Method for identifying lithology by inverting particle size median of clastic rock reservoir based on XGBoost algorithm

Abstract

The invention relates to the technical field of geophysical logging, in particular to an XGBoost algorithm-based clastic rock reservoir granularity median inversion lithology identification method, which comprises the following steps of S1, constructing a granularity median M-logging multivariable data set; s2, dividing the granularity median M in the data set into a plurality of granularity types; s3, establishing a granularity type identification model, and then judging the granularity type to which the unknown depth point on the shaft section belongs; and S4, establishing a granularity median logging intelligent calculation model based on an XGBoost algorithm under granularity classification, then calculating a continuous granularity median curve, and then identifying the lithology of the clastic rock reservoir according to the corresponding relationship between the lithology and the granularity median. According to the method, granularity analysis and conventional logging information are fully mined, geological significance is combined, the granularity samples are classified, then the calculation model of the granularity median is established through the XGBoost algorithm, low-cost and high-precision granularity median continuous calculation can be achieved, and therefore low-cost and high-precision identification of the lithology of the clastic rock reservoir is achieved.

Description

Technical Field

[0001] The present invention relates to the field of geophysical logging technology, and more specifically, to a method for identifying lithology by inverting the median grain size of a clastic rock reservoir based on an XGBoost algorithm. Background Art

[0002] In conventional logging grain size inversion, Zhao Jun et al. (2016) established a multi-parameter grain size median calculation model based on natural gamma, resistivity curve and lithology index M and N based on the analysis of the relationship between lithology index parameters and grain size median. Wang Lihua (2016), Liu Shanshan et al. (2022) established a vertical grain size median curve prediction model based on machine learning and neural network technology using natural gamma and density logging curves. Yang Ning et al. (2012) calculated grain size parameters such as grain size median using binary wavelet transform method based on natural gamma curve. The prior art with publication number "CN116411958A" discloses a grain size median calculation method, system, equipment and storage medium based on logging data. The prior art is based on logging data such as acoustic time difference, neutron and density, and establishes skeleton index parameters to eliminate the influence of pore fluid on logging response, and then improves the grain size median calculation model. The prior art with publication number "CN107688037B" discloses a method for determining the downhole rock grain size curve using nuclear magnetic logging T2 distribution. The prior art realizes continuous calculation of rock grain size after determining the grain size conversion distribution coefficient in combination with nuclear magnetic experiment.

[0003] The above-mentioned prior art methods all combine well logging curves to obtain the median grain size, but all of the above methods have certain deficiencies. The accuracy of the calculation model established by multivariate regression is often low; machine learning directly uses grain size parameters as input, ignoring the geological significance behind the relationship between well logging curves and grain size parameters; the model established based on the wavelet transform mathematical method is relatively simple and there is a difference with the actual formation situation, which is not suitable for the fine identification of complex lithology; the physical model construction determines that the skeleton parameters are different for different wells and different lithologies, and the accuracy is difficult to guarantee; the cost of combining nuclear magnetic resonance experiments to determine the grain size conversion distribution coefficient is high. Summary of the invention

[0004] In view of the problem that the identification methods in the above-mentioned prior art have high identification cost and low accuracy, the present invention provides a method for identifying lithology by inverting the median grain size of clastic reservoirs based on the XGBoost algorithm, which can achieve low-cost and high-precision identification of clastic reservoir lithology.

[0005] In order to solve the above technical problems, the technical solution provided by the present invention is:

[0006] A method for identifying lithology by inverting the median grain size of clastic reservoirs based on an XGBoost algorithm comprises the following steps:

[0007] S1: Determine the area to be identified, perform drilling and logging in the area to be identified, and then construct a granularity median M-logging multivariate data set;

[0008] S2: dividing the particle size median M in the data set into multiple particle size types;

[0009] S3: Establish a particle size type recognition model with well logging as input, determine the well logging boundary, and then identify the particle size type to which the unknown depth point on the wellbore profile belongs;

[0010] S4: Establish a grain size median logging intelligent calculation model based on the XGBoost algorithm under grain size classification, then calculate the continuous grain size median curve, and finally identify the lithology of the clastic reservoir in the area to be identified based on the corresponding relationship between lithology and grain size median.

[0011] In the above technical scheme, after the particle size samples are classified, the XGBoost algorithm is used to establish a calculation model for the median particle size, which can realize low-cost and high-precision continuous calculation of the median particle size, thereby realizing low-cost and high-precision identification of the lithology of the clastic reservoir.

[0012] Preferably, in step S1, the variables in the data set include natural gamma ray (GR), resistivity (P40H), density (RHOB), neutron porosity (TNPH), and time difference of acoustic wave (DTCO).

[0013] Preferably, in step S1, core observation, grain size analysis and logging curve data are first used to analyze the relationship between the median grain size M, natural gamma (GR), resistivity (P40H), density (RHOB), neutron porosity (TNPH), and time difference of acoustic wave (DTCO) and the lithology type and logging response respectively; after the sensitive logging curve variables reflecting the median grain size are clearly defined, the data set is constructed.

[0014] Preferably, in step S2, the K-MEANS clustering method is used to determine the optimal number of clusters, and then the granularity median M in the data set is divided into multiple granularity types according to the optimal relationship between the sum of squares of cluster number errors and the number of clusters, wherein the number of granularity types is equal to the optimal number of clusters.

[0015] Preferably, in step S2, when n (n≤5) principal components are extracted and the cumulative variance contribution rate is greater than 85%, the number of principal components extracted is determined to be n; then each principal component is used to replace the 5 logging curve variables sensitive to the median particle size as the clustering input, and the relationship between the sum of squares of cluster errors SSE and the number of clusters k is obtained, and then the optimal number of clusters is determined based on the relationship.

[0016] Preferably, in step S2, when the cluster number k exceeds a certain recent value and the cluster number error sum of squares SSE tends to be flat, the recent value is used as the optimal cluster number.

[0017] Preferably, in step S3, there are four types of particle sizes, namely siltstone, fine sandstone, medium sandstone and coarse sandstone.

[0018] Preferably, in step S3, the identification method adopted by the particle size type identification model is Fisher discriminant analysis. The aforementioned clustering method divides the known particle size median-logging response (5 curve variables) sample data into 4 categories, but it is impossible to discriminate any depth sample point on the unknown wellbore profile. The basic idea of ​​Fisher analysis and discrimination is to use projection technology to project points in multidimensional space into one-dimensional direction, and then discriminate the category of unknown sample points based on the degree of closeness between unknown sample points and known sample types. It can optimize the separation between categories, improve classification accuracy, and provide a linear discriminant function, making the classification results easier to understand and explain.

[0019] Preferably, in step S3, the identification step of the particle size type identification model is to calculate the Mahalanobis distance between the sample to be identified and the projections of the remaining samples, and use the particle size median-logging response sample data to obtain the discriminant function, and then select part of the discriminant function in combination with the Mahalanobis distance data to determine which category the sample to be identified belongs to.

[0020] Preferably, in step S3, a discriminant function with a cumulative contribution rate greater than 85% is selected in combination with Mahalanobis distance data to discriminate which category the sample to be identified belongs to. Selecting a discriminant function with a cumulative contribution rate greater than 85% is a strategy for data dimensionality reduction and feature selection. By selecting a discriminant function with a high cumulative contribution rate, redundant features can be effectively reduced, and features with a small contribution to classification can be removed, thereby simplifying the model, improving computational efficiency, and enhancing the interpretability of the model.

[0021] Beneficial effects of the present invention:

[0022] (1) By fully exploring the grain size analysis and conventional logging data, combined with geological significance, the grain size samples are classified and then the calculation model of the median grain size is established using the XGBoost algorithm. This can achieve low-cost, high-precision continuous calculation of the median grain size, thereby achieving low-cost and high-precision identification of the lithology of clastic reservoirs.

[0023] (2) Before establishing the recognition model, the particle size samples are effectively classified. Based on the particle size median M-logging 5-variable data set, the K-MEANS clustering method is used to divide the data set into four categories according to the optimal relationship between the sum of squares of the sample cluster number errors and the number of clusters. After classification, the correlation between the particle size median M and the logging response can be optimized.

[0024] (3) In the actual well data processing process, the Fisher discriminant equation is used to discriminate the grain type of the unknown depth point, and then the grain size median logging intelligent calculation model based on the XGBoost algorithm is established after grain size classification; by inverting the grain size median M curve based on logging on the wellbore profile, the purpose of fine lithology identification is achieved by using the grain size median range corresponding to different lithologies, which also lays the foundation for sedimentary analysis on the vertical profile and fine interpretation and effectiveness evaluation of reservoir architecture. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is a flow chart of a method for identifying lithology by inverting the median grain size of clastic reservoirs based on the XGBoost algorithm;

[0026] Figure 2 It is a schematic diagram of the response relationship between the median particle size M and natural gamma (GR);

[0027] Figure 3 It is a schematic diagram of the response relationship between the median particle size M and the neutron porosity (TNPH);

[0028] Figure 4 It is a schematic diagram of the response relationship between the median particle size M and the resistivity (P40H);

[0029] Figure 5 It is a schematic diagram of the response relationship between the median particle size M and density (RHOB);

[0030] Figure 6 It is a schematic diagram of the response relationship between the median particle size M and the time difference of sound wave (DTCO);

[0031] Figure 7 It is a schematic diagram of the calculation results of the eigenvalue and eigenvector of principal component analysis;

[0032] Figure 8 It is a schematic diagram of the relationship between the sum of squared cluster errors SSE and the number of clusters k;

[0033] Fig. 9 It is a schematic diagram of the correspondence between the four types of lithology and cluster categories;

[0034] Fig.10 This is a schematic diagram of the comparison between the XGBoost intelligent computing model and the multivariate regression computing model for the second segment of the Huangliu Formation in Well L-1;

[0035] Fig.11 It is a schematic diagram of the corresponding relationship between the median grain size M and lithology;

[0036] Fig.12 This is a schematic diagram of the lithology results of grain size classification in Well L-1 in the study area;

[0037] Fig.13It is a schematic diagram of the calculation results of Fisher discriminant eigenvalue and contribution rate. DETAILED DESCRIPTION

[0038] The technical solution of the present invention is further described in detail below through specific embodiments and in conjunction with the accompanying drawings:

[0039] Example 1

[0040] In this example, the Huangliu Formation in the L area of ​​the Y basin is used as the area to be identified, and a method for identifying lithology by inverting the median grain size of clastic reservoirs based on the XGBoost algorithm is used to identify its lithology. Figure 1 As shown, the method comprises the following steps:

[0041] S1: Determine the area to be identified, drill and log wells in the area to be identified, and then construct a granular median M-logging multivariate dataset;

[0042] S2: Divide the median particle size M in the data set into multiple particle size types;

[0043] S3: Establish a particle size type recognition model with well logging as input, determine the well logging boundary, and then identify the particle size type to which the unknown depth point on the wellbore profile belongs;

[0044] S4: Establish an intelligent calculation model for median grain size logging based on the XGBoost algorithm under grain size classification, then calculate the continuous median grain size curve, and finally identify the lithology of the clastic reservoir in the area to be identified based on the correspondence between lithology and median grain size.

[0045] Further, in step S1, the parameters in the data set include median grain size M, natural gamma (GR), resistivity (P40H), density (RHOB), neutron porosity (TNPH), and time difference of sonic wave (DTCO), wherein the variables include natural gamma (GR), resistivity (P40H), density (RHOB), neutron porosity (TNPH), and time difference of sonic wave (DTCO).

[0046] Furthermore, in step S1, core observation, grain size analysis and logging data are first used to analyze the relationship between the median grain size M, natural gamma (GR), resistivity (P40H), density (RHOB), neutron porosity (TNPH), and acoustic time difference (DTCO) and the lithology type and logging response respectively; after clarifying the sensitive logging variables that reflect the median grain size, the data set is constructed. Take the scatter plot of the median grain size data of 8 wells in area L and the corresponding 164 sample points of the logging curve value as an example, as shown in Figures 2 to 6As shown, it can be seen that the change trends and correlations of the median grain size M and natural gamma (GR), resistivity (P40H), density (RHOB), neutron porosity (TNPH), and time difference of sonic wave (DTCO) are as follows. With the increase of the median grain size M, GR, TNPH, RHOB, and DTCO decrease, and P40H increases. Based on this, a data set of GR, TNPH, RHOB, DTCO, P40H and median grain size is established.

[0047] Furthermore, in step S2, the K-MEANS clustering method is used to determine the optimal number of clusters, and then the granularity median M in the data set is divided into multiple granularity types according to the optimal relationship between the sum of squares of cluster number errors and the number of clusters, where the number of granularity types is equal to the optimal number of clusters.

[0048] Furthermore, in step S2, when n (n≤5) principal components are extracted and the cumulative variance contribution rate is greater than 85%, the number of principal components extracted is determined to be n; then each principal component is used to replace the five logging curve variables sensitive to the median particle size as clustering input, and the relationship between the sum of squares of cluster errors SSE and the number of clusters k is obtained, and then the optimal number of clusters is determined based on the relationship.

[0049] Further, in step S2, when the cluster number k exceeds a certain recent value, the cluster number error sum of squares SSE tends to be flat, and the recent value is used as the optimal cluster number. Taking the L area as an example, after principal component analysis, when three principal components (Y1, Y2 and Y3) are selected, the cumulative variance contribution rate is greater than 85%, so the number of principal components extracted is determined to be 3, and Y1, Y2, and Y3 are used to replace the five logging curve variables sensitive to the median particle size as clustering inputs, where the principal component analysis eigenvalue and eigenvector calculation results are as follows: Figure 7 For the sample data set of 164 groups of median grain size-logging response (5 curve variables) from 8 wells, the above three principal component variables are extracted as input for clustering, and the relationship between the sum of squares of cluster number errors SSE and the value of cluster number k is obtained as shown in Figure 8 When the number of clusters k exceeds 4, the sum of squares of cluster errors (SSE) tends to be flat, so 4 is selected as the optimal number of clusters, and the median particle size characteristics and logging response laws of the four categories are obtained, as shown in the following figure: Fig. 9 shown.

[0050] Specifically, in step S3, there are four types of particle sizes, namely siltstone, fine sandstone, medium sandstone and coarse sandstone.

[0051] Specifically, in step S3, the identification method adopted by the particle size type identification model is Fisher discriminant analysis. Fisher discriminant analysis is to analyze and identify unknown samples based on known sample category data sets, and determine the category to which they belong. The aforementioned clustering method divides the known particle size median-well logging response (5 curve variables) sample data into 4 categories, but it is impossible to distinguish any depth sample point on the unknown wellbore profile. The basic idea of ​​Fisher analysis and discrimination is to use projection technology to project points in multidimensional space into one-dimensional direction, and then discriminate the category of unknown sample points based on the degree of closeness between unknown sample points and known sample types. Therefore, the discriminant function is still obtained for the above 164 groups of particle size median-well logging response (5 curve variables) sample data as follows:

[0052] F1=-0.0719*GR+0.0074*P40H-0.6627*RHOB+0.0129*TNPH+0.0064*DTCO+6.1172;

[0053] F2=0.0326*GR+0.105*P40H+0.5282*RHOB+0.0075*TNPH+0.0016*DTCO-5.5673;

[0054] F3=-0.009*GR+0.0186*P40H-0.7833*RHOB-0.0231*TNPH+0.0703*DTCO-2.6305;

[0055] F4=0.0307*GR-0.0025*P40H-4.984*RHOB-0.1554*TNPH+0.0064*DTCO+10.848;

[0056] F5=-0.005*GR-0.0156*P40H+5.3668*RHOB-0.1113*TNPH+0.0168*DTCO-13.2858.

[0057] The effect verification of this embodiment: Taking the second section of the Huangliu Formation in the L-1 well as an example, the XGBoost intelligent calculation model for particle size median classification and the multivariate regression model for particle size classification are used to calculate the particle size median curve, see Fig.10, Track 7 is a comparison between the calculated results of the multivariate regression model and the measured values ​​of the median particle size, and Track 8 is a comparison between the calculated results of the XGBoost intelligent model and the measured values ​​of the median particle size. The comparison of the calculated results of the two models shows that at the specific measured depth point, the calculated values ​​of the XGBoost model are in better agreement with the measured values ​​of the median particle size than the multivariate regression model, and the median particle size curve after high-frequency filtering eliminates some unreasonable burrs and tends to be smoother, indicating that the median particle size logging intelligent calculation model based on the XGBoost algorithm has good learning efficiency, the prediction results have high accuracy, and can invert to obtain an accurate median particle size curve.

[0058] Combination Fig.11 The corresponding relationship between the median grain size of the L curve and the lithology shown can realize the fine identification of the lithology on the wellbore section. The identification results are shown in Fig.12 As shown. Among them, Fig.11 Track 7 is the logging lithology, Track 8 is the Fisher discriminant classification, Track 9 is the comparison between the calculated granularity curve and the actual granularity of the wall core based on the median granularity calculation model of the XGBoost algorithm for grain size classification, and Track 10 is the classification of lithology according to the median granularity curve. The median granularity M calculated in Track 9 is in good agreement with the actual median granularity M and has a small error value. The logging lithology in Track 7 shows that this section is a large section of medium sandstone, and the lithology analysis divided by the median granularity in Track 10 shows that the 4 types of sandstone lithologies in this section exist and the lithology changes are complex. The different lithologies identified by the median granularity curve are compared with the electrical imaging lithology after color calibration. The four lithologies correspond well, which confirms the feasibility of the model to achieve fine identification of lithology by granularity.

[0059] Beneficial effects of this embodiment:

[0060] (1) By fully exploring the grain size analysis and conventional logging data, combined with geological significance, the grain size samples are classified and then the calculation model of the median grain size is established using the XGBoost algorithm. This can achieve low-cost, high-precision continuous calculation of the median grain size, thereby achieving low-cost and high-precision identification of the lithology of clastic reservoirs.

[0061] (2) Before establishing the recognition model, the particle size samples are effectively classified. Based on the particle size median M-logging 5-variable data set, the K-MEANS clustering method is used to divide the data set into four categories according to the optimal relationship between the sum of squares of the sample cluster number errors and the number of clusters. After classification, the correlation between the particle size median M and the logging response can be optimized.

[0062] (3) In the actual well data processing process, the Fisher discriminant equation is used to discriminate the grain type of the unknown depth point, and then the grain size median logging intelligent calculation model based on the XGBoost algorithm is established after grain size classification; by inverting the grain size median M curve based on logging on the wellbore profile, the purpose of fine lithology identification is achieved by using the grain size median range corresponding to different lithologies, which also lays the foundation for sedimentary analysis on the vertical profile and fine interpretation and effectiveness evaluation of reservoir architecture.

[0063] Example 2

[0064] This embodiment further explains step S3 based on embodiment 1. In step S3, the identification step of the particle size type identification model is to calculate the Mahalanobis distance between the sample to be identified and the projections of the remaining samples, and use the particle size median-logging response sample data to obtain the discriminant function, and then select part of the discriminant function in combination with the Mahalanobis distance data to determine which category the sample to be identified belongs to.

[0065] The other features, working principles and beneficial effects of this embodiment are consistent with those of Embodiment 1.

[0066] Example 3

[0067] This embodiment further explains step S3 based on embodiment 2. In step S3, a discriminant function with a cumulative contribution rate of > 85% is selected and combined with Mahalanobis distance data to determine which category the sample to be identified belongs to. Fig.13 The calculation results of Fisher discriminant eigenvalue and contribution rate are shown. It can be seen that the cumulative contribution rate of the first and second discriminant functions is 99.01%>85.0%. Therefore, the first and second discriminant equations combined with Mahalanobis distance can achieve accurate discrimination of unknown samples. Selecting a discriminant function with a cumulative contribution rate exceeding 85% is a strategy for data dimensionality reduction and feature selection. By selecting a discriminant function with a high cumulative contribution rate, redundant features can be effectively reduced, and features with less contribution to classification can be removed, thereby simplifying the model, improving computational efficiency, and enhancing the interpretability of the model.

[0068] Obviously, the above embodiments of the present invention are merely examples for clearly explaining the present invention, and are not intended to limit the implementation methods of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description, and it is not necessary and impossible to list all the implementation methods here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A method for identifying lithology by inverting the median grain size of clastic reservoirs based on the XGBoost algorithm. It is characterized in that The following steps are involved: S1: Determine the area to be identified, perform drilling and logging in the area to be identified, and then construct a granularity median-logging multivariate data set; S2: dividing the particle size medians in the data set into multiple particle size types; S3: Establish a particle size type recognition model with well logging as input, determine the well logging boundary, and then identify the particle size type to which the unknown depth point on the wellbore profile belongs; S4: Establish a grain size median logging intelligent calculation model based on the XGBoost algorithm under grain size classification, then calculate the continuous grain size median curve, and finally identify the lithology of the clastic reservoir in the area to be identified based on the corresponding relationship between lithology and grain size median.

2. According to claim 1, a method for identifying lithology by inverting the median grain size of clastic reservoirs based on the XGBoost algorithm, It is characterized in that In the step S1, the variables in the data set include natural gamma ray (GR), resistivity (P40H), density (RHOB), neutron porosity (TNPH), and time difference of acoustic wave (DTCO).

3. According to claim 2, a method for identifying lithology by inverting the median grain size of clastic reservoirs based on the XGBoost algorithm, It is characterized in that In the step S1, core observation, grain size analysis and logging curve data are first used to analyze the relationship between the median grain size M, natural gamma (GR), resistivity (P40H), density (RHOB), neutron porosity (TNPH), and time difference of acoustic wave (DTCO) and the lithology type and logging response respectively; after the sensitive logging curve variables reflecting the median grain size are clearly defined, the data set is constructed.

4. According to claim 3, a method for identifying lithology by inverting the median grain size of clastic reservoirs based on the XGBoost algorithm, It is characterized in that In step S2, the K-MEANS clustering method is used to determine the optimal number of clusters, and then the granularity median M in the data set is divided into multiple granularity types according to the optimal relationship between the sum of squares of cluster number errors and the number of clusters, wherein the number of granularity types is equal to the optimal number of clusters.

5. According to claim 4, a method for identifying lithology by inverting the median grain size of clastic reservoirs based on the XGBoost algorithm, It is characterized in that In step S2, when n (n≤5) principal components are extracted and the cumulative variance contribution rate is greater than 85%, the number of principal components extracted is determined to be n; then each principal component is used to replace the five logging curve variables sensitive to the median particle size as clustering input, and the relationship between the sum of squares of cluster number errors and the number of clusters is obtained, and then the optimal number of clusters is determined based on the relationship.

6. According to claim 5, a method for identifying lithology by inverting the median grain size of clastic reservoirs based on the XGBoost algorithm, It is characterized in that In step S2, when the cluster number k exceeds a certain recent value and the cluster number error sum of squares tends to be flat, the recent value is used as the optimal cluster number.

7. The method for identifying lithology by inverting the median grain size of clastic reservoirs based on the XGBoost algorithm according to claim 5, It is characterized in that In step S3, there are four types of particle sizes, namely siltstone, fine sandstone, medium sandstone and coarse sandstone.

8. According to claim 3, a method for identifying lithology by inverting the median grain size of clastic reservoirs based on the XGBoost algorithm, It is characterized in that In the step S3, the particle size type recognition model adopts the Fisher discriminant analysis method.

9. The method for identifying lithology by inverting the median grain size of clastic reservoirs based on the XGBoost algorithm according to claim 8, It is characterized in that In step S3, the identification step of the particle size type identification model is to calculate the Mahalanobis distance between the sample to be identified and the projections of the remaining samples, and use the particle size median-logging response sample data to obtain the discriminant function, and then select part of the discriminant function in combination with the Mahalanobis distance data to determine which category the sample to be identified belongs to.

10. The method for identifying lithology by inverting the median grain size of clastic reservoirs based on the XGBoost algorithm according to claim 9, It is characterized in that In the step S3, a discriminant function with a cumulative contribution rate greater than 85% is selected in combination with Mahalanobis distance data to discriminate which category the sample to be identified belongs to.

Images