Element sensitive analysis method for shale sublayer intelligent identification by element logging

Abstract

This invention discloses a method for intelligent identification of elemental sensitivity analysis in shale sub-layers using elemental logging, comprising the following steps: Step S1, mining the identification features of shale sub-layers with multiple elemental single-double composite intersections; Step S2, screening homogeneous elements based on Siamese neural networks; Step S3, heterogeneous element sensitivity analysis based on the generalization property of residual neural networks. Using the method designed in this invention, the systematic, objective, and comprehensive extraction of sensitive elements from shale sub-layers can be achieved, automating the analysis of sensitive elements in shale sub-layers.

Description

Technical Field

[0001] This invention belongs to the field of shale oil and gas horizontal well drilling technology, and in particular relates to an element-sensitive analysis method for intelligently identifying shale sub-layers using element logging. Background Technology

[0002] Shale oil and gas, as an important unconventional oil and gas resource, has relatively thin high-quality reservoirs. Therefore, during horizontal well drilling in shale oil and gas, it is necessary to determine the relative position of the well bottom to the high-quality reservoir in real time in order to adjust the well trajectory in a timely manner and improve the encounter rate of high-quality shale oil and gas reservoirs. Currently, the data basis for determining the relative position of the well bottom to the high-quality reservoir is mainly based on data such as elemental logging. On-site engineers have established elemental characteristic models of high-quality reservoirs by using elemental logging data of high-quality reservoirs from adjacent completed horizontal and vertical wells, which serve as the identification basis for the current horizontal well drilling.

[0003] The key to accurately determining the relative position of the well bottom and high-quality reservoir lies in establishing an accurate, objective, and reliable elemental characteristic model of the high-quality reservoir. The first step in establishing this model is to analyze the sensitivity of each element to the high-quality reservoir, selecting the more sensitive element logging curves to reduce interference from insensitive elements in subsequent analyses. During horizontal well drilling, the number of element curves obtained through logging is typically as high as thirty or more. Manual analysis is insufficient to simultaneously capture the variation characteristics of all element curves. Furthermore, due to measurement errors and non-standard operations by on-site workers, element content curves often contain noise. Therefore, analyzing and selecting the sensitivity of each element to changes in shale sublayers is necessary and essential. Retaining sensitive element curves can eliminate the interference of noise from insensitive curves, reduce the amount of data for manual analysis, and thus improve the reliability of the results of manual analysis of the relative position of the well bottom and high-quality reservoir.

[0004] Currently, the main method for determining the sensitivity of element curves to high-quality reservoirs is for field engineers to rely on past horizontal well drilling experience, combined with the relevant characteristics of the corresponding appraisal wells (vertical wells) in high-quality reservoirs. Typically, a few curves are also compared and intersected based on experience to generate new characteristic curves. However, generating intersecting curves based on experience for a few elements cannot fully uncover the elemental intersection characteristics of the well. Furthermore, due to the homogeneity of some elements, manual analysis is prone to many meaningless calculations. This analysis process lacks systematicity, leading to unnecessary calculations and incomplete results. In addition to this lack of systematicity, the automation level of the entire analysis process is low; most data calculations, processing, and visualization still require manual intervention, resulting in low overall efficiency.

[0005] In summary, current sensitivity analysis of elemental identification in shale sublayers suffers from problems such as a lack of systematic analysis process, low degree of automation, and subjective and biased results. Summary of the Invention

[0006] To address the shortcomings of the aforementioned background technology, this invention provides an element-sensitive analysis method for intelligently identifying shale sub-layers using element logging.

[0007] The objective of this invention is achieved through the following technical solution:

[0008] An element-sensitive analysis method for intelligently identifying shale sub-layers using elemental logging includes the following steps:

[0009] Step S1: Mining the identification features of shale sublayers with multiple elemental single and double composite intersections;

[0010] Step S2: Homogeneous element screening based on Siamese neural network;

[0011] Step S3: Heterogeneous element sensitivity analysis based on the generalization property of residual neural networks.

[0012] Specifically, step S1 further includes the step of constructing a basic dataset of shale sublayer elements, which includes the following sub-steps:

[0013] Step S11: All single-element curves obtained through element logging are intersected pairwise to generate a bi-element curve;

[0014] Step S12: Intersect the single-element curve and the two-element curve to generate a three-element curve;

[0015] Step S13: Generate a multi-element curve obtained by any number of intersections;

[0016] Step S14: Combine single-element curves, bi-element curves, and multi-element curves to generate a basic dataset of shale sublayer elements.

[0017] Specifically, step S2 optimizes the basic dataset of shale sublayer elements into a heterogeneous element dataset, including the following sub-steps:

[0018] Step S21: Construct homogeneous element samples with the same shape and information;

[0019] Step S22, Design and optimization of the twin neural network structure;

[0020] Step S23: Calculation of homogeneity parameters and screening of heterogeneous elements in the twin neural network.

[0021] Specifically, step S21 includes the following sub-steps:

[0022] Perform a normalization operation on the curves, and retain only one curve with a similar curve shape.

[0023] When identifying shale sublayers, only one curve with the same information is retained.

[0024] Specifically, the twin neural network structure includes a multilayer perceptron or a convolutional neural network;

[0025] Preferably, the twin neural network structure is a multilayer perceptron structure with 4 layers and 1.5 times the number of nodes.

[0026] Specifically, step S23 calculates the homogeneity parameter of all curves in the shale sublayer elemental basic dataset, filters out all homogeneous single-element curves, and the remaining heterogeneous single-element curves intersect to generate a set of bi-element curves.

[0027] Step S23 also includes the step of continuing to generate multiple element curve sets.

[0028] Specifically, step S3 involves designing and optimizing the residual multilayer perceptron structure based on the heterogeneous element dataset, training and predicting shale sublayers with heterogeneous elements in segments, and using the differences in generalization performance of the residual multilayer perceptron for different heterogeneous element curves during prediction as a sensitivity evaluation index for heterogeneous elements. The heterogeneous element dataset is then further optimized into a sensitive element dataset, ultimately forming a sensitive element identification analysis method for shale sublayers, including the following sub-steps:

[0029] Step S31, Residual multilayer perceptron structure design and optimization;

[0030] Step S32: Comparison of generalization performance of residual multilayer perceptron for heterogeneous element curves and screening of sensitive elements;

[0031] Step S33: The heterogeneous element sensitivity analysis method is formed;

[0032] Step S31 involves introducing a residual module and establishing a residual multilayer perceptron model as a model for evaluating sensitivity.

[0033] Specifically, in step S32, at the same depth, the heterogeneous element dataset is divided into a training segment and a validation segment. After learning the features of the training segment using a residual multilayer perceptron, the generalization performance of the validation segment is used as an evaluation index of element sensitivity. Insensitive elements in the heterogeneous element dataset are removed, and the heterogeneous element dataset is further optimized into a sensitive element dataset.

[0034] Specifically, step S33 integrates the generation of multi-element single-double composite intersection curves, the evaluation of element homogeneity using twin neural networks, and the evaluation of element sensitivity using residual multilayer perceptron generalization, to form a heterogeneous element sensitivity analysis method for identifying shale sublayers at the bottom of the well.

[0035] The beneficial effects of this invention are:

[0036] The method designed in this invention can achieve systematic, objective, and comprehensive extraction of sensitive elements in shale sublayers, and automate the analysis of sensitive elements in shale sublayers. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0038] Figure 1 This is a technical flowchart of the present invention;

[0039] Figure 2 This is a schematic diagram of a single-element curve obtained by elemental logging in a horizontal well according to an example of the present invention;

[0040] Figure 3 This is a schematic diagram of the bi-element intersection curve of a horizontal well in this invention (10 curves are randomly selected from all bi-element curves);

[0041] Figure 4 This is a schematic diagram of the three-element intersection curves of a horizontal well in this invention (10 curves are randomly selected from all three-element curves);

[0042] Figure 5 This is a comparison diagram of the normalized speciation of iron (Fe) and cobalt (Co) elements in the horizontal well of this invention.

[0043] Figure 6 This is a comparison diagram of the morphology of silicon (Si) and phosphorus (P) elements in different sublayers in a horizontal well, as described in this invention.

[0044] Figure 7 This is a flowchart of the curve homogeneity parameter calculation based on the Siamese neural network of the present invention;

[0045] Figure 8 This is a graph showing the screening results of heterogeneous single-element curves in the horizontal well section of this invention.

[0046] Figure 9 This is a graph showing the results of heterogeneous dual-element curve screening for a horizontal well section in this invention.

[0047] Figure 10 This is a flowchart of the sensitivity evaluation based on the generalization property of residual multilayer perceptron of the present invention;

[0048] Figure 11This is a diagram showing the results of shale sub-layer division in a horizontal well, as described in this invention.

[0049] Figure 12 This is a schematic diagram of the heterogeneous element sensitivity analysis method of the present invention. Detailed Implementation

[0050] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0051] To provide a clearer understanding of the technical features, objectives, and beneficial effects of this invention, the following detailed description of the technical solution is provided. Obviously, the described embodiments are only a portion of the embodiments of this invention, not all of them, and should not be construed as limiting the scope of implementation of this invention. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without inventive effort are within the protection scope of this invention.

[0052] The solution adopted in this invention is:

[0053] like Figure 1 As shown, an element-sensitive analysis method for intelligently identifying shale sub-layers using elemental logging includes the following steps:

[0054] Step S1: Mining the identification features of shale sublayers with multiple elemental single and double composite intersections;

[0055] Step S2: Homogeneous element screening based on Siamese neural network;

[0056] Step S3: Heterogeneous element sensitivity analysis based on the generalization property of residual neural networks.

[0057] Specifically, step S1 further includes the step of constructing a basic dataset of shale sublayer elements, which includes the following sub-steps:

[0058] Step S11: All single-element curves obtained through element logging are intersected pairwise to generate a bi-element curve;

[0059] Step S12: Intersect the single-element curve and the two-element curve to generate a three-element curve;

[0060] Step S13: Generate a multi-element curve obtained by any number of intersections;

[0061] Step S14: Combine single-element curves, bi-element curves, and multi-element curves to generate a basic dataset of shale sublayer elements.

[0062] Specifically, step S2 optimizes the basic dataset of shale sublayer elements into a heterogeneous element dataset, including the following sub-steps:

[0063] Step S21: Construct homogeneous element samples with the same shape and information;

[0064] Step S22, Design and optimization of the twin neural network structure;

[0065] Step S23: Calculation of homogeneity parameters and screening of heterogeneous elements in the twin neural network.

[0066] Specifically, step S21 includes the following sub-steps:

[0067] Perform a normalization operation on the curves, and retain only one curve with a similar curve shape.

[0068] When identifying shale sublayers, only one curve with the same information is retained.

[0069] Specifically, the twin neural network structure includes a multilayer perceptron or a convolutional neural network:

[0070] The multilayer sensor has 4 or 6 layers and the number of nodes is 1, 1.5, or 2 times.

[0071] The convolutional neural network has 2, 4, or 6 layers, and the kernel size is 1×3, 1×5, or 1×7.

[0072] Preferably, the twin neural network structure is a multilayer perceptron structure with 4 layers and 1.5 times the number of nodes.

[0073] Specifically, step S23 calculates the homogeneity parameter of all curves in the shale sublayer elemental basic dataset, filters out all homogeneous single-element curves, and the remaining heterogeneous single-element curves intersect to generate a set of bi-element curves.

[0074] The heterogeneous single-element curves include curves for magnesium, aluminum, silicon, calcium, titanium, uranium, iron, thorium, sodium, and barium.

[0075] The heterogeneous bi-element curves include curves for magnesium to calcium, magnesium to uranium, aluminum to silicon, aluminum to uranium, and silicon to thorium.

[0076] Step S23 also includes the step of continuing to generate multiple element curve sets.

[0077] Specifically, step S3 involves designing and optimizing the residual multilayer perceptron structure based on the heterogeneous element dataset, training and predicting shale sublayers with heterogeneous elements in segments, and using the differences in generalization performance of the residual multilayer perceptron for different heterogeneous element curves during prediction as a sensitivity evaluation index for heterogeneous elements. The heterogeneous element dataset is then further optimized into a sensitive element dataset, ultimately forming a sensitive element identification analysis method for shale sublayers, including the following sub-steps:

[0078] Step S31, Residual multilayer perceptron structure design and optimization;

[0079] Step S32: Comparison of generalization performance of residual multilayer perceptron for heterogeneous element curves and screening of sensitive elements;

[0080] Step S33: The heterogeneous element sensitivity analysis method is formed;

[0081] Step S31 involves introducing a residual module and establishing a residual multilayer perceptron model as a model for evaluating sensitivity.

[0082] Specifically, in step S32, at the same depth, the heterogeneous element dataset is divided into a training segment and a validation segment. After learning the features of the training segment using a residual multilayer perceptron, the generalization performance of the validation segment is used as an evaluation index of element sensitivity. Insensitive elements in the heterogeneous element dataset are removed, and the heterogeneous element dataset is further optimized into a sensitive element dataset.

[0083] Specifically, step S33 integrates the generation of multi-element single-double composite intersection curves, the evaluation of element homogeneity using twin neural networks, and the evaluation of element sensitivity using residual multilayer perceptron generalization, to form a heterogeneous element sensitivity analysis method for identifying shale sublayers at the bottom of the well.

[0084] Example 1:

[0085] 1. Mining the identification characteristics of shale sublayers with multiple elements and single / double / composite features

[0086] By using the ratio and difference of two curves, a bi-element intersection curve is generated by the pairwise intersection of the original single-element curves. Furthermore, by using the original single-element curves and the bi-element intersection curves, a multi-element curve set is generated through composite intersection. This fully explores the elemental characteristics of shale sublayers and integrates single, bi, and multi-element curves to form a relatively complete basic dataset of shale sublayer elements.

[0087] The case study is the Luzhou shale gas block in southern Sichuan. During the drilling process of a horizontal well in this block, a total of 31 single-element curves were obtained through elemental logging. Figure 2 The elements in the yttrium ore include sodium (Na), magnesium (Mg), aluminum (Al), silicon (Si), and phosphorus (P), while the trace elements include yttrium (Y), uranium (U), gallium (Ga), and rubidium (Rb).

[0088] Engineers typically use geological engineering experience to generate new bi-element curves by comparing or differing two specified element curves. Common bi-element curves in engineering include "uranium to thorium (U / Th)" and "silicon to calcium (Si / Ca)". Inspired by this, bi-element curves are generated by intersecting all single-element curves pairwise (mainly by subtraction or quotient). Figure 3 Furthermore, the intersection of single-element curves and two-element curves can generate a three-element curve. Figure 4This process can be repeated to generate any number of intersecting multi-element curves. By combining single, double, and multi-element curves, a relatively complete shale sublayer element dataset can be generated. This dataset can fully explore the elemental characteristics used to identify shale sublayers. Typically, when single-element curves are combined to form three-element curves, the resulting element dataset can fully explore the distinguishing features of different shale sublayers.

[0089] 2. Homogeneous element screening based on Siamese neural network

[0090] By analyzing the basic element dataset, element curves with similar morphologies that provide the same information for identifying shale sublayers are selected to form a homogeneous element sample set. Based on this sample set, the structure of a Siamese neural network is designed and optimized to form an optimal Siamese neural network for homogeneity evaluation. This Siamese network is used to calculate the homogeneity parameters of any two curves in the basic dataset, filtering out homogeneous element curves and optimizing the basic dataset into a heterogeneous element dataset.

[0091] The initial design of the Siamese neural network was to compare whether two images depict the same object or person. For example, inputting two images into the network determines if the person in the images is the same person, or inputting two scanned images of handwritten text determines if the handwriting is from the same person. Thus, the characteristic of this network model is its ability to compare a certain "consistency" between two inputs. Therefore, this invention utilizes this characteristic of the Siamese neural network to compare the homogeneity of two curves. A homogeneous curve refers to a curve that provides the same useful information when identifying shale sublayers; that is, only one homogeneous curve needs to be retained, while the others do not provide additional information for identifying shale sublayers.

[0092] To train a Siamese neural network, the first step is to construct a set of homogeneous element samples for training. This set can be constructed in two ways: firstly, using curves with similar or identical shapes; and secondly, using a set of identical information provided when identifying shale sublayers. Secondly, the specific structure of the Siamese neural network needs to be considered. Unlike convolutional neural networks or recurrent neural networks, Siamese neural networks only share the weight values ​​of two network models without specifying a concrete network structure. Therefore, the network structure can be designed and optimized from the perspectives of module type, number of layers, and number of nodes. Finally, the optimized Siamese neural network is trained using the constructed set of homogeneous element curves. This algorithm is then used to calculate the homogeneity parameter of all curve pairs in the basic element dataset obtained from the intersection of multiple single and double elements, thereby eliminating homogeneous curves.

[0093] ① Construction of homogeneous element samples with the same shape and information

[0094] By comparing the shapes of element curves in the baseline dataset pairwise, isomorphic element curves with the same or similar shapes are selected. The feature differences of element curves in the baseline dataset across each shale sublayer are analyzed to extract the distinguishing information of curves in different shale sublayers, and isomorphic curves with the same information are selected. By combining isomorphic and isomorphic curves, a homogeneous element sample set is constructed, laying the data foundation for the subsequent design, optimization, and computation of the Siamese neural network.

[0095] The basic dataset contains tens of thousands of curves, some of which have similar morphological variations. Obviously, these curves provide the same information for identifying shale sublayers. To compare the morphology of the curves, a normalization operation can be performed on each curve. The normalization operation can map the numerical range of the curve to the interval [0, 1] without changing the curve's morphology, making it easier to compare two element curves on the same graph. Figure 5 The normalized forms of iron (Fe) and cobalt (Co) are shown. Obviously, the two curves are similar in shape and provide the same information for identifying shale sublayers. Therefore, either one can be used.

[0096] In addition to the aforementioned isomorphic curves, there is another type of curve that provides the same information when identifying shale sublayers. Figure 6 The diagram shows a comparison of the morphological characteristics of silicon and phosphorus in the three sublayers 2, 1, and 0. As shown in the figure, the well's horizontal section, approximately 1600m long, is mostly located in sublayers 2 and 1. Near a depth of approximately 5400m, the well trajectory encounters sublayer 0. These two curves exhibit distinct distinguishing features in sublayers 2 and 1, and respectively show local minimum and maximum values ​​in sublayer 0. Although the shapes of these two curves are not similar, they both effectively identify sublayer 0. Therefore, they provide the same information for identifying shale sublayers; removing one element curve does not affect the overall shale sublayer identification process. This invention refers to this type of curve as the homogeneous information curve for shale sublayer identification, which, together with the homomorphic curve, constitutes the homogeneous curve for shale sublayer identification.

[0097] Since the basic dataset contains a large number of curve elements (several thousand), only a small number of homogeneous curve pairs need to be selected (usually a few dozen pairs). To train the subsequently designed Siamese neural network, an equal number of non-homogeneous curve pairs also need to be selected.

[0098] ② Design and optimization of twin neural network structure

[0099] Based on the structure and hyperparameters of the Siamese neural network, the direction of structural adjustment and the range of hyperparameter adjustment are defined, a design scheme for the Siamese network structure and hyperparameters is formulated, the performance of each scheme is evaluated using a homogeneous element sample set, and a set of structure and hyperparameter design schemes with better performance of Siamese neural network is obtained through optimization.

[0100] A Siamese neural network (also known as a twin neural network) transforms two inputs into the same feature space using two neural networks, facilitating the comparison of the similarity between the two inputs. If the two neural networks share parameters, it is called a Siamese neural network; otherwise, it is called a pseudo-Siamese neural network. In the task of homogeneous curve screening, both inputs are curves with the same length; therefore, a Siamese neural network with shared parameters is used as the basic algorithm framework.

[0101] Current deep neural networks are mainly composed of basic computing units such as perceptron layers, recurrent layers, and convolutional layers. Among them, recurrent layers mainly handle tasks with sequential features between learning samples, which are obviously not suitable for homogeneous curve screening tasks. Therefore, for homogeneous curve screening tasks, multilayer perceptrons or convolutional neural networks are mainly considered as the core framework.

[0102] For multilayer perceptrons, the algorithm model mainly considers three schemes: 4 and 6 layers, and the number of nodes per layer is mainly considered to be 1, 1.5, and 2 times the number of input nodes. For convolutional neural networks, the number of convolutional layers is mainly considered to be three schemes: 2, 4, and 6, and the size of the convolutional kernel is mainly considered to be three schemes: 1×3, 1×5, and 1×7.

[0103] Table 1. Performance Comparison of Horizontal Well Twin Neural Network Structure Design Schemes in Case Studies

[0104]

[0105] As shown in Table 1, a total of 15 schemes were designed. Each scheme was trained 5 times, with 50 iterations per cycle. The average fitting rate of the sample set was taken as the evaluation index. The final results are shown in Table 1. Obviously, the difference between the multilayer perceptron and the convolutional neural network is small. There are 4 schemes with relatively high performance: 2, 8, 12, and 15. Scheme 2 has a relatively simple structure, so the second structure design scheme was finally adopted.

[0106] ③ Homogeneity parameter calculation and heterogeneous element screening of twin neural networks

[0107] Using the designed and optimized Siamese neural network, the homogeneity parameter of the element curves in the basic dataset is calculated pairwise. Based on the parameter value, the homogeneous element curves are filtered out, and the heterogeneous element curves are retained, thus optimizing the basic dataset into a heterogeneous element dataset.

[0108] The designed Siamese neural network is trained using the homogeneous sample set constructed in step ①, and then the homogeneity parameters of all curves in the basic dataset are calculated pairwise. Figure 7To reduce computational load, a Siamese neural network can be used to first filter out all homogeneous single-element curves. Heterogeneous single-element curves are then intersected pairwise to generate a set of bi-element curves. This process is repeated to filter out heterogeneous bi-element curves, further generating ternary curves, and then performing homogeneity screening on these ternary curves. In practical applications, a curve intersection count of 2 is usually sufficient for most application scenarios.

[0109] All single-element curves are input into a Siamese neural network in pairs to obtain homogeneity evaluation parameters among them. After screening, approximately 20 heterogeneous single-element curves are obtained, including those for magnesium, aluminum, silicon, calcium, titanium, uranium, iron, thorium, sodium, and barium (some of which are listed below). Figure 8 The heterogeneous bi-element curves include approximately 70 curves, such as magnesium to calcium, magnesium to uranium, aluminum to silicon, aluminum to uranium, and silicon to thorium (some of which are as follows). Figure 9 (As shown).

[0110] 3. Heterogeneous element sensitivity analysis based on the generalization property of residual neural networks

[0111] Based on a heterogeneous element dataset, a residual multilayer perceptron structure was designed and optimized. The optimized residual multilayer perceptron was then used to train and predict shale sublayers with heterogeneous elements in segments. The difference in generalization performance of the residual multilayer perceptron to different heterogeneous element curves during prediction was used as a sensitivity evaluation index for heterogeneous elements. The heterogeneous element dataset was further optimized into a sensitive element dataset, and finally a sensitive analysis method for element identification of shale sublayers was formed.

[0112] The most direct way to evaluate element sensitivity is to examine how easily the element can identify shale sub-layers; that is, if the element can easily identify each shale sub-layer, then the element is sensitive to sub-layer changes, and vice versa. Obviously, this method has a certain degree of subjectivity and the evaluation index is difficult to quantify. Therefore, a multilayer perceptron is used as the algorithm for identifying shale sub-layers. Based on the input element curve value, the algorithm predicts the sub-layer to which the element belongs, and the prediction effect of different curves (i.e., generalization performance) is used as the basis for evaluating the element curve sensitivity.

[0113] To avoid the algorithm's structure affecting generalization and thus failing to effectively distinguish element curves with similar sensitivities, different multilayer perceptron (MLP) schemes were first designed. The performance of these schemes was compared using a few element curves and layering results to determine the optimal MLP design. A residual MLP was then established by introducing a residual MLP module. Based on the residual MLP model, a segment of each element curve was used for training, and the remaining segments of shale sublayers were predicted. The prediction accuracy was used to evaluate the sensitivity of each curve to the shale sublayers.

[0114] ① Residual Multilayer Perceptron Structural Design and Optimization

[0115] Design adjustment schemes for the number of layers, the number of nodes per layer, and the connection method of residual modules in a residual multilayer perceptron. Select a small number of heterogeneous element curves and compare the fitting degree and generalization of the residual multilayer perceptrons under different schemes to these heterogeneous element curves. Optimize to obtain the residual multilayer perceptron with the best performance.

[0116] The multilayer perceptron (MLP) algorithm model is simple in structure and fast in computation, and has certain applicability to various types of tasks. Therefore, it is the best deep learning algorithm model for comparing the sensitivity of various element curves to shale sublayer identification. The main structural factors affecting the performance of the MLP model are the number of perceptron layers and the number of perceptrons per layer. The design schemes for these two factors still follow the design scheme of the second technical step of the second technique, that is, the number of layers is mainly considered to be 4 or 6, and the number of perceptrons per layer is mainly considered to be 1, 1.5, and 2 times the input. In the previous technique, the MLP was used to perform feature transformation on the element curve, while the MLP in this technique identifies the shale sublayer to which each depth position on the curve belongs based on the element curve. Therefore, the performance comparison results of the previous technique are not applicable to this technique, and recalculation is required using a few curves and their shale sublayer division results. The final calculation results show that the performance of the MLPs represented by these 6 schemes is similar, with no significant difference. Therefore, the simplest MLP design scheme is selected, and a residual MLP model can be established after introducing a residual module as the algorithm model for evaluating sensitivity.

[0117] ② Comparison of generalization performance of residual multilayer perceptrons for heterogeneous element curves and screening of sensitive elements

[0118] At the same depth, the heterogeneous element dataset is divided into training and validation segments. After learning the features of the training segment using a residual multilayer perceptron, the generalization performance on the validation segment is used as an evaluation index of element sensitivity. Figure 10 The heterogeneous element dataset is further optimized into a sensitive element dataset by removing insensitive elements from the heterogeneous element dataset.

[0119] The shale sub-layer division results of the case horizontal well are as follows: Figure 11 As shown, the shale sublayer division results and elemental data before a well depth of 4750 m were used as the training segment, and the data after a well depth of 4750 m were used as the validation segment to verify the generalization performance of the algorithm. The specific calculation process is as follows: the training segment is input into a residual multilayer perceptron for training; the trained model is used to predict the shale sublayer division results of the validation segment, and the results are compared with the actual division results. The accuracy of this comparison is used as the basis for evaluating the sensitivity of the curve. By calculating all heterogeneous single-element and dual-element curves, the top five most sensitive single-element curves are: magnesium, calcium, aluminum, silicon, and uranium; the dual-element curves are aluminum to silicon, magnesium to calcium, aluminum to iron, and thorium to uranium.

[0120] ③ Formation of Heterogeneous Element Sensitivity Analysis Method

[0121] By integrating three major technical aspects—generation of multi-element single-double composite intersection curves, evaluation of element homogeneity using twin neural networks, and evaluation of element sensitivity using residual multilayer perceptrons—a method for heterogeneous element sensitivity analysis is formed.

[0122] The final analytical methodology architecture is as follows Figure 12 As shown, this analysis method can be used to screen out single and double multi-element curves that are more sensitive to changes in shale sublayers based on the element curve data of completed horizontal wells and the shale sublayer division results in the block, thereby assisting in the identification of bottom shale sublayers during the drilling of new horizontal wells.

[0123] This invention designs a three-stage sensitivity analysis and identification method for shale sub-layers using elemental logging: "feature mining – homogeneous element screening – heterogeneous element sensitivity analysis". Specifically, it includes: 1) feature mining of shale sub-layers with multi-element single / double / composite intersections; 2) homogeneous element screening based on Siamese neural networks; and 3) heterogeneous element sensitivity analysis based on the generalization property of residual neural networks. Using the identification method designed in this invention, the systematic, objective, and comprehensive extraction of sensitive elements from shale sub-layers can be achieved, automating the analysis of sensitive elements in shale sub-layers.

[0124] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

[0125] It should be noted that, for the sake of simplicity, the aforementioned method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, because according to this application, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and units involved are not necessarily essential to this application.

[0126] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0127] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, ROM, RAM, etc.

[0128] The above description discloses only preferred embodiments of the present invention and should not be construed as limiting the scope of the present invention. Therefore, equivalent variations made in accordance with the claims of the present invention are still within the scope of the present invention.

Claims

1. A method for intelligently identifying shale sub-layers using elemental logging, characterized in that, Includes the following steps: Step S1, the feature mining of multi-element single-double-compound intersection shale sublayer identification, also includes constructing a basic dataset of shale sublayer elements, including: Step S11: All single-element curves obtained through element logging are intersected pairwise to generate a bi-element curve; Step S12: Intersect the single-element curve and the two-element curve to generate a three-element curve; Step S13: Generate a multi-element curve obtained by any number of intersections; Step S14: Combine single-element curves, bi-element curves, and multi-element curves to generate a basic dataset of shale sublayer elements; Step S2, based on the Siamese neural network, optimizes the basic dataset of shale sublayer elements into a heterogeneous dataset, including the following sub-steps: Step S21, constructing homogeneous element samples with the same shape and information, includes the following sub-steps: Perform a normalization operation on the curves, and retain only one curve with a similar curve shape. When identifying shale sub-layers, only one curve with the same information is retained; Step S22, design and optimization of the twin neural network structure; Step S23: Calculation of homogeneity parameters and screening of heterogeneous elements in the twin neural network. Calculate the homogeneity parameters of all curves in the basic dataset of shale sublayer elements, filter out all homogeneous single-element curves, and merge the remaining heterogeneous single-element curves to generate a set of bi-element curves; Step S23 also includes the step of generating multiple sets of element curves. Step S3: Heterogeneous element sensitivity analysis based on the generalization property of residual neural networks.

2. The element-sensitive analysis method for intelligent identification of shale sub-layers using elemental logging as described in claim 1, characterized in that, The twin neural network structure includes a multilayer perceptron or a convolutional neural network.

3. The element-sensitive analysis method for intelligently identifying shale sub-layers using elemental logging as described in claim 1, characterized in that, Step S3 involves designing and optimizing a residual multilayer perceptron structure based on a heterogeneous element dataset, training and predicting shale sublayers containing heterogeneous elements in segments, and using the differences in generalization performance of the residual multilayer perceptron for different heterogeneous element curves during prediction as a sensitivity evaluation index for heterogeneous elements. The heterogeneous element dataset is then further optimized into a sensitive element dataset, ultimately forming a shale sublayer element identification sensitivity analysis method, which includes the following sub-steps: Step S31, residual multilayer perceptron structure design and optimization; Step S32: Comparison of generalization performance of residual multilayer perceptron for heterogeneous element curves and screening of sensitive elements; Step S33: The heterogeneous element sensitivity analysis method is formed; Step S31 involves introducing a residual module and establishing a residual multilayer perceptron model as a model for evaluating sensitivity.

4. The element-sensitive analysis method for intelligent identification of shale sub-layers using elemental logging as described in claim 3, characterized in that, In step S32, at the same depth, the heterogeneous element dataset is divided into a training segment and a validation segment. After learning the features of the training segment using a residual multilayer perceptron, the generalization performance of the validation segment is used as an evaluation index of element sensitivity. Insensitive elements in the heterogeneous element dataset are removed, and the heterogeneous element dataset is further optimized into a sensitive element dataset.

5. The element-sensitive analysis method for intelligent identification of shale sub-layers using elemental logging as described in claim 4, characterized in that, Step S33 integrates the generation of multi-element single-double composite intersection curves, the evaluation of element homogeneity using twin neural networks, and the evaluation of element sensitivity using residual multilayer perceptron generalization, to form a heterogeneous element sensitivity analysis method for identifying shale sub-layers at the bottom of the well.

Images