Method and system for predicting pore throat structure by fusing various logging curve data

Abstract

The invention relates to a method and a system for predicting a pore throat structure by fusing various logging curve data, and relates to the field of structure prediction.A pore throat structure topology library is constructed, cross plot analysis of various conventional logging curves is combined, lithology distribution probability is predicted by utilizing a machine learning model, and pore throat structure topology and lithology probability distribution are fused, so that the lithology distribution probability is predicted. According to the method, the probability distribution matrix of the pore throat structure is accurately calculated on each depth point along the well track, and finally verification and optimization are performed through nuclear magnetic logging data, so that the technical problems of difficulty in accurate and efficient prediction of the reservoir pore throat structure, high oil-gas exploration cost and low efficiency are solved, accurate prediction of the reservoir pore throat structure is realized, and the prediction efficiency of the reservoir pore throat structure is improved. And the efficiency and the accuracy of oil-gas exploration and development are improved.

Description

Technical Field

[0001] The present invention relates to the field of structural prediction, and particularly to a method and system for predicting pore-throat structure by integrating multiple logging curve data. Background Art

[0002] In the field of oil and gas exploration and development, the pore-throat structure of a reservoir is one of the important parameters for evaluating reservoir quality, predicting oil and gas production capacity, and formulating development plans. The pore-throat structure refers to the distribution, morphology, size, and connectivity of pores and throats in the reservoir, which directly affects the seepage ability of oil and gas and the storage performance of the reservoir. Therefore, accurately obtaining the pore-throat structure information of the reservoir is of great significance for improving the success rate of oil and gas exploration and optimizing the development plan.

[0003] Traditionally, the methods for obtaining the pore-throat structure of a reservoir mainly include two categories: laboratory analysis and logging technology. Laboratory analysis, such as mercury injection method, nuclear magnetic resonance (NMR) method, etc., can provide high-precision pore-throat structure information, but these methods are usually limited to core samples, difficult to be extended to the entire oil and gas reservoir, and have high costs and long time consumption. Logging technology, especially nuclear magnetic logging, although it can obtain continuous pore-throat structure information in the wellbore, its equipment is expensive, the operation is complex, and not all wells will perform nuclear magnetic logging, so the data coverage is limited.

[0004] In recent years, with the rapid development of machine learning technology, its application in geological data processing and interpretation has become increasingly widespread. Machine learning algorithms can process complex data relationships, mine potential laws in data, and provide new ideas and methods for solving geological problems. In the prediction of reservoir pore-throat structure, conventional logging curves, such as natural gamma, resistivity, acoustic travel time, etc., have the advantages of low cost and wide coverage. They are widely used in oil and gas exploration and development and mainly reflect the basic characteristics of the reservoir, such as lithology, physical properties, and electrical properties. If the conventional logging data can be correlated with the pore-throat structure information through an effective method, the efficiency and accuracy of pore-throat structure prediction will be greatly improved. Thus, using machine learning algorithms to integrate multiple logging curve data, combined with laboratory analysis and historical sample information, is expected to achieve accurate and efficient prediction of the reservoir pore-throat structure. At present, there is no mature method that can comprehensively integrate multiple logging curve data, combine laboratory analysis and historical sample information to predict the pore-throat structure of the reservoir. Therefore, developing a method for predicting pore-throat structure by integrating multiple logging curve data is of great significance for improving the efficiency of oil and gas exploration and development and reducing development costs. Summary of the Invention

[0005] Aiming at the technical problems in the prior art that it is difficult to accurately and efficiently predict the pore-throat structure of a reservoir, and the high cost and low efficiency of oil and gas exploration, the present invention provides a method and system for predicting pore-throat structure by integrating multiple logging curve data to solve the problems.

[0006] The technical solution of the present invention to solve the above technical problems is as follows: The present invention provides a method for predicting pore-throat structure by fusing various logging curve data, including: constructing a pore-throat structure topology library based on historical samples according to pore-throat structure test data, wherein any pore-throat structure topology in the pore-throat structure topology library has a lithology type identifier; obtaining multiple lithology logging response characteristics by using cross-plot analysis in combination with several attribute conventional logging curves; processing the multiple lithology logging response characteristics according to a lithology distribution probability prediction model, and outputting a lithology joint probability distribution matrix, wherein the lithology distribution probability prediction model is generated by machine learning training using multiple groups of data, and any group of the multiple groups of data includes a lithology logging response characteristic record value and a label identifying the lithology joint probability distribution matrix; based on multiple lithology types, comparing with the lithology type identifier, extracting pore-throat structure topologies from the pore-throat structure topology library, generating a pore-throat structure distribution matrix, and taking the matrix dot product at each depth point along the wellbore trajectory in combination with the lithology joint probability distribution matrix to obtain a pore-throat structure probability distribution matrix; verifying and optimizing the pore-throat structure probability distribution matrix through nuclear magnetic logging data to obtain a target pore-throat structure.

[0007] In a second aspect, the present invention provides a system for predicting pore-throat structure by fusing various logging curve data. The system includes: a pore-throat structure topology library construction module for constructing a pore-throat structure topology library based on historical samples according to pore-throat structure test data, wherein any pore-throat structure topology in the pore-throat structure topology library has a lithology type identifier; a cross-plot analysis module for obtaining multiple lithology logging response characteristics by using cross-plot analysis in combination with several attribute conventional logging curves; a model prediction module for processing the multiple lithology logging response characteristics according to a lithology distribution probability prediction model and outputting a lithology joint probability distribution matrix, wherein the lithology distribution probability prediction model is generated by machine learning training using multiple groups of data, and any group of the multiple groups of data includes a lithology logging response characteristic record value and a label identifying the lithology joint probability distribution matrix; a comparison and analysis module for comparing with the lithology type identifier based on multiple lithology types, extracting pore-throat structure topologies from the pore-throat structure topology library, generating a pore-throat structure distribution matrix, and taking the matrix dot product at each depth point along the wellbore trajectory in combination with the lithology joint probability distribution matrix to obtain a pore-throat structure probability distribution matrix; a verification and optimization module for verifying and optimizing the pore-throat structure probability distribution matrix through nuclear magnetic logging data to obtain a target pore-throat structure.

[0008] The beneficial effects of the present invention are as follows: By constructing a pore-throat structure topology library, combining the cross-plot analysis of multiple conventional logging curves, using a machine learning model to predict the lithology distribution probability, and fusing the pore-throat structure topology with the lithology probability distribution, the pore-throat structure probability distribution matrix is accurately calculated at each depth point along the wellbore trajectory. Finally, through verification and optimization using nuclear magnetic logging data, the accurate prediction of the reservoir pore-throat structure is achieved, improving the efficiency and accuracy of oil and gas exploration and development. Description of the Drawings

[0009] Figure 1 It is a schematic flow chart of a method for predicting pore-throat structure by fusing multiple logging curve data provided by the present invention.

[0010] Figure 2 It is a schematic flow chart of constructing a pore-throat structure topology library in a method for predicting pore-throat structure by fusing multiple logging curve data provided by the present invention.

[0011] Figure 3 It is a schematic structural diagram of a system for predicting pore-throat structure by fusing multiple logging curve data provided by the present invention.

[0012] Description of the reference numerals: Pore-throat structure topology library construction module 11, cross-plot analysis module 12, model prediction module 13, comparison and analysis module 14, verification and optimization module 15. Detailed Embodiments

[0013] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative efforts fall within the scope of protection of the present invention.

[0014] In the description of the present invention, the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of the described features. In the description of the present invention, "a plurality" means two or more, unless otherwise specifically defined.

[0015] In the description of the present invention, the term "for example" is used to mean "serving as an example, illustration, or explanation". Any embodiment described as "for example" in the present invention is not necessarily construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to implement and use the present invention. In the following description, details are set forth for purposes of explanation. It should be understood that those of ordinary skill in the art can recognize that the present invention can be implemented without using these specific details. In other instances, well-known structures and processes are not elaborated in detail to avoid obscuring the description of the present invention with unnecessary details. Therefore, the present invention is not intended to be limited to the embodiments shown, but rather to be in line with the broadest scope consistent with the principles and features disclosed in the present invention.

[0016] Embodiment 1:

[0017] As Figure 1 shown, an embodiment of the present invention provides a method for predicting pore-throat structure by fusing multiple logging curve data, including: S10: According to the pore-throat structure test data, construct a pore-throat structure topology library based on historical samples, wherein any pore-throat structure topology in the pore-throat structure topology library has a lithology type identifier.

[0018] S20: Combine several attribute conventional logging curves and use cross-plot analysis to obtain multiple lithology logging response characteristics.

[0019] S30: According to the lithology distribution probability prediction model, process the multiple lithology logging response characteristics and output a lithology joint probability distribution matrix, wherein the lithology distribution probability prediction model is generated by machine learning training using multiple groups of data, and any one of the multiple groups of data includes a lithology logging response characteristic record value and a label identifying the lithology joint probability distribution matrix.

[0020] S40: Based on multiple lithology types, compare with the lithology type identifier, extract pore-throat structure topologies from the pore-throat structure topology library, generate a pore-throat structure distribution matrix, and at each depth point along the wellbore trajectory, take the matrix dot product in combination with the lithology joint probability distribution matrix to obtain a pore-throat structure probability distribution matrix.

[0021] S50: Verify and optimize the pore-throat structure probability distribution matrix through nuclear magnetic logging data to obtain the target pore-throat structure.

[0022] Exemplarily, in the process of constructing the pore-throat structure topology library, a large number of pore-throat structure test data are first collected. These data mainly come from mercury injection curves or laboratory nuclear magnetic resonance analysis experimental results, which accurately reflect the pore-throat structure characteristics of different lithologic reservoirs. Subsequently, the pore-throat structure characteristics are systematically analyzed and extracted based on these historical samples. Specifically, the collected pore-throat structure test data are preprocessed to ensure the accuracy and consistency of the data. Subsequently, for each pore-throat structure characteristic, feature extraction techniques are used to identify and quantify its key parameters, such as pore radius distribution, throat connectivity, etc. These characteristics form the basis of the pore-throat structure topology. To establish the pore-throat structure topology library, the extracted pore-throat structure characteristics are associated with the corresponding lithologic types. Each pore-throat structure topology is assigned a unique lithologic type identifier, which ensures that the pore-throat structure can be accurately matched with a specific lithology during subsequent prediction. For example, for a reservoir with a specific lithology, its pore-throat structure topology may exhibit characteristics such as the pore radius distribution being concentrated in a specific range and good throat connectivity. These characteristics and their corresponding lithologic type identifiers will be stored in the pore-throat structure topology library together. In this way, a library containing multiple pore-throat structure topologies is constructed, and each topology is associated with its specific lithologic type. This library provides a basis for predicting the pore-throat structure of reservoirs using various logging curve data subsequently, enabling a more accurate understanding and prediction of the pore-throat structure characteristics of different lithologic reservoirs.

[0023] Meanwhile, multiple lithologic logging response characteristics are obtained. In this process, the conventional logging curves of several attributes are first analyzed in depth. These conventional logging curves, such as natural gamma, resistivity, acoustic travel time, etc., each contain rich geological information and are important indicators reflecting various characteristics of the reservoir, such as lithology, physical properties, oil-bearing properties, etc. The classic geostatistical method of crossplot analysis is used to cross two or more logging curves in the same coordinate system to construct a two-dimensional or higher-dimensional scatter plot. For example, the natural gamma curve can be crossed with the resistivity curve, and by observing the distribution characteristics of different lithologic samples in the scatter plot, the unique patterns of different lithologies in the logging response can be identified. Shale may show a combination of low natural gamma values and high resistivity values, while sandstone may show the opposite trend. Through the analysis of these crossplot scatter plots, the logging response characteristics of multiple lithologies can be extracted. These characteristics not only include the specific numerical ranges of the logging curves but also cover statistical information such as the distribution form and density of different lithologic samples in the crossplot space. These characteristics provide valuable data support for predicting the lithology distribution probability using machine learning models subsequently, enabling a more accurate understanding of the distribution laws of different lithologies in the reservoir.

[0024] Subsequently, a lithology distribution probability prediction model is constructed. Multiple groups of data are collected, and each group of data contains the recorded values of lithology logging response characteristics and the corresponding lithology joint probability distribution matrix labels. These data are obtained based on a large number of actual logging data through manual annotation or historical sample analysis, and they provide the basis for model training. Next, machine learning algorithms such as the Gaussian mixture model (GMM), kernel density estimation (KDE), etc. are selected as the training framework for the model. These algorithms can effectively handle high-dimensional data and complex probability distributions, and are suitable for processing multi-dimensional and non-linear geological data such as logging curves. In the model training stage, the recorded values of lithology logging response characteristics are used as input features, and the lithology joint probability distribution matrix labels are used as output targets. The parameters of the model are continuously adjusted through iterative optimization algorithms, enabling the model to accurately learn and simulate the complex relationship between lithology distribution and logging response characteristics. For example, for a specific combination of logging curves, the model may learn the joint occurrence probabilities of different lithologies such as mudstone and sandstone within a specific logging response range. After sufficient training, a lithology distribution probability prediction model is obtained. This model can accept new lithology logging response characteristics as input and quickly output the lithology joint probability distribution matrix corresponding to this feature. The output matrix intuitively shows the probability of the coexistence of different lithologies under the given logging response, providing key lithology probability distribution information for subsequent pore-throat structure prediction. Through this method, the distribution law of lithology in the reservoir can be understood and predicted more accurately, providing strong support for oil and gas exploration and development.

[0025] In the process of specifically predicting the pore-throat structure of a reservoir, multiple lithology types are compared with the lithology type identifiers in the pore-throat structure topology library. The pore-throat structure topology library is constructed based on historical samples, where each pore-throat structure topology is associated with a specific lithology type identifier, providing a rich pore-throat structure information library. Through comparison, the pore-throat structure topology that matches the lithology type encountered in the current wellbore trajectory can be extracted from the pore-throat structure topology library. For example, if the lithology at the current depth point is identified as sandstone, the pore-throat structure topology corresponding to sandstone is extracted from the topology library, and this topology details the pore and throat distribution characteristics of the sandstone reservoir. Next, a pore-throat structure distribution matrix is generated based on the extracted pore-throat structure topology, which quantifies the probability of different pore and throat sizes occurring in the sandstone reservoir. At the same time, using the lithology distribution probability prediction model, the lithology joint probability distribution matrix at the current depth point is calculated based on the logging curve data, which reflects the probability of various lithologies coexisting under the current logging response. Finally, a matrix dot product operation is performed on the pore-throat structure distribution matrix and the lithology joint probability distribution matrix. The core lies in combining the pore-throat structure characteristics with the lithology probability distribution, thereby calculating the pore-throat structure probability distribution matrix at each depth point along the wellbore trajectory. This matrix not only considers the pore-throat structure characteristics of different lithologies but also incorporates the lithology probability distribution information at the current depth point, thus being able to more comprehensively reflect the actual situation of the reservoir pore-throat structure. In this way, the pore-throat structure of the reservoir can be predicted more accurately, providing a strong geological basis for subsequent oil and gas exploration and development.

[0026] After obtaining the pore-throat structure probability distribution matrix, in order to further verify its accuracy and seek the optimal solution, nuclear magnetic logging data is introduced for comparative verification. As a high-precision logging method, nuclear magnetic logging can directly measure the size distribution of pores and throats in the reservoir, so it is regarded as the standard for verifying the prediction results of pore-throat structure. The specific process is as follows: First, analyze the nuclear magnetic logging data to extract the pore-throat structure curve array data set, which details the actual distribution of pores and throats at different depth points in the reservoir. Then, compare and analyze these actual data with the pore-throat structure probability distribution matrix predicted by multi-source logging curve fusion before, and calculate the sum of squared residuals between the two as a quantitative index of the fitness of pore-throat structure prediction. Next, set a fitness threshold to judge the quality of the prediction results. If the fitness of pore-throat structure prediction is greater than or equal to this threshold, it means that there is a large deviation between the prediction results and the actual nuclear magnetic logging data. At this time, start the model parameter update process to optimize the prediction results by adjusting the parameters of the lithology distribution probability prediction model. This process may go through multiple iterations until a set of model parameters is found such that the prediction fitness is lower than the threshold. Finally, when the fitness of pore-throat structure prediction is less than the fitness threshold, it is considered that the pore-throat structure probability distribution matrix at this time is close enough to the actual nuclear magnetic logging data, so it is determined as the target pore-throat structure. This process not only verifies the reliability of the pore-throat structure prediction method, but also improves the prediction accuracy through continuous iterative optimization, providing more accurate geological information support for subsequent oil and gas exploration and development. For example, in a certain wellbore, through the verification and optimization of nuclear magnetic logging data, it may be found that there is a deviation between the predicted pore-throat structure and the actual data in a certain depth section, and then the prediction model is fine-tuned to more accurately describe the pore-throat structure characteristics of this depth section.

[0027] In a preferred embodiment, as Figure 2 shown, according to the pore-throat structure test data, construct a pore-throat structure topology library based on historical samples, including: obtaining the pore-throat structure test data of the confidence historical samples with preset lithology type identifiers; evaluating the pairwise similarity of several pore-throat structures in the pore-throat structure test data to obtain a pore-throat structure similarity set; sorting the frequent pore-throat structures according to the pore-throat structure similarity set to obtain the selected pore-throat structure topology, associating and storing it with the preset lithology type identifier, and adding it to the pore-throat structure topology library.

[0028] Preferably, in the process of constructing the pore-throat structure topology library, first screen out the confidence historical samples with preset lithology type identifiers from a large amount of pore-throat structure test data. These confidence historical samples have been screened and verified, and their accuracy and representativeness can be highly trusted. They constitute the basic data for constructing the topology library. Next, for the pore-throat structure test data in these confidence historical samples, use the cosine similarity algorithm to evaluate the pairwise similarity.

[0029] The cosine similarity algorithm quantifies the similarity between two vectors by calculating the cosine value of the angle between them. The value range is between [-1, 1], where 1 indicates complete similarity, 0 indicates orthogonality (no correlation), and -1 indicates complete opposition. In the evaluation of pore-throat structure similarity, the characteristic parameters of each pore-throat structure (such as pore radius, throat width, connectivity, etc.) are regarded as a point in the vector space. By calculating the cosine similarity between these points, a similarity set of pore-throat structures can be obtained. For example, for two pore-throat structures A and B, their characteristic parameter vectors are extracted respectively, and then the cosine similarity of these two vectors is calculated. If the calculation result is 0.9, it indicates that A and B are highly similar in structure; if the result is 0.1, it indicates that the similarity between them is low.

[0030] Sort the frequent pore-throat structures according to the calculated pore-throat structure similarity set. The purpose of this step is to identify the frequently occurring and representative structure types from a large number of pore-throat structures. Set a similarity threshold, and classify the pore-throat structures with similarity higher than this threshold into one category to form the selected pore-throat structure topology. Finally, associate and store these selected pore-throat structure topologies with the preset lithology type identifiers and add them to the pore-throat structure topology library. In this way, in subsequent pore-throat structure prediction, the corresponding pore-throat structure topology can be quickly retrieved according to the lithology type, providing strong support for the prediction. For example, in a certain sandstone reservoir, several representative pore-throat structure types may be found through similarity evaluation and associated with the lithology type identifier of sandstone for direct use in subsequent predictions.

[0031] In a preferred embodiment, sorting the frequent pore-throat structures according to the pore-throat structure similarity set to obtain the selected pore-throat structure topology includes: configuring a neighborhood scale threshold, where the neighborhood scale threshold is greater than or equal to the ceiling value of 5% of the total number of the pore-throat structure similarity set and less than or equal to the floor value of 15% of the total number; according to the neighborhood scale threshold, taking the first pore-throat structure of the several pore-throat structures as a reference, sorting the first pore-throat structure neighborhood similarity set from large to small in the pore-throat structure similarity set, calculating the mean value, setting it as the local density of the first pore-throat structure, and adding it to the local density of the several pore-throat structures; calculating the mean value of the local density of the several pore-throat structures, setting it as the frequency evaluation factor, traversing the local density of the several pore-throat structures, respectively calculating the ratio with the frequency evaluation factor to obtain the frequent coefficients of the several pore-throat structures, and extracting the pore-throat structure topology with the maximum value as the selected pore-throat structure topology.

[0032] Specifically, in the process of sorting frequent pore-throat structures based on the pore-throat structure similarity set, a neighborhood scale threshold needs to be configured. The setting of this threshold determines the similarity range considered when sorting pore-throat structures. Specifically, the neighborhood scale threshold is set to the ceiling value greater than or equal to 5% of the total number of the pore-throat structure similarity set and the floor value less than or equal to 15% of the total number. Such a setting not only ensures the accuracy of sorting but also avoids overfitting or missing important information. Next, taking the first pore-throat structure among several pore-throat structures as a reference, the neighborhood similarity set of the first pore-throat structure is sorted from large to small from the pore-throat structure similarity set. This neighborhood similarity set contains other pore-throat structures with relatively high similarity to the first pore-throat structure. Then, the mean value of this neighborhood similarity set is calculated, set as the local density of the first pore-throat structure, and added to the local density set of several pore-throat structures.

[0033] Meanwhile, to evaluate the frequency of pore-throat structures, the mean value of the local densities of several pore-throat structures is calculated and set as the frequency evaluation factor. Traverse the local densities of several pore-throat structures, and calculate the ratio with the frequency evaluation factor respectively to obtain the frequency coefficients of several pore-throat structures. This frequency coefficient reflects the occurrence frequency and density of each pore-throat structure in the similarity set. Finally, extract the pore-throat structure topology with the largest frequency coefficient and set it as the selected pore-throat structure topology. This selected pore-throat structure topology represents the most representative and frequently occurring pore-throat structure type in the current similarity set. For example, in the pore-throat structure analysis of a certain sandstone reservoir, a similarity set containing multiple pore-throat structures may be obtained through similarity evaluation. By configuring the neighborhood scale threshold and sorting with the first pore-throat structure as a reference, it may be found that a specific pore-throat structure frequently appears in the similarity set, and its local density and frequency coefficient are both relatively high. Therefore, this pore-throat structure topology is selected as the selected pore-throat structure topology for subsequent pore-throat structure prediction and modeling.

[0034] In a preferred embodiment, combined with several attribute conventional logging curves, crossplot analysis is used to obtain multiple lithology logging response characteristics, including: performing pairwise crossplots on the several attribute conventional logging curves to construct multiple two-dimensional scatter plots; according to the multiple two-dimensional scatter plots, extracting the first two-dimensional scatter plot, performing frequent distribution coordinate matching with the first lithology type to obtain the first scatter plot logging response characteristic of the first lithology, and adding it to the first lithology logging response characteristic; adding the first lithology logging response characteristic to the multiple lithology logging response characteristics.

[0035] Furthermore, in the process of using crossplot analysis on several conventional logging curves combined with multiple attributes to obtain multiple lithologic logging response characteristics, these logging curves are first crossplotted pairwise to construct multiple two-dimensional scatter plots. These scatter plots intuitively show the correlation between different logging attributes and are the basis for identifying lithologic logging response characteristics. Analyzing these two-dimensional scatter plots, taking the first two-dimensional scatter plot as an example, it is matched with the known first lithologic type for frequently distributed coordinates. The purpose of this step is to identify the specific response pattern shown by the logging curves under the first lithologic type. Through matching, the logging response characteristics of the first lithology in the first scatter plot can be extracted, and these characteristics may include specific numerical ranges, distribution patterns, or differences from other lithologies, etc. Subsequently, the extracted logging response characteristics of the first lithology are added to the first lithologic logging response characteristic set for more comprehensive analysis later. At the same time, this characteristic set is also incorporated into the overall framework of multiple lithologic logging response characteristics and jointly constitutes a complete characteristic library with the logging response characteristics of other lithologies.

[0036] For example, in the analysis of logging data in a certain well section, through crossplot analysis, it is found that when the natural gamma curve intersects with the resistivity curve, the shale sample points show the characteristics of low natural gamma values and high resistivity values, and these points are relatively concentrated in the first two-dimensional scatter plot. Through frequently distributed coordinate matching, the logging response characteristics of shale in this scatter plot can be accurately extracted and used as an important identification marker for shale lithology. Subsequently, this characteristic is used together with the logging response characteristics of other lithologies for subsequent lithology identification and reservoir evaluation work.

[0037] In a preferred embodiment, the first two-dimensional scatter plot is extracted and matched with the first lithologic type for frequently distributed coordinates to obtain the first scatter plot logging response characteristics of the first lithology, including: extracting the first coordinate of the first two-dimensional scatter plot, where the first coordinate includes the first logging attribute characteristic value and the second logging attribute characteristic value, and the first coordinate has a depth identifier; using the first logging attribute characteristic value, the second logging attribute characteristic value, and the depth identifier as query constraints to retrieve the lithologic detection type data set that meets the query constraints; analyzing the trigger frequency ratio of the first lithologic type in the lithologic detection type data set, and when the trigger frequency ratio is greater than or equal to the trigger frequency threshold, adding the first coordinate to the initial first scatter plot logging response characteristics of the first lithology; when each coordinate of the first two-dimensional scatter plot is traversed, taking the first coordinate as the center, counting the number of the initial first scatter plot logging response characteristics within a preset radius, and when the number is greater than or equal to the number threshold, adding the first coordinate to the first scatter plot logging response characteristics of the first lithology, otherwise, deleting the first coordinate.

[0038] Optionally, in the process of extracting the first two-dimensional scatter plot and performing frequent distribution coordinate matching with the first lithology type to obtain the logging response characteristics of the first scatter plot of the first lithology, focus on each coordinate point in the first two-dimensional scatter plot. These coordinate points are composed of the first logging attribute eigenvalue and the second logging attribute eigenvalue, and each coordinate point is marked with a depth identifier, ensuring that its specific position in the wellbore can be accurately traced. Next, use the first logging attribute eigenvalue, the second logging attribute eigenvalue, and the depth identifier as query constraints to retrieve the lithology detection type dataset that meets these conditions. The purpose of this step is to find out which coordinate points in the first two-dimensional scatter plot are associated with the first lithology type.

[0039] Subsequently, analyze the trigger frequency ratio of the first lithology type in the lithology detection type dataset. If the trigger frequency ratio of a certain coordinate point is greater than or equal to the preset trigger frequency threshold, then add this coordinate point to the initial first scatter plot logging response feature set of the first lithology. This step screens out those coordinate points that are highly correlated with the first lithology type.

[0040] After each coordinate point in the first two-dimensional scatter plot has been traversed, further screen the coordinate points in the initial first scatter plot logging response feature set. Taking each coordinate point as the center, count the number of coordinate points belonging to the initial first scatter plot logging response within the preset radius. If the number is greater than or equal to the preset number threshold, then keep this coordinate point in the first scatter plot logging response feature of the first lithology; otherwise, delete this coordinate point. The purpose of this step is to ensure that the coordinate points in the finally obtained logging response feature set are highly aggregated and representative.

[0041] For example, in the analysis of logging data in a certain well section, it may be found that a certain coordinate point (composed of the natural gamma value and the resistivity value) in the first two-dimensional scatter plot is highly correlated with the shale lithology type. By retrieving the lithology detection type dataset, it is found that the trigger frequency ratio of this coordinate point in the shale sample is very high. Further counting the number of coordinate points within the preset radius, it is found that these points are also highly aggregated. Therefore, keep this coordinate point in the first scatter plot logging response feature of the shale as an important sign for identifying the shale lithology.

[0042] In a preferred embodiment, according to the lithology distribution probability prediction model, the multiple lithology logging response characteristics are processed to output a lithology joint probability distribution matrix, including: configuring lithology logging response characteristic record values, extracting the set of lithology types at the first measurement point, and constructing a lithology distribution probability identifier for the first measurement point based on the proportion of lithology types; until the lithology distribution probability identifier for the L-th measurement point is obtained; combining the lithology distribution probability identifiers from the first measurement point to the L-th measurement point to construct a label for the identified lithology joint probability distribution matrix; using the label of the identified lithology joint probability distribution matrix as supervision and the lithology logging response characteristic record values as input to train the lithology distribution probability prediction model.

[0043] Exemplarily, in the process of processing multiple lithology logging response characteristics according to the lithology distribution probability prediction model and outputting a lithology joint probability distribution matrix, it is necessary to configure lithology logging response characteristic record values, which are extracted based on actual logging data and reflect the characteristics of different lithologies in logging responses. For each measurement point (such as the first measurement point), the set of lithology types is extracted, and this set contains all possible lithology types at this measurement point. Then, based on the proportion of these lithology types at the measurement point, a lithology distribution probability identifier for the first measurement point is constructed, and this identifier is actually a vector or matrix indicating the probability of each lithology occurring at this measurement point.

[0044] Repeat the above steps until the lithology distribution probability identifier for the L-th measurement point is obtained. In this way, all lithology distribution probability identifiers from the first measurement point to the L-th measurement point are obtained. Combine these lithology distribution probability identifiers to construct a label for the identified lithology joint probability distribution matrix, and this label is a higher-dimensional matrix that reflects the probability of different lithologies jointly occurring at each measurement point along the wellbore trajectory. Finally, using the label of the identified lithology joint probability distribution matrix as a supervision signal and the lithology logging response characteristic record values as input, train the lithology distribution probability prediction model. During the training process, the model will learn how to predict the joint probability distribution of lithologies based on logging response characteristics.

[0045] For example, in the analysis of logging data in a certain well section, lithology logging response characteristic record values at multiple measurement points may be obtained. By extracting the set of lithology types at each measurement point and calculating their distribution probabilities, a label for a lithology joint probability distribution matrix is constructed. Then, this label is used to train the lithology distribution probability prediction model so that it can accurately predict the joint probability distribution of lithologies based on logging response characteristics. In this way, this model can be used to more accurately identify lithologies and evaluate reservoir properties in subsequent logging data interpretation.

[0046] In a preferred embodiment, the pore-throat structure probability distribution matrix is verified and optimized through nuclear magnetic logging data to obtain a target pore-throat structure, including: analyzing the nuclear magnetic logging data to obtain a pore-throat structure curve array data set; comparing the sum of squared residuals between the pore-throat structure curve array data set and the pore-throat structure probability distribution matrix, which is set as the pore-throat structure prediction fitness; when the pore-throat structure prediction fitness is greater than or equal to the fitness threshold, update the model parameters of the lithology distribution probability prediction model and execute a loop; when the pore-throat structure prediction fitness is less than the fitness threshold, set the pore-throat structure probability distribution matrix as the target pore-throat structure.

[0047] Preferably, the nuclear magnetic logging data is analyzed and a pore-throat structure curve array data set is extracted therefrom. This data set details the actual distribution of the pore-throat structure in the reservoir and is the key basis for verifying the accuracy of the pore-throat structure probability distribution matrix. The pore-throat structure curve array data set is compared with the pore-throat structure probability distribution matrix, and the sum of squared residuals between the two is calculated. This sum of squared residuals reflects the deviation degree between the prediction result and the actual data, and is set as the pore-throat structure prediction fitness. The smaller the fitness value, the closer the prediction result is to the actual data, and the higher the prediction accuracy.

[0048] Subsequently, a fitness threshold is set as the criterion for judging the quality of the prediction result. When the pore-throat structure prediction fitness is greater than or equal to this threshold, it indicates that there is a large deviation between the current prediction result and the actual data, and further optimization is required. At this time, the model parameters of the lithology distribution probability prediction model are updated, and the prediction and verification of the pore-throat structure probability distribution matrix are performed again, forming a cyclic optimization process. This cyclic optimization process will be continuously iterated until the pore-throat structure prediction fitness is less than the fitness threshold. At this time, it is considered that the current pore-throat structure probability distribution matrix is close enough to the actual nuclear magnetic logging data, so it is set as the target pore-throat structure.

[0049] For example, in the pore-throat structure prediction of a certain well section, there may be a large deviation between the initially obtained pore-throat structure probability distribution matrix and the actual nuclear magnetic logging data. By continuously updating the model parameters and re-predicting and verifying, the prediction fitness value is gradually reduced. Finally, when the prediction fitness is less than the set threshold, the current pore-throat structure probability distribution matrix is determined as the target pore-throat structure, providing more accurate geological information support for subsequent oil and gas exploration and development.

[0050] In a preferred embodiment, when the pore-throat structure prediction fitness is greater than or equal to the fitness threshold, updating the model parameters of the lithology distribution probability prediction model performs a loop, including: when the number of model parameter updates is greater than or equal to a preset number, obtaining a set of model parameters and a set of pore-throat structure prediction fitness; sorting the set of model parameters according to the ascending order of the set of pore-throat structure prediction fitness, and extracting the mean value of the model parameters with the top three serial numbers as the target model parameters; based on the target model parameters, performing a same-dimensional model parameter distance indentation update on 30% of the sorted model parameters to obtain extended model parameters; performing a loop based on the extended model parameters.

[0051] Specifically, when the pore-throat structure prediction fitness is greater than or equal to the fitness threshold, it is necessary to update the model parameters of the lithology distribution probability prediction model and perform loop optimization. First, a preset number is set to control the maximum number of iterations for model parameter updates. After each update of the model parameters, the current model parameters and the corresponding pore-throat structure prediction fitness are recorded. When the number of model parameter updates reaches or exceeds the preset number, a set of model parameters and a set of pore-throat structure prediction fitness are obtained.

[0052] Next, the set of model parameters is sorted in ascending order according to the set of pore-throat structure prediction fitness. The purpose of this is to find the model parameters that can make the prediction fitness smaller, that is, the model parameters closer to the actual data. Then, the model parameters with the top three serial numbers after sorting are extracted, and their mean value is calculated. This mean value is set as the target model parameter. Based on this target model parameter, a same-dimensional model parameter distance indentation update is performed on 30% of the sorted model parameters. Specifically, the distance between these model parameters and the target model parameter is calculated, and these model parameters are updated according to a certain rule (such as linear interpolation or proportional scaling) to make them closer to the target model parameter. In this way, a set of extended model parameters is obtained. Finally, loop optimization continues based on the set of extended model parameters. In subsequent iterations, these extended model parameters will be used for the prediction and verification of the pore-throat structure probability distribution matrix until the pore-throat structure prediction fitness is less than the fitness threshold.

[0053] For example, in the prediction of pore-throat structure in a certain well section, multiple sets of model parameters may be initially used for prediction, but the prediction fitness is not ideal. By continuously updating the model parameters and recording the prediction fitness each time, a set of model parameters and the corresponding pore-throat structure prediction fitness set are finally obtained. After sorting in ascending order of the prediction fitness, the mean value of the first three serial numbers of the model parameters is extracted as the target model parameter, and the 30% of the sorted model parameters are updated by indenting the distance of the same-dimensional model parameters. In this way, a set of augmented model parameters closer to the actual data is obtained, and the loop optimization is continued based on these parameters, and finally a satisfactory pore-throat structure prediction result is obtained.

[0054] The method for predicting pore-throat structure by fusing multiple logging curve data provided by the embodiments of the present invention has at least the following technical effects: 1. By constructing a pore-throat structure topology library based on historical samples and introducing a confidence historical sample and a similarity evaluation mechanism, the pore-throat structure topologies associated with different lithology type identifiers can be accurately identified and stored. This high-precision topology library not only improves the accuracy of pore-throat structure prediction but also provides a basis for the subsequent generation of the lithology joint probability distribution matrix. In particular, by configuring the neighborhood scale threshold and the frequent pore-throat structure sorting strategy, the most representative pore-throat structure topologies can be automatically selected, thereby further improving the prediction accuracy.

[0055] 2. Multiple logging curve data are fused, and the lithology logging response characteristics are extracted through crossplot analysis. The fusion of this multi-source data not only enriches the information source for lithology identification but also improves the accuracy of lithology identification. In particular, through frequent distribution coordinate matching and quantity threshold screening, the logging response characteristics highly correlated with specific lithology types can be accurately extracted, providing high-quality data input for the subsequent generation of the lithology joint probability distribution matrix.

[0056] 3. A dynamic model parameter optimization mechanism is introduced. By continuously updating the model parameters of the lithology distribution probability prediction model and aiming at minimizing the pore-throat structure prediction fitness, the verification and optimization of the pore-throat structure probability distribution matrix are realized. This dynamic optimization strategy not only improves the adaptability and prediction accuracy of the model but also enables the model to automatically adapt to logging data under different geological conditions. In particular, when the pore-throat structure prediction fitness is greater than or equal to the fitness threshold, the optimal model parameter combination can be efficiently found through steps such as model parameter set sorting, target model parameter extraction, and same-dimensional model parameter distance indentation update, so as to quickly converge to the target pore-throat structure.

[0057] Embodiment 2:

[0058] As Figure 3As shown, based on the same inventive concept as the method for predicting pore-throat structure by integrating multiple logging curve data provided in the first embodiment, the embodiment of the present invention further provides a system for predicting pore-throat structure by integrating multiple logging curve data. The system includes: A pore-throat structure topology library construction module 11, configured to construct a pore-throat structure topology library based on historical samples according to pore-throat structure test data. Among them, any pore-throat structure topology in the pore-throat structure topology library has a lithology type identifier.

[0059] An intersection plot analysis module 12, configured to perform intersection plot analysis by combining several attribute conventional logging curves to obtain multiple lithology logging response characteristics.

[0060] A model prediction module 13, configured to process the multiple lithology logging response characteristics according to a lithology distribution probability prediction model and output a lithology joint probability distribution matrix. Among them, the lithology distribution probability prediction model is generated by machine learning training using multiple groups of data. Any one of the multiple groups of data includes a lithology logging response characteristic record value and a label identifying the lithology joint probability distribution matrix.

[0061] A comparison and analysis module 14, configured to compare with the lithology type identifier based on multiple lithology types, extract pore-throat structure topologies from the pore-throat structure topology library, generate a pore-throat structure distribution matrix, and take the matrix dot product at each depth point along the wellbore trajectory in combination with the lithology joint probability distribution matrix to obtain a pore-throat structure probability distribution matrix.

[0062] A verification and optimization module 15, configured to verify and optimize the pore-throat structure probability distribution matrix through nuclear magnetic logging data to obtain a target pore-throat structure.

[0063] Furthermore, the pore-throat structure topology library construction module 11 is further configured to perform the following steps: Obtain pore-throat structure test data of confidence historical samples with preset lithology type identifiers; perform pairwise similarity evaluation on several pore-throat structures of the pore-throat structure test data to obtain a pore-throat structure similarity set; perform frequent pore-throat structure sorting according to the pore-throat structure similarity set to obtain selected pore-throat structure topologies, associate and store them with the preset lithology type identifiers, and add them to the pore-throat structure topology library.

[0064] Furthermore, the pore-throat structure topology library construction module 11 is further configured to perform the following steps: Configure the neighborhood scale threshold, where the neighborhood scale threshold is greater than or equal to the ceiling value of 5% of the total number of the pore-throat structure similarity set and less than or equal to the floor value of 15% of the total number; according to the neighborhood scale threshold, taking the first pore-throat structure of the several pore-throat structures as a reference, sort the first pore-throat structure neighborhood similarity set from large to small in the pore-throat structure similarity set, calculate the mean value, set it as the local density of the first pore-throat structure, and add it to the local densities of the several pore-throat structures; calculate the mean value of the local densities of the several pore-throat structures, set it as the frequency evaluation factor, traverse the local densities of the several pore-throat structures, calculate the ratio with the frequency evaluation factor respectively to obtain the frequent coefficients of the several pore-throat structures, and extract the pore-throat structure topology with the maximum value, set it as the selected pore-throat structure topology.

[0065] Furthermore, the cross-plot analysis module 12 is further configured to perform the following steps: Perform pairwise intersections on the several conventional logging curves of attributes to construct a plurality of two-dimensional scatter plots; according to the plurality of two-dimensional scatter plots, extract the first two-dimensional scatter plot, perform frequent distribution coordinate matching with the first lithology type to obtain the first scatter plot logging response characteristics of the first lithology, and add them to the logging response characteristics of the first lithology; add the logging response characteristics of the first lithology to the plurality of lithology logging response characteristics.

[0066] Furthermore, the cross-plot analysis module 12 is further configured to perform the following steps: Extract the first coordinates of the first two-dimensional scatter plot, where the first coordinates include the first logging attribute characteristic value and the second logging attribute characteristic value, and the first coordinates have depth identifiers; use the first logging attribute characteristic value, the second logging attribute characteristic value and the depth identifier as query constraints to retrieve the lithology detection type data set that meets the query constraints; analyze the trigger frequency ratio of the first lithology type in the lithology detection type data set. When the trigger frequency ratio is greater than or equal to the trigger frequency threshold, add the first coordinates to the initial first scatter plot logging response characteristics of the first lithology; when each coordinate of the first two-dimensional scatter plot is traversed, count the number of the initial first scatter plot logging response characteristics within a preset radius centered on the first coordinates. When the number is greater than or equal to the quantity threshold, add the first coordinates to the first scatter plot logging response characteristics of the first lithology, otherwise, delete the first coordinates.

[0067] Furthermore, the model prediction module 13 is further configured to perform the following steps: Configure the recorded values of the lithology logging response characteristics, extract the set of lithology types at the first measurement point, and construct the lithology distribution probability identifier at the first measurement point based on the proportion of lithology types; until the lithology distribution probability identifier at the L-th measurement point is obtained; combine the lithology distribution probability identifier at the first measurement point until the lithology distribution probability identifier at the L-th measurement point to construct the label of the joint probability distribution matrix of the identified lithology; use the label of the joint probability distribution matrix of the identified lithology as the supervision and the lithology logging response characteristic recorded values as the input to train the lithology distribution probability prediction model.

[0068] Furthermore, the verification and optimization module 15 is further configured to perform the following steps: Analyze the nuclear magnetic logging data to obtain the pore-throat structure curve array data set; compare the sum of squared residuals between the pore-throat structure curve array data set and the pore-throat structure probability distribution matrix, and set it as the pore-throat structure prediction fitness; when the pore-throat structure prediction fitness is greater than or equal to the fitness threshold, update the model parameters of the lithology distribution probability prediction model and execute a loop; when the pore-throat structure prediction fitness is less than the fitness threshold, set the pore-throat structure probability distribution matrix as the target pore-throat structure.

[0069] Furthermore, the verification and optimization module 15 is further configured to perform the following steps: When the number of model parameter updates is greater than or equal to the preset number, obtain the set of model parameters and the set of pore-throat structure prediction fitness; sort the set of model parameters in ascending order according to the set of pore-throat structure prediction fitness, and extract the mean values of the model parameters with the top three serial numbers, and set them as the target model parameters; based on the target model parameters, perform a same-dimensional model parameter distance indentation update on 30% of the sorted model parameters to obtain the extended model parameters; execute a loop based on the extended model parameters.

[0070] Through the foregoing detailed description of a method for predicting pore-throat structure by fusing multiple logging curve data in this specification, those skilled in the art can clearly know a system for predicting pore-throat structure by fusing multiple logging curve data in this embodiment. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the description of the method part.

[0071] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to these embodiments shown herein, but will be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for predicting pore throat structure by integrating multiple logging curve data, characterized in that: include: According to the pore-throat structure test data, a pore-throat structure topology library is constructed based on historical samples, wherein any pore-throat structure topology in the pore-throat structure topology library has a lithology type identifier; Combine several attribute conventional logging curves and use cross plot analysis to obtain multiple lithology logging response characteristics; According to the lithology distribution probability prediction model, the multiple lithology logging response characteristics are processed to output a lithology joint probability distribution matrix, wherein the lithology distribution probability prediction model is generated by using machine learning training through multiple groups of data, and any group of the multiple groups of data includes a lithology logging response characteristic record value and a label identifying the lithology joint probability distribution matrix; Based on multiple lithology types, the lithology type identifiers are compared, the pore throat structure topology is extracted from the pore throat structure topology library, a pore throat structure distribution matrix is ​​generated, and the matrix dot product is taken at each depth point along the wellbore trajectory in combination with the lithology joint probability distribution matrix to obtain a pore throat structure probability distribution matrix; The pore-throat structure probability distribution matrix is ​​verified and optimized through nuclear magnetic logging data to obtain the target pore-throat structure.

2. The method according to claim 1, characterized in that According to the pore-throat structure test data, a pore-throat structure topology library is constructed based on historical samples, including: Obtain pore-throat structure test data of trusted historical samples with preset lithology type identification; Performing pairwise similarity evaluation on a plurality of pore-throat structures of the pore-throat structure test data to obtain a pore-throat structure similarity set; Frequent pore-throat structures are sorted according to the pore-throat structure similarity set to obtain a selected pore-throat structure topology, which is associated with the preset lithology type identifier and stored, and added to the pore-throat structure topology library.

3. The method according to claim 2, characterized in that Frequent pore throat structures are sorted according to the pore throat structure similarity set to obtain a selected pore throat structure topology, including: Configuring a neighborhood scale threshold, wherein the neighborhood scale threshold is greater than or equal to an upward rounded value of 5% of the total number of the pore throat structure similarity set, and less than or equal to a downward rounded value of 15% of the total number; According to the neighborhood scale threshold, taking the first pore throat structure of the plurality of pore throat structures as a benchmark, sorting the first pore throat structure neighborhood similarity set from large to small among the pore throat structure similarity sets, calculating the mean, setting it as the first pore throat structure local density, and adding the plurality of pore throat structure local densities; The mean values ​​of the local densities of the pore-throat structures are calculated and set as the frequency evaluation factor. The local densities of the pore-throat structures are traversed and compared with the frequency evaluation factors respectively to obtain the frequent coefficients of the pore-throat structures. The pore-throat structure topology with the maximum value is extracted and set as the selected pore-throat structure topology.

4. The method according to claim 1, characterized in that Combined with several conventional logging curves of attributes, cross-plot analysis is used to obtain multiple lithology logging response characteristics, including: Intersecting the conventional well logging curves of the plurality of attributes in pairs to construct a plurality of two-dimensional scatter plots; Extracting a first two-dimensional scatter plot according to the plurality of two-dimensional scatter plots, performing frequent distribution coordinate matching with the first lithology type, obtaining a first scatter plot well logging response feature of the first lithology, and adding the first lithology well logging response feature; The first lithology logging response feature is added to the plurality of lithology logging response features.

5. The method according to claim 4, characterized in that Extract the first two-dimensional scatter plot, match the frequent distribution coordinates with the first lithology type, and obtain the first scatter plot logging response characteristics of the first lithology, including: Extracting a first coordinate of the first two-dimensional scatter plot, wherein the first coordinate includes a first well logging attribute characteristic value and a second well logging attribute characteristic value, and the first coordinate has a depth identifier; Using the first well logging attribute characteristic value, the second well logging attribute characteristic value and the depth identifier as query constraints, searching for a lithology detection type data set that meets the query constraints; Analyzing the trigger frequency ratio of the first lithology type in the lithology detection type data set, and when the trigger frequency ratio is greater than or equal to a trigger frequency threshold, adding the first coordinate to an initial first scatter plot logging response feature of the first lithology; When each coordinate traversal of the first two-dimensional scatter plot is completed, the number of logging response features belonging to the initial first scatter plot within a preset radius is counted with the first coordinate as the center. When the number is greater than or equal to a quantity threshold, the first coordinate is added to the first scatter plot logging response feature of the first lithology; otherwise, the first coordinate is deleted.

6. The method according to claim 1, characterized in that According to the lithology distribution probability prediction model, the plurality of lithology logging response characteristics are processed to output a lithology joint probability distribution matrix, including: Configure the lithology logging response characteristic record value, extract the lithology type set of the first measurement point, and construct the lithology distribution probability mark of the first measurement point based on the lithology type proportion; Until the lithology distribution probability mark of the Lth measuring point is obtained; Combining the lithology distribution probability identification of the first measurement point to the lithology distribution probability identification of the Lth measurement point, constructing a label identifying a lithology joint probability distribution matrix; The lithology distribution probability prediction model is trained by taking the labels of the lithology joint probability distribution matrix as supervision and the lithology logging response characteristic record values ​​as input.

7. The method according to claim 1, characterized in that The pore-throat structure probability distribution matrix is ​​verified and optimized through nuclear magnetic logging data to obtain the target pore-throat structure, including: Analyzing the nuclear magnetic logging data to obtain a pore throat structure curve array data set; Comparing the residual sum of squares of the pore-throat structure curve array data set and the pore-throat structure probability distribution matrix, setting it as the pore-throat structure prediction fitness; When the pore-throat structure prediction fitness is greater than or equal to a fitness threshold, updating the model parameters of the lithology distribution probability prediction model to execute a loop; When the predicted fitness of the pore-throat structure is less than the fitness threshold, the pore-throat structure probability distribution matrix is ​​set to the target pore-throat structure.

8. The method according to claim 7, characterized in that When the pore-throat structure prediction fitness is greater than or equal to the fitness threshold, the model parameters of the lithology distribution probability prediction model are updated and a cycle is executed, including: When the number of model parameter updates is greater than or equal to the preset number, a model parameter set and a pore throat structure prediction fitness set are obtained; According to the pore-throat structure prediction fitness set, the model parameter set is sorted from small to large, and the mean of the model parameters of the first three numbers is extracted and set as the target model parameter; Based on the target model parameters, the sorted 30% model parameters are updated by indenting the same-dimensional model parameters to obtain expanded model parameters; A loop is performed based on the augmented model parameters.

9. A system for predicting pore throat structure by integrating multiple logging curve data, characterized in that: A method for predicting pore throat structure by fusing multiple logging curve data as described in any one of claims 1 to 8, the system comprising: A pore-throat structure topology library construction module is used to construct a pore-throat structure topology library based on pore-throat structure test data and historical samples, wherein any pore-throat structure topology of the pore-throat structure topology library has a lithology type identifier; The cross-plot analysis module is used to combine several attribute conventional logging curves to obtain multiple lithology logging response characteristics by cross-plot analysis; A model prediction module, used for processing the plurality of lithology logging response characteristics according to a lithology distribution probability prediction model, and outputting a lithology joint probability distribution matrix, wherein the lithology distribution probability prediction model is generated by training multiple sets of data using machine learning, and any set of the multiple sets of data includes a lithology logging response characteristic record value and a label identifying the lithology joint probability distribution matrix; A comparison and analysis module is used to compare the lithology type identifier with multiple lithology types, extract the pore-throat structure topology from the pore-throat structure topology library, generate a pore-throat structure distribution matrix, and combine the lithology joint probability distribution matrix at each depth point along the wellbore trajectory to obtain a pore-throat structure probability distribution matrix by taking the matrix dot product; The verification and optimization module is used to verify and optimize the pore-throat structure probability distribution matrix through nuclear magnetic logging data to obtain the target pore-throat structure.

Images