A robust method for lithology identification

Abstract

The present invention discloses a robust lithology prediction method, comprising the following steps: 1) constructing a labeled sample set of existing logging data, 2) system initialization, 3) generating a kernel matrix, 4) calculating an optimal output weight matrix, and 5) predicting the lithology of a new well. Compared with the existing technology, the method has the following advantages: 1) the model parameters are analytical, so it has extremely high operating efficiency, and 2) the introduction of sparsity constraints forces the output weights to weaken the weights of secondary samples and highlight the main samples, so it has strong robustness.

Description

Technical Field

[0001] The invention relates to the fields of structural geology and petroleum geology, and in particular to a robust lithology identification method. Background Art

[0002] Lithology identification has always been a key issue in geological engineering research. The existing technologies for lithology identification mainly include geological research methods, intersection diagram methods, and lithology identification using mathematical algorithms. The methods involved are artificial experience methods that rely on expert knowledge, and the prediction results given by different experts may be inconsistent.

[0003] The machine learning methods that have emerged in recent years have injected new blood into lithology identification. For example, a method for lithology identification is disclosed in the prior art. This method first determines the target logging data file and performs format conversion and normalization preprocessing. Then, according to the known lithology of the key coring wells in the whole area in the coring section, the logging curve data is subjected to feature screening and / or feature combination expansion to obtain logging curve data sensitive to lithology. Then, the logging curve data sensitive to lithology response is labeled and calibrated to form a sample database. At the same time, the unlabeled logging curve data in the whole area is formed into a database to be tested. Then, the data in the sample database is used in combination with several machine learning algorithms for machine learning training to automatically establish several lithology identification models. The optimal model is selected through the classification performance evaluation rule, and the optimal model is used to predict the lithology of the data in the database to be tested to realize intelligent identification of complex lithology of multiple wells in the whole area. It is efficient and convenient, applicable to the whole area, and the automatic intelligent prediction results are very accurate.

[0004] However, these machine learning-based lithology identification methods do not propose specific models for lithology identification, but only apply existing general models. There are some problems in the actual application of this method. For example, the output weight of the model is weighted for all training samples, which does not reflect the value of key samples, so its robustness is often poor. Summary of the invention

[0005] The purpose of the present invention is to solve at least one of the technical problems existing in the prior art and to provide a robust lithology identification method with extremely high operating efficiency and strong robustness.

[0006] According to an embodiment of the present invention, a robust lithology identification method includes the following steps:

[0007] Step 1: Construct a labeled sample set of existing logging data, obtain existing logging data, and all logging values ​​at each depth form a feature vector, i.e., a sample. The logging values ​​include natural gamma, natural potential, well diameter, acoustic wave time difference, density, compensated neutron, deep lateral resistivity, and shallow lateral resistivity. A sample is expressed as x∈R d×1, where R represents the real number domain and d is the sample dimension; the sample set consisting of the existing logging data is Where n is the total number of samples in the set; the lithology of the existing logging sample set is marked, Lithology labeling y i For x i Corresponding lithology label, y i ∈R 1 ×c is the unique hot code of the lithology, c is the total number of lithology types; y i ∈y={y1,y2,...,y c}, y1, y2, ..., y c Indicates the 1st to cth lithologies;

[0008] Step 2: Initialize the system, set the model complexity coefficient λ>0, and set the maximum number of iterations K>0 to a positive integer;

[0009] Step 3: First, generate the kernel matrix of the basic layer in, is the kernel function of the base layer, ρ1 is the kernel coefficient of the base layer; then, the kernel matrix of the enhancement layer is generated in,

[0010]

[0011] ρ2 is the kernel coefficient of the enhancement layer; Φ1 and Φ2 are combined to obtain the kernel matrix Φ = [Φ1, Φ2];

[0012] Step 4: Calculate the optimal output weight matrix β∈R 2n×c , including the following steps:

[0013] Step 401: Let the initial value of the iteration number k be 1, let the matrix G be an n-dimensional diagonal matrix, and all diagonal elements g i is 1 / n;

[0014] Step 402: Calculate the optimal output weight matrix

[0015]

[0016] Where Y = [y1; y2; ...; y n ] is a label matrix, which means that y1, y2, ..., y n Vertical combination;

[0017] Step 403: Calculate in, Among them, ∈∈(-∞, 1), ||·||2 is the 2-norm, β iis the i-th row vector of β, i=1, 2, ..., n; the number of iterations k is increased by 1, if k≤K, jump to step 402, otherwise jump to step 5;

[0018] Step 5: Predict the lithology of the new well, form a sample set with the new logging data, and calculate the following formula:

[0019]

[0020] in, is a sample in the new sample set; then calculate the following formula:

[0021]

[0022] in, Will and Merge Then calculate for The prediction results are The index number of the maximum value in is Corresponding to the lithology number, the lithology distribution of the new logging is obtained.

[0023] Preferably, in step 403, ∈=10 -8 .

[0024] Preferably, in step 2, K∈[5,10].

[0025] The embodiments of the present invention have the following beneficial technical effects: 1) the model parameters are analytical and therefore have extremely high operating efficiency; 2) the introduction of sparsity constraints enables the output weights to weaken the weights of secondary samples and highlight the main samples, thus having strong robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The present invention is further described below in conjunction with the accompanying drawings and embodiments;

[0027] Figure 1 This is a robust lithology identification method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0028] This section will describe in detail the specific embodiments of the present invention. The preferred embodiments of the present invention are shown in the accompanying drawings. The purpose of the accompanying drawings is to supplement the description of the text part of the specification with graphics, so that people can intuitively and vividly understand each technical feature and the overall technical solution of the present invention, but it cannot be understood as a limitation on the scope of protection of the present invention.

[0029] Reference Figure 1 FIG. 1 is a flowchart of a robust lithology identification method according to an embodiment of the present invention, and the method comprises the following steps:

[0030] Step 1: Construct a labeled sample set of existing logging data, obtain existing logging data, and all logging values ​​at each depth form a feature vector, i.e., a sample. The logging values ​​include natural gamma, natural potential, well diameter, acoustic wave time difference, density, compensated neutron, deep lateral resistivity, and shallow lateral resistivity. A sample is expressed as x∈R d×1 , where R represents the real number domain and d is the sample dimension; the sample set consisting of the existing logging data is Where n is the total number of samples in the set; the lithology of the existing logging sample set is marked, Lithology labeling y i For x i Corresponding lithology label, y i ∈R 1×c is the unique hot code of the lithology, c is the total number of lithology types; y i ∈y={y1,y2,...,y c}, y1, y2, ..., y c Indicates the 1st to cth lithologies.

[0031] Step 2: Initialize the system, set the model complexity coefficient λ>0, and set the maximum number of iterations K>0 to a positive integer.

[0032] Step 3: First, generate the kernel matrix of the basic layer in, is the kernel function of the base layer, ρ1 is the kernel coefficient of the base layer; then, the kernel matrix of the enhancement layer is generated in,

[0033]

[0034] ρ2 is the kernel coefficient of the enhancement layer; Φ1 and Φ2 are combined to obtain the kernel matrix Φ = [Φ1, Φ2];

[0035] Step 4: Calculate the optimal output weight matrix β∈R 2n×c , including the following steps:

[0036] Step 401: Let the initial value of the iteration number k be 1, let the matrix G be an n-dimensional diagonal matrix, and all diagonal elements g i is 1 / n;

[0037] Step 402: Calculate the optimal output weight matrix

[0038]

[0039] Where Y = [y1; y2; ...; y n] is a label matrix, which means that y1, y2, ..., y n Vertical combination;

[0040] Step 403: Calculate in, Among them, ∈∈(-∞, 1), ||·||2 is the 2-norm, β i is the i-th row vector of β, i=1, 2, ..., n; the number of iterations k is increased by 1, if k≤K, jump to step 402, otherwise jump to step 5;

[0041] Step 5: Predict the lithology of the new well, form a sample set with the new logging data, and calculate the following formula:

[0042]

[0043] in, is a sample in the new sample set; then calculate the following formula:

[0044]

[0045] in, Will and Merge Then calculate for The prediction results are The index number of the maximum value in is Corresponding to the lithology number, the lithology distribution of the new logging is obtained.

[0046] Preferably, in step 403, ∈=10 -8 .

[0047] Preferably, in step 2, K∈[5,10].

[0048] The embodiments of the present invention are described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Various changes can be made within the knowledge scope of ordinary technicians in the technical field without departing from the purpose of the present invention.

Claims

1. A robust lithology identification method, characterized in that: The following steps are involved: Step 1: Construct a labeled sample set of existing logging data, obtain existing logging data, and all logging values ​​at each depth form a feature vector, i.e., a sample. The logging values ​​include natural gamma, natural potential, well diameter, acoustic wave time difference, density, compensated neutron, deep lateral resistivity, and shallow lateral resistivity. A sample is expressed as x∈R d×1 , where R represents the real number domain and d is the sample dimension; the sample set consisting of the existing logging data is Where n is the total number of samples in the set; the lithology of the existing logging sample set is marked, Lithology labeling y i For x i Corresponding lithology label, y i ∈R 1×c is the unique hot code of the lithology, c is the total number of lithology types; Indicates the 1st to cth lithologies; Step 2: Initialize the system, set the model complexity coefficient λ>0, and set the maximum number of iterations K>0 to a positive integer; Step 3: First, generate the kernel matrix of the basic layer in, is the kernel function of the base layer, ρ1 is the kernel coefficient of the base layer; then, the kernel matrix of the enhancement layer is generated in, ρ2 is the kernel coefficient of the enhancement layer; Φ1 and Φ2 are combined to obtain the kernel matrix Φ = [Φ1, Φ2]; Step 4: Calculate the optimal output weight matrix β∈R 2n×c , including the following steps: Step 401: Let the initial value of the iteration number k be 1, let the matrix G be an n-dimensional diagonal matrix, and all diagonal elements g i is 1 / n; Step 402: Calculate the optimal output weight matrix Where Y = [y1; y2; ...; y n ] is a label matrix, which means that y1, y2, ..., y n Vertical combination; Step 403: Calculate in, Among them, ε∈(-∞, 1), ||·||2 is the 2-norm, β i is the i-th row vector of β, i=1, 2, ..., n; the number of iterations k is increased by 1, if k≤K, jump to step 402, otherwise jump to step 5; Step 5: Predict the lithology of the new well, form a sample set with the new logging data, and calculate the following formula: in, is a sample in the new sample set; then calculate the following formula: in, Will and Merge Then calculate for The prediction results are The index number of the maximum value in is Corresponding to the lithology number, the lithology distribution of the new logging is obtained.

2. A robust lithology identification method according to claim 1, characterized in that: In step 403, ε=10 -8 .

3. A robust lithology identification method according to any one of claims 1 and 2, characterized in that: In step 2, K∈[5,10].

Images