Method and device for constructing nuclear magnetic resonance T2 spectrum using physical constraints and data drive

By preprocessing conventional logging data and inputting it into the pre-trained NMR T2 spectrum prediction model, the NMR T2 spectrum is generated, which solves the problems of complex and high cost of NMR logging instruments, and realizes the use of oil exploration in The effect of reducing costs.

CN114594523BActive Publication Date: 2024-11-19CHINA UNIV OF PETROLEUM (BEIJING)
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210005106.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-01-04
Publication Date
2024-11-19
Estimated Expiration
2042-01-04

AI Technical Summary

Technical Problem

The complex operation procedures of the NMR logger and the slow logging speeds result in high logging work in petroleum exploration and cannot be applied to wells in all reservoirs.

Method used

By obtaining conventional well logging data for preprocessing and inputting into the pre-trained NMR T2 spectrum prediction model, NMR T2 spectrum is generated, including data cleaning, standardization processing and main feature analysis, and prediction using LSTM network and fully connected layer composition The model makes predictions.

Benefits of technology

Without using a NMR logger, NMR T2 spectrum containing internal characteristic information of the formation is generated to estimate pore size distribution and fluid flow, reducing exploration and development costs.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114594523B_ABST
    Figure CN114594523B_ABST
Patent Text Reader

Abstract

The present application provides a method and device for constructing a nuclear magnetic resonance T2 spectrum by physical constraints and data drive, which obtains the target conventional logging data corresponding to the well section to be measured in the lateral well position, and then inputs the target conventional logging data into the pre-trained nuclear magnetic resonance T2 spectrum prediction model to obtain the target nuclear magnetic resonance data corresponding to the target conventional logging data, and then generates a target nuclear magnetic resonance T2 spectrum according to the target nuclear magnetic resonance data. The target nuclear magnetic resonance T2 spectrum includes internal formation feature information, which can be used by mining personnel to estimate the pore size distribution and fluid mobility of the formation, thereby reducing the cost of exploration and development.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to oil exploration and development technology, and in particular to a method and device for constructing a nuclear magnetic resonance T2 spectrum by physical constraints and data drive. Background Art

[0002] Petroleum exploration refers to the process of using various exploration methods to understand the underground geological conditions, oil and gas layer conditions and production capacity in order to find and identify oil and gas resources. In the process of petroleum geology and oil and gas field exploration and development, the economic development of reservoirs depends on accurate geological and petrophysical characteristics.

[0003] At present, geological and rock physical characteristics can be analyzed by nuclear magnetic resonance logging. Nuclear magnetic resonance logging instruments excite hydrogen nuclei in underground fluids through an external magnetic field. After removing the external magnetic field, the relaxation of hydrogen nuclei produces a relaxation signal, which is inverted to obtain a nuclear magnetic resonance T2 spectrum containing 64 T2 amplitudes corresponding to 64 T2 wave numbers. The nuclear magnetic resonance T2 spectrum is the relaxation time distribution of hydrogen nuclei in formation fluids at a specific formation depth, which can be used to estimate the pore size distribution and fluid mobility of near-well formations.

[0004] However, due to the complex operating procedures and slow logging speed of the nuclear magnetic resonance logging instrument, the cost of logging work is greatly increased and it cannot be applied to wells drilled in all reservoirs. Summary of the invention

[0005] The present application provides a method and device for constructing a nuclear magnetic resonance T2 spectrum by physical constraints and data driving, so as to solve the problem that the cost of well logging is high when using a nuclear magnetic resonance logging instrument to obtain a nuclear magnetic resonance T2 spectrum.

[0006] In a first aspect, the present application provides a method for constructing a nuclear magnetic resonance T2 spectrum using physical constraints and data-driven methods, comprising:

[0007] Obtain target conventional logging data corresponding to the well section to be tested in the well position to be tested;

[0008] Performing one or more of the following preprocessing on the target conventional logging data: data cleaning, standardization processing and main feature analysis;

[0009] Inputting the preprocessed target conventional logging data into a pre-trained nuclear magnetic resonance T2 spectrum prediction model to obtain target nuclear magnetic resonance data, wherein the target nuclear magnetic resonance data includes echo data in different formats;

[0010] A target nuclear magnetic resonance T2 spectrum is generated according to the target nuclear magnetic resonance data.

[0011] Optionally, the target conventional logging data includes the following parameter groups: natural gamma value, natural potential, well diameter, deep lateral resistivity, medium lateral resistivity, shallow lateral resistivity, acoustic wave time difference, compensated neutron and logging density, and each parameter group includes parameter values ​​corresponding to each depth in the well section to be logged.

[0012] Optionally, also include:

[0013] Acquire training data, the training data including feature data and label data, the feature data including conventional logging data corresponding to each depth, and the label data including nuclear magnetic resonance data corresponding to the conventional logging data;

[0014] The NMR T2 spectrum prediction model is obtained by training the preset model using the training data.

[0015] Optionally, the acquiring of training data includes:

[0016] Acquire a large amount of original conventional logging data and original nuclear magnetic resonance data, wherein the original conventional logging data includes the following parameter groups: natural gamma value, natural potential, well diameter, deep lateral resistivity, medium lateral resistivity, shallow lateral resistivity, acoustic wave time difference, compensated neutron, logging density photoelectric cross-section index, density after environment correction and compensated neutron after vertical correction, the original nuclear magnetic resonance data includes original echo data of different formats measured by a nuclear magnetic resonance logging instrument in a target well section of a data extraction well location at a first fixed depth interval, and the original conventional logging data is measured by a logging tool corresponding to each parameter group in the target well section of the data extraction well location at a second fixed depth interval, and the first fixed depth interval is not equal to the second fixed depth interval;

[0017] Calculating the original conventional logging data according to preset depth intervals using logging data processing software to obtain calculated original conventional logging data;

[0018] Calculating the original nuclear magnetic resonance data according to the preset depth interval using logging data processing software to obtain calculated original nuclear magnetic resonance data, and using the calculated original nuclear magnetic resonance data as the label data;

[0019] Preprocessing the calculated original conventional logging data to obtain preprocessed original conventional logging data, wherein the preprocessing includes data cleaning and / or standardization processing;

[0020] A grey correlation analysis is performed on the preprocessed original conventional logging data to obtain the correlation between each parameter group in the preprocessed original conventional logging data and the natural gamma parameter group in the preprocessed original conventional logging data, and a parameter group with a correlation greater than or equal to a preset threshold is selected to form the conventional logging data.

[0021] Optionally, the selecting a parameter group whose correlation is greater than or equal to a preset threshold to form the conventional well logging data includes:

[0022] Selecting a parameter group whose correlation is greater than or equal to a preset threshold;

[0023] A principal component analysis is performed on the selected parameter groups to obtain principal component information corresponding to each parameter group, and the principal component information corresponding to each parameter group forms the conventional well logging data.

[0024] Optionally, the preset model consists of three long short-term memory neural network LSTM layers and two fully connected layers.

[0025] In a second aspect, the present application provides a device for constructing a nuclear magnetic resonance T2 spectrum by physical constraints and data driving, comprising:

[0026] An acquisition module is used to acquire target conventional logging data corresponding to the well section to be logged in the well position to be logged;

[0027] A preprocessing module, used to perform one or more of the following preprocessing on the target conventional logging data: data cleaning, standardization processing and main feature analysis;

[0028] A prediction module, used for inputting the pre-processed target conventional logging data into a pre-trained nuclear magnetic resonance T2 spectrum prediction model to obtain target nuclear magnetic resonance data, wherein the target nuclear magnetic resonance data includes echo data in different formats;

[0029] A generating module is used to generate a target nuclear magnetic resonance T2 spectrum according to the target nuclear magnetic resonance data.

[0030] Optionally, the target conventional logging data includes the following parameter groups: natural gamma value, natural potential, well diameter, deep lateral resistivity, medium lateral resistivity, shallow lateral resistivity, acoustic wave time difference, compensated neutron and logging density, and each parameter group includes parameter values ​​corresponding to each depth in the well section to be logged.

[0031] In a third aspect, the present application provides an electronic device, comprising: a processor, and a memory communicatively connected to the processor;

[0032] The memory stores computer-executable instructions;

[0033] The processor executes the computer-executable instructions stored in the memory to implement the method according to the first aspect.

[0034] In a fourth aspect, the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer execution instructions, and when the computer execution instructions are executed by a processor, they are used to implement a method for constructing a nuclear magnetic resonance T2 spectrum by physical constraints and data-driven as described in the first aspect.

[0035] In a fifth aspect, the present application provides a computer program product, including a computer program, which implements the method described in the first aspect when executed by a processor.

[0036] The present application provides a method and device for constructing a nuclear magnetic resonance T2 spectrum by physical constraints and data drive, which obtains the target conventional logging data corresponding to the well section to be measured in the lateral well position, and then inputs the target conventional logging data into the pre-trained nuclear magnetic resonance T2 spectrum prediction model to obtain the target nuclear magnetic resonance data corresponding to the target conventional logging data, and then generates a target nuclear magnetic resonance T2 spectrum according to the target nuclear magnetic resonance data. The target nuclear magnetic resonance T2 spectrum includes internal formation feature information, which can be used by mining personnel to estimate the pore size distribution and fluid mobility of the formation, thereby reducing the cost of exploration and development. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0038] Figure 1 is a schematic diagram of a nuclear magnetic resonance T2 spectrum;

[0039] Figure 2 A schematic diagram of a network structure of a nuclear magnetic resonance T2 spectrum prediction model used in this application;

[0040] Figure 3 A schematic diagram of a process for constructing a nuclear magnetic resonance T2 spectrum using physical constraints and data-driven methods provided in Example 1 of the present application;

[0041] Figure 4 A schematic diagram of the curve formed by the natural potential connection at various depths in a well section;

[0042] Figure 5 A schematic diagram of a process for constructing a nuclear magnetic resonance T2 spectrum using physical constraints and data-driven methods provided in Example 2 of the present application;

[0043] Figure 6 This is the error change diagram of the above preset model training and testing process;

[0044] Figure 7 Schematic diagram of the network structure of RNN;

[0045] Figure 8 Preset the error change diagram of the RNN model training and testing process;

[0046] Fig. 9 A schematic diagram of the structure of a device for constructing a nuclear magnetic resonance T2 spectrum by physical constraints and data drive provided in Example 3 of the present application;

[0047] Fig.10 A schematic diagram of the structure of an electronic device provided in Example 4 of the present application.

[0048] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0049] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0050] At present, geological and rock physical characteristics can be analyzed by nuclear magnetic resonance logging. Nuclear magnetic resonance logging uses an external magnetic field to excite hydrogen nuclei in underground fluids. After removing the external magnetic field, the relaxation of hydrogen nuclei produces a relaxation signal, which is inverted to obtain a nuclear magnetic resonance T2 spectrum containing 64 T2 amplitudes corresponding to 64 T2 wave numbers. Figure 1 As shown, Figure 1 The figure is a schematic diagram of a nuclear magnetic resonance T2 spectrum, with the horizontal axis representing the relaxation time distribution and the vertical axis representing the pore size distribution. The nuclear magnetic resonance T2 spectrum is the relaxation time distribution of hydrogen nuclei in the formation fluid at a specific formation depth, which can be used to estimate the pore size distribution and fluid mobility of the near-well formation. The pore size distribution information contained in the nuclear magnetic resonance T2 spectrum is the key to the average formation fluid pore size, and data obtained by other logging methods cannot provide this information. However, due to the complex operating procedures and slow logging speed of the nuclear magnetic resonance logging instrument, the cost of logging work is greatly increased, and it cannot be applied to wells drilled in all reservoirs.

[0051] Therefore, the present application provides a method and device for constructing a nuclear magnetic resonance T2 spectrum with physical constraints and data-driven, and the acquired conventional logging data is input into a pre-trained nuclear magnetic resonance T2 spectrum prediction model, so that the nuclear magnetic resonance T2 spectrum can be obtained without using a nuclear magnetic resonance logging instrument. The internal characteristic information of the formation in the nuclear magnetic resonance T2 spectrum can be used by mining personnel to estimate formation characteristics such as pore size distribution and fluid fluidity, thereby reducing exploration and development costs.

[0052] Figure 2 A schematic diagram of a network structure of the NMR T2 spectrum prediction model used in this application is shown in FIG. Figure 2 As shown, the prediction model 20 is composed of five network layers, wherein the fully connected layers (FC) 201 and the fully connected layers 205 are used as network input and output, and three LSTM network layers are used as network hidden layers. The LSTM network layer can handle the long-term dependency problem of data during network training, and the prediction accuracy of the network is higher. In addition, the modification of the corresponding time series information can be realized during the transmission of data in the network layer, and the restriction of the time series on a certain feature is eliminated, so that the feature learning can be affected by global training. A dropout function is added after each network layer to alleviate the occurrence of overfitting during the training process and achieve the effect of regularization. The number of neurons of the LSTM network layer 202, the LSTM network layer 203 and the LSTM network layer 204 can be reasonably set according to the input and output dimensions of the network. For example, it can be set to 30, 20 and 10, so that the training effect of the prediction model is better.

[0053] The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments may exist independently or in combination with each other, and the same or similar concepts or processes may not be described in detail in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0054] refer to Figure 3 , Figure 3 A flow chart of a method for constructing a nuclear magnetic resonance T2 spectrum by physical constraints and data-driven means provided in Example 1 of the present application. The method can be executed by a device for constructing a nuclear magnetic resonance T2 spectrum, which can be a server, dedicated equipment, personal computer or other electronic device. The method can include the following steps.

[0055] S301, obtaining target conventional logging data corresponding to the well section to be tested in the well location to be tested.

[0056] During the oil production process, the mining personnel can further analyze the geological and rock characteristics by obtaining the nuclear magnetic resonance T2 spectrum of the well section to be measured at a certain well location. The server can obtain the target conventional data corresponding to the well section to be measured in the well location to be measured. The target conventional logging data includes the following parameter groups: natural gamma value, natural potential, well diameter, deep lateral resistivity, medium lateral resistivity, shallow lateral resistivity, acoustic wave time difference, compensated neutron and logging density, etc. Each parameter group includes parameter values ​​corresponding to each depth in the well section to be measured. Exemplarily, the target conventional logging data is shown in Table 1 (the values ​​in the table are standardized values).

[0057] Table 1

[0058]

[0059] The above parameters are measured by the mining personnel using their corresponding logging tools to go deep into the well section to be tested at the fixed depth interval of the tool, and then the measured parameters are input into the server. Exemplarily, the well section to be tested is the well section from 200 meters underground to 180 meters underground. Taking natural potential as an example, the mining personnel use the natural potential measurement tool to measure in this well section at a depth interval of 0.0768 meters, that is, measure once every 0.0768 meters to obtain the natural potential data corresponding to each depth in this well section. The natural potential data corresponding to each depth can be processed in the logging data processing software to form an intuitive longitudinal natural potential curve SP. The parameters corresponding to each depth can form a longitudinal logging curve. The natural potential curve SP can be understood as a curve formed by connecting the natural potentials at each depth in this well section, such as Figure 4 shown.

[0060] S302. Perform one or more of the following processing on the target conventional logging data: data cleaning, standardization processing, and main feature analysis.

[0061] After the server obtains the target conventional logging data, it can perform one or more of the following operations on the target conventional logging data: data cleaning, standardization processing, and main feature analysis.

[0062] Data cleaning is to delete the abnormal values ​​of the parameter group in the target conventional logging data. For example, in the natural potential parameter group, the natural potential data at a depth of 181.0786 meters underground is 500, while the natural potentials at depths of 181.000 meters, 181.8646 meters and 182.6506 meters are all about 100. Therefore, the natural potential data at a depth of 181.0786 meters is considered to be an abnormal value. The corresponding parameters at a depth of 181.0786 meters in the target conventional logging data can be deleted, that is, the natural gamma value, well diameter, deep lateral resistivity, medium lateral resistivity, shallow lateral resistivity, acoustic wave time difference, compensated neutron and logging density in each parameter group at a depth of 181.0786 meters are deleted. In general, it can be seen from the intuitive vertical curve that when a logging curve has little overall volatility, but the abnormal parameters at a certain depth cause the curve to have a large volatility, it can be judged that the parameter corresponding to the depth is an abnormal value. Data cleaning can improve the accuracy of the prediction results.

[0063] Standardization is to unify the values ​​of each parameter group to between 0 and 1, which can be standardized by the following formula:

[0064]

[0065] Wherein, X is the value to be standardized, X′ is the value of X after standardization, min is the minimum value of the parameter group to which X belongs, and max is the maximum value of the parameter group.

[0066] The server can perform a main feature analysis on the target conventional logging data. The main feature analysis mainly extracts the common features of each parameter group in the target conventional logging data. For example, when the target conventional logging data includes 9 parameter groups, the main feature analysis of the target conventional logging data is performed to obtain 9 principal component data, and the principal component data is represented in the form of a matrix. After performing the main feature analysis on the target conventional logging data, the server can use the principal component information as the input of the pre-trained model for prediction. The specific implementation of the main feature analysis can refer to Example 2.

[0067] It should be noted that when the server is training the nuclear magnetic resonance T2 spectrum prediction model, if the training data is standardized and the main features are analyzed, then when the target conventional logging data is preprocessed as mentioned above, the target conventional logging data also needs to be standardized and the main features are analyzed to make the input of model training and prediction consistent.

[0068] S303, inputting the pre-processed target conventional logging data into the pre-trained nuclear magnetic resonance T2 spectrum prediction model to obtain target nuclear magnetic resonance data.

[0069] After the server obtains the target conventional logging data, it can input the target conventional logging data into the pre-trained nuclear magnetic resonance T2 spectrum prediction model for prediction to obtain target nuclear magnetic resonance data corresponding to the target conventional logging data, wherein the target nuclear magnetic resonance data includes predicted echo data in different formats.

[0070] The NMR T2 spectrum prediction model pre-trained by the server is obtained by training a preset model with training data. The specific training process and acquisition of training data are described in Example 2.

[0071] S304 , generating a target nuclear magnetic resonance T2 spectrum according to the target nuclear magnetic resonance data.

[0072] After the server predicts the target NMR data, it saves the predicted target NMR data. The server can then use the logging data processing software to perform inversion calculations on the target NMR data to generate a target NMR T2 spectrum. For example, a piece of NMR data can be composed of 32 echo data, so the generated NMR T2 spectrum is a curve graph composed of 32 points. Since the target NMR data is the data corresponding to each depth of the well section to be measured, the target NMR T2 spectrum obtained by inversion calculation is formed by the superposition of the NMR T2 spectra corresponding to each depth in the well section to be measured, so that the mining personnel can further analyze the reservoir according to the NMR T2 spectrum. For example, the porosity, rock permeability, water saturation and oil saturation can be further calculated based on the data corresponding to the NMR T2 spectrum. Different parameters correspond to different calculation formulas.

[0073] In this embodiment, the server obtains the target conventional logging data corresponding to the well section to be logged in the lateral well position, pre-processes the target conventional logging data, and then inputs the pre-processed target conventional logging data into the pre-trained nuclear magnetic resonance T2 spectrum prediction model to obtain the target nuclear magnetic resonance data corresponding to the target conventional logging data, and then generates a target nuclear magnetic resonance T2 spectrum according to the target nuclear magnetic resonance data. The target nuclear magnetic resonance T2 spectrum includes the internal characteristic information of the formation, which can be used by mining personnel to estimate the characteristics of the formation such as the pore size distribution and fluid mobility, thereby reducing the cost of exploration and development.

[0074] Example 1 describes the prediction of the nuclear magnetic resonance T2 spectrum. Before using the nuclear magnetic resonance T2 spectrum prediction model for prediction, the preset model needs to be trained to obtain the nuclear magnetic resonance T2 spectrum prediction model. The training of the preset model is explained below through Example 2.

[0075] refer to Figure 5 , Figure 5A flow chart of a method for constructing a nuclear magnetic resonance T2 spectrum by physical constraints and data-driven means provided in Example 2 of the present application. The method can be executed by a device for constructing a nuclear magnetic resonance T2 spectrum, which can be a server, dedicated equipment, personal computer or other electronic device. The method can include the following steps.

[0076] S501: Obtain training data, where the training data includes feature data and label data.

[0077] The feature data includes conventional logging data corresponding to each depth, and the label data includes nuclear magnetic resonance data corresponding to the conventional logging data.

[0078] The acquisition of training data is specifically implemented as follows:

[0079] 1. The server obtains a large amount of original conventional logging data and original nuclear magnetic resonance data.

[0080] The original conventional logging data includes the following parameter groups: natural gamma value, natural potential, borehole diameter, deep lateral resistivity, medium lateral resistivity, shallow lateral resistivity, acoustic time difference, compensated neutron, logging density, photoelectric cross-section index, density after environmental correction and compensated neutron after vertical correction. Each parameter group includes parameter values ​​corresponding to each depth. These parameter groups can be processed in the logging data processing software, and each parameter group can form a longitudinal logging curve, namely, natural gamma curve GR, natural potential curve SP, borehole diameter logging curve CAL, deep lateral resistivity curve RD, medium lateral resistivity curve RMLL, shallow lateral resistivity curve RS, acoustic time difference curve AC, compensated neutron curve CNL, density logging curve DEN, photoelectric cross-section index curve PE, density curve after environmental correction DENC and compensated neutron curve after vertical correction VCNL.

[0081] The original conventional logging data is obtained by mining personnel using the logging tools corresponding to each parameter group in the target well section of the data extraction well location according to the second fixed depth interval. The original nuclear magnetic resonance data includes raw echo data of different formats measured by mining personnel using the nuclear magnetic resonance logging instrument in the target well section of the data extraction well location according to the first fixed depth interval, such as data in the formats of DLIS, XTF and CLS. One piece of original nuclear magnetic resonance data can be composed of 32 echo data. The first fixed depth interval is not equal to the second fixed depth interval. For example, the first fixed depth interval is 0.8 meters, and the second fixed depth interval is 0.0768 meters, that is, the original nuclear magnetic resonance data is measured every 0.8 meters, and the original conventional logging data is measured every 0.0768 meters.

[0082] The original conventional logging data and original nuclear magnetic resonance data acquired by the server are obtained at different depth intervals, that is, the original conventional logging data and the original nuclear magnetic resonance data are not measured at the same depth, and the original conventional logging data and the original nuclear magnetic resonance data need to be aligned at the same depth.

[0083] Second, the server can use the logging data processing software to calculate the original conventional logging data according to the preset depth interval to obtain the calculated original conventional logging data.

[0084] 3. The server can use the logging data processing software to calculate the nuclear magnetic resonance data according to the preset depth interval to obtain the calculated nuclear magnetic resonance data. The calculated nuclear magnetic resonance data is used as label data. The logging data processing software can be CIFLOG.

[0085] The above calculation of the original nuclear magnetic resonance data and the original conventional logging data according to the preset depth interval is to align the original nuclear magnetic resonance data and the original conventional logging data in depth, that is, to obtain the original conventional logging data of the same depth and the original nuclear magnetic resonance data corresponding to the original conventional logging data. For example, the preset depth interval is set to 0.1 meter, then every 0.1 meter, a piece of original conventional logging data and the original nuclear magnetic resonance data corresponding to the original conventional logging data are calculated.

[0086] Fourth, the server can preprocess the calculated original conventional logging data to obtain the preprocessed original conventional logging data. The preprocessing includes data cleaning and / or standardization processing to make the training of the neural network more stable.

[0087] The data cleaning and standardization processing of the original conventional logging data may refer to step S302 in the first embodiment, which will not be described in detail here.

[0088] 5. The server performs grey correlation analysis on the preprocessed original conventional logging data to obtain the correlation between each parameter group in the preprocessed original conventional logging data and the natural gamma parameter group in the preprocessed original conventional logging data, and selects the parameter group with a correlation greater than or equal to a preset threshold to form conventional logging data. This can be understood as calculating the correlation between each logging curve and the natural gamma curve. The correlation is actually the degree of difference in the geometric shapes between the curves. The difference between the curves can be used as a measure of the degree of correlation.

[0089] The grey relational analysis is as follows:

[0090] The server can use the natural gamma parameter group in the original conventional logging data as the reference series for grey correlation analysis, and use other parameter groups in the original conventional logging data as comparison series. The reference series reflects the data series of the overall characteristics of the data and can be understood as the dependent variable. The comparison series is a data series composed of factors that affect data changes and can be understood as the independent variable. The reference series and the comparison series are both one.

[0091] The grey correlation coefficient ξ of the reference series and the comparison series can be calculated by the following formula i (k):

[0092]

[0093] Among them, ρ is the resolution coefficient, ρ>0, the smaller ρ is, the greater the resolution is, generally ρ is in the range of (0, 1), usually 0.5. min i min k is the second-order minimum difference, denoted as Δmin, |x0(k)-x i (k)| is the comparison sequence x i The absolute difference between each parameter value and the parameter value corresponding to the same depth on the reference sequence x0 is denoted as Δ0i(k), where i is used to distinguish each comparison sequence.

[0094] The above calculation formula can also be simplified as:

[0095]

[0096] According to formula (2), the correlation between each parameter value of the comparison series and the parameter value of the corresponding depth in the reference series can be obtained, that is, the correlation value between the parameter value of each parameter group and the parameter value of the corresponding depth in the natural gamma parameter group can be obtained. The information is too scattered to facilitate the overall comparison of the curve. The average of each correlation value can be calculated as the quantitative representation of the correlation between the comparison series and the reference series. The calculation formula of the average correlation is as follows:

[0097]

[0098] Among them, r i is the average correlation, ξ i (k) is the correlation between each comparison sequence and the reference sequence. The parameters corresponding to the comparison sequence with a preset correlation threshold can be selected to form conventional logging data.

[0099] Exemplarily, the original conventional logging data include parameters corresponding to each depth: natural potential, well diameter, deep lateral resistivity, medium lateral resistivity, shallow lateral resistivity, acoustic wave time difference, compensated neutron, logging density, photoelectric cross-section index, density after environmental correction and compensated neutron after vertical correction. The correlation between the comparison series corresponding to each parameter and the reference series is shown in Table 2.

[0100] Table 2

[0101] Comparison series Relevance Natural Potential 0.882146 Well diameter 0.873648 Deep lateral resistivity 0.865874 Middle lateral resistivity 0.759745 Shallow lateral resistivity 0.747859 Sonic Time Difference 0.732150 Compensation Neutron 0.721056 Logging density 0.672678 Photoelectric cross section index 0.564597 Density after environmental correction 0.596002 Compensated neutrons after vertical correction 0.456942

[0102] The parameter groups with correlation greater than 0.69 are selected to form conventional logging data. The conventional logging data include parameter groups: natural gamma value, natural potential, well diameter, deep lateral resistivity, medium lateral resistivity, shallow lateral resistivity, acoustic wave time difference, compensated neutron and logging density. Each parameter group includes parameter values ​​at each depth.

[0103] Optionally, the server can also perform principal component analysis (PCA) on the conventional logging data selected above to obtain principal component information corresponding to each parameter group, that is, to obtain the common features of each parameter group. The principal component information is represented in the form of a matrix to achieve dimensionality reduction processing of conventional logging data. The principal component information is formed into conventional logging data, that is, characteristic data in the training data, and the common features of the parameter groups are used for training to improve the accuracy of the model.

[0104] S502, training a preset model using training data to obtain a nuclear magnetic resonance T2 spectrum prediction model.

[0105] After the server obtains the training data, it can train the preset model through the training data. The preset model can be composed of three long short-term memory neural network LSTM layers and two fully connected layers. Before model training, the model weight initialization uses uniform distribution initialization, and the bias is initialized to 0.1. The activation function in the network model uses linear activation function, the optimization method of the training model uses stochastic gradient descent method, the optimizer uses Adam optimizer, and the model training uses batch training method. Every 20 samples form a batch, the step size is 10, and the dimension of each data is determined by the number of points of the nuclear magnetic resonance T2 spectrum. The loss function uses the mean absolute error (MAE) loss function, and the learning rate is set to 0.02. The training process makes the loss function value loss continuously decrease until it reaches the preset loss function threshold or the number of iterations reaches the set number. The preset loss function threshold can be set to 0.0001, and the number of iterations can be set to 10,000 times.

[0106] Figure 6 This is the error change diagram of the above preset model training and testing process. The horizontal axis is the number of training iterations, and the vertical axis is the mean absolute error. Figure 6 As shown in the figure, the mean absolute error decreases with the increase of the number of iterations, so that the accuracy rate gradually improves. After the training is completed, the training results can be tested using the test set, which includes conventional logging data. In the test of the training results in the figure, the mean absolute error decreases with the increase of the number of iterations. The server saves the mean absolute error and bias parameters after the preset model training to obtain the nuclear magnetic resonance T2 spectrum prediction model.

[0107] Optionally, the preset model can also be a recurrent neural network RNN. Figure 7 The following is a schematic diagram of the RNN network structure, which includes an input layer, three hidden layers, and an output layer. The number of neurons in the hidden layers is 20, 15, and 10, respectively. The weights are initialized using Gaussian initialization, and the bias is initialized to 0.1. The network training adopts the batch training method, with 50 samples forming a batch, and all batches completing one training is an iteration process. The loss function uses the mean absolute error loss function, and the learning rate is set to 0.02. The training process causes the loss function to decrease continuously until it reaches the set threshold or the number of iterations reaches the set number.

[0108] Figure 8 This is the error change diagram of the RNN preset model training and testing process. The horizontal axis is the number of training iterations, and the vertical axis is the mean absolute error. Figure 8 As shown in the figure, the mean absolute error decreases with the increase of iteration number, which also makes the accuracy gradually improve. After training, the training results can be tested using the test set, which includes conventional logging data. In the test of the training results in the figure, the mean absolute error decreases with the increase of iteration number. The server saves the RNN weight and bias parameters to obtain the nuclear magnetic resonance T2 spectrum prediction model.

[0109] The preset model composed of three long short-term memory neural network LSTM layers and two fully connected layers has a higher accuracy than the NMR T2 spectrum prediction model obtained after training the RNN preset model. The obtained NMR T2 spectrum prediction model can be used for measurement in wells where the NMR T2 spectrum needs to be measured. It only needs to obtain the conventional logging data in the well and input it into the model to obtain the predicted NMR T2 spectrum. It does not require the NMR logging instrument measurement method to estimate the pore size distribution and fluid mobility of the formation based on the predicted NMR T2 spectrum, thereby reducing the cost of exploration and development.

[0110] In this embodiment, the server obtains a training data set and trains a preset model to obtain a nuclear magnetic resonance T2 spectrum prediction model. According to the model, nuclear magnetic resonance data can be predicted for some well locations that cannot be measured by nuclear magnetic resonance, thereby obtaining a nuclear magnetic resonance T2 spectrum, providing interpretation data and theoretical basis for the formation, which can be used by mining personnel to estimate the pore size distribution and fluid fluidity and other characteristics of the formation, thereby reducing exploration and development costs.

[0111] refer to Fig. 9 , Fig. 9 This is a schematic diagram of the structure of a device for constructing a nuclear magnetic resonance T2 spectrum by physical constraints and data drive provided in Example 3 of this application. Fig. 9 As shown, the device 90 includes: an acquisition module 901, a preprocessing module 902, a prediction module 903 and a generation module 904.

[0112] The acquisition module 901 is used to acquire target conventional logging data corresponding to the well section to be logged in the well position to be logged.

[0113] The preprocessing module 902 is used to perform one or more of the following preprocessing on the target conventional logging data: data cleaning, standardization processing and main feature analysis.

[0114] The prediction module 903 is used to input the pre-processed target conventional logging data into the pre-trained nuclear magnetic resonance T2 spectrum prediction model to obtain target nuclear magnetic resonance data, which includes echo data in different formats.

[0115] The generating module 904 is used to generate a target NMR T2 spectrum according to the target NMR data.

[0116] Optionally, the target conventional logging data includes the following parameter groups: natural gamma value, natural potential, well diameter, deep lateral resistivity, medium lateral resistivity, shallow lateral resistivity, acoustic wave time difference, compensated neutron and logging density, each parameter group includes parameter values ​​corresponding to each depth in the well section to be logged.

[0117] Also includes:

[0118] Optionally, also include:

[0119] Used to obtain training data, the training data includes feature data and label data, the feature data includes conventional logging data corresponding to each depth, and the label data includes nuclear magnetic resonance data corresponding to the conventional logging data.

[0120] The preset model is trained by training data to obtain the nuclear magnetic resonance T2 spectrum prediction model.

[0121] Optionally, obtain training data, including:

[0122] A large amount of original conventional logging data and original nuclear magnetic resonance data are obtained. The original conventional logging data include the following parameter groups: natural gamma value, natural potential, well diameter, deep lateral resistivity, medium lateral resistivity, shallow lateral resistivity, acoustic wave time difference, compensated neutron, logging density photoelectric cross-section index, density after environment correction and compensated neutron after vertical correction. The original nuclear magnetic resonance data include original echo data of different formats measured by the nuclear magnetic resonance logging instrument in the target well section of the data extraction well location according to the first fixed depth interval. The original conventional logging data are measured by the logging tools corresponding to each parameter group in the target well section of the data extraction well location according to the second fixed depth interval. The first fixed depth interval is not equal to the second fixed depth interval.

[0123] The original conventional logging data is calculated according to the preset depth interval using the logging data processing software to obtain the calculated original conventional logging data.

[0124] The original nuclear magnetic resonance data are calculated according to the preset depth intervals by using the logging data processing software to obtain the calculated original nuclear magnetic resonance data, and the calculated original nuclear magnetic resonance data are used as label data.

[0125] The calculated original conventional logging data is preprocessed to obtain preprocessed original conventional logging data, and the preprocessing includes data cleaning and / or standardization processing.

[0126] A grey correlation analysis is performed on the preprocessed original conventional logging data to obtain the correlation between each parameter group in the preprocessed original conventional logging data and the natural gamma parameter group in the preprocessed original conventional logging data, and a parameter group with a correlation greater than or equal to a preset threshold is selected to form conventional logging data.

[0127] Optionally, selecting a parameter group whose correlation is greater than or equal to a preset threshold to form conventional logging data includes:

[0128] A parameter group whose correlation is greater than or equal to a preset threshold is selected.

[0129] The principal component analysis is performed on the selected parameter groups to obtain the principal component information corresponding to each parameter group, and the principal component information corresponding to each parameter group forms conventional logging data.

[0130] Optionally, the preset model consists of three long short-term memory neural network LSTM layers and two fully connected layers.

[0131] The device of this embodiment can be used to execute a method of physically constraining and data-driven construction of a nuclear magnetic resonance T2 spectrum in embodiment 1. The specific implementation method and technical effect are similar and will not be repeated here.

[0132] refer to Fig.10 , Fig.10A structural schematic diagram of an electronic device provided in Example 4 of the present application, the device includes 100: a processor 1001, a memory 1002, and a transceiver 1003. The processor 1001 executes computer execution instructions stored in the memory 1002, and controls the receiving action and the sending action of the transceiver 1003, so that at least one processor executes the method steps of physically constraining and data-driven construction of a nuclear magnetic resonance T2 spectrum in Example 1. The specific implementation method and technical effect are similar and will not be repeated here.

[0133] A computer-readable storage medium is provided in the fifth embodiment of the present application, and computer execution instructions are stored in the computer-readable storage medium. When the computer execution instructions are executed by the processor, they are used to implement the method steps of constructing a nuclear magnetic resonance T2 spectrum by physical constraints and data-driven as in the first embodiment above. The specific implementation method and technical effect are similar and will not be repeated here.

[0134] A computer program product provided in Embodiment 6 of the present invention includes a computer program. When the computer program is executed by a processor, it implements the method steps of constructing a nuclear magnetic resonance T2 spectrum by physical constraints and data-driven as in Embodiment 1 above. The specific implementation method and technical effect are similar and will not be repeated here.

[0135] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the invention disclosed herein. The present application is intended to cover any modification, use or adaptation of the present application, which follows the general principles of the present application and includes common knowledge or customary techniques in the art that are not disclosed in the present application. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present application are indicated by the following claims.

[0136] It should be understood that the present application is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.

Claims

1. A method for constructing a nuclear magnetic resonance T2 spectrum using physical constraints and data-driven methods, characterized in that: include: Obtain target conventional logging data corresponding to the well section to be tested in the well position to be tested; Performing one or more of the following preprocessing on the target conventional well logging data: data cleaning, standardization processing and main feature analysis, wherein the main feature analysis is used to extract common features of various parameter groups in the target conventional well logging data; Inputting the preprocessed target conventional logging data into a pre-trained nuclear magnetic resonance T2 spectrum prediction model to obtain target nuclear magnetic resonance data, wherein the target nuclear magnetic resonance data includes echo data in different formats; Generate a target nuclear magnetic resonance T2 spectrum according to the target nuclear magnetic resonance data; The method further comprises: Acquiring training data, including: acquiring a large amount of original conventional logging data and original nuclear magnetic resonance data, wherein the original conventional logging data includes the following parameter groups: natural gamma value, natural potential, well diameter, deep lateral resistivity, medium lateral resistivity, shallow lateral resistivity, acoustic wave time difference, compensated neutron, logging density photoelectric cross-section index, density after environment correction and compensated neutron after vertical correction, the original nuclear magnetic resonance data includes original echo data of different formats measured by a nuclear magnetic resonance logging instrument in a target well section of a data extraction well location at a first fixed depth interval, and the original conventional logging data is measured by a logging tool corresponding to each parameter group in the target well section of the data extraction well location at a second fixed depth interval, and the first fixed depth interval is not equal to the second fixed depth interval; Calculating the original conventional well logging data according to preset depth intervals using well logging data processing software to obtain calculated original conventional well logging data; Calculating the original nuclear magnetic resonance data according to the preset depth interval using logging data processing software to obtain calculated original nuclear magnetic resonance data, and using the calculated original nuclear magnetic resonance data as label data; Preprocessing the calculated original conventional logging data to obtain preprocessed original conventional logging data, wherein the preprocessing includes data cleaning and / or standardization processing; Performing grey correlation analysis on the preprocessed original conventional logging data to obtain correlations between each parameter group in the preprocessed original conventional logging data and a natural gamma parameter group in the preprocessed original conventional logging data, and selecting a parameter group whose correlation is greater than or equal to a preset threshold to form the conventional logging data; The reference series of the grey correlation analysis is the natural gamma parameter group in the original conventional logging data, and the comparison series is the other parameter groups in the original conventional logging data. The grey correlation coefficient ξ of the reference series and the comparison series is calculated. i (k): Among them, ρ is the resolution coefficient, min i min k is the second-order minimum difference, |x0(k)-x i (k)| is the comparison sequence x i The absolute difference between each parameter value and the parameter value corresponding to the same depth on the reference sequence x0, i is used to distinguish each comparison sequence.

2. The method according to claim 1, characterized in that The target conventional logging data includes the following parameter groups: natural gamma value, natural potential, well diameter, deep lateral resistivity, medium lateral resistivity, shallow lateral resistivity, acoustic wave time difference, compensated neutron and logging density, and each parameter group includes parameter values ​​corresponding to each depth in the well section to be logged.

3. The method according to claim 2, characterized in that The training data includes feature data and label data, wherein the feature data includes conventional logging data corresponding to each depth, and the label data includes nuclear magnetic resonance data corresponding to the conventional logging data; The NMR T2 spectrum prediction model is obtained by training the preset model using the training data.

4. The method according to claim 1, characterized in that: The selecting a parameter group whose correlation is greater than or equal to a preset threshold to form the conventional logging data includes: Selecting a parameter group whose correlation is greater than or equal to a preset threshold; A principal component analysis is performed on the selected parameter groups to obtain principal component information corresponding to each parameter group, and the principal component information corresponding to each parameter group forms the conventional well logging data.

5. The method according to claim 3, characterized in that: The preset model consists of three long short-term memory neural network LSTM layers and two fully connected layers.

6. A device for constructing a nuclear magnetic resonance T2 spectrum using physical constraints and data drive, characterized in that: include: An acquisition module is used to acquire target conventional logging data corresponding to the well section to be logged in the well position to be logged; A preprocessing module, used to perform one or more of the following preprocessing on the target conventional well logging data: data cleaning, standardization processing and main feature analysis, wherein the main feature analysis is used to extract common features of various parameter groups in the target conventional well logging data; A prediction module, used for inputting the pre-processed target conventional logging data into a pre-trained nuclear magnetic resonance T2 spectrum prediction model to obtain target nuclear magnetic resonance data, wherein the target nuclear magnetic resonance data includes echo data in different formats; A generating module, used for generating a target nuclear magnetic resonance T2 spectrum according to the target nuclear magnetic resonance data; The acquisition module is also used to acquire training data, including: acquiring a large amount of original conventional logging data and original nuclear magnetic resonance data, wherein the original conventional logging data includes the following parameter groups: natural gamma value, natural potential, well diameter, deep lateral resistivity, medium lateral resistivity, shallow lateral resistivity, acoustic wave time difference, compensated neutron, logging density photoelectric cross-section index, density after environment correction and compensated neutron after vertical correction, the original nuclear magnetic resonance data includes original echo data of different formats measured by the nuclear magnetic resonance logging instrument in the target well section of the data extraction well location according to the first fixed depth interval, and the original conventional logging data is measured by the logging tools corresponding to each parameter group in the target well section of the data extraction well location according to the second fixed depth interval, and the first fixed depth interval is not equal to the second fixed depth interval; Calculating the original conventional well logging data according to preset depth intervals using well logging data processing software to obtain calculated original conventional well logging data; Calculating the original nuclear magnetic resonance data according to the preset depth interval using logging data processing software to obtain calculated original nuclear magnetic resonance data, and using the calculated original nuclear magnetic resonance data as label data; Preprocessing the calculated original conventional logging data to obtain preprocessed original conventional logging data, wherein the preprocessing includes data cleaning and / or standardization processing; Performing grey correlation analysis on the preprocessed original conventional logging data to obtain correlations between each parameter group in the preprocessed original conventional logging data and a natural gamma parameter group in the preprocessed original conventional logging data, and selecting a parameter group whose correlation is greater than or equal to a preset threshold to form the conventional logging data; The reference series of the grey correlation analysis is the natural gamma parameter group in the original conventional logging data, and the comparison series is the other parameter groups in the original conventional logging data. The grey correlation coefficient ξ of the reference series and the comparison series is calculated. i (k): Among them, ρ is the resolution coefficient, min i min k is the second-order minimum difference, |x0(k)-x i (k)| is the comparison sequence x i The absolute difference between each parameter value and the parameter value corresponding to the same depth on the reference sequence x0, i is used to distinguish each comparison sequence.

7. The device according to claim 6, characterized in that The target conventional logging data includes the following parameter groups: natural gamma value, natural potential, well diameter, deep lateral resistivity, medium lateral resistivity, shallow lateral resistivity, acoustic wave time difference, compensated neutron and logging density, and each parameter group includes parameter values ​​corresponding to each depth in the well section to be logged.

8. An electronic device, characterized in that: include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the method according to any one of claims 1 to 5.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement a method for constructing a nuclear magnetic resonance T2 spectrum using physical constraints and data-driven methods as described in any one of claims 1 to 5.