CN105467449B - Deep layer Thin oil sandwich quantitatively characterizing method based on earthquake classification Sensitive Attributes fusion - Google Patents
Deep layer Thin oil sandwich quantitatively characterizing method based on earthquake classification Sensitive Attributes fusion Download PDFInfo
- Publication number
- CN105467449B CN105467449B CN201410449177.XA CN201410449177A CN105467449B CN 105467449 B CN105467449 B CN 105467449B CN 201410449177 A CN201410449177 A CN 201410449177A CN 105467449 B CN105467449 B CN 105467449B
- Authority
- CN
- China
- Prior art keywords
- classification
- attribute
- fusion
- earthquake
- deep layer
- 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.)
- Active
Links
Landscapes
- Geophysics And Detection Of Objects (AREA)
Abstract
A kind of deep layer Thin oil sandwich quantitatively characterizing method based on earthquake classification Sensitive Attributes fusion, effectively increases the efficiency for deep layer Thin oil sandwich seismic attributes analysis, substantially increases precision of prediction.This programme includes：When big under the geologic setting of window section, for the reasonable hour window section attribute of the geologic objective subregion extraction of interval of interest, Favorable Reservoir finely carved, finally merged with overall background, obtains the attribute forecast figure for geologic objective.This method includes two committed steps：When under big geologic setting, using the more attribute correlation dimensionality reductions of common seismic, for interval of interest selected objective target Sensitive Attributes.First it is classified second, being taken based on earthquake classification attribute：Determine the rational hour window section in target area under different background；Merge again：Divide Sensitive Attributes reservoir scope in fusion quantitative forecast research area under sequence level hour window parameter.Wherein described classification and fusion are using the real SVMs bored under data constraint（SVM）Algorithm is realized.
Description
Technical field
The present invention relates to geophysical exploration and comprehensive study field, more particularly to a kind of earthquake classification sensitivity that is based on to belong to
Property fusion method prediction deep layer Thin oil sandwich.
Background technology
Seismic properties are to characterize the physical quantity of seismic wave geometric shape, kinematics character, dynamic characteristic and statistical nature,
It is an importance for carrying out reservoir prediction in current oil exploration comprehensive study.Due to being confined to use single attribute
Predicting reservoir, often multi-solution is very strong for prediction result, and this multiresolution issue can be improved by more attribute fusion methods.
On the basis of well-log information, Study of Seismic attributive character, sensitiveness of each attribute to reservoir is analyzed, using more attribute fusion sides
Several properties are blended in one by method, are recycled well location to calculate the fusion proportion of various seismic properties, are predicted Favorable Reservoir
Development position.The thinking organically combines the advantages of each attribute, can effectively reduce the multi-solution of reservoir prediction result.But with
The intensification of depth of stratum, the change of occurrence and reservoir thickness it is thinning, the precision of prediction of more attributes fusions can substantially reduce.
The content of the invention
The purpose of the present invention is to be directed to the problem of current art is relatively low to deep layer thin-inter bed reservoir prediction precision, there is provided a kind of
Be classified using earthquake (zone be classified and when window be classified) the deep layer Thin oil sandwich based on earthquake classification Sensitive Attributes fusion quantify
Characterizing method, the efficiency for deep layer Thin oil sandwich seismic attributes analysis is effectively increased, substantially increases precision of prediction.Especially
Bury that depth, fault block are complicated, reservoir thickness is thin, border is difficult to the area that determines suitable for reservoir.
The purpose of the present invention can be achieved by the following technical measures：
Deep layer Thin oil sandwich quantitatively characterizing method based on earthquake classification Sensitive Attributes fusion：The geologic setting of window section when big
Under, for the reasonable hour window section attribute of the geologic objective subregion extraction of interval of interest, Favorable Reservoir is finely carved
Carve, finally merged with overall background, obtain the attribute forecast figure for geologic objective.This method includes two committed steps：When
Under big geologic setting, using the more attribute correlation dimensionality reductions of common seismic, for interval of interest selected objective target Sensitive Attributes.Second, it is based on
Earthquake classification, which is taken attribute, to be first classified：Determine the rational hour window section in target area under different background；Merge again：Divide sequence level small
When window parameter under Sensitive Attributes reservoir scope in fusion quantitative forecast research area.Wherein described classification and fusion is bored using real
SVMs (SVM) algorithm under data constraint is realized.
The concrete technical scheme of invention：
1st, the sedimentation setting of surge well shake high frequency sequence division, to the more attribute correlation dimensionality reductions of common seismic, to interval of interest
It is preferred to carry out attribute.Conventional more attribute fusion figures are completed, predict deep layer Thin oil sandwich.
2. on the basis of conventional more attribute forecast reservoirs, classification processing is carried out to whole district's reservoir.It is a wide range of large-scale
Region, the occurrence of reservoir change greatly, and the prediction of window can not meet the fine engraving of small range sand body when big, therefore the whole district is entered
Row classification is handled, including：Zone be classified and when window be classified.Zone classification is special with reference to deposition mainly using well logging, drilling data
Sign, on the basis of drilling well, it would be desirable to which the reservoir portrayed in detail is determined, marks off region in the plane；When window be classified
Principle be that the place for comparing concentration in sand-body distribution can use same hour window, can be further in the place do not concentrated
Target area true timing window again is divided, until can be sand body meticulous depiction.But the scope of hour window must when big window scope
It is interior, it is impossible to exceed, main purpose is for the border of finer description sand body.
3rd, it is determined that on the basis of classification, reservoir when to each graded region band and being classified in window carries out classification prediction.Its
Algorithm includes two stages:Training stage and Forecasting recognition stage.Training stage is based on SVMs primarily to establishing
(SVM) more attribute situation identification models, are mainly comprised the following steps：To seismic properties training, attribute information is concentrated to be normalized：
Extract characteristic vector.Establish seismic properties characteristic vector training set：Thus SVMs (SVM) classification mould is established based on
Type；Cognitive phase mainly comprises the following steps：The seismic attributes information to be classified is normalized；Extract its characteristic vector.Wherein divide
Level establishes during SVMs (SVM) identification model the constraint for being also added into real drilling well：Add when extracting attribute feature vector
Enter the constraint of drilling well reservoir parameter, be that prediction result and the real drilling well goodness of fit are higher.
4th, carry out second to attribute of the subregion with timesharing window to merge, obtain whole district's hierarchical nature fusion attribute.In computing
Take some sampled points at random to preferred attribute, carry out classification function f (x) computing, calculate classification threshold values, it is former according to optimizing
Then, i.e., the threshold values can correctly separate the different classifications in attribute, and consideration of compromising makes sample mistake point rate minimum and classification
Interval is maximum.Constantly modification classification threshold values, until the sample point tolerance in different classifications is all converged in acceptable scope.This
The fusion results that sample obtains avoid the multi-solution of more attribute fusions, make prediction result more accurate.
5. the attributive classification result being classified based on earthquake under sensitive algorithm overall background completed：It is further for classification results
The attribute forecast that target carries out hour window is segmented, is melted again with reacting the primitive attribute information of background by the means of attribute optimization
Close, obtain earthquake hierarchical nature fusion forecasting figure.
6. combining the real reservoir parameter for boring data using earthquake hierarchical nature fusion reservoir prediction map, semidefinite quantifies pre-
The areal extent of the thin reservoir of deep layer is measured, makes prediction result and the real drilling well goodness of fit higher, more accurately, as a result can be applicable.
The seismic attributes information to be classified is normalized in such scheme step (3) cognitive phase, extracts its feature
Vector；The computing for carrying out classification function is inputted using the vector as SVMs (SVM), each supporting vector corresponds to one point
Class threshold values b, one kind is classified as less than special value to classification threshold values b, determines categorised demarcation line according to this；Extract attribute feature vector
When add the constraint of drilling well reservoir parameter, compare analysis by seismic properties under seismic attributes information near well point and overall background,
Constantly calculate and restrain, the extraction of characteristic vector in Statistical error modeling process, model is reduced closer to actual geological condition
Multi-solution.
Such scheme step (4) carries out second to attribute of the subregion with timesharing window and merged, and obtains whole district's hierarchical nature and melts
Attribute is closed, takes some sampled points at random to preferred attribute, carries out classification function f (x) computing, classification threshold values is calculated, according to most
Optimization principles, the i.e. threshold values can correctly separate the different classifications in attribute, and consideration of compromising makes sample mistake point rate minimum
It is maximum with class interval, classification threshold values is constantly changed, until the sample point tolerance in different classifications all converges to acceptable model
In enclosing；In this process, it is trained by inspection of hierarchy model, if obtaining characteristic vector of the Ganlei based on classification function, often
The all corresponding classification set in seismic properties of category feature vector, so far assorting process terminates；Whole process is with supporting vector
Machine (SVM) is core algorithm, and the implementation of algorithm is as follows：
H is classification line, meets linear equation：W*x+b=0, H1, H2 be respectively it is all kinds of in from the nearest sample of classification line
And it is referred to as class interval parallel to classification line H straight line, the distance between they；So-called optimal classification line is exactly to require classification
Line not only can correctly separate two classes, that is, it is 0 to train error rate, and make the ‖ w ‖ of class interval 2/ maximum；Linear separable sets
Close (xi,yi) to classification line equation be standardized after meet yi (w*xi+b) -1 >=0i=1 ... n y ∈ { -1,1 }, make
The ‖ w ‖ of class interval 2/ are maximum of equal value with making ‖ w ‖2Minimum, formula and make ‖ w ‖ before satisfaction2/ 2 minimum classifying faces are referred to as optimal classification
Face, the sample point on H1, H2 are referred to as supporting vector, ask for optimal classification surface and namely ask for optimal (w, b) in linear equation.I.e.
In yi (w*xi+b) -1 >=0i=1 ... ‖ w ‖ are sought under n y ∈ { -1,1 } constraint2/ 2 minimum value；It is defined as follows
Lagrange functions：
Wherein ai>=0 is Lagrange coefficient；Seek ‖ w ‖2/ 2 minimum is exactly to w, and b seeks the minimum of Lagranian, right
L functions do differential and make it be equal to 0, Lagranian and are converted to：In ∑ yiai=0, ai>=0, i=1 ... seek glug under n constraints
Bright day coefficient aiTo function：
Its maximum optimal solution meets ai(yi(w*xi+ b) -1)=0, i=1 ... n, a in solutioniThe sample x being not zeroi,yi
That is the sample point in supporting vector, that is, H1, H2, therefore W can be reduced to：
The optimization class function obtained after solution is：
Wherein sgn is classification function, b is classification threshold values, and x is given sample, xi, yi are linear separability sample set；For
Unknown sample x need to only calculate classification function, you can judge to classify belonging to x, in the case of linearly inseparable：1 only need to be added
Relaxation option ξi>=0, sample set meets yi (w*xi+b) -1+ ξi>=0i=1 ... n, the conversion of function and calculating process with it is linear
It is similar, but constraints is changed into：∑yiai=0,0≤ai≤ c, i=1 ... n, c are the constant that control mistake divides that sample punishes degree,
Also respective change is classification function
K is interior Product function appropriate in optimal classification surface, and other specification is identical with linear case.
The effect of invention：This method buries depth for reservoir, and occurrence changes greatly, and fault block is complicated, the thin sand body of reservoir thickness,
Propose zone classification and when window classification method, by after classification it is multiattribute extraction and addition drilling well information melt again
Close, accurate can realize the prediction of deep layer Thin oil sandwich, solve seismic properties and predict difficulty for deep layer thin interbed
Problem, more can accurately obtain deep seismic attribute information realize prediction, preferably to oil exploration seismic interpretation provide according to
According to there is good application effect and promote future, effectively increase the precision of deep layer Thin oil sandwich seismic attributes analysis, instruct
Exploration and exploitation.
Brief description of the drawings
Flow chart is embodied in a kind of embodiment that Fig. 1 is the present invention；
Fig. 2 illustrates for a kind of optimal classification surface；
Fig. 3 is the hierarchical objectives area schematic marked off using zone stage division；
The hour window for the target area that window stage division marks off when Fig. 4 is utilized；
Hierarchical nature fusion figures of the Fig. 5 to hour window section in the target area that is obtained using hierarchical nature；
Attribute of Fig. 6 subregions with timesharing window carries out merging obtained whole district's earthquake classification Sensitive Attributes fusion figure for the second time.
Embodiment
It is cited below particularly go out preferred embodiment, and coordinate institute's accompanying drawings, be described in detail below.
Based on the deep layer Thin oil sandwich quantitatively characterizing method of earthquake classification Sensitive Attributes fusion, its specific implementation process is：
1. the sedimentation setting of surge well shake high frequency sequence division, to the more attribute correlation dimensionality reductions of common seismic, to interval of interest
It is preferred to carry out attribute.Conventional more attribute fusion figures are completed, predict deep layer Thin oil sandwich.
2. on the basis of conventional more attribute forecast reservoirs, classification processing is carried out to whole district's reservoir.It is a wide range of large-scale
Region, the occurrence of reservoir change greatly, and the prediction of window can not meet the fine engraving of small range sand body when big, therefore the whole district is entered
Row classification is handled, including：Zone be classified and when window be classified.Zone classification is special with reference to deposition mainly using well logging, drilling data
Sign, on the basis of drilling well, it would be desirable to which the reservoir portrayed in detail is determined, marks off region in the plane；When window be classified
Principle be that the place for comparing concentration in sand-body distribution can use same hour window, can be further in the place do not concentrated
Target area true timing window again is divided, until can be sand body meticulous depiction.But the scope of hour window must when big window scope
It is interior, it is impossible to exceed, main purpose is for the border of finer description sand body.
3. it is determined that on the basis of classification, reservoir when to each graded region band and being classified in window carries out classification prediction.Its
Algorithm includes two stages:Training stage and Forecasting recognition stage.Training stage is based on SVMs primarily to establishing
(SVM) more attribute situation identification models, are mainly comprised the following steps：To seismic properties training, attribute information is concentrated to be normalized：Carry
Take characteristic vector.Establish seismic properties characteristic vector training set：Thus SVMs (SVM) disaggregated model is established based on.
Cognitive phase mainly comprises the following steps：The seismic attributes information to be classified is normalized；Extract its characteristic vector.The vector is made
The computing for carrying out classification function, the corresponding classification threshold values b of each supporting vector are inputted for SVMs (SVM), to classification
Threshold values b is classified as one kind less than special value, determines categorised demarcation line according to this.Wherein SVMs (SVM) knowledge is established in classification
The constraint of real drilling well is also added into other model process：The constraint of drilling well reservoir parameter is added when extracting attribute feature vector, is led to
Crossing well point, nearby seismic attributes information compares analysis with seismic properties under overall background, constantly calculates convergence, Statistical error modeled
The extraction of characteristic vector in journey, model is set to reduce multi-solution closer to actual geological condition.The core in two stages is all branch
Hold vector machine (SVM).
Merged 4. attribute of pair subregion with timesharing window carries out second, obtain whole district's hierarchical nature fusion attribute.Its algorithm
It is as follows：Take some sampled points at random to preferred attribute, carry out classification function f (x) computing, classification threshold values is calculated, according to optimal
Change principle, i.e. the threshold values can correctly separate the different classifications in attribute, and consideration of compromising make sample mistake point rate minimum with
Class interval is maximum.Constantly modification classification threshold values, until the sample point tolerance in different classifications all converges to acceptable scope
It is interior.In this process, it is trained by inspection of hierarchy model, if obtaining characteristic vector of the Ganlei based on classification function, per class
Characteristic vector all corresponding classification set in seismic properties, so far assorting process terminates.Whole process is with SVMs
(SVM) it is core algorithm：It is developed from the optimal classification surface in the case of linear separability, mainly to pattern classification and non-
Linear regression, it overcomes such as local optimum present in conventional neutral net, overlearning, the problems such as unstable networks,
Be FA algorithm in statistical learning and artificial intelligence, it is heterogeneous linear divide with can not be in two kinds of situation.The realization of algorithm
Journey is as follows：
Fig. 2 is a two-dimensional diagram：In figure ▲ and ● 2 class samples are represented, H is classification line, meets linear equation：w*x+b
=0.(i.e. under the conditions of linear separability) H1, H2 be respectively it is all kinds of in from the nearest sample of classification line and parallel to classification line H
Straight line, the distance between they are referred to as class interval (margin).So-called optimal classification line is exactly to require that classification line not only can
Two classes are correctly separated, that is, it is 0 to train error rate；And make the ‖ w ‖ of class interval 2/ maximum.Linear separable sets closes (xi,yi)
Meet after being standardized to classification line equation：Yi (w*xi+b) -1 >=0i=1 ... n y ∈ { -1,1 }, make class interval
2/ ‖ w ‖ are maximum of equal value with making ‖ w ‖2It is minimum.Formula and make ‖ w ‖ before satisfaction2/ 2 minimum classifying faces we be referred to as optimal classification surface.H1、
Sample point on H2 is referred to as supporting vector.Ask for optimal classification surface and namely ask for optimal (w, b) in linear equation.I.e. in yi
(w*xi+b) -1 >=0i=1 ... ‖ w ‖ are sought under n y ∈ { -1,1 } constraint2/ 2 minimum value.The Lagrange letters being defined as follows
Number：
Wherein ai>=0 is Lagrange coefficient.Seek ‖ w ‖2/ 2 minimum is exactly to w, and b seeks the minimum of Lagranian.It is right
L functions do differential and make it be equal to 0, Lagranian and are converted to：In ∑ yiai=0, ai>=0, i=1 ... seek glug under n constraints
Bright day coefficient aiTo function：
Maximum.Its optimal solution meets ai(yi(w*xi+ b) -1)=0, i=1 ... n.A in solutioniThe sample x being not zeroi,
yiThat is the sample point in supporting vector, that is, H1, H2.Therefore W can be reduced to：
The optimization class function obtained after solution is：
Wherein sgn is classification function, b is classification threshold values, and x is given sample, xi, yi are linear separability sample set, for
Unknown sample x need to only calculate classification function.It can determine whether to classify belonging to x.In the case of linearly inseparable：1 only need to be added
Relaxation option ξi>=0, sample set meets yi (w*xi+b) -1+ ξi>=0i=1 ... n, the conversion of function and calculating process with it is linear
It is similar, but constraints is changed into：∑yiai=0,0≤ai≤ c, i=1 ... n, c are the constant that control mistake divides that sample punishes degree,
Also respective change is classification function:
K is interior Product function appropriate in optimal classification surface, and other specification is identical with linear case.
5. the attributive classification result being classified based on earthquake under sensitive algorithm overall background completed：It is further for classification results
The attribute forecast that target carries out hour window is segmented, is melted again with reacting the primitive attribute information of background by the means of attribute optimization
Close, obtain earthquake hierarchical nature fusion forecasting figure.
6. combining the real reservoir parameter for boring data using earthquake hierarchical nature fusion reservoir prediction map, semidefinite quantifies pre-
The areal extent of the thin reservoir of deep layer is measured, makes prediction result and the real drilling well goodness of fit higher, more accurately, as a result can be applicable.
According to the method described above, depth is buried for reservoir, occurrence changes greatly, and fault block is complicated, the thin sand body of reservoir thickness, proposes
Zone classification and when window classification method, obtained hierarchical objectives region (Fig. 3) and classification hour window (Fig. 4).By dividing
The information of multiattribute extraction and addition drilling well merges again after level, effectively increases the precision of reservoir prediction, has obtained subregion
Band attribute forecast figure (Fig. 5).By the reservoir parameter quantitatively characterizing method of neural network function approximate algorithm, in subregion band attribute
On the basis of prediction, whole district's attribute merge, finally give whole district's earthquake classification Sensitive Attributes prognostic chart (Fig. 6), partly
The border for predicting the thin reservoir of deep layer of quantification, the deployment of the oil field development well for instructing next step are relatively reliable.
Claims (3)
1. based on the deep layer Thin oil sandwich quantitatively characterizing method of earthquake classification Sensitive Attributes fusion, it is characterized in that：
（1）The sedimentation setting of surge well shake high frequency sequence division, related dimensionality reduction is carried out to the more attributes of common seismic, to interval of interest
It is preferred to carry out attribute, completes conventional more attribute fusion figures, predicts deep layer Thin oil sandwich；
（2）On the basis of conventional more attribute forecast reservoirs, classification processing is carried out to whole district's reservoir, including：Zone be classified and when
Window is classified, wherein, zone classification is using well logging, drilling data, with reference to deposition characteristicses, on the basis of drilling well, it would be desirable to
The reservoir portrayed in detail is determined, marks off region in the plane；When window classification be the place of comparing concentration in sand-body distribution
With same hour window, in the further division target area in place do not concentrated true timing window again, until can sand body is fine
Portray；
（3）It is determined that on the basis of classification, reservoir when to each graded region band and being classified in window carries out classification prediction, its algorithm
Include training stage and Forecasting recognition stage；Wherein, the training stage mainly comprises the following steps：To seismic properties training, concentrate attribute letter
Breath is normalized, extract characteristic vector, establish seismic properties characteristic vector training set, thus based on establish SVMs
(SVM) disaggregated model；Cognitive phase mainly comprises the following steps：The seismic attributes information to be classified is normalized, extracts its feature
Vector；Wherein the constraint for being also added into real drilling well is established during SVMs (SVM) identification model in classification；
（4）Second is carried out to attribute of the subregion with timesharing window to merge, and whole district's hierarchical nature fusion attribute is obtained, to excellent in computing
Select attribute to take some sampled points at random, carry out classification function f (x) computing, according to optimization criteria, calculate classification threshold values, make
Threshold values of classifying can correctly separate the different classifications in attribute, and consideration of compromising makes sample mistake point rate minimum, class interval
It is maximum；Constantly modification classification threshold values, until the sample point tolerance in different classifications is all converged in acceptable scope；
（5）The attributive classification result being classified based on earthquake under sensitive algorithm overall background completed：It is further thin for classification results
Partial objectives for, the attribute forecast of hour window is carried out, melted again with reacting the primitive attribute information of overall background by the means of attribute optimization
Close, obtain earthquake hierarchical nature fusion forecasting figure；
（6）The real reservoir parameter for boring data is combined using earthquake hierarchical nature fusion forecasting figure, what semidefinite quantified predicts deep layer
The areal extent of Thin oil sandwich.
2. the deep layer Thin oil sandwich quantitatively characterizing method according to claim 1 based on earthquake classification Sensitive Attributes fusion,
It is characterized in that：
By step（3）The characteristic vector of middle cognitive phase extraction is as supporting vector input, progress classification function computing, Mei Gezhi
The corresponding classification threshold values b of vector is held, classification threshold values b is classified as one kind less than special value, determines categorised demarcation line according to this；
The constraint of drilling well reservoir parameter is added when extracting characteristic vector, passes through earthquake category under seismic attributes information near well point and overall background
Property compare analysis, constantly calculate convergence, the extraction of characteristic vector in Statistical error modeling process makes model closer to practically
Matter situation, reduce multi-solution.
3. the deep layer Thin oil sandwich quantitatively characterizing side according to claim 1 or 2 based on earthquake classification Sensitive Attributes fusion
Method, it is characterized in that：
Performing step（4）During, it is trained by inspection of hierarchy model, if obtaining spy of the Ganlei based on classification function
Sign vector, all corresponding classification set in seismic properties per category feature vector, so far assorting process terminates；Whole process with
SVMs（SVM）It is as follows for core algorithm, specific implementation：
H is classification line, meets linear equation：w*x+b=0；H1, H2 be respectively it is all kinds of in from nearest samples of classification line H and flat
The distance between H1, H2 are referred to as class interval by row in classification line H straight line；Wherein, optimal classification line refers to classify line H not
But two classes can correctly be separated, it is 0 to realize training error rate, and makes the ‖ w ‖ of class interval 2/ maximum；Linear separable sets closes
(xi,yi), meet yi (w*xi+b) -1 >=0, y ∈ { -1,1 }, i=1 ... n after classification line equation is standardized, point
The maximum of equal value and ‖ w ‖ of the ‖ w ‖ of class interval 2/2Minimum, or meet formula yi (w*xi+b) -1 >=0, y ∈ { -1,1 }, i=1 ... n and
Make ‖ w ‖2For/2 minimum classifying faces as optimal classification surface, the sample point wherein on H1, H2 is referred to as supporting vector；It will ask for optimal
Classifying face is converted into asking in linear equation most preferably（w,b）；I.e. in yi (w*xi+b) -1 >=0, y ∈ { -1,1 }, i=1 ... n's
‖ w ‖ are sought under constraint2/ 2 minimum value；The Lagrangian being defined as follows：
Wherein ai>=0 is Lagrange coefficient；Seek ‖ w ‖2/ 2 minimum is exactly to w, and b seeks the minimum of Lagrangian, right
L functions do differential and make it be equal to 0, Lagrangian and are converted to：In ∑ yiai=0,ai>=0, i=1 ... ask drawing under n constraints
Ge Lang coefficients aiTo function：
Its maximum optimal solution meets ai(yi(w*xi+ b) -1)=0, i=1 ... n, a in solutioniThe sample x being not zeroi,yiSupport
Sample point on vector, that is, H1, H2, therefore W can be reduced to：
The optimization class function obtained after solution is：
Wherein sgn is classification function, b is classification threshold values, and x is given sample, xi, yi are linear separability sample set；For unknown
Sample x need to only calculate classification function, you can judge to classify belonging to x, in the case of linearly inseparable：1 relaxation only need to be added
Option ξi>=0, sample set meets yi (w*xi+b) -1+ ξi>=0 i=1 ... n, the conversion of function and calculating process with linear similar,
But constraints is changed into：∑yiai=0,0≤ai≤ c, i=1 ... n, c are the constant that control mistake divides that sample punishes degree, letter of classifying
Also respective change is number
K is interior Product function appropriate in optimal classification surface, and other specification is identical with linear case.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410449177.XA CN105467449B (en) | 2014-09-04 | 2014-09-04 | Deep layer Thin oil sandwich quantitatively characterizing method based on earthquake classification Sensitive Attributes fusion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410449177.XA CN105467449B (en) | 2014-09-04 | 2014-09-04 | Deep layer Thin oil sandwich quantitatively characterizing method based on earthquake classification Sensitive Attributes fusion |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105467449A CN105467449A (en) | 2016-04-06 |
| CN105467449B true CN105467449B (en) | 2018-01-05 |
Family
ID=55605346
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410449177.XA Active CN105467449B (en) | 2014-09-04 | 2014-09-04 | Deep layer Thin oil sandwich quantitatively characterizing method based on earthquake classification Sensitive Attributes fusion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105467449B (en) |
Families Citing this family (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106353811B (en) * | 2016-08-12 | 2019-04-30 | 北京天工新源科技有限公司 | Based on well information attribute, preferably more attributes merge fluid identification of reservoir method to one kind |
| CN107885966B (en) * | 2017-10-23 | 2020-06-26 | 中国地质大学（武汉） | Abnormal chemical single element classification method containing constraint SVM |
| CN110274859A (en) * | 2018-03-14 | 2019-09-24 | 中国石油化工股份有限公司 | Carbonate reservoir control methods |
| CN108957534B (en) * | 2018-06-12 | 2020-02-14 | 中国石油天然气股份有限公司 | Method and device for predicting gas saturation |
| CN109902390B (en) * | 2018-12-13 | 2023-10-24 | 中国石油大学(华东) | Method for predicting development area of favorable reservoir based on small sample expansion |
| CN112346117B (en) * | 2019-08-09 | 2023-06-30 | 中国石油天然气集团有限公司 | Reservoir characteristic prediction method and device based on seismic attribute fusion |
| CN113419271B (en) * | 2021-05-07 | 2023-08-29 | 中铁二院工程集团有限责任公司 | Earthquake magnitude prediction method, device, equipment and readable storage medium |
| CN113341459B (en) * | 2021-05-12 | 2022-04-12 | 北京大学 | Earthquake positioning method and device based on machine learning and dynamics calculation fusion |
| CN115857047B (en) * | 2022-09-28 | 2023-07-21 | 北京中恒利华石油技术研究所 | Comprehensive prediction method for earthquake reservoir |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103412332A (en) * | 2013-01-22 | 2013-11-27 | 中国地质大学（北京） | Method for quantitative calculation of thickness of thin reservoir layer |
| CN103675906A (en) * | 2012-09-19 | 2014-03-26 | 中国石油化工股份有限公司 | Multi-control reservoir prediction method used for improving prediction accuracy of complex clastic rock reservoir |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6438493B1 (en) * | 2000-09-29 | 2002-08-20 | Exxonmobil Upstream Research Co. | Method for seismic facies interpretation using textural analysis and neural networks |
| US6560540B2 (en) * | 2000-09-29 | 2003-05-06 | Exxonmobil Upstream Research Company | Method for mapping seismic attributes using neural networks |
| US9128203B2 (en) * | 2011-09-28 | 2015-09-08 | Saudi Arabian Oil Company | Reservoir properties prediction with least square support vector machine |
-
2014
- 2014-09-04 CN CN201410449177.XA patent/CN105467449B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103675906A (en) * | 2012-09-19 | 2014-03-26 | 中国石油化工股份有限公司 | Multi-control reservoir prediction method used for improving prediction accuracy of complex clastic rock reservoir |
| CN103412332A (en) * | 2013-01-22 | 2013-11-27 | 中国地质大学（北京） | Method for quantitative calculation of thickness of thin reservoir layer |
Non-Patent Citations (2)
| Title |
|---|
| 塔里木盆地哈拉哈塘地区三维叠前深度偏移与储层定量雕刻;彭更新 等;《中国石油勘探》;20110531(第5-6期);第49-56页 * |
| 砂泥岩薄互层储层预测技术在NP区的应用;柴巧英 等;《石油地球物理勘探》;20120831;第102-105页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105467449A (en) | 2016-04-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105467449B (en) | Deep layer Thin oil sandwich quantitatively characterizing method based on earthquake classification Sensitive Attributes fusion | |
| CN106803253B (en) | A kind of three-dimensional rock image crack identification method | |
| CN101894275B (en) | Weakly supervised method for classifying SAR images | |
| CN107341517A (en) | The multiple dimensioned wisp detection method of Fusion Features between a kind of level based on deep learning | |
| Ren et al. | A novel hybrid method of lithology identification based on k-means++ algorithm and fuzzy decision tree | |
| US8090538B2 (en) | System and method for interpretation of well data | |
| CN106096561A (en) | Infrared pedestrian detection method based on image block degree of depth learning characteristic | |
| CN106778804A (en) | The zero sample image sorting technique based on category attribute transfer learning | |
| CN105069779A (en) | Building ceramic surface pattern quality detection method | |
| CN107301221A (en) | A kind of data digging method of multiple features dimension heap fusion | |
| CN109345007B (en) | Advantageous reservoir development area prediction method based on XGboost feature selection | |
| CN112684497B (en) | Seismic waveform clustering method and device | |
| CN107798340B (en) | Multiple Geochemical abnormality recognition method based on the more self-encoding encoders of space constraint | |
| CN107038505A (en) | Ore-search models Forecasting Methodology based on machine learning | |
| Eskandari et al. | Reservoir Modelling of Complex Geological Systems--A Multiple-Point Perspective | |
| CN109683871A (en) | Code automatically generating device and method based on image object detection method | |
| CN106803010A (en) | For the Fuzzy Grey comprehensive evaluation method and device of low permeability reservoir quantitative assessment | |
| CN108985145A (en) | The Opposite direction connection deep neural network model method of small size road traffic sign detection identification | |
| Hu et al. | Water flooding flowing area identification for oil reservoirs based on the method of streamline clustering artificial intelligence | |
| CN106569272A (en) | Earthquake attribute fusion method based on data property space ascending dimension | |
| Li et al. | Research on reservoir lithology prediction method based on convolutional recurrent neural network | |
| Wei et al. | Characterization of saturation and pressure distribution based on deep learning for a typical carbonate reservoir in the Middle East | |
| CN111079783A (en) | Method for identifying stratum lithology parameters based on multi-core ensemble learning | |
| CN107167852A (en) | Tunnel geology synthetical forecast method based on Multi-source Information Fusion | |
| Lu et al. | Identifying flow units by FA-assisted SSOM—An example from the Eocene basin-floor-fan turbidite reservoirs in the Daluhu Oilfield, Dongying Depression, Bohai Bay Basin, China |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |