of Radial Basis Function Network and Fuzzy Inference System. IEEE Trans. On NN. 1996 (3): 776-781
[22]鲍鸿, 黄心汉,李锡雄. 模糊RBF网络简化模型研究. 华中理工大学学报. 第26卷. 第9期. 1998
[23]Isaka S. On input space clustering by fuzzy systems and neural networks. Proceedings of 1992 IEEE Int. Conf. on systems, Man, and Cybernetics. 1992. 1597-1602
[24]Hayashi Y, Buckley J J. Fuzzy neural networks: A survy. Fuzzy Sets and Systems,1994,66(1):1-13
[25]Buckley J J, Hayashi Y. Fuzzy neural nets and applications. Fuzzy Systems and AI. 1992. 5(3): 11-41
[26]Tagaki H, Suzuki N, T Koda et al. Neural networks designed on approximate reasoning architecture and their applications. IEEE Tran NN. 1992. 3(5): 752-760
[27]刘普寅, 汪浩. 正则模糊神经网络对于连续模糊函数的近似能力研究. 中国科学（E辑）.1999. 2: 54-60；
[28]黄道鑫, 提钒炼钢. 冶金工业出版社.2000:21-48
[29]陈家祥, 钢铁冶金学. 冶金工业出版社.1990: 140-176
[30]杨素波, 黎之政, 杜德信. 转炉炼钢过程自动控制实现方法. 钢铁钒钛. 1996. 3: 47-52
[31]陶钧, 谢书明, 柴天佑. 转炉炼钢控制模型的研究与展望. 钢铁. 1999. 8: 69-72
[32]丁容, 刘浏. 转炉炼钢过程人工智能静态控制模型. 钢铁. 1997. 1: 22-27
[33]喻淑仁. 转炉炼钢过程静态控制及其数学模型. 炼钢. 1995. 6: 55-60
[34]邹喜正, 朱明华. 对模糊聚类分析中原始数据预处理研究. 中国矿业大学学报.1998: 24-29
[35]苗立靖, 杨杰，黄欣. 模糊规则提取的两种方法性能分析. 模糊系统与数学. 1999. 9: 16-21
[36]Nie J H , Linkens D A. Back-propagation neural network based fuzzy controller with a self-learning teacher. Int J Control .1994.60(1) : 17-39
[37]苏彩虹, 谢海鸿，赵海洋.正规化模糊神经网络在自动车控制中的应用. 佛山技术学院学报（自然版）.1999.6: 19-23
[38]王士同. 神经模糊系统及其应用. 北京航空航天大学出版社. 1998:147-156
[39]赵振宇,徐用懋 . 模糊理论和神经网络的基础与应用. 清华大学出版社. 1996:160-161
[2]Austrom K J.Directions inIntelligent Control.Plenary Session.IFAC International Symposium. ITAC91 （Intelligent Tuning and Adaptive Control .1991: 15-17
[3]Mamdani E H, Assilian S. An Experiment in Linguistic Synthesis with a Fuzzy Logic Controller. Int J Man-Machine Studies. 1975. 7:1-13
[4]Holmblad L P, Ostergaard J J. Control of a Kiln by Fuzzy Logic. In: Gupta M M, Sanchez E, Eds. Fuzzy Information and Decision Problems. Amsterdam: North-Holland.1982: 25-29
[5]谢康林, 傅劲游. 神经模糊逻辑控制系统隶属函数和推理规则的确定.上海交通大学学报.1997.8: 28-33
[6]Procyk T J，Mamdani E H. A Linguistic Self-Organizing Process Controller.Automatic.1979.15: 15-30
[7]Kosko B. Neural networks and fuzzy systems: A dynamical systems approach to intelligence. NJ: Prentice-Hall. 1992
[8]Isaka S. On Neural Approximation of Fuzzy Systems. IJCNN'92. Baltimore.1992. 1: 263—268
[9]Kawamurn A, et al. A Prototype of Neural-Fuzzy Cooperation Systems. IEEE Fuzzy'92.1992. 1:1276-1282
[10]Okada H, et al. Initializing Multilayer Neural Networks with Fuzzy Logic.IJCNN'92. 1992 :239-244
[11]Dumitrache I, et al. A Real Time Control Experiment Using a Neural Network Based Linguistical Approach. IEEE Fuzzy'92,.1992: 123-127
[12]Nagao T, et al. A Genetic Method for Optimization of Asynchronous Random Neural Networks and its Application to Action Control. IJCNN'93. 1993: 1893-1896
[13]Arabshahi P, et al. Fuzzy Control of Backpropagation, IEEE Fuzzy'92.1992: 967-972
[14]Choi Jai, et al. Fuzzy parameter adaptation in neural systems. IJCNN'92.Baltimore.1992:232-238
[15]张良杰, 李衍达. 基于模糊逻辑与组合插值技术的新型一维全局优化算法. 人工智能与模式识别专辑. 清华大学学报.1994.Sup4
[16]张良杰, 李衍达. 模糊神经网络技术的新近发展. 信息与控制. 1995. 2: 39-44
[17]阎平凡, 张长水. 人工神经网络与模拟进化计算. 2000. 清华大学出版社
[18]Rumelhart D E, Hinton G E, Williams R J. Learning Representation by Backpropagation Erros[J].Nature .1986. 323(6188): 533-536
[19]Wang L X，Mendel J M. Fuzzy basis functions. universal approximation and orthogonal least-squares learning. IEEE Trans. On Neural Networks. 1992. vol. 3. 807-814
[20]Jang J S R. Functional Equivalence Between Radial Basis Function Networks and Fuzzy Inference System. IEEE Trans. On NN. 1993 (1): 156-158
[21]Hunt K J.Extending the Functional Equivalence of Radial Basis Function Network and Fuzzy Inference System. IEEE Trans. On NN. 1996 (3): 776-781
[22]鲍鸿, 黄心汉,李锡雄. 模糊RBF网络简化模型研究. 华中理工大学学报. 第26卷. 第9期. 1998
[23]Isaka S. On input space clustering by fuzzy systems and neural networks. Proceedings of 1992 IEEE Int. Conf. on systems, Man, and Cybernetics. 1992. 1597-1602
[24]Hayashi Y, Buckley J J. Fuzzy neural networks: A survy. Fuzzy Sets and Systems,1994,66(1):1-13
[25]Buckley J J, Hayashi Y. Fuzzy neural nets and applications. Fuzzy Systems and AI. 1992. 5(3): 11-41
[26]Tagaki H, Suzuki N, T Koda et al. Neural networks designed on approximate reasoning architecture and their applications. IEEE Tran NN. 1992. 3(5): 752-760
[27]刘普寅, 汪浩. 正则模糊神经网络对于连续模糊函数的近似能力研究. 中国科学（E辑）.1999. 2: 54-60；
[28]黄道鑫, 提钒炼钢. 冶金工业出版社.2000:21-48
[29]陈家祥, 钢铁冶金学. 冶金工业出版社.1990: 140-176
[30]杨素波, 黎之政, 杜德信. 转炉炼钢过程自动控制实现方法. 钢铁钒钛. 1996. 3: 47-52
[31]陶钧, 谢书明, 柴天佑. 转炉炼钢控制模型的研究与展望. 钢铁. 1999. 8: 69-72
[32]丁容, 刘浏. 转炉炼钢过程人工智能静态控制模型. 钢铁. 1997. 1: 22-27
[33]喻淑仁. 转炉炼钢过程静态控制及其数学模型. 炼钢. 1995. 6: 55-60
[34]邹喜正, 朱明华. 对模糊聚类分析中原始数据预处理研究. 中国矿业大学学报.1998: 24-29
[35]苗立靖, 杨杰，黄欣. 模糊规则提取的两种方法性能分析. 模糊系统与数学. 1999. 9: 16-21
[36]Nie J H , Linkens D A. Back-propagation neural network based fuzzy controller with a self-learning teacher. Int J Control .1994.60(1) : 17-39
[37]苏彩虹, 谢海鸿，赵海洋.正规化模糊神经网络在自动车控制中的应用. 佛山技术学院学报（自然版）.1999.6: 19-23
[38]王士同. 神经模糊系统及其应用. 北京航空航天大学出版社. 1998:147-156
[39]赵振宇,徐用懋 . 模糊理论和神经网络的基础与应用. 清华大学出版社. 1996:160-161