Data-driven sand shale rock lithology prediction and uncertainty estimation method

Abstract

The invention provides a data-driven sand shale rock lithology prediction and uncertainty estimation method, and the method comprises the steps: 1, carrying out the rock physical analysis, and selecting a proper rock physical model; 2, a fusion method of data generated by forward modeling of the rock physical model and actual data under the physical property parameter disturbance condition is researched, and an elastic parameter data set is generated to serve as an expanded simulation training sample; 3, establishing a statistical rock physical mapping relation between reservoir sand shale and elastic parameters; and step 4, the improved deep neural network no longer outputs the lithology result but outputs the lithology probability. According to the data-driven sand shale mass lithology prediction and uncertainty estimation method, a more accurate sand shale prediction model is established by using a deep learning technology, the probability of a sand shale prediction result is given, and finally a fine statistical physical model for sand shale stratum lithology body prediction and uncertainty estimation is obtained.

Description

Technical Field

[0001] The invention relates to the technical field of geophysical exploration seismic interpretation and comprehensive research, and in particular to a data-driven sandstone-mudstone body lithology prediction and uncertainty estimation method. Background Art

[0002] Establishing the relationship between sandstone physical properties and elastic parameters and applying quantitative interpretation of actual seismic data are important links in guiding oil and gas exploration and development.

[0003] Seismic sandstone and mudstone lithology prediction is the key to finding favorable reservoir distribution in reservoir prediction, and it is also the primary research content of reservoir prediction. Traditional sandstone and mudstone lithology prediction methods have certain limitations. It is difficult to accurately describe the complex nonlinear relationship between seismic elastic parameters (P-wave velocity, S-wave velocity, density) and physical parameters (such as lithology, physical properties), and the prediction results are difficult to evaluate. Therefore, the development of high-precision sandstone and mudstone lithology prediction technology and the evaluation of prediction results are key research topics in reservoir prediction.

[0004] The traditional sandstone and mudstone lithology prediction method is to determine sensitive parameters by multi-parameter intersection analysis based on rock physical quantity plates, and use the sensitive parameters obtained by prestack seismic inversion to predict sandstone and mudstone lithology. An elastic parameter template is drawn according to the influence of reservoir physical properties, and the range of sandstone reservoirs is defined by the different mud contents in the intersection template. Due to the complex nonlinear relationship between lithology and seismic attributes and seismic parameters, precise sandstone and mudstone prediction requires the establishment of a reliable nonlinear model to accurately characterize the mapping relationship.

[0005] In recent years, data-driven deep learning technology has achieved initial results in the field of seismic exploration in seismic data processing, fault detection, seismic inversion, etc., and is an important means to solve nonlinear problems in reservoir prediction. Deep neural networks can theoretically approximate any function through data-driven methods. Since deep learning is a pure data-driven method, when applied to seismic exploration, it often faces problems such as actual data containing noise, lack of learning labels, and lack of logging data. Limited noisy data-driven networks often have low accuracy, confusing physical relationships, and large uncertainty in prediction results.

[0006] In the Chinese patent application with application number: CN202111562006.4, a method, device and equipment for determining reservoir porosity based on rock physical knowledge are involved. The method includes obtaining a well logging data set of a target lithology category; screening a porosity index from the well logging data set using a preset method; wherein the porosity index represents a well logging curve that affects the reservoir porosity; based on the data and porosity curve corresponding to the specified porosity index in the well logging data set, a porosity estimation curve is obtained; the data and porosity estimation curve corresponding to the porosity index in the well logging data set are input into an integrated porosity prediction model to obtain the porosity of the reservoir corresponding to the target lithology category; wherein the integrated porosity prediction model is obtained by constraining the training of multiple decision regressors based on rock physical information. Using this embodiment of the specification can not only better predict the porosity of complex reservoirs and the porosity of different lithologies, but also greatly improve the prediction accuracy of rock porosity.

[0007] In the Chinese patent application with application number: CN201711088665.2, a method for uncertainty modeling based on earthquake multi-attributes is involved. In view of the uncertainty problem of earthquake multi-attributes, the data points of the boundary of the cave body corresponding to the earthquake attribute are taken as the research object of uncertainty analysis, and reasonable probability distribution functions are selected by probability theory and statistical methods to study the uncertainty of the boundary data points of the cave body corresponding to a single earthquake attribute. The uncertainty of the scattered points of the boundary of the cave body corresponding to each earthquake attribute is obtained according to the probability density function, and the corresponding scattered point uncertainty field is established; then, the uncertainty field of the scattered points of the boundary of the cave body corresponding to each earthquake attribute is fused respectively by using the evidence theory, and the fused scattered point uncertainty is obtained, and the uncertainty field of the earthquake multi-attributes is fused according to the method of the evidence theory, and finally the uncertainty of these attributes is visually analyzed.

[0008] In the Chinese patent application with application number: CN201510386046.6, a seismic prediction method for the porosity and mud content of a sandstone mudstone reservoir is involved, which includes: determining the test data of the core sample; obtaining matrix elastic parameters according to rock physics theory calculation; obtaining the theoretical relationship between longitudinal and transverse wave velocities, porosity and mud content of the sandstone mudstone reservoir according to the sandstone mudstone matrix elastic parameters and the test data of the core sample; calculating the logging porosity and logging mud content according to the actual well logging data collected; performing parameter intersection analysis based on the theoretical relationship between longitudinal and transverse wave velocities, porosity and mud content, calibrating and correcting the logging velocity, logging porosity and logging mud content; establishing a quantitative prediction model of well logging longitudinal and transverse wave impedance, porosity and mud content; using seismic pre-stack inversion to obtain seismic longitudinal wave impedance data body, transverse wave impedance data body and density data body; performing seismic porosity and mud content data inversion to obtain the porosity data body and mud content data body for seismic prediction.

[0009] The above existing technologies are all significantly different from the present invention and fail to solve the technical problem we want to solve. For this reason, we have invented a new data-driven sandstone and mudstone lithology prediction and uncertainty estimation method. Summary of the invention

[0010] The purpose of the present invention is to provide a data-driven sandstone and mudstone body lithology prediction and uncertainty estimation method that uses deep learning technology to establish a more accurate sandstone and mudstone prediction model and gives the probability of sandstone and mudstone prediction results.

[0011] The object of the present invention can be achieved by the following technical measures: a data-driven sand-mudstone body lithology prediction and uncertainty estimation method, the data-driven sand-mudstone body lithology prediction and uncertainty estimation method comprising:

[0012] Step 1: Conduct rock physics analysis and select appropriate rock physics models;

[0013] Step 2: Study the fusion method of the data generated by the forward modeling of the rock physics model under the condition of physical property parameter disturbance and the actual data, and generate an elastic parameter data set as an expanded simulation training sample;

[0014] Step 3: Establish statistical rock physics mapping relationship between reservoir sandstone and elastic parameters;

[0015] Step 4: The improved deep neural network no longer outputs lithology results but lithology probabilities.

[0016] The purpose of the present invention can also be achieved by the following technical measures:

[0017] In step 1, rock physics analysis is carried out based on experimental data and logging data, and a suitable rock physics model is selected.

[0018] In step 1, it is clear that the main reservoir lithology of the geological background of the work area is sandstone, and the theory used is the differential equivalent medium theory:

[0019]

[0020]

[0021] The initial condition K * (0) = K1, μ * (0) = μ1; K1 and μ1 are the bulk modulus and shear modulus of the initial main phase material (phase 1); K2 and μ2 are the bulk modulus and shear modulus of the gradually added inclusions (phase 2); y is the content of phase 2; P and Q represent the geometric factors of the pore shape, and the calibration of P and Q can be achieved through the calibration of logging data.

[0022] The rock physics model used is the Gassmann equation:

[0023]

[0024] In the above formula, K sat is the bulk modulus of the fluid saturation; K dry is the bulk modulus of dry rock; K0 is the bulk modulus of matrix; K fl is the bulk modulus of the fluid; ф represents the porosity.

[0025] In step 2, the fusion method of the data generated by the forward modeling of the rock physics model and the actual data under the disturbance of different physical parameters such as porosity, shale content, pore shape, and contact relationship is studied to generate an elastic parameter data set as an expanded simulation training sample.

[0026] In step 2, the target reservoir is relatively deeply buried, the sand body thickness is thin, the porosity is relatively low, and the rock physics model parameter disturbance range is set to: porosity: 0-20%, mud content: 0-40%, fracture aspect ratio: 0.1-0.8.

[0027] In step 3, the simulated sample data and the actual logging data are fused to form a training set for the deep neural network to establish a statistical rock physics mapping relationship between reservoir sandstone and elastic parameters.

[0028] In step 3, the simulated data and the actual data are fused, and the deep learning network used is a deep fully connected network.

[0029] In step 3, the loss function used is:

[0030]

[0031] Where M is the number of categories, p is the predicted probability that the observed sample i belongs to category c, and y is a sign function that takes 1 if the true category of sample i is c, otherwise it takes 0. L represents the value of the loss function; N represents the number of all samples.

[0032] In step 4, a statistical rock physics model suitable for sandstone and mudstone prediction and uncertainty evaluation is constructed; the activation function of the output layer of the neural network is improved to obtain the lithology probability prediction result.

[0033] In step 4, the well-trained model is applied to the prediction of actual profile data in the work area. In the profile results, the predicted sand body part is in good agreement with the well, and the lateral continuity is also high. The probability distribution result of the prediction of sandstone is also given. The darker the color of the area, the greater the possibility of sandstone.

[0034] The data-driven sandstone and mudstone lithology prediction and uncertainty estimation method in the present invention mainly relates to a sandstone and mudstone lithology prediction and uncertainty estimation method, which combines logging data with theoretical simulation data for network driving to overcome data problems, uses deep learning technology to establish a more accurate sandstone and mudstone prediction model and gives the probability of sandstone and mudstone prediction results, and finally obtains a detailed statistical physical model for sandstone and mudstone formation lithology prediction and uncertainty estimation.

[0035] The present invention is based on deep learning technology and uses simulation data and logging data to construct a statistical rock physics model suitable for sandstone and mudstone prediction and uncertainty evaluation. The model combines the advantages of model-driven and data-driven. Since the model-generated data and actual data are used together, both the theoretical physical model of the simulation data and the actual physical mechanism behind the actual data are introduced. By jointly driving the deep neural network with the model-data, a statistical rock physics model for sandstone and mudstone prediction and uncertainty evaluation is constructed. The model can avoid the empirical risk of the prediction results due to insufficient actual data, and to a certain extent solves the problem of insufficient machine learning training samples. It not only has a high prediction accuracy for the prediction of reservoir sandstone and mudstone lithology, but also can give the probability of sandstone and mudstone prediction results. It has a strong generalization ability and can obtain a higher confidence reservoir lithology-elasticity relationship. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a schematic diagram of data expansion of rock physics model in a work area according to a specific embodiment of the present invention;

[0037] Figure 2 It is a schematic diagram of deep learning lithology classification and probability prediction of a work area in a specific embodiment of the present invention;

[0038] Figure 3 It is a schematic diagram of the lithology prediction results of the well F168 in the work area in a specific embodiment of the present invention;

[0039] Figure 4 This is a bar graph showing the accuracy of lithology prediction results for 7 wells in a work area in a specific embodiment of the present invention;

[0040] Figure 5 It is a graph of the lithology probability prediction result of the actual cross section of the work area in a specific embodiment of the present invention;

[0041] Figure 6 The flowchart is a specific embodiment of the data-driven sandstone-mudstone body lithology prediction and uncertainty estimation method of the present invention. DETAILED DESCRIPTION

[0042] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0043] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations and / or combinations thereof.

[0044] Select a suitable rock physics model for the sandstone and mudstone in the work area, and conduct rock physics analysis based on experimental data and logging data to determine the main influencing parameters of the model and the value range of the corresponding parameters. That is, by adjusting parameters such as porosity, skeleton properties, fluid properties, and fluid saturation, an elastic parameter data set is generated as an expanded simulation training sample. The simulation sample data and the actual logging data are combined as a training set for deep learning technology to establish a statistical rock physics mapping relationship between reservoir sandstone and elastic parameters, and the network is adjusted to no longer output lithology results but lithology probabilities. That is, a statistical rock physics model suitable for sandstone and mudstone prediction and uncertainty evaluation is constructed.

[0045] The following are several specific embodiments of the present invention.

[0046] Example 1

[0047] In a specific embodiment 1 of the present invention, the data-driven sandstone and mudstone lithology prediction and uncertainty estimation method of the present invention is specifically implemented as follows:

[0048] Step 1: First, conduct rock physics analysis based on experimental data and logging data, and select a suitable rock physics model.

[0049] Step 2: Study the fusion method of the data generated by the forward modeling of the rock physics model and the actual data under different perturbations of physical parameters such as porosity, shale content, pore shape, and contact relationship, and generate an elastic parameter data set as an expanded simulation training sample.

[0050] Step 3: The simulated sample data and the actual logging data are integrated and used together as the training set of the deep neural network to establish the statistical rock physics mapping relationship between the reservoir sandstone and the elastic parameters.

[0051] Step 4: The improved deep neural network no longer outputs lithology results but lithology probabilities.

[0052] Example 2

[0053] In a specific embodiment 2 of the present invention, as Figure 6 As shown, Figure 6 The flowchart of the data-driven sand-mudstone lithology prediction and uncertainty estimation method of the present invention is as follows. The data-driven sand-mudstone lithology prediction and uncertainty estimation method includes:

[0054] Step 101: First, perform rock physics analysis based on experimental data and logging data, and select a suitable rock physics model.

[0055] It is clear that the main reservoir lithology of the geological background of the work area is sandstone, and the theory used is the differential equivalent medium theory:

[0056]

[0057]

[0058] The initial condition K * (0) = K1, μ * (0) = μ1; K1 and μ1 are the bulk modulus and shear modulus of the initial main phase material (phase 1); K2 and μ2 are the bulk modulus and shear modulus of the gradually added inclusions (phase 2); y is the content of phase 2; P and Q represent the geometric factors of the pore shape, and the calibration of P and Q can be achieved through the calibration of logging data.

[0059] The rock physics model used is the Gassmann equation:

[0060]

[0061] In the above formula, K sat is the bulk modulus of the fluid saturation; K dry is the bulk modulus of dry rock; K0 is the bulk modulus of matrix; K fl is the bulk modulus of the fluid; ф represents the porosity.

[0062] Step 102: Study the fusion method of the data generated by the rock physics model forward modeling and the actual data under the conditions of different porosity, shale content, pore shape, contact relationship and other physical property parameter disturbances, and generate an elastic parameter data set as an expanded simulation training sample.

[0063] The target reservoir is relatively deep (~3000m), the sand body is thin, the porosity is relatively low (<15%), and the rock physics model parameter disturbance range is set as: porosity: 0-20%, mud content: 0-40%, fracture aspect ratio: 0.1-0.8.

[0064] Step 103: The simulated sample data and the actual logging data are integrated to form a training set of a deep neural network, so as to establish a statistical rock physics mapping relationship between reservoir sandstone and elastic parameters.

[0065] The simulated data and actual data are integrated. The deep learning network used is a deep fully connected network, and the loss function is:

[0066]

[0067] Where M is the number of categories, p is the predicted probability that the observed sample i belongs to category c, and y is a sign function that takes 1 if the true category of sample i is c, otherwise it takes 0. L represents the value of the loss function; N represents the number of all samples.

[0068] Step 104: Improve the deep neural network to output lithology probabilities instead of lithology results, and construct a statistical rock physics model suitable for sandstone and mudstone prediction and uncertainty assessment. Improve the activation function of the output layer of the neural network to obtain lithology probability prediction results.

[0069] Example 3

[0070] In a specific embodiment 3 of the present invention, according to the sandstone-mudstone body lithology prediction and uncertainty estimation method proposed in the present invention, an embodiment is given as follows: In this example, the work area is the Daluhu Oilfield in the Fanbei work area of ​​the Shengli Oilfield.

[0071] Step 1: First, conduct rock physics analysis based on experimental data and logging data, and select a suitable rock physics model;

[0072] In this example, the target reservoir is relatively deep (~3000m), the sand body is turbidite sand body, the sand body thickness is thin, the porosity is relatively low (<15%), and its lithology is structural muddy sandstone. In this example, we use the differential equivalent medium model and Gassmann equation to calculate the elastic parameters. The differential equivalent medium theory is as follows:

[0073]

[0074]

[0075] Gassmann equation:

[0076]

[0077] The empirical parameters of P and Q reflecting the pore shape are obtained by fitting the well logging data. The shape expressions of P and Q are detailed in the paper of Berryman (1980), which is not the content of the present invention and will not be described in detail.

[0078] Step 2: Study the fusion method of the data generated by the forward modeling of the rock physics model and the actual data under different perturbations of physical parameters such as porosity, shale content, pore shape, and contact relationship, and generate an elastic parameter data set as an expanded simulation training sample.

[0079] In this embodiment, the rock physics model parameter perturbation range is set to: porosity: 0-20%, shale content: 0-40%, fracture aspect ratio: 0.1-0.8. The model perturbation parameter results are shown in the attached Figure 1 As shown. The horizontal axis is density, the vertical axis is shear wave velocity, and the color represents porosity. The scattered points are actual logging data, and the background is the rock physics model augmented data. It can be seen that the augmented data of the rock physics model is well integrated with the actual logging data. The scattered points are actual logging data, and the background is augmented data.

[0080] Step 3: Fusion of simulated sample data and actual logging data, both of which are used as the training set of the deep neural network to establish the statistical rock physics mapping relationship between reservoir sandstone and elastic parameters. See the attached figure for the deep learning lithology classification and probability prediction process. Figure 2 The network is trained jointly with simulated data and actual data. The lithological sensitivity factors input to the network are λ, μ, and vp / vs. The post-stack impedance is added as the final input based on the pre-stack inversion information.

[0081] In this embodiment, after a large number of experiments, the fused simulated data and actual data are used as the network training set. The lithology sensitivity factors of the network input are λ, μ, and vp / vs. At the same time, the post-stack impedance is added as the final input based on the pre-stack inversion information to make the final lithology prediction. There are 7 wells in the work area. We use 6 wells for training and the remaining 1 well for blind well testing. The results are shown in the attached figure. Figure 3 , the left side is the real label, the right side is the network prediction result, and AC is the prediction accuracy. We define the accuracy as the proportion of correctly predicted samples to the total predicted samples. Taking F168 as an example, the accuracy (AC) of the deep learning lithology blind well prediction result can reach 0.99. In the blind well test of the other 6 wells, as shown in the attached Figure 4 As shown in the figure, dark colors represent only pre-stack information as network input. Gray colors represent both pre-stack and post-stack information as input. The lithology prediction accuracy of each well remains at a high level. At the same time, we also found that compared with only inputting pre-stack information, the sandstone prediction accuracy is higher when the pre-stack and post-stack multi-dimensional information is fused and input.

[0082] Step 4: Improve the deep neural network to output lithology probabilities instead of lithology results.

[0083] By improving the activation function of the output layer of the neural network, we can get the lithology probability prediction results. We applied the well-trained model to the actual profile data prediction of the work area. In the profile results, the predicted sand body part is well consistent with the well, and the lateral continuity is also high. The probability distribution results of the prediction of sandstone are given. The darker the color, the greater the possibility of sandstone.

[0084] Figure 5 This is the result of the probability prediction of the lithology of the actual section of the working area in the embodiment. This probability is the probability that the lithology of the sample point is sandstone. The gray part represents that the probability of sandstone is less than 50%, and the bright part represents that the probability of sandstone is higher than 50%. The brighter the color, the greater the possibility of sandstone.

[0085] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions recorded in the aforementioned embodiments or replace some of the technical features therein with equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0086] Except for the technical features described in the specification, all other technical features are known technologies to those skilled in the art.

Claims

1. A data-driven sandstone-mudstone lithology prediction and uncertainty estimation method, characterized in that: The data-driven sandstone-mudstone lithology prediction and uncertainty estimation method includes: Step 1: Conduct rock physics analysis and select appropriate rock physics models; Step 2: Study the fusion method of the data generated by the forward modeling of the rock physics model under the condition of physical property parameter disturbance and the actual data, and generate an elastic parameter data set as an expanded simulation training sample; Step 3: Establish statistical rock physics mapping relationship between reservoir sandstone and elastic parameters; Step 4: The improved deep neural network no longer outputs lithology results but lithology probabilities.

2. The data-driven sandstone-mudstone lithology prediction and uncertainty estimation method according to claim 1, characterized in that: In step 1, rock physics analysis is carried out based on experimental data and logging data, and a suitable rock physics model is selected.

3. The data-driven sandstone-mudstone lithology prediction and uncertainty estimation method according to claim 2, characterized in that: In step 1, it is clear that the main reservoir lithology of the geological background of the work area is sandstone, and the theory used is the differential equivalent medium theory: The initial condition K * (0) = K1, μ * (0) = μ1; K1 and μ1 are the bulk modulus and shear modulus of the initial main phase material, i.e., phase 1; K2 and μ2 are the bulk modulus and shear modulus of the gradually added inclusions, i.e., phase 2; y is the content of phase 2; P and Q represent the geometric factors of the pore shape, which can be calibrated by logging data calibration; The rock physics model used is the Gassmann equation: In the above formula, K sat is the bulk modulus of the fluid saturation; K dry is the bulk modulus of dry rock; K0 is the bulk modulus of matrix; K fl is the bulk modulus of the fluid; ф represents the porosity.

4. The data-driven sandstone-mudstone lithology prediction and uncertainty estimation method according to claim 1, characterized in that: In step 2, the fusion method of the data generated by the forward modeling of the rock physics model and the actual data under the disturbance of different physical parameters such as porosity, shale content, pore shape, and contact relationship is studied to generate an elastic parameter data set as an expanded simulation training sample.

5. The data-driven sandstone-mudstone lithology prediction and uncertainty estimation method according to claim 1, characterized in that: In step 2, the target reservoir is relatively deeply buried, the sand body thickness is thin, the porosity is relatively low, and the rock physics model parameter disturbance range is set to: porosity: 0-20%, mud content: 0-40%, fracture aspect ratio: 0.1-0.

8.

6. The data-driven sandstone-mudstone lithology prediction and uncertainty estimation method according to claim 1, characterized in that: In step 3, the simulated sample data and the actual logging data are fused to form a training set for the deep neural network to establish a statistical rock physics mapping relationship between reservoir sandstone and elastic parameters.

7. The data-driven sandstone-mudstone lithology prediction and uncertainty estimation method according to claim 6, characterized in that: In step 3, the simulated data and the actual data are fused, and the deep learning network used is a deep fully connected network.

8. The data-driven sandstone-mudstone lithology prediction and uncertainty estimation method according to claim 7, characterized in that: In step 3, the loss function used is: Where M is the number of categories, p is the predicted probability that the observed sample i belongs to category c, y is the sign function, which takes 1 if the true category of sample i is c, otherwise it takes 0; L represents the value of the loss function; N represents the number of all samples.

9. The data-driven sandstone-mudstone lithology prediction and uncertainty estimation method according to claim 1, characterized in that: In step 4, a statistical rock physics model suitable for sandstone and mudstone prediction and uncertainty evaluation is constructed; the activation function of the output layer of the neural network is improved to obtain the lithology probability prediction result.

10. The data-driven sandstone-mudstone body lithology prediction and uncertainty estimation method according to claim 9, characterized in that: In step 4, the well-trained model is applied to the prediction of actual profile data in the work area. In the profile results, the predicted sand body part is in good agreement with the well, and the lateral continuity is also high. The probability distribution result of the prediction of sandstone is also given. The darker the color of the area, the greater the possibility of sandstone.

Images