A low permeability reservoir classification method, device and medium

Abstract

The application relates to a low-permeability reservoir classification method, which comprises the following steps: drilling a limited number of core samples distributed at different well depths, and performing physical property testing on each core sample to obtain a corresponding core classification result, which is used as a sample label; acquiring parameter values of each well logging curve in preferred well logging curves, collecting core sample labels and corresponding physical parameters at different depths as labeled sample data, and acquiring well logging curve parameter values at a set well depth according to each well logging parameter curve with a set step, and taking well logging curve parameter values at depths without core samples as unlabeled sample data; based on the labeled sample data and the unlabeled sample data, a semi-supervised machine learning model is trained; and based on the trained semi-supervised machine learning model, the core classification result at the set well depth of each well logging is predicted. The scheme improves the accuracy of reservoir classification.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of oil and gas development, and particularly relates to a low-permeability reservoir classification method, device and medium. BACKGROUND

[0002] Reservoir classification is an important part of reservoir characterization technology and a core difficulty in reservoir research.

[0003] At present, a method for classifying reservoirs by using a machine learning algorithm has been reported, and the existing classification method can obtain good classification results when applied to conventional reservoirs.

[0004] However, the inventors of the present application found in research that, compared with conventional reservoirs, low-permeability reservoirs have different physical properties, and the technical difficulty and cost of sampling (core) are high. Compared with conventional reservoirs, the number of samples obtained under reasonable conditions is much smaller than that of conventional reservoirs. Under the condition of a small number of samples, if the existing machine learning model is used for training and the trained model is used for classification prediction, the accuracy of the classification effect will be greatly reduced. Therefore, it is necessary to propose a corresponding reservoir classification technology for low-permeability reservoirs. SUMMARY

[0005] To solve the above problems, the purpose of the present application is to provide a low-permeability reservoir classification method, which can construct a semi-supervised machine learning model based on a small number of samples obtained by a set number of well logging and well logging parameter curves, form a reservoir classification quantitative method, and improve the accuracy of reservoir classification.

[0006] To achieve the above purpose, the present application adopts the following technical solutions:

[0007] In a first aspect, the present application provides a low-permeability reservoir classification method, comprising:

[0008] A set number of coresamples distributed at different well depths are extracted by a set number of well logging, and physical property testing is performed on each of the core samples to obtain corresponding core classification results, which are used as sample labels;

[0009] Well logging parameter curves of each of the well logging are obtained, and physical parameters at a set well depth are obtained from each of the well logging parameter curves by using a set step, and the sample labels of the core samples at different depths and the corresponding physical parameters are further combined as labeled sample data, and the physical parameters at the remaining depths are used as unlabeled sample data;

[0010] Based on the labeled sample data and the unlabeled sample data, a semi-supervised machine learning model is trained;

[0011] Based on the trained semi-supervised machine learning model, a core classification result at a set depth of each of the well logs is predicted.

[0012] In an implementation of the present application, the physical property test on each of the core samples to obtain a corresponding core classification result comprises:

[0013] The physical property test is performed on each of the core samples to obtain a corresponding porosity and permeability;

[0014] The porosity and the permeability are taken as two-dimensional coordinates of each of the core samples, fitting is performed in a corresponding coordinate system, and a core classification result of each of the core samples is determined according to a fitting result.

[0015] In an implementation of the present application, the fitting adopts linear fitting, the fitting result is a straight line with two different slopes, and the corresponding core classification result comprises a first type of core, a second type of core and a third type of core.

[0016] In an implementation of the present application, the well log parameter curve of each of the well logs is obtained by:

[0017] A predefined number of basic physical parameters are screened out;

[0018] Each well log is detected according to each of the basic physical parameters to obtain a test curve corresponding to each of the basic physical parameters;

[0019] A well log parameter curve of each of the well logs is obtained according to the test curve corresponding to each of the basic physical parameters and a combined curve after data processing of at least two test curves.

[0020] In an implementation of the present application, the basic physical parameters comprise acoustic parameters, lithology photoelectric parameters, density parameters, shale content parameters and neutron parameters.

[0021] In an implementation of the present application, the well log parameter curve comprises an acoustic curve, a lithology photoelectric curve, a shallow lateral resistivity curve, a deep lateral resistivity curve, a density curve, a shale content curve, a neutron curve and a deep / shallow lateral resistivity ratio curve.

[0022] The acoustic curve is obtained by detecting acoustic parameters by each well log and reflects formation porosity, lithology and pore fluid properties.

[0023] The lithology photoelectric curve is obtained by detecting lithology photoelectric parameters by each well log and reflects formation lithology.

[0024] The shallow lateral resistivity curve is obtained by detecting shallow lateral resistivity parameters by each well log and reflects formation resistivity.

[0025] Deep lateral resistivity curve, which is obtained by each well logging detecting deep lateral resistivity parameter, reflects the high and low of formation resistivity;

[0026] Density curve, which is obtained by each well logging detecting density parameter, reflects the formation lithology and porosity;

[0027] Shale content curve, which is obtained by calculating natural gamma curve, reflects the size of formation shale content;

[0028] Neutron curve, which is obtained by each well logging detecting neutron parameter, reflects the formation porosity;

[0029] Deep / shallow lateral resistivity ratio curve, which is obtained by dividing deep lateral resistivity by shallow lateral resistivity, reflects the formation permeability.

[0030] In an implementation form of the present application, the physical parameters corresponding to different well logging parameter curves correspond to different parameter sensitivities.

[0031] In an implementation form of the present application, the training of the semi-supervised machine learning model based on the labeled sample data and the unlabeled sample data comprises:

[0032] Inputting a set proportion of the labeled sample data and the unlabeled sample data as a training set into the semi-supervised machine learning model;

[0033] Using the set proportion of the labeled sample data to construct an initial model to predict the core classification of the unlabeled sample data until the prediction accuracy is higher than a first set probability, and taking the predicted core classification of the unlabeled sample data as pseudo-labels;

[0034] Continuing to train the model using the set proportion of the labeled sample data and the sample data with pseudo-labels until the prediction accuracy on a validation set is higher than a second set probability within a set number of iterations, wherein the validation set is the remaining part of the labeled sample data excluding the set proportion.

[0035] In a second aspect, the present application provides a low-permeability reservoir classification device, which comprises:

[0036] A label acquisition module drills out a limited number of core samples distributed at different well depths, and performs physical property testing on each core sample to obtain a corresponding core classification result, which is taken as a sample label;

[0037] a sample acquisition module, configured to acquire parameter values of each of the well logging curves in the preferred well logging curves, to take the core sample labels and corresponding physical parameter sets at different depths as labeled sample data, and to acquire parameter values of the well logging curves at the set well depths according to each of the well logging parameter curves, and to take the parameter values of the well logging curves at the depths without core samples as unlabeled sample data;

[0038] a training module, configured to train a semi-supervised machine learning model based on the labeled sample data and the unlabeled sample data;

[0039] a prediction module, configured to predict the core classification results at the set well depths of each of the well logs based on the trained semi-supervised machine learning model.

[0040] In a third aspect, the present application provides a computer readable storage medium, wherein a computer program is stored in the computer readable storage medium, and the computer program controls a device where the computer readable storage medium is located to execute the low-permeability reservoir classification method in the first aspect.

[0041] The present application has the following advantages due to the above technical solutions: a predetermined number of core samples distributed at different well depths are extracted through a set number of well logs, and corresponding core classification results are obtained as sample labels; well logging parameter curves are acquired, and physical parameters at the set well depths are acquired, and the sample labels of the core samples and the corresponding physical parameter sets are taken as labeled sample data, and the remaining physical parameters are taken as unlabeled sample data; further, a semi-supervised machine learning model is trained based on the labeled sample data and the unlabeled sample data, so that when the trained model is used to predict the core classification results, the prediction accuracy can be improved. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 is a schematic diagram of core classification provided by an embodiment of the present application;

[0043] Figure 2 is a comparison schematic diagram of the parameter importance of physical parameters related by an embodiment of the present application;

[0044] Figure 3 is a schematic diagram of well logging curves in an embodiment of the present application;

[0045] Figure 4 is a flowchart of model training in an embodiment of the present application;

[0046] Figure 5 is a schematic diagram of model stability;

[0047] Figure 6 is a schematic diagram of application effects of an embodiment of the present application. DETAILED DESCRIPTION

[0048] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions of the embodiments of the present application will be described clearly and completely below with reference to the drawings of the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments of the present application. Based on the described embodiments of the present application, all other embodiments obtained by a person of ordinary skill in the art belong to the scope of protection of the present application.

[0049] In view of the low classification effect of the low-permeability reservoir in the prior art, the present application provides a low-permeability reservoir classification method, device and medium. The method comprises the following steps: a limited number of core samples distributed at different depths are drilled out, and physical property tests are performed on each core sample to obtain corresponding core classification results, which are used as sample labels; well logging parameter curves of each well logging are obtained, and physical parameters at a set depth are obtained from each well logging parameter curve by using a set step, and the sample labels of the core samples at different depths and the corresponding physical parameters are further combined as labeled sample data, and the physical parameters at the remaining depths are used as unlabeled sample data; a semi-supervised machine learning model is trained based on the labeled sample data and the unlabeled sample data; and the core classification results at the set depth of each well logging are predicted based on the trained semi-supervised machine learning model. The present application can construct a semi-supervised machine learning model based on a small number of samples and well logging curve data obtained from a limited number of well loggings, form a reservoir classification quantitative method, and improve the accuracy of reservoir classification.

[0050] Reference should also be made to Figures 1 to 5 In one aspect of the present application, a low-permeability reservoir classification method is provided.

[0051] The low-permeability reservoir classification method of the present application comprises the following steps:

[0052] S1, a limited number of core samples distributed at different depths are drilled out, and physical property tests are performed on each core sample to obtain corresponding core classification results, which are used as sample labels.

[0053] Specifically, the present application is used to predict the core classification results at different depths in the low-permeability reservoir.

[0054] For low permeability reservoirs, coring is difficult and costly. In the embodiment, 4 wells can be used to obtain 60 core samples of a predetermined number in an oilfield. The core samples can be distributed at different depths in each well, for example, the maximum depth of the well is 3000 meters, and the core samples can be distributed at different depths of 1200 meters, 1800 meters, 2000 meters, etc. according to the distribution characteristics of the reservoir.

[0055] Further physical property testing is performed on the core samples to classify them. Specifically, the physical property of each core sample is tested to obtain the corresponding porosity and permeability; the porosity and permeability are used as two-dimensional coordinates of each core sample, and fitting is performed in the corresponding coordinate system; and according to the fitting result, the core classification result of each core sample is determined.

[0056] The schematic diagram is shown in Figure 1 where the horizontal axis of the coordinate system is porosity and the vertical axis is permeability, and each point in the coordinate system represents a core sample. Linear fitting is performed on the core samples to obtain three straight lines, and the fitting result of different core samples corresponds to a core classification result. In the embodiment, the results are classified into one type of core, two types of core and three types of core, as shown in Figure 1 The fitting slopes of the two types of core are different, representing the difference in physical properties of the two types of core.

[0057] In the embodiment, the core classification result of the core sample is used as the sample label Y.

[0058] S2, obtain the well logging parameter curve of each well, and obtain the physical parameter at the set depth using a set step length according to each well logging parameter curve, and further collect the sample label of the core sample at different depths and the corresponding physical parameter as labeled sample data, and the physical parameter at the remaining depth as unlabeled sample data.

[0059] Specifically, the process of obtaining the well logging parameter curve includes: first, screening a predetermined number of basic physical parameters. Then, according to each of the basic physical parameters, detecting each well to obtain a test curve corresponding to each of the basic physical parameters; further, according to each of the test curves corresponding to the basic physical parameters and the combined curve after data processing of at least two of the test curves, obtaining the well logging parameter curve of each well.

[0060] For example, according to the cognition of geologists, a predetermined number of basic physical parameters can be screened, which can include acoustic parameters, lithology photoelectric parameters, density parameters, shale content parameters and neutron parameters.

[0061] The well logging parameter curves include an acoustic curve, a lithology photoelectric curve, a shallow lateral resistivity curve, a deep lateral resistivity curve, a density curve, a shale content curve, a neutron curve, and a deep / shallow lateral resistivity ratio curve.

[0062] The acoustic curve is obtained by detecting an acoustic parameter by each well logging, and reflects formation porosity, lithology, and pore fluid properties.

[0063] The lithology photoelectric curve is obtained by detecting a lithology photoelectric parameter by each well logging, and reflects formation lithology.

[0064] The shallow lateral resistivity curve is obtained by detecting a shallow lateral resistivity parameter by each well logging, and reflects formation resistivity.

[0065] The deep lateral resistivity curve is obtained by detecting a deep lateral resistivity parameter by each well logging, and reflects formation resistivity.

[0066] The density curve is obtained by detecting a density parameter by each well logging, and reflects formation lithology and porosity.

[0067] The shale content curve is obtained by calculating a natural gamma ray curve, and reflects formation shale content.

[0068] The neutron curve is obtained by detecting a neutron parameter by each well logging, and reflects formation porosity.

[0069] The deep / shallow lateral resistivity ratio curve is obtained by dividing the deep lateral resistivity by the shallow lateral resistivity, and reflects formation permeability.

[0070] In the embodiments of the present application, different parameters have different importance, and a schematic diagram is as shown in Figure 2 .

[0071] Further, in the embodiments of the present application, corresponding physical parameters of sample data can be obtained according to the well logging parameter curves.

[0072] For the above eight well logging parameter curves, 0.1 meters, 0.5 meters, etc. can be taken as a step to obtain physical parameters at different depths. It is easy to understand that, for sample data at the depth of a core sample, not only has physical parameters, but also has labels, forming labeled sample data, and sample data at other depths only contains physical parameters, but does not have labels, forming unlabeled sample data. For all sample data, the physical parameters are taken as independent variables X in the embodiments of the present application. A schematic diagram of obtaining sample data through well logging parameter curves is as shown in Figure 3 .

[0073] S3, based on the labeled sample data and the unlabeled sample data, training a semi-supervised machine learning model.

[0074] Specifically, the schematic diagram of model training is as Figure 4 .

[0075] Specifically, a proportion (for example, 80%) of the labeled sample data and the unlabeled sample data are input into the semi-supervised machine learning model as a training set;

[0076] The initial model is constructed based on the proportion of the labeled sample data, and the core classification of the unlabeled sample data is predicted until the predicted probability is higher than a first set probability (0.9), and the predicted core classification of the unlabeled sample data is used as a pseudo label.

[0077] The model is further trained based on the proportion of the labeled sample data and the sample data with the pseudo label until the prediction accuracy of the validation set is higher than a set probability within a set number of iterations, wherein the validation set is the remaining part (i.e., 20%) of the labeled sample data excluding the proportion of the labeled sample data.

[0078] The accuracy of the model is verified by using the validation data set (20% of the labeled core data), and the stability of the model is evaluated by the cross-validation method. The principle is to select the data range of the validation set in turn, and the test results show that the prediction accuracy of the validation set is more than 80%, and the stability of the model is high as Figure 5 indicated.

[0079] S4, based on the trained semi-supervised machine learning model, predicting the core classification result at the set depth of each well log.

[0080] The existing technology is based on large-scale sample training to obtain high accuracy for predicting conventional reservoirs. For low-permeability reservoirs, the prediction accuracy is low due to the small number of samples. The semi-supervised machine learning model trained by the present application is based on labeled samples and pseudo-labeled samples obtained from well logs to train the model, which can improve the prediction accuracy compared with the existing technology.

[0081] Compared with other classification methods such as support vector machine, the model of the present application is more accurate in predicting the physical property changes of fine interlayers, and the comparison effect diagram is as Figure 6 indicated.

[0082] In summary, the present application extracts a predetermined number of core samples distributed at different well depths through a set number of well logs, and obtains corresponding core classification results as sample labels; then acquires well log parameter curves and physical parameters at the set well depths, and collects the sample labels of the core samples and the corresponding physical parameters as labeled sample data, and the remaining physical parameters as unlabeled sample data; further based on the labeled sample data and the unlabeled sample data, a semi-supervised machine learning model is trained, and thus, when the trained model is used to predict core classification results, the prediction accuracy can be improved.

[0083] In several embodiments provided by the present application, it should be understood that the disclosed method can be implemented by other ways. For example, the above-described device embodiments are merely illustrative, for example, the division of the above-mentioned units is merely a logical function division, and actual implementation can have another division manner, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the displayed or discussed units can be indirect coupling or communication connection through some interfaces, devices or units, and can be electrical, mechanical or other forms.

[0084] The integrated unit in the form of the software function unit described above can be stored in a computer readable storage medium. The software function unit described above is stored in a storage medium, including a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor (Processor) to execute part of the steps of the method described above in various embodiments of the present application. The storage medium described above includes a U disk, a mobile hard disk, a read-only memory (Read-Only Memory, ROM), a random access memory (Random Access Memory, RAM), a magnetic disk or an optical disk, and various program code storage media.

[0085] The above is only a preferred embodiment of the present application, and is not used to limit the present application, and any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method of low permeability reservoir classification characterized by, The method comprises the following steps: cores are drilled out from a limited number of core samples distributed at different well depths, and physical property tests are performed on each of the core samples to obtain corresponding core classification results, which are used as sample labels; logging parameter curves of each well log are obtained, and physical parameters at a set depth are obtained from each of the logging parameter curves using a set step size, and sample labels of core samples at different depths are combined with corresponding physical parameters to obtain labeled sample data, and physical parameters at other depths are used as unlabeled sample data; based on the labeled sample data and the unlabeled sample data, a semi-supervised machine learning model is trained; based on the trained semi-supervised machine learning model, core classification results at the set depth of each well log are predicted; the physical property tests performed on each of the core samples to obtain corresponding core classification results comprise: physical property tests performed on each of the core samples to obtain corresponding porosity and permeability; the porosity and the permeability are used as two-dimensional coordinates of each of the core samples, fitting is performed in a corresponding coordinate system, and core classification results of each of the core samples are determined according to the fitting results; the training of the semi-supervised machine learning model based on the labeled sample data and the unlabeled sample data comprises: a set proportion of the labeled sample data and the unlabeled sample data are used as a training set and input into the semi-supervised machine learning model; an initial model is constructed using the set proportion of the labeled sample data, core classification of the unlabeled sample data is predicted, until the accuracy of the prediction is higher than a first set probability, and the predicted core classification of the unlabeled sample data is used as pseudo-labels; the training of the model is continued using the set proportion of the labeled sample data and the sample data with the pseudo-labels, until the accuracy of the prediction on a validation set is higher than a second set probability within a set number of iterations, wherein the validation set is the remaining part of the labeled sample data excluding the set proportion.

2. The low permeability reservoir classification method of claim 1, wherein, the logging parameter curves of each well log are obtained by: a predefined number of basic physical parameters are screened out; each well log is detected according to each of the basic physical parameters to obtain a logging curve corresponding to each of the basic physical parameters; logging parameter curves of each well log are obtained according to the test curve corresponding to each of the basic physical parameters and a combined curve obtained by combining at least two test curves after data processing.

3. The low permeability reservoir classification method of claim 2, wherein, The basic physical parameters include acoustic parameters, lithology photoelectric parameters, density parameters, shale content parameters, and neutron parameters.

4. The method of low permeability reservoir classification of claim 3, wherein, The logging parameter curves include acoustic curves, lithology photoelectric curves, shallow lateral resistivity curves, deep lateral resistivity curves, density curves, shale content curves, neutron curves, and deep / shallow lateral resistivity ratio curves; The acoustic curve is obtained by detecting acoustic parameters by each well log, and reflects formation porosity, lithology, and pore fluid properties. The lithology photoelectric curve is obtained by detecting lithology photoelectric parameters by each well log, and reflects formation lithology. The shallow lateral resistivity curve is obtained by detecting the shallow lateral resistivity parameter of each well logging, and reflects the high and low of the formation resistivity. The deep lateral resistivity curve is obtained by detecting the deep lateral resistivity parameter of each well logging, and reflects the high and low of the formation resistivity. The density curve is obtained by detecting the density parameter of each well logging, and reflects the formation lithology and porosity. The shale content curve is obtained by calculating the natural gamma ray curve, and reflects the shale content of the formation. The neutron curve is obtained by detecting the neutron parameter of each well logging, and reflects the formation porosity. The deep / shallow lateral resistivity ratio curve is obtained by dividing the deep lateral resistivity by the shallow lateral resistivity, and reflects the formation permeability.

5. The method of low permeability reservoir classification of claim 4, wherein, The physical parameters corresponding to different well logging parameter curves correspond to different parameter sensitivities.

6. A low permeability reservoir classification device characterized by, The device comprises: A label acquisition module is configured to extract a predetermined number of core samples distributed at different well depths through a set number of well loggings, and perform physical property testing on each core sample to obtain a corresponding core classification result, which is used as a sample label. A sample acquisition module is configured to acquire well logging parameter curves of each well logging, and acquire physical parameters at a set well depth according to each well logging parameter curve with a set step size, and further collect sample labels of core samples at different depths and corresponding physical parameters as labeled sample data, and physical parameters at the rest of the depths as unlabeled sample data. A training module is configured to train a semi-supervised machine learning model based on the labeled sample data and the unlabeled sample data. A prediction module is configured to predict the core classification result at the set well depth of each well logging based on the trained semi-supervised machine learning model. The device implements the method of claim 1.

7. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program controls the device where the computer readable storage medium is located to execute the low-permeability reservoir classification method of any one of claims 1 to 5 when running. The computer readable storage medium stores a computer program, and the computer program controls the device where the computer readable storage medium is located to execute the low-permeability reservoir classification method of any one of claims 1 to 5 when running.

Images