Source rock organic carbon content logging quantitative prediction method based on KNN machine learning algorithm

Abstract

The invention provides a hydrocarbon source rock organic carbon content logging quantitative prediction method based on a KNN machine learning algorithm, and the prediction method fully considers the advantages of the KNN machine learning algorithm, builds a recessive relation between a logging curve and TOC analysis data, i.e., obtains continuous hydrocarbon source rock TOC data in the longitudinal direction through the logging curve. And compared with three conventional methods, the precision of the calculation result is obviously improved.

Description

Technical Field

[0001] The invention belongs to the technical field of oil and gas exploration, and relates to a well logging quantitative prediction method of organic carbon content in source rocks based on a KNN machine learning algorithm. Background Art

[0002] Source rocks have a significant control over the distribution of oil and gas. How to quickly and accurately identify and evaluate source rocks has always been one of the most important basic tasks. The organic carbon content (Total Organic Carbon, referred to as TOC) of source rocks is the most commonly used and most important parameter, and it is a hot topic in oil and gas exploration research. Although accurate TOC data can be obtained based on coring data analysis, due to factors such as high costs, drilling coring is rare in most cases, and it is difficult to obtain continuous analysis data. The quality of source rocks is often evaluated based on limited TOC analysis data, and the results are greatly affected by the representativeness of the core samples; when the core is contaminated, the results are significantly biased and often misleading. Although vertically continuous TOC data can be obtained based on the logging curve meter, due to the limitations of the method itself, the calculation results have large errors and are difficult to meet the needs of actual work.

[0003] Conventional TOC prediction based on logging curves mainly includes single factor method, multiple regression method and ΔLgR method. The single factor method is to calculate TOC through a single logging curve. It appeared the earliest and has the characteristics of fast, intuitive, simple and efficient. Except for the good effect in some areas, the calculation error is large in most cases. The multiple regression method calculates TOC through multiple logging curves. It is more commonly used and convenient to operate. In most cases, the effect is good; but the estimation result will tend to the average value, that is, the lower value is overestimated and the high value is underestimated, and the regionality is obvious, and the difference between different regions is large. The ΔLgR method calculates TOC through resistivity curve and acoustic wave curve (density curve or neutron curve can also be used). It is the most widely used and can be applied to different lithologies (both carbonate and clastic rocks) and different maturity. The superposition of acoustic wave curve and resistivity curve can largely eliminate the influence of porosity; but there are difficulties such as large subjective factors in baseline determination and difficulty in determining maturity parameters such as LOM, which limit the application of this method. Summary of the invention

[0004] In order to solve the technical problems existing in the prior art, the present invention provides a quantitative prediction method for organic carbon content in source rocks by well logging based on the KNN machine learning algorithm. The prediction method fully considers the advantages of the KNN machine learning algorithm and establishes an implicit relationship between the well logging curve and the TOC analysis data, that is, vertically continuous TOC data of source rocks is obtained through the well logging curve, and the calculation results are significantly more accurate than the three conventional methods.

[0005] In order to achieve the above technical effects, the present invention adopts the following technical solutions:

[0006] The present invention provides a method for quantitative prediction of organic carbon content in source rocks by well logging based on a KNN machine learning algorithm, characterized in that the prediction method comprises:

[0007] Perform depth correction on well logging curves;

[0008] For the logging curve, the depth of the logging curve and the core analysis data are matched by geological condition constraints to perform depth return of the TOC analysis data;

[0009] performing normalization processing on the well logging curve;

[0010] By numerically analyzing the TOC analysis data and the well logging curve at the same depth, the relationship between the single condition well logging curve and TOC is established;

[0011] The TOC analysis data after depth homing and the well logging curve after normalization are used as label data samples for machine learning, and a KNN algorithm is applied to establish a quantitative calculation model of the well logging curve and TOC analysis data;

[0012] Determine the parameter K of the KNN algorithm;

[0013] The quantitative calculation model of the logging curve and TOC analysis data and the parameter K are applied to perform TOC logging quantitative prediction to obtain vertically continuous TOC data.

[0014] In the present invention, a new method for quantitative prediction of organic carbon content in source rocks by logging based on a KNN (K Nearest Neighbor) machine learning algorithm fully utilizes KNN as a non-parametric model, has high result accuracy, does not make any assumptions about data, and the model structure is determined according to the data; the inert algorithm has high efficiency, low requirements for the amount of training data, and fast machine learning and training processes; and has three advantages of being insensitive to abnormal points based on the distance judgment result; in view of the difficulty that most conventional methods for quantitatively calculating TOC by logging have low accuracy, full consideration is given to the characteristics that TOC in source rocks is generally low and the logging response is weak, which makes it difficult to express the relationship between TOC and the logging curve using an explicit relationship formula; an implicit relationship expression between TOC analysis data with high accuracy but limited quantity and the logging curve is established through the KNN machine learning algorithm; under the constraints of geological conditions, vertically continuous and high-precision TOC data is obtained through the logging curve, providing more reliable data support for oil and gas exploration and development.

[0015] As a preferred technical solution of the present invention, the depth correction method includes manual correction and function correction.

[0016] Preferably, the manual correction method comprises: utilizing the high correlation of the GR curves of each logging zone to align the GR curves and determine the depth correction amount of each curve.

[0017] Preferably, the formula for the function correction is:

[0018]

[0019] In the formula, x(i) is the i-th logging data in the reference curve, y(i) is the i-th logging data in the comparison curve, and x a is the average value of the x curve in the comparison window, y a It is the average value of the y curve within the comparison time window.

[0020] As a preferred technical solution of the present invention, the basic formula of the relationship between the single condition logging curve and TOC is:

[0021] TOC = f(log) Formula 2

[0022] Wherein, log refers to a single logging curve, and f(log) refers to the relationship between a single logging curve and TOC.

[0023] As a preferred technical solution of the present invention, the method for normalizing the well logging curve includes a trend surface analysis method.

[0024] As a preferred technical solution of the present invention, the trend establishment formula in the trend surface analysis method is:

[0025]

[0026] Where Q is the residual, and its value is the minimum; i is a specific data point, i = 1, 2, 3...N, N is the total number of data points; z(x i ,y i ) is the characteristic value of the i-th point, f(x i ,y i ) is the trend surface.

[0027] As a preferred technical solution of the present invention, the formula for normalization processing in the trend surface analysis method is:

[0028] Δz=z(x i ,y i )-z(x i ,y i ) Formula 4

[0029] Where Δz is the correction amount.

[0030] As a preferred technical solution of the present invention, the method for determining the parameter K includes calculating the Euclidean distance between the prediction point and all points, as shown in Formula 5, sorting and storing, classifying the attributes of the first K values, and assigning the attribute with the largest number of points to the prediction point;

[0031]

[0032] Where d(x,y) is the Euclidean distance between two points in multidimensional space; (x 1 ,y 1 )、(x 2 ,y 2 )...(x n ,y n ) is the coordinate in n-dimensional space; n is the number of spatial dimensions, that is, the number of logging curves used.

[0033] As a preferred technical solution of the present invention, the data length and resolution of the TOC data are the same as those of the well logging curve.

[0034] As a preferred technical solution of the present invention, the prediction results of the TOC data are tested.

[0035] As a preferred technical solution of the present invention, the inspection method includes comparing the prediction results with the results of the multiple regression method and the ΔLgR method, and statistically analyzing the errors between the experimental analysis data and the calculated data at the same depth.

[0036] Compared with the prior art, the present invention has at least the following beneficial effects:

[0037] (1) The present invention addresses the problems of high cost of obtaining TOC based on core analysis data and low accuracy of obtaining TOC based on conventional methods of well logging curves in most cases. By establishing an implicit relationship expression between TOC analysis data and well logging curves, vertically continuous and high-precision TOC data are obtained under the constraints of geological conditions, thus realizing a new method for quantitative prediction of TOC in source rocks by well logging.

[0038] (2) The present invention proposes a new method for quantitative prediction of source rock TOC logging, which is independent of lithology and maturity, that is, it is applicable to all types of source rocks and all maturity states of source rocks; and it is not limited by region, and can be used for source rock evaluation in land and ocean areas;

[0039] (3) The quantitative prediction accuracy of TOC provided by the present invention is significantly improved compared with conventional methods, and the effect of field application is obvious;

[0040] (4) The present invention obtains TOC data based on logging data, which significantly improves work efficiency and can also achieve quantitative prediction of TOC in drilling without coring data. It is expected to have broad application prospects in source rock evaluation, exploration target evaluation and optimization, and block evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 A schematic diagram of a flow chart of a method for quantitatively predicting the organic carbon content of source rocks by well logging based on a KNN machine learning algorithm provided in Example 1 of the present invention;

[0042] Figure 2 It is a schematic diagram of the KNN machine learning algorithm in a specific implementation manner of the present invention;

[0043] Figure 3 This is a schematic diagram of the K parameter determination principle in Embodiment 1 of the present invention;

[0044] Figure 4 It is a quantitative prediction diagram of source rock TOC logging based on KNN machine learning algorithm in Example 1 of the present invention;

[0045] Figure 5 This is a quantitative effect evaluation diagram of source rock TOC logging based on the KNN machine learning algorithm in Example 1 of the present invention.

[0046] The present invention is further described in detail below. However, the following examples are only simplified examples of the present invention and do not represent or limit the scope of protection of the present invention. The scope of protection of the present invention shall be subject to the claims. DETAILED DESCRIPTION

[0047] The technical solution of the present invention is further described below with reference to the accompanying drawings and through specific implementation methods.

[0048] The embodiment of the present invention provides a method for quantitative prediction of organic carbon content in source rocks by well logging based on a KNN machine learning algorithm, the prediction method comprising:

[0049] Perform depth correction on well logging curves;

[0050] For the logging curve, the depth of the logging curve and the core analysis data are matched by geological condition constraints to perform depth return of the TOC analysis data;

[0051] performing normalization processing on the well logging curve;

[0052] By numerically analyzing the TOC analysis data and the well logging curve at the same depth, the relationship between the single condition well logging curve and TOC is established;

[0053] The TOC analysis data after depth homing and the well logging curve after normalization are used as label data samples for machine learning, and a KNN algorithm is applied to establish a quantitative calculation model of the well logging curve and TOC analysis data;

[0054] Determine the parameter K of the KNN algorithm;

[0055] The quantitative calculation model of the logging curve and TOC analysis data and the parameter K are applied to perform TOC logging quantitative prediction to obtain vertically continuous TOC data.

[0056] In a specific embodiment of the present invention, depth correction of the logging curve can eliminate the depth error of the logging curve caused by reasons such as wellbore conditions and velocity measurement as much as possible, which is an extremely important link in the preprocessing of logging data.

[0057] In a specific implementation of the present invention, since the coring depth is determined according to the cumulative length of the drill pipe, and the logging depth is determined according to the lowering depth of the cable, the two are determined by different methods, resulting in a certain error in the depth, so it is necessary to perform depth homing of the TOC analysis data.

[0058] In one specific embodiment of the present invention, the principle of applying the KNN machine learning algorithm is as follows Figure 2 As shown in the figure, the green box is the point to be found. According to the distance, if the distance is equal to 3 units (k=3 black dotted circles), there are 2 blue triangles and 1 red circle, then the point to be found is assigned to the blue triangle; if the distance is equal to 5 units (k=5 black dotted circles), there are 2 blue triangles and 3 red circles, then the point to be found is assigned to the red circle.

[0059] In a specific embodiment of the present invention, the relationship between the single-condition logging curve and TOC is established, which is a single factor analysis method; the single factor method has the advantages of being fast, intuitive, simple and efficient, and is often used for optimizing curves.

[0060] In a specific embodiment of the present invention, the well logging curve is normalized to eliminate the impact of the well logging curve dimension, numerical range and differences in different well logging series as much as possible, so that the well logging curve can reflect the geological information to the maximum extent. The main basis of the trend surface analysis method is that the geological parameters have certain changes in the horizontal direction, but have a certain regular gradual change.

[0061] In a specific embodiment of the present invention, a quantitative calculation model of logging curves and TOC analysis data is established, and the quantitative calculation model is tested. If an abnormal situation occurs, the model needs to be adjusted and optimized. The specific test method is a common test method of the KNN algorithm and is not specifically limited here.

[0062] In a specific embodiment of the present invention, in the process of TOC logging quantitative prediction, if the source rock layer and type are the same but there is no TOC data constraint from core analysis, the model can also be extended and applied to obtain TOC quantitative prediction data through logging curves.

[0063] In a specific embodiment of the present invention, the multiple regression method specifically uses the established multiple regression formula to perform quantitative prediction of TOC. The LgR method specifically determines the calculation baseline by superimposing the acoustic time difference and the resistivity curve; at the same time, key parameters are determined based on regional geological data to perform quantitative prediction of TOC.

[0064] In a specific embodiment of the present invention, the test of the prediction structure of the TOC data further includes comparing the TOC quantitative prediction results with the regional geological characteristics of the source rock, and evaluating the reliability and rationality of the results through geological condition constraints. The geological characteristics include distribution horizons, sedimentary environments, lithological characteristics, type characteristics, and distribution characteristics.

[0065] To better illustrate the present invention and facilitate understanding of the technical solution of the present invention, typical but non-limiting embodiments of the present invention are as follows:

[0066] Example 1

[0067] This embodiment provides a method for quantitative prediction of organic carbon content in source rocks by well logging based on the KNN machine learning algorithm. The process is as follows: Figure 1 As shown, the prediction method includes:

[0068] ① Well logging curve depth correction: Using the high correlation of the GR curves in each logging band, align the GR curves and determine the depth correction of each curve. If manual correction is difficult, use the commonly used correlation function method (see formula 1) to determine the depth correction of each curve, and try to eliminate the depth error of the well logging curve caused by the wellbore condition, velocity measurement and other reasons. Obtaining a well logging curve with a unified depth is the basis for conducting quantitative prediction of TOC;

[0069]

[0070] ② TOC analysis data depth restoration: through geological conditions, that is, the TOC of pure mudstone section is relatively high, and the TOC of sandstone section is relatively low, the coring depth is matched with the logging curve, and the real depth of the core is restored, providing a solid data foundation for TOC quantitative prediction;

[0071] ③ Logging curve optimization: Through the numerical intersection analysis of TOC analysis data and a logging curve at the same depth, the relationship between the single-condition logging curve and TOC was established (see formula 2), and five curves with high correlation were optimized, namely, the acoustic time difference curve, the neutron curve TNPH, the deep resistivity curve RT, the natural gamma curve GR and the radio and television index curve PEFZ;

[0072] TOC = f(log) Formula 2

[0073] ④ Normalization of logging curves: Eliminate the dimensions of logging curves as much as possible (e.g. density is usually g / cm 3 and kg / m 3 , the values ​​of the two differ by 1000 times), value range (such as the GR curve value range is mainly concentrated in 20-200, while the TNPH curve value range is mainly concentrated in 0.05-0.55) and the differences between different logging series, so that the logging curve can reflect the geological information to the maximum extent. The trend surface analysis method (see formula 3 and formula 4) is used. After normalization, the value range of all logging curves is between 0 and 1;

[0074]

[0075] Δz=z(x i ,y i )-z(x i ,y i ) Formula 4

[0076] ⑤ Establishment of calculation model: Using the depth-corrected TOC analysis data and the optimized logging curve (after depth correction, normalization and other preprocessing) as the labeled data samples for machine learning, the KNN machine learning algorithm is applied to establish the quantitative calculation model of the logging curve and TOC analysis data, and then test it;

[0077] ⑥ Determination of the key parameter K: Calculate the Euclidean distance between the predicted point and all points (Formula 5), ​​sort and store, classify the attributes of the first K value attributes, and assign the attribute with the most points to the predicted point. In actual operation, let K = 1, 3, 5, 10, 15, 20 and 30, scatter the predicted results and TOC analysis data, and determine the optimal K value; the results are as follows Figure 3 , let K = 1, 3, 5, 10, 15, 20 and 30, and intersect the prediction results with the TOC analysis data. If the distribution is concentrated and close to the 45-degree line, it means that the prediction is closer to the actual situation, that is, the prediction result is better. The prediction result of the red box in the figure is relatively the best;

[0078]

[0079] ⑦TOC logging quantitative prediction: Apply the calculation model established in step ⑤ and the key parameter K determined in step ⑥ to perform TOC logging quantitative prediction, obtain vertically continuous TOC data, and verify the data;

[0080] ⑧TOC logging quantitative prediction result inspection: According to the results of curve optimization, the conventional multiple regression method TOC = (0.778952-0.00492714*(DTCO)-1.06788*(TNPH)+0.0745007*(RHOZ)+0.041983*(PEFZ)+0.00263219*(GR)+5.69985e-005*(RT)) prediction formula is established to perform TOC quantitative prediction; the ΔLgR method is used to perform TOC quantitative prediction through the acoustic time difference and resistivity curve; the TOC quantitative prediction results of this embodiment are compared with the conventional multiple regression method and the ΔLgR method. Figure 4 As shown, the first track is the natural gamma curve (GR), the second track is the TOC curve predicted by the multivariate regression method, which has a high correlation with the logging curve; the third track is the TOC curve predicted by the ΔLgR method, with a large upper deviation; the fourth track is the TOC curve predicted by this embodiment; the gray rod-shaped points from the second to the fourth track are TOC data from core sample analysis, and the values ​​are the same, indicating that the prediction results of this embodiment are more consistent with the TOC analysis data, indicating that the prediction accuracy is higher. The error between the experimental analysis data and the calculated data at the same depth is statistically analyzed to verify the results. Figure 5 As shown, the left figure is the TOC predicted by the multivariate regression method, with an average absolute error of 28%; the middle figure is the TOC predicted by the ΔLgR method, with an average absolute error of 46%; the right figure is the TOC predicted by this embodiment, with an average absolute error of 15%, which also shows that the prediction error of this patent is smaller and the accuracy is higher.

[0081] After verifying the prediction results through TOC analysis data, it is also necessary to geologically constrain and verify the prediction results through regional source rock geological knowledge, including distribution layers, sedimentary environment, lithological characteristics, type characteristics and distribution characteristics, so as to further implement the reliability and rationality of the prediction results.

[0082] The applicant declares that the present invention illustrates the detailed structural features of the present invention through the above-mentioned embodiments, but the present invention is not limited to the above-mentioned detailed structural features, that is, it does not mean that the present invention must rely on the above-mentioned detailed structural features to be implemented. Those skilled in the art should understand that any improvement of the present invention, equivalent replacement of the components selected by the present invention, addition of auxiliary components, selection of specific methods, etc., all fall within the protection scope and disclosure scope of the present invention.

[0083] The preferred embodiments of the present invention are described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, a variety of simple modifications can be made to the technical solution of the present invention, and these simple modifications all belong to the protection scope of the present invention.

[0084] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present invention will not further describe various possible combinations.

[0085] In addition, various embodiments of the present invention may be arbitrarily combined, and as long as they do not violate the concept of the present invention, they should also be regarded as the contents disclosed by the present invention.

Claims

1. A method for quantitative prediction of organic carbon content in source rocks by well logging based on KNN machine learning algorithm, characterized in that: The prediction method comprises: Perform depth correction on well logging curves; For the logging curve, the depth of the logging curve and the core analysis data are matched by geological condition constraints to perform depth return of the TOC analysis data; performing normalization processing on the logging curve; By numerically analyzing the TOC analysis data and the well logging curve at the same depth, the relationship between the single condition well logging curve and TOC is established; The TOC analysis data after depth homing and the well logging curve after normalization are used as label data samples for machine learning, and a KNN algorithm is applied to establish a quantitative calculation model of the well logging curve and TOC analysis data; Determine the parameter K of the KNN algorithm; The quantitative calculation model of the logging curve and TOC analysis data and the parameter K are applied to perform TOC logging quantitative prediction to obtain vertically continuous TOC data.

2. The prediction method according to claim 1, characterized in that: The depth correction method includes manual correction and function correction; Preferably, the manual correction method comprises: utilizing the high correlation of the GR curves of each logging zone to align the GR curves and determine the depth correction amount of each curve; Preferably, the formula for the function correction is: In the formula, x(i) is the i-th logging data in the reference curve, y(i) is the i-th logging data in the comparison curve, and x a is the average value of the x curve in the comparison window, y a It is the average value of the y curve within the comparison time window.

3. The prediction method according to claim 1, characterized in that: The basic formula for the relationship between the single condition logging curve and TOC is: TOC = f(log) Formula 2 Wherein, log refers to a single logging curve, and f(log) refers to the relationship between a single logging curve and TOC.

4. The prediction method according to claim 1, characterized in that: The method for normalizing the logging curve includes a trend surface analysis method.

5. The prediction method according to claim 4, characterized in that: The trend establishment formula in the trend surface analysis method is: Where Q is the residual, and its value is the minimum; i is a specific data point, i = 1, 2, 3...N, N is the total number of data points; z(x i ,y i ) is the characteristic value of the i-th point, f(x i ,y i ) is the trend surface.

6. The prediction method according to claim 4, characterized in that: The formula for normalization in the trend surface analysis method is: Δz = z(x i , y i ) - z(x i , y i ) Equation 4 Where Δz is the correction amount.

7. The prediction method according to claim 1, characterized in that: The method for determining the parameter K includes calculating the Euclidean distance between the prediction point and all points, as shown in Formula 5, sorting and storing, classifying the attributes of the first K values, and assigning the attribute with the most points to the prediction point; Where d(x,y) is the Euclidean distance between two points in multidimensional space; (x1,y1), (x2,y2)...(x n ,y n ) is the coordinate in n-dimensional space; n is the number of spatial dimensions, that is, the number of logging curves used.

8. The prediction method according to claim 1, characterized in that: The data length and resolution of the TOC data are the same as those of the well logging curve.

9. The prediction method according to claim 1, characterized in that: The prediction results of the TOC data were tested.

10. The prediction method according to claim 9, characterized in that: The test method includes comparing the prediction results with the results of the multiple regression method and the ΔLgR method, and statistically analyzing the errors between the experimental analysis data and the calculated data at the same depth.

Images