ANN-GA based optimization of a high ash coal-fired supercritical power plant

合集下载

基于GA—ANN的建筑工程质量评价模型研究

文章编号：０９６２（０６１．３７０１０．８５２０）００５ —２
基于ＧＡＡＮＮ的建筑工程质量评价模型研究 —
王伟江
摘
邢凤强孔秋生
要：利用人工神经网络强大的非线性函数逼近能力，实现了对建筑工程质量水平的评价。针对Ａｘ＃￣网络的不足，－
Ｏｎｆｏｔｄｍａｅｏｒｇｔｏｉｃｍｐｒｉｕｒｓａｇｆｉｉａｉｎｄｔｈｉｅｖｏｓｒ
ｌｉｇｐｏｅｔｎｓｒｖｎｉｄｃｒａｕｅｉｎｒｊｃａｄｉｅｅｔｎａｕｅｎｔｐｏｎｍｅｓｒ
人工神经网络（ｒｆｉｅｒｌｔｏｋ简称Ａ是２ＡｔｉａＮｕａＮｅｗｒｓｉｃｌ『Ｎ）０世网络进行改进。
纪８年代后期发展起来的一门新兴学科，０它具有较强的非线性、１３引入遗传算法的Ｂ．Ｐ神经网络大规模并行处理能力。许多学者致力于将其引入自己的专业研遗传算法是模拟生物界在自然环境中的遗传和进化过程而
在设计渠道断面和衬砌时采用合理的形式和尺寸，其具有参考文献：使
１谷党国．全断面现浇混凝土衬砌渠道施工技术［］山西建筑，Ｊ．削减适应或回避冻胀的能力，常用的有梯形。弧形坡脚梯形、弧［］２０，（１：３．３．０５３１）１０１１形、矩形和ｕ型等，还有浆砌石弧形渠底，混凝土板砌渠坡等综合
究领域。取得了令人瞩目的成果，具有广泛的应用前景。
形成的一种自适应全局概率搜索算法。作为一种全局优化算法，典型的ＢＰ神经网络是三层，输入层、即：隐含层和输出层，各由于其简单通用、鲁棒性强、应用范围广等特点，因此被广泛应用

韩国先进科技学院机械工程系Ikjin Lee,副教授简介

Ikjin Lee, Assistant Professor7109, N7-4, Mechanical Engineering DepartmentKorea Advanced Institute of Science and Technology (KAIST)291 Daehak-ro, Yuseong-gu, Daejeon 305-701, Republic of KoreaTel: +82-42-350-3041, Fax: +82-42-350-3210, Email: ikjin.lee@kaist.ac.kr_______________________________________________________________________EDUCATIONB.S. Mechanical Engineering Seoul National University, Korea 1993-2001 M.S. Mechanical Engineering Seoul National University, Korea 2001-2003 Ph.D. Mechanical Engineering University of Iowa 2003-2008RESEARCH AREAReliability-Based Design Optimization (RBDO)Reliability-Based Robust Design Optimization (RBRDO)System Reliability Analysis and Design OptimizationDesign under Uncertainties with Lack of InformationDesign under Uncertainties with Correlated Input VariablesSampling-Based RBDO with Parallel ComputingSurrogate Model Generation (Meta-modeling)PROFESSIONAL EXPERIENCESUniversity of Iowa Research Assistant 2003~2008 University of Iowa Teaching Assistant 2006~2007 University of Iowa Postdoctoral Research Scholar 2008.8~2011.7 University of Iowa Adjunct Professor 2010.7~2011.7 University of Connecticut Assistant Professor 2011.8~2013.7 KAIST Assistant Professor 2013.8~ TEACHING EXPERIENCESME3224 Analysis and Design of Mechanisms Fall 2011 ME3227 Design of Machine Elements Spring 2012 ME5895/ME3295 Probabilistic Engineering Design Fall 2012, Spring 2014 ME5511/ME3295 Principles of Optimum Design Spring 2013 MAE340 Engineering Design Fall 2013 MAE370 Understanding of Materials & Manufacturing Spring 2014 MAE475 Applied Mathematics Fall 2014 MAE 400 Capstone Design I Spring 2015 CD401 Multidisciplinary Capstone Design I Spring 2015 AWARDS1.ISSMO/Springer Prize for a young scientist, International Society of Structural andMultidisciplinary Optimization (ISSMO), 2009.2.Cited in Marquis Who’s Who in America, 64th Edition, 2010.PROJECT ACTIVITYFINSIHEDU.S. Army Tank-Automotive Command (TACOM)Caterpillar 994F Axle Pad ProjectU.S. Army Automotive Research Center (ARC) on FMTVWind Turbine Optimization with Clippers supported by Iowa Wind Project (IAWIND) Dynamic Analysis Software Development & Design Optimization by Stanley Black & DeckerIN PROGRESSLaunching Plug-in Digital Analysis Framework for Modular System DesignDevelopment of sensor-based virtual plant engineering technology for the support of plant O&MPROFESSIONAL SERVICESJournal Editor1.Associate Editor, Trans. Korean Soc. Mech. Eng. A, 2014~PresentPaper Reviewer (Reviewed more than 85 Journal Papers and 33 Conference Papers)1.Structural and Multidisciplinary Optimization (SMO)2.ASME Journal of Mechanical Design (JMD)3.Journal of Soils and Sediments (JSSS)4.Mechanism and Machine Theory (MECHMT)5.Journal of Optimization Theory and Applications (JOTA)6.International Journal of Vehicle Design (IJVD)7.Entropy8.Mechanics Based Design of Structures and Machines9.Probabilistic Engineering Mechanics (PREM)puters and Industrial Engineering (CAIE)11.Engineering Optimization (GENO)puter Methods in Applied Mechanics and Engineering (CMAME)13.Applied Mathematical Modeling (APM)14.Mechanical Systems and Signal Processing (MSSP)15.Ain Shams Engineering Journal (ASEJ)16.Journal of Mechanical Science and Technology (JMST)17.ASME (IDETC/CIE) Conference18.AIAA/MAO ConferencePaper Review Coordinator1.Paper review coordinator for the Design Automation Conference of ASMEInternational Design Engineering Technical Conferences (IDETC), 2011~2015.Program Committee1.Local organizing committee for 10th World Congress of Structural MultidisciplinaryOptimization (WCSMO), 2013.2.International scientific and organizing committee for 5th International Conference onComputational Methods (ICCM), 2014.3.Scientific committee for 8th China-Japan-Korea Joint Symposium on Optimization ofStructural and Mechanical Systems (CJK-OSM8), 2014.4.Local organizing committee for 12th World Congress on Computational Mechanics(WCCM), 2016Other Professional Activities1.Chaired sessions at the 38th Design Automation Conference of 2012 ASMEInternational Design Engineering Technical Conferences (IDETC), August 2012.2.Chaired sessions at the World Congress of Structural Multidisciplinary Optimization(WCSMO) 10, May 2013.3.Chaired sessions at the 39th Design Automation Conference of 2013 ASMEInternational Design Engineering Technical Conferences (IDETC), August 2013.4.Chaired sessions at the 40th Design Automation Conference of 2014 ASMEInternational Design Engineering Technical Conferences (IDETC), August 2014.MembershipAmerican Society of Mechanical Engineers (ASME)American Institute of Aeronautics and Astronautics (AIAA)International Society for Structural and Multidisciplinary Optimization (ISSMO)Korea Society of Mechanical Engineers (KSME)Korea Society of Computational Mechanics (KSCM)Korea Society of Design Optimization (KSDO)Korean Society of Precision Engineering (KSPE)Proposal Review Panel1.Romanian Executive Agency for Higher Education, Research, Development andInnovation Funding (UEFISCDI) Proposal Review Panel (2011,2013,2015)2.Kazakhstan National Center for Science and Technology Evaluation (NCSTE)Proposal Review Panel (2014)CONFERENCE SEMINAR PRESENTATIONS1.Presented a seminar “Alternative Methods for Reliability-Based Robust DesignOptimization Including Dimension Reduction Method,” at ARC conference, Ann Arbor, MI, May 24, 2006.2.Presented a seminar “Alternative Methods for Reliability-Based Robust DesignOptimization Including Dimension Reduction Method,” at 2006 ASME IDETC, Philadelphia, Pennsylvania, September 10-13, 2006.3.Presented a seminar “RBDO Using MPP-Based Dimension Reduction Method(DRM) for Multidimensional Highly Nonlinear Systems,” at ARC conference, Ann Arbor, MI, May 16, 2007.4.Presented a seminar “RBDO Using MPP-Based Dimension Reduction Method(DRM) for Multidimensional Highly Nonlinear Systems” at WCSMO7 conference, Seoul, Korea, May 22, 2007.5.Presented a seminar “A New Inverse Reliability Analysis Method Using MPP-BasedDimension Reduction Method (DRM),” at 2007 ASME IDETC, Las Vegas, Nevada, September 4-7, 2007.6.Presented a seminar “System Reliability-Based Design Optimization Using MPP-BasedDimension Reduction Method,” at ARC conference, Ann Arbor, MI, May 21, 2008.7.Presented a seminar “Sensitivity Analyses of FORM-Based and DRM-BasedPerformance Measure Approach for Reliability-Based Design Optimization,” at 2008 ASME IDETC, New York City, New York, August 3-6, 2008.8.Presented a seminar “Comparison Study between Probabilistic and PossibilisticApproach for Problems with Correlated Input and Lack of Input Statistical Information” at ARC conference, Ann Arbor, MI, May 13, 2009.9.Presented a seminar “Comparison Study between Probabilistic and PossibilisticApproach for Problems with Correlated Input and Lack of Input Statistical Information” at WCSMO8 conference, Lisbon, Portugal, June 1-5, 2009.10.Presented an award speech “RBDO Using MPP-Based Dimension Reduction Method(DRM) for Multidimensional Highly Nonlinear Systems” at WCSMO8 conference, Lisbon, Portugal, June 1-5, 2009.11.Presented a seminar “Comparison Study between Probabilistic and PossibilisticApproach for Problems with Correlated Input and Lack of Input Statistical Information” at 2009 ASME IDETC, San Diego, California, August 31-September 2, 2009.12.Presented a seminar “Sampling-Based Stochastic Sensitivity Analysis Using Scorefunctions for RBDO problems with Correlated Random Variables” at ARC conference, Ann Arbor, MI, May 11, 2010.13.Presented a seminar “Sampling-Based Stochastic Sensitivity Analysis Using Scorefunctions for RBDO problems with Correlated Random Variables” at 2010 ASME IDETC, Montreal, Canada, August 16, 2010.14.Presented a seminar “Equivalent Standard Deviation to Convert High-ReliabilityModel to Low-Reliability Model for Efficiency of Sampling-Based RBDO” at ARC conference, Ann Arbor, MI, May 24, 2011.15.Presented a seminar “Equivalent Standard Deviation to Convert High-ReliabilityModel to Low-Reliability Model for Efficiency of Sampling-Based RBDO” at 2011 ASME IDETC, Washington, D.C., August 28-31, 2011.16.Presented a seminar “A Novel Second-Order Reliability Method (SORM) Using Non-Central or Generalized Chi-Squared Distributions” at 2012 ASME IDETC, Chicago, Illinois, August 13-15, 2012.17.Presented a seminar“Probabilistic Sensitivity Analysis for Novel Second-OrderReliability Method (SORM) Using Generalized Chi-squared Distribution” at WCSMO10 conference, Orlando, FL, May 19-24, 2013.18.Presented a seminar “Sampling-Based Design Optimization in the Presence ofInterval Variables” at APCOM&ISCM 2013, Singapore, December 12, 2013, Keynote Speech.19.Presented a seminar “Reliability-Based Vehicle Safety Assessment and DesignOptimization of Roadway Radius and Speed Limit in Windy Environments” at KSME conference, Jeongseon, Korea, December 19, 2013.20.Presented a seminar “Inverse Reliability Analysis for Approximated Second-OrderReliability Method Using Hessian Update” at 2014 ASME IDETC, Buffalo, New York, August 17-20, 2014.21.Presented a seminar “Enhanced Second-Order Reliability Method and StochasticSensitivity Analysis Using Importance Sampling” at WCSMO11 conference, Sydney, Australia, June 7-12, 2015.INVITED SEMINAR PRESENTATIONS1.Presented a seminar “Reliability-based Design Optimization: The Past, Present, andFuture,” at the University of Iowa, October 1, 2009.2.Provided a lecture on “Sampling-based RBDO using the Dynamic Kriging andStochastic Sensitivity Analysis” to John Deere, August, 2010.3.Presented a seminar “Sampling-Based RBDO Using the Dynamic Kriging (D-Kriging)Method and Stochastic Sensitivity Analysis” at ARC seminar, the University of Michigan, Ann Arbor, MI, October 29, 2010.4.Provided a seminar “Recent Improvements on Reliability-Based Design Optimization(RBDO) Methodology,” at the University of Connecticut, March 2, 2011.5.Provided a seminar “Recent Improvements on Reliability-Based Design Optimization(RBDO) Methodology,” at the Korea Advanced Institute of Science and Technology (KAIST), April 8, 2011.6.Provided a workshop on “Sampling-Based RBDO Using Dynamic Kriging Methodand Stochastic Sensitivity Analysis” to Army TARDEC members, Warren, MI, April 19, 2011.7.Presented a seminar “Reliability-Based Design Optimization,” at Hanyang University,August 20, 2013.8.Presented a seminar “Application of RBDO to Vehicle Design,” at Hyundai Motors,October 25, 2013.9.Presented a seminar “Application of RBDO to Vehicle Design,” at Doosan Infracore,November 22, 2013.10.Presented a seminar “Reliability Assessment and its Application to Shipbuilding andOcean Plant Design,” at Samsung Heavy Industry, June 20, 2014.11.Presented a seminar “Simulation-Based Design under Uncertainties: Theory &Application,” at Harbin Institute of Technology, January 19, 2015.12.Presented an invited lecture “Simulation-based Design Under Uncertainties: Theory& Application”, 2nd Annual Conference of Korea Society for Design Optimization, 2015.13.Presented a seminar “Simulation-based Design Under Uncertainties: Theory &Application”, at Korea Maritime University, 2015.14.Will present a seminar at Dalian University of Technology, July, 2015.15.Will present a seminar at NYU POLY, August, 2015.PUBLICATIONSBooks1.Lee, I.,Dimension Reduction Method for Design under Uncertainty: Applications ofDimension Reduction Method to Reliability-Based Design Optimization and Robust Design Optimization, LAP LAMBERT Academic Publishing, 2010.Ph. D. Thesis1.“Reliability-Based Design Optimization and Robust Design Optimization UsingUnivariate Dimension Reduction Method,” University of Iowa, 2008.Papers in Technical Journals (International)1.Lee, I., Choi, K.K., Du, L., and Gorsich, D., “Dimension Reduction Method forReliability-Based Robust Design Optimization,” Special Issue of Computers & Structures: Structural and Multidisciplinary Optimization, Vol. 86, pp. 1550–1562, 2008. (IF: 2.134)2.Lee, I., Choi, K.K., Du, L., and Gorsich, D., “Inverse Analysis Method Using MPP-Based Dimension Reduction for Reliability-Based Design Optimization of Nonlinear and Multi-Dimensional Systems,” Special Issue of Computer Methods in Applied Mechanics and Engineering: Computational Methods in Optimization Considering Uncertainties, Vol. 198, No. 1, pp. 14-27, 2008. (IF: 2.959)3.Noh, Y., Choi, K.K., and Lee, I., “Reduction of Ordering Effect in RBDO UsingDimension Reduction Method,” AIAA Journal, Vol. 47, No. 4, pp. 994-1004, 2009.(IF: 1.207)4.Lee, I., Choi, K.K., and Gorsich, D., “Sensitivity Analyses of FORM-Based andDRM-Based Performance Measure Approach (PMA) for Reliability-Based Design Optimization (RBDO),” International Journal for Numerical Methods in Engineering, Vol. 82, No.1, pp. 26-46, 2010. (IF: 2.055)5.Lee, I., Choi, K.K., and Gorsich, D., “System Reliability-Based Design OptimizationUsing the MPP-Based Dimension Reduction Method,” Journal of Structural and Multidisciplinary Optimization, Vol. 41, No. 6, pp. 823-839, 2010. (IF: 1.974)6.Noh, Y., Choi, K.K., and Lee, I., “Identification of Marginal and Joint CDFs UsingBayesian Method for RBDO,” Journal of Structural and Multidisciplinary Optimization, Vol. 40, No. 1, pp. 35-51, 2010.(IF: 1.974)7.Noh, Y., Choi, K.K., and Lee, I., “Comparison Study between MCMC-based andWeight-based Bayesian Methods for Identifications of Joint Distribution,” Journal of Structural and Multidisciplinary Optimization, Vol. 42, No. 6, pp. 823-833, 2010.(IF: 1.974)8.Lee, I., Choi, K.K., Noh, Y. Zhao, L., and Gorsich D., “Sampling-Based StochasticSensitivity Analysis Using Score Functions for RBDO Problems with CorrelatedRandom Variables,” Journal of Mechanical Design, Vol. 133, No. 2, 21003, 2011.(IF: 1.250)9.Noh, Y., Choi, K.K., and Lee, I., “Reliability-Based Design Optimization withConfidence Level under Input Model Uncertainty Due to Limited Test Data,” Journal of Structural and Multidisciplinary Optimization, Vol. 43, No. 4, pp. 443-458, 2011.(IF: 1.974)10.Zhao, L., Choi, K.K., and Lee, I., “Metamodeling Method Using Dynamic Krigingfor Design Optimization,” AIAA Journal, Vol. 49, No. 9, pp. 2034-2046, 2011. (IF:1.207)11.Noh, Y., Choi, K.K., and Lee, I., “Reliability-based Design Optimization withConfidence Level for Non-Gaussian Distributions Using Bootstrap Method,” Journal of Mechanical Design, Vol. 133, No. 9, 91001, 2011. (IF: 1.250)12.Lee, I., Choi, K.K., and Zhao, L., “Sampling-Based RBDO Using the StochasticSensitivity Analysis and Dynamic Kriging Method,” Journal of Structural and Multidisciplinary Optimization, Vol. 44, No. 3, pp. 299-317, 2011. (IF: 1.974)13.Lee, I., Noh, Y., and Yoo, D., “A Novel Second-Order Reliability Method (SORM)Using Non-Central or Generalized Chi-Squared Distributions,” Special Issue of Journal of Mechanical Design on Design under Uncertainty, Vol. 134, No. 10, 100912, 2012. (IF: 1.250)14.Lee, I., Choi, K.K., Noh, Y., and Lamb, D., “Comparison Study betweenProbabilistic and Possibilistic Methods for Problems under a Lack of Correlated Input Statistical Information,” Journal of Structural and Multidisciplinary Optimization, Vol. 47, No. 2, pp. 175-189, 2013. (IF: 1.974)15.Song, H., Choi, K.K., Lee, I., Zhao, L., and Gorsich, D., “Adaptive Virtual SupportVector Machine for Reliability Analysis of High-Dimensional Problems,” Journal of Structural and Multidisciplinary Optimization,Vol. 47, No. 4, pp. 479-491, 2013.(IF: 1.974)16.Lee, I., Choi, K.K., and Shin, J., “Equivalent Target Probability of Failure to ConvertHigh-reliability Model to Low-reliability Model for Efficiency of Sampling-based RBDO,” Journal of Structural and Multidisciplinary Optimization, Vol. 48, No. 2, pp.235-248, 2013. (IF: 1.974)17.Zhao, L., Choi, K.K., Lee, I., and Gorsich, D., “Conservative Surrogate Model usingWeighted Kriging Variance for Sampling-based RBDO,”Journal of Mechanical Design, Vol. 135, No. 9, 091003, 2013. (IF: 1.250)18.Yoo, D., and Lee, I., “Sampling-based Approach for Design Optimization in thePresence of Interval Variables,” Journal of Structural and Multidisciplinary Optimization, Vol. 49, No. 2, pp. 253-266, 2014. (IF: 1.974)19.Shin, J., and Lee, I., “Reliability-Based Vehicle Safety Assessment and DesignOptimization of Roadway Radius and Speed Limit in Windy Environments,” Journal of Mechanical Design, Vol. 136. No. 8, 081006, 2014. (IF: 1.250)20.Yoo, D., Lee, I., and Cho, H., “Probabilistic Sensitivity Analysis for Novel Second-Order Reliability Method using Generalized Chi-Squared Distribution,” Journal of Structural and Multidisciplinary Optimization,Vol. 50, No. 5, pp. 787-797, 2014.(IF: 1.974)21.Lim, J., Lee, B., and Lee, I., “SORM-based Inverse Reliability Analysis UsingHessian Update for Accurate and Efficient Reliability-based Design Optimization,”International Journal for Numerical Methods in Engineering, Vol. 100, No. 10, pp.773-792, 2014. (IF: 2.055)22.Shin, J., and Lee, I., “Reliability Analysis and Reliability-Based Design Optimizationof Roadway Horizontal Curves Using a First-Order Reliability Method (FORM),”Engineering Optimization, Vol. 47, No. 5, pp. 622-641, 2015. (IF: 1.076)23.Lim, J., Lee, B., and Lee, I., “Sequential Optimization and Reliability Assessmentbased on Dimension Reduction Method for Accurate and Efficient Reliability-based Design Optimization,” Journal of Mechanical Science and Technology, Vol. 29, No.4, pp. 1349-1354, 2015. (IF: 0.838)24.Cho, H., Choi, K.K., and Lee, I., “Design Sensitivity Method for Sampling-BasedRBDO with Fixed COV,” submitted to Journal of Mechanical Design, 2015.Technical Notes1.Zhao, L., Choi, K.K., and Lee, I., “Reply by the Authors to the Comment by H. Liangand M. Zhu,” AIAA Journal, Vol. 51, No. 12, pp. 2989-2990, 2013. (IF: 1.207)International Conference Proceedings1.Choi, K.K., Lee, I., and Gorsich, D., “Dimension Reduction Method for Reliability-Based Robust Design Optimization,” III European Conference on Computational Mechanics, Lisbon, Portugal, June 5-8, 2006.2.Lee, I., Choi, K.K., and Du, L., “Alternative Methods for Reliability-Based RobustDesign Optimization Including Dimension Reduction Method,” 32nd ASME Design Automation Conference, Philadelphia, Pennsylvania, September 10-13, 2006.3.Lee, I., Choi, K.K., and Du, L., “Dimension Reduction Method (DRM) Based RBDOfor Highly Nonlinear Systems,” WCSMO7, COEX Seoul, Korea, May 21-25, 2007, Received the ISSMO-Springer Prize.4.Choi, K.K., Du, L., Lee, I., and Gorsich, D., “A New Robust Concept in PossibilityTheory for Possibility-Based Robust Design Optimization,” WCSMO7, COEX Seoul, Korea, May 21-25, 2007.5.Lee, I., Choi, K.K., Du, L., and Gorsich, D., “A New Inverse Reliability AnalysisMethod Using MPP-Based Dimension Reduction Method (DRM),” 33rd ASME Design Automation Conference, Las Vegas, Nevada, September 4-7, 2007.6.Du, L., Choi, K.K., and Lee, I., “Robust Design Concept in Possibility Theory AndOptimization For System With Both Random And Fuzzy Input Variables,” the 2007 ASME International Design Engineering Technical Conferences (IDETC), Las Vegas, Nevada, September 4-7, 2007.7.Lee, I., Choi, K.K., Du, L., and Gorsich, D., “Sensitivity Analyses of FORM-Basedand DRM-Based Performance Measure Approach for Reliability-Based Design Optimization,” 34th ASME Design Automation Conference, New York City, New York, August 3-6, 2008.8.Lee, I., Choi, K.K., Du, L., and Gorsich, D., “System Reliability-Based DesignOptimization Using MPP-Based Dimension Reduction Method,” 12th AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference, Victoria, British Columbia, September 10-12, 2008.9.Noh, Y., Choi, K.K., and Lee, I., “MPP-Based Dimension Reduction Method forRBDO Problems with Correlated Input Variables,” 12th AIAA/ISSMOMultidisciplinary Analysis and Optimization Conference, Victoria, British Columbia, September 10-12, 2008.mb, D., Gorsich, D., Choi, K.K., Noh, Y., and Lee, I., “The Use of Copulas andMPP-based Dimension Reduction Method (DRM) to Assess and Mitigate Engineering Risk in the Army Ground Vehicle,” 26th Army Science Conference, Orlando, Florida, December, 1-4, 2008.11.Lee, I., Choi, K.K., and Noh, Y., “Comparison Study between Probabilistic andPossibilistic Approach for Problems with Correlated Input and Lack of Input Statistical Information,” WCSMO8, Lisbon, Portugal, June 1-5, 2009.12.Noh, Y., Choi, K.K., Lee, I., and Gorsich, D., “Reliability-Based DesignOptimization with Confidence Level using Copula under Input Model Uncertainty,”WCSMO8, Lisbon, Portugal, June 1-5, 2009.13.Zhao, L., Choi, K.K., Lee, I., and Gorsich, D., “Sequential Sampling-Based KrigingMethod with Dynamic Basis Selection,” WCSMO8, Lisbon, Portugal, June 1-5, 2009.14.Lee, I., Choi, K.K., and Noh, Y., “Comparison Study between Probabilistic andPossibilistic Approach for Problems with Correlated Input and Lack of Input Statistical Information,” 35th ASME Design Automation Conference, San Diego, California, August 31-September 2, 2009.15.Noh, Y., Choi, K.K., Lee, I., Gorsich, D., and Lamb, D., “Reliability-Based DesignOptimization with Confidence Level using Copula under Input Model Uncertainty,”35th ASME Design Automation Conference, San Diego, California, August 31-September 2, 2009.16.Zhao, L., Choi, K.K., Lee, I., and Du, L., “Response Surface Method usingSequential Sampling for Reliability-Based Design Optimization,” 35th ASME Design Automation Conference, San Diego, California, August 31-September 2, 2009.17.Noh, Y., Choi, K.K., Lee, I., Gorsich, D., and Lamb, D., “Reliability-Based DesignOptimization with Confidence Level for Non-Gaussian Distributions Using Bootstrap Method,” 6th China-Japan-Korea Joint Symposium on Optimization of Structural and Mechanical Systems, Kyoto, Japan, June 22-25, 2010.18.Lee, I., Choi, K.K., Noh, Y., Zhao, L., and Gorsich, D., “Sampling-Based StochasticSensitivity Analysis Using Score Functions for RBDO Problems with Correlated Random Variables,” 36th ASME Design Automation Conference, Montreal, Canada, August 16-18, 2010.19.Noh, Y., Choi, K.K., Lee, I., and Gorsich, D., “Reliability-Based DesignOptimization with Confidence Level for Non-Gaussian Distributions Using Bootstrap Method,” 36th ASME Design Automation Conference, Montreal, Canada, August 16-18, 2010.20.Lee, I., Choi, K.K., and Zhao, L., “Sampling-Based RBDO Using the DynamicKriging (D-Kriging) Method and Stochastic Sensitivity Analysis,” 13th AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference, Fort Worth, Texas, September 13-15, 2010.21.Zhao, L., Choi, K.K., Lee, I., and Gorsich, D., “A Metamodeling Method UsingDynamic Kriging and Sequential Sampling,”13th AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference, Fort Worth, Texas, September 13-15, 2010. 22.Zhao, L., Choi, K.K., Lee, I., and Gorsich, D., “Conservative Surrogate Model usingWeighted Krigin Variance for Sampling-Based RBDO,” WCSMO9, Shizuoka, Japan,June 13-17, 2011.23.Choi, K.K., Lee, I., Zhao, L., Noh, Y., Lamb, D., and Gorsich, D., “Sampling-BasedRBDO Using Stochastic Sensitivity and Dynamic Kriging for Broader Army Applications,” NDIA Ground Vehicle Systems Engineering And Technology Symposium, Dearborn, Michigan, August 9-11, 2011.24.Lee, I., Choi, K.K., and Gorsich, D., “Equivalent Standard Deviation to ConvertHigh-reliability Model to Low-reliability Model for Efficiency of Sampling-based RBDO,” 37th ASME Design Automation Conference, Washington, D.C., August 28-31, 2011.25.Song, H., Choi, K.K., Lee, I., Zhao, L., and Lamb, D., “Adaptive Virtual SupportVector Machine for the Reliability Analysis of High-Dimensional Problems,”37th ASME Design Automation Conference, Washington, D.C., August 28-31, 2011.26.Lee, I., Noh, Y., and Yoo, D., “A Novel Second-Order Reliability Method (SORM)Using Non-Central or Generalized Chi-Squared Distributions,” 38th ASME Design Automation Conference, Chicago, Illinois, August 13-15, 2012.27.Cho, H., Choi, K.K., Lee, I., and Gorsich, D., “Confidence Level Estimation andDesign Sensitivity Analysis for Confidence-Based RBDO,” 38th ASME Design Automation Conference,Chicago, Illinois, August 13-15, 2012.28.Song, H., Choi, K.K., Lee, I., Zhao, L., and Gorsich, D., “Sampling-based RBDOUsing Stochastic Sensitivity-based Analysis and Virtual Support,” 38th ASME Design Automation Conference, Chicago, Illinois, August 13-15, 2012.29.Yoo, D., Lee, I., and Cho, H., “Probabilistic Sensitivity Analysis for Novel Second-Order Reliability Method using Generalized Chi-Squared Distribution,” WCSMO10, Orlando, Florida, May 19-24, 2013.30.Shin, J., and Lee, I., “Reliability-Based Design Optimization of Highway HorizontalCurves Based on first-Order Reliability Method,” WCSMO10, Orlando, Florida, May 19-24, 2013.31.Yoo, D., and Lee, I., “Sampling-Based Approach for Design Optimization in thePresence of Interval Variables,” WCSMO10, Orlando, Florida, May 19-24, 2013.32.Shin, J., and Lee, I., “First-Order Reliability Analysis of Vehicle Safety in HighwayHorizontal Curves,” 39th ASME Design Automation Conference, Portland, Oregon, August 4-7, 2013.33.Yoo, D., and Lee, I., “Sampling-based Approach for Design Optimization in thePresence of Interval Variables,” 39th ASME Design Automation Conference, Portland, Oregon, August 4-7, 2013.34.Yoon, G., Hur, J., Lee, I., and Youn, S., “Efficiency Improvement Approach ofSurrogate-model Based BLISS,” 8th China-Japan-Korea Joint Symposium on Optimization of Structural and Mechanical Systems (CJK-OSM8), Gyeongju, Korea, May 25-29, 2014.35.Lim, J., Lee, I., and Lee, B., “Sequential Optimization and Reliability AssessmentBased on Dimension Reduction Method for Accurate and Efficient Reliability-based Design Optimization,” 8th China-Japan-Korea Joint Symposium on Optimization of Structural and Mechanical Systems (CJK-OSM8),Gyeongju, Korea, May 25-29, 2014.36.Lim, J., Lee, I., and Lee, B., “Inverse Reliability Analysis for Approximated Second-Order Reliability Method Using Hessian Update” 40th ASME Design AutomationConference, Buffalo, New York,August 17-20, 2014.37.Piao, M.J., Park, C.H., Huh, H., and Lee, I., “Validation of Dynamic HardeningModels with Taylor Impact Tests at High Strain Rates,” 12th Asia-Pacific Conference on Engineering Plasticity and Its Application (AEPA), Taiwan, September 1-5, 2014.38.Lim, J., Lee, B., and Lee, I., “Enhanced second-order reliability method andstochastic sensitivity analysis using importance sampling” 11th World Congress of Structural and Multidisciplinary Optimization (WCSMO), Sydney, Australia, June 7-12, 2015.39.Kang, S.,Lee, I., and Lim, J., “Accuracy Improvement of MPP-Based DimensionReduction Method Using the Eigenvectors of the Hessian Matrix” 11th World Congress of Structural and Multidisciplinary Optimization, Sydney (WCSMO), Australia, June 7-12, 2015.Domestic Conference Proceedings1.Song, J., and Lee, I., “Accurate and Efficient Dimension Reduction Method UsingEigenvectors of the Hessian Matrix,” Proceedings of the KSME Fall Annual Meeting, Gwangju, Korea, November 11-14, 2014.2.Kang, K., and Lee, I., “A study on Efficiency Improvement of Kriging using cross-validationand process variance,” Proceedings of the KSME Fall Annual Meeting, Gwangju, Korea, November 11-14, 2014.3.Kang, S., and Lee, I., “Efficient reliability-based design optimization using approximation ofmost probable point,” Proceedings of the KSME Fall Annual Meeting, Gwangju, Korea, November 11-14, 20144.Kang, S., and Lee, I., “Accuracy Improvement of MPP-Based Dimension Reduction MethodUsing the Eigenvectors of Hessian Matrix” Proceedings of the KSME Spring Annual Meeting, Jeju, Korea, April 15-18, 2015.5.Song, J., Lee, I., and Lee, B., “Accurate and Efficient Dimension Reduction Method UsingEigenvectors of the Hessian Matrix” Proceedings of the KSME Spring Annual Meeting, Jeju, Korea, April 15-18, 2015.6.Kang, K., and Lee, I., “Basis Screening Kriging : Efficient and accurate surrogate modeling”Proceedings of the KSME Spring Annual Meeting, Jeju, Korea, April 15-18, 2015.7.Lee, I., and Kang, K., “Design Optimization of Subframe of Chassis Using SurrogateModeling” Proceedings of the KSME Spring Annual Meeting, Jeju, Korea, April 15-18, 2015.11。

GA-BP神经网络在罗源湾口波浪模拟研究中的应用

收稿日期：０ｌ６１２１－ —４０
权值和隐层、出层节点的阈值为（一１１之间的输，）
本研究将遗传算法优化的ＢＰ神经网络技术应
用到罗源湾口的波浪模拟中，并分析输入因子和训练样本数对网络性能的影响，讨观测的风速、探风向、气温差和有效波高（ｉｉｃｎｗｖｅｈ，海ｓｎｆａｔａｅｈｉｔｇｉｇＨ）间的关系．ｓ之
第３卷１
第２期
台
湾
海
峡
Ｖ０．１３１．Ｎｏ２．Ｍａｙ，２０１２
２１０２年５月
ＪＯＵＲＮＡＬＯＦＣＯＥＡＮＯＧＲＨＹＮＡＩＡＮＴＡＩＡＰＩＴＷＳＲＴ
ＧＡＢ — Ｐ神经网络在罗源湾口波浪模拟研究中的应用
海洋环境保护具有重要意义 ¨ ．Ｊ
神经网络方法近几年发展较快，明显的优点它
１Ｇ —Ｐ神经网络原理ＡＢ
１１Ｂ．Ｐ神经网络
是具有较好的自适应学习和非线性映射能力，适较合于处理物理机制复杂、因果关系和推理规则难以确定的非线性问题ｊ这些特点都为其在海浪模拟，和预报中的运用提供可能，内外很多学者将其付国诸实践．ｎａ等（０６采用动态神经网络预报海Ｍａｄｌ２０）浪的有效波高；ｓｉ（０２将两个站点的观测Ｔａ等２０）波高作为输入因子预测附近另一个站点的波高＿；４齐义泉等（０５利用神经网络技术改进传统数值２０）

乙烯装置裂解气压缩机性能预测模型研究

文章编号：1006-3080(2021)02-0177-06DOI: 10.14135/ki.1006-3080.20191226002乙烯装置裂解气压缩机性能预测模型研究马芳芳1，熊达1，孙铁栋2，欧阳福生1（1. 华东理工大学石油加工研究所，上海 200237；2. 石化盈科信息技术有限责任公司，北京 100007）摘要：以乙烯装置裂解气压缩机的设计信息为基础，通过修正设计数据建立了压缩机性能模型；基于学习速率自适应误差变化思想并结合遗传算法（GA ）的全局寻优特性，提出了一种改进BP 算法LR-GA-BP 进行压缩机性能预测；用模型对某乙烯装置四级压缩系统进行了模拟计算，压缩机第四段出口气体主要组分的预测值与实测值的相对误差均小于2%，压缩机各段出口温度和压力的预测值与实测值的相对误差均小于1%，说明裂解气压缩机性能预测模型是可靠的。

根据热力学原理，将四级压缩过程等价为绝热压缩，分析了压缩机第四段出口温度较高的影响因素，提出了降温措施，并进行模拟验证。

结果表明，适当增加压缩机段间返回流量可以降低压缩机第四段出口温度。

本文结果对于减缓压缩系统结焦、优化压缩机操作具有重要参考价值。

关键词：裂解气压缩机；性能预测模型；模拟计算；LR-GA-BP 算法中图分类号：TQ 052文献标志码：A作为乙烯装置的关键设备，裂解气压缩机多采用多级离心式压缩机。

在压缩机运转过程中，其性能信息的完备性至关重要。

通常压缩机在设计工况下的性能信息由制造商提供，但根据实际生产需要，压缩机常在偏离设计工况的条件下运行。

由于实际工况下的工艺参数不可避免会发生变化，所以需要经常对压缩机进行频繁的手动调节，人工干预程度较高，且存在一定的安全风险。

为保证压缩机在变工况条件下安全、平稳运行，研究其变工况性能十分必要。

为获取压缩机在各类工况下的性能信息，Sieros 等[1]提出了压缩机和涡轮机性能图的解析表示法，该方法的有效性已在发动机试验中得到验证，但为达到高的精度要求，该方法通常需要引入大量的过程参数，导致模型较为复杂；褚菲等[2]利用热力学定理和压缩机能量损失机理建立了机理模型，并通过BP （Back Propagation ）神经网络修正机理模型，虽然该模型的精度较高，但由于离心机操作条件复杂，导致机理模型中的一些重要参数如冲击损失系数等难以准确获取。

改进小生境差分进化算法在配网无功优化中的应用

改进小生境差分进化算法在配网无功优化中的应用黄俊辉1，李琥1，衣涛2，元梨花3，韩俊1 （1.国网江苏省电力公司经济技术研究院，江苏南京210000；2.上海交通大学电气工程系，上海200240；3.上海博英信息科技有限公司，上海200240）【摘要】摘要：配网无功优化是一类非线性的整数规划问题，通过调整变压器的变比，改变发电机的端电压和连接补偿电容来改变电力网络中的无功，减小系统网损。

差分进化算法是一种收敛速度快，收敛精度高的智能进化算法，针对无功优化模型对差分进化算法做出改进，引入小生境思想。

通过实例验证了小生境粒子群算法（NPSO）和改进小生境差分进化算法（FERDE）对无功补偿装置布点优化规划的有效性。

结果表明，增强算法的局部搜索能力和扩宽搜索范围，在收敛速度和精度上都有不同程度的提高。

【期刊名称】电网与清洁能源【年(卷),期】2015(031)006【总页数】5【关键词】配网无功优化；差分进化算法；小生境；粒子群优化配网无功优化问题是一个多变量、多约束的混合非线性规划问题，其控制变量既有连续变量，又有离散变量，整个优化过程十分复杂，计算规模大。

从传统的算法，如线性规划法、非线性规划法等，到人工智能算法，如粒子群优化算法、遗传算法等，都在不同程度上对无功优化做出贡献。

随着智能启发式优化算法的发展，差分进化算法逐步被应用到电力系统中，该算法具有易理解、并行处理、鲁棒性好等特点，能以较大概率找到问题的全局最优解，且计算效率比传统的进化规划等算法高。

其最大的优势在于简单易实现、收敛速度快、搜索精度高，不但适合科学研究，而且适合工程应用。

因此，差分进化算法（Differential evolution algorithm，DE）一经提出，立刻引起了演化计算领域研究者的广泛关注，并涌现出大量的研究成果，已经在函数优化、神经网络设计、分类、模式识别、信号处理、机器人技术等应用领域取得了成功应用[1-2]。

基于神经网络的微生物生长预测模型

基于神经网络的微生物生长预测模型侯奇;刘静;管骁【摘要】鉴于现有大多数预测模型都是经验型模型,含有过多没有生物解释的参数,提出一个基于神经网络的非经验型的微生物生长预测模型,并以李斯特菌为研究实例,利用其试验环境的温度、pH 值和Aw值建立BP神经网络二级生长模型,在不同环境条件下拟合微生物的生长速率和倍增时间,结合微生物初始浓度对一级模型的时间与微生物生长情况进行预测,最后利用李斯特菌生长数据对模型进行仿真测试.试验结果证明,该模型可以对微生物生长的各个时期进行有效预测,相对于经验模型,该模型更加适用于微生物生长动力学预测,有效地解决了经验型模型的参数问题.%Most of the existing predictive models are empirical models which contain too many parameters without biological explanations. In this study,a non-empirical growth prediction model based on neu-ral network was proposed.A BP neural network secondary growth model was established by using Listeria monocytogenes as an exam-ple,using the temperature,pH value and Aw value of the experi-mental environment.The growth rate and double time of microbes were fitted in different environments.Subsequently,combining with the initial concentration of microorganisms,the primary model of mi-croorganism growth with time was predicted.Finally,the growth data of Listeria monocytogenes were tested,and the experimental re-sults showed that the model could predict the growth period of pared with the empirical model,this non-empirical pre-diction one was more suitable for predicting themicrobial growth dy-namics,and also the parameters of the empirical model could be solved effectively.【期刊名称】《食品与机械》【年(卷),期】2018(034)002【总页数】4页(P120-123)【关键词】微生物;生长预测模型;神经网络【作者】侯奇;刘静;管骁【作者单位】上海海事大学信息工程学院,上海 201306;上海海事大学信息工程学院,上海 201306;上海理工大学医疗器械与食品学院,上海 200093【正文语种】中文预测微生物学本质上是基于微生物群体对环境因素反应的可重现，利用过去观察到的试验数据通过数学模型预测食物环境中的微生物行为，并用试验结果证明模型所得到的误差不大于微生物试验所带来的误差[1]。

TFT-LCD 概况及技术原理

➢驱动方式:主动矩阵 TFT 单纯矩阵 TN/STN
➢Array制程:需要 TFT 不需要 TN/STN
➢液晶材料:Nematic/Chiral dopant TFT/TN Cholesteric(Chiral Nematic) STN
➢配向方向: Twist 90o TFT/TN Twist 180o~270o STN
数据源:工研院IEK,台湾TFT-LCD产业协会
投資金額
(億台幣) 200 238 200 250 200 250 38 160 160 155 200 150 75 200 250
基板總產能
(萬片/月) 3.5 3 5 3 5 3 -3.75 3.75 2.5 4 4 3.6 3 --
量產時間
1999/5 2001/5 1999/7 2001/1 1999/12 2001/3Q 1995/10 2000/1 2000/4Q (已暫停) 1999/12 2001/2Q 1995/10
AAttttaacchhmmeenntt
大片组合
UV胶
CF基板
TFT基板
力盛光电
HHoott Press
气囊热压
压力表大片组合之基板
力盛光电
CCeell l液液晶晶段段制制程程
Scribing
Breaking 切裂
Inspection
Injection
End Seal 注入＆封口
Cleaning
4-Mask
力盛光电
AArrrraayy ((55 MMask)
Cr
SiNx;a-Si; n+-Si
Cr/Al Passivation
ITO
Gate端子部
Gate 力盛光电

基于JITL的多模态工业数据预测

基于JITL的多模态工业数据预测发布时间：2021-10-14T07:32:32.120Z 来源：《科学与技术》2021年17期作者：陈雨杉[导读] 在工业过程中，由于产品变化、工况切换或控制器调整等原因，过程数据往往呈现多模态特征。

数据驱动方法通常基于单模态假设，这可能无法描述过程特征。

陈雨杉杭州电子科技大学浙江杭州 310018摘要：在工业过程中，由于产品变化、工况切换或控制器调整等原因，过程数据往往呈现多模态特征。

数据驱动方法通常基于单模态假设，这可能无法描述过程特征。

传统的实时学习（JITL）方法能够不断更新模型来描述多模态数据，但耗时长，不能满足实时性要求。

本文提出了一种改进的JITL方法来快速发现相似样本。

首先将新样本划分为主类别，然后查找相似样本，提高搜索效率。

通过一个工业软测量实例与偏最小二乘法（PLS）相结合，证明了该方法的有效性，与基本JITL相比，该方法的均方根误差（RMSE）降低了0.09，运行速度提高了8.8倍。

关键词：软测量、即时学习、多模式、偏最小二乘、数据驱动一、背景介绍在实际的工业过程中，追求产品质量改进是一项长期且具有工业价值的任务。

然而，由于设备的成本或环境的复杂性，许多关键的过程变量很难获得。

随着人工智能和数据存储技术的发展，软测量越来越受到人们的重视。

数据驱动的软测量方法有许多吸引人的特性：（1）它们为昂贵的硬件传感器提供了一种低成本的替代方案(2）它们允许实时估计数据，克服了缓慢的硬件传感器带来的时间延迟，从而提高了控制算法的性能(3）它们在质量控制中起着不可或缺的作用。

在过去的几十年中，基于数据的软测量建模方法已经得到了广泛的研究，如支持向量机（SVM）[1]，人工神经网络（ANN）[2]，偏最小二乘（PLS）[3]。

支持向量机被定义为一个凸二次优化问题，它具有计算量小、优化选择方便等优点。

然而，当输入大规模样本时，模型的构建很难实现。

神经网络通过建立数据之间的关系和调整各种网络参数来建立模型。

GA与ANN的结合

GA与ANN的结合作者：吴伟来源：《吉林省教育学院学报·上旬刊》 2011年第9期吴伟（1．苏州市职业大学，江苏苏州 215104 2．苏州大学计算机科学与技术学院，江苏苏州 215104）摘要：本文主要讲述了GA算法的特点和ANN的优点并说明了GA与ANN结合的必要性，同时对今后的研究前景作了具体的展望。

关键词：遗传算法GA；人工神经网络ANN；结合中图分类号：G642文献标识码：A文章编号：1671—1580（2011）09—0111—02一、遗传算法(Genetic Algorithms，GA)遗传算法是一类借鉴生物界自然选择和自然遗传机理的随机优化算法，是模拟达尔文遗传选择和自然淘汰生物进化过程的计算模型。

主要特点是群体搜索策略和群体中个体的信息交换，搜索不依赖于梯度信息。

尤其适用于处理传统搜索方法难以解决的复杂和非线性问题。

可广泛用于组合优化，机器学习，自适应控制，规划设计和人工生命等领域。

随着问题种类的不同以及问题规模的扩大，要寻求一种能以有限的代价来解决搜索和优化的通用方法，GA正是为我们提供的一个有效的途径，它不同于传统的搜索和优化方法。

主要区别在于：一是自组织、自适应和自学习性(智能性)。

应用GA求解问题时，在编码方案、适应度函数及遗传算子确定后，算法将利用进化过程中获得的信息自行组织搜索。

通常，适应度大的个体具有更适应环境的基因结构，再通过基因重组和基因突变等遗传操作，就可能产生更适应环境的后代。

进化算法的这种自组织、自适应特征，使它同时具有能根据环境变化来自动发现环境的特性和规律的能力。

自然选择消除了算法设计过程中的一个最大障碍，即需事先描述问题的全部特点。

因此，利用GA的方法，可以解决复杂的非结构化问题。

二是GA的本质并行性。

GA按并行方式搜索一个种群数目的点，而不是单点。

它的并行性表现在两个方面，一方面遗传算法是内在并行的(inherent parallelism)，即GA本身非常适合大规模并行；另一方面是GA的内涵并行性(implicit parallelism)。

建筑工程毕业设计外文翻译英文原文

建筑工程毕业设计外文翻译英文原文The effects of surface preparation on the fracture behavior ofECC/concrete repair systemToshiro Kamada a,*, Victor C. Li ba Department of Civil Engineering, Gifu University, Yanagido, Gifu 501-1193, Japanb Advanced Civil Engineering Materials Research Laboratory, Department of Civil and Environmental Engineering,University of Michigan, Ann Arbor, Michigan, MI 48109-2125, USAReceived 7 July 1999; accepted 15 May 2000AbstractThis paper presents the influence of surface preparation on thekink-crack trapping mechanism of engineered cementitious composite (ECC)/concrete repair system. In general,surfacepreparation of the substrate concrete is considered essential to achieve a durable repair. In thisexperiment, the ``smooth sur face’’ system showed more desirable behavior in the crack pattern and the crack widths than the ``rough surface’’ system. This demonstrates that the smooth surface system is preferable to the rough surface system, from the view point of obtaining durable repair structure. The special phenomenon of kink-crack trapping which prevents the typical failuremodes of delamination or spalling in repaired systems is best revealed when the substrate concrete is prepared to have a smooth surface prior to repair. This is in contrast to the standard approach when the substrate concrete is deliberately roughened to create better bonding to the new concrete. Ó 2000 Elsevier Science Ltd. All rights reserved.Keywords: ECC repair system; Kink-crack trapping mechanism; Surface preparation; Durable repair1. IntroductionEngineered cementitious composites (ECCs) [1,2] are high performance fiber-reinforced cement based composite materials designed with micromechanical principles. Micromechanicalparameters associated with fiber, matrix and interface are combined to satisfy a pair of criteria, the first crack stress criterion and steady state cracking criterion [3] to achieve the strain hardening behavior. Micromechanics allows optimization of the composite for high performance while minimizing the amount of reinforcing fibers (generally less than 2-3%). ECC has a tensile strain capacity of up to 6% and exhibits pseudo-strain hardening behavior accompanied by multiple cracking. It also has high ultimate tensile strength (5-10 MPa), modulus of rupture (8-25 MPa), fracture toughness (25-30 kJ/m2) and compressive strength (up to 80 MPa) and strain (0.6%). A typical tensile stress-strain curve is shown in Fig. 1. ECC has its uniqueness not only insuperior mechanical properties in tension or in relatively small amount ofchopped fiber usage but also in micromechanical methodology in material design.The use of ECC for concrete repair was proposed by Li et al. [4], and Lim and Li [5]. In theseexperiments, specimens representative of an actual repair system - bonded overlay of a concrete pavement above a joint, were used. It was shown that the common failure phenomenona ofspalling or delamination in repaired concrete systems were eliminated. Instead, microcracksemanated from the tips of defects on the ECC/concrete interface, kinked into and subsequently were arrested in the ECC material (see Fig. 2, [5]). The tendency for the interface crack to kink into the ECC material depends on the competing driving force for crack extension at differentorientations, and on the competing crack extension resistance along the interface and into the ECC material. A low initial toughness of ECC combined with a high Mode II loading configuration tends to favor kinking. However, if the toughness of ECC remains low after crack kinking, this crack will propagate unstably to the surface, forming a surface spall. This is the typically observed phenomenon associated with brittle concrete and even fiber-reinforced concrete (FRC). In the case of ECC, the kinked crack is trapped or arrested in the ECC material, dueto the rapidly rising toughness of the ECC material. Conceptually, the ECC behaves like a material with strong R-curve behavior, with lowinitial toughness similar to that of cement (0.01 kJ/m2) and high plateau toughness (25-30 kJ/m2). After kinked crack arrest,additional load can drive further crackextension into the interface, followed by subsequent kinking and arrest.Details of the energetics of kink-crack trapping mechanism can befound in [5]. It was pointed out that this kink-crack trapping mechanism could serve as a means for enhancing repaired concrete system durability.In standard concrete repair, surface preparation of the substrate concrete is considered critical in achieving a durable repair [6]. Inthe study of Lim and Li [5], the ECC is cast onto a diamond saw cut surface of the concrete. Hence, the concrete surface is smooth and is expected as a result to produce a low toughness interface. Higherinterface roughness has been associated with higher interface toughnessin bi-material systems [7].In this paper, this particular aspect of the influence of surface preparation on the kink-crack trapping phenomenon is investigated. Specifically, the base concrete surfaces were prepared by threedifferent methods. The first surface was obtained as cut surface byusing a diamond saw (smooth surface), similar to that used in theprevious study [5]. The second one was obtained by applying a lubricanton the smooth surface of the concrete to decrease the bond between thebase concrete and the repair material. This surface was applied only in one test case to examine the effect of weak bond of interface on the fracture behavior of the repaired specimen. The third surface was prepared with a portable scarifier to produce a roughened surface (rough surface) from a diamond saw-cut surface.Regarding the repair materials, the water/cement ratio of ECC was varied to control its toughness and strength. Thus, two different mixtures of ECC were used for the comparison of fracture behavior in both smooth and rough surface case. Concrete and steel fiber-reinforced concrete (SFRC) were also used as control repair materials instead of ECC.2. Experimental procedure2.1. Specimens and test methodsThe specimens in this experiment were designed to induce a defect in the form of aninterfacial crack between the repair material and the base concrete, as well as a joint in thesubstrate. Fig. 3 shows the dimensions of the designed specimen and the loading configuration, and these were the same as those of the previous experiment [5]. This loading condition can provide a stable interface crack propagation condition, when the crack propagates along the interface [8].In this experiment, concrete, SFRC and ECC (with two different W/C ratios) were used as therepair materials. Table 1 illustrates the combinations of the repair material and the surface condition of test specimens. The numbers of specimens are also shown in Table 1. Only in the concrete overlay specimens, a special case where lubricant was smeared on the concrete smooth surface was used.The mix proportions of materials are shown in Table 2. Ordinary mixture proportions wereadopted in concrete and SFRC as controls for comparisons with ECC overlay specimens. The steel fiber for SFRC was ``I.S fiber’’, straight with indented surfaceand rectangular cross-section (0.5* 0.5 mm2), 30 mm in length. An investigation using a steel fiber with hooked ends had already been performed in the previous study [5]. Polyethylene fiber (Trade name Spectra 900) with 19 mm length and 0.038 mm diameter was used for ECC. The elastic modulus, the tensile strength and the fiber density of Spectra 900 were 120 GPa, 2700 MPa and 0.98 g/cm3, respectively. Two different ECCs were used with different water/cement ratios. The mechanical properties of the base concrete and the repair materials are shown in Table 3. The tensile strain capacity of the ECC materials are not measured, but are estimated to be in excess of 3% based on test results of similar materials [2].An MTS machine was used for loading. Load and load point displacement were recorded. The loading rate in this experiment was0.005 mm/s. After the final failure of specimens, interface crack (extension) lengths were measured at both (left and right) sides of a specimen as the distance from a initial notch tip to a propagated crack tip along the interface between the base concrete and the repair material.2.2. Specimen preparationMost of the specimen preparation procedures followed those of the previous work [5]. The base concrete was prepared by cutting a concrete block (see Fig. 4(a)) into four pieces (see Fig. 4(b)) using a diamond saw. Two out of the four pieces were usedfor one smooth surface repairspecimen. In order to make a rough surface, a cut surface was roughened uniformly with ascarifier for 30 s. To prepare a repair specimen in the form of an overlay system, a repair material was cast against either the smooth surface or the rough surface of the base concrete blocks (see Fig. 5). Special attention was paid both to maintain cleanliness and to provide adequate moisture on the base concrete surface just before the casting. In two of the concrete overlay specimens, lubricant was sprayed on the smooth surface just before concrete casting. The initial notch and joint were made by applying a smooth tape on the base concrete before casting the repair materials(see Fig. 4(c)).The specimens were cured for 4 weeks in water. Eventually, the base concrete was cured for a total of 8 weeks, and repair materials were cured for 4 weeks in water. The specimens were dried for 24 h before testing.3. Results and discussion3.1. Comparison of the ECC overlay system with the control systemsFig. 6 shows the representative load-deflection curves in each test case. The overall peak load and deflection at peak load are recorded in Table 4. In the ECC overlay system, the deflections at peak load, which reflect the system ductility, are considerably larger than those of both theconcrete overlay (about one order of magnitude higher) and the SFRC overlay system (over five times). These results show good agreement with the previous results [5]. Moreover, it is clear fromFig. 6 that the energy absorption capacity in the ECC overlay system is much enhanced when it is compared with the other systems. This significant improvement in ductility and in energyabsorption capacity of the ECC overlay system is expected to enhance the durability of repaired structures by resisting brittle failure. The ECC overlay system failed without spalling ordelamination of the interface, whereas, both the concrete and SFRC overlay systems failed by spalling in these experiments (Fig. 7).3.2. Influence of surface preparationBoth in the concrete overlay system and the SFRC overlay system, the peak load and thedeflection at peak load do not show significant differences between smooth surface specimen and rough surface specimen (Table 4). Thetypical failure mode for both overlay systems (for smooth surface) is shown in Fig. 7. In the concrete overlay specimen with lubricant on the interface, delamination between repair concrete and substrate occurred first, followed by a kinked crack which propagates unstably to the surface of the repair concrete. On the other hand, in the concrete overlay system without lubricant, the initial interface crack kinked out from the interface into the repair concrete with a sudden load drop, without any interface delamination. The fractured halves of the specimens separated completely in both smooth surface specimens and rough surfacespecimens. In the SFRC overlay system, the initial interface crack also kinked out into the SFRC and the load decreased gradually in both surface conditions of specimen. In all these repairsystems, a single kink-crack always leads to final failure, and the influence of surface preparation is not reflected in the experimental data. Instead, only the fracture behavior of the repair material (concrete versus SFRC) are revealed in the test data. These specimen failures are characterized bya single kinked crack with immediate softening following elastic response.。

布拉酵母高密度发酵培养基及发酵工艺优化

yield of S. boulardii reached 51.21 g/L in the large scale experiment.
Keywords: Saccharomyces boulardii; neural network; genetic algorithm; enrichment medium; high cell density fermentation
56 2019, Vol.40, No.08
食品科学
※生物工程
布拉酵母高密度发酵培养基及发酵工艺优化
刘开放，席志文，黄林娜，惠丰立*
（南阳师范学院生命科学与技术学院，河南南阳 473061）
摘要：为实现布拉酵母高密度培养，对其高密度发酵培养基和发酵工艺进行优化。采用Plackett-Burman试验筛选培养基中的显著因素，并进行中心组合设计。通过人工神经网络（artificial neural network，ANN）和响应面试验建立菌体布拉酵母产量与培养基之间的关系模型，利用遗传算法（genetic algorithm，GA）进行全局寻优。结果表明，ANN模型有较好的数据拟合能力和预测能力，更适合处理复杂的非线性问题。GA优化获得最佳培养基组合：葡萄糖40.52 g/L、蛋白胨36.8 g/L、玉米浆17.32 g/L、硝酸钾14 g/L、酵母营养盐1.5 g/L、磷酸二氢钾0.6 g/L、硫酸镁0.8 g/L。利用该培养基进行摇瓶培养，菌体布拉酵母产量可达到8.21 g/L，比优化前提高1.39 倍。在此基础上利用1 L发酵罐培养确定最佳发酵工艺：温度30 ℃、接种量10%、pH 5.0、溶氧40%。利用50 L发酵罐进行扩大培养，流加葡萄糖和蛋白胨控制发酵液中葡萄糖3 g/L、氨氮0.06 g/L，菌体布拉酵母产量达到51.21 g/L。关键词：布拉酵母；神经网络；遗传算法；增殖培养基；高密度培养

中英文双语外文文献翻译：一种基于...

中英⽂双语外⽂⽂献翻译：⼀种基于...此⽂档是毕业设计外⽂翻译成品（含英⽂原⽂+中⽂翻译），⽆需调整复杂的格式！下载之后直接可⽤，⽅便快捷！本⽂价格不贵,也就⼏⼗块钱！⼀辈⼦也就⼀次的事！英⽂3890单词，20217字符(字符就是印刷符)，中⽂6398汉字。

A Novel Divide-and-Conquer Model for CPI Prediction UsingARIMA, Gray Model and BPNNAbstract:This paper proposes a novel divide-and-conquer model for CPI prediction with the existing compilation method of the Consumer Price Index (CPI) in China. Historical national CPI time series is preliminary divided into eight sub-indexes including food, articles for smoking and drinking, clothing, household facilities, articles and maintenance services, health care and personal articles, transportation and communication, recreation, education and culture articles and services, and residence. Three models including back propagation neural network (BPNN) model, grey forecasting model (GM (1, 1)) and autoregressive integrated moving average (ARIMA) model are established to predict each sub-index, respectively. Then the best predicting result among the three models’for each sub-index is identified. To further improve the performance, special modification in predicting method is done to sub-CPIs whose forecasting results are not satisfying enough. After improvement and error adjustment, we get the advanced predicting results of the sub-CPIs. Eventually, the best predicting results of each sub-index are integrated to form the forecasting results of the national CPI. Empirical analysis demonstrates that the accuracy and stability of the introduced method in this paper is better than many commonly adopted forecasting methods, which indicates the proposed method is an effective and alternative one for national CPI prediction in China.1.IntroductionThe Consumer Price Index (CPI) is a widely used measurement of cost of living. It not only affects the government monetary, fiscal, consumption, prices, wages, social security, but also closely relates to the residents’daily life. As an indicator of inflation in China economy, the change of CPI undergoes intense scrutiny. For instance, The People's Bank of China raised the deposit reserve ratio in January, 2008 before the CPI of 2007 was announced, for it is estimated that the CPI in 2008 will increase significantly if no action is taken. Therefore, precisely forecasting the change of CPI is significant to many aspects of economics, some examples include fiscal policy, financial markets and productivity. Also, building a stable and accurate model to forecast the CPI will have great significance for the public, policymakers and research scholars.Previous studies have already proposed many methods and models to predict economic time series or indexes such as CPI. Some previous studies make use of factors that influence the value of the index and forecast it by investigating the relationship between the data of those factors and the index. These forecasts are realized by models such as Vector autoregressive (VAR)model1 and genetic algorithms-support vector machine (GA-SVM) 2.However, these factor-based methods, although effective to some extent, simply rely on the correlation between the value of the index and limited number of exogenous variables (factors) and basically ignore the inherent rules of the variation of the time series. As a time series itself contains significant amount of information3, often more than a limited number of factors can do, time series-based models are often more effective in the field of prediction than factor-based models.Various time series models have been proposed to find the inherent rules of the variation in the series. Many researchers have applied different time series models to forecasting the CPI and other time series data. For example, the ARIMA model once served as a practical method in predicting the CPI4. It was also applied to predict submicron particle concentrations frommeteorological factors at a busy roadside in Hangzhou, China5. What’s more, the ARIMA model was adopted to analyse the trend of pre-monsoon rainfall data forwestern India6. Besides the ARIMA model, other models such as the neural network, gray model are also widely used in the field of prediction. Hwang used the neural-network to forecast time series corresponding to ARMA (p, q) structures and found that the BPNNs generally perform well and consistently when a particular noise level is considered during the network training7. Aiken also used a neural network to predict the level of CPI and reached a high degree of accuracy8. Apart from the neural network models, a seasonal discrete grey forecasting model for fashion retailing was proposed and was found practical for fashion retail sales forecasting with short historical data and better than other state-of-art forecastingtechniques9. Similarly, a discrete Grey Correlation Model was also used in CPI prediction10. Also, Ma et al. used gray model optimized by particle swarm optimization algorithm to forecast iron ore import and consumption of China11. Furthermore, to deal with the nonlinear condition, a modified Radial Basis Function (RBF) was proposed by researchers.In this paper, we propose a new method called “divide-and-conquer model”for the prediction of the CPI.We divide the total CPI into eight categories according to the CPI construction and then forecast the eight sub- CPIs using the GM (1, 1) model, the ARIMA model and the BPNN. To further improve the performance, we again make prediction of the sub-CPIs whoseforecasting results are not satisfying enough by adopting new forecasting methods. After improvement and error adjustment, we get the advanced predicting results of the sub-CPIs. Finally we get the total CPI prediction by integrating the best forecasting results of each sub-CPI.The rest of this paper is organized as follows. In section 2, we give a brief introduction of the three models mentioned above. And then the proposed model will be demonstrated in the section 3. In section 4 we provide the forecasting results of our model and in section 5 we make special improvement by adjusting the forecasting methods of sub-CPIs whose predicting results are not satisfying enough. And in section 6 we give elaborate discussion and evaluation of the proposed model. Finally, the conclusion is summarized in section 7.2.Introduction to GM(1,1), ARIMA & BPNNIntroduction to GM(1,1)The grey system theory is first presented by Deng in 1980s. In the grey forecasting model, the time series can be predicted accurately even with a small sample by directly estimating the interrelation of data. The GM(1,1) model is one type of the grey forecasting which is widely adopted. It is a differential equation model of which the order is 1 and the number of variable is 1, too. The differential equation is:Introduction to ARIMAAutoregressive Integrated Moving Average (ARIMA) model was first put forward by Box and Jenkins in 1970. The model has been very successful by taking full advantage of time series data in the past and present. ARIMA model is usually described as ARIMA (p, d, q), p refers to the order of the autoregressive variable, while d and q refer to integrated, and moving average parts of the model respectively. When one of the three parameters is zero, the model is changed to model “AR”, “MR”or “ARMR”. When none of the three parameters is zero, the model is given by:where L is the lag number,?t is the error term.Introduction to BPNNArtificial Neural Network (ANN) is a mathematical and computational model which imitates the operation of neural networks of human brain. ANN consists of several layers of neurons. Neurons of contiguous layers are connected with each other. The values of connections between neurons are called “weight”. Back Propagation Neural Network (BPNN) is one of the most widely employed neural network among various types of ANN. BPNN was put forward by Rumelhart and McClelland in 1985. It is a common supervised learning network well suited for prediction. BPNN consists of three parts including one input layer, several hidden layers and one output layer, as is demonstrated in Fig 1. The learning process of BPNN is modifying the weights of connections between neurons based on the deviation between the actual output and the target output until the overall error is in the acceptable range.Fig. 1. Back-propagation Neural Network3.The Proposed MethodThe framework of the dividing-integration modelThe process of forecasting national CPI using the dividing-integration model is demonstrated in Fig 2.Fig. 2.The framework of the dividing-integration modelAs can be seen from Fig. 2, the process of the proposed method can be divided into the following steps: Step1: Data collection. The monthly CPI data including total CPI and eight sub-CPIs are collected from the official website of China’s State Statistics Bureau (/doc/d62de4b46d175f0e7cd184254b35eefdc9d31514.html /).Step2: Dividing the total CPI into eight sub-CPIs. In this step, the respective weight coefficient of eight sub- CPIs in forming the total CPI is decided by consulting authoritative source .(/doc/d62de4b46d175f0e7cd184254b35eefdc9d31514.html /). The eight sub-CPIs are as follows: 1. Food CPI; 2. Articles for Smoking and Drinking CPI; 3. Clothing CPI; 4. Household Facilities, Articles and Maintenance Services CPI; 5. Health Care and Personal Articles CPI; 6. Transportation and Communication CPI;7. Recreation, Education and Culture Articles and Services CPI; 8. Residence CPI. The weight coefficient of each sub-CPI is shown in Table 8.Table 1. 8 sub-CPIs weight coefficient in the total indexNote: The index number stands for the corresponding type of sub-CPI mentioned before. Other indexes appearing in this paper in such form have the same meaning as this one.So the decomposition formula is presented as follows:where TI is the total index; Ii (i 1,2, ,8) are eight sub-CPIs. To verify the formula, we substitute historical numeric CPI and sub-CPI values obtained in Step1 into the formula and find the formula is accurate.Step3: The construction of the GM (1, 1) model, the ARIMA (p, d, q) model and the BPNN model. The three models are established to predict the eight sub-CPIs respectively.Step4: Forecasting the eight sub-CPIs using the three models mentioned in Step3 and choosing the best forecasting result for each sub-CPI based on the errors of the data obtained from the three models.Step5: Making special improvement by adjusting the forecasting methods of sub-CPIs whose predicting results are not satisfying enough and get advanced predicting results of total CPI. Step6: Integrating the best forecasting results of 8 sub-CPIs to form the prediction of total CPI with the decomposition formula in Step2.In this way, the whole process of the prediction by the dividing-integration model is accomplished.3.2. The construction of the GM(1,1) modelThe process of GM (1, 1) model is represented in the following steps:Step1: The original sequence:Step2: Estimate the parameters a and u using the ordinary least square (OLS). Step3: Solve equation as follows.Step4: Test the model using the variance ratio and small error possibility.The construction of the ARIMA modelFirstly, ADF unit root test is used to test the stationarity of the time series. If the initial time series is not stationary, a differencing transformation of the data is necessary to make it stationary. Then the values of p and q are determined by observing the autocorrelation graph, partial correlation graph and the R-squared value.After the model is built, additional judge should be done to guarantee that the residual error is white noise through hypothesis testing. Finally the model is used to forecast the future trend ofthe variable.The construction of the BPNN modelThe first thing is to decide the basic structure of BP neural network. After experiments, we consider 3 input nodes and 1 output nodes to be the best for the BPNN model. This means we use the CPI data of time , ,toforecast the CPI of time .The hidden layer level and the number of hidden neurons should also be defined. Since the single-hidden- layer BPNN are very good at non-liner mapping, the model is adopted in this paper. Based on the Kolmogorov theorem and testing results, we define 5 to be the best number of hidden neurons. Thus the 3-5-1 BPNN structure is determined.As for transferring function and training algorithm, we select ‘tansig’as the transferring function for middle layer, ‘logsig’for input layer and ‘traingd’as training algorithm. The selection is based on the actual performance of these functions, as there are no existing standards to decide which ones are definitely better than others.Eventually, we decide the training times to be 35000 and the goal or the acceptable error to be 0.01.4.Empirical AnalysisCPI data from Jan. 2012 to Mar. 2013 are used to build the three models and the data from Apr. 2013 to Sept. 2013 are used to test the accuracy and stability of these models. What’s more, the MAPE is adopted to evaluate the performance of models. The MAPE is calculated by the equation:Data sourceAn appropriate empirical analysis based on the above discussion can be performed using suitably disaggregated data. We collect the monthly data of sub-CPIs from the website of National Bureau of Statistics of China(/doc/d62de4b46d175f0e7cd184254b35eefdc9d31514.html /).Particularly, sub-CPI data from Jan. 2012 to Mar. 2013 are used to build the three models and the data from Apr. 2013 to Sept. 2013 are used to test the accuracy and stability of these models.Experimental resultsWe use MATLAB to build the GM (1,1) model and the BPNN model, and Eviews 6.0 to build the ARIMA model. The relative predicting errors of sub-CPIs are shown in Table 2.Table 2.Error of Sub-CPIs of the 3 ModelsFrom the table above, we find that the performance of different models varies a lot, because the characteristic of the sub-CPIs are different. Some sub-CPIs like the Food CPI changes drastically with time while some do not have much fluctuation, like the Clothing CPI. We use different models to predict the sub- CPIs and combine them by equation 7.Where Y refers to the predicted rate of the total CPI, is the weight of the sub-CPI which has already been shown in Table1and is the predicted value of the sub-CPI which has the minimum error among the three models mentioned above. The model chosen will be demonstrated in Table 3:Table 3.The model used to forecastAfter calculating, the error of the total CPI forecasting by the dividing-integration model is 0.0034.5.Model Improvement & Error AdjustmentAs we can see from Table 3, the prediction errors of sub-CPIs are mostly below 0.004 except for two sub- CPIs: Food CPI whose error reaches 0.0059 and Transportation & Communication CPI 0.0047.In order to further improve our forecasting results, we modify the prediction errors of the two aforementioned sub-CPIs by adopting other forecasting methods or models to predict them. The specific methods are as follows.Error adjustment of food CPIIn previous prediction, we predict the Food CPI using the BPNN model directly. However, the BPNN model is not sensitive enough to investigate the variation in the values of the data. For instance, although the Food CPI varies a lot from month to month, the forecasting values of it are nearly all around 103.5, which fails to make meaningful prediction.We ascribe this problem to the feature of the training data. As we can see from the original sub-CPI data on the website of National Bureau of Statistics of China, nearly all values of sub-CPIs are around 100. As for Food CPI, although it does have more absolute variations than others, its changes are still very small relative to the large magnitude of the data (100). Thus it will be more difficult for the BPNN model to detect the rules of variations in training data and the forecastingresults are marred.Therefore, we use the first-order difference series of Food CPI instead of the original series to magnify the relative variation of the series forecasted by the BPNN. The training data and testing data are the same as that in previous prediction. The parameters and functions of BPNN are automatically decided by the software, SPSS.We make 100 tests and find the average forecasting error of Food CPI by this method is 0.0028. The part of the forecasting errors in our tests is shown as follows in Table 4:Table 4.The forecasting errors in BPNN testError adjustment of transportation &communication CPIWe use the Moving Average (MA) model to make new prediction of the Transportation and Communication CPI because the curve of the series is quite smooth with only a few fluctuations. We have the following equation(s):where X1, X2…Xn is the time series of the Transportation and Communication CPI, is the value of moving average at time t, is a free parameter which should be decided through experiment.To get the optimal model, we range the value of from 0 to 1. Finally we find that when the value of a is 0.95, the forecasting error is the smallest, which is 0.0039.The predicting outcomes are shown as follows in Table5:Table 5.The Predicting Outcomes of MA modelAdvanced results after adjustment to the modelsAfter making some adjustment to our previous model, we obtain the advanced results as follows in Table 6: Table 6.The model used to forecast and the Relative ErrorAfter calculating, the error of the total CPI forecasting by the dividing-integration model is 0.2359.6.Further DiscussionTo validate the dividing-integration model proposed in this paper, we compare the results of our model with the forecasting results of models that do not adopt the dividing-integration method. For instance, we use the ARIMA model, the GM (1, 1) model, the SARIMA model, the BRF neural network (BRFNN) model, the Verhulst model and the Vector Autoregression (VAR) model respectively to forecast the total CPI directly without the process of decomposition and integration. The forecasting results are shown as follows in Table7.From Table 7, we come to the conclusion that the introduction of dividing-integration method enhances the accuracy of prediction to a great extent. The results of model comparison indicate that the proposed method is not only novel but also valid and effective.The strengths of the proposed forecasting model are obvious. Every sub-CPI time series have different fluctuation characteristics. Some are relatively volatile and have sharp fluctuations such as the Food CPI while others are relatively gentle and quiet such as the Clothing CPI. As a result, by dividing the total CPI into several sub-CPIs, we are able to make use of the characteristics of each sub-CPI series and choose the best forecasting model among several models for every sub-CPI’s prediction. Moreover, the overall prediction error is provided in the following formula:where TE refers to the overall prediction error of the total CPI, is the weight of the sub-CPI shown in table 1 and is the forecasting error of corresponding sub-CPI.In conclusion, the dividing-integration model aims at minimizing the overall prediction errors by minimizing the forecasting errors of sub-CPIs.7.Conclusions and future workThis paper creatively transforms the forecasting of national CPI into the forecasting of 8 sub-CPIs. In the prediction of 8 sub-CPIs, we adopt three widely used models: the GM (1, 1) model, the ARIMA model and the BPNN model. Thus we can obtain the best forecasting results for each sub-CPI. Furthermore, we make special improvement by adjusting the forecasting methods of sub-CPIs whose predicting results are not satisfying enough and get the advanced predicting results of them. Finally, the advanced predicting results of the 8 sub- CPIs are integrated to formthe forecasting results of the total CPI.Furthermore, the proposed method also has several weaknesses and needs improving. Firstly, The proposed model only uses the information of the CPI time series itself. If the model can make use of other information such as the information provided by factors which make great impact on the fluctuation of sub-CPIs, we have every reason to believe that the accuracy and stability of the model can be enhanced. For instance, the price of pork is a major factor in shaping the Food CPI. If this factor is taken into consideration in the prediction of Food CPI, the forecasting results will probably be improved to a great extent. Second, since these models forecast the future by looking at the past, they are not able to sense the sudden or recent change of the environment. So if the model can take web news or quick public reactions with account, it will react much faster to sudden incidence and affairs. Finally, the performance of sub-CPIs prediction can be higher. In this paper we use GM (1, 1), ARIMA and BPNN to forecast sub-CPIs. Some new method for prediction can be used. For instance, besides BPNN, there are other neural networks like genetic algorithm neural network (GANN) and wavelet neural network (WNN), which might have better performance in prediction of sub-CPIs. Other methods such as the VAR model and the SARIMA model should also be taken into consideration so as to enhance the accuracy of prediction.References1.Wang W, Wang T, and Shi Y. Factor analysis on consumer price index rising in China from 2005 to 2008. Management and service science 2009; p. 1-4.2.Qin F, Ma T, and Wang J. The CPI forecast based on GA-SVM. Information networking and automation 2010; p. 142-147.3.George EPB, Gwilym MJ, and Gregory CR. Time series analysis: forecasting and control. 4th ed. Canada: Wiley; 20084.Weng D. The consumer price index forecast based on ARIMA model. WASE International conferenceon information engineering 2010;p. 307-310.5.Jian L, Zhao Y, Zhu YP, Zhang MB, Bertolatti D. An application of ARIMA model to predict submicron particle concentrations from meteorological factors at a busy roadside in Hangzhou, China. Science of total enviroment2012;426:336-345.6.Priya N, Ashoke B, Sumana S, Kamna S. Trend analysis and ARIMA modelling of pre-monsoon rainfall data forwestern India. Comptesrendus geoscience 2013;345:22-27.7.Hwang HB. Insights into neural-network forecasting of time seriescorresponding to ARMA(p; q) structures. Omega2001;29:273-289./doc/d62de4b46d175f0e7cd184254b35eefdc9d31514.html am A. Using a neural network to forecast inflation. Industrial management & data systems 1999;7:296-301.9.Min X, Wong WK. A seasonal discrete grey forecasting model for fashion retailing. Knowledge based systems 2014;57:119-126.11. Weimin M, Xiaoxi Z, Miaomiao W. Forecasting iron ore import and consumption of China using grey model optimized by particleswarm optimization algorithm. Resources policy 2013;38:613-620.12. Zhen D, and Feng S. A novel DGM (1, 1) model for consumer price index forecasting. Greysystems and intelligent services (GSIS)2009; p. 303-307.13. Yu W, and Xu D. Prediction and analysis of Chinese CPI based on RBF neural network. Information technology and applications2009;3:530-533.14. Zhang GP. Time series forecasting using a hybrid ARIMA and neural network model. Neurocomputing 2003;50:159-175.15. Pai PF, Lin CS. A hybrid ARIMA and support vector machines model in stock price forecasting. Omega 2005;33(6):497-505.16. Tseng FM, Yu HC, Tzeng GH. Combining neural network model with seasonal time series ARIMA model. Technological forecastingand social change 2002;69(1):71-87.17.Cho MY, Hwang JC, Chen CS. Customer short term load forecasting by using ARIMA transfer function model. Energy management and power delivery, proceedings of EMPD'95. 1995 international conference on IEEE, 1995;1:317-322.译⽂：⼀种基于ARIMA、灰⾊模型和BPNN对CPI（消费物价指数）进⾏预测的新型分治模型摘要:在本⽂中，利⽤我国现有的消费者价格指数（CPI）的计算⽅法，提出了⼀种新的CPI预测分治模型。

韩国先进科技学院机械工程系Ikjin Lee,副教授简介

基于改进GA算法的高速公路交通拥堵状况预测

第40卷　第2期吉林大学学报(信息科学版)Vol.40　No.22022年3月Journal of Jilin University (Information Science Edition)Mar.2022文章编号:1671⁃5896(2022)02⁃0198⁃08基于改进GA 算法的高速公路交通拥堵状况预测收稿日期:2021⁃07⁃09基金项目:国家社会科学重点基金资助项目(16AGJ007)作者简介:黄承锋(1965 ),男,重庆人,重庆交通大学教授,博士,主要从事交通运输经济研究,(Tel)86⁃199****9076(E⁃mail)huang18601@㊂黄承锋,陈一铭,李元龙(重庆交通大学经济与管理学院,重庆400074)摘要:目前高速公路拥堵状况发生频率越来越高,为给驾驶者提供便利的出行路径,减缓道路交通拥堵状态,以流量统计为基础,设计了基于改进遗传算法的高速公路交通拥堵状况预测模型㊂利用固定与移动检测技术采集流量㊁密度以及速度等宏观交通流量数据,针对冗余数据㊁缺失数据以及错误数据等异常类参数,采取不同的识别与处理方法,得到有效且完整的流量数据;利用反向传播神经网络与支持向量机回归网络改进遗传算法,建立两个子预测模型;通过加权处理两个模型权重构建混合预测模型,根据子预测模型拥堵预测偏差,结合最优权值组合策略修正混合预测模型的权值系数㊂实验结果表明,设计模型能划分目标高速公路的交通拥堵状况等级,可依据流量㊁速度以及占有率等数据预测拥堵状况,且模型预测精度较高,具有理想的预测有效性与准确性㊂关键词:交通流量统计;高速公路交通;拥堵状况预测;改进遗传算法;支持向量机回归;反向传播神经网络中图分类号:TP391.7文献标识码:AHighway Traffic Congestion Prediction Model Based on Improved Genetic AlgorithmHUANG Chengfeng,CHEN Yiming,LI Yuanlong(School of Economics and Management,Chongqing Jiaotong University,Chongqing 400074,China)Abstract :The frequency of highway congestion is increasing.In order to provide convenient travel path for drivers and slow down road traffic congestion,according to traffic statistics a highway traffic congestion prediction model based on improved genetic algorithm is designed.The fixed and mobile detection technology is used to collect macro traffic flow data such as flow,density and speed.Different identification and processing methodsare adopted for abnormal parameters such as redundant data,missing data and error data to obtain effective and complete traffic flow data.The back propagation neural network and support vector machine regression network are used to improve the genetic algorithm.Two sub prediction models are established,and a hybrid prediction model is constructed by weighting the weights of the two models.According to the congestion prediction deviation of the sub prediction model,the weight coefficient of the hybrid prediction model is modified combined with the optimal weight combination strategy.The experimental results show that the design model can divide the traffic congestion level of the target expressway,and predict the congestion status according to the data of traffic flow,speed and occupancy rate,and the model has high prediction accuracy and ideal prediction effectiveness andaccuracy.Key words :traffic flow statistics;highway traffic;congestion prediction;improved genetic algorithm;support vector machine regression;back propagation neural network0　引　言近年来高速公路覆盖率日益增大,交通管理难度越来越高,同时随着社会经济总量不断提升,车辆保有量逐年递增,导致交通拥堵情况日益严峻,这对行车成本㊁环境污染㊁事故发生概率㊁出行时长均有巨大影响,甚至会阻碍㊁制约城际之间的发展协调性[1]㊂因此,交通拥堵状况预测方法成为目前重点研究课题之一,科学预测拥堵状况不仅能为及时疏通车流提供可靠的参考数据,从而减少发生道路交通拥堵㊂戢晓峰等[2]基于深度学习理论,建立一种长短期记忆⁃支持向量机回归预测模型,在预测节假日不同时段㊁天气以及流量状态下的交通流时,表现出较好的适用性;蔡延光等[3]根据暴雨天气特点,通过融合布谷鸟搜索改进算法与径向基函数神经网络,设计出一种收敛速度较快㊁预测精度较高的交通流预测模型;温惠英等[4]在长短期记忆神经网络中融入遗传算法,架构出预测性能较好的交通流预测模型㊂虽然上述方法均有效预测出了高速公路的交通流,但不能满足当前的预测精度需求,因此,笔者基于流量统计理念,利用改进遗传算法设计了一种交通拥堵状况预测模型㊂将交通流量参数作为描述拥堵状况特征的物理量,其具有客观反映拥堵状况的能力;将固定检测技术与移动检测技术相结合,使流量数据更准确;为提升预测精度,采取不同策略识别㊁处理不同种类的异常数据;设计的预测模型中归一化处理有助于提升网络训练与收敛速度㊂1　交通流量统计1.1　流量数据采集在高速公路附近与行驶在该条公路上的移动车辆中分别安装多个固定交通检测设备[5]与移动图形处理器[6],采用互补性较强的固定与移动检测技术完成采集任务,以取得完善㊁精准的交通流量数据,为后续拥堵状况预测奠定良好的数据基础㊂因单一数据无法准确描述高速公路交通拥堵状况,故选取与交通流量息息相关的3个参数,从宏观角度反映公路交通流量特征,参数流量q㊁密度k以及速度v分别为交通系统负荷㊁交通流量内行车自由度㊁交通系统服务水平,三者之间的关系如下q=kv(1) 参数界定内容与计算方法如下㊂1)流量q㊂其反映某地单位时间内的行车数量㊂假设在T时长的检测时段内,总共行驶过n台车辆,则交通流量为q=n T(2) 2)密度k㊂其反映某地单位距离内的现有车辆个数㊂由于该参数属于瞬时值,受距离㊁时间干扰较大,故替换成由时间占有率R t与空间占有率R s构成的占有率参数,其中前者表示交通流量状态,后者表示交通密度,计算公式如下R t=1n∑n i=1t i(3)R s=1d∑n i=1l i(4)其中高速公路的检测距离为d,通过固定交通检测器测量长度为l i的第i台车辆需要花费t i时长㊂3)速度v㊂其反映单位时间里车辆的行驶长度㊂通常由区段均速v s与时段均速v t组成,两类速度均值分别指代固定路段内与固定检测位置的交通流量,计算公式如下v s=d 1 n∑n i=1t i (5) 991第2期黄承锋,等:基于改进GA算法的高速公路交通拥堵状况预测v t =1n ∑n i =1v i (6)其中v i 为第i 台车辆通过检测位置时的速度㊂1.2　流量数据预处理为保证数据质量,识别㊁处理具有不同异常情况的交通流量数据㊂1)冗余数据识别与处理㊂采用图1所示的流程识别冗余数据,直接去除识别结果㊂2)缺失数据识别与处理㊂采用图2所示流程识别缺失的交通流量数据㊂图1　冗余数据识别流程图图2　缺失数据识别流程图 Fig.1　Redundancy data identification flow chart Fig.2　Missing data identification flow chart 针对未更新或少更新与字段信息丢失等两类缺失数据,采取不同的补偿策略㊂1)当缺失数据属于未更新或少更新情况时,分为部分缺失与大量缺失两种状况,前者丢失的是不大于3个连续采样周期的数据量,后者丢失的则是超过该采样周期的数据量㊂对部分缺失数据量,利用K 近邻非参数回归补偿算法[7],依据图3流程进行完善㊂图3　补偿算法流程图Fig.3　Flow chart of the compensation algorithm 2)字段信息丢失㊂若极少量数据缺失,则忽略不计;若大量数据缺失,则补偿方法同上㊂3)错误数据识别与处理㊂假定样本均值与标准差分别为μ㊁δ,则车辆在单位时段中行车时间的正态分布[8]范围为[μ-2δ,μ+2δ],当行车时间比合理时限小时,认为对应数据为错误数据并去除该数据;对合理时限内的数据,通过方程组μ′=1n ∑n i =1t iδ′=∑n i =1(t i -μ′)2n -ìîíïïïïïï1(7)求解均值与标准差㊂若将解值代入[μ-2δ,μ+2δ]后超出给定的正态分布范围,则去除该错误数据;待全部数据均位于正态分布范围内,停止迭代,错误数据全部被处理㊂002吉林大学学报(信息科学版)第40卷2　高速公路交通拥堵状况预测模型构建依据高速公路基本服务水平分析指标与分级标准,划分交通拥堵状况如表1所示的6个等级㊂在拥堵状况判定阶段设定阈值T thr ,根据交通流量的速度标准化函数v′与占有率标准化函数R′,通过T thr =βv′+γR′(8)求解出对应等级的交通拥堵状况阈值㊂其中β㊁γ分别为对应参数权值㊂表1　高速公路交通拥堵状况等级利用反向传播神经网络改进遗传算法[9],构建拥堵状况预测模型如图4所示㊂图4　基于遗传算法⁃反向传播神经网络预测模型结构图Fig.4　Structural diagram of the prediction model based on the genetic algorithm⁃backpropagation neural network该模型的预测过程如下㊂1)经过加载样本数据,实施归一化训练,令取值范围为[0,1]㊂2)明确网络层数及其节点个数,若输入层有p 个节点,隐藏层有θ个节点,输出层有h 个节点,则通过不等式θ≤p +h +a (9)明确隐藏层节点个数㊂其中变量因子a 的取值范围为[1,10]㊂3)为避免快速饱和,从[0,1]中选取较小值,赋值初始网络权重与阈值㊂4)在遗传算法内输入初始权重与阈值,获取最优解㊂5)输出结果,通过误差函数不断训练网络,以取得最佳权重与阈值,待满足预设条件后终止训练㊂6)基于训练过的预测模型,把测试样本数据输入模型,输出最终的流量预测结果㊂2.2　基于遗传算法⁃支持向量机回归的预测模型图5为遗传算法与支持向量机回归[10]相结合的拥堵状况预测模型㊂该模型的预测过程如下㊂1)输入流量数据,将数据归一化至0~1之间,取得训练与测试数据;2)编码处理模型的不敏感损失ε㊁惩罚参数C ㊁核函数参数σ,设定筛选参数标准为K⁃交叉验证[11]的均方偏差均值;3)针对支持向量机回归核函数[12],其表达式由高斯径向基核函数实现界定,即102第2期黄承锋,等:基于改进GA 算法的高速公路交通拥堵状况预测K (x ,x i )=exp ‖x -x i ‖22σæèçöø÷2(10) 4)将得到的不敏感损失ε㊁惩罚参数C 以及核函数参数σ的最优解代入遗传算法⁃支持向量机回归预测模型,预测出交通流量参数数据㊂图5　基于遗传算法⁃支持向量机回归的预测模型示意图Fig.5　Schematic diagram of a predictive model based on genetic algorithm⁃support vector machine regression 2.3　混合预测模型基于上述两种预测模型,根据确定的各模型权重,经加权处理构建出一种混合预测模型,如图6所示㊂图6　混合预测模型示意图Fig.6　Schematic diagram of the hybrid prediction model 假设遗传算法⁃反向传播神经网络预测模型的预测结果为f 1,遗传算法⁃支持向量机回归预测模型的预测结果为f 2,w 1㊁w 2分别为两模型的对应权值系数,由此得到混合预测模型的交通流量数据预测表达式如下f b =w 1f 1+w 2f 2(11)其中b 为正整数,取值为1或2㊂为提升模型预测准度,采用最优权值组合策略获取模型权值系数,该策略目标函数及约束条件如下max(min)Q =Q (w 1,w 2)(12)∑2b =1w b =1w b ≥ìîíïïïï0(13) 假设时刻t 的模型预测偏差为e bt ,混合预测模型的偏差平方和为S =∑2t =1e 2t =∑2t =1∑2t =1w b e ()bt 2(14) 结合偏差平方和S 与目标函数Q ,推导出最小化偏差平方和的目标函数表达式如下202吉林大学学报(信息科学版)第40卷min S =∑2t =1∑2t =1w b e ()bt 2(15)式(15)的约束条件同式(13)㊂若两预测模型在t 时刻预测的拥堵偏差分别为e 1t ㊁e 2t ,则构建下列方程组w 1=∑2t =1e 22t -∑2t =1e 1t e 2t ∑2t =1e 21t +∑2t =1e 22t -2∑2t =1e 1t e 2t w 2=∑2t =1e 21t -∑2t =1e 1t e 2t ∑2t =1e 21t +∑2t =1e 22t -2∑2t =1e 1t e 2ìîíïïïïïïïïïït (16) 通过式(16)修正混合预测模型的权值系数㊂对遗传算法⁃反向传播神经网络与遗传算法⁃支持向量机回归预测模型,通过训练数据并经组合后得到混合预测模型,根据测试数据预测交通拥堵状况,设定该混合模型的预测时间间隔为10min,即每隔10min 可以获得一个新的预测结果㊂3　高速公路交通拥堵状况预测实验选取某地区的一条高速公路,在其周边与移动车辆上安装多个检测器,每隔60s 采集一次公路上的行驶车辆流量㊁速度以及占有率等交通流量参数数据,经预处理后,将连续10d 内得到的13482条流量数据划分为两种数据集:训练数据集与测试数据集㊂前者用于模型训练,后者用于检测模型性能㊂为检验预测有效性与精准度,将测试数据集输入已训练的模型中,得到如图7所示的交通流量数据预测结果㊂通过对比图7中的实际交通流量参数数据可以看出,笔者模型针对不同种类的异常数据,分别采取了不同的识别与处理策略,使数据更加完整,预测质量得到保证,神经网络模型根据误差函数不断训练网络,取得了最佳权重与阈值,因此,具有较高的拟合度与预测精度㊂图7　交通流量参数数据预测结果拟合示意图Fig.7　Drawing diagram of data prediction results of traffic flow parameters 图8　交通拥堵状况预测结果 Fig.8　Prediction result of traffic congestion 将标准化处理测得的交通流量参数数据,代入阈值计算式(8)中,得到图8所示的交通拥堵状况预测结果㊂从图8可以看出,该模型能有效预测出高速公路的交通拥堵状况,且预测精准度较为理想㊂这是因为笔者利用与交通流量息息相关的3个参数,从宏观角度全面㊁准确地反映了高速公路的交通流量特征,不同种类异常数据的识别与预处理阶段,为拥堵状况预测提供了高质量数据,将两个预测模型融合,有助于使两模型得到互补,经加权处理得到混合预测模型,并采用最优权值组合策略获取模型权值系数,302第2期黄承锋,等:基于改进GA 算法的高速公路交通拥堵状况预测402吉林大学学报(信息科学版)第40卷在一定程度上提升了模型的预测精准度㊂4　结　语随着国民经济水平迅猛上升,城市化进程快速发展,高速公路领域日益繁荣,但同时也大幅增加了拥堵的概率㊂交通拥堵不仅会增加车辆运营成本,而且还将导致交通事故频发,延长行车时长,加剧污染㊂为缓解高速道路拥堵问题,交通状况预测技术受重视程度越来越高,因此,笔者从宏观的交通流量数据入手,构建出基于改进遗传算法的交通拥堵状况预测模型㊂虽然目前研究已经取得了一定的成果,但仍需针对以下方面加以完善,进一步加强模型的预测精度与适用性:需根据公路的重要与非重要时段㊁优劣天气条件等影响因素,综合性地预测拥堵状况;应基于笔者预测模型,采用浏览器/服务器模式(B/S:Browser/Server)结构建立一种自动预测系统,使交通状况预测方法趋向更便捷㊁更人性化的方向发展㊂参考文献:[1]熊亭,戚湧,张伟斌.基于DCGRU⁃RF模型的路网短时交通流预测[J].计算机科学,2020,47(5):84⁃89. XIONG Ting,QI Yong,ZHANG Weibin.Short⁃Term Traffic Flow Prediction Based on DCGRU⁃RF Model for Road Network [J].Computer Science,2020,47(5):84⁃89.[2]戢晓峰,戈艺澄.基于深度学习的节假日高速公路交通流预测方法[J].系统仿真学报,2020,32(6):1164⁃1171. JI Xiaofeng,GE Yicheng.Holiday Highway Traffic Flow Prediction Method Based on Deep Learning[J].Journal of System Simulation,2020,32(6):1164⁃1171.[3]蔡延光,乐冰,蔡颢,等.暴雨天气下高速公路短时交通流预测[J].计算机工程,2020,46(6):34⁃39.CAI Yanguang,LE Bing,CAI Hao,et al.Short⁃Term Traffic Flow Forecast of Expressway under Heavy Rain[J].Computer Engineering,2020,46(6):34⁃39.[4]温惠英,张东冉,陆思园.GA⁃LSTM模型在高速公路交通流预测中的应用[J].哈尔滨工业大学学报,2019,51(9): 81⁃87,95.WEN Huiying,ZHANG Dongran,LU Siyuan.Application of GA⁃LSTM Model in Highway Traffic Flow Prediction[J]. Journal of Harbin Institute of Technology,2019,51(9):81⁃87,95.[5]苗旭,王忠宇,邹亚杰,等.改进的固定交通检测器缺失数据综合修复方法[J].同济大学学报(自然科学版),2019, 47(10):1477⁃1484.MIAO Xu,WANG Zhongyu,ZOU Yajie,et al.Improved Modification Method of Missing Data for Location⁃Specific Detector [J].Journal of Tongji University(Natural Science),2019,47(10):1477⁃1484.[6]全美娟,陈庆奎.移动GPU上基于轻量级卷积神经网络的道路拥堵检测方法[J].小型微型计算机系统,2019, 40(12):2667⁃2672.QUAN Meijuan,CHEN Qingkui.Road Congestion Detection Based on Lightweight Convolutional Neural Network under Mobile GPU[J].Journal of Chinese Computer Systems,2019,40(12):2667⁃2672.[7]杨志强,史丰收,黄俊达,等.基于土地利用的城轨新站点客流预测方法研究[J].都市快轨交通,2020, 33(2):70⁃74.YANG Zhiqiang,SHI Fengshou,HUANG Junda,et al.Passenger Flow Forecast of New Urban Rail Transit Stations Based on Land Use[J].Urban Rapid Rail Transit,2020,33(2):70⁃74.[8]符锌砂,胡嘉诚,何石坚.基于交通状况及行驶速度的高速公路换道时间研究[J].公路交通科技,2020,37(4): 133⁃139.FU Xinsha,HU Jiacheng,HE Shijian.Study on Expressway Lane⁃Changing Time Based on Traffic Condition and Driving Speed[J].Journal of Highway and Transportation Research and Development,2020,37(4):133⁃139.[9]YAN Chun,LI Meixuan,LIU Wei,et al.Improved Adaptive Genetic Algorithm for the Vehicle Insurance Fraud Identification Model Based on a BP Neural Network⁃Science Direct[J].Theoretical Computer Science,2020,817:12⁃23. [10]安磊,赵书良,武永亮,等.基于Recurrent Neural Networks的网约车供需预测方法[J].计算机应用研究,2019, 36(3):756⁃761.AN Lei,ZHAO Shuliang,WU Yongliang,et al.Prediction Method of Supply and Demand for Online Car based on RecurrentNeural Networks [J].Application Research of Computers,2019,36(3):756⁃761.[11]戚文正,李优华,刘杰,等.城市轨道交通专用齿轮润滑油台架试验研究[J].润滑油,2020,35(2):50⁃53.QI Wenzheng,LI Youhua,LIU Jie,et al.Study on the Bench Test of Special Gear Lubricating Oils for Urban Rail Transit [J].Lubricating Oil,2020,35(2):50⁃53.[12]梁礼明,陈明理,刘博文,等.基于图论的支持向量机核函数选择[J].计算机工程与设计,2019,40(5):1316⁃1321.LIANG Liming,CHEN Mingli,LIU Bowen,et al.Selection of SVM Kernel Function Based on Graph Theory [J].Computer Engineering and Design,2019,40(5):1316⁃1321.(责任编辑:刘东亮)502第2期黄承锋,等:基于改进GA 算法的高速公路交通拥堵状况预测。

喷射成形TiCp

2023 年第 43 卷航　空　材　料　学　报2023，Vol. 43第 2 期第 59 – 65 页JOURNAL OF AERONAUTICAL MATERIALS No.2 pp.59 – 65喷射成形TiC p/ZA35复合材料热挤压工艺的ANN优化和组织研究刘敬福1*, 叶建军1, 周祥春1, 庄伟彬1, 王一1,2（1．辽宁工程技术大学材料科学与工程学院，辽宁阜新 123000；2．沈阳工业大学材料科学与工程学院，沈阳 110870）摘要：采用人工神经网络（ANN）的方法，研究挤压比、挤压比压、挤压温度和挤压速率对喷射成形TiC p/ZA35复合材料力学性能的影响，建立了TiC p/ZA35复合材料热挤压的人工神经网络模型。

模型的输入参数为挤压比、挤压比压、挤压温度和挤压速率，输出参数为复合材料的抗拉强度。

该模型可以仿真TiC p/ZA35复合材料在不同热挤压工艺参数下的力学性能，也可以优化热挤压工艺参数，模型结果与实验结果误差小于1.8%，拟合率为0.986。

推荐热挤压工艺优化参数为：挤压比22，挤压比压415 MPa，挤压温度315 ℃，挤压速率8 mm•s−1，此工艺条件下复合材料的抗拉强度为486.7 MPa。

热挤压间接对复合材料进行了时效处理，材料晶内析出晶须状和颗粒状的MnAl6强化相。

弥散强化和位错强化作用使热挤压喷射沉积TiC p/ZA35复合材料较未挤压复合材料抗拉强度提高38.3%。

关键词：喷射成形TiC P/ZA35复合材料；热挤压；人工神经网络；优化；强化机制doi：10.11868/j.issn.1005-5053.2021.000200中图分类号：TG249.2 文献标识码：A 文章编号：1005-5053(2023)02-0059-07Heat extrusion processing ANN optimization and microstructure of sprayforming TiC P/ZA35 compositesLIU Jingfu1*, YE Jianjun1, ZHOU Xiangchun1, ZHUANG Weibin1, WANG Yi1,2（1. College of Material Science and Engineering , Liaoning Technical University, Fuxin 123000，Liaoning，China；2. College of Material Science and Engineering, Shenyang University of Technology, Shenyang 110870，China）Abstract: The effects of heat extrusion processing of spray forming TiC p/ZA35 composites on extrusion ratio, extrusion specific pressure, extrusion temperature and extrusion rate had been studied by artificial neural network (ANN). The artificial neural network model was created for heat extrusion processing. The input parameters of the ANN model were extrusion ratio, extrusion specific pressure, extrusion temperature and extrusion rate. The output of the ANN model was ultimate tensile strength. The model can be used for the prediction of properties of spray forming TiC p/ZA35 composites as functions of processing parameters. It can also be used for the optimization of the processing parameters. The ANN results are in good agreement with experimental phenomena, the biggest relative error and coincidence rate is less than 1.8% and 0.986. The optimized heat extrusion ratio, extrusion specific pressure, extrusion temperature and extrusion rate are 22415 MPa, 315 ℃ and 8 mm•s−1 respectively, and the tensile strength of spray forming TiC p/ZA35 composites is 486.7 MPa. The reinforcement phase MnAl6 whisker or particle is precipitated in the grains due to the indirect aging treatment of composites by hot extrusion. Dispersion strengthen and dislocation strengthen contribute a combination factor to increase the room temperature mechanical properties of the hot extruded TiC p/ZA35 composites, which is 38.3% higher than that of TiC p/ZA35 composites without heat extrusion.Key words: spray forming TiC p/ZA35 composites；heat extrusion；artificial neural network(ANN)；optimization；strengthen mechan-ism喷射成形制备的锌基复合材料坯体存在一定的孔隙率。

修正的基于广义Gamma语音模型语音增强算法

修正的基于广义Gamma语音模型语音增强算法赵改华;周彬;张雄伟【期刊名称】《计算机工程与应用》【年(卷),期】2014(000)018【摘要】广义Gamma模型是近年来新提出的一种语音分布模型，相对于传统的高斯或超高斯模型具有更好的普适性和灵活性，提出一种基于广义Gamma语音模型和语音存在概率修正的语音增强算法。

在假设语音和噪声的幅度谱系数分别服从广义Gamma分布和Gaussian分布的基础上，推导了语音信号对数谱的最小均方误差估计式；在该模型下进一步推导了语音存在概率，对最小均方误差估计进行修正。

仿真结果表明，与传统的短时谱估计算法相比，该算法不仅能够进一步提高增强语音的信噪比，而且可以有效减小增强语音的失真度，提高增强语音的主观感知质量。

%This paper presents a modified speech enhancement algorithm under signal presence probability. Generalized Gamma distribution priors are assumed for speech short-time spectral amplitudes, which is more flexible in capturing the statistical behavior of speech signals. It derives a Minimum Mean-Square Error(MMSE)estimator of the log-spectra am-plitude for speech signals, under the assumption of a generalized Gamma speech priors and additive Gaussian noise priors. Furthermore, modification under signal presence probability is obtained, which is estimated for each frequency bin and each frame consistent with the new model. The simulation results show that the proposed algorithm achieves better noise suppression and lower speech distortion compared to theconventional short-time spectral amplitude estimators, which are based on Gaussian and super-Gaussian speech model.【总页数】6页(P230-235)【作者】赵改华;周彬;张雄伟【作者单位】解放军理工大学指挥信息系统学院，南京 210007;解放军理工大学指挥信息系统学院，南京 210007;解放军理工大学指挥信息系统学院，南京210007【正文语种】中文【中图分类】TP912.3【相关文献】1.一种基于修正倒谱平滑技术改进的维纳滤波语音增强算法 [J], 李季碧;马永保;夏杰;刘金刚2.基于广义特征子空间快速估计的语音增强算法 [J], 朱义勇;姚富强;赵丽屏;朱勇刚3.基于Gamma语音模型的语音增强算法 [J], 邹霞;陈亮;张雄伟4.基于广义加权贝叶斯估计的语音增强算法研究 [J], 黄张翼;周翊;刘金刚;刘宏清5.一种基于广义奇异值分解的语音增强算法 [J], 曹梅双;曾庆宁;陈芙蓉因版权原因，仅展示原文概要，查看原文内容请购买。

一种基于标杆管理的优化算法

软件学报ISSN 1000-9825, CODEN RUXUEW E-mail: jos@Journal of Software,2014,25(5):953−969 [doi: 10.13328/ki.jos.004428] +86-10-62562563 ©中国科学院软件研究所版权所有. Tel/Fax:∗一种基于标杆管理的优化算法谢安世1,2, 于永达1, 黄思明21(清华大学公共管理学院,北京 100084)2(中国科学院科技政策与管理科学研究所,北京 100190)通讯作者: 谢安世, E-mail: shermanxas@163com摘要: 借鉴标杆管理理念,提出了一种基于标杆管理的优化算法(benchmarking-based optimization algorithm,简称BOA).根据标杆管理的核心价值观,设计了一套基于动态小生境的竞争性学习机制,针对常用的编码方案,设计出了具体可行的执行方法.种群内个体执行方向明确的主动学习式搜索,通过对标杆的模仿学习,能够快速搜索到解空间内的目标区域内,具有较好的智能性.此外,整个小生境种群系统通过自组织学习实现与环境的友好交互,较好地解决了保持种群的多样性的难题.分析了BOA算法与遗传算法等现代智能优化方法在搜索模式上的重要区别,并通过对比仿真实验,表明算法能够与环境进行稳定而友好的交互,表现出较强的鲁棒性,其搜索速度和寻优能力在实验中均有较好的表现.关键词: 标杆管理;优化算法;智能计算;搜索模式中图法分类号: TP181中文引用格式: 谢安世,于永达,黄思明.一种基于标杆管理的优化算法.软件学报,2014,25(5):953−969./1000-9825/4428.htm英文引用格式: Xie AS, Yu YD, Huang SM. Optimization algorithm based on benchmarking. Ruan Jian Xue Bao/Journal ofSoftware, 2014,25(5):953−969 (in Chinese)./1000-9825/4428.htmOptimization Algorithm Based on BenchmarkingXIE An-Shi1,2, YU Yong-Da1, HUANG Si-Ming21(School of Public Policy and Management, Tsinghua University, Beijing 100084, China)2(Institute of Policy and Management, The Chinese Academy of Sciences, Beijing 100190, China)Corresponding author: XIE An-Shi, E-mail: shermanxas@163comAbstract: Drawing on the benchmarking theory in the business management, a new search method, benchmarking-based optimizationalgorithm (BOA), is proposed in this paper. BOA provides a competitive learning mechanism based on dynamic niche according to thecore values of benchmarking. Through imitation and learning, all the individuals within a population are able to approach to the highyielding regions in the solution space and seek out the optimal solutions quickly. Further, the formidable problem of maintaining thediversity of population is effectively resolved through the self-organizing learning process of the niche system and its friendly interactionwith the environment. In this paper, the main differences between BLA and the existing intelligent optimization methods, sush as geneticalgorithm (GA), are analyzed. The comparative experiments show that BLA is robust and able to perform friendly interactive learningwith the environment, and its search speed and optimization ability is far superior to the existing intelligent optimization methods.Key words: benchmarking; optimization algorithm; intelligence computation; search mode优化问题的求解技术一般以数学为基础来获取各种工程问题的最优解或者满意解.在实践中,绝大多数问题都可以转换为最优化问题来求解.如0-1背包、组合优化、任务指派等许多存在于数学及工程领域中的NP难题,一般很难找到精确的数学解法,有的也不需要精确解,只要寻找到近似最优解即可.针对优化问题,寻求除∗收稿时间:2012-05-29; 定稿时间: 2013-05-07954 Journal of Software软件学报 V ol.25, No.5, May 2014基于严格数学逻辑的运筹学方法之外的方法,国内外相关学者做出了很大努力,且成果显著.在国外,1975年,美国的Holland教授提出了遗传算法.1982年,Kirkpatrick将模拟退火思想引入到了优化研究领域,提出了模拟退火算法(simulating annealing,简称SA).1983年,Creutz提出了微正则退火算法(micro- canonical annealing,简称MA)[1].它和SA都是模仿退火机制,但在退火过程中的状态转移不是用Metropolis准则,而是用一种确定性方式.目前,关于MA的研究不太多,已有的研究主要应用在图像处理领域.1989年,Moscato 首次提出了模拟文化进化过程的文化基因算法(memetic algorithm,简称MA)[2].1991年,意大利学者Dorigo根据蚂蚁觅食的群体行为,提出了蚁群算法的基本模型.1995年,美国心理学家Kennedy及其合作者Eberhart博士根据鸟类捕食行为模型提出了粒子群算法.1997年,Storn等人在遗传算法等进化算法思想上提出了差分进化算法.1999年,巴西的Castro最早在其论文中总结了人工免疫系统(artificial immune system,简称AIS)[3].后来,他基于克隆选择基本原理提出了著名的克隆选择算法(clonal selection algorithm,简称CSA)[4].进入新世纪以来,各类新式算法层出不穷,仅2000年就出现了3种新算法,如Eusuff等人提出了一种后启发式的混合蛙跳算法(shuffled frog leaping algorithm,简称SFLA)[5],此后有若干改进.同年,日本的Murase提出了一种模拟植物光合作用原理的光合算法(photosynthetic algorithm,简称PA)[6],在N皇后问题、有限元分析等问题上取得了较好的效果.在同一年,Zelinka和Lampinen基于物种合作竞争策略提出了自组织迁移算法(self-organizing migrating algorithm,简称SOMA)[7].2001年,Geem等人根据音乐演奏中的和声原理提出了一种和声搜索算法(harmony search,简称HS)[8].2002年,Passino提出了一种模拟人类大肠杆菌觅食行为的细菌觅食优化算法(bacteria foraging optimization algorithm,简称BFOA)[9].2004年,Nakrani首次根据蜜蜂的觅食原理提出了蜂群算法(bee algorithm,简称BA)[10].后来,Karabog根据蜂群的群居行为模型提出了系统的人工蜂群算法(artificial bee colony algorithm,简称ABC)[11].剑桥大学的Yang在2005年提出了一种模拟生化酶抑制和催化作用的酶算法(enzyme algorithm,简称EA)[12].同年,印度学者Krishnanand和Ghose根据萤火虫的求偶行为提出了萤火虫群优化算法(glowworm swarm optimization,简称GSO)[13].2007年,Mucherino等人根据猴子的攀树觅食模式提出了猴子搜索算法(monkey search,简称MS)[14].2008年,学者们又提出了3种新算法,如Havens等人提出了一种模拟蟑螂群居行为的蟑螂算法[15];同年,剑桥大学的Yang又提出了萤火虫算法(firefly algorithm,简称FA)[16];Simon根据生物物种迁移数学模型提出了一种生物地理学优化算法(biogeography-based optimization,简称BBO)[17].2010年1月,日本的Tero等人在《Science》上发表了使用粘菌(俗称鼻涕虫,英文名为Slime mold,一种粘菌门组织)设计连通东京及其附近城市的铁路网的研究成果,这项研究展现了仿生计算的若干特征[18].在国内,2002年,清华大学的谢晓峰基于人类智能中的社会认知理论(social cognitive theory)提出了一种社会认知优化算法(social cognitive optimization,简称SCO)[19].2003年,浙江大学的李晓磊提出了模拟鱼群行为模式的人工鱼群算法(artificial fishswarm algorithm,简称AFSA)[20].同年,周永华等人提出了一种模拟人口迁移机理的人口迁移算法(population migration algorithm,简称PMA)[21].2005年,李彤等人提出了基于植物向光性机理的模拟植物生长算法[22].2008年,最早有国内匿名学者利用费马原理模拟自然界光线折射现象,提出了光线寻优算法(light ray optimization,简称LRO),后来有一部分学者进行了应用和发展[23].2009年,马海平等人提出了一种基于物种迁移优化的进化算法[[24].2011年,国内相关学者提出了4种新颖的优化搜索算法,如:谢丽萍等人提出了一种基于拟态物理学方法的全局优化算法[25];曹炬等人受烟花爆炸现象的启发,提出了一种新的并行弥漫式搜索的优化算法——爆炸优化算法[26];谭世恒等人根据细胞膜的特性及其物质转动方式,提出了细胞膜优化算法(cell membrane optimization,简称CMO)[27];台湾地区的潘文超根据果蝇的觅食原理,提出了果蝇优化算法(fruit fly optimization algorithm,简称FOA)[28,29].这些方法一般受自然界规律和生物群体智能行为的启发,具有较新颖的设计思路,即将问题域中随机采样的样本点看作粒子,粒子具有适应值、速度和位置等属性.这些粒子在某种智能搜索策略的引导下,通过速度和位置的更新迭代逐步求得给定问题的最优解.虽然实践中这些方法都取得了令人满意的应用效果,但与基于严格数学逻辑的运筹学方法相比,目前尚无真正严格而完备的数理逻辑能够证明这些方法的全局收敛性.因此,可以将这些方法统一称为基于概率的智能算法.谢安世等:一种基于标杆管理的优化算法 955与模拟自然界生物活动规律的做法不同,本文借鉴企业管理领域中的标杆管理方法,提出一种基于标杆管理的寻优算法(benchmarking-based optimization algorithm,简称BOA).下面将介绍标杆管理的基本理念和基于这种理念的寻优方法,分析其不同于现有智能优化算法的特性,并通过实验验证这种新模式的有效性.1 标杆管理优化算法的基本内容标杆管理(benchmarking)一词来源于企业管理界,首次出现于施乐公司(Xerox corporation)的前身——位于美国康涅狄格州(Connecticut)罗切斯特市的哈洛伊德(Haloid)公司.标杆原意为“固定对象的标记,诸如用石柱来说明高出海平面之高度,作为调查中的参考点”,有基准之意,其构想是寻找学习的对象,以他们的既有成就为基准,透过合法管道学习,以“见贤思齐”的方式,达到改善自己经营品质的目的.换句话说,它是指一家公司可以就某一特定过程,将本身的绩效与其他公司的绩效相比较,然后学习其中绩效最佳者的做法,以达到提升自己绩效的目的,已经成为企业界通行的管理理念和管理工具.然而,标杆管理并不只是简单地向他人学习,而是包括4项基本原则,即标杆管理的4项核心价值观:全面品质观、流程观、衡量标准观和学习观.全面品质观是指达成顾客的全面性满意;流程观是指标杆管理涵盖学习对象的运营流程及组织内部的计划和运作流程;衡量标准观是指标杆管理须订出某些组织功能上共同绩效衡量标准,作为比较的依据;学习观则同时强调向他人学习与自我学习的精神.在管理学中,标杆管理既是一种态度也是一种行动,表现为一个持续的学习过程,不断地向标杆迈进,不断地创新与改善,不断地提升发展优势,不断地提高组织效能.通过Web of knowledge,Google Scholar 等检索,目前尚未发现国内外有学者将标杆管理应用于优化算法领域.标杆管理,简而言之,即找出与最佳个案的差距,并通过模仿学习,快速缩小这个差距乃至超越对手.本文提出的基于标杆管理的寻优算法(BOA),其总体框架是:整个生态系统(解空间)内分布着若干小生境种群,相当于全球市场上各大企业法人主体,种群内的个体相当于企业内部各部门员工;根据优化目的,以目标值大小为衡量标准,找出各小生境种群内的最佳个体(即局部最佳个体)和整个生态系统内的最佳个体(即全局最佳个体),相当于树立内部标杆和外部标杆;小生境种群中的个体有选择地进行标杆学习,此外还会进行自我学习;通过对标杆进行模仿学习,迅速超越学习对象,进而成为其他个体学习的对象.因而,BOA 是一个学习性竞争和竞争性学习的寻优模型.1.1 自组织学习设计在搜索学习的过程中,各小生境种群的自组织学习是这样进行的:种群内个体首先进行外部标杆学习,即向整个生态系统内具有最佳目标值的个体学习,参照该最佳个体来调整自己的搜索方向和搜索步长,即主动拉近与外部标杆的距离;如果目标值没有得到改善,则该个体进行内部标杆学习,即向该个体所在种群内具有最佳目标值的个体学习,参照本种群内最优个体来调整自己的搜索方法和搜索步长,即主动拉近与内部标杆的距离;如果目标值仍然没有得到改善,则该个体继续进行自我学习.此外,各小生境种群在学习过程中会相互交换最佳个体,即各小生境种群内个体的学习对象(内部标杆)不断发生改变.个体的上述3个学习行为并不是按顺序执行的,而是有选择地执行的,只有在执行前一个学习策略没有得到改善时,才会执行下一个学习策略.BOA 运用标杆学习的理念来获取候选解,如何实现这种学习理念,本文针对常用的浮点数(实数)编码方式和0-1编码方式(包括Binary Code & Gary Code),分别设计了相应的实现方法. 1.1.1 外部标杆学习设计设bestEX 是整个生态系统内具有最佳目标值(根据优化目的,取最大或最小,下同)的个体,即全局最佳个体,也即外部标杆,其对应的基因表达式为;best E G 种群P K 所属的第i 个个体i K X 所对应的基因表达式为,i K G 则个体iK X的外部学习率为max ():1min ():1i i K K Ki i K K K f x Grate Grate f f f x Grate Grate f f ⎧′=+−⎪⎨′=+−⎪⎩(1)956 Journal of Software 软件学报 V ol.25, No.5, May 2014其中,max f (x )表示求解目标函数的最大值,min f (x )表示求解目标函数的最小值,Grate ′表示外部学习率的初始值,i Kf 表示个体i KX 的目标值,Kf表示该个体所在种群P K 的平均目标值.可以看到:如果某个体的目标值越符合优化目的(根据优化目的,以求解目标函数的最大值为例),当其目标值大于所在种群的平均目标值时,其学习欲望会变得强烈,其外部学习率会增大.这样,生态系统内那些有前途的个体主动聚集到全局最佳个体所在的搜索邻域内,因而能起到协助搜索的作用.同理,当其目标值小于所在种群的平均目标值时,其学习欲望会降低,其外部学习率会减小,以等待下一个学习对象.下面针对常用的两种不同编码方案,设计了相应的学习方法:如果采用0-1编码方案,个体i K X 进行外部标杆学习,是指其基因表达式i K G 中的基因位值以iK Grate 的概率被best E G 中相应的基因位值所取代,即个体i K X 主动缩小与全局最佳个体bestE X 的海明距离(Hamming distance).如果采用浮点数编码方案,个体i K X 进行外部标杆学习,是指其对应的基因表达式i K G 以iK Grate 的概率按公式(2)进行更新,即个体i KX 主动缩小与全局最佳个体bestE X 的欧氏距离(Euclidean distance): ()i i best iK K E K G G G G λ=+− (2) 其中,λ∈[0,1],是iK X 进行外部标杆学习时的移动步长因子.实验结果表明:当λ与搜索空间的大小成正比例时,优化效果较好;当然,也可以引入目标值等要素,使λ在学习过程中动态自适应地改变,优化效果更好.这不是本文研究重点,留待以后再做深入研究.1.1.2 内部标杆学习设计设bestE X 是小生境种群P K 内具有最佳目标值的个体(即局部最佳个体,也即内部标杆),其对应的基因表达式为;best E G 该种群内第i 个个体i K X 所对应的基因表达式为,i K G 则个体i K X 的内部学习率iK Brate 为,,01:1:1iK k h iK k h Brate Brate HD Length Brate Brate ED Radius ⎧′−=−+⎪⎨′=−+⎪⎩编码实数编码 (3) 其中,Brate ′表示内部学习率的初始值;HD k ,h 为该个体与bestE X 的海明距离(Hamming distance),Length 为种群内个体的基因表达式编码长度;ED k ,h 为该个体与bestEX 的欧氏距离(Euclidean distance),即2,1()nbest k h i i ED x x −∑Radius 为搜索空间的直径,即21()ni i Radius b a =−∑其中,x i 是个体基因表达式中的第i 维分量,且x i ∈[a i ,b i ].由公式(3)可知,当小生境种群中某个体与该种群中最佳个体的海明距离或欧氏距离较小时,其学习欲望会自动升高,从而迅速聚拢到该局部最佳个体的搜索邻域内,以协助其进行密集搜索.与外部标杆学习类似,当采用0-1编码方案时,个体i KX 进行内部标杆学习,是指i K G 中与bestE G 相异的基因位值以i KBrate 的概率被best E G 中相应的基因位值替换,即个体i K X 主动缩小与局部最佳个体bestE X 的海明距离;当采用浮点数编码方案时,个体i KX 进行内部标杆学习,指i K G 以i K Brate 的概率按公式(4)进行更新,即个体iK X 主动缩小与局部最佳个体bestEX 的欧氏距离. ()i i best iK K K K G G G G λ=+− (4)同理,λ∈[0,1],是iK X 进行内部标杆管理时的移动步长因子.这里的外部标杆学习和内部标杆学习都是缩小个体之间的海明(或欧氏)距离,看似相同,但实际上两者却有很大的差异,这也是本算法的核心思想之一:这种主动缩小与最优个体距离的行为,既有利于种群进行密集搜索,从而形成群集效应,快速搜索到全局最优解,同时也是保持种群多样性的最好方式——因为每个个体的学习对象是不断地动态变化的,因此生态系统内个体群集的层次也是动态变化的(详见后文算法分析和仿真实验部分内容).1.1.3 自我学习设计如果采用0-1编码方案,则个体的自我学习是指进行对偶映射(dual mapping)[30],即该个体的基因表达式中每个基因位都执行对偶映射,即0↔1,如图1所示.谢安世等:一种基于标杆管理的优化算法 957Fig.1 Dual mapping based on 0-1 encoding图1 基于0-1编码的对偶映射如果采用浮点数编码方案,则个体的自我学习是指进行类逻辑斯蒂混沌映射(logistic chaos mapping),即个体i KX 的基因表达式iK G 按公式(5)进行更新.利用混沌运动对初始状态的敏感性和非重复的遍历性,使个体及时跳出当前所在区域,以便对解空间的其他区域展开搜索.设121[,,...,,]iKn n G x x x x −=,x i ∈[a i ,b i ]:(0)(1)()(1())()()()[2,4], 1,2,3,...,i i i i i i i i ii i i i x a b a t t t x t a t b a i n λλδλλλδ−⎧=⎪−⎪⎪+=−⎨⎪=+−⎪⎪∈=⎩ (5) 设i K X 是隶属小生境种群P K 的个体,则个体iK X 的自我学习率为max ():min ():i i K K Ki i K K K f x Srate Srate f f f x Srate Srate f f ⎧′=×⎪⎨′=×⎪⎩(6) 其中,max f (x )表示求解目标函数的最大值,min f (x )表示求解目标函数的最小值,Srate ′表示自我学习率的初始值,i Kf 表示个体i KX 的目标值,Kf表示该个体所在种群P K 的平均目标值.由公式(6)可以看到:当优化的目的是实现目标最大化时,如果个体的目标值小于其所在种群的平均目标值,则其学习率将迅速增大,其学习欲望会迅速提高,于是有较多机会得到其对偶个体,以提高其自身的目标值;但如果该个体的目标值已经比其所在种群的平均目标值大了,则其学习欲望会快速消退,则其学习率将快速降低,这样可保护优良基因以免遭破坏.同理,当优化的目的是实现目标最小化时,其自学习欲望也会相应地自动调整,以符合优化目的. 1.2 BOA 算法伪代码设E ={P 1,P 2,…,P np }是包含np 个小生境种群的生态系统,N i 表示种群P i 中个体的数目,j i P 表示种群P i 中第j 个个体,best iP 表示种群P i 中具有最佳目标值的个体,if表示种群P i 在当前代的平均目标值,Ef 表示整个生态系统当前代的平均目标值,P best 表示当前代整个生态系统中具有最佳目标值的个体,max gen 表示最大迭代次数.则BOA 算法的伪代码可表示如下:1. 种群等相关参数初始化2. for gen =1:max gen , do(a) for i =1:np , doi. 评估种群P i 中每个个体的目标值ii. 计算种群P i 的平均目标值ifiii. 找出并记录种群P i 中具有最佳目标值的个体best i P (树立内部标杆) (b) 找出并记录整个生态系统中具有最佳目标值的个体P best (树立外部标杆) (c) 记录并更新整个生态系统中具有最佳目标值的个体,即全局最优解(d) 评估整个生态系统在当前代的平均目标值()Eif fnp =∑ (e) for i =1:np , doi. 种群p i 中个体j i P 进行外部标杆学习ii. 如果个体j i P 的目标值没有得到改善,则进行内部标杆学习 iii. 如果个体j i P 的目标值仍然没有得到改善,则进行自我学习1 0 1 0 0 1 0 1 1 0 0 0 1 0 00 1 0 1 1 0 1 0 0 1 1 1 0 1 1对偶基因表达式:原基因表达式:958 Journal of Software 软件学报 V ol.25, No.5, May 2014(f) if 与上一代相比,平均目标值Ef 没有得到改善或最佳个体P best 没有发生改变 do 各小生境种群之间相互交换具有最佳目标值的个体(即各种群重新树立新的内部标杆)3. 对全局最佳个体解码并输出全局最优解2 仿真实验目前,在数值优化领域已经出现了多达几十个经典测试函数,这里摘取15个求解最小值的测试函数.考虑到实际优化问题中,最优解与所在搜索空间的相对位置是非确定性的,因此,借鉴文献[31]的思想设计3种位置关系:最优解恰好在搜索空间边界上(BOUNDARY)、最优解靠近搜索空间边界(CLOSE)、最优解在搜索空间正中心位置(CENTRE),具体设置见表1.Table 1 Testbed functions 表1 标准测试函数列表搜索空间名称函数Boundary Close Centre 最优解的位置全局最优解目标值精度Sphere 211()Di i f x x ==∑[0,200]D [−5,100]D[−100,100]D0.0D 0 0.01 Griewank 22111()cos 14000D D i i i i x f x x i ==⎛⎞=−+⎜⎟⎝⎠∑∏ [0,1200]D [−5,1195]D [−600,600]D 0.0D 0 0.01 Rastrigin 231()(10cos(2)10)D i i i f x x x ==−π+∑ [0,10]D[−1.12,9.12]D [−5.12,5.12]D0.0D 0 0.01 Rosenbroke 1222411()(100()(1))D i i i i f x x x x −+==−+−∑[0,60]D [0,60]D [−30,30]D 1.0D 0 0.01 Ackley 211110.2cos(2)5()20eee 20DDi i i i x x D D f x ==−π∑∑=−−++[0,64]D [−5,59]D [−32,32]D 0.0D 0 0.01 Schwefel ()()61()418.9829sin||Di i i f x D x x ==−∑[420.9687,500]D [415,500]D [340,500]D 420.9687D 0 0.01 Step 271()(|0.5|)Di i f x x ==+∑[0,200]D [−5,195]D [−100,100]D −0.5D0 0.01Schwefel’sP2.22 811()||||DDi i i i f x x x ===+∑∏[0,20]D [−1,19]D [−10,10]D 0.0D 0 0.01 Quadric 2911()Di j i j f x x ==⎛⎞=⎜⎟⎜⎟⎝⎠∑∑[0,200]D [−5,195]D[−100,100]D0.0D 0 0.01 Quadric Noise 410()[0,1)Di if x ix rand =+∑[0,2.56]D [0.28,2.28]D [−1.28,1.28]D 0.0D 0 0.01 Schwefel’s P2.21 11()max {||,1}i i f x x i D =≤≤[0,200]D [−5,195]D[−100,100]D0.0D 0 0.01 Unnamed2221211()sin 501DDi i i i f x xx ⎛⎞⎛⎞⎜⎟⎜⎟⎜⎟⎜⎟⎜⎟⎜⎟⎜⎟⎜⎟==⎝⎠⎝⎠=+∑∑[0,10]D[−1.12, 9.12]D [−5.12,5.12]D0.0D 0 0.01Schaffer22113221sin 0.5()0.510.001Di i Di i x f x x ==−=+⎡⎤+⎢⎥⎣⎦∑∑[0,200]D[−5,195]D [−100,100]D0.0D 0 0.01谢安世等:一种基于标杆管理的优化算法959Table 1 Tested functions (Continued)表1标准测试函数列表(续)为了验证BOA算法的性能,与近年来出现的几种有代表的算法进行对比,共设计了两组实验:一组测试算法的全局收敛能力,一组用来测试算法的综合性能.分别使用近年来出现的几种典型的群体智能优化方法,如人工蜂群算法ABC[11]、人工鱼群算法AFSA[20]、改进的遗传算法SAMGA[32]、改进的粒子群优化算法SzAPSO[32]与本文所提出的标杆学习算法BOA进行测试比较.2.1 全局收敛能力测试算法的控制参数对算法性能影响非常大,各算法涉及到的控制参数众多且不尽相同,为了保证实验测试的客观、公正,除了相同的软硬件实验平台以外(硬件CPU:AMD Athlon(tm) 64 X2 Dual Core Processor 3600+,1.91GHz; DDR:667MHz,1024MB,软件Matlab7.12),在此实验中,各算法统一采用浮点数编码方案,种群中个体的总数目统一设为100,进行一次搜索时迭代次数统一设为3 000,每个实验重复执行30次;各算法的其他重要控制参数分别按照各自参考文献中提出的最佳方案配置.为了缩短篇幅,此实验只测试15个函数的最优解在搜索空间正中心位置的情形.另外,为求客观、公正,各算法在初始化时所有个体皆被置于边界上同一点,以消除初始化时个体的位置优劣对算法搜索性能的影响,同时,搜索空间维数设定为50维.图2是由5种算法分别在15个测试函数上实验30次,取其中表现最好的一次和表现最差的一次,共两次实验结果平均值绘制的收敛曲线,其中,横轴是实验时的迭代次数,纵轴是对应的目标函数值取自然对数之后的函数值.观察其中各图可以发现:5种算法都能搜索到各测试函数的全局最优解附近,在决策变量范围不大时,5种算法搜索结果的精度相差不大;但当决策变量的变化范围较大时,5种算法搜索结果的精度有较大差别, SzAPSO和BOA表现较为突出,SzAPSO有时比BOA表现更优异,如f4,f7,f8和f11.960 Journal of Software 软件学报 V ol.25, No.5, May 2014f 1: Sphere f 2: Griewank f 3: Rastriginf 4: Rosenbrock f 5: Ackley f 6: Schwefelf 7: Step f 8: Schwefel’sP2.22 f 9: Quadricf 10: QuadricNoise f 11: Schwefel’sP2.21 f 12: Unnamedf 13: Schaffer f 14: GeneralizedPenalized01 f 15: GeneralizedPenalized02Fig.2 Test result of the 15 functions for the 5 algorithms while D =50图2 D =50时,5种算法在15个标准测试函数上的收敛曲线迭代次数目标函数值−−−−迭代次数−−−3000迭代次数 −−−200010000 迭代次数目标函数值−−ABC AFSA SAMGA SzAPSO OAB迭代次数−−−−ABC AFSA SAMGA SzAPSO OAB迭代次数−−ABC AFSA SAMGA SzAPSO OAB迭代次数目标函数值−−ABCAFSA SAMGA SzAPSO OAB 15迭代次数−−−ABC AFSA SAMGA SzAPSO OAB20迭代次数目标函数值−−−ABC AFSA SAMGA SzAPSO OAB 迭代次数目标函数值QuadricNoise 的寻优过程−−−−ABC AFSASAMGA SzAPSO OAB 迭代次数Schwefel’sP2.21的寻优过程−−−ABC AFSA SAMGA SzAPSO OAB 迭代次数−−−ABCAFSASAMGASzAPSOOAB迭代次数ABCAFSASAMGASzAPSO OAB −−−迭代次数目标函数值−−−−ABCAFSA SAMGA SzAPSO OAB迭代次数GeneralizedPenalized01的寻优过程−−−ABC AFSA SAMGA SzAPSO OAB谢安世等:一种基于标杆管理的优化算法9612.2 算法综合性能比较在此实验中,各算法的参数设置与上一节实验相同.表2是问题最优解与所在搜索空间处于3种相对位置、搜索空间维数为50时,5种算法在以上15个测试函数上分别实验30次的平均目标值和方差.结合图2,由表2中的数据可以看到最解位置对算法性能的影响:ABC,AFSA和SAMGA对最优解的位置敏感程度不一,对有些问题,最优解在边界时优化效果较好,有时在中心位置时优化效果更好,从具体数据看,关系较为复杂;而SzAPSO和BOA总体上对最优解位置不敏感,无论最优解在何处,其收敛速度和收敛精度均表现较好,具有较强的鲁棒性.Table 2Mean and variance of test result of 5 algorithms for 3 situations while D=50表2当D=50时,5种算法在3种位置关系下收敛精度测试结果的均值和方差f1: Sphere f2: Griewank f3: RastriginM SD M SD M SDABC 4.31E−02 8.53E−01 4.17E−017.80E−01 6.35E+00 5.47E−02AFSA 9.11E−01 6.22E−01 4.97E−01 3.90E−01 3.53E+00 2.96E−02 BOUNDARYSAMGA8.18E−02 3.51E−019.03E−01 2.42E−018.21E+00 7.45E−03SzAPSO 2.64E−03 5.13E−029.45E−10 4.04E−03 1.54E−08 1.89E−04OAB 1.46E−07 4.02E−02 4.91E−109.65E−04 4.30E−10 6.87E−04ABC 1.69E−02 1.84E−019.29E−02 6.44E−01 2.08E−03 3.11E−02AFSA 6.49E−01 3.68E−017.76E−01 3.79E−02 3.01E+00 9.23E−02CLOSESAMGA 3.17E−02 6.26E−02 4.87E−028.12E−02 4.71E+00 4.30E−02SzAPSO 6.48E−03 7.80E−01 4.36E−10 5.33E−04 2.30E−09 1.85E−05OAB 4.51E−07 8.11E−03 4.47E−12 3.51E−038.44E−10 9.05E−05ABC 2.35E−03 5.24E−01 4.84E−01 3.53E−018.85E+00 1.20E−02AFSA 6.50E−01 3.88E−01 3.25E+00 5.42E+007.55E+00 1.21E−01CENTRESAMGA7.70E−01 9.61E−01 6.65E−01 3.23E−018.03E+00 1.20E−02SzAPSO 1.55E−03 3.30E−038.20E−12 5.48E−04 3.62E−11 6.57E−06OAB 1.01E−08 5.80E−04 2.55E−129.95E−06 3.12E−11 4.54E−07f4: Rosenbrock f5: Ackley f6: SchwefelM SD M SD M SDABC 1.36E+01 7.60E+00 4.89E−02 1.32E−01 5.95E−02 8.55E−01AFSA 8.69E+01 2.40E+01 3.38E−039.42E−02 6.22E−02 2.62E−01 BOUNDARYSAMGA 5.97E−01 2.33E−019.00E−049.56E−02 6.03E−02 8.01E−01SzAPSO 5.50E−04 1.84E−04 3.69E−09 5.75E−047.11E−04 2.92E−02OAB 1.45E−04 2.40E−03 1.11E−09 5.98E−04 2.22E−03 9.29E−02ABC 3.06E+01 9.39E−01 1.95E−039.80E−02 1.17E−02 7.30E−01AFSA 5.09E+00 8.76E−01 2.26E−02 4.39E−02 2.97E−02 4.89E−01CLOSESAMGA 5.11E−02 5.50E−01 1.71E−03 1.11E−01 3.19E−02 5.79E−02SzAPSO8.18E−05 6.22E−04 2.28E−09 2.58E−04 4.24E−03 2.37E−02OAB 7.95E−03 5.87E−02 4.36E−10 4.09E−03 5.08E−03 4.59E−02ABC 3.88E+02 5.64E+03 3.24E−038.60E−02 2.67E−02 5.20E−01AFSA 2.35E+02 1.20E+02 1.74E−02 4.50E−01 3.25E−02 3.84E−01CENTRESAMGA 1.07E−02 8.10E−01 1.03E−04 2.06E+00 2.25E−02 6.56E−01SzAPSO 1.22E−05 2.51E−03 2.20E−11 5.40E−04 3.51E−03 1.52E−02OAB 2.56E−03 3.65E−01 3.32E−12 6.24E−03 1.90E−03 6.52E−03f7: Step f8: Schwefel’sP2.22f9: QuadricM SD M SD M SDABC 9.63E−01 3.77E−02 1.07E−02 3.05E−01 1.83E+01 5.96E+00AFSA 5.47E+01 8.85E−01 6.54E−027.44E−01 2.40E+01 6.82E+00 BOUNDARYSAMGA 5.21E+00 9.13E−01 4.94E−03 5.00E−028.87E+01 4.24E−01SzAPSO 2.32E−05 7.96E−027.79E−10 4.80E−03 2.87E−02 7.14E−03OAB 4.89E−03 9.87E−037.15E−129.05E−05 4.90E−04 5.22E−03ABC 6.24E+00 2.62E−019.04E−02 6.10E−02 1.68E+01 9.67E−02AFSA 6.79E+01 3.35E+008.91E−02 6.18E−039.79E+00 8.18E−01CLOSESAMGA 3.96E+00 6.80E+00 3.34E−038.59E−027.13E+01 8.18E−01SzAPSO 3.67E−05 1.37E−04 6.99E−108.05E−04 5.00E−02 7.22E−03OAB 9.88E−03 7.21E−02 1.98E−11 5.77E−03 4.71E−04 1.50E−04962 Journal of Software软件学报 V ol.25, No.5, May 2014Table 2Mean and variance of test result of 5 algorithms for 3 situations while D=50 (Continued)表2当D=50时,5种算法在3种位置关系下收敛精度测试结果的均值和方差(续)f7: Step f8: Schwefel’sP2.22f9: QuadricM SD M SD M SDABC 1.13E+00 3.87E−01 1.85E−02 2.45E−01 2.02E+02 1.10E+01AFSA 1.65E+02 3.50E+01 5.75E−03 5.42E−02 1.52E+01 8.00E+00 CENTRESAMGA 2.35E+00 1.08E+01 1.22E−03 8.89E−02 1.98E+02 1.14E+01SzAPSO7.54E-04 6.24E−03 6.45E−12 5.84E−03 2.00E−02 6.55E−02OAB 1.08E−02 1.35E−02 2.09E−13 1.02E−05 8.80E−05 5.45E−03f10: QuadricNoise f11: Schwefel’sP2.21f12: UnnamedM SD M SD M SDABC 6.60E−02 1.73E−02 8.15E−02 1.58E−01 6.56E−02 7.06E−02AFSA 5.19E−02 3.91E−01 9.06E−01 9.71E−01 3.57E−01 3.18E−02 BOUNDARYSAMGA9.73E−038.31E−01 1.27E−02 9.57E−02 8.49E−02 2.77E−02SzAPSO 6.49E−048.03E−02 9.13E−07 4.85E−03 9.34E−05 4.62E−04OAB 8.00E−05 6.05E−02 6.32E−05 8.00E−02 6.79E−07 9.71E−05ABC 4.54E−02 3.99E−02 9.75E−02 1.42E−01 7.58E−02 8.23E−02AFSA 4.32E−03 5.27E−01 2.78E−02 4.22E−02 7.43E−02 6.95E−02 CLOSESAMGA8.25E−02 1.68E−01 5.47E−02 9.16E−02 3.92E−02 3.17E−02SzAPSO8.35E−04 6.57E−02 9.58E−07 7.92E−03 6.55E−04 9.50E−03OAB 1.33E−03 6.28E−03 9.65E−04 9.59E−03 1.71E−03 3.44E−04ABC 1.24E−02 2.54E−02 4.20E−01 8.40E−01 2.65E−01 6.24E−01AFSA 1.34E−02 2.15E−01 5.45E−01 1.01E+00 2.98E−01 8.54E−01 CENTRESAMGA 5.54E−02 3.22E−01 3.07E−01 1.07E+00 1.89E−01 5.21E−01SzAPSO7.50E−04 6.86E−02 1.04E−06 6.71E−02 5.33E−05 2.54E−03OAB 1.24E−04 4.44E−02 2.56E−04 8.46E−03 2.55E−06 8.87E−03f13: Schaffer f14: GeneralizedPenalized01f15: GeneralizedPenalized02M SD M SD M SDABC 4.39E−01 2.76E−01 7.51E+00 8.41E−02 3.52E+00 7.59E−01AFSA 3.82E−01 6.80E−01 2.55E+00 2.54E−02 8.31E−01 5.40E−02 BOUNDARYSAMGA7.66E−02 6.55E−02 5.06E−02 8.14E−01 5.85E−03 5.31E−03SzAPSO 2.80E−06 1.63E−04 6.99E−08 2.44E−03 5.50E−05 7.79E−04OAB 1.87E−08 1.19E−03 8.91E−07 9.29E−04 9.17E−06 9.34E−03ABC 4.90E−01 4.98E−02 9.59E+00 3.50E−02 2.86E+00 1.30E−01AFSA 4.46E−019.60E−01 5.47E+00 1.97E−03 7.57E−02 5.69E−02 CLOSESAMGA 6.46E−01 3.40E−02 1.39E−02 2.51E−02 7.54E−02 4.69E−02SzAPSO7.09E−07 5.85E−04 1.49E−06 6.16E−03 3.80E−06 1.19E−04OAB 7.55E−07 2.24E−04 2.58E−07 4.73E−04 5.68E−05 3.37E−03ABC 3.56E−01 5.64E−01 1.60E+01 6.41E−02 1.47E+01 2.45E−01AFSA 2.22E−01 4.76E−01 1.20E+01 5.41E−02 2.46E−03 8.99E−02 CENTRESAMGA 4.26E−01 2.05E−01 5.66E−03 8.72E−02 8.81E−02 2.04E−01SzAPSO8.89E−07 6.52E−03 7.57E−07 5.24E−03 4.15E−06 2.45E−03OAB 5.43E−088.82E−03 2.85E−06 7.65E−03 6.56E−07 2.36E−03 为了测试各算法的可靠性,每种算法在15个函数上分别实验30次,设定的终止条件是达到如表1中所示的指定精度或达到最大迭代次数5×104.表3是5种算法在指定精度下对表1中15个函数的测试结果的均值.由表3可以看到:BOA对15个函数在3种位置关系下,大部分能以100%的概率成功收敛到指定目标精度,因其收敛速度较快,因此所记录的成功达到指定目标精度时的迭代次数和CPU运行时间相对ABC,AFSA和SAMGA少了很多;SzAPSO也有类似的结果,这是由于其空间缩放和吸引子策略的成功应用所致.在上述对比实验中产生了大量的实验数据,有时并不利于用来分析对比算法的性能,因此采用配对双侧T 检验方法对所得的实验数据进行量化统计分析,以便客观评价BOA算法与其他各算法之间的性能差别程度.在此次检验中,显著水平α设为0.05,自由度V=30−1=29,查表可知,t0.05(29)=1.699,根据公式容易计算出统计量t值,其计算结果见表4.根据双侧T检验规则,当|t|<t0.05(29)时,表明BOA算法与所配对算法无显著性差异,表4最后一列列出了在15个测试函数中,优化结果与BOA有显著差异的个数.由表中数据可知:BOA与ABC,AFSA及SAMGA这3种算法对绝大多数测试函数在优化性能上有显著差异,而与SzAPSO在优化性能上差异并不显著,这与前面实验结果的若干直观印象是一致的.。

考虑多因素的一步矿房采场跨度优化

考虑多因素的一步矿房采场跨度优化东龙宾;王少泉;张华;周育;何祥【摘要】合理采场结构参数的确定是控制地压危害、实现矿体安全高效开采的重要措施之一.不同采场结构参数的优劣由多个指标判定,而各项参数的指标优劣趋势往往不统一.针对张庄铁矿的生产实际,通过FLAC3D数值模拟,获取能量释放率、屈服率以及顶板拉应力区率,结合胶结充填率,利用层次分析法和模糊综合评判法获取综合评价指标,对一步矿房采场跨度优劣进行评价,再利用遗传算法优化出最优的采场跨度,为非等宽采场结构参数的选择提供理论基础.研究成果和技术方法可以运用于其他采用充填法开采的大型深部金属矿山.【期刊名称】《金属矿山》【年(卷),期】2018(000)010【总页数】5页(P31-35)【关键词】采场跨度;数值模拟;层次分析法;模糊综合评价;遗传算法【作者】东龙宾;王少泉;张华;周育;何祥【作者单位】中冶北方(大连)工程技术有限公司,辽宁大连116000;中冶北方(大连)工程技术有限公司,辽宁大连116000;安徽马钢张庄矿业有限责任公司,安徽六安237400;中冶北方(大连)工程技术有限公司,辽宁大连116000;中冶北方(大连)工程技术有限公司,辽宁大连116000【正文语种】中文【中图分类】TD313;TD823采用充填法进行开采可有效控制地压、防止地表沉降和预防岩爆。

目前越来越多的大型深部金属矿山采用充填法进行开采，而采用充填法开采时，如何确定合理的采场结构参数，一直是现场工程技术人员关注的热点，国内外学者也对采场结构参数优化进行了多方面的研究。

许多研究者采用层次分析法与模糊数学综合评价方法对采场结构参数进行评价［1-5］，但未涉及遗传算法的研究。

例如李洁慧等［1］针对影响采场结构参数选择的复杂因素，提出了一种基于层次分析法和模糊数学的综合评价方法。

杨建［4］根据残矿资源技术分类和技术条件，运用模糊数学和层次分析法的基本理论，对4种备选开采模式方案进行综合评判优选。

智能优化方法

15
二.传统优化方法的局限性（2）
3. 最优性达到的条件太苛刻——目标函数为凸，可行域为凸 4. 在非双凸条件下，没有跳出局部最优解的能力
16
三.实际问题中对最优化方法的要求（1）
1. 对问题的描述要宽松（目标和约束函数）— —可以用一段程序来描述(程序中带判断、循环)，函数可以非连续、非凸、非可微、非显式 2. 并不苛求最优解——通常满意解、理想解，甚至可行解就可以
选初始解
停止判据
Y
停止
N
改进解
11
一.传统优化方法的基本步骤—三步曲（2） 2. 停止判据——最优性检验
T 1 T C B N C ① LP：检验数 B N
C C B | C N
T
A B | N
当∏≥0时有可能减小
②
NLP： f ( x) 0
12
一.传统优化方法的基本步骤—三步曲（3） 3. 向改进方向移动——改进解
（从不同的随机种子出发）达优率——客观公正
与良心！
24
六.优化领域的新进展
1. 随着人们关注的系统越来越复杂，最优化技术也相应不断发展 2. 最优化方法的发展
1940s－1970s：数学规划阶段——目标和约束是解析函数 1970s—2000s：智能优化阶段——目标和约束可以放宽为含有判断逻辑的计算机程序 2000s之后：基于仿真优化阶段
25
七.学习这门课需要具备的基础
1. 信心、决心、热情 2. 良好的外语能力
3. 较为熟练的计算机编程能力
4. 一定的数学基础
26
1995年 Dorigo提出蚁群算法(ACO)
20
四.智能优化算法的产生与发展（3）

基于高斯过程优化与FLAC3D数值计算的岩体力学参数反分析方法

基于高斯过程优化与FLAC3D数值计算的岩体力学参数反分析方法龚杨凯; 卢翠芳; 黄杰; 苏国韶【期刊名称】《《广西大学学报（自然科学版）》》【年(卷),期】2019(044)004【总页数】6页(P1038-1043)【关键词】岩体力学参数; 地下工程; 反分析; 高斯过程优化【作者】龚杨凯; 卢翠芳; 黄杰; 苏国韶【作者单位】广西大学土木建筑工程学院广西南宁530004; 工程防灾与结构安全教育部重点实验室广西南宁 530004【正文语种】中文【中图分类】TU450 引言在地下工程稳定性分析中，由于岩体介质的高度复杂性和显著的尺度效应，室内及现场的岩石力学试验往往不能够合理地获得岩体力学参数，如何合理地确定岩体力学参数一直是一个比较棘手的现实问题[1]。

利用岩体开挖过程监测到的位移或破坏区等实测信息进行反分析，进而推求定岩体参数的岩体参数分析方法是解决上述问题的有效途径。

但是，对于复杂岩体工程，反分析的目标优化函数常具有表达式未知、高度非线性、多极值等特征[2]，传统优化方法难以获取全局最优解。

近年来，学者们采用的遗传算法(GA)、粒子群算法(PSO)、人工蜂群算法(ABC)等随机全局优化算法进行反分析，取得了良好成效[3-5]。

但对于洞室群等大型岩体工程的参数反分析，为保证数值计算的精度，计算单元致密且数量庞大，导致单次数值计算的耗时较大，若采用随机全局优化算法进行反分析，常需要成千上万次地进行数值计算，因计算耗时巨大导致所谓的高计算代价问题。

将机器学习模型与随机全局优化算法相结合是解决高计算代价问题的有效途径，利用机器学习模型替代数值计算模型，并建立岩体参数与数值计算结果的非线性映射关系，可显著提高计算效率，其中，基于神经网络—遗传算法(ANN-GA)以及支持向量机—遗传算法(SVM-GA)的反分析方法应用较为广泛[1、6-11]，但这些方法尚存在着神经网络不适用于小样本、合理的网络结构与超参数难以确定、易限于局部最优解等局限性问题。

1、下载文档前请自行甄别文档内容的完整性，平台不提供额外的编辑、内容补充、找答案等附加服务。
2、"仅部分预览"的文档,不可在线预览部分如存在完整性等问题,可反馈申请退款(可完整预览的文档不适用该条件!)。
3、如文档侵犯您的权益，请联系客服反馈,我们会尽快为您处理(人工客服工作时间：9:00-18:30)。

ANN-GA based optimization of a high ash coal-ﬁred supercritical power plantM.V.J.J.Suresh,K.S.Reddy ⇑,Ajit Kumar KolarHeat Transfer and Thermal Power Laboratory,Department of Mechanical Engineering,Indian Institute of Technology Madras,Chennai 600036,Indiaa r t i c l e i n f o Article history:Received 27February 2011Received in revised form 12June 2011Accepted 18June 2011Available online 20July 2011Keywords:Artiﬁcial neural network Genetic algorithmSupercritical power plant High ash coal Energy Exergya b s t r a c tThe efﬁciency of coal-ﬁred power plant depends on various operating parameters such as main steam/reheat steam pressures and temperatures,turbine extraction pressures,and excess air ratio for a given fuel.However,simultaneous optimization of all these operating parameters to achieve the maximum plant efﬁ-ciency is a challenging task.This study deals with the coupled ANN and GA based (neuro-genetic)optimi-zation of a high ash coal-ﬁred supercritical power plant in Indian climatic condition to determine the maximum possible plant efﬁciency.The power plant simulation data obtained from a ﬂow-sheet program,‘‘Cycle-Tempo’’is used to train the artiﬁcial neural network (ANN)to predict the energy input through fuel (coal).The optimum set of various operating parameters that result in the minimum energy input to the power plant is then determined by coupling the trained ANN model as a ﬁtness function with the genetic algorithm (GA).A unit size of 800MWe currently under development in India is considered to carry out the thermodynamic analysis based on energy and exergy.Apart from optimizing the design parameters,the developed model can also be used for on-line optimization when quick response is required .Further-more,the effect of various coals on the thermodynamic performance of the optimized power plant is also determined.Ó2011Elsevier Ltd.All rights reserved.1.IntroductionElectricity drives the economic growth of a developing country like India,which is witnessing a robust economic growth rate of 8%and above.India has huge coal reserves –about 7.1%of the world’s total [1]and thus,coal-ﬁred power plants contribute to about 70%of the total power generation [2].Currently all the coal-ﬁred power plants in India operate on subcritical (SubC)steam parameters with the exception of two recent plants that use supercritical (SupC)steam parameters.Most of the coal-ﬁred power plants that use indigenous high ash (HA)($45%)coal have plant efﬁciencies (net)less than 35%(based on HHV of coal).Rapid depletion of fossil fuel resources and consequent increase in CO 2emissions necessi-tate installation and operation of more efﬁcient power plants.The ﬁrst coal-ﬁred SupC power plant recently commissioned by National Thermal Power Corporation (NTPC)in India has a gross power output of 660MWe with steam parameters of 242.2bar/537°C/565°C [3].However,the steam parameters adopted for the new SupC units in India are on the lower range of SupC condi-tions compared to the state-of-the-art power plants elsewhere.Hence,there is an ample scope to optimize the operating parame-ters of the SupC power plants further to improve the plant efﬁcien-cies signiﬁcantly.The efﬁciency of a power plant depends on various operating parameters such as main steam/reheat steam pressures and temperatures,turbine extraction pressures,and excess air ratio for a given fuel.But simultaneous optimization of these operating parameters to achieve the maximum plant efﬁciency is a challeng-ing task.The use of Artiﬁcial Intelligence (AI)-based tools like arti-ﬁcial neural networks (ANN)and genetic algorithms (GA)have been found very promising to solve a variety of such complex/ill-deﬁned problems [4–8].ANN is widely applied in design,optimiza-tion,classiﬁcation,forecasting,and control systems.De et al.[9]developed an ANN model for the steam process of a coal biomass co-ﬁred combined heat and power plant to quickly predict the per-formance with good accuracy.Reddy and Ranjan [10]used ANN to estimate solar resource in India.The performance parameters of a solar-driven ejector-absorption cycle were modeled as functions of only the working temperature using ANN by Sözen and Akçayol [11].GA is a stochastic global search method that simulates the natural biological evolution.It searches from a population of solu-tions rather than from a single point and thus prevents the conver-gence to suboptimal solutions.Sacco et al.[4]applied GA to optimize turbine extraction in a secondary side of pressurized-water reactor.Mohagheghi and Shayegan [12]applied GA to calcu-late the optimal thermodynamic performance conditions for heat recovery steam generators.The optimization of thermodynamic parameters of the supercritical CO 2power cycle was reported by Wang et al.[8]using ANN and GA.Kalogirou [13]optimized a so-lar-energy system to maximize its economic beneﬁts using ANN and GA.This study presents a coupled neuro-genetic optimization methodology involving ANN and GA to determine the maximum0306-2619/$-see front matter Ó2011Elsevier Ltd.All rights reserved.doi:10.1016/j.apenergy.2011.06.029Corresponding author.Tel.:+914422574702;fax:+914422574652.E-mail address:ksreddy@iitm.ac.in (K.S.Reddy).possible plant efﬁciency of a high ash coal-ﬁred SupC power plant in Indian climatic condition where the design ambient tempera-ture is considered as 33°C.A unit size of 800MWe currently under development in India is considered for the neuro-genetic optimiza-tion.Furthermore,the effect of various coals on the thermody-namic performance of the optimized plant is also determined based on energy and exergy analysis.It is to be noted that the majority of the causes of irreversibilities like heat transfer through a ﬁnite temperature difference,chemical reactions,friction,and mixing are accounted by only exergy analysis [14].2.MethodologyPower plant is a complex system that involves various intercon-nected circuits each of which consists of different components.Hence,a ﬂow-sheeting computer program,‘Cycle-Tempo’is used to perform a component-wise modeling followed by a systemsimulation.‘Cycle-Tempo’is a well-structured package for the steady state thermodynamic modeling and analysis of systems for the production of electricity,heat and refrigeration [15].The power plant simulation data obtained from ‘Cycle-Tempo’is used to train the ANN to predict the energy input through fuel (coal).The optimum set of various operating parameters that result in the minimum energy input to the power plant is then determined by using the trained ANN model as a ﬁtness function with the GA.The maximum plant efﬁciency is then ﬁnally obtained from the power plant simulation in ‘Cycle-Tempo’using the set of optimum parameters.The neuro-genetic optimization of the entire plant is carried out in two stages.In the ﬁrst stage,optimal excess air ratio,intermediate pressure turbine (IP)steam parameters (reheat pres-sure and temperature),and low pressure (LP)turbine inlet steam pressure are calculated assuming high pressure (HP)turbine steam parameters (main steam temperature and pressure).Once the HP,IP,and LP turbine steam parameters are determined,then the tur-bine extraction steam pressures are calculated for the individual feedwater heaters as a part of the second stage.3.Power plant simulationThe conﬁguration of the ﬁrst 660MWe SupC power plant com-missioned by NTPC in India is considered for optimizing the vari-ous operating parameters [3].Also,the simulations were carried out for higher capacity of 800MWe for the same plant conﬁgura-tion which is currently under development in India.The process ﬂow diagram of the power plant is prepared in ‘Cycle-Tempo’and the required operating parameters (such as pressures,temper-atures,and efﬁciencies)for individual components are speciﬁed.Table 2Characteristics of Indian coals.Reference high ash (HA)Sample-1Sample-2Sample-3Sample-4As-received (wt.%)Dry (wt.%)As-received (wt.%)Dry (wt.%)As-received (wt.%)Dry (wt.%)As-received (wt.%)Dry (wt.%)As-received (wt.%)Dry (wt.%)Proximate analysis Fixed carbon 24.0027.2730.0031.7132.8035.7342.8047.4048.3049.19Volatile matter 21.0023.8623.9025.2727.3029.7426.4029.2434.1034.73Ash43.0048.8740.7043.0231.7034.5321.1023.3615.8016.08Moisture 12.00– 5.40–8.20–9.70– 1.80–Ultimate analysis Carbon 34.4639.1640.4042.7146.3050.4454.6060.4766.5067.72Hydrogen 2.43 2.76 2.60 2.75 2.70 2.94 3.00 3.32 4.10 4.18Oxygen (by difference) 6.977.929.5010.049.7010.5610.0011.079.709.88Nitrogen 0.690.78 1.00 1.06 1.00 1.09 1.20 1.33 1.70 1.73Sulfur 0.450.510.400.420.400.440.400.440.400.41Ash43.0048.8740.7043.0231.7034.5321.1023.3715.8016.08Moisture 12.00–5.40–8.20–9.70–1.80–HHV (MJ/kg)13.9615.8315.7916.6417.9019.4421.1023.3026.7827.20Exergy (MJ/kg)15.2617.3017.1418.0819.1120.7722.1424.4527.6428.08Table 1Major assumptions for the SupC power plant simulation.Ambient pressure of the reference environment (bar) 1.013Ambient temperature of the reference environment (°C)33Relative humidity of the ambient air (%)60Chemical composition of the reference-environment model:(mole fraction)N 20.7562O 20.2030H 2O 0.0312CO 20.0003Others0.0093Ash composition:(by weight)SiO 270Al 2O 330Bottom to ﬂy ash ratio 20:80Excess air (%)20Condenser pressure (kPa)10.3Temperature gain of the condenser cooling water (°C)10Final feedwater temperature (°C)305Terminal temperature difference (TTD):(°C)Low pressure (LP)closed feedwater heaters (FWHs)3High pressure (HP)closed FWHs0Drain cooler approach (DCA)temperature of closed FWHs (°C)5Isentropic efﬁciencies:(%)High pressure (HP)turbine90Intermediate pressure (IP)turbine 92Low pressure (LP)turbine90Turbine driven boiler feed pump (BFP)80Fans 80Pumps85Generator efﬁciency (%)98.7Table 3Assumed ranges of the operating parameters to be optimized.ParameterRangeExcess airUp to 25%IP turbine (RH)steam pressure 15–25%of the HP turbine (main)steam pressure IP turbine (RH)steam temperature580–620°C LP turbine steam pressure 3–5bar De-aerator pressure 9–12barLP FWH10.103–0.42bar LP FWH20.42–1.19bar LP FWH43–6.1bar HP FWH111–30.35bar4868M.V.J.J.Suresh et al./Applied Energy 88(2011)4867–4873Fig.1.Schematic representation of the800MWe supercritical power plant.Fig.2.Schematic of the ANN architecture.Plant energy efficiency;g¼Net electricity outputMass flow rate of coalÂHHVðdry basisÞof the coalð1ÞPlant exergy efficiency;e¼Net electricity outputð2ÞIn India,as a normal practice,power plant industry quotes the power plant efﬁciencies on the basis of higher heating value (HHV)of fuel.Hence,to reﬂect the typical values of power plant efﬁciencies in India,HHV(dry basis)has been used throughout the study instead of LHV.tion.The optimized parameters of the studied power plant conﬁg-uration are determined in two stages using neuro-genetic approach.In theﬁrst stage,optimized values of operating parame-ters such as excess air ratio,IP turbine(reheat steam)pressure/ temperature,and LP turbine pressure are determined assuming the typical ranges as shown in Table3[16]whereas in the second stage,the optimized extraction pressures of turbine bleed streams to feedwater heaters(FWHs)are determined.Fig.1shows the schematic of the SupC power plant conﬁgura-tion and the typical ANN architecture considered for the present study is shown in Fig.2.The neuro-genetic optimization approach shown in Fig.3is applied using MATLAB’s Neural Network and Ge-netic Algorithm toolbox[19].The neural network is trained using Levenberg–Marquardt backpropagation algorithm with four and6.Regressionﬁt based on the ANN model of power plant including FWHs.4870M.V.J.J.Suresh et al./Applied Energy88(2011)4867–4873operating variables.The objective function is to minimize the en-ergy input to the power plant without the feedwater heaters (FWHs)and subject to the constraints considered in Table3.The corresponding convergence of the GA is shown in Fig.5.Once the optimized turbine parameters are identiﬁed,the neuro-genetic optimization approach is repeated for the entire plant including the FWHs.In order to identify the optimized extraction pressures for FWHs,an equal temperature distribution is assumed for indi-vidual FWHs(wherever applicable)after determining the de-aera-tor pressure.The corresponding dataﬁt and GA convergence curves for the total plant including FWHs are shown in Figs.6 and7,respectively.Furthermore,the comparison of results obtained with the coupled neuro-genetic optimization and the‘Cy-cle-Tempo’simulation is also carried out to determine the accuracy of the adopted methodology.The variation in the output of the objective function,i.e.the minimum energy input to the power plant using reference HA Indian coal was less than1%.The stream data of the optimized power plant conﬁguration is shown in Table4.The comparison of results of neuro-genetic optimization and the parametric optimization reported by the authors in their ear-lier work[16]is shown in Table5.It is observed that neuro-genetic optimization results in almost the same plant energy and exergy efﬁciencies.Moreover,the variations in optimized operating parameters obtained using both the methods are very minimal. The neuro-genetic optimization methodology results in the signif-icant reduction of computation effort compared to the parametric optimization wherein a number of cases are required to be simu-lated corresponding to the variations in individual operating parameters.The major advantage of the neuro-genetic algorithm is the possibility of on-line optimization when quick response is re-quired.However,the physical model of the power plant needs to be built prior to the on-line optimization.5.Effect of various coals on the thermodynamic performance of the optimized plantThe power plant efﬁciency gets affected considerably by the variation in fuel composition and it is difﬁcult to account the loss that involves unburnts without using any assumptions that in turn may lead to uncertainties.It is to be noted that the energy loss in the steam generator due to the combustibles in ash,radiation and convection losses,and unaccounted losses is consideredasTable4Stream data of the optimized SupC power plant.Stream No.(as indicated in Fig.1)Pressure(bar)Temperature(°C)Massﬂow rate(kg/s)Energyﬂow rate(MW th)Exergyﬂow rate(MW th)Coal/bottom ash1 1.03033.0118.21870.82044.49 1.0131050.011.615.49.2Air/ﬂue gas2 1.01333.0687.031.203 1.04035.9687.033.2 1.64 1.030273.9687.0202.646.25 1.0101784.9793.62030.01404.46 1.000320.0793.6351.9137.77 1.000122.7793.6182.574.48 1.060130.0793.6188.679.6Water/steam10290.0600.0636.92110.9983.71162.0620.0523.61865.6787.612 3.0209.5190.3522.8122.213 3.0209.5190.3522.8122.21492.2409.844.8134.854.61562.0353.9523.61525.5586.61662.0353.968.5199.776.81725.6480.329.897.635.81811.0362.320.562.519.61911.0362.345.9139.743.920 6.1288.821.461.917.221 3.0209.525.469.716.322 1.1114.223.760.810.5230.369.117.140.8 4.2240.10346.4339.9772.932.0250.10346.4473.326.50.52611.046.5473.327.1 1.027360.0191.2636.9440.6104.828360.0305.0636.9772.9241.729342.5340.0636.9893.2299.430 1.01333.020808.30031 2.03033.020808.3 2.7 2.132 1.03043.020808.3870.314.0Energy88(2011)4867–487348711.5%of energy input through the coal for the optimized power plant conﬁguration.Since exergy analysis gives more insights into the process,the present study is extended to determine the effect of coal composition on the thermodynamic performance of the optimized power plant based on both energy and exergy.Different coal samples considered in Table 2are used to evalu-ate the performance.The results of energy and exergy balance are shown in Tables 6and 7,respectively with cases representing the values corresponding to respective coal samples.The energy losses are calculated as the ratio of energy rejected to the energy content of input fuel whereas the exergy losses are calculated as the ratio of irreversibilities to the exergy content of the fuel.It is observed that there is an increase of 1.2%points in plant energy efﬁciency using coal with an ash content of 16%compared with the reference coal with an ash content of 49%(dry-basis).The corresponding in-crease in plant exergy efﬁciency is 3.3%.The variation of fuel consumption with different coal samples is shown in Fig.8.A signiﬁcant reduction of about 42%in coal con-sumption is observed using coal with an ash content of 16%(sam-ple-4)compared to the reference coal that in turn results in a reduction of auxiliary power consumption.The reduction of energy loss through the bottom ash also contributes to the increase in plant energy efﬁciency.However,exergy balance gives additional insights into the process.There is a signiﬁcant reduction in exergy loss in the combustor with the decrease in ash content of coals which is due to the increase in combustibles.However,the heat transfer irreversibility in the steam generator increases for the plant using relatively low ash coals compared to the reference HA coal.This is due to the relatively higher ﬂue gas temperature using low ash coals (higher reaction temperature)compared to the reference coal and hence higher temperature difference be-tween the ﬂue gas and the steam for the same excess air ratio and steam parameters of the turbine cycle.6.ConclusionsThermodynamic optimization of power plant based on coupled artiﬁcial neural network and genetic algorithm (neuro-genetic)is found to be an efﬁcient methodology compared to the routine parametric optimization.Neuro-genetic optimization methodol-ogy signiﬁcantly reduces the computational effort without com-promising the accuracy of the results along with the major advantage of on-line optimization.Furthermore,the thermody-namic analysis carried out to study the effect of coal composition on the power plant performance shows a reduction of about 42%in fuel consumption using coal with 16%ash compared with the coal having 49%ash.The corresponding increase in plant energy and exergy efﬁciencies are 1.2%and 3.3%points,respectively.It is also observed that the exergy loss in the combustor may be a suitable indicator to determine the effect of variation in coal com-position on the power plant performance.References[1]BP Statistical Review of World Energy June 2010.British Petroleum;2010.</productlanding.do?categoryId=6929&contentId=7044622>(accessed 12.02.11).[2]Central Electricity Authority (CEA),2010.Monthly generation report;March2010.<http://www.cea.nic.in/god/opm/Monthly_Generation_Report/18col_A_10_03/actual-mar10.htm >(accessed 22.09.10).[3]National Thermal Power Corporation Limited (NTPC).Power Plant Data.Engineering Ofﬁce Complex,Noida,India:Private communication;2008.[4]Sacco WF,Pereira CMNA,Soares PPM,Schirru R.Genetic algorithms applied toturbine extraction optimization of a pressurized-water reactor.Appl Energy 2002;73:217–22.[5]Yanmin W,Pingjing Y.Simulation and optimization for thermally coupleddistillation using artiﬁcial neural network and genetic algorithm.Chin J Chem Eng 2003;11(3):307–11.Table 6Comparison of energy ponents (%)Reference high ash (HA)Case-1Case-2Case-3Case-4Power (efﬁciency)39.639.840.340.740.8Heat rejected in cooling water 46.546.646.947.046.8Heat rejected through stack 10.110.09.89.810.2Heat rejected through bottom ash0.80.70.50.30.2Other losses (by difference)3.02.92.52.22.0Plant energy efﬁciency (%)39.539.6Plant exergy efﬁciency (%)36.136.2Table 7Comparison of exergy ponents (%)Reference high ash (HA)Case-1Case-2Case-3Case-4Power (efﬁciency)36.236.637.738.739.5Loss in combustor 33.132.531.029.428.3Loss in steam generator (excluding combustor)16.816.917.417.918.3Loss through stack 3.9 4.0 4.2 4.4 4.5Loss in turbine 3.5 3.5 3.6 3.7 3.7Loss in condenser and cooling water1.9 1.9 1.92.0 2.0Loss in feed water heaters 1.0 1.0 1.0 1.0 1.1Loss through bottom ash 0.50.40.30.20.1Other losses (by difference)3.13.22.92.72.5Fig.8.Variation of coal consumption with different samples.[6]Changyu S,Lixia W,Qian L.Optimization of injection molding processparameters using combination of artiﬁcial neural network and genetic algorithm method.J Mater Process Technol2007;183:412–8.[7]Qdais HA,Hani KB,Shatnawi N.Modeling and optimization of biogasproduction from a waste digester using artiﬁcial neural network and genetic algorithm.Resour Conserv Recycl2010;54:359–63.[8]Wang J,Sun Z,Dai Y,Ma S.Parametric optimization design for supercritical CO2power cycle using genetic algorithm and artiﬁcial neural network.Appl Energy 2010;87:1317–24.[9]De S,Kaiadi M,Fast M,Assadi M.Development of an artiﬁcial neural networkmodel for the steam process of a coal biomass coﬁred combined heat and power(CHP)plant in Sweden.Energy2007;32:2099–109.[10]Reddy KS,Ranjan M.Solar resource estimation using artiﬁcial neural networksand comparison with other correlation models.Energy Convers Manage 2003;44:2519–30.[11]Sözen A,Akçayol MA.Modelling(using artiﬁcial neural-networks)theperformance parameters of a solar-driven ejector-absorption cycle.Appl Energy2004;79:309–25.[12]Mohagheghi M,Shayegan J.Thermodynamic optimization of design variablesand heat exchangers layout in HRSGs for CCGT,using genetic algorithm.Appl Therm Eng2009;29:290–9.[13]Kalogirou SA.Optimization of solar systems using artiﬁcial neural-networksand genetic algorithms.Appl Energy2004;77:383–405.[14]Cengel YA,Boles MA.Thermodynamics:an engineering approach.4th ed.NewDelhi:Tata McGraw-Hill;2004.[15]Cycle-Tempo release 5.0,Delft University of Technology;2007.<http://www.3me.tudelft.nl/live/pagina.jsp?id=8c53f82e-a500-41f1-971b-629e832bfbef&lang=en>(accessed06.08.09).[16]Suresh MVJJ,Reddy KS,Kolar AK.Thermodynamic optimization of advancedsteam power plants retroﬁtted for oxy-coal combustion.J Eng Gas Turbines Power2011;133.p.063001-1–063001-12.[17]US Department of Energy.Market-Based Advanced Coal Power Systems,Washington;1999.</technologies/coalpower/refshelf/ marketbased_systems_report.pdf>(accessed11.09.08).[18]Choudhury N,Boral P,Mitra T,Adak AK,Choudhury A,Sarkar P.Assessment ofnature and distribution of inertinite in Indian coals for burning characteristics.Int J Coal Geol2007;72:141–52.[19]MATLAB version7.9(R2009b);2009.The MathWorks Inc.,Natick,Massachusetts,USA.M.V.J.J.Suresh et al./Applied Energy88(2011)4867–48734873。