CN114755744A

CN114755744A - Total organic carbon well logging interpretation method and system based on mud shale heterogeneity characteristics

Info

Publication number: CN114755744A
Application number: CN202210249260.7A
Authority: CN
Inventors: 程明; 罗晓容; 雷裕红; 张立宽; 李超; 刘乃贵
Original assignee: Institute of Geology and Geophysics of CAS
Current assignee: Institute of Geology and Geophysics of CAS
Priority date: 2022-03-14
Filing date: 2022-03-14
Publication date: 2022-07-15

Abstract

The embodiment of the invention provides a total organic carbon well logging interpretation method and a system based on the heterogeneity characteristic of shale, wherein the method comprises the following steps: acquiring relevant data of shale layer sections; obtaining the distribution of the type of the homogeneous unit of each single-well core-taking section along with the depth according to the core description result and the core sample test data in the modeling data; aiming at each homogeneous unit type, establishing a homogeneous unit logging identification model by adopting a machine learning algorithm; respectively establishing respective TOC logging interpretation models for each homogeneous unit type by adopting a machine learning algorithm; and identifying the types of the homogeneous units of different depth sections of the single well by using the well drilling data set to be predicted and the homogeneous unit logging identification model, and carrying out TOC logging interpretation of shale layer sections according to the TOC logging interpretation model of each homogeneous unit. By constructing rock facies types with relatively homogeneous rock components except TOC, TOC well logging interpretation models are respectively established for different rock types, so that TOC interpretation is more accurate.

Description

Total organic carbon well logging interpretation method and system based on mud shale heterogeneity characteristics

Technical Field

The invention relates to logging evaluation of key parameters of a shale reservoir, in particular to a total organic carbon logging interpretation method and system based on shale heterogeneity characteristics.

Background

The organic matter content is essential basic data for the work of hydrocarbon source rock evaluation, shale oil and gas resource evaluation, dessert prediction and the like. At present, a laboratory core test analysis technology is the most direct and accurate means for obtaining the organic matter content of shale, wherein Total Organic Carbon (TOC) is the most common organic matter content characterization index. The TOC interpretation is developed by high-resolution and high-coverage well logging data under the limitation of most of drilling non-core data or incomplete coring, and is an important means for evaluating the integral organic matter content of the shale section.

As shale oil and gas reservoir exploration and development progresses, a number of TOC well logging interpretation methods, techniques or models have been proposed. These methods can be divided into two broad categories, model-driven and data-driven. In the early development stage of TOC well logging interpretation technology, model driving type interpretation methods are developed in an important mode and comprise a stratum density curve method, a GR curve method, a delta logR and other classical interpretation methods. In the last two decades, with the development of artificial intelligence technology, data-driven methods gradually become research hotspots, and TOC well logging interpretation technologies based on algorithms such as BP neural network, SVM, naive bayes and the like are proposed and widely applied successively.

The purpose of model-driven or data-driven TOC well logging interpretation techniques is to find models that establish statistical relationships between log response values and TOC. Generally, a model-driven interpretation method is used for constructing a model (or assisting other parameters) by taking the logging response characteristic of an organic matter poor interval as a datum line and taking the relation between the amplitude of a logging response value offset from the datum line and the TOC content, for example, a delta logR method is used for calculating a delta LogR value by taking a water-containing sandstone segment as a datum interval and by the amplitude of the deviation of the time difference of resistivity and sound wave from the datum, and then a relevant model of the value and the TOC is established on the basis of considering the maturity of a hydrocarbon source rock; the data driving type interpretation method does not depend on a datum line of the poor organic matter interval, and after a well logging series closely related to the TOC is optimized, a statistical relationship between a logging response and the TOC is excavated through an artificial intelligence algorithm, and then the statistical relationship is utilized to carry out the well logging interpretation of the TOC.

However, in practice, the log response of a shale section is affected by, in addition to the organic matter content, other reservoir rock formations, including particle size, mineral content (especially heavy mineral and radioactive element content), degree of pore development, pore fluid properties, organic matter occurrence, and the like. These rock formation characteristics, together with the organic matter content, affect the log response values, thereby making the statistical relationship between TOC and log response values exceptionally complex. However, in conventional model-driven and data-driven TOC interpretation methods, the rock formation characterization data described above is not used to constrain modeling of log response-TOC relationships. The above situations are caused by a plurality of reasons, for example, the TOC data acquisition cost is low, the acquisition cost of other rock texture data is high, and a large-scale data set required by modeling cannot be formed; in the rock texture information, partial qualitative or semi-quantitative data exist, such as an organic texture occurrence form, and cannot be used for quantitative modeling.

In addition, TOC test samples are often sampled at equal intervals or randomly, which inevitably results in fewer samples in certain value intervals, and makes the data set characterized by unbalanced data distribution. Generally, data driving methods focus on data of samples with high distribution frequency, and the data driving methods have the characteristic of low generalization capability due to low prediction accuracy of data with few samples. Unfortunately, the TOC interval with lower data density may be the key interval of interest for shale reservoir research, such as the high value organic carbon content interval.

Disclosure of Invention

The invention provides a total organic carbon logging interpretation method and system based on shale heterogeneity characteristics, starts from the heterogeneity of a shale stratum, establishes a TOC logging interpretation model for different homogeneous unit types by constructing homogeneous unit types with relatively homogeneous rock components except TOC, avoids the condition of under-fitting or over-fitting of the TOC interpretation model, enhances the generalization capability of the interpretation model, enables the TOC interpretation to be more accurate, and is beneficial to fine evaluation of a shale reservoir.

In order to achieve the above object, in one aspect, an embodiment of the present invention provides a total organic carbon well logging interpretation method based on shale heterogeneity characteristics, including:

Obtaining shale interval related data, wherein the shale interval related data comprises two types: modeling a data set and a drilling data set to be predicted; the modeling data set comprises a core description result of a core taking section, core sample testing data and logging data, and the drilling data set to be predicted comprises the logging data; in the modeling data set, the test data of the rock core sample comprises characteristic parameters after the TOC test analysis of the total organic carbon content of the mineralogy and the organic petrology, wherein the characteristic parameters are quantitative data and typed qualitative data; the total organic carbon content is an index for characterizing the content of organic matters in the shale;

establishing a dividing standard of the type of the homogeneous unit according to a rock core description result and rock core sample test data in the modeling data, and dividing the type of the homogeneous unit of the coring section by using the dividing standard of the type of the homogeneous unit to obtain the distribution of the type of the homogeneous unit of each single-well coring section along with the depth;

aiming at each homogeneous unit type, establishing a homogeneous unit logging identification model by adopting a machine learning algorithm, and realizing prediction of the homogeneous unit type by utilizing shale stratum interval logging response;

respectively establishing respective TOC logging interpretation models for each homogeneous unit type by adopting a machine learning algorithm;

And identifying the types of the homogeneous units of different depth sections of a single well by using the well drilling data set to be predicted and the homogeneous unit logging identification model, and carrying out TOC logging interpretation of shale layer sections according to the TOC logging interpretation model of each homogeneous unit.

In another aspect, an embodiment of the present invention provides a total organic carbon well logging interpretation system based on a characteristic of heterogeneity of shale, including:

the data acquisition unit is used for acquiring the related data of the shale interval, and the related data of the shale interval comprises two types: modeling a data set and a drilling data set to be predicted; the modeling data set comprises a core description result of a coring section, core sample testing data and logging data, and the drilling data set to be predicted comprises the logging data; in the modeling data set, the test data of the rock core sample comprises characteristic parameters after the TOC test analysis of the total organic carbon content of the mineralogy and the organic petrology, wherein the characteristic parameters are quantitative data and typed qualitative data; the total organic carbon content is an index for characterizing the content of organic matters in the shale;

the homogeneous unit type dividing unit is used for establishing a dividing benchmark of the homogeneous unit type according to a rock core description result and rock core sample test data in the modeling data, and dividing the type of the homogeneous unit of the coring section by using the dividing benchmark of the homogeneous unit type to obtain the distribution of the type of the homogeneous unit of the coring section of each single well along with the depth;

The homogeneous unit type prediction unit is used for establishing a homogeneous unit logging identification model by adopting a machine learning algorithm aiming at each homogeneous unit type so as to realize prediction of the homogeneous unit type by using shale interval logging response;

the TOC well logging interpretation model establishing unit is used for respectively establishing respective TOC well logging interpretation models for each homogeneous unit type by adopting a machine learning algorithm;

and the TOC logging interpretation unit is used for identifying the types of the homogeneous units of different depth sections of the single well by using the drilling data set to be predicted and the homogeneous unit logging identification model, and carrying out TOC logging interpretation of the shale interval according to the TOC logging interpretation model of each homogeneous unit.

The technical scheme has the following beneficial effects: starting from the view of the heterogeneity of the shale stratum, the TOC well logging interpretation model is respectively established for different rock types by constructing the rock facies types with relatively homogeneous rock components except the TOC, so that the condition of under-fitting or over-fitting of the TOC interpretation model is avoided, the generalization capability of the interpretation model is enhanced, the TOC interpretation is more accurate, and the fine evaluation of the shale reservoir stratum is facilitated.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.

FIG. 1 is a flow chart of a total organic carbon well logging interpretation method based on shale heterogeneity characteristics according to an embodiment of the present invention;

FIG. 2 is a block diagram of a total organic carbon log interpretation system based on mud shale heterogeneity characteristics according to an embodiment of the invention;

FIG. 3 is a characterization of different homogeneous cell types within the domain of an embodiment of the invention;

FIG. 4 is a YY22 well homogenization unit type, vertical distribution of TOC, and log response thereof for an embodiment of the invention;

FIG. 5 is a YY22 well homogeneous unit logging identification model construction and its logging response of an embodiment of the invention;

FIG. 6 is a process of SVM algorithm hyperparameter optimization for an embodiment of the invention;

FIG. 7 is a graph of measured TOC and predicted TOC intersections for an embodiment of the invention;

FIG. 8 is an FY3 well homogenization unit type and TOC prediction of an embodiment of the invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

As shown in fig. 1, in conjunction with an embodiment of the present invention, there is provided a total organic carbon well logging interpretation method based on shale heterogeneity characteristics, including:

s101: obtaining shale interval related data, wherein the shale interval related data comprises two types: modeling a data set and a drilling data set to be predicted; the modeling data set comprises a core description result of a core taking section, core sample testing data and logging data, and the drilling data set to be predicted comprises the logging data; in the modeling data set, the test data of the rock core sample comprises characteristic parameters after the TOC test analysis of the total organic carbon content of the mineralogy and the organic petrology, wherein the characteristic parameters are quantitative data and typed qualitative data; the total organic carbon content is an index for characterizing the content of organic matters in the shale;

s102: establishing a dividing standard of the type of the homogeneous unit according to a rock core description result and rock core sample test data in the modeling data, and dividing the type of the homogeneous unit of the coring section by using the dividing standard of the type of the homogeneous unit to obtain the distribution of the type of the homogeneous unit of each single-well coring section along with the depth;

s103: aiming at each homogeneous unit type, establishing a homogeneous unit logging identification model by adopting a machine learning algorithm, and realizing prediction of the homogeneous unit type by utilizing shale stratum interval logging response;

S104: respectively establishing respective TOC well logging interpretation models for each homogeneous unit type by adopting a machine learning algorithm;

s105: and identifying the types of the homogeneous units of different depth sections of the single well by using the well drilling data set to be predicted and the homogeneous unit logging identification model, and carrying out TOC logging interpretation of shale layer sections according to the TOC logging interpretation model of each homogeneous unit.

Preferably, the core description result of the shale comprises: lithology, sedimentary structure, and sedimentary conformation;

step 102 specifically includes:

the lithology, the sedimentary structure and the sedimentary structure of the shale are used as basic bases for the partition of the homogenizing unit, and the test data of the core sample is used as judgment, so that any homogenizing unit to be partitioned has the following characteristics at the same time:

in addition to TOC, there is a relative homogeneity of formation characteristic parameters, i.e. other formation characteristic parameters have a relatively small distribution space;

the homogeneous unit thickness should be at least a first thickness that is greater than the minimum vertical resolution of conventional logging;

using any homogenizing unit as a dividing reference of the type of the homogenizing unit;

the well logs corresponding to the modeling data set are divided into a plurality of different homogeneous cells according to the characteristics of any homogeneous cell.

Preferably, step 103 specifically includes:

obtaining the homogeneous unit type of each logging response point from the modeling data set in the coring segment with homogeneous unit type calibration;

respectively establishing a respective logging response-homogeneous unit type data set A aiming at each homogeneous unit type;

using all logging responses and homogeneous unit type data in the logging response-homogeneous unit dataset A as a supervision dataset, and constructing a logging identification method F of the homogeneous unit type on the basis of the supervision dataset by using a data mining algorithm_lithoObtaining a homogeneous unit logging identification model F after ensuring that the identification precision of all logging response and homogeneous unit type data reaches the preset precision_litho。

That is, a sample is taken in the coring segment, the total organic carbon content of the sample is obtained through testing, then the mean unit type of the sample is obtained according to the previous mean unit distribution, the logging response corresponding to the depth is obtained according to the depth, so that a data pair corresponding to TOC-mean unit-logging is formed, and the data pairs are combined to form a TOC-mean unit type-logging response data set. Wherein, the TOC test sample is the sample with depth information in the acquisition core section, and then the sample is tested by using a TOC test means to obtain a corresponding TOC value.

Preferably, all the well-log response, homogeneous unit type data in the well-log response-homogeneous unit data set A are used asMonitoring data set, and constructing well logging identification method F of homogeneous unit type based on the monitoring data set by using data mining algorithm_lithoObtaining a homogeneous unit logging identification model F after ensuring that the identification precision of all logging response and homogeneous unit type data reaches the preset precision_lithoThe method specifically comprises the following steps:

establishing a well logging identification model of a homogeneous unit type by adopting an extreme gradient lifting tree XGboost algorithm;

taking the well logging response-rock phase data of the well logging response-mean unit data set A as a supervision data set, taking the supervision data set as input data of a well logging identification model of a homogeneous unit type, training the XGboost classification model, and performing K-fold cross operation during training;

setting a fitness function of a genetic algorithm to be XGboost cross-validation accuracy; optimizing by using a genetic algorithm, so that the cross validation precision of the XGboost reaches the minimum super-parameter series, and the recognition precision is ensured to reach the preset precision; the method specifically comprises the following steps:

after a population of an initial hyper-parameter series is given, XGBoost cross verification precision corresponding to each individual in the population is obtained, the individual in the initial population is selected according to an elite retention mode, a new generation population is generated through operations such as cross and variation, the process is iterated repeatedly until set iteration times are reached or a fitness function reaches a set threshold, a hyper-parameter system enabling the fitness function to reach the lowest is selected, and the recognition precision is guaranteed to reach the preset precision.

Preferably, step 104 specifically includes:

acquiring TOC data of a core section, acquiring a corresponding mean value unit type by using depth information of each TOC test sample, and acquiring a logging response value corresponding to each TOC logging sample to form a TOC-homogeneous unit type-logging response data set B; the TOC data of the core section is from a plurality of TOC test samples, the TOC data of the core section is core sample test data, and the logging response value is logging data;

log TOC-homogeneous cell type-according to the number m of all mean cell types that all TOC log samples haveThe response data set B is divided into m TOC-log response data sets B_iWherein each TOC-log response data set B_iAll samples in (a) are from the same homogeneous cell type, where i e [1, m ∈](ii) a TOC-log response data set B of target lithology for non-shale hydrocarbons_iSetting as an empty set;

using a data mining algorithm, each TOC-log response data set B_iTaking the data in the well logging interpretation method as a supervision data set, and constructing respective TOC well logging interpretation method F of each homogeneous unit type_TOC，iAnd after the recognition precision is ensured to reach the preset precision, obtaining a TOC logging interpretation model F corresponding to each homogeneous unit type _TOC，i。

Preferably, each TOC-log response data set B is analyzed using a data mining algorithm_iTaking the data in the well logging interpretation method as a supervision data set, and constructing the respective TOC well logging interpretation method F of each homogeneous unit type_TOC，iObtaining a TOC logging interpretation model F corresponding to each homogeneous unit type after ensuring that the identification precision reaches the preset precision_TOC，iThe method specifically comprises the following steps:

taking data in each TOC-logging response data set Bi as a supervision data set, taking each supervision data set as input data, training an SVM regression algorithm, and adopting K-fold cross validation in the training of the SVM regression algorithm;

setting a fitness function of a genetic algorithm as a cross square error of the SVM under cross validation; optimizing by using a genetic algorithm to enable the cross square error to reach the minimum super-parameter series, and taking the super-parameter series with the cross square error reaching the minimum as the super-parameters in the SVM regression algorithm to obtain TOC logging data interpretation models corresponding to various homogeneous unit types;

setting a fitness function of the genetic algorithm as a cross square error of the SVM under cross validation; optimizing by using a genetic algorithm to enable a cross square error to reach a minimum hyperparameter series, which specifically comprises the following steps:

After a population of an initial hyper-parameter series is given, a cross square error corresponding to each individual in the population is obtained, the individual in the initial population is selected according to an elite retention mode, a new-generation population is generated through operations such as cross and variation, the process is iterated repeatedly until set iteration times are reached or a fitness function reaches a set threshold, the hyper-parameter series enabling the fitness function to reach the lowest is selected, and the recognition accuracy is guaranteed to reach preset accuracy.

Preferably, step 105 specifically comprises:

carrying out homogeneous unit type identification on a drilling data set to be predicted by using a homogeneous unit logging identification model to obtain the distribution of the homogeneous unit type of the drilling to be predicted along with the depth;

and respectively adopting respective TOC logging explanation models to perform TOC explanation on each type of the homogeneous unit of the well to be predicted, obtaining TOC explanation corresponding to each type of the homogeneous unit, and taking all TOC explanations corresponding to the types of the homogeneous units as TOC logging explanations of the well shale to be predicted.

As shown in fig. 2, in connection with an embodiment of the present invention, there is provided a total organic carbon logging interpretation system based on shale heterogeneity characteristics, including:

the data obtaining unit 21 is configured to obtain data related to a shale interval, where the data related to the shale interval includes two types: modeling a data set and a drilling data set to be predicted; the modeling data set comprises a core description result of a core taking section, core sample testing data and logging data, and the drilling data set to be predicted comprises the logging data; in the modeling data set, the test data of the rock core sample comprises characteristic parameters after the TOC test analysis of the total organic carbon content of the mineralogy and the organic petrology, wherein the characteristic parameters are quantitative data and typed qualitative data; the total organic carbon content is an index for characterizing the content of organic matters in the shale;

The homogeneous unit type dividing unit 22 is used for establishing a dividing standard of the homogeneous unit type according to a rock core description result and rock core sample test data in the modeling data, and dividing the homogeneous unit type of the coring section by using the dividing standard of the homogeneous unit type to obtain the distribution of the homogeneous unit type of each single-well coring section along with the depth;

the homogenizing unit type predicting unit 23 is used for establishing a homogenizing unit logging identification model by adopting a machine learning algorithm aiming at each homogenizing unit type to realize prediction of the homogenizing unit type by using shale interval logging response;

the TOC well logging interpretation model establishing unit 24 is used for respectively establishing respective TOC well logging interpretation models for each homogeneous unit type by adopting a machine learning algorithm;

and the TOC logging interpretation unit 25 is used for identifying the types of the homogeneous units of different depth sections of a single well by using the drilling data set to be predicted and the homogeneous unit logging identification model, and carrying out TOC logging interpretation of shale layer sections according to the TOC logging interpretation model of each homogeneous unit.

the homogenizing unit type dividing unit 22 is specifically configured to:

Preferably, the homogeneous unit type prediction unit 23 is specifically configured to:

obtaining the homogeneous unit type of each logging response point from the modeling data set in a coring segment with homogeneous unit type calibration;

using all logging responses and homogeneous unit type data in the logging response-homogeneous unit dataset A as a supervision dataset, and constructing a logging identification method F of the homogeneous unit type on the basis of the supervision dataset by using a data mining algorithm _lithoObtaining a homogeneous unit logging identification model F after ensuring that the identification precision of all logging response and homogeneous unit type data reaches the preset precision_litho。

Preferably, the TOC well logging interpretation model building unit 24 is specifically configured to:

acquiring TOC data of a core section, acquiring a corresponding mean value unit type by using the depth information of each TOC test sample, and acquiring a logging response value corresponding to each TOC logging sample to form a TOC-homogeneous unit type-logging response data set B; the TOC data of the core section is from a plurality of TOC test samples, the TOC data of the core section is core sample test data, and the logging response value is logging data;

dividing the TOC-homogeneous unit type-logging response data set B into m TOC-logging response data sets B according to the number m of all mean unit types of all TOC logging samples_iWherein each TOC-log response data set B_iAll samples in (a) are from the same homogeneous cell type, where i e [1, m ∈](ii) a TOC-log response data set B of target lithology for non-shale hydrocarbons_iSetting as an empty set;

using a data mining algorithm, each TOC-log response data set B_iTaking the data in the well logging interpretation method as a supervision data set, and constructing respective TOC well logging interpretation method F of each homogeneous unit type _TOC，iObtaining a TOC logging interpretation model F corresponding to each homogeneous unit type after ensuring that the identification precision reaches the preset precision_TOC，i。

Preferably, the TOC logging interpretation unit 25 is specifically configured to:

identifying the type of a homogeneous unit of the drilling data set to be predicted by using the homogeneous unit logging identification model to obtain the distribution of the type of the homogeneous unit of the drilling to be predicted along with the depth;

and respectively adopting respective TOC logging interpretation models to perform TOC interpretation for each type of the homogeneous unit of the drilling well to be predicted to obtain TOC interpretations corresponding to each type of the homogeneous unit, and taking all the TOC interpretations corresponding to the types of the homogeneous units as TOC logging interpretations of the drilling shale to be predicted.

The invention has the following beneficial effects: starting from the perspective of heterogeneity of the shale stratum, the TOC well logging interpretation model is respectively established for different rock types by constructing rock phase types with relatively homogeneous rock components except for TOC, so that the condition that the TOC interpretation model is under-fit or over-fit is avoided, the generalization capability of the interpretation model is enhanced, the TOC interpretation is more accurate, and the fine evaluation of the shale reservoir is facilitated.

Therefore, the petroleum company isophase shutdown mechanism can utilize the method to perform fine TOC interpretation and evaluation on the shale reservoir with strong heterogeneity, thereby greatly reducing the risk of the shale oil and shale gas reservoir in exploration and development. The method has wide application prospect.

The above technical solutions of the embodiments of the present invention are described in detail below with reference to specific application examples, and reference may be made to the foregoing related descriptions for technical details that are not described in the implementation process.

The invention provides a total organic carbon logging interpretation method based on shale heterogeneity characteristics, and provides an organic carbon logging interpretation technology based on the shale heterogeneity characteristics in view of the problem that strong shale heterogeneity characteristics have large influence on organic carbon logging interpretation results. The basic idea is as follows: the shale interval is divided into a plurality of intervals, so that in the longitudinal dimension of conventional logging (about 30-80 cm), other structure information except TOC has relatively homogeneous characteristics, the variation range is small, the part of structure information becomes redundant information, and the influence on logging response is relatively small; carrying out homogeneous type division on a plurality of intervals according to the similarity of the structure information, respectively establishing a TOC (total organic carbon) interpretation model of logging response for each type, and ensuring that the interpretation model in each homogeneous unit type has better generalization capability; carrying out homogeneous unit type identification on the shale interval to be predicted, dividing different types of intervals according to the homogeneous unit type, and then carrying out TOC logging interpretation on each interval by using the established TOC interpretation model.

The principle of the invention is as follows:

in essence, TOC interpretation is a typical regression problem. In the ideal case, it is assumed that X is the logging information set and Y is the formation information set (which contains all the formation information). Let X ∈ X, Y ∈ Y, all come from the joint distribution P_X×YThe fundamental goal of TOC well logging interpretation is to establish a mapping F ∈ F: x->Y, such that the desired error ε_ex＝E_{(x，y)～P(X×Y)}L(f(x)，y_toc) Minimization, wherein, L (f (x), y_toc) Is a loss function representing the difference between the predicted TOC and the measured TOC.

In practical cases, P is distributed jointly_X×YAre often unknown. x and y are both from a finite discrete supervised data set

Wherein x is_i∈X，y_iE.g. Y, n represents D_tNumber of discrete data in the data set. In this case, the TOC log interpretation is targeted to establish a mapping F ∈ F: d_t，x->D_t，ySo that epsilon_em＝E_(x，y)～DtL(f(x)，y_toc) And (4) minimizing. Therefore, to enable the TOC interpretation results to be less erroneous, the most straightforward approach is to increase the Dt data set size (i.e., the value of n) so that D is_tCapable of covering joint distribution P_X×Y. This method can ensure that epsilon ═ epsilon_ex-ε_emAnd | is small enough, so that the interpretation model has better generalization capability, and the interpretation result is ensured to have smaller error.

However, the supervised data set D contains complete geological information _tThis is not given. In most cases, the supervised data set, which often contains only TOC information in well log information and formation information, may be expressed as

Compare with D_tData set, D_t' A data set is a partially complete data set that lacks formation information y other than the TOC_-toc. Thus, the goal of the TOC log interpretation is set to: establishing a mapping relation F 'belongs to F': d_t，x->D_t，ytocSo that the expected error ε_em’＝E_(x，y)～Dt，L(f(x)，y_toc) And (4) minimizing.

The prerequisite that the interpretation model has generalization capability is: and the structure information except the organic matter content in the shale interval is more homogeneous, and the variation range is smaller. For TOC interpretation construction, this part of the rock texture information is redundant information, data set D_t' can provide with D_tInformation equivalent to a data set. Obviously, this is not compatible with the extremely strong heterogeneity of the rock structure of the shale interval, increasing D anyway_t' the size of the data set, neither can guarantee ═ ε_ex-ε_em' I is small enough, so that the interpretation model is easy to have poor generalization capability due to under-fitting or over-fitting, and the TOC interpretation result has larger error.

The method constructs a gene different from D_t' supervision data set D_t，j', which is a reaction of D_t'division into m subsets D' _t，j，

j is a non-zero natural number less than m. These data sets satisfy 3 basic conditions:

1. all subsets DD't, j_t，jIs equal to D_t', i.e. U_j＝1 ^mD_t，j’＝D_t'(note: D'_t，jCan be a null set)

2. Any two subsets are disjoint sets, i.e. D_tj＝a1’∩D_t，j＝a2' phi, a1 and a2 are non-zero natural numbers less than m.

3. Data set consisting of TOC data within a single subset

Each corresponding to a data set containing information of all the strata

Wherein, y_-tocRepresenting other stratigraphic configuration information in addition to the TOC. Then y within a single subset_-tocFeatures having low dispersion, i.e. σ (y)_-toc)＜ε。

Therefore, in the case where the above-mentioned 3 basic conditions are satisfied, the goal of building the TOC well interpretation model is set to: establishing m mapping relations f_j∈F：D_t，j，x->D_t，j，ySo that epsilon_em，j＝E_{(x，y)～Dt，j’}L(f(x)，y_toc) Thus, each ε is guaranteed_j＝/ε_em，j-ε_ex，jSufficiently small and has better generalization.

The method comprises the following specific steps:

1. obtaining relevant data of shale layer sections, wherein the sample data set comprises two types: the method comprises the steps that a modeling data set (of multiple wells) and a drilling data set to be predicted need to contain a core description result of a core section, core sample test data and logging data, and the drilling data set to be predicted only needs to contain the logging data; the core description result comprises: lithology, sedimentary structure, and mineral composition (sedimentary structure);

2. The test data of the core sample in the modeling data set comprises characteristic parameters after test analysis of mineralogy, organic petrography, physical properties, TOC and the like, wherein the characteristic parameters are quantitative data and typed qualitative data;

3. establishing a dividing scheme of a homogeneous unit according to differences in lithology, sedimentary structure and mineral composition in the core description result according to the core description result and the core sample test data in the modeling data, dividing the type of the core sample homogeneous unit by using the scheme, and establishing vertical distribution of the type of the single-well homogeneous unit (the vertical distribution refers to the distribution of the type of the homogeneous unit along with the depth); the dividing standard of the type of the homogenizing unit is determined according to the coring data of all the drilled wells, and after the determination is finished, the coring of each well is divided into single wells according to the standard.

4. In a coring segment with homogeneous unit type calibration, obtaining the homogeneous unit type of each logging response point, and establishing a logging response-mean unit data set A (also called homogeneous unit-logging response data set A);

5. taking the logging response-homogeneous unit type data (rock facies logging response-rock facies data) in the data set A as a supervision data set, and constructing a logging identification method F of the homogeneous unit type on the basis of the supervision data set by using a data mining algorithm _lithoAnd ensuring that the identification precision reaches the preset precision; and obtaining a homogeneous unit logging identification model.

6. Obtaining a homogeneous unit type corresponding to each TOC test sample, collecting a logging response value corresponding to each TOC logging sample to form a TOC-homogeneous unit type-logging response data set (the data set is represented by B), dividing the data set B into m TOC-logging response data sets according to the number m of mean unit types, wherein the samples in each TOC-logging response data set are from the same homogeneous unit type, and the TOC-logging response data sets are respectively represented by B_iIs represented by where i ∈ [1, m ]]；

7. Constructing respective TOC logging interpretation method F of each homogeneous unit type by using data mining algorithm and taking data in each TOC-logging response data set Bi as a supervision data set_TOC，iAnd ensuring that the identification precision reaches the preset precision;

8. using the drilling data set to be predicted, first using F_lithoCarrying out identification of the type of the homogenizing unit, and then respectively adopting independent F for each rock facies type on the basis of the explained result_TOC，iAnd the explanation model carries out TOC explanation to obtain a TOC explanation result of the single well and obtain a TOC logging explanation model.

In conclusion, (1) the types of the homogenizing units are divided; (2) establishing a homogeneous unit logging identification model by adopting a machine learning algorithm; (3) establishing a TOC logging data interpretation model of each homogenizing unit; (4) and (3) carrying out TOC logging interpretation of the whole well section of the shale interval by using the types (2) and (3). Geology has certain complexity, and there may be different mean cell types in each region, so (1) and (2) are necessary.

In combination with the principles of the present invention, the specific example of TOCC well logging interpretation by selecting 10 wells in the south of the Ordors basin is as follows:

1. 10 wells were selected in the southern region of the orldos basin, 9 of which were used as the modeled well dataset and 1 as the validated well dataset to be predicted, as shown in table 1. Table 1 shows that the modeling data set includes core description results of 9 well shale core-taking sections, 429 core test data and 7 basic well logs, and the verification well includes 33 core data and 7 basic well logs.

Table 1 example data set presentation

2. And establishing a homogeneous unit division standard, wherein the division basis is derived from core observation, in order to ensure that the homogeneous unit type is easy to obtain, the lithology, the sedimentary structure and the sedimentary structure of the shale are used as basic basis for the homogeneous unit division, and the test data is used as basis for judging the reasonable division. The partition criteria include two categories, one being that the homogeneous cells have a relative homogeneity of formation characteristics other than the TOC, i.e., other formation characteristics have a relatively small distribution space; the second is that the homogenizing unit should correspond to the logging dimension, and the thickness of the homogenizing unit should be above the logging vertical resolution, so the DEN with the minimum vertical resolution (the minimum vertical resolution also refers to a preset thickness) in the logging series is used as the standard in core observation, that is, the homogenizing unit should maintain homogeneity within 30cm of the core. Based on the core observation of 9 wells, five rock facies types, namely silty/fine sandstone facies, clayey shale facies, tuff facies, silty and clayey interbedded shale, and tufy and clayey interbedded shale, are distinguished in the target zone. The first three are composed of a single rock type, the second two are composed of two rock types, and the second two have relatively homogeneous characteristics on the logging scale. Fig. 3 is a statistical result of the 5 homogeneous units in the states of core, slice, mineral composition, organic content and organic occurrence, showing that there is a significant difference between the different homogeneous unit types, and the formation characteristics except TOC within the homogeneous unit types are relatively close.

3. According to the homogenization unit partition standard, the vertical homogenization unit distribution of 9 core observation wells is determined, and the mean unit attribute of the core data is determined (fig. 4 shows the vertical distribution of YY22 well homogenization units, TOC data points and logging information).

4. On the basis of the division of the single-well homogeneous unit, the logging attributes of the homogeneous unit (here, the logging attributes are logging response values, including AC, CAL, DEN, GR, K, Rt, PE, RD, RS, Th, U) are extracted first to form a homogeneous unit type-logging response data set, in this embodiment, the observed core length is 230m, and 1836 sets of homogeneous unit type-logging response data set data are obtained. The mean unit types are coded numerically, wherein tuff is 1, interbedded shale of tuff and clay layers is coded 2, argillaceous shale is coded 3, interbedded shale of silty and argillaceous is coded 4, and siltstone/fine sandstone is coded 5. Table 2 shows a portion of the data in the homogeneous unit-log response dataset.

5. And then, extracting logging attributes corresponding to core TOC data points in different homogeneous units (the logging attributes are logging response values, including values of AC, CAL, DEN, GR, PE and RT. except TOC as shown in Table 3 and form a modeling supervision data set of a TOC logging interpretation model together with the TOC) to form TOC-logging response data sets in different homogeneous unit types, wherein the TOC-logging response data sets in the tuff are empty sets because the tuff is not the target lithology of the shale oil and gas and has less data and the TOC values are all below 1.0 wt.%. In the data set construction, three comprehensive parameters of delta 1gR, U/Thh and Th/K are added as additional logging parameters, and the three parameters are often used by the predecessor for well logging TOC interpretation. Through the division process, a TOC-test response data set of 4 homogeneous units of pozzolanic and clayey interbedded shale, clayey shale, silty and clayey interbedded shale, siltstone/fine sandstone and the like is obtained, and the data volume is 171 groups, 214 groups, 72 groups and 55 groups respectively. Table 3 shows the data set for clayey shale in the TOC-log response data set.

6. The homogeneous unit-log response dataset was divided into two parts, and the corresponding data of WY1 and FY3 were extracted as the validation well dataset (number of data set 144 groups), and the other well data were assigned as the training dataset (number of data set 1353 groups).

7. And predicting the type of the homogeneous unit by using the logging information by adopting a machine learning algorithm. In the present embodiment, the XGboost algorithm (extreme gradient lifting tree) is used to build a well logging identification model of homogeneous cell type. And training the XGboost classification model by taking the modeling data set as input data, and ensuring that the trained model has the optimal robustness by using K-fold cross operation. For the super-parameter selection problem in XGboost, the present embodiment performs optimization by using a genetic algorithm, where the fitness function of the genetic algorithm is set to the cross-validation precision (mlogloss is used herein) of XGboost, and the goal of the genetic algorithm is to find a super-parameter series that minimizes the cross-validation precision. After a population of an initial hyper-parameter series is given, XGBoost cross validation precision corresponding to each individual in the population is obtained, the individual in the initial population is selected according to an elite retention mode, a new generation population is generated through operations such as cross and variation, the process is iterated repeatedly until set iteration times are reached or a fitness function reaches a set threshold, and a hyper-parameter system enabling the fitness function to reach the lowest is selected.

FIG. 5(a) shows the hyper-parameter setting of the XGBOSt algorithm and the hyper-parameter range to be optimized by the genetic algorithm in the embodiment, wherein K is 7 in K-fold; fig. 5(b) is an iterative process of average cross-validation accuracy and optimal cross-validation accuracy of XGboost when the genetic algorithm optimization algorithm is used (population number per generation is 150, and genetic iteration number is 200). The optimal hyperparametric series and XGboost model shown in fig. 5(a) were determined by the above procedure. Fig. 5(c) shows a comparison graph of the true homogeneous unit type in the WY1 well and the predicted homogeneous unit type, which shows that the homogeneous unit well logging identification method has excellent identification precision, and the identification accuracy can reach 88.3%.

8. The TOC-logging response data set is divided, firstly, 20% of data are extracted from each homogeneous unit type TOC-logging response data set to respectively form verification data sets (the data quantity is 86) of each homogeneous unit type, and the rest data respectively form modeling supervision data sets of each homogeneous unit to construct 8 sub-data sets.

9. And respectively establishing respective TOC well logging data interpretation models (TOC well logging interpretation models) for each homogeneous unit type by adopting a machine learning algorithm. In this embodiment, an SVM regression algorithm is selected to construct a TOC log data prediction model (TOC log interpretation model). Training an SVM regression algorithm by taking a training data set formed by each mean unit type as input data so as to train and obtain respective TOC logging data prediction models of the 4 unit types, and improving the robustness of an interpretation model by adopting K-fold cross validation in the training. For the hyperparameters in the SVM, the genetic algorithm is selected for optimization in the embodiment, and the fitness function of the genetic algorithm is set as the cross square error of the SVM under cross validation. The goal of the genetic algorithm is to find the series of hyperparameters that minimize cross-squared error. After a population of an initial hyper-parameter series is given, a cross square error corresponding to each individual in the population is obtained, the individuals in the initial population are selected according to an elite retention mode, a new generation population is generated through operations such as crossing and variation, the process is iterated repeatedly until set iteration times are reached or a fitness function reaches a set threshold, and a hyper-parameter system enabling the fitness function to reach the lowest is selected, as shown in fig. 6.

Table 4 shows the main parameter settings in the SVM algorithm and the genetic algorithm in the embodiment, and fig. 5 is an iterative process of the average recognition accuracy and the optimal recognition accuracy of the SVM when the genetic algorithm optimization algorithm is adopted (the population number of each generation is 150, the number of genetic iterations is 200, and K is 7). The optimal hyperparameter series and SVM interpretation model shown in Table 5 were determined by the above process. As shown in fig. 7, the accuracy index MSE 1.18 and MAPE 20.57% in the verification set by using the method is much higher than the accuracy index MSE 3.27 and MAPE 29.54% predicted by using the conventional method.

10. On the basis of establishing a homogeneous unit logging identification method and a homogeneous unit TOC logging interpretation method, performing TOC logging interpretation of a whole shale interval well section, firstly identifying the homogeneous unit by using the homogeneous unit logging identification method and logging data, and then interpreting TOC in each homogeneous unit by using the homogeneous unit TOC logging interpretation method, as shown in FIG. 8.

The invention has the following beneficial effects:

starting from the perspective of heterogeneity of the shale stratum, the TOC well logging interpretation model is respectively established for different rock types by constructing rock phase types with relatively homogeneous rock components except for TOC, so that the condition that the TOC interpretation model is under-fit or over-fit is avoided, the generalization capability of the interpretation model is enhanced, the TOC interpretation is more accurate, and the fine evaluation of the shale reservoir is facilitated.

It should be understood that the specific order or hierarchy of steps in the processes disclosed is an example of exemplary approaches. Based upon design preferences, it is understood that the specific order or hierarchy of steps in the processes may be rearranged without departing from the scope of the present disclosure. The accompanying method claims present elements of the various steps in a sample order, and are not meant to be limited to the specific order or hierarchy presented.

In the foregoing detailed description, various features are grouped together in a single embodiment for the purpose of streamlining the disclosure. This method of disclosure is not to be interpreted as reflecting an intention that the claimed embodiments of the subject matter require more features than are expressly recited in each claim. Rather, as the following claims reflect, invention lies in less than all features of a single disclosed embodiment. Thus, the following claims are hereby expressly incorporated into the detailed description, with each claim standing on its own as a separate preferred embodiment of the invention.

The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. To those skilled in the art; various modifications to these embodiments will be readily apparent, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the disclosure. Thus, the present disclosure is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

What has been described above includes examples of one or more embodiments. It is, of course, not possible to describe every conceivable combination of components or methodologies for purposes of describing the aforementioned embodiments, but one of ordinary skill in the art may recognize that many further combinations and permutations of various embodiments are possible. Accordingly, the embodiments described herein are intended to embrace all such alterations, modifications and variations that fall within the scope of the appended claims. Furthermore, to the extent that the term "includes" is used in either the detailed description or the claims, such term is intended to be inclusive in a manner similar to the term "comprising" as "comprising" is interpreted when employed as a transitional word in a claim. Furthermore, any use of the term "or" in the specification of the claims is intended to mean a "non-exclusive or".

Those of skill in the art will further appreciate that the various illustrative logical blocks, elements, and steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or combinations of both. To clearly illustrate the interchangeability of hardware and software, various illustrative components, elements, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented as hardware or software depends upon the particular application and design requirements of the overall system. Those skilled in the art may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the embodiments of the invention.

The above-mentioned embodiments are intended to illustrate the objects, technical solutions and advantages of the present invention in further detail, and it should be understood that the above-mentioned embodiments are merely exemplary embodiments of the present invention, and are not intended to limit the scope of the present invention, and any modifications, equivalent substitutions, improvements and the like made within the spirit and principle of the present invention should be included in the scope of the present invention.

Claims

1. A total organic carbon well logging interpretation method based on shale heterogeneity characteristics is characterized by comprising the following steps:

Respectively establishing respective TOC well logging interpretation models for each homogeneous unit type by adopting a machine learning algorithm;

and identifying the types of the homogeneous units of different depth sections of the single well by using the well drilling data set to be predicted and the homogeneous unit logging identification model, and carrying out TOC logging interpretation of shale layer sections according to the TOC logging interpretation model of each homogeneous unit.

2. The method for explaining total organic carbon logging based on the shale heterogeneity characteristics according to claim 1, wherein the core description results of the shale comprise lithology, sedimentary structure and sedimentary structure;

establishing a dividing benchmark of the homogeneous unit type according to a rock core description result and rock core sample test data in the modeling data; dividing the core sample homogenizing unit type by using the dividing standard of the homogenizing unit type, wherein the distribution of each single-well homogenizing unit type along with the depth specifically comprises the following steps:

in addition to TOC, the method has the relative homogeneity of the formation characteristic parameters, namely other formation characteristic parameters have relatively small distribution spaces;

The homogeneous unit thickness should be at least a first thickness that is greater than the minimum vertical resolution of conventional logging; using any homogenizing unit as a dividing reference of the type of the homogenizing unit;

and dividing the logging corresponding to the modeling data set into a plurality of different homogenizing units according to the characteristics of any homogenizing unit.

3. The shale heterogeneity characteristic-based total organic carbon well logging interpretation method according to claim 2, wherein for each homogeneous unit type, a machine learning algorithm is used to build a homogeneous unit well logging identification model, so as to predict the homogeneous unit type by using shale interval well logging responses, specifically comprising:

taking all logging responses and homogeneous unit type data in the logging response-homogeneous unit dataset A as a supervision dataset, and constructing a logging identification method F of the homogeneous unit type on the basis of the supervision dataset by using a data mining algorithm _lithoObtaining a homogeneous unit logging identification model F after ensuring that the identification precision of all logging response and homogeneous unit type data reaches the preset precision_litho。

4. The method for total organic carbon well logging interpretation based on shale heterogeneity characteristics of claim 3, wherein the establishing respective TOC well logging interpretation models for each homogeneous unit type by using a machine learning algorithm comprises:

acquiring TOC data of a core section, acquiring a corresponding mean value unit type by using depth information of each TOC test sample, and acquiring a logging response value corresponding to each TOC logging sample to form a TOC-homogeneous unit type-logging response data set B; the TOC data of the core section is from a plurality of TOC test samples, the TOC data of the core section is core sample test data, and the logging response value refers to logging data;

dividing the TOC-homogeneous unit type-logging response data set B into m TOC-logging response data sets B according to the number m of all mean unit types of all TOC logging samples_i(ii) a Wherein each TOC-log response data set B_iAll samples in the same sampleA homogeneous cell type where i ∈ [1, m ]]TOC-log response data set B of target lithology of non-shale hydrocarbons _iSetting as an empty set;

using a data mining algorithm to extract each TOC-log response data set B_iTaking the data in the well logging interpretation method as a supervision data set, and constructing the respective TOC well logging interpretation method F of each homogeneous unit type_TOC，iObtaining a TOC logging interpretation model F corresponding to each homogeneous unit type after ensuring that the identification precision reaches the preset precision_TOC，i。

5. The method for total organic carbon logging interpretation based on shale heterogeneity characteristics as claimed in claim 1, wherein the using the drilling data set to be predicted, using the homogeneous unit logging identification model to identify the homogeneous unit types of different depth segments of a single well, and performing TOC logging interpretation of shale intervals according to each homogeneous unit TOC logging interpretation model specifically comprises:

6. A total organic carbon well logging interpretation system based on shale heterogeneity characteristics is characterized by comprising:

the data acquisition unit is used for acquiring the related data of the shale interval, and the related data of the shale interval comprises two types: modeling a data set and a drilling data set to be predicted; the modeling data set comprises a core description result of a coring section, core sample testing data and logging data, and the drilling data set to be predicted comprises the logging data; in the modeling data set, the test data of the rock core sample comprises characteristic parameters after the total organic carbon content TOC test analysis of the petrochemistry of minerals and the facichemistry of organic rocks, wherein the characteristic parameters are quantitative data and typed qualitative data; the total organic carbon content is an index for characterizing the content of organic matters in the shale;

the homogeneous unit type dividing unit is used for establishing a dividing standard of the homogeneous unit type according to a rock core description result and rock core sample test data in the modeling data, and dividing the homogeneous unit type of the coring section by using the dividing standard of the homogeneous unit type to obtain the distribution of the homogeneous unit type of each single-well coring section along with the depth;

and the TOC logging interpretation unit is used for identifying the types of the homogeneous units of different depth sections of a single well by using the drilling data set to be predicted and the homogeneous unit logging identification model, and carrying out TOC logging interpretation of shale layer sections according to the homogeneous unit TOC logging interpretation model.

7. The shale heterogeneity characteristic-based total organic carbon logging interpretation system of claim 6, wherein the shale core description results comprise lithology, sedimentary structure and sedimentary configuration;

the homogenizing unit type dividing unit is specifically configured to:

the homogeneous cell thickness should be at least a first thickness that is greater than the minimum vertical resolution of conventional logging; using any homogenizing unit as a dividing reference of the homogenizing unit type;

8. The shale heterogeneity feature based total organic carbon logging interpretation system of claim 7, wherein the homogeneous unit type prediction unit is specifically configured to:

9. The shale heterogeneity characteristic-based total organic carbon logging interpretation method of claim 8, wherein the TOC logging interpretation model building unit is specifically configured to:

partitioning the TOC-homogeneous cell type-log response dataset B into a number m of all mean cell types that all TOC log samples havem TOC-log response datasets B_i(ii) a Wherein each TOC-log response data set B_iAll samples in (a) are from the same homogeneous cell type, where i e [1, m ∈]TOC-log response data set B of target lithology of non-shale hydrocarbons_iSetting as an empty set;

using a data mining algorithm, each TOC-log response data set B_iTaking the data in the well logging interpretation method as a supervision data set, and constructing respective TOC well logging interpretation method F of each homogeneous unit type_TOC，iAnd after the recognition precision is ensured to reach the preset precision, obtaining a TOC logging interpretation model F corresponding to each homogeneous unit type_TOC，i。

10. The method for interpreting total organic carbon logs based on the heterogeneity characteristics of shale as claimed in claim 6, wherein the TOC log interpretation unit is specifically configured to: