Seismic attribute analysis method and device based on machine learning, equipment and medium

Abstract

The invention discloses a seismic attribute analysis method and device based on machine learning, equipment and a medium. The method comprises the following steps: establishing a training data set according to actual data of a plurality of attribute parameters, and dividing the training data set into training data and test data; sequentially adopting the training data of the single attribute to carry out machine learning on each machine to obtain a learning model of the corresponding attribute and the corresponding machine; testing and evaluating the learning model of each machine by adopting the test data of the corresponding attribute to obtain an evaluation index corresponding to each machine; and calculating a final evaluation index of the attribute. According to the method, the evaluation result of each attribute relative to each target is formed, and quantitative attribute analysis and optimization are realized.

Description

Technical Field

[0001] The present invention relates to the field of earthquake attribute analysis, and more specifically, to an earthquake attribute analysis method, device, equipment and medium based on machine learning. Background Art

[0002] Seismic exploration stimulates seismic waves through artificial methods, studies the propagation of seismic waves in underground strata, identifies the properties of underground geological structures and rock formations, and thus searches for oil and gas fields or other exploration targets. It is an important means of surveying solid resources such as oil, natural gas, and coal. Seismic reservoir prediction is based on seismic data, guided by geological principles, and constrained by drilling and logging data. It is a process of studying the spatial variation characteristics of the lithology and reservoir physical properties of oil and gas reservoirs. It is an important technology for interpreting seismic data, and the technologies it includes mainly include: seismic inversion technology, attribute analysis technology, and visualization analysis technology.

[0003] Seismic attributes help to gain insight into data, and seismic attribute extraction technology has long been the main research content of seismic special processing and interpretation. Seismic attributes refer to the geometric, kinematic, dynamic or statistical characteristics of seismic waves obtained by mathematical transformation of pre-stack or post-stack seismic data.

[0004] The earliest attribute analysis technology was to directly use bright spots (strong amplitude attributes) for hydrocarbon detection. Later, instantaneous attributes based on complex channel analysis and multi-attribute analysis based on attribute fusion appeared. Geometric attributes such as dip and azimuth and discontinuous attributes such as coherence that appeared in the 1990s can be directly used for the interpretation of stratigraphic structures and faults. Prestack seismic attributes that appeared later include AVO gradient, intercept and other attributes that can be used to analyze lithology and oil and gas properties, and azimuth AVO analysis can also be used to predict fracture reservoirs.

[0005] Seismic attribute analysis technology can extract hidden information from seismic data and transform this information into information related to lithology, physical properties or reservoir parameters, thereby guiding interpreters to correctly understand geological phenomena and increasing the application value of seismic methods. However, with the introduction of new knowledge in the fields of mathematics and information science, the seismic attributes extracted from seismic data are becoming more and more abundant. There are hundreds of seismic attributes related to time, amplitude, frequency, absorption attenuation, etc., and new attributes are constantly emerging. On the other hand, not all seismic attributes have clear geological meanings, and seismic attributes that have a one-to-one correspondence with rock geological characteristics do not exist obviously, and most effective attributes do not exist alone. In fact, the sensitivity of seismic attribute information to reflect reservoir or oil-bearing characteristics is different, and there are also large differences in the combination of seismic attributes that are sensitive to oil and gas or reservoir characteristics. Some attributes also interfere with prediction and classification, and some are related to each other or even repeated, thus generating a large amount of redundant information. On the other hand, the relationship between seismic attributes and reservoir lithology, fluid properties, and reservoir parameters is complex. Using a single attribute to analyze and predict reservoirs often leads to multiple solutions, so it is necessary to select useful information from a large number of seismic attributes. The key to the successful application of attributes is to select the most appropriate attributes. The seismic attribute analysis method is a method that uses a variety of mathematical methods to extract various seismic attributes from seismic data bodies, and combines geological, drilling, and logging data to conduct a comprehensive analysis and study of the characteristics of the target layer.

[0006] The most commonly used optimization method is to use logging data to analyze the correlation between attributes and reservoir characteristics through intersection diagrams, select the attribute with the largest correlation coefficient as the attribute most sensitive to geological conditions, and establish a certain functional relationship between the selected attribute and reservoir characteristics. This method uses the correlation coefficient, so it can be quantitative. However, this method is often performed on two variables, so there can only be one seismic attribute and one reservoir characteristic. In addition, many seismic attributes and reservoir characteristics need to be distributed one by one for intersection analysis, which is a huge workload, and the correlation coefficient also depends on the function type. Another is a multi-attribute fusion technology based on multivariate linear regression, neural network or color space. This method uses visualization analysis technology to fuse multiple attributes to characterize reservoir characteristics, such as lithology or fluid. On the one hand, this method can use multiple attributes, and on the other hand, it can also adjust the weights of each attribute when the attribute is fused. However, this method is finally explained by displaying the fused attributes in a visual way. The weight of each attribute or the optimization of the attribute is determined by interpreting the visualization map. Not only is the workload huge, but it is also very subjective.

[0007] There is still a need to develop a seismic attribute analysis method based on machine learning.

[0008] The information disclosed in the background technology section of the present invention is only intended to deepen the understanding of the general background technology of the present invention, and should not be regarded as acknowledging or suggesting in any form that the information constitutes the prior art already known to those skilled in the art. Summary of the invention

[0009] The present invention proposes a seismic attribute analysis method, device, equipment and medium based on machine learning, which form evaluation results of various attributes relative to various targets and realize quantitative attribute analysis and optimization.

[0010] In a first aspect, an embodiment of the present disclosure provides a seismic attribute analysis method based on machine learning, comprising:

[0011] Establishing a training data set according to actual data of multiple attribute parameters, and dividing the training data set into training data and test data;

[0012] Use the training data of a single attribute to perform machine learning on each machine in turn to obtain a learning model of the corresponding attribute and the corresponding machine;

[0013] Use test data with corresponding attributes to test and evaluate the learning model of each machine, and obtain the corresponding evaluation index of each machine;

[0014] Calculate the final evaluation index of the attribute.

[0015] As a specific implementation method of an embodiment of the present disclosure, the attribute parameters include longitudinal wave velocity, shear wave velocity, density logging data, longitudinal wave impedance, shear wave impedance, longitudinal and shear wave velocity ratio, λρ, μρ, Poisson's ratio, bulk modulus, shear modulus, Young's modulus, elastic impedance, porosity, saturation, and lithology.

[0016] As a specific implementation method of the embodiment of the present disclosure, the training data set is D(X ij ,Y i ), where X ij represents attribute data, i=1,2,…,N represents the data point number, j=1,2,…,P represents the variable number of the data at each data point, Y i Indicates the target.

[0017] As a specific implementation of the embodiment of the present disclosure, if the target is a non-continuous target, using test data with corresponding attributes to test and evaluate the learning model of each machine includes:

[0018] Test the trained machine learning model one by one with test data of a single attribute r to obtain the test result matrix C. The element c of the matrix kl Indicates that when the training sample X ir The real target is Y kWhen the test result is Y l probability;

[0019] Based on the test result matrix C, the following parameters are calculated for each target object:

[0020] TP k =c kk

[0021]

[0022]

[0023]

[0024] Then, an evaluation parameter is calculated when the attribute is used to predict the target, and then the sensitivity of the attribute and the target is calculated through the evaluation parameter.

[0025] As a specific implementation of the embodiment of the present disclosure, the evaluation parameters are:

[0026]

[0027]

[0028] The sensitivity levels are:

[0029]

[0030] Among them, α is a parameter.

[0031] As a specific implementation of the embodiment of the present disclosure, if the target is a continuous target, using test data with corresponding attributes to test and evaluate the learning model of each machine includes:

[0032] The trained machine learning model is tested one by one with the test data of a single attribute r.

[0033] Training sample X ir The real target is Y i When the average value is Indicates that the test result is The following evaluation parameters are used to evaluate the attributes:

[0034]

[0035]

[0036]

[0037]

[0038] AIC=N·log(MSE)+2·P

[0039] BIC = N·log(MSE)+P·log(N)

[0040] The smaller the above evaluation index is, the more sensitive the attribute is.

[0041] As a specific implementation of the embodiment of the present disclosure, the final evaluation index of the attribute is:

[0042]

[0043] Among them, J(r) is the final evaluation index of attribute r, H i (r) is the attribute r in machine H i The evaluation index obtained above, T is the total number of machines.

[0044] In a second aspect, the present disclosure also provides a seismic attribute analysis device based on machine learning, comprising:

[0045] A data set establishment module, establishing a training data set according to actual data of a plurality of attribute parameters, and dividing the training data set into training data and test data;

[0046] The learning module uses the training data of a single attribute to perform machine learning on each machine in turn to obtain a learning model of the corresponding attribute and the corresponding machine;

[0047] The testing module uses the test data with corresponding attributes to test and evaluate the learning model of each machine and obtain the evaluation index corresponding to each machine;

[0048] The calculation module calculates the final evaluation index of the attribute.

[0049] As a specific implementation method of an embodiment of the present disclosure, the attribute parameters include longitudinal wave velocity, shear wave velocity, density logging data, longitudinal wave impedance, shear wave impedance, longitudinal and shear wave velocity ratio, λρ, μρ, Poisson's ratio, bulk modulus, shear modulus, Young's modulus, elastic impedance, porosity, saturation, and lithology.

[0050] As a specific implementation method of the embodiment of the present disclosure, the training data set is D(X ij ,Y i ), where X ij represents attribute data, i=1,2,…,N represents the data point number, j=1,2,…,P represents the variable number of the data at each data point, Y i Indicates the target.

[0051] As a specific implementation of the embodiment of the present disclosure, if the target is a non-continuous target, using test data with corresponding attributes to test and evaluate the learning model of each machine includes:

[0052] Test the trained machine learning model one by one with test data of a single attribute r to obtain the test result matrix C. The element c of the matrix kl Indicates that when the training sample X ir The real target is Y k When the test result is Y l probability;

[0053] Based on the test result matrix C, the following parameters are calculated for each target object:

[0054] TP k =c kk

[0055]

[0056]

[0057]

[0058] Then, an evaluation parameter is calculated when the attribute is used to predict the target, and then the sensitivity of the attribute and the target is calculated through the evaluation parameter.

[0059] As a specific implementation of the embodiment of the present disclosure, the evaluation parameters are:

[0060]

[0061]

[0062] The sensitivity levels are:

[0063]

[0064] Among them, α is a parameter.

[0065] As a specific implementation of the embodiment of the present disclosure, if the target is a continuous target, using test data with corresponding attributes to test and evaluate the learning model of each machine includes:

[0066] The trained machine learning model is tested one by one with the test data of a single attribute r.

[0067] Training sample X ir The real target is Y i When the average value is Indicates that the test result is The following evaluation parameters are used to evaluate the attributes:

[0068]

[0069]

[0070]

[0071]

[0072] AIC=N·log(MSE)+2·P

[0073] BIC = N·log(MSE)+P·log(N)

[0074] The smaller the above evaluation index is, the more sensitive the attribute is.

[0075] As a specific implementation of the embodiment of the present disclosure, the final evaluation index of the attribute is:

[0076]

[0077] Among them, J(r) is the final evaluation index of attribute r, H i (r) is the attribute r in machine H i The evaluation index obtained above, T is the total number of machines.

[0078] In a third aspect, an embodiment of the present disclosure further provides an electronic device, the electronic device comprising:

[0079] A memory storing executable instructions;

[0080] A processor runs the executable instructions in the memory to implement the seismic attribute analysis method based on machine learning.

[0081] In a fourth aspect, an embodiment of the present disclosure further provides a computer-readable storage medium, which stores a computer program, and when the computer program is executed by a processor, the seismic attribute analysis method based on machine learning is implemented.

[0082] The methods and apparatus of the present invention have other features and advantages that will be apparent from, or will be described in detail in, the accompanying drawings and subsequent detailed descriptions incorporated herein, which together serve to explain the specific principles of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0083] The above and other objects, features and advantages of the present invention will become more apparent through a more detailed description of exemplary embodiments of the present invention in conjunction with the accompanying drawings, wherein like reference numerals generally represent like components throughout the exemplary embodiments of the present invention.

[0084] Figure 1 A flowchart showing the steps of a seismic attribute analysis method based on machine learning according to an embodiment of the present invention.

[0085] Figure 2 A schematic diagram of a test result matrix according to an embodiment of the present invention is shown.

[0086] Figure 3 A schematic diagram of a histogram of evaluation results according to an embodiment of the present invention is shown.

[0087] Figure 4 A schematic diagram of a target-attribute sensitivity matrix according to an embodiment of the present invention is shown.

[0088] Figure 5 A block diagram of a seismic attribute analysis device based on machine learning according to an embodiment of the present invention is shown.

[0089] Description of reference numerals:

[0090] 201, data set establishment module; 202, learning module; 203, testing module; 204, calculation module. DETAILED DESCRIPTION

[0091] The preferred embodiments of the present invention will be described in more detail below. Although the preferred embodiments of the present invention are described below, it should be understood that the present invention can be implemented in various forms and should not be limited to the embodiments set forth herein.

[0092] To facilitate understanding of the solutions and effects of the embodiments of the present invention, six specific application examples are given below. Those skilled in the art should understand that the examples are only for facilitating understanding of the present invention, and any specific details thereof are not intended to limit the present invention in any way.

[0093] Example 1

[0094] Figure 1 A flowchart showing the steps of a seismic attribute analysis method based on machine learning according to an embodiment of the present invention.

[0095] like Figure 1As shown, the seismic attribute analysis method based on machine learning includes: step 101, establishing a training data set according to actual data of multiple attribute parameters, and dividing the training data set into training data and test data; step 102, using the training data of a single attribute to perform machine learning on each machine in turn, and obtaining a learning model of the corresponding attribute and the corresponding machine; step 103, using the test data of the corresponding attribute to test and evaluate the learning model of each machine, and obtain the evaluation index corresponding to each machine; step 104, calculating the final evaluation index of the attribute.

[0096] In one example, the attribute parameters include P-wave velocity, S-wave velocity, density logging data, P-wave impedance, S-wave impedance, P-wave velocity ratio, λρ, μρ, Poisson's ratio, bulk modulus, shear modulus, Young's modulus, elastic impedance, porosity, saturation, and lithology.

[0097] In one example, the training dataset is D(X ij ,Y i ), where X ij represents attribute data, i=1,2,…,N represents the data point number, j=1,2,…,P represents the variable number of the data at each data point, Y i Indicates the target.

[0098] In one example, if the target is a non-continuous target, the test data with corresponding attributes is used to test and evaluate the learning model of each machine, including:

[0099] Test the trained machine learning model one by one with test data of a single attribute r to obtain the test result matrix C. The element c of the matrix kl Indicates that when the training sample X ir The real target is Y k When the test result is Y l probability;

[0100] Based on the test result matrix C, the following parameters are calculated for each target object:

[0101] TP k =c kk

[0102]

[0103]

[0104]

[0105] Then, an evaluation parameter is calculated when the attribute is used to predict the target, and then the sensitivity of the attribute and the target is calculated through the evaluation parameter.

[0106] In one example, the evaluation parameters are:

[0107]

[0108]

[0109] The sensitivity levels are:

[0110]

[0111] Among them, α is a parameter.

[0112] In one example, if the target is a continuous target, the test data with corresponding attributes is used to test and evaluate each machine learning model, including:

[0113] The trained machine learning model is tested one by one with the test data of a single attribute r.

[0114] Training sample X ir The real target is Y i When the average value is Indicates that the test result is The following evaluation parameters are used to evaluate the attributes:

[0115]

[0116]

[0117]

[0118]

[0119] AIC=N·log(MSE)+2·P

[0120] BIC = N·log(MSE)+P·log(N)

[0121] The smaller the above evaluation index is, the more sensitive the attribute is.

[0122] In one example, the final evaluation index of the attribute is:

[0123]

[0124] Among them, J(r) is the final evaluation index of attribute r, H i (r) is the attribute r in machine H i The evaluation index obtained above, T is the total number of machines.

[0125] Specifically, a training data set is established based on actual data of multiple attribute parameters, and the training data set is divided into training data and test data; the attribute parameters include P-wave velocity, S-wave velocity, density logging data, P-wave impedance, S-wave impedance, P-wave velocity ratio, λρ, μρ, Poisson's ratio, bulk modulus, shear modulus, Young's modulus, elastic impedance, porosity, saturation, and lithology.

[0126] The training data set is D(X ij ,Y i ), where X ij represents attribute data, i=1,2,…,N represents the data point number, j=1,2,…,P represents the variable number of the data at each data point, Y i Indicates the target.

[0127] The training data of a single attribute is used to perform machine learning on each machine in turn to obtain a learning model of the corresponding attribute and the corresponding machine.

[0128] Use the test data of the corresponding attributes to test and evaluate the learning model of each machine, and obtain the evaluation index corresponding to each machine; if the target is a non-continuous target, use the test data of a single attribute r to test the trained machine learning model one by one, and obtain the test result matrix C. The element c of the matrix kl Indicates that when the training sample X ir The real target is Y k When the test result is Y l probability;

[0129] Based on the test result matrix C, the following parameters are calculated for each target object:

[0130] TP k =c kk

[0131]

[0132]

[0133]

[0134] Then calculate the evaluation parameters when using this attribute to predict a target:

[0135]

[0136]

[0137] Finally, use the above evaluation parameters to score the sensitivity of the attribute to the target object, the higher the score, the better:

[0138]

[0139] Where α is a parameter, which can be 0.5, 1 or 2;

[0140] If the target is a continuous target, the trained machine learning model is tested one by one with the test data of a single attribute r. ir The real target is Y i When the average value is Indicates that the test result is The following evaluation parameters are used to evaluate the attributes:

[0141]

[0142]

[0143]

[0144]

[0145] AIC=N·log(MSE)+2·P

[0146] BIC = N·log(MSE)+P·log(N)

[0147] The smaller the above evaluation index is, the more sensitive the attribute is.

[0148] The above process can be performed on all machines one by one.

[0149] Assume there are T machines: H 1 , H 2 ,…,H T , each machine obtains the evaluation index of each attribute through the above method, where a certain attribute r is i The evaluation index obtained is H i (r), then the final evaluation index of this attribute is:

[0150]

[0151] Example 2

[0152] The present invention also provides a seismic attribute analysis device based on machine learning, comprising:

[0153] A data set establishment module establishes a training data set according to actual data of multiple attribute parameters, and divides the training data set into training data and test data;

[0154] The learning module uses the training data of a single attribute to perform machine learning on each machine in turn to obtain a learning model of the corresponding attribute and the corresponding machine;

[0155] The testing module uses the test data with corresponding attributes to test and evaluate the learning model of each machine and obtain the evaluation index corresponding to each machine;

[0156] The calculation module calculates the final evaluation index of the attribute.

[0157] In one example, the attribute parameters include P-wave velocity, S-wave velocity, density logging data, P-wave impedance, S-wave impedance, P-wave velocity ratio, λρ, μρ, Poisson's ratio, bulk modulus, shear modulus, Young's modulus, elastic impedance, porosity, saturation, and lithology.

[0158] In one example, the training dataset is D(X ij ,Y i ), where X ij represents attribute data, i=1,2,…,N represents the data point number, j=1,2,…,P represents the variable number of the data at each data point, Y i Indicates the target.

[0159] In one example, if the target is a non-continuous target, the test data with corresponding attributes is used to test and evaluate the learning model of each machine, including:

[0160] Test the trained machine learning model one by one with test data of a single attribute r to obtain the test result matrix C. The element c of the matrix kl Indicates that when the training sample X ir The real target is Y k When the test result is Y l probability;

[0161] Based on the test result matrix C, the following parameters are calculated for each target object:

[0162] TP k =c kk

[0163]

[0164]

[0165]

[0166] Then, an evaluation parameter is calculated when the attribute is used to predict the target, and then the sensitivity of the attribute and the target is calculated through the evaluation parameter.

[0167] In one example, the evaluation parameters are:

[0168]

[0169]

[0170] The sensitivity levels are:

[0171]

[0172] Among them, α is a parameter.

[0173] In one example, if the target is a continuous target, the test data with corresponding attributes is used to test and evaluate each machine learning model, including:

[0174] The trained machine learning model is tested one by one with the test data of a single attribute r.

[0175] Training sample X ir The real target is Y i When the average value is Indicates that the test result is The following evaluation parameters are used to evaluate the attributes:

[0176]

[0177]

[0178]

[0179]

[0180] AIC=N·log(MSE)+2·P

[0181] BIC = N·log(MSE)+P·log(N)

[0182] The smaller the above evaluation index is, the more sensitive the attribute is.

[0183] In one example, the final evaluation index of the attribute is:

[0184]

[0185] Among them, J(r) is the final evaluation index of attribute r, H i (r) is the attribute r in machine H i The evaluation index obtained above, T is the total number of machines.

[0186] Specifically, a training data set is established based on actual data of multiple attribute parameters, and the training data set is divided into training data and test data; the attribute parameters include P-wave velocity, S-wave velocity, density logging data, P-wave impedance, S-wave impedance, P-wave velocity ratio, λρ, μρ, Poisson's ratio, bulk modulus, shear modulus, Young's modulus, elastic impedance, porosity, saturation, and lithology.

[0187] The training data set is D(X ij ,Y i ), where X ij represents attribute data, i=1,2,…,N represents the data point number, j=1,2,…,P represents the variable number of the data at each data point, Y i Indicates the target.

[0188] The training data of a single attribute is used to perform machine learning on each machine in turn to obtain a learning model of the corresponding attribute and the corresponding machine.

[0189] Use the test data of the corresponding attributes to test and evaluate the learning model of each machine, and obtain the evaluation index corresponding to each machine; if the target is a non-continuous target, use the test data of a single attribute r to test the trained machine learning model one by one, and obtain the test result matrix C. The element c of the matrix kl Indicates that when the training sample X ir The real target is Y k When the test result is Y l probability;

[0190] Based on the test result matrix C, the following parameters are calculated for each target object:

[0191] TP k =c kk

[0192]

[0193]

[0194]

[0195] Then calculate the evaluation parameters when using this attribute to predict a target:

[0196]

[0197]

[0198] Finally, use the above evaluation parameters to score the sensitivity of the attribute to the target object, the higher the score, the better:

[0199]

[0200] Where α is a parameter, which can be 0.5, 1 or 2;

[0201] If the target is a continuous target, the trained machine learning model is tested one by one with the test data of a single attribute r. irThe real target is Y i When the average value is Indicates that the test result is The following evaluation parameters are used to evaluate the attributes:

[0202]

[0203]

[0204]

[0205]

[0206] AIC=N·log(MSE)+2·P

[0207] BIC = N·log(MSE)+P·log(N)

[0208] The smaller the above evaluation index is, the more sensitive the attribute is.

[0209] The above process can be performed on all machines one by one.

[0210] Assume there are T machines: H 1 , H 2 ,…,H T , each machine obtains the evaluation index of each attribute through the above method, where a certain attribute r is i The evaluation index obtained is H i (r), then the final evaluation index of this attribute is:

[0211]

[0212] Example 3

[0213] Data preparation: logging data and its interpretation results, pre-stack and post-stack seismic data, and various attribute data obtained based on pre-stack and post-stack seismic data, including geometric attributes such as coherence and curvature, instantaneous attributes such as instantaneous amplitude, amplitude attributes such as root mean square amplitude and amplitude change rate, and frequency attributes such as frequency division. Well logging data mainly include various elastic parameters and logging interpretation results, such as reservoir lithology, porosity, oil and gas content, and sedimentary phases.

[0214] From the well logging data, mainly the P-wave velocity V p , shear wave velocity V s , density logging data ρ, and calculate various other elastic parameters, including the following elastic parameters:

[0215] Longitudinal wave impedance: I p =V p ·ρ

[0216] Shear wave impedance: I s =V s ·ρ

[0217] P-wave velocity ratio:

[0218] λρ:

[0219] μρ:

[0220] Poisson's ratio:

[0221] Bulk modulus:

[0222] Shear modulus:

[0223] Young's modulus: E = 3·K·(1-2·σ)

[0224] Elastic impedance: in,

[0225] Of course, you can also define some calculation parameters yourself, for example:

[0226] The above logging data (including elastic parameters calculated from logging data) and their interpretation results (porosity, saturation, lithology, etc.) and the calibrated wellside seismic attribute data together constitute the training data set D(X ij ,Y i ), where X ij Represents various attribute data, where i = 1, 2, ..., N represents the data point number, j = 1, 2, ..., P represents the variable number (data dimension) of the data at each data point, and Y i Indicates the target.

[0227] Whether it is K-nearest neighbor, random forest or neural network, they are collectively referred to as machines, and the applicable machine can be any machine based on supervised learning.

[0228] The training data set is divided into two parts: training data and test data; then the machines (all available machines) are trained one by one with the training data of a single attribute r, that is, machine learning is performed.

[0229] Figure 2 A schematic diagram of a test result matrix according to an embodiment of the present invention is shown.

[0230] If the target is a non-continuous target, for all attributes, the trained machine is tested one by one with the test data of a single attribute r, and the following is formed: Figure 2 The test result matrix C is shown, and the element c of the matrix kl Indicates that when the training sample X ir The real target is Y k When the test result is Y l The probability (frequency) of . Based on the test result matrix C, the following parameters are calculated for each target object:

[0231] TP k =c kk

[0232]

[0233]

[0234]

[0235] Then calculate the evaluation parameters when using this attribute to predict a target:

[0236]

[0237]

[0238] Finally, use the above evaluation parameters to score the sensitivity of the attribute to the target object, the higher the score, the better:

[0239]

[0240] Where α is a parameter and can be 0.5, 1 or 2.

[0241] If the prediction target is a continuous object, the trained machine is tested one by one with the test data of a single attribute r. ir The real target is Y i When the average value is Indicates that the test result is If , the following evaluation parameters can be used to evaluate the attribute. The smaller these indicators are, the more sensitive the attribute is:

[0242]

[0243]

[0244]

[0245]

[0246] AIC=N·log(MSE)+2·P

[0247] BIC = N·log(MSE)+P·log(N)

[0248] Figure 3 A schematic diagram of an evaluation result histogram according to an embodiment of the present invention is shown, where the horizontal axis represents the attribute and the vertical axis represents the attribute score.

[0249] All attributes are tested and evaluated one by one, and the evaluation results are displayed in the form of a histogram, such as Figure 3 shown.

[0250] Figure 4 A schematic diagram of a target-attribute sensitivity matrix according to an embodiment of the present invention is shown.

[0251] For a specific machine, after testing, all its target classes and attribute sensitivities form a target-attribute sensitivity matrix and are displayed in the form of a matrix diagram, such as Figure 4 shown.

[0252] The above process can be performed on all machines one by one. There are T machines: H 1 , H 2 ,…,H T , each machine obtains the evaluation index of each attribute through the above method, where a certain attribute r is i The evaluation index obtained is H i (r), then the final evaluation index of this attribute is:

[0253] Example 4

[0254] Figure 5 A block diagram of a seismic attribute analysis device based on machine learning according to an embodiment of the present invention is shown.

[0255] like Figure 5 As shown, the earthquake attribute analysis device based on machine learning includes:

[0256] A data set establishment module 201 establishes a training data set according to actual data of a plurality of attribute parameters, and divides the training data set into training data and test data;

[0257] The learning module 202 sequentially uses the training data of a single attribute to perform machine learning on each machine to obtain a learning model of the corresponding attribute and the corresponding machine;

[0258] Testing module 203, using test data of corresponding attributes to test and evaluate the learning model of each machine, and obtain the evaluation index corresponding to each machine;

[0259] The calculation module 204 calculates the final evaluation index of the attribute.

[0260] As an option, the attribute parameters include P-wave velocity, S-wave velocity, density logging data, P-wave impedance, S-wave impedance, P-wave velocity ratio, λρ, μρ, Poisson's ratio, bulk modulus, shear modulus, Young's modulus, elastic impedance, porosity, saturation, and lithology.

[0261] As an alternative, the training data set is D(X ij ,Y i ), where X ij represents attribute data, i=1,2,…,N represents the data point number, j=1,2,…,P represents the variable number of the data at each data point, Y i Indicates the target.

[0262] As an optional solution, if the target is a non-continuous target, the test data with corresponding attributes is used to test and evaluate each machine learning model, including:

[0263] Test the trained machine learning model one by one with test data of a single attribute r to obtain the test result matrix C. The element c of the matrix kl Indicates that when the training sample X ir The real target is Y k When the test result is Y l probability;

[0264] Based on the test result matrix C, the following parameters are calculated for each target object:

[0265] TP k =c kk

[0266]

[0267]

[0268]

[0269] Then, an evaluation parameter is calculated when the attribute is used to predict the target, and then the sensitivity of the attribute and the target is calculated through the evaluation parameter.

[0270] As an option, the evaluation parameters are:

[0271]

[0272]

[0273] The sensitivity levels are:

[0274]

[0275] Among them, α is a parameter.

[0276] As an optional solution, if the target is a continuous target, the test data with corresponding attributes is used to test and evaluate each machine learning model, including:

[0277] The trained machine learning model is tested one by one with test data of a single attribute r. ir The real target is Y i When the average value is Indicates that the test result is The following evaluation parameters are used to evaluate the attributes:

[0278]

[0279]

[0280]

[0281]

[0282] AIC=N·log(MSE)+2·P

[0283] BIC = N·log(MSE)+P·log(N)

[0284] The smaller the above evaluation index is, the more sensitive the attribute is.

[0285] As an optional solution, the final evaluation index of the attribute is:

[0286]

[0287] Among them, J(r) is the final evaluation index of attribute r, H i (r) is the attribute r in machine H i The evaluation index obtained above, T is the total number of machines.

[0288] Example 5

[0289] This embodiment provides an electronic device, which includes: a memory storing executable instructions; and a processor running the executable instructions in the memory to implement the above-mentioned earthquake attribute analysis method based on machine learning.

[0290] An electronic device according to an embodiment of the present disclosure includes a memory and a processor.

[0291] The memory is used to store non-temporary computer-readable instructions. Specifically, the memory may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may, for example, include random access memory (RAM) and / or cache memory (cache), etc. The non-volatile memory may, for example, include read-only memory (ROM), hard disk, flash memory, etc.

[0292] The processor may be a central processing unit (CPU) or other forms of processing units having data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device to perform desired functions. In one embodiment of the present disclosure, the processor is used to run the computer-readable instructions stored in the memory.

[0293] Those skilled in the art should be able to understand that in order to solve the technical problem of how to obtain a good user experience, the present embodiment may also include well-known structures such as a communication bus and an interface, and these well-known structures should also be included in the protection scope of the present disclosure.

[0294] For detailed description of this embodiment, reference may be made to the corresponding descriptions in the aforementioned embodiments, which will not be repeated here.

[0295] Example 6

[0296] This embodiment provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the seismic attribute analysis method based on machine learning is implemented.

[0297] According to the computer-readable storage medium of the embodiment of the present disclosure, non-transitory computer-readable instructions are stored thereon. When the non-transitory computer-readable instructions are executed by a processor, all or part of the steps of the above-mentioned methods of each embodiment of the present disclosure are executed.

[0298] The above-mentioned computer-readable storage media include, but are not limited to: optical storage media (e.g., CD-ROM and DVD), magneto-optical storage media (e.g., MO), magnetic storage media (e.g., magnetic tape or mobile hard disk), media with built-in rewritable non-volatile memory (e.g., memory card) and media with built-in ROM (e.g., ROM box).

[0299] Those skilled in the art should understand that the purpose of the above description of the embodiments of the present invention is only to exemplarily illustrate the beneficial effects of the embodiments of the present invention, and is not intended to limit the embodiments of the present invention to any given examples.

[0300] The embodiments of the present invention have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments.

Claims

1. A seismic attribute analysis method based on machine learning, It is characterized in that include: Establishing a training data set according to actual data of multiple attribute parameters, and dividing the training data set into training data and test data; Use the training data of a single attribute to perform machine learning on each machine in turn to obtain a learning model of the corresponding attribute and the corresponding machine; Use test data with corresponding attributes to test and evaluate the learning model of each machine, and obtain the corresponding evaluation index of each machine; Calculate the final evaluation index of the attribute.

2. The seismic attribute analysis method based on machine learning according to claim 1, in, The attribute parameters include P-wave velocity, S-wave velocity, density logging data, P-wave impedance, S-wave impedance, P-wave velocity ratio, λρ, μρ, Poisson's ratio, bulk modulus, shear modulus, Young's modulus, elastic impedance, porosity, saturation, and lithology.

3. The earthquake attribute analysis method based on machine learning according to claim 1, in, The training data set is D(X ij ,Y i ), where X ij represents attribute data, i=1,2,…,N represents the data point number, j=1,2,…,P represents the variable number of the data at each data point, Y i Indicates the target.

4. The seismic attribute analysis method based on machine learning according to claim 3, in, If the target is a non-continuous target, the test data with corresponding attributes is used to test and evaluate the learning model of each machine, including: Test the trained machine learning model one by one with test data of a single attribute r to obtain the test result matrix C. The element c of the matrix kl Indicates that when the training sample X ir The real target is Y k When the test result is Y l The probability of Based on the test result matrix C, the following parameters are calculated for each target object: TP k =c kk Then, an evaluation parameter is calculated when the attribute is used to predict the target, and then the sensitivity of the attribute and the target is calculated through the evaluation parameter.

5. The seismic attribute analysis method based on machine learning according to claim 4, in, The evaluation parameters are: The sensitivity levels are: Among them, α is a parameter.

6. The seismic attribute analysis method based on machine learning according to claim 1, in, If the target is a continuous target, the test data with corresponding attributes is used to test and evaluate the learning model of each machine, including: The trained machine learning model is tested one by one with test data of a single attribute r. ir The real target is Y i When the average value is Indicates that the test result is The following evaluation parameters are used to evaluate the attributes: AIC=N·log(MSE)+2·P BIC = N·log(MSE)+P·log(N) The smaller the above evaluation index is, the more sensitive the attribute is.

7. The seismic attribute analysis method based on machine learning according to claim 1, in, The final evaluation index of the attribute is: Among them, J(r) is the final evaluation index of attribute r, H i (r) is the attribute r in machine H i The evaluation index obtained above, T is the total number of machines.

8. A seismic attribute analysis device based on machine learning, It is characterized in that include: A data set establishment module, establishing a training data set according to actual data of a plurality of attribute parameters, and dividing the training data set into training data and test data; The learning module uses the training data of a single attribute to perform machine learning on each machine in turn to obtain a learning model of the corresponding attribute and the corresponding machine; The testing module uses the test data with corresponding attributes to test and evaluate the learning model of each machine and obtain the evaluation index corresponding to each machine; The calculation module calculates the final evaluation index of the attribute.

9. An electronic device, It is characterized in that The electronic device comprises: A memory storing executable instructions; A processor, wherein the processor runs the executable instructions in the memory to implement the seismic attribute analysis method based on machine learning as described in any one of claims 1-7.

10. A computer-readable storage medium, It is characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the seismic attribute analysis method based on machine learning described in any one of claims 1 to 7 is implemented.

Images