基于神经网络的转炉炼钢终点控制的研究
摘要
转炉炼钢是生产钢铁的一道关键工序,它所要完成的是冶炼出温度和各种成分(关键是含碳量)都达到预期所需要的钢水。因为钢水的温度及碳含量无法连续检测,并且冶炼过程的边界条件变化复杂,这就导致冶炼过程的终点控制有很大困难,在实际生产中,很难精准掌控钢水碳、温从而导致重复吹炼,因此提高转炉炼钢的终点命中率是非常有必要的。因为转炉炼钢过程是一个相当复杂的多元多相高温物理化学过程,他的机理的解析还不是很透彻,输入输出之间的非线性化非常突出,常规建模方法总是无法满足生产要求。因此,研究基于神经网络的转炉炼钢控制问题是必要的。
本文,首先简单介绍了转炉炼钢原理和设备状况和转炉冶炼终点控制技术的概况和现状。其次,通过对几种BP算法之间特性归纳和比较,得出Levenberg ?Marquardt(LM)算法拥有收敛速度快,并且学习性能好的特性。再次,简要介绍了神经网络的方法及几种神经网络转炉炼钢终点预报模型。最后,提出转炉炼钢基于LM算法的神经网络智能预报模型和控制模型。
本文研究的方法有充实的理论基础,通过了理论和试验的双重验证。在研究过程中选用某转炉炼钢厂60炉实际生产数据为样本,用对转炉冶炼终点温度和碳有影响因素作为输入参数,提出了有三层网络结构的BP神经网络预报和控制模型,也就是先对终点温度和含碳量进行预报,然后以此为前提决定补吹时期所需的吹氧量和辅助原料量。通过仿真研究表明了此方法有效,能够在实际转炉炼钢使用。
BOF is a key process to produce steel, whose task is smelting steel that meets the desired requirements about the temperature and various components. In the actual production process, because the temperature of molten steel can be measured continuouly, and the boundary condition of smelting process is very complex, the terminal control of smelting process is achieved difficultly. Because the precise control of carbon and temperature of steel is very difficult, it is very necessary to improve the terminal control efficiency of BOF. Meantime, because the process of BOF is a very complex physical and chemical processes with multi-component multi-phase and high temperature, whose mechanism about the input and output is nonlinear and not very clear , the conventional modeling methods can not always meet the production requirements.
In this paper, firstly, the principle and equipment and the overview and status about terminal control technology of BOF are introduced briefly. Secondly, the neural network and several predicting methods of terminal model of BOF by using the neural network are introduced briefly. Thirdly, the result that the Levenberg-Marquardt(LM) method has fast convergence and good learning performance is got by summarizing and comparing features between several BP algorithm. Finaly, the intelligent forecast and control model about BOF besed on LM neural network algorithm is proposed.
In this paper the researched method has a substantial theoretic base and has been validated by the theory and experiment. The actual data of continuous 60 batches from a converter steel mill are chosen as example, input parameter that influence and endpoint steel temperatures in converter are determined. In this paper establish three-layers BP prediction neural networks, endpoint various and steeltemperature content. The model of endpoint various and steel temperature prediction have established. On the basis of this, the method based on BP neural network is proposed so as to determine the blown oxygen and auxiliary raw materials. Simulationed showed the method is effective and can be used to practical process to produce steel.
本文,首先简单介绍了转炉炼钢原理和设备状况和转炉冶炼终点控制技术的概况和现状。其次,通过对几种BP算法之间特性归纳和比较,得出Levenberg ?Marquardt(LM)算法拥有收敛速度快,并且学习性能好的特性。再次,简要介绍了神经网络的方法及几种神经网络转炉炼钢终点预报模型。最后,提出转炉炼钢基于LM算法的神经网络智能预报模型和控制模型。
本文研究的方法有充实的理论基础,通过了理论和试验的双重验证。在研究过程中选用某转炉炼钢厂60炉实际生产数据为样本,用对转炉冶炼终点温度和碳有影响因素作为输入参数,提出了有三层网络结构的BP神经网络预报和控制模型,也就是先对终点温度和含碳量进行预报,然后以此为前提决定补吹时期所需的吹氧量和辅助原料量。通过仿真研究表明了此方法有效,能够在实际转炉炼钢使用。
BOF is a key process to produce steel, whose task is smelting steel that meets the desired requirements about the temperature and various components. In the actual production process, because the temperature of molten steel can be measured continuouly, and the boundary condition of smelting process is very complex, the terminal control of smelting process is achieved difficultly. Because the precise control of carbon and temperature of steel is very difficult, it is very necessary to improve the terminal control efficiency of BOF. Meantime, because the process of BOF is a very complex physical and chemical processes with multi-component multi-phase and high temperature, whose mechanism about the input and output is nonlinear and not very clear , the conventional modeling methods can not always meet the production requirements.
In this paper, firstly, the principle and equipment and the overview and status about terminal control technology of BOF are introduced briefly. Secondly, the neural network and several predicting methods of terminal model of BOF by using the neural network are introduced briefly. Thirdly, the result that the Levenberg-Marquardt(LM) method has fast convergence and good learning performance is got by summarizing and comparing features between several BP algorithm. Finaly, the intelligent forecast and control model about BOF besed on LM neural network algorithm is proposed.
In this paper the researched method has a substantial theoretic base and has been validated by the theory and experiment. The actual data of continuous 60 batches from a converter steel mill are chosen as example, input parameter that influence and endpoint steel temperatures in converter are determined. In this paper establish three-layers BP prediction neural networks, endpoint various and steeltemperature content. The model of endpoint various and steel temperature prediction have established. On the basis of this, the method based on BP neural network is proposed so as to determine the blown oxygen and auxiliary raw materials. Simulationed showed the method is effective and can be used to practical process to produce steel.
引文
[1]李润生,李延辉,胡学军.神经网络在冶金中的应用.钢铁研究,2004(2):48-50
[2]张昆龙.模糊控制理论研究及在转炉炼钢过程控制中的应用[D].北京科技大学,2004.
[3]陶钧转炉炼钢过程智能控制方法及其应用的研究[D].东北大学,2001.
[4]孟劲松,讥勇.复吹转炉终点控制的实践和理论[J].鞍钢技术,1995(8) :7-10
[5]孙晓凌.转炉炼钢终点控制的模型研究[J].辽宁冶金,1990(1):30-32
[6]王茂华,惠志刚,施雄梁.转炉终点控制技术[J],鞍钢技术,No.3 2005 :6-10
[7]刘浏.中国转炉炼钢技术的进步[J] .钢铁,2005(2):1-5
[8]刘春荣.宝钢转炉复吹改造后的计算机控制[J].冶金自动化,1991(4) :9-11
[9]K.lwamura等。转炉吹炼控制新方法[J].国外钢铁,1997(4):27-30
[10]R Boom.B Shangen, W Raaijen, etal.Developments in Sublance Design and Measurement Techniques at Hoogovens.Ijmuiden Plant[J].I&SM, 1998 (15):25-28
[11] Lwamura.K,解四海译。一种控制氧气顶吹转炉吹炼的新方法[J].武钢技术,1997(2):22-25
[12]宋卫良.分析仪器的新进展及其在钢铁分析中的应用[J].武钢技术,1996(5):23-25
[13]张鉴,伶福生,成国光.终点碳控制的现状和前景[J].特殊钢,1995(4):18-21
[14]张旭升.质谱仪在转炉炼钢终点控制中的应用[J].鞍钢技术,2003(6):8-10
[15]Yun S Y, Chang K S. Dynamic Predication Using Neural Network for Automation of BOF Process in Steel Industry. I&SM,1996(8):37-42
[16]丁容,刘浏.转炉炼钢过程人工智能静态控制模型.钢铁,1997,32(1):22-26
[17]屠海,洪新.转炉炼钢终点锰磷动态控制技术的开发.上海金属,2002,24(2):27-30
[18]谢书明,陶钧,柴天佑.基于神经网络的转炉炼钢终点控制.控制理论与应用,2003,20(6) :903-907
[19]王中杰,柴天佑,邵诚.基于RBF神经网络的加热炉钢温预报模型.系统仿真学报,1998,11(3):181-184
[20]谢书明,高宪文,柴天佑.基于灰色模型的转炉炼钢终点预报研究.钢铁研究学报,1999, 11(4):9-12
[21]谢书明,柴天佑,陶钧.一种转炉炼钢动态终点预报的新方法.自动化学报,2001,27(1):136-139
[22]陶钧,谢书明,柴天佑.基于遗传算法和径向基函数神经网络的转炉炼钢模型.系统仿真学报,2000 , 12(3):241-244
[23]陶钧,柴天佑,李小平等.转炉炼钢智能控制方法及应用.控制理论与应用,2001, 18(增刊):129-133
[24]王永富,李小平,柴天佑,谢书明.转炉炼钢动态过程预设定模型的混合建模与预报.东北大学学报,2003,24(8):715-718
[25]韩力群.人下神经网络理论、设计及应用[M],北京化学下业出版社,2002
[26]康德山.人工神经网络简介.在南京炼油厂为计算机站科技人员的讲课稿,1993,6:1-22
[27]王洪元,史国栋编著.人工神经网络技术及其应用[M].第1版.中国石化出版,2002
[28]王旭.人工神经元网络原理与应用[M].第1版.东北大学出版社,2000:21-25
[29]易继楷,侯媛彬编著.智能控制技术.北京工业大学出版社,2003
[30]R.A.Jacobs.Neural Networks,Increased rates of convergence through learning rate adaptation.vol.l, no.4, pp.295-308,1988
[31] T. Tollenaere.SuperSAB:Fast adaptive Back Propagation with good scaling Properties,Neural Networks, vo1.3,no.5,pp.561-572,1990
[32] A.K.Rigler,J.M.Irvine and T. P. Vogl.Rescaling of variables in back propagation learning.Neural Networks, vo1.3,no.5, pp.561-573,1990
[33] E. Barnard.Optimization for training neural nets , IEEE Trans. on Neural Networks,1992,vo1.3, no.2 :232-240, 1992
[34] D. F. Shanno.Recent advances in numerical techniques for large-scale optimization.Neural Networks for Control, Miller, Sutton and Werbos, Eds. Cambridge MA:MIT press, 1990
[35] R.attiti.First-and second-order methods for learning:Between steepest descent and Newton’s method,Neural Computation,vo1.4,no.2,pp.141-166,1992.
[36] C.Charalambous. Conjugate gradient algorithm for efficient training of artificial neural networks.IEEE Proceedings,vol.139,no.3,pp.301-310,992
[37]远祯,罗波.BP网络的改进研究[J].信息技术,2006, 2 : 88-91
[38]胡耀该,李伟,胡继明一种改进激活函数的人工神经网络及其应用[J].武汉大学学报,2004,29 (10):9-16
[39] Haran M T, Menhaj M B. Training feedforward networks with the marquardt algorithm [J].IEEE Trans Neural Net,1994,5(6):989-993
[40]赵弘,周瑞祥,林廷沂.基于LM算法的神经网络监督控制.西安交通大学学报,2002,36(5 ):523-527
[30]Robert H ,Theory of the Back-Propagation Neural Network. Proc.UCNN Washington.C,1989(1):593-605
[41] Wei Wu, Guorui Feng, and Xin Li..Training multilayer perceptions via minimization of sum of ridge functions, Advances in Computational Mathematics. 2002,17 :331-347
[42] Wei Wu,Yuesheng Xu.Deterministic convergence of an online gradient method for neural networks. Advances in Computational Mathematics, 2002, 144(1-2):335-347
[43] Fernando M.Silva,Luis B.Almeida. Acceleration Techniques for the Back propagation Algorithm. Neural Networks. EUTSDII' Workshop 1990, Feb.15-17:110-119
[44] Ralf Salomon, J.Leovan Hermmen. Accelerating Back Propagation through Dynamic Self-Adaptation. Neural Networks, 1996, 9 (4):589-601
[45] D.R.Wilson,T.R. Martinez. The need for small learning rates on large problems.Proceedings of the 2001 International Joint Conference on Neural Networks, IJCNNOI,115-119
[46]T.Chen et al.Approximation capability in C by multilayer feed-forward networks. IEEE Trans.Neural Networks,1995, 6 (1):25-30
[47] Mackay D J C. Bayesian Interpolation. Neural Computation, 1992,4(3):415-447.
[48] Ito Y.Representation of functions by superposition of a step or sigmoidal function and their application to neural network theory.Neural Networks,1991(4):386-392
[49]Ito Y. Approximation of continuous function on Rd by linear combination of shifted rotations of a sigmoidalfunction with and without scaling. Neural Network, 1992(5):105-112
[50]R.A.Jacobs.Neural Networks,“Increased rates of convergence through learning rate adaptation".vol.l, no.4:295-308,1988
[51]闻新. MATLAB神经网络仿真与应用(M).科学出版社,2003
[52]张宏斌.训练多层网络的样本问题.自动化学报,1993, 19 (1):71-77
[53]李俊国.高炉铁水硅含量神经网络预测模型研究.河北理工学院学报,2002
[54]Rumelhart D E,hinton G E and Williams R G. Learning representations by back-propagation errors. Nature, 1986(323):533-536.
[55]Michele De Laurentiis, Peeter M.Ravdin. Survival analysis of censored data: Neural network analysis detection of complex interactions between variable, Breast Cancer Research and Treatment, 1994(32):113-118
[56]戴云阁,李文秀,龙腾春编.现代转炉炼钢.东北大学出版社,1998: 148-150
[2]张昆龙.模糊控制理论研究及在转炉炼钢过程控制中的应用[D].北京科技大学,2004.
[3]陶钧转炉炼钢过程智能控制方法及其应用的研究[D].东北大学,2001.
[4]孟劲松,讥勇.复吹转炉终点控制的实践和理论[J].鞍钢技术,1995(8) :7-10
[5]孙晓凌.转炉炼钢终点控制的模型研究[J].辽宁冶金,1990(1):30-32
[6]王茂华,惠志刚,施雄梁.转炉终点控制技术[J],鞍钢技术,No.3 2005 :6-10
[7]刘浏.中国转炉炼钢技术的进步[J] .钢铁,2005(2):1-5
[8]刘春荣.宝钢转炉复吹改造后的计算机控制[J].冶金自动化,1991(4) :9-11
[9]K.lwamura等。转炉吹炼控制新方法[J].国外钢铁,1997(4):27-30
[10]R Boom.B Shangen, W Raaijen, etal.Developments in Sublance Design and Measurement Techniques at Hoogovens.Ijmuiden Plant[J].I&SM, 1998 (15):25-28
[11] Lwamura.K,解四海译。一种控制氧气顶吹转炉吹炼的新方法[J].武钢技术,1997(2):22-25
[12]宋卫良.分析仪器的新进展及其在钢铁分析中的应用[J].武钢技术,1996(5):23-25
[13]张鉴,伶福生,成国光.终点碳控制的现状和前景[J].特殊钢,1995(4):18-21
[14]张旭升.质谱仪在转炉炼钢终点控制中的应用[J].鞍钢技术,2003(6):8-10
[15]Yun S Y, Chang K S. Dynamic Predication Using Neural Network for Automation of BOF Process in Steel Industry. I&SM,1996(8):37-42
[16]丁容,刘浏.转炉炼钢过程人工智能静态控制模型.钢铁,1997,32(1):22-26
[17]屠海,洪新.转炉炼钢终点锰磷动态控制技术的开发.上海金属,2002,24(2):27-30
[18]谢书明,陶钧,柴天佑.基于神经网络的转炉炼钢终点控制.控制理论与应用,2003,20(6) :903-907
[19]王中杰,柴天佑,邵诚.基于RBF神经网络的加热炉钢温预报模型.系统仿真学报,1998,11(3):181-184
[20]谢书明,高宪文,柴天佑.基于灰色模型的转炉炼钢终点预报研究.钢铁研究学报,1999, 11(4):9-12
[21]谢书明,柴天佑,陶钧.一种转炉炼钢动态终点预报的新方法.自动化学报,2001,27(1):136-139
[22]陶钧,谢书明,柴天佑.基于遗传算法和径向基函数神经网络的转炉炼钢模型.系统仿真学报,2000 , 12(3):241-244
[23]陶钧,柴天佑,李小平等.转炉炼钢智能控制方法及应用.控制理论与应用,2001, 18(增刊):129-133
[24]王永富,李小平,柴天佑,谢书明.转炉炼钢动态过程预设定模型的混合建模与预报.东北大学学报,2003,24(8):715-718
[25]韩力群.人下神经网络理论、设计及应用[M],北京化学下业出版社,2002
[26]康德山.人工神经网络简介.在南京炼油厂为计算机站科技人员的讲课稿,1993,6:1-22
[27]王洪元,史国栋编著.人工神经网络技术及其应用[M].第1版.中国石化出版,2002
[28]王旭.人工神经元网络原理与应用[M].第1版.东北大学出版社,2000:21-25
[29]易继楷,侯媛彬编著.智能控制技术.北京工业大学出版社,2003
[30]R.A.Jacobs.Neural Networks,Increased rates of convergence through learning rate adaptation.vol.l, no.4, pp.295-308,1988
[31] T. Tollenaere.SuperSAB:Fast adaptive Back Propagation with good scaling Properties,Neural Networks, vo1.3,no.5,pp.561-572,1990
[32] A.K.Rigler,J.M.Irvine and T. P. Vogl.Rescaling of variables in back propagation learning.Neural Networks, vo1.3,no.5, pp.561-573,1990
[33] E. Barnard.Optimization for training neural nets , IEEE Trans. on Neural Networks,1992,vo1.3, no.2 :232-240, 1992
[34] D. F. Shanno.Recent advances in numerical techniques for large-scale optimization.Neural Networks for Control, Miller, Sutton and Werbos, Eds. Cambridge MA:MIT press, 1990
[35] R.attiti.First-and second-order methods for learning:Between steepest descent and Newton’s method,Neural Computation,vo1.4,no.2,pp.141-166,1992.
[36] C.Charalambous. Conjugate gradient algorithm for efficient training of artificial neural networks.IEEE Proceedings,vol.139,no.3,pp.301-310,992
[37]远祯,罗波.BP网络的改进研究[J].信息技术,2006, 2 : 88-91
[38]胡耀该,李伟,胡继明一种改进激活函数的人工神经网络及其应用[J].武汉大学学报,2004,29 (10):9-16
[39] Haran M T, Menhaj M B. Training feedforward networks with the marquardt algorithm [J].IEEE Trans Neural Net,1994,5(6):989-993
[40]赵弘,周瑞祥,林廷沂.基于LM算法的神经网络监督控制.西安交通大学学报,2002,36(5 ):523-527
[30]Robert H ,Theory of the Back-Propagation Neural Network. Proc.UCNN Washington.C,1989(1):593-605
[41] Wei Wu, Guorui Feng, and Xin Li..Training multilayer perceptions via minimization of sum of ridge functions, Advances in Computational Mathematics. 2002,17 :331-347
[42] Wei Wu,Yuesheng Xu.Deterministic convergence of an online gradient method for neural networks. Advances in Computational Mathematics, 2002, 144(1-2):335-347
[43] Fernando M.Silva,Luis B.Almeida. Acceleration Techniques for the Back propagation Algorithm. Neural Networks. EUTSDII' Workshop 1990, Feb.15-17:110-119
[44] Ralf Salomon, J.Leovan Hermmen. Accelerating Back Propagation through Dynamic Self-Adaptation. Neural Networks, 1996, 9 (4):589-601
[45] D.R.Wilson,T.R. Martinez. The need for small learning rates on large problems.Proceedings of the 2001 International Joint Conference on Neural Networks, IJCNNOI,115-119
[46]T.Chen et al.Approximation capability in C by multilayer feed-forward networks. IEEE Trans.Neural Networks,1995, 6 (1):25-30
[47] Mackay D J C. Bayesian Interpolation. Neural Computation, 1992,4(3):415-447.
[48] Ito Y.Representation of functions by superposition of a step or sigmoidal function and their application to neural network theory.Neural Networks,1991(4):386-392
[49]Ito Y. Approximation of continuous function on Rd by linear combination of shifted rotations of a sigmoidalfunction with and without scaling. Neural Network, 1992(5):105-112
[50]R.A.Jacobs.Neural Networks,“Increased rates of convergence through learning rate adaptation".vol.l, no.4:295-308,1988
[51]闻新. MATLAB神经网络仿真与应用(M).科学出版社,2003
[52]张宏斌.训练多层网络的样本问题.自动化学报,1993, 19 (1):71-77
[53]李俊国.高炉铁水硅含量神经网络预测模型研究.河北理工学院学报,2002
[54]Rumelhart D E,hinton G E and Williams R G. Learning representations by back-propagation errors. Nature, 1986(323):533-536.
[55]Michele De Laurentiis, Peeter M.Ravdin. Survival analysis of censored data: Neural network analysis detection of complex interactions between variable, Breast Cancer Research and Treatment, 1994(32):113-118
[56]戴云阁,李文秀,龙腾春编.现代转炉炼钢.东北大学出版社,1998: 148-150
© 2004-2018 中国地质图书馆版权所有 京ICP备05064691号 京公网安备11010802017129号 地址：北京市海淀区学院路29号 邮编：100083 电话：办公室：(+86 10)66554848；文献借阅、咨询服务、科技查新：66554700 |