重力异常的BP神经网络三维物性反演
|
摘要
三维物性反演参数多,计算量巨大,传统的方法难以实现.本文使用BP神经网络实现重力三维物性反演,介绍了BP神经网络的基本原理及特性,并构造一个适用于重力位场反演的BP神经网络.并用其对模型进行反演计算,结果表明:BP网络具有较好的泛化能力和容错能力,反演速度快、准确,并且较好的反应了场源的分布情况.
The traditional methods are hardly used in 3-D inversion for physical properties,because of the large number of parameters and the quantities.The BP artificial neural network is widely used in geophysical inversion,but it's mainly used in the inversions which have few inversion parameters.In this paper,we used BP artificial neural network to develop a method of 3-D gravity inversion for physical properties which has the large number of parameters and the quantities.In this paper,we introduced the principle of BP artificial neural network,it's characteristic and the characteristic of different learning algorithms,built kinds of BP artificial neural network and analysed that how the structure of BP artificial neural network,the scalar of the samples and the learning algorithm effect the inversion reaults when there is lots of parameters.Also,we inroduced that how to built a suitable BP artificial neural network using structure method and abridgment.Then,we built a suitable BP artificial neural network for 3-D gravity physical inversion with the matlab neural network toolbox which is aslo introduced in the paper.Compared the inversion results of different date produced by different models,We'll analysis the generalization ability of BPnetwork.Compared with inversion result of the noise date,we can analysis the fault-tolerant ability of BPnetwork.The result proved that the inversion using a suitable BP artificial neural network is fast and exact.Even there is a large number of parameters,we can well obtained the distribution of source.It's have a good generalization ability,and it's fault-tolerant ability is still good.
The traditional methods are hardly used in 3-D inversion for physical properties,because of the large number of parameters and the quantities.The BP artificial neural network is widely used in geophysical inversion,but it's mainly used in the inversions which have few inversion parameters.In this paper,we used BP artificial neural network to develop a method of 3-D gravity inversion for physical properties which has the large number of parameters and the quantities.In this paper,we introduced the principle of BP artificial neural network,it's characteristic and the characteristic of different learning algorithms,built kinds of BP artificial neural network and analysed that how the structure of BP artificial neural network,the scalar of the samples and the learning algorithm effect the inversion reaults when there is lots of parameters.Also,we inroduced that how to built a suitable BP artificial neural network using structure method and abridgment.Then,we built a suitable BP artificial neural network for 3-D gravity physical inversion with the matlab neural network toolbox which is aslo introduced in the paper.Compared the inversion results of different date produced by different models,We'll analysis the generalization ability of BPnetwork.Compared with inversion result of the noise date,we can analysis the fault-tolerant ability of BPnetwork.The result proved that the inversion using a suitable BP artificial neural network is fast and exact.Even there is a large number of parameters,we can well obtained the distribution of source.It's have a good generalization ability,and it's fault-tolerant ability is still good.
引文
[1]姚长利,郑元满,张聿文.重磁异常三维物性反演随机子域法方法技术.地球物理学报,2007,50(5):1576-1583.YAO Chang-Li,ZHENG Yuan-Man,ZHANG Yu-Wen.3-Dgravity and magnetic inversion for physical properties usingstochastic subspaces.Chinese J.Geophys.(in Chinese),2007,50(5):1576-1583.
[2]杨立强,宋海滨,郝天珧.基于BP神经网络的波阻抗反演及应用.地球物理学进展,2005,20(1):34-37.YANG Li-qiang,SONG Hai-bin,HAO Tian-yao.Applicationof impedance inversion based on BP neural network.PROGRESS IN GEOPHYSICS,2005,20(1):34-37.
[3]Jordi,Turiel A,Garcia Ladona E,Font.Microcanonicalmultifractal formalism:Application to the estimation of oceansurface velocities.Journal of Geophysical Research,2007,112(5):80-97.
[4]张艺峰,姚道平,谢志招等.爆破震害预测的神经网络模型.地球物理学进展,2009,24(2):692-698.ZHANG Yi-feng,YAO Dao-ping,XIE Zhi-zhao.Neuralnetwork model for forecast of hazard by harm of blastingvibration.PROGRESS IN GEOPHYSICS,2009,24(2):692-698.
[5]谢林涛,付志红,谢品芳等.基于神经网络的视电阻率快速算法.地球物理学进展,2009,24(4):1527-1532.XIE Lin-tao,FU Zhi-hong,XIE Pin-fang.A fast algorithm ofapparent resistivity based on neural networks.PROGRESS INGEOPHYSICS,2009,24(2):692-698.
[6]唐秋华,刘保华,陈永奇等.结合遗传算法的LVQ神经网络在声学底质分类中的应用.地球物理学报,2007,50(1):313-319.TANG Qiu-Hua,LIU Bao-Hua,CHEN Yong-Qi.Applicationof LVQ neural network combined with the genetic algorithm inacoustic seafioor classification.Chinese J.Geophys.(inChinese),2007,50(1):313-319.
[7]张治国,杨毅恒,夏立显.自组织特征映射神经网络在测井岩性识别中的应用.地球物理学进展,2005,20(2):332-336.ZHANG Zhi-guo,YANG Yi-heng,XIA Li-xian.Theapplication of self-organizing feature map neural network tologging lithological identification.PROGRESS IN GEOPHYSICS,2005,20(2):332-336.
[8]李琼,李勇,李正文等.基于GA-BP理论的储层视裂缝密度地震非线性反演方法.地球物理学进展,2006,21(2):465-471.LI Qiong,LI Yong,LI Zheng-wen.A seismic nonlinearinversion for apparent fracture density of hydrocarbonreservoir based on GA-BP theory.PROGRESS INGEOPHYSICS,2006,21(2):465-471.
[9]徐海浪,吴小平.电阻率二维神经网络反演.地球物理学报,2006,49(2):584-589.XU Hai-Lang,WU Xiao-Ping.2-D resistivity inversion usingthe neural network method.Chinese J.Geophys.(in Chinese),2006,49(2):584-589.
[10]Gad El-Qady,Keisuke Ushijima.Inversion of DC resistivitydata using neural networks.Geophysical Prospecting,2001,49:417-430.
[11]朱自强,程方道,黄国祥.同时反演两个三维密度界面的拟神经网BP算法.石油物探,1995,34(1):76-85.Zhu Zi-Qiang,Cheng Fang-Dao,Huang Guo-xiang.Quasineural network BP algorithm for simultaneous inverstion oftwo 3-D density interfaces.geophysical prospecting forpetroleum,1995,34(1):76-85.
[12]朱晓军,申宁华.BP网络在位场解释中的应用.长春地质学院学报,1995,25(1):81-86.Zhu Xiao-jun,Shen Ning-hua.The application of BP networkto the interpretation of potential field data.Journal ofChangchun university of earth sciences.1995,25(1):81-86.
[13]张新兵,王家林.BP、Hopfield神经网络在位场反演中的应用比较.物探与化探计算技术,2007,29(2):161-166.Zhang Xin-bing,Wang Jia-lin.The application comparisons ofBPNN and HNN in potential field inversion.COMPUTINGTECHNIQUES FOR GEOPHYSICAL AND GEOCHEMICALEXPLORATION,2007,29(2):161-166.
[14]朱凯光,林君,韩悦慧等.基于神经网络的时间域直升机电磁数据电导率深度成像.地球物理学报,2010,53(5):743-750.ZHU Kai-Guang,LIN Jun,HAN Yue-Hui.Research onconductivity depth imaging of time domain helicopter-borneelectromagnetic data based on neural network.Chinese J.Geophys.(in Chinese),2010,53(5):743-750.
[15]Baan m,jutten C.Nueral networks in geophysical applications.Geophysics,2000,65(4):1032-1047.
[16]管志宁,侯俊生,黄临平.重磁异常反演的拟BP神经网络方法及其应用.地球物理学报,1998,41(2):242-251.Guan Zhi-ning,Hou Jun-sheng,Huang Lin-ping.Inversion ofgravity and magnetic anomalies using pseudo-BP neuralnetwork method and it’s application.Chinese J.Geophys.(inChinese),1998,41(2):242-251.
[17]Zhang Lin,Poulton M M,Wang T.Borehole electrical resistivitymodeling using neural networks.Geophysics,2002,67(6):1790-1791.
[18]Vjacheslav Spichak,Irina Popova.Artificial neural networkinversion of magnetotelluric data in termsof three-dimensionalearth macroparameters.Geophys.J.Int,2000(142):15-26.
[19]Carlos Caldero-Macas,Mrinal K.Sen,and Paul L.Stoffa.Articial neural networks for parameter estimationingeophysics.Geophysical Prospecting,2000,48(1):21-47.
[20]McCormack M,Zaucha D,Dushek D.First-break refractionevent picking and seismic data trace editing using neuralnetworks.Geophysics,1993,58:67-78.
[21]O.Sarzeaud,M.-F.LeQuentrec-Lalancette,D.Rouxel.Optimal Interpolation of Gravity Maps Using a Modi?edNeural Network.Math Geosci,2009(41):379-395.
[22]董长虹.Matlab神经网络与应用[M].北京:国防工业出版社,2007.Dong Chang-hong.Neural Networks and Applications forMatlab[M].Beijing:National Defence Industry Press,2007.
[23]王伟.人工神经网络原理——入门与应用[M].北京:北京航空航天大学出版社,1995:52-62.Wang Wei.Artificial Neural Networks——Introduction andApplications[M].Beijing:Beijing Aerospace UniversityPress,1995:52-62.
[24]叶斌,雷燕.关于BP网中隐含层层数及其节点数选取方法浅析.商丘职业技术学院学报,2004,15(6):52-60.YE Bin,Lei Yan.Analysis of Choosing the Number of theHiddenlayers and Its Nodes Number in Back PropagationNetwork.JOURNAL OF SHANGQIU VOCATIONAL ANDTECHNICAL COLLEGE,2004,15(6):52-60.
[25]李士勇.模糊控制、神经控制和智能控制论[M].哈尔滨:哈尔滨工业大学出版社,98-99.Li Shi-Yong.Fuzzy control,neural control and intelligentcontrol theory[M].Harbin:Harbin Institute of TechnologyPress,98-99.
[26]吕砚山.赵正琦.BP神经网络的优化及应用研究.北京化工大学学报,2001,28(1):67-69.LU Yan-shan,ZHAO Zheng-qi.Optimization and applicationresearch of BP neural network.JOURNAL OF BEIJINGUNIVERSITY OF CHEMICAL THECHNOLOGY,2001,28(1):67-69.
[2]杨立强,宋海滨,郝天珧.基于BP神经网络的波阻抗反演及应用.地球物理学进展,2005,20(1):34-37.YANG Li-qiang,SONG Hai-bin,HAO Tian-yao.Applicationof impedance inversion based on BP neural network.PROGRESS IN GEOPHYSICS,2005,20(1):34-37.
[3]Jordi,Turiel A,Garcia Ladona E,Font.Microcanonicalmultifractal formalism:Application to the estimation of oceansurface velocities.Journal of Geophysical Research,2007,112(5):80-97.
[4]张艺峰,姚道平,谢志招等.爆破震害预测的神经网络模型.地球物理学进展,2009,24(2):692-698.ZHANG Yi-feng,YAO Dao-ping,XIE Zhi-zhao.Neuralnetwork model for forecast of hazard by harm of blastingvibration.PROGRESS IN GEOPHYSICS,2009,24(2):692-698.
[5]谢林涛,付志红,谢品芳等.基于神经网络的视电阻率快速算法.地球物理学进展,2009,24(4):1527-1532.XIE Lin-tao,FU Zhi-hong,XIE Pin-fang.A fast algorithm ofapparent resistivity based on neural networks.PROGRESS INGEOPHYSICS,2009,24(2):692-698.
[6]唐秋华,刘保华,陈永奇等.结合遗传算法的LVQ神经网络在声学底质分类中的应用.地球物理学报,2007,50(1):313-319.TANG Qiu-Hua,LIU Bao-Hua,CHEN Yong-Qi.Applicationof LVQ neural network combined with the genetic algorithm inacoustic seafioor classification.Chinese J.Geophys.(inChinese),2007,50(1):313-319.
[7]张治国,杨毅恒,夏立显.自组织特征映射神经网络在测井岩性识别中的应用.地球物理学进展,2005,20(2):332-336.ZHANG Zhi-guo,YANG Yi-heng,XIA Li-xian.Theapplication of self-organizing feature map neural network tologging lithological identification.PROGRESS IN GEOPHYSICS,2005,20(2):332-336.
[8]李琼,李勇,李正文等.基于GA-BP理论的储层视裂缝密度地震非线性反演方法.地球物理学进展,2006,21(2):465-471.LI Qiong,LI Yong,LI Zheng-wen.A seismic nonlinearinversion for apparent fracture density of hydrocarbonreservoir based on GA-BP theory.PROGRESS INGEOPHYSICS,2006,21(2):465-471.
[9]徐海浪,吴小平.电阻率二维神经网络反演.地球物理学报,2006,49(2):584-589.XU Hai-Lang,WU Xiao-Ping.2-D resistivity inversion usingthe neural network method.Chinese J.Geophys.(in Chinese),2006,49(2):584-589.
[10]Gad El-Qady,Keisuke Ushijima.Inversion of DC resistivitydata using neural networks.Geophysical Prospecting,2001,49:417-430.
[11]朱自强,程方道,黄国祥.同时反演两个三维密度界面的拟神经网BP算法.石油物探,1995,34(1):76-85.Zhu Zi-Qiang,Cheng Fang-Dao,Huang Guo-xiang.Quasineural network BP algorithm for simultaneous inverstion oftwo 3-D density interfaces.geophysical prospecting forpetroleum,1995,34(1):76-85.
[12]朱晓军,申宁华.BP网络在位场解释中的应用.长春地质学院学报,1995,25(1):81-86.Zhu Xiao-jun,Shen Ning-hua.The application of BP networkto the interpretation of potential field data.Journal ofChangchun university of earth sciences.1995,25(1):81-86.
[13]张新兵,王家林.BP、Hopfield神经网络在位场反演中的应用比较.物探与化探计算技术,2007,29(2):161-166.Zhang Xin-bing,Wang Jia-lin.The application comparisons ofBPNN and HNN in potential field inversion.COMPUTINGTECHNIQUES FOR GEOPHYSICAL AND GEOCHEMICALEXPLORATION,2007,29(2):161-166.
[14]朱凯光,林君,韩悦慧等.基于神经网络的时间域直升机电磁数据电导率深度成像.地球物理学报,2010,53(5):743-750.ZHU Kai-Guang,LIN Jun,HAN Yue-Hui.Research onconductivity depth imaging of time domain helicopter-borneelectromagnetic data based on neural network.Chinese J.Geophys.(in Chinese),2010,53(5):743-750.
[15]Baan m,jutten C.Nueral networks in geophysical applications.Geophysics,2000,65(4):1032-1047.
[16]管志宁,侯俊生,黄临平.重磁异常反演的拟BP神经网络方法及其应用.地球物理学报,1998,41(2):242-251.Guan Zhi-ning,Hou Jun-sheng,Huang Lin-ping.Inversion ofgravity and magnetic anomalies using pseudo-BP neuralnetwork method and it’s application.Chinese J.Geophys.(inChinese),1998,41(2):242-251.
[17]Zhang Lin,Poulton M M,Wang T.Borehole electrical resistivitymodeling using neural networks.Geophysics,2002,67(6):1790-1791.
[18]Vjacheslav Spichak,Irina Popova.Artificial neural networkinversion of magnetotelluric data in termsof three-dimensionalearth macroparameters.Geophys.J.Int,2000(142):15-26.
[19]Carlos Caldero-Macas,Mrinal K.Sen,and Paul L.Stoffa.Articial neural networks for parameter estimationingeophysics.Geophysical Prospecting,2000,48(1):21-47.
[20]McCormack M,Zaucha D,Dushek D.First-break refractionevent picking and seismic data trace editing using neuralnetworks.Geophysics,1993,58:67-78.
[21]O.Sarzeaud,M.-F.LeQuentrec-Lalancette,D.Rouxel.Optimal Interpolation of Gravity Maps Using a Modi?edNeural Network.Math Geosci,2009(41):379-395.
[22]董长虹.Matlab神经网络与应用[M].北京:国防工业出版社,2007.Dong Chang-hong.Neural Networks and Applications forMatlab[M].Beijing:National Defence Industry Press,2007.
[23]王伟.人工神经网络原理——入门与应用[M].北京:北京航空航天大学出版社,1995:52-62.Wang Wei.Artificial Neural Networks——Introduction andApplications[M].Beijing:Beijing Aerospace UniversityPress,1995:52-62.
[24]叶斌,雷燕.关于BP网中隐含层层数及其节点数选取方法浅析.商丘职业技术学院学报,2004,15(6):52-60.YE Bin,Lei Yan.Analysis of Choosing the Number of theHiddenlayers and Its Nodes Number in Back PropagationNetwork.JOURNAL OF SHANGQIU VOCATIONAL ANDTECHNICAL COLLEGE,2004,15(6):52-60.
[25]李士勇.模糊控制、神经控制和智能控制论[M].哈尔滨:哈尔滨工业大学出版社,98-99.Li Shi-Yong.Fuzzy control,neural control and intelligentcontrol theory[M].Harbin:Harbin Institute of TechnologyPress,98-99.
[26]吕砚山.赵正琦.BP神经网络的优化及应用研究.北京化工大学学报,2001,28(1):67-69.LU Yan-shan,ZHAO Zheng-qi.Optimization and applicationresearch of BP neural network.JOURNAL OF BEIJINGUNIVERSITY OF CHEMICAL THECHNOLOGY,2001,28(1):67-69.
版权所有：© 2023 中国地质图书馆 中国地质调查局地学文献中心