CN108596780A - A kind of Reservoir type division methods and system based on multiple information - Google Patents
A kind of Reservoir type division methods and system based on multiple information Download PDFInfo
- Publication number
- CN108596780A CN108596780A CN201810397193.7A CN201810397193A CN108596780A CN 108596780 A CN108596780 A CN 108596780A CN 201810397193 A CN201810397193 A CN 201810397193A CN 108596780 A CN108596780 A CN 108596780A
- Authority
- CN
- China
- Prior art keywords
- data
- reservoir type
- neural network
- reservoir
- variance
- 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
- 238000000034 method Methods 0.000 title claims abstract description 43
- 238000013528 artificial neural network Methods 0.000 claims abstract description 92
- 238000005259 measurement Methods 0.000 claims abstract description 75
- 238000012549 training Methods 0.000 claims description 28
- 230000004913 activation Effects 0.000 claims description 10
- 230000008569 process Effects 0.000 claims description 7
- 230000035945 sensitivity Effects 0.000 claims description 7
- 230000001537 neural effect Effects 0.000 claims description 5
- 210000004218 nerve net Anatomy 0.000 claims description 3
- 230000006870 function Effects 0.000 description 17
- 238000009826 distribution Methods 0.000 description 8
- 230000035699 permeability Effects 0.000 description 6
- 210000004027 cell Anatomy 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 238000004519 manufacturing process Methods 0.000 description 4
- 230000008859 change Effects 0.000 description 3
- 238000010276 construction Methods 0.000 description 3
- 210000002569 neuron Anatomy 0.000 description 3
- 241001269238 Data Species 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000012804 iterative process Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 238000002844 melting Methods 0.000 description 2
- 230000008018 melting Effects 0.000 description 2
- 210000005036 nerve Anatomy 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 239000004744 fabric Substances 0.000 description 1
- 239000012530 fluid Substances 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 230000008595 infiltration Effects 0.000 description 1
- 238000001764 infiltration Methods 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 239000003208 petroleum Substances 0.000 description 1
- 230000000704 physical effect Effects 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 230000003252 repetitive effect Effects 0.000 description 1
- 239000011435 rock Substances 0.000 description 1
- 238000005507 spraying Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/02—Agriculture; Fishing; Forestry; Mining
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- Tourism & Hospitality (AREA)
- Artificial Intelligence (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- Mining & Mineral Resources (AREA)
- Marine Sciences & Fisheries (AREA)
- General Business, Economics & Management (AREA)
- Animal Husbandry (AREA)
- Agronomy & Crop Science (AREA)
- Economics (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Measurement Of Radiation (AREA)
Abstract
The present invention discloses a kind of Reservoir type division methods and system based on multiple information.This method includes：The data sensitive to Reservoir type are obtained from the log data of multiple sample points；The data sensitive to Reservoir type include neutron data, resistivity, porosity, interval transit time, natural gamma and compensation density；Obtain the actual reservoir type that each sample point that artificial reservoir division obtains is carried out to each sample point；The data sensitive to Reservoir type of each sample point are inputted into deep neural network, adjusting the structural parameters of neural network makes neural network export corresponding Reservoir type, to be trained to deep neural network；Trained deep neural network is used to realize the division to Reservoir type according to the data sensitive to Reservoir type of actual measurement.Method and system disclosed by the invention can improve the precision of reservoir division.
Description
Technical field
The present invention relates to field of petroleum geophysical exploration, are drawn more particularly to a kind of Reservoir type based on multiple information
Divide method and system.
Background technology
Reservoir type division is one of important content of reservoir description, is the key that carry out oil field development Potentials work
Make.Generally use shale content, permeability etc. carry out Reservoir type with the closely related parameter of reservoir characteristic in actual production
Division, wherein shale content can reflect that reservoir lithology and physical property characteristic, permeability can characterize fluid in rock interior stream
Dynamic complexity.But the parameters such as the shale content divided for Reservoir type and permeability can not be directly from the measurement number of well logging
It directly acquires in, but is calculated using cross-plot or empirical equation to obtain based on practical logging curve, these methods
Accuracy depends on the selection of aspect of model parameter, Regional Geological Characteristics and layered characteristic, is differentiated by expert artificial with experience etc.
Factor is affected.Since each log parameter has it to be specifically applicable in stratum, and can be complete without which kind of log parameter
It is suitable for all stratum entirely, limited well logging information is often only utilized when carrying out Reservoir type and dividing, leads to reservoir division
Precision it is not high.
Invention content
The object of the present invention is to provide a kind of Reservoir type division methods and system based on multiple information improve reservoir and draw
The precision divided.
To achieve the above object, the present invention provides following schemes：
A kind of Reservoir type division methods based on multiple information, including：
The data sensitive to Reservoir type are obtained from the log data of multiple sample points；It is described sensitive to Reservoir type
Data include neutron data, resistivity, porosity, interval transit time, natural gamma and compensation density；
Obtain the actual reservoir type that each sample point that artificial reservoir division obtains is carried out to each sample point；
The data sensitive to Reservoir type of each sample point are inputted into deep neural network, adjust the structure of neural network
Parameter makes neural network export corresponding Reservoir type, to be trained to the deep neural network；It is trained described
Deep neural network is used to realize the division to Reservoir type according to the data sensitive to Reservoir type of actual measurement.
Optionally, described that the data sensitive to Reservoir type of each sample point are inputted into deep neural network, adjustment god
Structural parameters through network make neural network export corresponding Reservoir type, to be trained to the deep neural network,
It specifically includes：
Feature is carried out to data sensitive to Reservoir type described in each respectively using cross-plot and kernel density estimation method
Statistics, determines the mean value and variance of each data sensitive to Reservoir type, obtains the sample average and sample of each data
This variance；
Using the sample average and sample variance of each data to each data sensitive to Reservoir type into rower
Standardization obtains standardized data；
Using the standardized data of each type of each sample point as the input of deep neural network, with linear
Activation primitive of the rectification function as the deep neural network makes by adjusting the structural parameters of the deep neural network
Difference between the Reservoir type data and corresponding actual reservoir type of deep neural network output preset range it
It is interior, to complete the training of deep neural network.
Optionally, the process divided to Reservoir type according to the data sensitive to Reservoir type of actual measurement includes：
The data sensitive to Reservoir type for obtaining actual measurement, obtain measured data；
The mean value and variance for calculating each measured data obtain actual measurement mean value and the actual measurement side of each measured data
Difference；
The mean value threshold range and variance threshold values of each data are calculated using the sample average and sample variance of each data
Range；
Judge whether the actual measurement mean value of each measured data is located in corresponding mean value threshold range and each surveys number
According to actual measurement variance whether be located at corresponding variance threshold values within the scope of, obtain the first judging result；
If first judging result indicates that the actual measurement mean value of any one measured data is located at corresponding mean value threshold value model
Except enclosing and/or actual measurement variance is located at outside corresponding variance threshold values range, then resurveys the data of corresponding types；
If first judging result indicates that the actual measurement mean value of each measured data is located in corresponding mean value threshold range
And actual measurement variance is located within the scope of corresponding variance threshold values, then is standardized to each measured data, obtains standard actual measurement number
According to；
Using the standard measured data of each type as the input of deep neural network, the output of neural network is obtained
Actual measurement Reservoir type.
Invention additionally discloses a kind of Reservoir type dividing system based on multiple information, including：
Sample data acquisition module, for obtaining the number sensitive to Reservoir type from the log data of multiple sample points
According to；The data sensitive to Reservoir type include neutron data, resistivity, porosity, interval transit time, natural gamma and compensation
Density；
Actual reservoir type acquisition module carries out each sample that artificial reservoir division obtains for obtaining to each sample point
The actual reservoir type of this point；
Neural metwork training module, for the data sensitive to Reservoir type of each sample point to be inputted depth nerve net
Network, adjusting the structural parameters of neural network makes neural network export corresponding Reservoir type, to the deep neural network
It is trained；The trained deep neural network is used to be realized to reservoir according to the data sensitive to Reservoir type of actual measurement
The division of type.
Optionally, the neural metwork training module, specifically includes：
Sample data mean value and variance computing unit, for utilizing cross-plot and kernel density estimation method respectively to each institute
It states the data sensitive to Reservoir type and carries out characteristic statistics, determine mean value and the side of each data sensitive to Reservoir type
Difference obtains the sample average and sample variance of each data；
Sample data Standardisation Cell, sample average and sample variance for utilizing each data store up each described pair
The data of channel type sensitivity are standardized, and obtain standardized data；
Training unit, for using the standardized data of each type of each sample point as deep neural network
Input, using line rectification function as the activation primitive of the deep neural network, by adjusting the deep neural network
Structural parameters, make the difference between the Reservoir type data and corresponding actual reservoir type of deep neural network output
Within preset range, to complete the training of deep neural network.
Optionally, the Reservoir type dividing system further includes reservoir division module, is used for according to actual measurement to reservoir class
The data of type sensitivity divide Reservoir type；The reservoir division module, specifically includes：
Measured data acquiring unit, the data sensitive to Reservoir type for obtaining actual measurement, obtains measured data；
Measured data mean value and variance computing unit, mean value and variance for calculating each measured data, obtain
The actual measurement mean value and actual measurement variance of each measured data；
Threshold range computing unit, for calculating the equal of each data using the sample average and sample variance of each data
It is worth threshold range and variance threshold values range；
Judging unit, for judge each measured data actual measurement mean value whether be located at corresponding mean value threshold range in
And whether the actual measurement variance of each measured data is located within the scope of corresponding variance threshold values, obtains judging result；
Data adopt unit again, are corresponded to if indicating that the actual measurement mean value of any one measured data is located at for the judging result
Mean value threshold range except and/or actual measurement variance be located at outside corresponding variance threshold values range, then resurvey corresponding types
Data；
Measured data Standardisation Cell, if for the judging result indicate each measured data actual measurement mean value be located at pair
In the mean value threshold range answered and actual measurement variance is located within the scope of corresponding variance threshold values, then carries out standard to each measured data
Change, obtains standard measured data；
Reservoir division unit, for using the standard measured data of each type as the input of deep neural network,
Obtain the actual measurement Reservoir type of the output of neural network.
According to specific embodiment provided by the invention, the invention discloses following technique effects：It is disclosed by the invention to be based on
The Reservoir type division methods and system of multiple information, by combining neutron data, resistivity, porosity, interval transit time, nature
A variety of data such as gamma and compensation density establish answering between a variety of data and Reservoir type using the study of deep neural network
Miscellaneous non-linear relation realizes that the automation and intelligentification of Reservoir type divides.The present invention is divided by automation and intelligentification, is avoided
The interference of human factor improves the precision of Reservoir type division, while utilizing a variety of data, realizes melting for multiple information
It closes, further improves reservoir division precision.
Description of the drawings
It in order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, below will be to institute in embodiment
Attached drawing to be used is needed to be briefly described, it should be apparent that, the accompanying drawings in the following description is only some implementations of the present invention
Example, for those of ordinary skill in the art, without having to pay creative labor, can also be according to these attached drawings
Obtain other attached drawings.
Fig. 1 is that the present invention is based on the method flow diagrams of the Reservoir type division methods embodiment of multiple information；
Fig. 2 is effective distribution figure of different log parameters in different type reservoir；
Fig. 3 is the network structure of deep neural network of the present invention；
Fig. 4 is the function curve schematic diagram that object function gradually becomes flat in deep neural network training process；
Fig. 5 is the reservoir division result figure that the Reservoir type division methods using the present invention based on multiple information obtain；
Fig. 6 is that the present invention is based on the system construction drawings of the Reservoir type dividing system embodiment of multiple information.
Specific implementation mode
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation describes, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
The purpose of the present invention is making full use of the polynary well logging informations such as sound electric discharge to carry out the direct division of Reservoir type, use
Well logging multiple information in actual production data establishes Reservoir type sample database, and polynary well logging is built using deep learning network
Complex nonlinear relationship between information and Reservoir type, the comprehensive utilization of the polynary well logging information such as development sound electric discharge, so as to
The direct division of Reservoir type in enough real logging datas, avoid calculating indirectly after the parameters such as shale content, permeability again into
Row Reservoir type existing uncertainty when dividing, the influence for avoiding human factor from dividing Reservoir type promote Reservoir type
The automation of prediction and intelligent level.
In order to make the foregoing objectives, features and advantages of the present invention clearer and more comprehensible, below in conjunction with the accompanying drawings and specific real
Applying mode, the present invention is described in further detail.
Fig. 1 is that the present invention is based on the method flow diagrams of the Reservoir type division methods embodiment of multiple information.
Referring to Fig. 1, the Reservoir type division methods based on multiple information are somebody's turn to do, including：
Step 101：The data sensitive to Reservoir type are obtained from the log data of multiple sample points；It is described to reservoir class
The data of type sensitivity include neutron data, resistivity, porosity, interval transit time, natural gamma and compensation density.
By to studying a variety of different types of well loggings such as sound wave, radioactivity, resistivity in the existing well-log information in work area
Information is screened, and the logs such as natural gamma, resistivity, natural potential, the neutron sensitive to lithology are sorted out, to infiltration
The various related logging curves such as neutron, density, the sound wave of rate sensitivity.The present invention in view of dividing Reservoir type in actual production
When Main Basiss shale content and Permeability Parameters, choose neutron data, resistivity, porosity, interval transit time, natural gamma,
The multiple information source that 6 kinds of log parameters such as compensation density are divided as Reservoir type.It can also further increase on this basis
Other parameter, but on condition that increased parameter has Effective Numerical on each Reservoir type, then rejecting hole collapse is tight
The data located again and the abnormal numerical value caused by logger or measurement error, for structure Reservoir type numeralization sample database
Basic data basis is provided.
Step 102：Obtain the actual reservoir class that each sample point that artificial reservoir division obtains is carried out to each sample point
Type.
To 6 kinds of neutron data of selection, resistivity, porosity, interval transit time, natural gamma, compensation density log parameters
It carries out building library, carries out environmental correction for the feature of each well data, and unit is all unified to SI units, it is ensured that no
The data that same time, different instruments, different operation personnel's measurement obtain can be efficiently used.On this basis according to existing
Shale content, Permeability Parameters and formation testing result at different depth data sample carry out Reservoir type artificial division,
By the data of each sample point be respectively divided into non-reservoir, I class reservoir (reservoir of oil and gas of relative ideal, potentiality to be exploited are big) and
The three types such as II class reservoir (possible reservoir of oil and gas, have certain potentiality to be exploited), and use respectively digital 0,1,2 as
Label, it is established that the Reservoir type numeralization sample database of polynary well logging information at each sample point.The non-reservoir contains for shale
Larger reservoir is measured, substantially without potentiality to be exploited；The I class reservoir is the reservoir of oil and gas of relative ideal, and potentiality to be exploited is big；It is described
II class reservoir is possible reservoir of oil and gas, has certain potentiality to be exploited.
Step 103：The data sensitive to Reservoir type of each sample point are inputted into deep neural network, adjust nerve net
The structural parameters of network make neural network export corresponding Reservoir type, to be trained to the deep neural network；Training
The good deep neural network is used to realize the division to Reservoir type according to the data sensitive to Reservoir type of actual measurement.
The step 103 specifically includes：
Feature is carried out to data sensitive to Reservoir type described in each respectively using cross-plot and kernel density estimation method
Statistics, determines the mean value and variance of each data sensitive to Reservoir type, obtains the sample average and sample of each data
This variance；
Using the sample average and sample variance of each data to each data sensitive to Reservoir type into rower
Standardization obtains standardized data；
Using the standardized data of each type of each sample point as the input of deep neural network, with linear
Activation primitive of the rectification function as the deep neural network makes by adjusting the structural parameters of the deep neural network
Difference between the Reservoir type data and corresponding actual reservoir type of deep neural network output preset range it
It is interior, to complete the training of deep neural network.
The step 103 specifically can be used following specific implementation mode and realize：
After setting up Reservoir type sample database, it is necessary to carry out characteristic statistics and standard to the multiple parameters in sample database
Change, it is intended to ensure that the distribution of each parameter in sample database is uniformly reasonable, while be the pre- of follow-up real data Reservoir type
It surveys and parameter foundation is provided.The present invention carries out the work of sample database characteristic statistics using cross-plot and Density Estimator figure method joint
(cross-plot is exactly that each two parameter all does intersection analysis with the distribution characteristics of statistical data respectively, and cuclear density method is in probability theory
In be used for estimating the density function of parameter, to obtain following required statistical nature), obtain each parameter in sample database
Mean μiAnd variances sigmai(i be data class number, share neutron data, resistivity, porosity, interval transit time, natural gamma,
6 kinds of log parameters of compensation density, therefore i=1,2 ... ..., 6), specify effective point of different log parameters in different type reservoir
Cloth range, Fig. 2 are effective distribution figure of different log parameters in different type reservoir.The distribution is used to carry out in fact
Judge whether measured data can be used for judging Reservoir type when survey.Then according to the Density Estimator peak of each parameter
Value and bandwidth determine each log parameter for discrimination possessed by Reservoir type classification.Especially work as parameter distribution range
It then needs to continue additional sample library information when uneven, Reservoir type sample number deficiency, i.e. sample number is unsatisfactory for training condition also
It needs to continue to collect more more fully data, the rational relation between structure Reservoir type and multiple information lays the foundation.
In view of the influence due to factors such as construction team, instrumentation and testings in actual production data, need to polynary ginseng
Number information is standardized, and optimizes the data structure of Reservoir type sample database.The present invention is using following formula to multiple information library
Data are standardized, it is ensured that the data fit mean value after standardizing is 0, the normal distribution that variance is 1, i.e.,：
Wherein μiFor corresponding log data mean value, σ in sample databaseiFor corresponding sample data variance, i is data class
It numbers (i=1,2 ... ..., 6), j represents different sample point (j=1,2 ... ..., n (n is sample point sum)), xijRepresent mark
Parameter before standardization, ZijRepresent the parameter after standardization.
After standardization is completed, the structure carried out to neural network is needed to be defined.In projected depth neural network
When need for Reservoir type sample database feature carry out Precise spraying, specify the number of plies of deep neural network, every layer of nerve
The network architecture parameters such as first number and activation primitive.Fig. 3 is the network structure of deep neural network of the present invention.Referring to
Fig. 3, the present invention constructs a deep neural network model for including 5 hidden layers, wherein the neuron number of each hidden layer
Mesh is respectively 20,40,80,40,20.Line rectification function (Relu functions) is used in network design is used as activation primitive,
To which overfitting problem be effectively relieved and accelerate network convergence process, and Softmax layers are added to before output layer.The network
Structure matrix form can be expressed as：
g(W(1)a(1)+b(1))=a(2)
g(W(2)a(2)+b(2))=a(3)
...
g(W(n-1)a(n-1)+b(n-1))=y
Wherein, g (x) indicates the operational form of activation primitive, and the activation primitive that the present invention uses is ReLu function, i.e. g (x)
=max (0, x), n indicate the number of plies (n=6, i.e. the hidden layer number of plies add the sum of output layer in the present invention) of deep neural network, b(i)Indicate bias term, wherein a(i)(i=1 ..., n-1) be middle layer feature vector (as i=1, a(1)Sample used by representing
Polynary well logging information in this library), W(i)Indicate the weight matrix for being mapped to i+1 layer in network structure from i-th layer, y representative samples
The Reservoir type marked in this library.
After network structure definition is completed, depth is input to using all data in multiple information Reservoir type sample database
It is trained in degree neural network, setting models parameter (the weighted value W of network random first when training(i)With bias term b(i)), model parameter numerical value is adjusted by iterating, after ensuring the multiple information in sample database by being exported after the network
Reservoir type y' and sample database in Reservoir type y between minimum, the specific object function expression that reaches object function error
Form is as follows：
The present invention uses regularization method to prevent over-fitting, i.e., in each repetitive exercise in the training process
A part of neuron is set to be not involved in training at random, to reduce the relation of interdependence between neuron, to improve neural network
Generalization ability；The optimization algorithm that declines based on gradient is used to update the model parameter in trained iterative process, i.e. basis
Object function calculates the gradient of each parameter in network model, is updated to network parameter using the Grad, and set
The maximum iteration of model parameter be 5000 times, when training reach maximum iteration or object function in an iterative process by
Grade when becoming flat can deconditioning, Fig. 4 is the letter that object function gradually becomes flat in deep neural network training process
Number curve schematic diagram, it can be seen that the slope of object function is gradually lower and represents object function and gradually become flat.When up to
It is the network model parameter W that can save training and obtain to this condition(i)And b(i)(i=1 ..., n-1).
After training is completed, so that it may to carry out the division of Reservoir type using training result.
The process divided to Reservoir type according to the data sensitive to Reservoir type of actual measurement includes：
The data sensitive to Reservoir type for obtaining actual measurement, obtain measured data；
The mean value and variance for calculating each measured data obtain actual measurement mean value and the actual measurement side of each measured data
Difference；
The mean value threshold range and variance threshold values of each data are calculated using the sample average and sample variance of each data
Range；
Judge whether the actual measurement mean value of each measured data is located in corresponding mean value threshold range and each surveys number
According to actual measurement variance whether be located at corresponding variance threshold values within the scope of, obtain the first judging result；
If first judging result indicates that the actual measurement mean value of any one measured data is located at corresponding mean value threshold value model
Except enclosing and/or actual measurement variance is located at outside corresponding variance threshold values range, then resurveys the data of corresponding types；
If first judging result indicates that the actual measurement mean value of each measured data is located in corresponding mean value threshold range
And actual measurement variance is located within the scope of corresponding variance threshold values, then is standardized to each measured data, obtains standard actual measurement number
According to；
Using the standard measured data of each type as the input of deep neural network, the output of neural network is obtained
Actual measurement Reservoir type.
The division that the mode of being implemented as follows carries out Reservoir type specifically may be used：
Filter out the multiple information of needs from numerous log datas first when carrying out real data Reservoir type and dividing
(i.e. 6 kinds of neutron data, resistivity, porosity, interval transit time, natural gamma, compensation density data) and data outliers are rejected,
Calculate the average value mu ' and variances sigma ' of each multiple information in real data, and by average value mu ' and variances sigma ' and present invention foundation
Sample database in multiple information scope of statistics be compared, in the comparison use following standard：
μ × (1- α) ＜ μ ' ＜ μ × (1+ α)
σ × (1-2 α) ＜ σ ' ＜ σ × (1+2 α)
Wherein, α is the confidence level parameter of prediction data, and the present invention sets α=0.1.When real data feature is unsatisfactory for
When formula, the distribution characteristics of the multivariate data and the information in sample database that show Reservoir type to be divided has big difference, and is unsatisfactory for storing up
The primary condition that channel type divides, whether reasonable, if unreasonable need to resurvey data if needing to reexamine the data.Work as input
When the statistical nature of real data meets above formula, then the data can be divided directly using this method to carry out Reservoir type.So
Afterwards, the multiple information treated in prediction data is standardized, and the network model parameter W and b that are obtained using training are to reality
The polynary well logging information of input is carried out Reservoir type and is divided, to obtain Reservoir type division result that the data correspond to and defeated
Go out into figure.
Fig. 5 is the reservoir division result figure that the Reservoir type division methods using the present invention based on multiple information obtain.
Referring to Fig. 5, the Reservoir type using the present invention based on multiple information divides the precision that can improve reservoir division.
Fig. 6 is that the present invention is based on the system construction drawings of the Reservoir type dividing system embodiment of multiple information.
Referring to Fig. 6, it is somebody's turn to do the Reservoir type dividing system based on multiple information, including：
Sample data acquisition module 601, it is sensitive to Reservoir type for being obtained from the log data of multiple sample points
Data；The data sensitive to Reservoir type include neutron data, resistivity, porosity, interval transit time, natural gamma and benefit
Repay density；
Actual reservoir type acquisition module 602, for obtain to each sample point carry out artificial reservoir division obtain it is each
The actual reservoir type of a sample point；
Neural metwork training module 603, for the data input depth sensitive to Reservoir type of each sample point is refreshing
Through network, adjusting the structural parameters of neural network makes neural network export corresponding Reservoir type, to depth nerve
Network is trained；The trained deep neural network is used for the data realization pair sensitive to Reservoir type according to actual measurement
The division of Reservoir type.
The neural metwork training module 603, specifically includes：
Sample data mean value and variance computing unit, for utilizing cross-plot and kernel density estimation method respectively to each institute
It states the data sensitive to Reservoir type and carries out characteristic statistics, determine mean value and the side of each data sensitive to Reservoir type
Difference obtains the sample average and sample variance of each data；
Sample data Standardisation Cell, sample average and sample variance for utilizing each data store up each described pair
The data of channel type sensitivity are standardized, and obtain standardized data；
Training unit, for using the standardized data of each type of each sample point as deep neural network
Input, using line rectification function as the activation primitive of the deep neural network, by adjusting the deep neural network
Structural parameters, make the difference between the Reservoir type data and corresponding actual reservoir type of deep neural network output
Within preset range, to complete the training of deep neural network.
Reservoir division module 604, for being divided to Reservoir type according to the data sensitive to Reservoir type of actual measurement；
The reservoir division module 604, specifically includes：
Measured data acquiring unit, the data sensitive to Reservoir type for obtaining actual measurement, obtains measured data；
Measured data mean value and variance computing unit, mean value and variance for calculating each measured data, obtain
The actual measurement mean value and actual measurement variance of each measured data；
Threshold range computing unit, for calculating the equal of each data using the sample average and sample variance of each data
It is worth threshold range and variance threshold values range；
Judging unit, for judge each measured data actual measurement mean value whether be located at corresponding mean value threshold range in
And whether the actual measurement variance of each measured data is located within the scope of corresponding variance threshold values, obtains judging result；
Data adopt unit again, are corresponded to if indicating that the actual measurement mean value of any one measured data is located at for the judging result
Mean value threshold range except and/or actual measurement variance be located at outside corresponding variance threshold values range, then resurvey corresponding types
Data；
Measured data Standardisation Cell, if for the judging result indicate each measured data actual measurement mean value be located at pair
In the mean value threshold range answered and actual measurement variance is located within the scope of corresponding variance threshold values, then carries out standard to each measured data
Change, obtains standard measured data；
Reservoir division unit, for using the standard measured data of each type as the input of deep neural network,
Obtain the actual measurement Reservoir type of the output of neural network.
According to specific embodiment provided by the invention, the invention discloses following technique effects：It is disclosed by the invention to be based on
The Reservoir type division methods and system of multiple information, by combining neutron data, resistivity, porosity, interval transit time, nature
A variety of data such as gamma and compensation density establish answering between a variety of data and Reservoir type using the study of deep neural network
Miscellaneous non-linear relation realizes that the automation and intelligentification of Reservoir type divides.The present invention is divided by automation and intelligentification, is avoided
The interference of human factor improves the precision of Reservoir type division, while utilizing a variety of data, realizes melting for multiple information
It closes, further improves reservoir division precision.
For system disclosed in embodiment, since it is corresponded to the methods disclosed in the examples, so the ratio of description
Relatively simple, reference may be made to the description of the method.
Principle and implementation of the present invention are described for specific case used herein, and above example is said
The bright method and its core concept for being merely used to help understand the present invention；Meanwhile for those of ordinary skill in the art, foundation
The thought of the present invention, there will be changes in the specific implementation manner and application range.In conclusion the content of the present specification is not
It is interpreted as limitation of the present invention.
Claims (6)
1. a kind of Reservoir type division methods based on multiple information, which is characterized in that including：
The data sensitive to Reservoir type are obtained from the log data of multiple sample points；The data sensitive to Reservoir type
Including neutron data, resistivity, porosity, interval transit time, natural gamma and compensation density；
Obtain the actual reservoir type that each sample point that artificial reservoir division obtains is carried out to each sample point；
The data sensitive to Reservoir type of each sample point are inputted into deep neural network, adjust the structural parameters of neural network
Neural network is set to export corresponding Reservoir type, to be trained to the deep neural network；The trained depth
Neural network is used to realize the division to Reservoir type according to the data sensitive to Reservoir type of actual measurement.
2. the Reservoir type division methods according to claim 1 based on multiple information, which is characterized in that it is described will be each
The data sensitive to Reservoir type of sample point input deep neural network, and adjusting the structural parameters of neural network makes neural network
Corresponding Reservoir type is exported, to be trained to the deep neural network, is specifically included：
Characteristic statistics are carried out to data sensitive to Reservoir type described in each respectively using cross-plot and kernel density estimation method,
The mean value and variance for determining each data sensitive to Reservoir type, obtain sample average and the sample side of each data
Difference；
Each data sensitive to Reservoir type are standardized using the sample average and sample variance of each data,
Obtain standardized data；
Using the standardized data of each type of each sample point as the input of deep neural network, with line rectification
Activation primitive of the function as the deep neural network makes described by adjusting the structural parameters of the deep neural network
Deep neural network output Reservoir type data and corresponding actual reservoir type between difference within preset range, from
And complete the training of deep neural network.
3. the Reservoir type division methods according to claim 2 based on multiple information, which is characterized in that according to actual measurement
The process that the data sensitive to Reservoir type divide Reservoir type includes：
The data sensitive to Reservoir type for obtaining actual measurement, obtain measured data；
The mean value and variance for calculating each measured data obtain the actual measurement mean value and actual measurement variance of each measured data；
The mean value threshold range and variance threshold values range of each data are calculated using the sample average and sample variance of each data；
Judge whether the actual measurement mean value of each measured data is located in corresponding mean value threshold range and each measured data
It surveys whether variance is located within the scope of corresponding variance threshold values, obtains the first judging result；
If first judging result indicate the actual measurement mean value of any one measured data be located at corresponding mean value threshold range it
Outer and/or actual measurement variance is located at outside corresponding variance threshold values range, then resurveys the data of corresponding types；
If it is in corresponding mean value threshold range and real that first judging result indicates that the actual measurement mean value of each measured data is located at
It surveys variance to be located within the scope of corresponding variance threshold values, then each measured data is standardized, obtains standard measured data；
Using the standard measured data of each type as the input of deep neural network, the reality of the output of neural network is obtained
Survey Reservoir type.
4. a kind of Reservoir type dividing system based on multiple information, which is characterized in that including：
Sample data acquisition module, for obtaining the data sensitive to Reservoir type from the log data of multiple sample points；Institute
It includes neutron data, resistivity, porosity, interval transit time, natural gamma and compensation density to state the data sensitive to Reservoir type；
Actual reservoir type acquisition module carries out each sample point that artificial reservoir division obtains for obtaining to each sample point
Actual reservoir type；
Neural metwork training module, for the data sensitive to Reservoir type of each sample point to be inputted deep neural network,
The structural parameters of adjustment neural network make neural network export corresponding Reservoir type, to be carried out to the deep neural network
Training；The trained deep neural network is used to be realized to Reservoir type according to the data sensitive to Reservoir type of actual measurement
Division.
5. the Reservoir type dividing system according to claim 4 based on multiple information, which is characterized in that the nerve net
Network training module, specifically includes：
Sample data mean value and variance computing unit, for utilizing cross-plot and kernel density estimation method respectively to right described in each
The data of Reservoir type sensitivity carry out characteristic statistics, determine the mean value and variance of each data sensitive to Reservoir type,
Obtain the sample average and sample variance of each data；
Sample data Standardisation Cell, for the sample average and sample variance using each data to each described to reservoir class
The data of type sensitivity are standardized, and obtain standardized data；
Training unit, for using the standardized data of each type of each sample point as the defeated of deep neural network
Enter, using line rectification function as the activation primitive of the deep neural network, by adjusting the knot of the deep neural network
Structure parameter makes the difference between the Reservoir type data and corresponding actual reservoir type of the deep neural network output pre-
Within the scope of if, to complete the training of deep neural network.
6. the Reservoir type dividing system according to claim 5 based on multiple information, which is characterized in that further include reservoir
Division module, for being divided to Reservoir type according to the data sensitive to Reservoir type of actual measurement；The reservoir division mould
Block specifically includes：
Measured data acquiring unit, the data sensitive to Reservoir type for obtaining actual measurement, obtains measured data；
Measured data mean value and variance computing unit, mean value and variance for calculating each measured data, obtain each
The actual measurement mean value and actual measurement variance of measured data；
Threshold range computing unit, the mean value threshold for calculating each data using the sample average and sample variance of each data
It is worth range and variance threshold values range；
Judging unit, for judging it is in corresponding mean value threshold range and every whether the actual measurement mean value of each measured data is located at
Whether the actual measurement variance of kind measured data is located within the scope of corresponding variance threshold values, obtains judging result；
Data adopt unit again, if it is corresponding equal to indicate that the actual measurement mean value of any one measured data is located at for the judging result
Except value threshold range and/or actual measurement variance is located at outside corresponding variance threshold values range, then resurveys the data of corresponding types；
Measured data Standardisation Cell, if it is corresponding to indicate that the actual measurement mean value of each measured data is located at for the judging result
In mean value threshold range and actual measurement variance is located within the scope of corresponding variance threshold values, then is standardized to each measured data,
Obtain standard measured data；
Reservoir division unit, for using the standard measured data of each type as the input of deep neural network, obtaining
The actual measurement Reservoir type of the output of neural network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810397193.7A CN108596780B (en) | 2018-04-28 | 2018-04-28 | A kind of Reservoir type division methods and system based on multiple information |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810397193.7A CN108596780B (en) | 2018-04-28 | 2018-04-28 | A kind of Reservoir type division methods and system based on multiple information |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108596780A true CN108596780A (en) | 2018-09-28 |
| CN108596780B CN108596780B (en) | 2019-05-21 |
Family
ID=63610592
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810397193.7A Active CN108596780B (en) | 2018-04-28 | 2018-04-28 | A kind of Reservoir type division methods and system based on multiple information |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108596780B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111474585A (en) * | 2020-03-16 | 2020-07-31 | 中国石油天然气集团有限公司 | Gas layer identification method based on deep learning and rock mechanical parameters |
| CN111694856A (en) * | 2020-06-11 | 2020-09-22 | 中国石油大学(北京) | Intelligent prediction method and device for reservoir sensitivity |
| WO2020222064A1 (en) * | 2019-05-01 | 2020-11-05 | International Business Machines Corporation | Dnn training with asymmetric rpu devices |
| CN112253087A (en) * | 2020-10-20 | 2021-01-22 | 河南理工大学 | Biological disturbance reservoir physical property calculation method based on multi-source logging data |
| CN113176608A (en) * | 2021-04-27 | 2021-07-27 | 中国石油大学(华东) | DAS six-component seismic signal decoupling and recovery method for spirally-wound optical fiber |
| CN114909125A (en) * | 2022-06-08 | 2022-08-16 | 中国地质大学（北京） | Text type rock debris logging data numeralization and stratum automatic comparison method and device |
| CN116757096A (en) * | 2023-08-16 | 2023-09-15 | 中国石油天然气集团有限公司 | Determination method, device, equipment and medium for deep, thin and compact sandstone gas reservoir development well |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1704770A (en) * | 2004-05-25 | 2005-12-07 | 张向军 | Dual fuzzy neural network reservoir bed oil gas prediction technique |
| CN1743872A (en) * | 2005-09-26 | 2006-03-08 | 大庆油田有限责任公司 | Multi-parameter dimension-reducing oil-gas-water-layer identifying method |
| US20120071333A1 (en) * | 2010-07-26 | 2012-03-22 | Tampere University Of Technology | Uses of systems with degrees of freedom poised between fully quantum and fully classical states |
| CN103306671A (en) * | 2013-05-17 | 2013-09-18 | 中国石油天然气股份有限公司 | Four-quadrant reservoir type identification method and system |
| CN103603794A (en) * | 2013-11-19 | 2014-02-26 | 中国石油天然气集团公司 | Method and device for adaptive fault diagnosis of gas storage injection-production compressor unit |
| CN106290105A (en) * | 2016-07-20 | 2017-01-04 | 中国石油大学(华东) | A kind of carbonate reservoir dissolution porosity volume content Forecasting Methodology |
-
2018
- 2018-04-28 CN CN201810397193.7A patent/CN108596780B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1704770A (en) * | 2004-05-25 | 2005-12-07 | 张向军 | Dual fuzzy neural network reservoir bed oil gas prediction technique |
| CN1743872A (en) * | 2005-09-26 | 2006-03-08 | 大庆油田有限责任公司 | Multi-parameter dimension-reducing oil-gas-water-layer identifying method |
| US20120071333A1 (en) * | 2010-07-26 | 2012-03-22 | Tampere University Of Technology | Uses of systems with degrees of freedom poised between fully quantum and fully classical states |
| CN103306671A (en) * | 2013-05-17 | 2013-09-18 | 中国石油天然气股份有限公司 | Four-quadrant reservoir type identification method and system |
| CN103603794A (en) * | 2013-11-19 | 2014-02-26 | 中国石油天然气集团公司 | Method and device for adaptive fault diagnosis of gas storage injection-production compressor unit |
| CN106290105A (en) * | 2016-07-20 | 2017-01-04 | 中国石油大学(华东) | A kind of carbonate reservoir dissolution porosity volume content Forecasting Methodology |
Non-Patent Citations (2)
| Title |
|---|
| 安鹏等: "基于深度学习的储层分类方法研究与应用", 《CPS/SEG北京2018国际地球物理会议暨展览电子论文集》 * |
| 安鹏等: "深度学习方法在泥质含量预测中的应用", 《CPS/SEG北京2018国际地球物理会议暨展览电子论文集》 * |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| GB2598856B (en) * | 2019-05-01 | 2022-12-07 | Ibm | DNN training with asymmetric RPU devices |
| WO2020222064A1 (en) * | 2019-05-01 | 2020-11-05 | International Business Machines Corporation | Dnn training with asymmetric rpu devices |
| CN113728338A (en) * | 2019-05-01 | 2021-11-30 | 国际商业机器公司 | DNN training with asymmetric RPU devices |
| GB2598856A (en) * | 2019-05-01 | 2022-03-16 | Ibm | DNN training with asymmetric RPU devices |
| US11562249B2 (en) | 2019-05-01 | 2023-01-24 | International Business Machines Corporation | DNN training with asymmetric RPU devices |
| CN111474585A (en) * | 2020-03-16 | 2020-07-31 | 中国石油天然气集团有限公司 | Gas layer identification method based on deep learning and rock mechanical parameters |
| CN111694856A (en) * | 2020-06-11 | 2020-09-22 | 中国石油大学(北京) | Intelligent prediction method and device for reservoir sensitivity |
| CN111694856B (en) * | 2020-06-11 | 2021-03-12 | 中国石油大学(北京) | Intelligent prediction method and device for reservoir sensitivity |
| CN112253087A (en) * | 2020-10-20 | 2021-01-22 | 河南理工大学 | Biological disturbance reservoir physical property calculation method based on multi-source logging data |
| CN113176608A (en) * | 2021-04-27 | 2021-07-27 | 中国石油大学(华东) | DAS six-component seismic signal decoupling and recovery method for spirally-wound optical fiber |
| CN114909125A (en) * | 2022-06-08 | 2022-08-16 | 中国地质大学（北京） | Text type rock debris logging data numeralization and stratum automatic comparison method and device |
| CN116757096A (en) * | 2023-08-16 | 2023-09-15 | 中国石油天然气集团有限公司 | Determination method, device, equipment and medium for deep, thin and compact sandstone gas reservoir development well |
| CN116757096B (en) * | 2023-08-16 | 2024-04-05 | 中国石油天然气集团有限公司 | Determination method, device, equipment and medium for deep, thin and compact sandstone gas reservoir development well |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108596780B (en) | 2019-05-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108596780B (en) | A kind of Reservoir type division methods and system based on multiple information | |
| CN110852527B (en) | Reservoir physical property parameter prediction method combining deep learning | |
| Coppola Jr et al. | Artificial neural network approach for predicting transient water levels in a multilayered groundwater system under variable state, pumping, and climate conditions | |
| US7970593B2 (en) | Method for gradually modifying lithologic facies proportions of a geological model | |
| US20030078901A1 (en) | Neural network based predication and optimization for groundwater / surface water system | |
| CN108897977A (en) | A kind of uncertain attribution method of Runoff Evolution based on the simulation of big Regional Hydrologic | |
| US8818781B2 (en) | Method for operating an oil pool based on a reservoir model gradually deformed by means of cosimulations | |
| Wong et al. | Multiple permeability predictions using an observational learning algorithm | |
| CN104063626A (en) | Predication method of outer-source entrapment fullness degree | |
| Soto et al. | Permeability prediction using hydraulic flow units and hybrid soft computing systems | |
| Wang et al. | Improved permeability prediction based on the feature engineering of petrophysics and fuzzy logic analysis in low porosity–permeability reservoir | |
| CN114862035A (en) | Combined bay water temperature prediction method based on transfer learning | |
| CN104948177B (en) | A kind of PRODUCTION FORECASTING METHODS based on rice liquid production index statistical nature rule | |
| CN107290800A (en) | Log Forecasting Methodology before practical brill | |
| CN114114414A (en) | Artificial intelligence prediction method for 'dessert' information of shale reservoir | |
| NO20200978A1 (en) | Optimized methodology for automatic history matching of a petroleum reservoir model with ensemble kalman filter | |
| Soto B et al. | Improved Reservoir permeability Models from Flow Units and Soft Computing Techniques: A Case Study, Suria and Reforma-Libertad Fields, Colombia | |
| CN108765562B (en) | Oil-gas productivity evaluation method based on three-dimensional geological model | |
| CN109655394B (en) | Nuclear magnetic resonance T2 spectrum permeability calculation method under constraint of throat structure parameters | |
| US20220390633A1 (en) | Fast, deep learning based, evaluation of physical parameters in the subsurface | |
| Ma et al. | Uncertainty Analysis | |
| CN117390970B (en) | Oil gas storage quantification method considering multi-element driving prediction | |
| Khodabakhshi et al. | Inference of Global Reservoir Connectivity from Static Pressure Data with Fast Coarse-Scale Simulation Models | |
| CN110261906B (en) | Multi-parameter lithologic information fusion reservoir prediction method | |
| An et al. | Reservoir classification with multi well log parameters based on deep learning technology |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |