CN104422969A - Method for reducing non-uniqueness of electromagnetic sounding inversion result - Google Patents
Method for reducing non-uniqueness of electromagnetic sounding inversion result Download PDFInfo
- Publication number
- CN104422969A CN104422969A CN201310397390.6A CN201310397390A CN104422969A CN 104422969 A CN104422969 A CN 104422969A CN 201310397390 A CN201310397390 A CN 201310397390A CN 104422969 A CN104422969 A CN 104422969A
- Authority
- CN
- China
- Prior art keywords
- depth
- inversion
- resistivity
- point
- feature
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Landscapes
- Geophysics And Detection Of Objects (AREA)
Abstract
The invention discloses a method for reducing the non-uniqueness of an electromagnetic sounding inversion result. The method comprises the following steps: acquiring the electromagnetic sounding data of an exploration area to obtain at least over two two-dimensional inversion depth-resistivity models; performing regularization to obtain regularized two-dimensional depth-resistivity model inversion results; determining a learning sample and a target sample; establishing a BP (Back Propagation) neural network system; inputting the inner section of the area into a neural network in sequence point after point and depth point after depth point to obtain the desired prediction results of each measuring point and each depth point; reducing the prediction results into a real inversion resistivity value to obtain an optimized two-dimensional inversion result. According to the processing method for rapidly optimizing the inversion non-uniqueness of electromagnetic sounding data, neural network processing is performed on a plurality of inversion results, and known information is learned, so that the non-uniqueness is reduced, and the method is practical and effective for obtaining a real model of a complex area target body.
Description
Technical field
The present invention relates to geophysical exploration method, is electromagnetic survey data treatment technology, specifically a kind of method reducing the nonuniqueness of electromagnetic sounding inversion result.
Background technology
Electromagnetic sounding is a kind of geophysical exploration method of the situation of change by understanding underground medium resistivity in the change of ground observation electromagnetic field.In recent years, theoretical research and computer technology fast development, the inversion method being converted to the degree of depth-resistivity data by the frequency-resistivity data gathered emerges in an endless stream, two-dimensional layered model inverting, the inverting of two dimension continuous medium, even 3-d inversion method etc. are also progressively grown up, computing velocity improves greatly, can than faster obtaining underground earth-electricity model, make electromagnetic sounding method in seismic prospecting difficulty district, as mountain front, pyrogenic rock exposure district, gravel area, top layer, carbonate rock area, loess tableland areal coverage achieves good exploration effects, compensate for the deficiency of seismic prospecting data.
But, no matter be two dimensional inversion or 3-d inversion, even if error of fitting is very little, the inversion result obtained often also not exclusively conforms to descending electrical structure practically, trace it to its cause or cause due to the nonuniqueness of Geophysics Inversion, namely geophysical survey data has multiple model corresponding with it, and the nonuniqueness of electromagnetic sounding inverting cannot be avoided especially.
The nonuniqueness how reducing inversion result is the direction that geophysicist makes great efforts always, wherein, most study be the optimization of inversion algorithm, but result can not be entirely satisfactory.This is the reason that electromagnetic sounding inverting achievement is difficult to meet needs of production always.
Summary of the invention
Be to provide order of the present invention a kind of obtain close to complex area objective body true model, reduce the method for electromagnetic sounding inversion result nonuniqueness.
The present invention is realized by following steps:
1) gather work area electromagnetic sounding data, different inversion procedure is done to every bar survey line, obtains more than at least 2 the two dimensional inversion degree of depth-resistivity models;
Described different inversion procedure adopt (Occam) inverting of two-dimentional Aukma or (RRI) inverting of two-dimentional fast relaxation or two-dimentional conjugate gradient inversion or two dimension to look mould inverting or two-dimentional regularization inverting or Two-Dimensional Generalized anti-inversion or two-dimensional analog annealing inverting.
2) regularization process is carried out to the step 1) two dimensional inversion degree of depth-resistivity models, obtain regularization two-dimensional depth-resistivity model inversion result;
Described regularization process be the two dimensional inversion degree of depth-resistivity models that all step 1) are obtained by ground within the scope of subterranean depth, according to identical depth interval, profile intervals, carry out interpolation processing, find out minimum value (RHOmin) and the maximal value (RHOmax) of all resistivity values, according to formula (RHOij-RHOmin)/(RHOmax-RHOmin), regularization is carried out to each resistivity value (RHOij).
Described subterranean depth can adjust depending on exploration targets body buried depth situation, is 5-20km; The best is 10km.
Described depth interval is 50-200m; The best is 100m.
Described section measuring point is spaced apart 50-200m; The best is 100m.
3) learning sample and target sample is determined;
Inversion result after regularization there is is the position of drilling well or seismic section data as given data point, in given data point environs, choose 1-3 point as learning sample;
Described learning sample given data point environs is 100-300m.
1-3 point of described learning sample is from the different inverse model of step 1).
If the well logging resistivity curve degree of depth is less than 10km, the mean value replacement of the inverse model that insufficient section step 1) is different, according to step 2) regularization process, according to step 2) depth range, step 2) depth interval carries out interpolation; Obtain well logging resistivity curve minimum value RHOmin, maximal value RHOmax, by each resistivity value RHOij according to step 2) formula carry out regularization, obtain the target sample of given data point.
If there is no drilling data, the seismic section after explaining is selected to set up strata division model, then formation resistivity master data is obtained according to exploratory area Physical Property Analysis, set up the resistivity models of layering, using on seismic section from the nearest point of electromagnetic sounding section or the resistivity hierarchical model of point that intersects with electromagnetic sounding as peg model, carry out regularization according to the process of above interpolation Sum fanction, obtain the target sample of given data point.
4) BP Neural Networks System is set up;
Described BP Neural Networks System of setting up is: the data of each for each learning sample depth point are input to neuroid successively, input target sample accordingly, start BP neuroid to train, if error of fitting reaches the error of setting, or reach maximum iterations and then stop training, just establish BP Neural Networks System.
Described BP Neural Networks System, the different two dimensional inversion degree of depth-resistivity models that different inversion method obtains, pattern number is input layer number; Hidden nodes is 10-15, and output layer neuron number is 1.
The data of described each depth point are the data of the different inverse model of step 1).
The error of described setting is 0.001.
Described maximum iterations is 1000 times.
5) by a section or many sections in work area, pointwise by depth point successively according to step 2) data after regularization process, be input to the neuroid set up by step 4) successively, obtain predict the outcome (Kij) of each depth point of each measuring point expected; RHOmin and the RHOmax value determined according to step 3), is reduced to real inverting resistivity value, that is, RHOij=RHOmin+Kij*(RHOmax-RHOmin), the two dimensional inversion result be finally optimized.
The present invention is a kind of disposal route of the inverting nonuniqueness rapid Optimum for electromagnetic sounding data, by the Processing with Neural Network to multiple inversion result, by the study to Given information, reduce nonuniqueness, for the true model obtaining complex area objective body provides the method for very practicability and effectiveness.
Accompanying drawing explanation
Fig. 1 is step schematic diagram of the present invention.
Embodiment
Below in conjunction with accompanying drawing and example in detail the present invention.
The concrete grammar step of the model optimization method of many inversion results is as follows:
1) exploratory area electromagnetic sounding data is gathered, following 5 kinds of invertings are done to every bar survey line: 1. two-dimentional Aukma (Occam) inverting, 2. two-dimentional fast relaxation (RRI) inverting, 3. two-dimentional conjugate gradient inversion, 4. two dimension looks mould inverting, 5. Two-Dimensional Generalized anti-inversion, obtains at least 5 the two dimensional inversion degree of depth-resistivity models;
2) regularization process is carried out to the 5 kinds of two dimensional inversion degree of depth-resistivity models of step 1), obtain the two-dimensional depth-resistivity model inversion result of 5 kinds of regularization;
Described regularization process be the 5 kinds of two dimensional inversion degree of depth-resistivity models that all step 1) are obtained by ground within the scope of subterranean depth, according to identical depth interval, profile intervals, carry out interpolation processing, find out minimum value (RHOmin) and the maximal value (RHOmax) of all resistivity values, according to formula (RHOij-RHOmin)/(RHOmax-RHOmin), regularization is carried out to each resistivity value (RHOij).
Described subterranean depth is chosen as 10km according to exploration targets body buried depth situation.
Described depth interval is 100m.
Described section measuring point is spaced apart 100m.
3) learning sample and target sample is determined;
Inversion result after regularization there is is the position of drilling well or seismic section data as given data point, in given data point environs, choose 3 points as learning sample;
Described learning sample given data point environs is 100-300m.
3 points of described learning sample are from the different inverse model of step 1).
If the well logging resistivity curve degree of depth is less than 10km, the mean value replacement of the inverse model that insufficient section step 1) is different, according to step 2) regularization process, according to step 2) depth range, step 2) depth interval carries out interpolation; Obtain well logging resistivity curve minimum value RHOmin, maximal value RHOmax, by each resistivity value RHOij according to step 2) formula carry out regularization, obtain the target sample of given data point.
If there is no drilling data, the seismic section after explaining is selected to set up strata division model, then formation resistivity master data is obtained according to exploratory area Physical Property Analysis, set up the resistivity models of layering, using on seismic section from the nearest point of electromagnetic sounding section or the resistivity hierarchical model of point that intersects with electromagnetic sounding as peg model, carry out regularization according to the process of above interpolation Sum fanction, obtain the target sample of given data point.
4) BP Neural Networks System is set up;
Described BP Neural Networks System of setting up is: the data of each for each learning sample depth point are input to neuroid successively, input target sample accordingly, start BP neuroid to train, if error of fitting reaches the error of setting, or reach maximum iterations and then stop training, just establish BP Neural Networks System.
Described BP Neural Networks System, the different two dimensional inversion degree of depth-resistivity models that 5 kinds of different inversion methods obtain, pattern number is input layer number; Hidden nodes is 15, and output layer neuron number is 1.
The data of described each depth point are the data of the different inverse model of step 1).
The error of described setting is 0.001.
Described maximum iterations is 1000 times.
Described BP(Back Propagation) network be 1986 by headed by Rumelhart and McCelland scientist group propose, being a kind of Multi-layered Feedforward Networks by Back Propagation Algorithm training, is one of current most widely used neural network model.BP network can learn and store a large amount of input-output mode map relations, and without the need to disclosing the math equation describing this mapping relations in advance.Its learning rules use method of steepest descent, constantly adjusted the weights and threshold of network, make the error sum of squares of network minimum by backpropagation.BP neural network model topological structure comprises input layer (input), hidden layer (hide layer) and output layer (output layer).In Matblab built-in function, there is neuron built-in function, can directly quote.
5) by a section or many sections in work area, pointwise by depth point successively according to step 2) data after regularization process, be input to the neuroid set up by step 4) successively, obtain predict the outcome (Kij) of each depth point of each measuring point expected; RHOmin and the RHOmax value determined according to step 3), is reduced to real inverting resistivity value, that is, RHOij=RHOmin+Kij*(RHOmax-RHOmin), the two dimensional inversion result be finally optimized.
Claims (13)
1. reduce a method for electromagnetic sounding inversion result nonuniqueness, feature is realized by following steps:
1) gather work area electromagnetic sounding data, different inversion procedure is done to every bar survey line, obtains more than at least 2 the two dimensional inversion degree of depth-resistivity models;
2) regularization process is carried out to the step 1) two dimensional inversion degree of depth-resistivity models, obtain regularization two-dimensional depth-resistivity model inversion result;
Described regularization process be the two dimensional inversion degree of depth-resistivity models that all step 1) are obtained by ground within the scope of subterranean depth, according to identical depth interval, profile intervals, carry out interpolation processing, find out minimum value RHOmin and the maximal value RHOmax of all resistivity values, according to following formula, regularization is carried out to each resistivity value RHOij: RHOij=(RHOij-RHOmin)/(RHOmax-RHOmin);
3) learning sample and target sample is determined;
The position of drilling well or seismic section data will be had in inversion result after regularization as given data point, in given data point environs, choose 1-3 point as learning sample;
4) BP Neural Networks System is set up;
Described BP Neural Networks System of setting up is: the data of each for each learning sample depth point are input to neuroid successively, input target sample accordingly, start BP neuroid to train, if error of fitting reaches the error of setting, or reach maximum iterations and then stop training, just establish BP Neural Networks System;
5) by a section or many sections in work area, pointwise by depth point according to step 2) data after regularization process, be input to the neuroid set up by step 4) successively, obtain the Kij that predicts the outcome of each depth point of each measuring point expected; RHOmin and the RHOmax value determined according to step 3), is reduced to real inverting resistivity value, the two dimensional inversion result be finally optimized.
2. method according to claim 1, to be the different inversion procedure described in step 1) be feature adopts the inverting of two-dimentional Aukma or two-dimentional fast relaxation inverting or two-dimentional conjugate gradient inversion or two dimension to look mould inverting or two-dimentional regularization inverting or Two-Dimensional Generalized anti-inversion or two-dimensional analog annealing inverting.
3. method according to claim 1, feature is step 2) described in subterranean depth can adjust depending on exploration targets body buried depth situation, be 5-20km; The best is 10km.
4. method according to claim 1, feature is step 2) described in depth interval be 50-200m; The best is 100m.
5. method according to claim 1, feature is step 2) described in section measuring point be spaced apart 50-200m; The best is 100m.
6. method according to claim 1, feature is the learning sample given data point environs described in step 3) is 100-300m.
7. method according to claim 1, feature is that 1-3 point of learning sample described in step 3) is from the different inverse model of step 1).
8. method according to claim 1, feature is step 2) the curve degree of depth of resistivity less than 10km time, the mean value of the inverse model that insufficient section step 1) is different replaces, according to step 2) regularization process, according to step 2) depth range and step 2) depth interval carries out interpolation; Obtain the well logging minimum value RHOmin of resistivity curve and maximal value RHOmax, by each resistivity value RHOij according to step 2) formula carry out regularization, obtain the target sample of given data point.
9. method according to claim 1, if feature is step 3), drilling data does not have, the seismic section after explaining is selected to set up strata division model, then formation resistivity master data is obtained according to exploratory area Physical Property Analysis, set up the resistivity models of layering, using on seismic section from the nearest point of electromagnetic sounding section or the resistivity hierarchical model of point that intersects with electromagnetic sounding as peg model, carry out regularization according to the process of above interpolation Sum fanction, obtain the target sample of given data point.
10. method according to claim 1, feature is the BP Neural Networks System described in step 4), the different two dimensional inversion degree of depth-resistivity models that different inversion method obtains, and pattern number is input layer number; Hidden nodes is 10-15, and output layer neuron number is 1.
11. methods according to claim 1, the data of feature to be the data of each depth point described in step 4) the be different inverse model of step 1).
12. methods according to claim 1, feature is the error of setting described in step 4) is 0.001.
13. methods according to claim 1, feature is the maximum iterations described in step 4) is 1000 times.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310397390.6A CN104422969B (en) | 2013-09-04 | 2013-09-04 | A kind of method for reducing electromagnetic sounding inversion result nonuniqueness |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310397390.6A CN104422969B (en) | 2013-09-04 | 2013-09-04 | A kind of method for reducing electromagnetic sounding inversion result nonuniqueness |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104422969A true CN104422969A (en) | 2015-03-18 |
| CN104422969B CN104422969B (en) | 2017-03-15 |
Family
ID=52972514
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310397390.6A Active CN104422969B (en) | 2013-09-04 | 2013-09-04 | A kind of method for reducing electromagnetic sounding inversion result nonuniqueness |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104422969B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104777516A (en) * | 2015-04-15 | 2015-07-15 | 国网重庆市电力公司电力科学研究院 | Apparent resistivity calculating method on basis of non-linear equation solution modular form |
| CN106842344A (en) * | 2017-04-24 | 2017-06-13 | 中国科学院电子学研究所 | The method of the unmanned plane boat magnetic holoaxial gradient magnetic disturbance compensation based on feedforward network |
| CN109188536A (en) * | 2018-09-20 | 2019-01-11 | 成都理工大学 | Time-frequency electromagnetism and magnetotelluric joint inversion method based on deep learning |
| CN109343134A (en) * | 2018-11-27 | 2019-02-15 | 中煤科工集团西安研究院有限公司 | A kind of Transient Electromagnetic Method in Mine data analysis interpretation method and system |
| CN109557601A (en) * | 2019-01-22 | 2019-04-02 | 青岛海洋地质研究所 | Reservoir parameter inversion method is combined in one-dimensional ocean controllable source electromagnetism and earthquake |
| CN111126591A (en) * | 2019-10-11 | 2020-05-08 | 重庆大学 | Magnetotelluric deep neural network inversion method based on space constraint technology |
| CN112699596A (en) * | 2020-12-04 | 2021-04-23 | 湖南工商大学 | Wide-area electromagnetic method induced polarization information nonlinear extraction method based on learning |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101008678A (en) * | 2006-01-26 | 2007-08-01 | 中国石油天然气集团公司 | 3D resistivity cascaded automatic imaging method |
| CA2659362A1 (en) * | 2006-08-22 | 2008-02-28 | Kjt Enterprises, Inc. | Fast 3d inversion of electromagnetic survey data using a trained neural network in the forward modeling branch |
| CN101650443A (en) * | 2009-09-15 | 2010-02-17 | 重庆大学 | Back-propagation network calculating method of apparent resistivity |
| US20120080197A1 (en) * | 2009-06-26 | 2012-04-05 | Dickens Thomas A | Constructing Resistivity Models From Stochastic Inversion |
-
2013
- 2013-09-04 CN CN201310397390.6A patent/CN104422969B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101008678A (en) * | 2006-01-26 | 2007-08-01 | 中国石油天然气集团公司 | 3D resistivity cascaded automatic imaging method |
| CA2659362A1 (en) * | 2006-08-22 | 2008-02-28 | Kjt Enterprises, Inc. | Fast 3d inversion of electromagnetic survey data using a trained neural network in the forward modeling branch |
| US20120080197A1 (en) * | 2009-06-26 | 2012-04-05 | Dickens Thomas A | Constructing Resistivity Models From Stochastic Inversion |
| CN101650443A (en) * | 2009-09-15 | 2010-02-17 | 重庆大学 | Back-propagation network calculating method of apparent resistivity |
Non-Patent Citations (4)
| Title |
|---|
| S.RATNAJEEVAN H. HOOLE: "Artificial Neural Networks in the Solution of Inverse Electromagnetic Field Problems", 《IEEE TRANSACTIONS ON MAGNETICS》 * |
| UPENDRA K. SINGH,ET AL.: "Neural network modeling and prediction of resistivity structures using VES Schlumberger data over a geothermal area", 《COMPUTERS & GEOSCIENCES》 * |
| 李创社等: "瞬变电磁勘探中的人工神经网络反演法", 《西安交通大学学报》 * |
| 谢品芳等: "基于BP网络的视电阻率快速计算", 《重庆大学学报》 * |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104777516A (en) * | 2015-04-15 | 2015-07-15 | 国网重庆市电力公司电力科学研究院 | Apparent resistivity calculating method on basis of non-linear equation solution modular form |
| CN106842344A (en) * | 2017-04-24 | 2017-06-13 | 中国科学院电子学研究所 | The method of the unmanned plane boat magnetic holoaxial gradient magnetic disturbance compensation based on feedforward network |
| CN106842344B (en) * | 2017-04-24 | 2018-10-12 | 中国科学院电子学研究所 | The method of unmanned plane boat magnetic holoaxial gradient magnetic disturbance compensation based on feedforward network |
| CN109188536A (en) * | 2018-09-20 | 2019-01-11 | 成都理工大学 | Time-frequency electromagnetism and magnetotelluric joint inversion method based on deep learning |
| CN109188536B (en) * | 2018-09-20 | 2020-06-05 | 成都理工大学 | Time-frequency electromagnetic and magnetotelluric joint inversion method based on deep learning |
| CN109343134A (en) * | 2018-11-27 | 2019-02-15 | 中煤科工集团西安研究院有限公司 | A kind of Transient Electromagnetic Method in Mine data analysis interpretation method and system |
| CN109557601A (en) * | 2019-01-22 | 2019-04-02 | 青岛海洋地质研究所 | Reservoir parameter inversion method is combined in one-dimensional ocean controllable source electromagnetism and earthquake |
| CN111126591A (en) * | 2019-10-11 | 2020-05-08 | 重庆大学 | Magnetotelluric deep neural network inversion method based on space constraint technology |
| CN112699596A (en) * | 2020-12-04 | 2021-04-23 | 湖南工商大学 | Wide-area electromagnetic method induced polarization information nonlinear extraction method based on learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104422969B (en) | 2017-03-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109709603B (en) | Seismic horizon identification and tracking method and system | |
| CN111126591B (en) | Magnetotelluric deep neural network inversion method based on space constraint technology | |
| CN104422969A (en) | Method for reducing non-uniqueness of electromagnetic sounding inversion result | |
| Jiang et al. | Electrical resistivity imaging inversion: An ISFLA trained kernel principal component wavelet neural network approach | |
| CN113610945B (en) | Ground stress curve prediction method based on hybrid neural network | |
| CN109541685B (en) | River channel sand body identification method | |
| CN103454677B (en) | Based on the earthquake data inversion method that population is combined with linear adder device | |
| Sarkheil et al. | Fractures distribution modeling using fractal and multi-fractal–neural network analysis in Tabnak hydrocarbon field, Fars, Iran | |
| CN115047531A (en) | Transient electromagnetic data inversion method based on ELM network | |
| CN113917560A (en) | Three-dimensional gravity magnetic-electric shock multi-parameter collaborative inversion method | |
| CN115508900A (en) | Ground dragging type transient electromagnetic imaging method and system | |
| Rastegarnia et al. | A comparative study of 3D FZI and electrofacies modeling using seismic attribute analysis and neural network technique: A case study of Cheshmeh-Khosh Oil field in Iran | |
| Ho | 3-D inversion of borehole-to-surface electrical data using a back-propagation neural network | |
| CN104297792B (en) | The phased inversion method of water channel reservoir is stacked on a kind of fan | |
| Azim | Estimation of fracture network properties from FMI and conventional well logs data using artificial neural network | |
| CN105589096B (en) | A kind of sedimentary facies belt division methods based on D-S evidence theory | |
| Shiroodi et al. | Shear wave prediction using committee fuzzy model constrained by lithofacies, Zagros basin, SW Iran | |
| Gyulai et al. | A quick 2-D geoelectric inversion method using series expansion | |
| CN115421181B (en) | Three-dimensional geological model phase control attribute modeling method based on deep learning | |
| Jia et al. | Magnetotelluric Closed-Loop Inversion | |
| Srinivas et al. | Estimation of subsurface strata of earth using adaptive neuro-fuzzy inference system (ANFIS) | |
| CN115407397A (en) | Rayleigh wave frequency dispersion curve supervised learning inversion method and system | |
| CN115880455A (en) | Three-dimensional intelligent interpolation method based on deep learning | |
| Isaev et al. | Neural network recovery of missing data of one geophysical method from known data of another one in solving inverse problems of exploration geophysics | |
| Malvić et al. | Neural networks in petroleum geology as interpretation tools |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant |