Lithology identification method and device based on logging information and electronic equipment
By employing convolutional neural networks and Markov prior probability correction methods, the accuracy problem of complex reservoir lithology identification was solved, achieving high-precision lithology identification, which is suitable for petroleum geological research and sweet spot selection.
Patent Information
- Application Number
- CN202410578057.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-05-10
- Publication Date
- 2025-11-11
AI Technical Summary
Existing technologies are ineffective in handling complex reservoirs in lithology identification, especially lacking consideration of surrounding rocks and vertical sedimentary patterns, resulting in low identification accuracy.
A lithology prediction model based on convolutional neural networks was adopted, combined with Markov prior probability correction. Through preprocessing of sensitive well logging information and transfer learning, a lithology identification model for the target study area was established, taking into account the prior probability of vertical lithology changes, and lithology identification was carried out.
It improves the accuracy of lithology identification, especially in complex reservoirs, where the accuracy is significantly enhanced, meeting the needs of petroleum geological research and sweet spot selection.
Smart Images
Figure CN120929941A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of well logging evaluation technology in oil and gas exploration, specifically to a lithology identification method, device, and electronic equipment based on well logging data, belonging to reservoir lithology identification technology. Background Technology
[0002] Currently, the main methods for lithological interpretation using well logging data are: spider diagrams, intersection plots, multi-resolution graph clustering, and machine learning methods such as decision trees and random forests.
[0003] Specifically, spider plots are a qualitative method, primarily classifying lithology based on differences in well logging response characteristics (Zhang et al., 2016). Cross-plotting methods can only utilize partial well logging information for classifying a specific lithology, and their accuracy is insufficient for complex reservoirs (Kuang et al., 2013). Multi-resolution plotting is an unsupervised / semi-supervised method, requiring reliance on the experience of the processing personnel when merging lithologies (Mao et al., 2019). Machine learning methods are characterized by high automation and intelligence, automatically extracting the implicit petrophysical features from well logging data, and can be well used for lithology identification. However, currently commonly used machine learning-based lithology identification methods, such as decision trees, support vector machines, random forests, neural networks, and one-dimensional convolutional neural networks, require a large number of training samples and do not consider the influence of surrounding rocks and vertical sedimentary patterns on the lithology identification results at the current depth point.
[0004] In summary, lithology identification is a crucial foundation for petroleum geological research and sweet spot selection. For complex lithological reservoirs, existing technologies face challenges in lithology identification. Summary of the Invention
[0005] This invention provides a lithology identification method, apparatus, and electronic device based on well logging data to solve the aforementioned technical problem of difficulty in lithology identification in existing technologies.
[0006] According to a first aspect of the present invention, a lithology identification method based on well logging data is provided, comprising:
[0007] Based on the analysis of lithological logging response characteristics in the source study area, sensitive logging information is determined;
[0008] The well logging information of the source study area and the target study area is preprocessed, and a sample set is created;
[0009] A lithological prediction model Mo based on a convolutional neural network was established for the source study area;
[0010] Based on a small number of samples from the target study area, the lithology prediction model Mo is finely adjusted to obtain the lithology identification model Mt for the target study area.
[0011] Calculate the Markov prior probability and the transition probability matrix between different lithologies for the sample set of the target study area;
[0012] The lithological prediction model Mt was used to predict the lithological results of the test set samples in the target study area.
[0013] The predicted lithology results are corrected using the Markov prior probability to obtain the posterior probability; and the final lithology category is determined based on the posterior probability.
[0014] Preferably, preprocessing the well logging information of the source study area and the target study area, and creating a sample set includes:
[0015] Normalize curve-type information and / or perform hot-coded category information;
[0016] A database was created based on the processed information and the corresponding lithology.
[0017] Datasets are created according to specific training set specifications.
[0018] Preferably, the fine-tuning of the lithological prediction model Mo based on a small number of samples from the target study area includes:
[0019] Using the established lithology prediction model Mo as the base model, and fixing the convolutional layer, we have:
[0020]
[0021] In the formula:
[0022] y represents the probability distribution of predicted lithology in the source region;
[0023] x is the characteristic matrix;
[0024] The activation function for the convolutional layer;
[0025] η F The activation function for the fully connected layer;
[0026] w C This is the weight coefficient matrix of the convolutional layer;
[0027] b C This is the bias matrix of the convolutional layer;
[0028] w F This is the weight coefficient matrix of the fully connected layer;
[0029] b F Here is the bias matrix of the fully connected layer;
[0030]
[0031] C represents the number of categories;
[0032] By fine-tuning the base model using training set samples from the target study area, we have:
[0033]
[0034] In the formula:
[0035] y t Predict the probability distribution of lithology for the target area;
[0036] x t The feature matrix of the target region;
[0037] w F * To fine-tune the weight coefficient matrix of the fully connected layers in the model;
[0038] b F * This is for fine-tuning the bias matrix of the fully connected layers in the model.
[0039] Preferably, in the lithology prediction results of the test set samples of the target study area using the lithology prediction model Mt, the lithology likelihood probability is obtained; and / or
[0040] In the lithological results predicted using the Markov prior probability correction, the lithological likelihood probability is corrected using the Markov prior probability, resulting in:
[0041]
[0042] In the formula:
[0043] The lithology identification results are after Markov a priori correction.
[0044] y mc For lithological Markov prior probabilities;
[0045] The lithological likelihood probability distribution for the target area is given.
[0046] when If i+1 > N, then the iteration stops.
[0047] Preferably, in determining the final lithology category based on the posterior probability, the maximum a posteriori probability criterion is used to determine the final lithology category.
[0048] Preferably, the sensitive logging information includes sensitive logging curves; and / or
[0049] The sensitive logging information includes at least one of the following: natural gamma ray (GR), logarithmically processed deep inductive resistivity logging value (LogRt), neutron-density porosity difference (DeltaPHI), neutron, average density porosity (PHIND), photoelectric absorption cross section index (PE), formation position (Formation), neutron logging (CNL), density logging (DEN), acoustic logging (AC), nuclear magnetic resonance (NMR) T2 distribution, and electrical imaging logging.
[0050] According to a second aspect of the present invention, a lithology identification device based on well logging data is provided, comprising:
[0051] The analysis module is used to determine sensitive logging information based on the lithological logging response characteristics of the source study area.
[0052] The preprocessing module is used to preprocess the well logging information of the source study area and the target study area, and to create a sample set;
[0053] The model building module is used to build a lithological prediction model Mo based on a convolutional neural network for the source study area;
[0054] The model fine-tuning module is used to fine-tune the lithology prediction model Mo based on a small number of samples from the target study area to obtain the lithology identification model Mt for the target study area.
[0055] The calculation module is used to calculate the Markov prior probability and the transition probability matrix between different lithologies for the sample set of the target study area;
[0056] The prediction module is used to predict lithological results for the test set samples of the target study area using the lithological prediction model Mt; and
[0057] The determination module is used to correct the predicted lithology results using the Markov prior probability to obtain the posterior probability; and to determine the final lithology category based on the posterior probability.
[0058] Preferably, the preprocessing module includes:
[0059] The processing module is used to normalize curve-type information and / or perform hot-coded category information;
[0060] The database creation module is used to create a database based on the processed information and corresponding lithology; and
[0061] The dataset creation module is used to create datasets according to specific training set specifications.
[0062] According to a third aspect of the present invention, an electronic device is provided, comprising:
[0063] Memory; and
[0064] processor;
[0065] The memory is used to store one or more computer instructions; the one or more computer instructions are executed by the processor to implement the method described in any of the above.
[0066] According to a fourth aspect of the present invention, a readable storage medium is provided, wherein computer instructions are stored thereon; wherein, when executed by a processor, the computer instructions implement the method described in any of the preceding claims.
[0067] The technical solution of this invention enables effective lithology identification in complex lithological reservoirs with high accuracy, which is beneficial for petroleum geological research and sweet spot selection. Specifically, a lithology prediction model based on a convolutional neural network is established using source domain samples. Then, for the target area, the source domain model is fine-tuned using a small number of samples to obtain a lithology prediction model for the target area. Furthermore, by analyzing the Markov properties of the vertical lithology variation in the target area, the prior probability of lithology variation is obtained. The model prediction results are then corrected using the Markov prior probability to obtain the posterior probability, which further improves the accuracy of lithology identification. Attached Figure Description
[0068] Figure 1 This is a flowchart illustrating a lithology identification method based on well logging data in one embodiment.
[0069] Figure 2 This is an iterative flowchart of Markov prior constraints in one embodiment;
[0070] Figure 3 This is a schematic diagram of lithology identification transfer learning in one embodiment.
[0071] Figure 4 This is a flowchart of transfer learning and Markov-constrained lithology identification (i.e., a lithology identification method based on well logging data) in one embodiment;
[0072] Figure 5 This is a lithology identification result map of well W in the Panoma gas field, as shown in one embodiment.
[0073] Figure 6 This is a lithological identification result diagram of the Longtan Formation in Well S in southwestern Sichuan, as described in one embodiment.
[0074] Figure 7 This is a lithological identification result diagram of the Shihezi Formation and Shanxi Formation in well J of the Ordos Basin in one embodiment;
[0075] Figure 8-10 It is the transition probability matrix between different lithologies in different embodiments;
[0076] Figure 11This is a schematic diagram of a lithology identification device based on well logging data in one embodiment;
[0077] Figure 12 This is a schematic diagram of the structure of a preprocessing module in one embodiment;
[0078] Figure 13 This is a schematic diagram of the model fine-tuning module in one embodiment. Detailed Implementation
[0079] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.
[0080] To enable those skilled in the art to better understand the present invention, the technical solution of one embodiment will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are merely some, not all, embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort should fall within the scope of protection of the present invention.
[0081] It should be noted that the terms "first," "second," etc., used in this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be used interchangeably where appropriate for the embodiments of the invention described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.
[0082] It should be understood that when an element (such as a layer, film, region, or substrate) is described as being "on" another element, the element may be directly on the other element, or there may be an intermediate element present. Moreover, in this invention, when an element is described as being "connected" to another element, the element may be "directly connected" to the other element, or "connected" to the other element via a third element.
[0083] Example 1
[0084] Please refer to Figure 1-10 This embodiment provides a lithology identification method based on well logging data. Starting with a case study, the technical solution will be explained through comparative analysis of several application cases to facilitate further understanding of its technical content.
[0085] Case 1
[0086] The Panoma Council Grove gas field in Kansas is a carbonate oil and gas reservoir located in southwestern Kansas. The Permian Wolfcampian Council Grove Formation within the Panoma field lies on a broad, shallow shelf or slope and can be longitudinally divided into seven sub-layers containing nine lithologies. Lithology identification using this patented technology process includes the following steps:
[0087] Step 1: Analyze the lithological logging response characteristics to determine sensitive logging information. Sensitive logging information for this area includes natural gamma ray (GR), logarithmically processed deep induction resistivity logging value (LogRt), neutron-density porosity difference (DeltaPHI), average neutron and density porosity (PHIND), photoelectric absorption cross section index (PE), marine-non-marine indicator (NM_M), and formation position (Formation).
[0088] Step 2: Preprocess the well logging curves in the study area and create a sample set. This includes standardizing curve type information and performing thermal coding on category information. Based on the processed information and corresponding lithology, a database is created, and the dataset is produced according to specific training set specifications.
[0089] Step 3: Establish a lithological prediction model based on convolutional neural networks for the source study area, such as... Figure 3 As shown.
[0090] Step 4: For the sample set of the study area, calculate the Markov prior probability (Table 1) and the transition probability matrix between different lithologies in the vertical direction. Figure 8 );
[0091] Table 1 shows the proportion and steady-state probability of different lithologies in the dataset.
[0092]
[0093] Step 5: For the test set samples, use the established lithology prediction model to predict the lithology results and obtain the lithology likelihood probability, such as... Figure 5 The penultimate track (2DCNN);
[0094] Step 6: According to Figure 2 The process involves using Markov prior constraints to correct the lithology likelihood probability, obtaining the posterior probability, and then employing the maximum a posteriori probability criterion to determine the final lithology category. The results are as follows: Figure 5 The third to last track (2DCNN+MC). The lithology identification accuracy improved from 86.8% to 89.5% after Markov constraint correction, demonstrating the effectiveness of this well logging-based lithology identification method.
[0095] Case 2
[0096] The Longtan Formation in southwestern Sichuan is a transitional marine-continental reservoir, comprising marine carbonate platform deposits and bay-tidal flat deposits. It exhibits complex lithology with well-developed thin interbedded layers and is vertically classified into eight lithologies. Lithology identification using this patented technology process includes the following steps:
[0097] Step 1: Analyze the lithological logging response characteristics to determine sensitive logging information. Sensitive logging information for this region includes natural gamma ray (GR), logarithmically processed deep induction resistivity (LogRt), neutron logging (CNL), density logging (DEN), and sonic logging (AC).
[0098] Step 2: Preprocess the logging curves in the study area and create a sample set. This involves standardizing the curve information and thermally encoding the category information. A database is then created based on the processed information and the corresponding lithology. The dataset is then created according to specific training set specifications.
[0099] Step 3: Establish a lithological prediction model based on convolutional neural networks for the source study area, such as... Figure 3 As shown.
[0100] Step 4: For the sample set of the study area, calculate the Markov prior probability in the vertical direction (Table 2) and the transition probability matrix between different lithologies ( Figure 9 );
[0101] Table 2 shows the proportion and steady-state probability of different lithologies in the dataset.
[0102]
[0103] Step 5: For the test set samples, use the established lithology prediction model to predict the lithology results and obtain the lithology likelihood probability, such as... Figure 6 The third to last track (2DCNN);
[0104] Step 6: According to Figure 2 The process involves using Markov prior constraints to correct the lithology likelihood probability, obtaining the posterior probability, and then employing the maximum a posteriori probability criterion to determine the final lithology category. The results are as follows: Figure 6 The final step (2DCNN+MC) improved the lithology identification accuracy from 82.4% to 83.1% after Markov constraint correction, and also reduced the uncertainty in lithology identification, demonstrating the effectiveness of this well logging-based lithology identification method.
[0105] Case 3
[0106] The Lower Permian strata in the Hangjin Banner study area of the Ordos Basin mainly consist of a series of braided river deposits from north to south, forming an alluvial plain. The main sedimentary formations are mudstone interbedded with thin layers of medium- to fine-grained lithic sandstone, conglomerate, and gravelly coarse sandstone, with a small amount of coal in the basal Shanxi Formation. Lithological identification using the technology described in this patented technique includes the following steps:
[0107] Step 1: Analyze the lithological logging response characteristics to determine sensitive logging information. Sensitive logging information for this region includes natural gamma ray (GR), logarithmically processed deep inductive resistivity (LogRt), neutron logging (CNL), density logging (DEN), acoustic logging (AC), and photoelectric absorption cross-section index (PE).
[0108] Step 2: Preprocess the logging curves in the study area and create a sample set. This involves standardizing the curve information and thermally encoding the category information. A database is then created based on the processed information and the corresponding lithology. The dataset is then created according to specific training set specifications.
[0109] Step 3: Establish a lithological prediction model based on convolutional neural networks for the source study area, such as... Figure 3 As shown.
[0110] Step 4: Using the established model as the base model, fix the convolutional layer, and fine-tune the base model using training set samples from the target study area to obtain the lithology prediction model for the target study area.
[0111] Step 5: For the sample set of the target study area, calculate the Markov prior probability in the vertical direction (Table 3) and the transition probability matrix between different lithologies ( Figure 10 );
[0112] Table 3 shows the proportion and steady-state probability of different lithologies in the dataset.
[0113]
[0114] Step 6: For the test set samples, use the established lithology prediction model to predict the lithology results and obtain the lithology likelihood probability, such as... Figure 7 The penultimate track (TL2DCNN);
[0115] Step 7: According to Figure 2 The process involves using Markov prior constraints to correct the lithology likelihood probability, obtaining the posterior probability, and then employing the maximum a posteriori probability criterion to determine the final lithology category. The results are as follows: Figure 7 The final step (TL2DCNN+MC) improved the lithology identification accuracy from 74.5% to 87% after applying transfer learning and Markov constraint correction, demonstrating the effectiveness of this well logging-based lithology identification method.
[0116] In summary, this invention establishes a lithology identification base model using a convolutional neural network, and then uses transfer learning and Markov constraints to establish a lithology prediction model for complex reservoirs that considers geological constraints. This enables effective lithology identification with high accuracy.
[0117] Example 2
[0118] Please refer to Figure 1-4 This embodiment provides a lithology identification method based on well logging data. This method can be summarized from the case of Embodiment 1, and includes the following steps:
[0119] S1. Based on the analysis of lithological logging response characteristics in the source study area, sensitive logging information is determined; the sensitive logging information includes sensitive logging curves; further, the sensitive logging information includes natural gamma ray (GR), logarithmically processed deep induction resistivity logging value LogRt, neutron-density porosity difference DeltaPHI, neutron and density porosity average PHIND, photoelectric absorption cross section index PE, formation position, neutron logging CNL, density logging DEN, acoustic logging AC, and nuclear magnetic resonance T2 fraction. At least one of the following: nuclear magnetic resonance T2 distribution, electrical imaging logging, etc.; among them, nuclear magnetic resonance T2 distribution and electrical imaging logging are special logging data; based on the natural gamma ray GR, logarithmically processed deep induction resistivity logging value LogRt, neutron-density porosity difference DeltaPHI, neutron, density porosity average PHIND, photoelectric absorption cross section index PE, formation position, neutron logging CNL, density logging DEN, and sonic logging AC, the constructed curves can be processed by gradient, wavelet transform, difference, quotient, etc.
[0120] S2. Preprocess the well logging information of the source study area and the target study area, and create a sample set;
[0121] S3. Establish a lithology prediction model Mo based on convolutional neural networks for the source study area;
[0122] S4. Based on a small number of samples from the target study area, fine-tune the lithology prediction model Mo to obtain the lithology identification model Mt for the target study area;
[0123] S5. Calculate the Markov prior probability and the transition probability matrix between different lithologies for the sample set of the target study area;
[0124] S6. Use the lithology prediction model Mt to predict the lithology results for the test set samples of the target study area; this step can obtain the lithology likelihood probability.
[0125] S7. The predicted lithology results are corrected using the Markov prior probability (based on step S6, specifically the corrected lithology likelihood probability), to obtain the posterior probability; wherein,
[0126]
[0127] In the formula:
[0128] The lithology identification results are after Markov a priori correction.
[0129] y mc For lithological Markov prior probabilities;
[0130] The lithological likelihood probability distribution for the target area is given.
[0131] when If i+1 > N, then the iteration stops.
[0132] The final lithology category is determined based on the posterior probability. Preferably, the final lithology category is determined using the maximum a posteriori probability criterion; such as... Figure 2 , Figure 3 .
[0133] In one embodiment, step S2 includes the following steps:
[0134] S21. Normalize curve-type information and / or perform hot-coded category information;
[0135] S22. Create a database based on the processed information and the corresponding lithology;
[0136] S23. Create datasets according to specific training set specifications.
[0137] In one embodiment, step S4 includes the following steps:
[0138] S41. Using the established lithology prediction model Mo as the base model, and fixing the convolutional layer, we have:
[0139]
[0140] In the formula:
[0141] y represents the probability distribution of predicted lithology in the source region;
[0142] x is the characteristic matrix;
[0143] The activation function for the convolutional layer;
[0144] η F The activation function for the fully connected layer;
[0145] w C This is the weight coefficient matrix of the convolutional layer;
[0146] b C This is the bias matrix of the convolutional layer;
[0147] w F This is the weight coefficient matrix of the fully connected layer;
[0148] b F Here is the bias matrix of the fully connected layer;
[0149]
[0150] C represents the number of categories;
[0151] S42. The base model is fine-tuned using training set samples from the target study area, resulting in:
[0152]
[0153] In the formula:
[0154] y t Predict the probability distribution of lithology for the target area;
[0155] x t The feature matrix of the target region;
[0156] w F * To fine-tune the weight coefficient matrix of the fully connected layers in the model;
[0157] b F * This is for fine-tuning the bias matrix of the fully connected layers in the model.
[0158] The lithology identification method based on well logging data of this invention establishes a lithology identification base model (i.e., lithology prediction model Mo) using a convolutional neural network, and then establishes a lithology prediction model Mo for complex reservoirs considering geological constraints using transfer learning and Markov constraints. This method enables effective lithology identification with high accuracy.
[0159] Example 3
[0160] Please refer to Figure 11-13 This embodiment provides a lithology identification device based on well logging data, which includes the following modules.
[0161] 1. Analysis Module
[0162] Analysis module 10 is used to determine sensitive logging information based on the lithological logging response characteristics of the source study area.
[0163] 2. Preprocessing module
[0164] The preprocessing module 20 is used to preprocess the well logging information of the source study area and the target study area, and to create a sample set;
[0165] 3. Model Building Module
[0166] The model building module 30 is used to build a lithological prediction model Mo based on a convolutional neural network for the source study area;
[0167] 4. Model fine-tuning module
[0168] Model fine-tuning module 40 is used to fine-tune the lithology prediction model Mo based on a small number of samples from the target study area to obtain the lithology identification model Mt for the target study area;
[0169] 5. Calculation Module
[0170] Calculation module 50 is used to calculate the Markov prior probability and the transition probability matrix between different lithologies for the sample set of the target study area;
[0171] 6. Prediction Module
[0172] Prediction module 60 is used to predict lithological results for the test set samples of the target study area using the lithological prediction model Mt;
[0173] 7. Determine the module
[0174] The determination module 70 is used to correct the predicted lithology results using the Markov prior probability to obtain the posterior probability; and to determine the final lithology category based on the posterior probability.
[0175] In one embodiment, the preprocessing module 20 includes a processing module 201, a database creation module 202, and a dataset creation module 203. The processing module 201 is used to normalize curve-type information and / or perform hot-coded category information; the database creation module 202 is used to create a database based on the processed information and corresponding lithology; and the dataset creation module 203 is used to create a dataset according to specific training set specifications.
[0176] In one embodiment, the model fine-tuning module 40 includes a fixing module 401 and a base model fine-tuning module 402. The fixing module 401 is used to fix the convolutional layer using the established lithology prediction model Mo as the base model; the base model fine-tuning module 402 is used to fine-tune the base model using training set samples from the target study area.
[0177] It should be noted that the lithology identification device based on well logging data described above is used to implement the lithology identification method based on well logging data in the above embodiments, and each module in the device corresponds to each step in the method.
[0178] Example 4
[0179] Based on the same inventive concept, one embodiment of the present invention provides an electronic device, including: a memory and a processor; wherein the memory is used to store one or more computer instructions; the one or more computer instructions are executed by the processor using any of the methods described in the above embodiments.
[0180] Example 5
[0181] Based on the same inventive concept, one embodiment of the present invention provides a readable storage medium storing computer instructions; wherein, when the computer instructions are executed by a processor, they implement the method of any one of the above embodiments.
[0182] One or more of the aforementioned computer instructions can form a program.
[0183] The aforementioned program can run on a processor or be stored in memory (or computer-readable medium). Computer-readable medium includes both permanent and non-permanent, removable and non-removable media, and information storage can be achieved by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable medium does not include transient computer-readable media, such as modulated data signals and carrier waves.
[0184] These computer programs may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes can be implemented using different modules, and different steps can be implemented using different modules.
[0185] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A lithology identification method based on well logging data, characterized in that, include: Based on the analysis of lithological logging response characteristics in the source study area, sensitive logging information is determined; The well logging information of the source study area and the target study area is preprocessed, and a sample set is created; A lithological prediction model Mo based on a convolutional neural network was established for the source study area; Based on a small number of samples from the target study area, the lithology prediction model Mo is finely adjusted to obtain the lithology identification model Mt for the target study area. Calculate the Markov prior probability and the transition probability matrix between different lithologies for the sample set of the target study area; The lithological prediction model Mt was used to predict the lithological results of the test set samples in the target study area. The predicted lithology results are corrected using the Markov prior probability to obtain the posterior probability; and the final lithology category is determined based on the posterior probability.
2. The lithology identification method based on well logging data according to claim 1, characterized in that, Preprocessing the well logging information of the source and target study areas and creating a sample set includes: Normalize curve-type information and / or perform hot-coded category information; A database was created based on the processed information and the corresponding lithology. Datasets are created according to specific training set specifications.
3. The lithology identification method based on well logging data according to claim 1, characterized in that, The fine-tuning of the lithological prediction model Mo based on a small number of samples from the target study area includes: Using the established lithology prediction model Mo as the base model, and fixing the convolutional layer, we have: In the formula: y represents the probability distribution of predicted lithology in the source region; x is the characteristic matrix; The activation function for the convolutional layer; η F The activation function for the fully connected layer; w C This is the weight coefficient matrix of the convolutional layer; b C This is the bias matrix of the convolutional layer; w F This is the weight coefficient matrix of the fully connected layer; b F Here is the bias matrix of the fully connected layer; C represents the number of categories; By fine-tuning the base model using training set samples from the target study area, we have: In the formula: y t Predict the probability distribution of lithology for the target area; x t The feature matrix of the target region; w F * To fine-tune the weight coefficient matrix of the fully connected layers in the model; b F * This is for fine-tuning the bias matrix of the fully connected layers in the model.
4. The lithology identification method based on well logging data according to claim 1, characterized in that, In the lithology prediction results obtained by using the lithology prediction model Mt on the test set samples of the target study area, the lithology likelihood probability is obtained; and / or In the lithological results predicted using the Markov prior probability correction, the lithological likelihood probability is corrected using the Markov prior probability, resulting in: In the formula: y t* The lithology identification results are after Markov a priori correction. y mc For lithological Markov prior probabilities; The lithological likelihood probability distribution for the target area is given. when If i+1 > N, then the iteration stops.
5. The lithology identification method based on well logging data according to claim 1, characterized in that, In determining the final lithology category based on the posterior probability, the maximum a posteriori probability criterion is used to determine the final lithology category.
6. The lithology identification method based on well logging data according to any one of claims 1-5, characterized in that, The sensitive logging information includes sensitive logging curves; and / or The sensitive logging information includes at least one of the following: natural gamma ray (GR), logarithmically processed deep inductive resistivity logging value (LogRt), neutron-density porosity difference (DeltaPHI), neutron, average density porosity (PHIND), photoelectric absorption cross section index (PE), formation position (Formation), neutron logging (CNL), density logging (DEN), acoustic logging (AC), nuclear magnetic resonance (NMR) T2 distribution, and electrical imaging logging.
7. A lithology identification device based on well logging data, characterized in that, include: The analysis module is used to determine sensitive logging information based on the lithological logging response characteristics of the source study area. The preprocessing module is used to preprocess the well logging information of the source study area and the target study area, and to create a sample set; The model building module is used to build a lithological prediction model Mo based on a convolutional neural network for the source study area; The model fine-tuning module is used to fine-tune the lithology prediction model Mo based on a small number of samples from the target study area to obtain the lithology identification model Mt for the target study area. The calculation module is used to calculate the Markov prior probability and the transition probability matrix between different lithologies for the sample set of the target study area; The prediction module is used to predict lithological results for the test set samples of the target study area using the lithological prediction model Mt. and The determination module is used to correct the predicted lithology results using the Markov prior probability to obtain the posterior probability; and to determine the final lithology category based on the posterior probability.
8. The lithology identification device based on well logging data according to claim 7, characterized in that, The preprocessing module includes: The processing module is used to normalize curve-type information and / or perform hot-coded category information; The database creation module is used to create a database based on the processed information and corresponding lithology; and The dataset creation module is used to create datasets according to specific training set specifications.
9. An electronic device, characterized in that, include: Memory; and processor; The memory is used to store one or more computer instructions; the one or more computer instructions are executed by the processor to implement the method according to any one of claims 1 to 6.
10. A readable storage medium, characterized in that, The readable storage medium stores computer instructions; wherein, when executed by a processor, the computer instructions implement the method described in any one of claims 1 to 6.