多智能体遗传算法在地球物理反演中的应用研究
|
摘要
大部分地球物理参数反演问题属于多极值的目标函数优化问题,使用基于线性或拟线性理论的反演方法往往导致反演结果陷入局部最优,造成反演结果的不可靠。常规的反演方法越来越难满足那些地下情况更复杂、处理标准更高的要求。所以,引入新的更优的反演思想和方法是改善这种状况的重要途径之一。本文结合前沿交叉学科的一些新研究成果,对简单遗传算法进行改进,将多智能体遗传算法用于地球物理参数反演。
本文在分析了当前地球物理反演的要求后,兼顾算法全局搜索和收敛速度两个核心要求,首次将已经在其他优化领域成功应用的多智能体遗传算法引入到地球物理反演中。该方法采用实数编码,利用均匀设计的方法产生初始种群,设计了四个智能体遗传算子,分别为邻域竞争算子,邻域正交交叉算子,变异算子和自学习算子。其中邻域竞争算子和邻域正交交叉算子分别实现智能体的竞争和合作行为,变异算子和自学习算子实现了智能体利用自身知识的行为。并用标准测试函数对算法进行测试,从理论模型开始,研究智能体遗传算法在大地电磁反演以及地震反射系数反演中的可行性,进行了大量理论模型计算试验,并对反演结果进行了分析。
本论文围绕多智能体遗传算法在地球物理反演中的应用这一前沿课题,分四个部分来论述:
第一部分首先简要阐述本文研究问题的提出,课题研究的意义,研究的思路和主要内容,主要创新和贡献等几个方面内容。
第二部分简要回顾了非线性反演方法的提出,非线性反演方法的发展和分类,并对非线性反演方法优势和局限性进行了评述,就反演问题的非线性与多极值、反演解的存在性、非唯一性和稳定性、计算量问题等进行了讨论。
第三部分和第四部分是本文的主要内容。第三部分结合多智能体遗传算法优化原理和算法实现过程中的一些关键技术,如邻域竞争,邻域正交交叉策略等,针对地球物理反演问题的非线性、多极值等特点提出一套可行的实现方案,并设计了多智能体遗传算法反演地球物理反演问题的详细实现流程。
第四部分对该优化算法的性能进行检验,先从优化标准测试函数开始,然后应用到地球物理理论模型数据反演,研究多智能体遗传方法在大地电磁反演以及地震反射系数反演中的可行性,并对其计算效率进行了大量计算试验,并对反演结果进行了分析。
多智能体遗传算法不但具有非线性优化算法的许多优点,如不受初始模型选取的限制、可以方便的与其它优化算法进行综合(联合或混合)反演,还可以实现并行计算,使其计算速度提升。这充分显示了方法的实用价值。其对地球物理理论模型正演数据反演结果及效率,也显示出多智能体遗传算法求解地球物理反演问题的巨大潜力。
Most of geophysical inversion belong to the optimazition of multi-stream cost function.However,conventional inversion methods such as generalized linear inversion and so on are based on linear theory or quasi-linear theory,which results in sensitive to initial model and local extreme and multi-extreme.The conventional methods can hardly meet the requirment for more complex subsurface conditions and sophisticated processing.So an important route to improve the status is to introduce newer and more outstanding ideas and algorithms in geophysical domain.In this paper, we use some new cutting-edge interdisciplinary research to improve on the simple genetic algorithm, developed the multi-agent genetic algorithm for geophysical inversion.
Based on analysis of prevailing requests of geophysical inversion and drawbacks of inversion methods today in this paper,the author first introduces multi-agent genetic algorithm in geophysical inversion field which is applied successfully in other optimal fields,considering the global searching and rapid convergence.
This algorithm involves several operators,including the initial population,neighborhood competition operator, neighborhood orthogonal crossover operator, mutation operator and self-learning operator.With the standard test functions to test the algorithm, and then,I study the feasibility of the multi-agent genetic algorithm to solve magnetotelluric inversion and seismic impedance inversion, carring out a large number of theoretical model testing, and inversion results are analyzed.Focused on the research of multi-agent genetic algorithm in finding a solution to the geophysical inversion problems,this paper consists of four parts as follow.
The first part reviews the study status of inversion problems briefly,and shows the significance,the train of thought,the major content,and the innovation of this project's rearch.
The second part gives the review of the proposed non-linear inversion,the development and classification ,the advantages and limitations ,non-linear inversion method.And discussing the characteristics of nonlinear,non-unisequeness,and multi-stream to the gephysical inversion problems.
The third part and the fourth part are the main work of this paper.In the third part,the author proposed a realizing scheme for geophysical inversion problems with nonliear and multi-stream properties,based on the fundaments of multi-agent genetic algorithm and the key step of the method,such as eighborhood competition operator, neighborhood orthogonal crossover operator.
In the forth part,I use the standard test functions to test the algorithm.And then,appling the algorithm to the MT and seismic acoustic impedance inversion.Results show that multi-agent genetic algorithm is a statable and effective nonliear inversion method with global convergence.
Multi-agent genetic algorithm not only has many advantages of some nonlinear optimization algorithm , such as the initial model selected from the restrictions, can easily be integrated with other optimization algorithms (joint or mixed) or parallel computation can be achieved, so that improve the calculation speed.This fully shows the method of practical value.The theoretical model of geophysical data inversion results of forward and efficiency, also shows that multi-agent genetic algorithm for solving geophysical inverse problem of great potential.
本文在分析了当前地球物理反演的要求后,兼顾算法全局搜索和收敛速度两个核心要求,首次将已经在其他优化领域成功应用的多智能体遗传算法引入到地球物理反演中。该方法采用实数编码,利用均匀设计的方法产生初始种群,设计了四个智能体遗传算子,分别为邻域竞争算子,邻域正交交叉算子,变异算子和自学习算子。其中邻域竞争算子和邻域正交交叉算子分别实现智能体的竞争和合作行为,变异算子和自学习算子实现了智能体利用自身知识的行为。并用标准测试函数对算法进行测试,从理论模型开始,研究智能体遗传算法在大地电磁反演以及地震反射系数反演中的可行性,进行了大量理论模型计算试验,并对反演结果进行了分析。
本论文围绕多智能体遗传算法在地球物理反演中的应用这一前沿课题,分四个部分来论述:
第一部分首先简要阐述本文研究问题的提出,课题研究的意义,研究的思路和主要内容,主要创新和贡献等几个方面内容。
第二部分简要回顾了非线性反演方法的提出,非线性反演方法的发展和分类,并对非线性反演方法优势和局限性进行了评述,就反演问题的非线性与多极值、反演解的存在性、非唯一性和稳定性、计算量问题等进行了讨论。
第三部分和第四部分是本文的主要内容。第三部分结合多智能体遗传算法优化原理和算法实现过程中的一些关键技术,如邻域竞争,邻域正交交叉策略等,针对地球物理反演问题的非线性、多极值等特点提出一套可行的实现方案,并设计了多智能体遗传算法反演地球物理反演问题的详细实现流程。
第四部分对该优化算法的性能进行检验,先从优化标准测试函数开始,然后应用到地球物理理论模型数据反演,研究多智能体遗传方法在大地电磁反演以及地震反射系数反演中的可行性,并对其计算效率进行了大量计算试验,并对反演结果进行了分析。
多智能体遗传算法不但具有非线性优化算法的许多优点,如不受初始模型选取的限制、可以方便的与其它优化算法进行综合(联合或混合)反演,还可以实现并行计算,使其计算速度提升。这充分显示了方法的实用价值。其对地球物理理论模型正演数据反演结果及效率,也显示出多智能体遗传算法求解地球物理反演问题的巨大潜力。
Most of geophysical inversion belong to the optimazition of multi-stream cost function.However,conventional inversion methods such as generalized linear inversion and so on are based on linear theory or quasi-linear theory,which results in sensitive to initial model and local extreme and multi-extreme.The conventional methods can hardly meet the requirment for more complex subsurface conditions and sophisticated processing.So an important route to improve the status is to introduce newer and more outstanding ideas and algorithms in geophysical domain.In this paper, we use some new cutting-edge interdisciplinary research to improve on the simple genetic algorithm, developed the multi-agent genetic algorithm for geophysical inversion.
Based on analysis of prevailing requests of geophysical inversion and drawbacks of inversion methods today in this paper,the author first introduces multi-agent genetic algorithm in geophysical inversion field which is applied successfully in other optimal fields,considering the global searching and rapid convergence.
This algorithm involves several operators,including the initial population,neighborhood competition operator, neighborhood orthogonal crossover operator, mutation operator and self-learning operator.With the standard test functions to test the algorithm, and then,I study the feasibility of the multi-agent genetic algorithm to solve magnetotelluric inversion and seismic impedance inversion, carring out a large number of theoretical model testing, and inversion results are analyzed.Focused on the research of multi-agent genetic algorithm in finding a solution to the geophysical inversion problems,this paper consists of four parts as follow.
The first part reviews the study status of inversion problems briefly,and shows the significance,the train of thought,the major content,and the innovation of this project's rearch.
The second part gives the review of the proposed non-linear inversion,the development and classification ,the advantages and limitations ,non-linear inversion method.And discussing the characteristics of nonlinear,non-unisequeness,and multi-stream to the gephysical inversion problems.
The third part and the fourth part are the main work of this paper.In the third part,the author proposed a realizing scheme for geophysical inversion problems with nonliear and multi-stream properties,based on the fundaments of multi-agent genetic algorithm and the key step of the method,such as eighborhood competition operator, neighborhood orthogonal crossover operator.
In the forth part,I use the standard test functions to test the algorithm.And then,appling the algorithm to the MT and seismic acoustic impedance inversion.Results show that multi-agent genetic algorithm is a statable and effective nonliear inversion method with global convergence.
Multi-agent genetic algorithm not only has many advantages of some nonlinear optimization algorithm , such as the initial model selected from the restrictions, can easily be integrated with other optimization algorithms (joint or mixed) or parallel computation can be achieved, so that improve the calculation speed.This fully shows the method of practical value.The theoretical model of geophysical data inversion results of forward and efficiency, also shows that multi-agent genetic algorithm for solving geophysical inverse problem of great potential.
引文
[1]杨文采.地球物理反演的理论与方法[M].北京:地质出版社,1997.
[2]王家映.地球物理反演理论(第二版)[M].北京:高等教育出版社,2002.
[3]罗红明.量子遗传算法及其在地球物理反演中的应用研究[D].武汉:中国地质大学,2007.
[4] Zhong Weicai,Liu Jing,Liu Fang,etal.Combinational optimization using multy-agent evolutionary algorithm[J].Chinese Journal of Computers,2004,27(10):1341-1353.
[5]白俊雨,赵俊省,潘凌飞.基于MPI并行实数编码混合遗传算法的波阻抗反演[J].勘探地球物理进展,2009,34(2):265-269.
[6]李志俊,程家兴.基于数论佳点集的遗传算法初始种群均匀设计[J].电脑与信息技术,2007,15(4):29-32.
[7] Leung,Yuping Wang.An Orthogonal Genetic Algorithm with Quantization for Global Numerical Optimization[J].IEEE Trans.Evolutionary Computation,2001,5(1):41-53.
[8]姚姚.地球物理反演-基本理论与应用方法[M].武汉:中国地质大学出版社,2002,8.
[9]王家映.地球物理反演问题概述[J].工程地球物理学报,2007,4(1):1-3.
[10]钟伟才,刘静,刘芳,等.组合优化多智能体进化算法[J].计算机学报,2004,27(10):1341—1353.
[11]白俊雨,王绪本,周军.大地电磁实码广义遗传反演算法研究[J].物探化探计算技术,2009,31(4):314-318.
[12]李志俊,程家兴.基于数论佳点集的遗传算法初始种群均匀设计[J].电脑与信息技术,2007,15(4):29-32.
[13] Deep Kusum, Thakur Manoj. A new crossover operator for real coded genetic algorithms [J]. Applied Mathematics and Computation (S0096-3003), 2006, 188(1): 895-911.
[14]丁承民,张传生,刘贵忠.正交试验遗传算法及其在函数优化中的应用[J].系统工程与电子技术,1997,19(10):57-60.
[15]史奎凡,董吉文,李金屏,等.正交遗传算法[J].电子学报,2002,30(10):1501-1504.
[16]何大阔,王福利贾明兴.遗传算法初始种群与操作参数的均匀设计[J].东北大学学报,2005,26(9):828-831
[17]李广文,章卫国,刘小雄,等.基于均匀设计的小生境遗传算法及其在飞控系统中的应用[J].航空学报,2008,29(增刊):73-78.
[18] Zhong W C,Liu J,Jiao L C .A multi-agent genetic algorithm for global optimization[J].IEEE Transactions on Systems,Man and Cybernetics,Part B:Cybernetics,2004,34(2):1128-1141.
[19]钟伟才,刘静,焦李成.多智能体遗传算法用于线性系统逼近[J].自动化学报,2004,30(6):933-938
[20]薛明志,钟伟才,刘静,焦李成.用于函数优化的正交Multi-Agent遗传算法[J].系统工程与电子技术,2004,26(9):1306-1311.
[21]鲁帆,蒋云钟,王浩,牛存稳.多智能体遗传算法用于马斯京根模型参数估计[J].水力学报,2007,38(3):289-294.
[22]曾孝平,张晓娟,李勇明.多子群协同链式智能体遗传算法分析[J].重庆大学学报,2008,31(7):781-785.
[23]杨文采.非线性地球物理反演方法:回顾与展望.地球物理学进展,2002,17(2):255-260.
[24]聂茹.煤炭高精度地震波阻抗反演智能算法研究与应用[D],北京:中国矿业大学,2009.
[25]王凌.智能优化算法及其应用[M].清华大学出版社,2001,10.
[26]钟伟才,薛明志,刘静,焦李成.多智能体遗传算法用于超高维函数优化[J].自然科学进展,2003,13(10).
[27] Liu Jiming,Hna Jing,TnagYunaY.Multi-agent oriented constarint satisafetion.Artifieial Intelligenee.2002,136l:101-144.
[28]卢虎生,朱林,高学东,著.多智能体计划调度系统的理论与应用[M].北京:冶金工业出版社.2003.
[29]潘正君,康立山,陈毓屏.演化计算[M].北京:清华大学出版社,1998.
[30]王元,方开泰.关于均匀分布与试验设计(数论方法)[J],科学通报,1981,26(2):65—70.
[31]方开泰.正交与均匀试验设计[M],科学出版社,2001,10.
[32]高齐圣,潘德惠.基于均匀设计的遗传算法及其应用[J],信息与控制,1999,28(3):236—240.
[33]纪晨,姚振兴.用于地球物理反演的均匀设计优化方法[J],地球物理学报,1996,39(2):233-242.
[34]钟伟才.多智能体进化模型和算法研究[D].西安,西安电子科技大学,2007.
[35]梁昌勇,张俊岭,杨善林.一种高效Multi-agent仿生算法用于设计优化,系统仿真学报,2009,21(2).
[36] Srinivas M,PatnaikL M.Adaptive probabilities of crossover and mutation in genetic algorithms[J].IEEE Transactions on Systems,Man and Cybernetics,1994,24(4):656-667.
[37]巩敦卫,孙晓燕,郭西进.一种新的优胜劣汰遗传算法[J].控制与决策,2002,(06).
[38] A R Mehrabian C. Lucas.A novel numerical optimization algorithm inspired from weed colonization[J].Ecological Informatics (S1574-9541),2006,1(4):355-366.
[39] Wei Lingyun,Zhao Mei.A niche hybrid genetic algorithm for global optimization of continuous multimodal functions[J].Applied Mathematics and Computation(S0096-3003), 2005,160(3): 649-661.
[40] Wooldridge M,石纯一,等译.多Agent系统引论[M].北京:北京电子工业出版社.2003.
[41] Bonabeau E,Dorigo M,Theraulaz G.Swarm Intelligence:From Natural to Artificial Systems [M].Oxford,,UK:Oxford University Press,1999.
[42] Gutowitz H. Cellular Automata: Theory and Experiment [M].Cambridge,UK: MIT Press,1991.
[43] Lachmann M,Sella G.The Computationally Complete Ant Colony:Global Coordination in a System with no Hierarchy [C]// Proc of European Conference on Artificial Life'95: Lecture in Artificial Intelligence,1995.New York,USA: Springer,1995,929: 784-800.
[44]刘文卿.实验设计[M].北京:清华大学出版社,2005.
[45]何大阔,王福利,贾明兴.遗传算法初始种群与操作参数的均匀设计[J].东北大学学报:自然科学版,2005,26(9):828-831.
[46]赵曙光,焦李成,王宇平,等.基于均匀设计的多目标自适应遗传算法及应用[J].电子学报,2004,32(10):1723-1725.
[2]王家映.地球物理反演理论(第二版)[M].北京:高等教育出版社,2002.
[3]罗红明.量子遗传算法及其在地球物理反演中的应用研究[D].武汉:中国地质大学,2007.
[4] Zhong Weicai,Liu Jing,Liu Fang,etal.Combinational optimization using multy-agent evolutionary algorithm[J].Chinese Journal of Computers,2004,27(10):1341-1353.
[5]白俊雨,赵俊省,潘凌飞.基于MPI并行实数编码混合遗传算法的波阻抗反演[J].勘探地球物理进展,2009,34(2):265-269.
[6]李志俊,程家兴.基于数论佳点集的遗传算法初始种群均匀设计[J].电脑与信息技术,2007,15(4):29-32.
[7] Leung,Yuping Wang.An Orthogonal Genetic Algorithm with Quantization for Global Numerical Optimization[J].IEEE Trans.Evolutionary Computation,2001,5(1):41-53.
[8]姚姚.地球物理反演-基本理论与应用方法[M].武汉:中国地质大学出版社,2002,8.
[9]王家映.地球物理反演问题概述[J].工程地球物理学报,2007,4(1):1-3.
[10]钟伟才,刘静,刘芳,等.组合优化多智能体进化算法[J].计算机学报,2004,27(10):1341—1353.
[11]白俊雨,王绪本,周军.大地电磁实码广义遗传反演算法研究[J].物探化探计算技术,2009,31(4):314-318.
[12]李志俊,程家兴.基于数论佳点集的遗传算法初始种群均匀设计[J].电脑与信息技术,2007,15(4):29-32.
[13] Deep Kusum, Thakur Manoj. A new crossover operator for real coded genetic algorithms [J]. Applied Mathematics and Computation (S0096-3003), 2006, 188(1): 895-911.
[14]丁承民,张传生,刘贵忠.正交试验遗传算法及其在函数优化中的应用[J].系统工程与电子技术,1997,19(10):57-60.
[15]史奎凡,董吉文,李金屏,等.正交遗传算法[J].电子学报,2002,30(10):1501-1504.
[16]何大阔,王福利贾明兴.遗传算法初始种群与操作参数的均匀设计[J].东北大学学报,2005,26(9):828-831
[17]李广文,章卫国,刘小雄,等.基于均匀设计的小生境遗传算法及其在飞控系统中的应用[J].航空学报,2008,29(增刊):73-78.
[18] Zhong W C,Liu J,Jiao L C .A multi-agent genetic algorithm for global optimization[J].IEEE Transactions on Systems,Man and Cybernetics,Part B:Cybernetics,2004,34(2):1128-1141.
[19]钟伟才,刘静,焦李成.多智能体遗传算法用于线性系统逼近[J].自动化学报,2004,30(6):933-938
[20]薛明志,钟伟才,刘静,焦李成.用于函数优化的正交Multi-Agent遗传算法[J].系统工程与电子技术,2004,26(9):1306-1311.
[21]鲁帆,蒋云钟,王浩,牛存稳.多智能体遗传算法用于马斯京根模型参数估计[J].水力学报,2007,38(3):289-294.
[22]曾孝平,张晓娟,李勇明.多子群协同链式智能体遗传算法分析[J].重庆大学学报,2008,31(7):781-785.
[23]杨文采.非线性地球物理反演方法:回顾与展望.地球物理学进展,2002,17(2):255-260.
[24]聂茹.煤炭高精度地震波阻抗反演智能算法研究与应用[D],北京:中国矿业大学,2009.
[25]王凌.智能优化算法及其应用[M].清华大学出版社,2001,10.
[26]钟伟才,薛明志,刘静,焦李成.多智能体遗传算法用于超高维函数优化[J].自然科学进展,2003,13(10).
[27] Liu Jiming,Hna Jing,TnagYunaY.Multi-agent oriented constarint satisafetion.Artifieial Intelligenee.2002,136l:101-144.
[28]卢虎生,朱林,高学东,著.多智能体计划调度系统的理论与应用[M].北京:冶金工业出版社.2003.
[29]潘正君,康立山,陈毓屏.演化计算[M].北京:清华大学出版社,1998.
[30]王元,方开泰.关于均匀分布与试验设计(数论方法)[J],科学通报,1981,26(2):65—70.
[31]方开泰.正交与均匀试验设计[M],科学出版社,2001,10.
[32]高齐圣,潘德惠.基于均匀设计的遗传算法及其应用[J],信息与控制,1999,28(3):236—240.
[33]纪晨,姚振兴.用于地球物理反演的均匀设计优化方法[J],地球物理学报,1996,39(2):233-242.
[34]钟伟才.多智能体进化模型和算法研究[D].西安,西安电子科技大学,2007.
[35]梁昌勇,张俊岭,杨善林.一种高效Multi-agent仿生算法用于设计优化,系统仿真学报,2009,21(2).
[36] Srinivas M,PatnaikL M.Adaptive probabilities of crossover and mutation in genetic algorithms[J].IEEE Transactions on Systems,Man and Cybernetics,1994,24(4):656-667.
[37]巩敦卫,孙晓燕,郭西进.一种新的优胜劣汰遗传算法[J].控制与决策,2002,(06).
[38] A R Mehrabian C. Lucas.A novel numerical optimization algorithm inspired from weed colonization[J].Ecological Informatics (S1574-9541),2006,1(4):355-366.
[39] Wei Lingyun,Zhao Mei.A niche hybrid genetic algorithm for global optimization of continuous multimodal functions[J].Applied Mathematics and Computation(S0096-3003), 2005,160(3): 649-661.
[40] Wooldridge M,石纯一,等译.多Agent系统引论[M].北京:北京电子工业出版社.2003.
[41] Bonabeau E,Dorigo M,Theraulaz G.Swarm Intelligence:From Natural to Artificial Systems [M].Oxford,,UK:Oxford University Press,1999.
[42] Gutowitz H. Cellular Automata: Theory and Experiment [M].Cambridge,UK: MIT Press,1991.
[43] Lachmann M,Sella G.The Computationally Complete Ant Colony:Global Coordination in a System with no Hierarchy [C]// Proc of European Conference on Artificial Life'95: Lecture in Artificial Intelligence,1995.New York,USA: Springer,1995,929: 784-800.
[44]刘文卿.实验设计[M].北京:清华大学出版社,2005.
[45]何大阔,王福利,贾明兴.遗传算法初始种群与操作参数的均匀设计[J].东北大学学报:自然科学版,2005,26(9):828-831.
[46]赵曙光,焦李成,王宇平,等.基于均匀设计的多目标自适应遗传算法及应用[J].电子学报,2004,32(10):1723-1725.