A reservoir permeability determination method, device, equipment and storage medium

Abstract

The application discloses a reservoir permeability determination method, device, equipment and storage medium. The technical scheme of the application constructs a permeability calculation model of different pore structure categories in a research area, determines the probability of each pore structure category to which a target well section belongs and a target average pore radius by using a trained target multi-task learning neural network model, determines a candidate permeability under different pore structure categories based on each permeability calculation model according to the target average pore radius, and finally determines a target permeability of the target well section according to the candidate permeability under the pore structure category and the pore structure category probability, thereby avoiding the problem of low permeability prediction accuracy of a traditional single model or hard classification model in a heterogeneous reservoir, and improving the prediction accuracy and geological coincidence degree of the reservoir permeability.

Description

Technical Field

[0001] This invention relates to the field of oil and gas exploration and development technology, and in particular to a method, apparatus, equipment and storage medium for determining reservoir permeability. Background Technology

[0002] In oil and gas exploration and development, the pore structure of the reservoir is a key factor controlling fluid permeability, directly determining the productivity and development effectiveness of oil and gas wells. Accurately evaluating reservoir permeability is one of the core tasks of oil and gas reservoir characterization.

[0003] Facing complex reservoirs with deep, low permeability and gravel content presents significant challenges, primarily due to the complex relationships among the four properties: lithology, physical properties, electrical properties, and hydrocarbon-bearing capacity. These relationships are not simple linear, especially in gravelly sandstones where the skeletal composition varies, leading to errors in porosity calculations based on volumetric models. Furthermore, the pore structure exhibits strong heterogeneity; even with similar porosities, differences in pore structure, such as pore radius, throat size, and coordination number, can result in permeability differences of several orders of magnitude. Traditional single-porosity-permeability relationships cannot characterize this complexity. Finally, the conductivity mechanism is complex: variations in pore structure directly affect the rock's conductivity, causing changes in rock electrical parameters and exhibiting "non-Archive" characteristics, reducing the accuracy of saturation assessments based on fixed parameters. Therefore, conventional methods often fall short of the accuracy required for refined permeability prediction in complex reservoirs.

[0004] In recent years, traditional machine learning methods, such as Support Vector Machines (SVM) and Random Forests, have been introduced into the field of well logging interpretation. By fusing multiple well logging curves to classify reservoirs or predict parameters, they have improved accuracy to some extent. However, these methods often employ a "hard classification" approach, where the model outputs only a single, definitive category label for each depth point. This approach fails to capture the inherent gradual changes and uncertainties at category boundaries in geological bodies. When samples are located in the transition zone between two types of pore structures, hard classification can lead to misclassification of the reservoir category, resulting in the application of incorrect porosity-permeability relationship models and causing abrupt or jumps in permeability calculations that do not match geological realities. Furthermore, these methods fail to deeply integrate the differentiated porosity-permeability mechanisms under different pore structures into the model framework, resulting in poor model interpretability.

[0005] Therefore, there is an urgent need for a geologically interpretable method for determining reservoir permeability with high precision. Summary of the Invention

[0006] This invention provides a method, apparatus, equipment, and storage medium for determining reservoir permeability, in order to improve the accuracy and geological consistency of reservoir permeability.

[0007] In a first aspect, embodiments of the present invention provide a method for determining reservoir permeability, the method comprising: Determine the target logging curve characteristics, target porosity, and permeability calculation model (represented by porosity and average pore radius) for each pore structure category in the study area; Using a target multi-task learning neural network model, the target logging curve features of the target well section are input to obtain the target average pore radius of the target well section and the probability of the target well section belonging to each pore structure category. The target multi-task learning neural network model includes a pore structure classification branch and an average pore radius regression branch. The pore structure classification branch is used to output the probability vector of the input sample belonging to each pore structure category, and the average pore radius regression branch is used to output the average pore radius of the input sample. Based on the permeability calculation model corresponding to each pore structure category, the target average pore radius and target porosity are substituted into the calculation model to obtain the candidate permeability corresponding to each pore structure category; The target permeability of the target well section is determined based on the probability of the pore structure category and the candidate permeability corresponding to each pore structure category.

[0008] Secondly, embodiments of the present invention also provide a reservoir permeability determination apparatus, the apparatus comprising: The data determination module is used to determine the target logging curve characteristics, target porosity, and permeability calculation model (represented by porosity and average pore radius) for each pore structure category in the study area. The pore radius and probability determination module is used to utilize a target multi-task learning neural network model, inputting the target logging curve features of the target well section, and outputting the target average pore radius of the target well section and the probability of the target well section belonging to each pore structure category; the target multi-task learning neural network model includes a pore structure classification branch and an average pore radius regression branch, the pore structure classification branch is used to output the probability vector of the input sample belonging to each pore structure category, and the average pore radius regression branch is used to output the average pore radius of the input sample; The candidate permeability determination module is used to calculate the candidate permeability corresponding to each pore structure category by substituting the target average pore radius and target porosity into the permeability calculation model corresponding to each pore structure category. The target permeability determination module is used to determine the target permeability of the target well section based on the probability of the pore structure category and the candidate permeability corresponding to each pore structure category.

[0009] Thirdly, embodiments of the present invention also provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the reservoir permeability determination method as described in any of the embodiments of the present invention.

[0010] Fourthly, embodiments of the present invention also provide a storage medium for storing computer-executable instructions, which, when executed by a computer processor, are used to perform the reservoir permeability determination method as described in any of the embodiments of the present invention.

[0011] The technical solution of this invention constructs a permeability calculation model for different pore structure categories in the study area, uses a trained target multi-task learning neural network model to determine the probability of the target well section belonging to each pore structure category and the target average pore radius, determines candidate permeabilities under different pore structure categories based on the target average pore radius and each permeability calculation model, and finally determines the target permeability of the target well section based on the candidate permeabilities under the pore structure category and the probability of the pore structure category. This avoids the problem of low permeability prediction accuracy of traditional single models or hard classification models in heterogeneous reservoirs, and improves the accuracy of reservoir permeability and geological consistency.

[0012] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 This is a flowchart of a method for determining reservoir permeability provided in Embodiment 1 of the present invention; Figure 2 This is a flowchart of a method for determining reservoir permeability provided in Embodiment 2 of the present invention; Figure 3 This is a schematic diagram of the architecture of the preset multi-task learning neural network model provided in Embodiment 2 of the present invention; Figure 4 This is a schematic diagram comparing the predicted permeability and the measured permeability of the entire target well section provided in Embodiment 2 of the present invention; Figure 5 This is a schematic diagram of a reservoir permeability determination device provided in Embodiment 3 of the present invention; Figure 6 This is a schematic diagram of the structure of an electronic device for implementing the reservoir permeability determination method of this invention. Detailed Implementation

[0015] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0016] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of 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 interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a 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.

[0017] Example 1 Figure 1 This is a flowchart illustrating a method for determining reservoir permeability according to Embodiment 1 of the present invention. This embodiment is applicable to reservoir permeability determination. The method can be executed by a reservoir permeability determination device, which can be implemented in hardware and / or software. This device can be configured in any electronic device with network communication and computing capabilities. Figure 1 As shown, the method includes: S110. Determine the target logging curve characteristics, target porosity, and permeability calculation model (represented by porosity and average pore radius) for the target well section in the study area.

[0018] In this embodiment, the target logging curve features are the logging curve features of the target well section. The logging curve features of this embodiment include at least the clay content, deep lateral resistivity, compensated density, compensated neutrons, P-wave transit time, S-wave transit time, and effective porosity. The target porosity is the porosity of the target well section.

[0019] In this embodiment, different pore structure categories are used to reflect the evolution of reservoir pore structure from high-energy sedimentary environments (such as delta front underwater distributary channels) to low-energy environments (such as delta plains and interchannel bays). The permeability calculation model is a mapping relationship with porosity and average pore radius as independent variables and permeability as the dependent variable. Each pore structure category corresponds to its own permeability calculation model.

[0020] In practical applications, combined logging is used to acquire natural gamma ray, deep lateral resistivity, compensated density, compensated neutron, and array acoustic full-wavelength logging data for the target well section. Depth alignment, environmental correction, and anomaly removal are then performed on each logging data set. Based on the corrected logging data, logging curve characteristics such as clay content, deep lateral resistivity, compensated density, compensated neutron, P-wave transit time, S-wave transit time, and effective porosity are calculated or directly extracted. Specifically, clay content can be calculated from natural gamma ray; deep lateral resistivity, compensated density, compensated neutron, P-wave transit time, and S-wave transit time can be obtained directly or through inversion from the deep lateral, density, neutron, and acoustic curves; and effective porosity can be calculated using a density-neutron-clay content joint model. Following these steps, accurate acquisition of logging curve characteristic parameters for the target well section can be achieved.

[0021] Optionally, before determining the permeability calculation model represented by porosity and average pore radius for each pore structure category in the study area, the following steps are included: Determine the high-pressure mercury intrusion curve morphology, displacement pressure, maximum mercury saturation, average pore radius, and pore throat sorting coefficient of core samples from each candidate well section in the study area; Based on at least two parameters among the high-pressure mercury intrusion curve morphology, displacement pressure, maximum mercury saturation, average pore radius, and pore throat sorting coefficient, and a preset parameter threshold, the reservoirs in the study area are classified into at least five pore structure categories, from high-quality reservoirs to non-reservoirs.

[0022] In this embodiment, the displacement pressure, maximum mercury ingress saturation, average pore radius, and pore throat sorting coefficient are quantitative characteristic parameters of pore structure, which can be directly obtained from the high-pressure mercury intrusion test of the core and calculated based on the measured mercury intrusion pressure-mercury ingress volume relationship curve of the core. The preset parameter threshold is a pre-set parameter threshold used to classify pore structure categories.

[0023] In this embodiment, core laboratory test data corresponding to the core wells in the study area are collected. The test data includes at least high-pressure mercury intrusion porosimetry data, nuclear magnetic resonance T2 spectrum test data, cast thin section microscopic image data, and porosity and permeability analysis data. Precise depth correction and spatial positioning processing are performed on all core sampling depths to complete the scale matching and position alignment between the core experimental depth and the field measured logging depth, ensuring that the core experimental samples and logging response data correspond one-to-one in the vertical depth dimension, and constructing a standardized core experimental database.

[0024] Furthermore, based on the overall morphology of the mercury intrusion curves obtained from high-pressure mercury intrusion experiments on core samples, multiple quantitative characterization parameters of pore structure, such as displacement pressure, maximum mercury saturation, average pore radius, and pore throat sorting coefficient, were extracted and extracted. According to the numerical distribution law of each quantitative characterization parameter of pore structure and the differences in reservoir storage and seepage performance, parameter thresholds were set to classify the reservoirs in the study area into hierarchical categories, resulting in multiple pore structure categories with clear sedimentary geological origins and differences in reservoir physical properties.

[0025] In this embodiment, five pore structure categories are preferably defined as Class I, Class II, Class III, Class IV, and Class V, with the classification levels corresponding to high-quality reservoirs, good reservoirs, medium reservoirs, poor reservoirs, and ineffective reservoirs, respectively, from high to low.

[0026] In this embodiment, by classifying the reservoirs in the study area from high-quality reservoirs to non-reservoirs into at least five categories of pore structure, this classification system can fully characterize the evolution characteristics of reservoir pore structure under the control of sedimentary paleoenergy environment. It can achieve a complete genetic evolution law corresponding to the development quality of reservoir pore structure from excellent to poor within the range from high-energy sedimentary favorable facies such as underwater distributary channels at the delta front to low-energy sedimentary unfavorable facies such as delta plains and interchannel bays.

[0027] Furthermore, in practical applications, systematic quantitative feature statistics and visualization feature map construction are carried out for each category of pore structure that has been classified, and the category-specific features are finely characterized from multiple dimensions, including mercury intrusion porosimetry feature characterization, nuclear magnetic resonance experiment feature characterization, microscopic feature characterization of cast thin sections, and macroscopic physical property characterization.

[0028] Specifically, the standard high-pressure mercury intrusion curve morphology for each type of pore structure is summarized and determined, the displacement pressure range and maximum mercury saturation value distribution range for each type of pore structure are clarified, and a category-specific mercury intrusion parameter judgment standard is established to achieve the characteristic characterization of mercury intrusion experiments.

[0029] The typical distribution patterns of nuclear magnetic resonance T2 spectra corresponding to various porosity structures were analyzed, different spectral structures such as single-peak and double-peak were distinguished, the relaxation time positions corresponding to the main peaks of each type of nuclear magnetic resonance T2 spectrum were marked, the overall relaxation time distribution range was delineated, and a basis for classifying and identifying nuclear magnetic resonance responses was formed to achieve the characterization of nuclear magnetic resonance experimental features.

[0030] Based on the microscopic observation results of cast thin sections, the pore types dominated by different types of pore structures are defined, including primary intergranular pores, intragranular dissolution pores, and casting pores, etc.; at the same time, the contact combination relationship of various reservoir rock particles and the occurrence and occurrence morphology of clay minerals in the formation are clarified, the differences in micropore development and rock-mineral combination are clarified, and the microscopic characteristics of cast thin sections are characterized.

[0031] The effective porosity and formation permeability of core samples of each type of pore structure were statistically analyzed and their measured values ​​were distributed. The mathematical correlation trend of porosity-permeability within a single category was established by fitting and establishing the differential characterization of the porosity-permeability response law of different types of reservoirs. This laid the data foundation for the subsequent construction of porosity-permeability prediction models by category and achieved the characterization of macroscopic physical properties.

[0032] By systematically analyzing the morphology of mercury intrusion porosimetry curves, T2 NMR spectra, thin-section micropore combinations, and porosity-permeability distribution ranges corresponding to each type of pore structure, a visualized and quantitative characteristic map is formed. This transforms manual qualitative classification based on experimental parameters into a standardized judgment criterion with a unified scale and quantifiable comparison, eliminating the subjectivity and experience bias of manual classification and ensuring that the pore structure classification results are rigorous, consistent, and reproducible. Simultaneously, the characteristic map comprehensively encompasses the full-dimensional characterization information of the reservoir, from micropore-throat structure to macroscopic properties. It can accurately anchor the comprehensive formation response patterns corresponding to different types of pore structures, providing complete geological prior knowledge and characteristic constraints for subsequent intelligent identification of pore structure types and prediction of average pore radius using multi-dimensional parameters of well logging curves as input. This improves the rationality of model training sample labels and the geological consistency of prediction results.

[0033] In this embodiment, by classifying the reservoirs in the study area from high-quality reservoirs to non-reservoir reservoirs into five types of pore structures, it is possible to achieve fine-grained classification of reservoir quality based on pore throat development characteristics and reservoir permeability. This classification system can effectively conform to the evolutionary law of reservoir pore structure from superior to inferior under different sedimentary energy environments, clearly distinguishing the differences in micropore combinations, pore throat configuration relationships, and macroscopic physical property development of various reservoirs. It can provide standardized and unified classification labels for multi-task learning neural network models, ensuring the accuracy of model training and intelligent recognition, and also lay the classification foundation for building a dedicated pore-permeability relationship model for the classification model, further improving the prediction accuracy of reservoir permeability.

[0034] Optionally, determine the permeability calculation model, represented by porosity and average pore radius, for each pore structure category in the study area, including: Porosity, average pore radius, and permeability of core samples corresponding to each pore structure type in the study area were obtained. Based on the porosity, average pore radius, and permeability, a permeability calculation model, represented by porosity and average pore radius, is determined for each pore structure category through nonlinear regression fitting.

[0035] In this embodiment, core samples belonging to different pore structure categories within the study area were screened, and the measured values ​​of porosity, average pore radius, and permeability for each category of core samples were statistically obtained. Using porosity and average pore radius as independent variables and permeability as the dependent variable within the same pore structure category, a nonlinear regression fitting method was employed to perform mathematical relationship fitting analysis. Specific quantitative permeability calculation models for each pore structure category were established, using porosity and average pore radius as input factors, thus achieving the technical objective of accurately calculating the permeability of the corresponding reservoir category based on porosity and average pore radius.

[0036] Optionally, the permeability calculation model is represented by the following formula: ; in, For penetration rate, Porosity The average pore radius, , , is the fitting coefficient corresponding to the i-th pore structure category.

[0037] In this embodiment, the fitting coefficients of the permeability calculation model corresponding to each type of pore structure may be different. The fitting permeability calculation model can solidify the control effect of pore structure on seepage capacity in a mechanistic way, which is convenient for subsequent calculations.

[0038] In this embodiment, by constructing permeability calculation models specific to different pore structure categories, the drawbacks of a uniform fitting model across the entire domain that cannot adapt to the different pore throat development characteristics and seepage patterns of reservoirs can be effectively avoided. By fully combining the differentiated characteristics of pore development morphology, pore throat connectivity, and pore size distribution of various reservoirs, the inherent nonlinear relationship between porosity, average pore radius, and permeability can be accurately clarified, significantly improving the accuracy and geological consistency of quantitative permeability calculation results.

[0039] S120. Using a target multi-task learning neural network model, input the target logging curve features of the target well section to obtain the target average pore radius of the target well section and the probability of the target well section belonging to each pore structure category.

[0040] In this embodiment, the target multi-task learning neural network model is a pre-trained multi-task learning neural network model. The target multi-task learning neural network model in this embodiment includes a pore structure classification branch and an average pore radius regression branch. The pore structure classification branch is used to output the probability vector of the input sample belonging to each type of pore structure, and the average pore radius regression branch is used to output the average pore radius of the input sample.

[0041] The target average pore radius is the average pore radius of the target well section output by the target multi-task learning neural network model based on the characteristics of the target logging curve. The pore structure category probability is the probability that the target well section belongs to each pore structure category, output by the target multi-task learning neural network model based on the characteristics of the target logging curve.

[0042] Specifically, relying on the constructed and trained target multi-task learning neural network model, the well logging curve features of the target well section after correction are used as input data for the target multi-task learning neural network model. The model performs feature analysis and calculation through the built-in shared feature extraction layer and dual-branch task network, and simultaneously outputs the continuous value of the predicted average pore radius corresponding to the formation of the target well section, as well as the probability distribution results of the target well section belonging to five different pore structure categories. In this way, intelligent identification of pore structure type and quantitative prediction of pore throat size parameters at the well section scale can be achieved based on measured well logging data.

[0043] In this embodiment, the target multi-task learning neural network model outputs the probability of the target well section belonging to each pore structure category. This not only provides the most likely pore structure category but also quantifies the uncertainty of the category boundary. This "soft" identification and classification result is more in line with the objective law of the gradual change of geological bodies in the vertical and horizontal directions, providing richer and more accurate information for subsequent permeability calculations and avoiding the step-like misjudgments that may be caused by "hard classification".

[0044] Optionally, before using a target multi-task learning neural network model, inputting the target logging curve features of the target well section, and determining the target average pore radius of the target well section and the probability of the target well section belonging to each pore structure category, the following steps are included: Obtain a core sample dataset, which includes the pore structure category, logging curve characteristics, and average pore radius for each core sample. The logging curve characteristics include at least the clay content, deep lateral resistivity, compensated density, compensated neutron, P-wave transit time, S-wave transit time, and effective porosity. The target multi-task learning neural network model is obtained by training a preset multi-task learning neural network model based on the core sample dataset.

[0045] In this embodiment, the core sample dataset includes the pore structure category, logging curve characteristics, and average pore radius of each core sample. The logging curve characteristics include at least the clay content, deep lateral resistivity, compensated density, compensated neutrons, P-wave transit time, S-wave transit time, and effective porosity.

[0046] In practical applications, based on the rock physics response mechanism, conventional logging curve characteristic parameters that are highly sensitive to the development characteristics of reservoir pore structure are selected as model input features. Specifically, these include seven logging curve characteristic parameters: clay content, deep lateral resistivity, compensated density, compensated neutron, P-wave transit time, S-wave transit time, and effective porosity.

[0047] Furthermore, the measured logging sequence data undergoes systematic correction processing, including outlier removal and correction for wellbore environment and mud invasion. To eliminate training interference caused by differences in the dimensions of various features, the Z-score standardization formula Xstd=(X-μ) / σ can be used to normalize all input features, where μ is the mean of the corresponding feature and σ is the standard deviation of the corresponding feature.

[0048] Furthermore, a model-supervised labeling system was constructed based on the pore structure classification results obtained from core experiments. The predetermined pore structure category of the core samples was used as the true hard label for the classification task, and the label was digitized using One-Hot encoding. For example, the label corresponding to pore structure category II was [0,1,0,0,0]. Simultaneously, the measured average pore radius value of the core samples was selected as the continuous training label for the regression task.

[0049] Furthermore, the sample dataset after preprocessing and label matching is divided into training set and test set according to a preset ratio. The division process is implemented by stratified sampling to strictly ensure that the proportion of various pore structure samples in the training set and test set is consistent with the distribution of the original overall dataset, thus avoiding model training bias caused by sample distribution imbalance.

[0050] In addition, a pre-defined multi-task learning neural network model with a feature sharing mechanism was constructed. The overall model consists of a shared feature extraction backbone network and two task-specific branch networks. The shared feature extraction layer adopts a two-layer fully connected network structure. It takes 7-dimensional standardized well logging curve feature data as input and extracts general reservoir high-level features through fully connected layers with 128 neurons and 64 neurons respectively. The ReLU linear rectified function is used as the activation function for each layer.

[0051] Two types of functional branch networks are built in parallel behind the shared feature layer. The first is a soft classification branch for pore structure, which consists of a multi-layer fully connected network. The number of neurons in the output layer is consistent with the number of pore structure categories. The output end can be equipped with a Softmax activation function. The final output is a probability distribution vector with the same dimension as the number of pore structure categories. Each element in the probability distribution vector represents the probability that the input sample belongs to the corresponding pore structure category, and the sum of all probability values ​​is 1. The second is a regression prediction branch for average pore radius, which is connected after the shared layer and consists of several fully connected layers. Its output layer is a single neuron and uses a linear activation function to predict the continuous value of average pore radius, so as to achieve accurate fitting and prediction of the continuous measured value of the formation average pore radius.

[0052] Furthermore, a core sample dataset was collected and constructed. Each core sample in this dataset was synchronously matched with the corresponding pore structure classification, seven logging curve feature parameters, and measured average pore radius data. Using the constructed core sample dataset, a pre-built multi-task learning neural network model was iteratively trained. Feature learning and dual-task fitting were completed by constraining the network through a joint loss function. After parameter updates and convergence determination, a target multi-task learning neural network model with the ability to predict the probability of pore structure categories and the quantitative prediction of average pore radius was obtained.

[0053] Optionally, a preset multi-task learning neural network model is trained based on the core sample dataset to obtain a target multi-task learning neural network model, including: The core sample dataset is divided into a training set and a test set according to a preset ratio; The well logging curve features are used as input features, the pore structure category is used as a classification label, and the average pore radius is used as a regression label. The preset multi-task learning neural network model is trained based on the training set. The classification loss and regression loss of the pre-trained multi-task learning neural network model are determined based on the test set, and the parameters of the pre-trained multi-task learning neural network model are updated according to the classification loss and regression loss until the classification loss and regression loss remain unchanged or the preset training rounds are reached, thus obtaining the target multi-task learning neural network model.

[0054] In this embodiment, the preset ratio is a pre-set ratio, and the preset training rounds are the maximum number of iterations of the model. The preset ratio and preset training rounds can be flexibly set. In this embodiment, the preset ratio can be 8:2. The training set is a core sample dataset of a preset ratio used to train the preset multi-task learning neural network model, and the test set is a core sample dataset of a preset ratio used to update the parameters of the trained preset multi-task learning neural network model.

[0055] The classification loss is the degree of deviation between the predicted pore structure category label output by the preset multi-task learning neural network model and the true label of the pore structure category of the core sample, while the regression loss is the degree of deviation between the predicted average pore radius output by the preset multi-task learning neural network model and the true average pore radius of the core sample.

[0056] In this embodiment, the classification loss can be the cross-entropy loss L_class = -Σ Yi * log(Pi), used to measure the difference between the predicted probability distribution and the true hard label. The regression loss can be the mean squared error loss function L_reg = MSE(Rm_pred, Rm_true). The total loss function is the weighted sum of the classification loss and the regression loss, and the total loss L_total = λ * L_class + (1-λ) * L_reg, where λ is the weight coefficient that balances the two tasks, such as λ = 0.7.

[0057] Specifically, well logging curve features are used as input features, pore structure category is used as classification label, and average pore radius is used as regression label. Based on the training set, the pre-set multi-task learning neural network model is trained in an end-to-end overall mode. The network parameters are trained based on the partitioned training set. The Adam optimizer is used in conjunction with a mini-batch gradient descent strategy to update the network weights and bias parameters. Hyperparameters such as the initial learning rate, weight decay coefficient, and maximum number of training iterations are reasonably configured. At the same time, an early stopping constraint mechanism is introduced. When the total loss value of the validation set does not show a significant decreasing trend for several consecutive iterations or reaches the preset number of training iterations, the training process is automatically terminated, effectively suppressing the overfitting phenomenon of the model.

[0058] In this embodiment, although the preset multi-task learning neural network model relies on discrete category hard labels to complete parameter convergence optimization during the training phase, it uses cross-entropy loss to constrain the network to learn the distribution patterns of samples of different pore structure categories in the high-dimensional feature space during the training process. When the logging curve features of the input sample are near the category decision boundary, the model can output multi-category probability soft classification results based on the Softmax layer during the actual application inference phase. For strata samples in the category division boundary area, the output probability distribution shows a smooth transition feature. While clarifying the optimal pore structure category, it completely preserves the probability information of the sample belonging to the adjacent category strata, which is consistent with the actual geological characteristics of the continuous and gradual change of underground reservoir pore structure with stratum deposition and diagenesis. Compared with the single hard category determination result, it has stronger geological applicability and interpretive rationality.

[0059] S130. Based on the permeability calculation model corresponding to each pore structure category, the target average pore radius and target porosity are substituted to calculate the candidate permeability corresponding to each pore structure category.

[0060] In this embodiment, the target porosity is the porosity of the target well section, and the candidate permeability is the permeability value calculated by substituting the target porosity and the target average pore radius into the permeability calculation model corresponding to each pore structure category. In this embodiment, the number of pore structure categories corresponds one-to-one with the number of candidate permeabilities. If there are five pore structure categories, then there are five candidate permeability values.

[0061] In this embodiment, based on the predicted target average pore radius and target porosity of the target well section, and combined with the pre-established permeability calculation models for various pore structure categories, the candidate permeability of the target well section under each pore structure category can be obtained.

[0062] Specifically, the target porosity and target average pore radius predicted for the target well section are substituted into the dedicated permeability calculation models corresponding to the five pore structure categories for calculation and solution, and the candidate permeability corresponding to different pore structure categories is obtained one by one.

[0063] S140. Determine the target permeability of the target well section based on the probability of the pore structure category and the candidate permeability corresponding to each pore structure category.

[0064] In this embodiment, the target permeability is the predicted permeability of the target well section.

[0065] This embodiment can determine the dominant pore structure type of the target well section based on the principle of maximum probability, match the permeability calculation formula of the corresponding category, and use the target average pore radius and target porosity as input parameters into the permeability calculation model formula. After calculation, the target permeability value corresponding to the formation of the target well section is obtained, and the reservoir seepage capacity parameter is quantitatively obtained.

[0066] Alternatively, this embodiment can also combine the probability of the target well section belonging to each pore structure category, and use the sum of the products of the pore structure category probability and the permeability of the target well section under each pore structure category as the target permeability of the target well section.

[0067] The technical solution of this invention constructs a permeability calculation model for different pore structure categories in the study area, uses a trained target multi-task learning neural network model to determine the probability of the target well section belonging to each pore structure category and the target average pore radius, determines candidate permeabilities under different pore structure categories based on the target average pore radius and each permeability calculation model, and finally determines the target permeability of the target well section based on the candidate permeabilities under the pore structure category and the probability of the pore structure category. This avoids the problem of low permeability prediction accuracy of traditional single models or hard classification models in heterogeneous reservoirs, and improves the prediction accuracy and geological consistency of reservoir permeability.

[0068] Example 2 Figure 2 This is a flowchart illustrating a method for determining reservoir permeability according to Embodiment 2 of the present invention. This embodiment further specifies the method based on the previous embodiment. This embodiment is applicable to reservoir permeability determination. The method can be executed by a reservoir permeability determination device, which can be implemented in hardware and / or software and can be configured in any electronic device with network communication and computing capabilities. Figure 2 As shown, the method includes: S210. Determine the target logging curve characteristics, target porosity, and permeability calculation model (represented by porosity and average pore radius) for the target well section in the study area.

[0069] S220. Using a target multi-task learning neural network model, the target logging curve features of the target well section are input to obtain the target average pore radius of the target well section and the probability of the target well section belonging to each pore structure category; the target multi-task learning neural network model includes a pore structure classification branch and an average pore radius regression branch, the pore structure classification branch is used to output the probability vector of the input sample belonging to each type of pore structure category, and the average pore radius regression branch is used to output the average pore radius of the input sample.

[0070] S230. Based on the permeability calculation model corresponding to each pore structure category, the target average pore radius and target porosity are substituted to calculate the candidate permeability corresponding to each pore structure category.

[0071] S240. The weighted sum of the pore structure category probability and candidate permeability for each pore structure category is used as the target permeability for the target well section.

[0072] In this embodiment, a weighted summation operation is performed based on the probability weights corresponding to various pore structure categories and the corresponding candidate permeability values. Using the pore structure category probability as the weighting coefficient, the candidate permeabilities of all pore structure categories are weighted and fused. The final weighted summation result is the target permeability corresponding to the formation in the target well section. The target permeability in this embodiment can be expressed by the following formula: ; Where P_i is the probability that the target well section belongs to the i-th pore structure category, and K_i is the candidate permeability calculated by the permeability calculation model for the i-th pore structure category. For the target porosity, The target average pore radius, To achieve the target penetration rate, The depth point of the target well section.

[0073] In this embodiment, based on the probability of the pore structure category to which the target well section belongs, the permeability calculation model specific to each pore structure category is dynamically and smoothly integrated. This allows the permeability curves of all subsequent well sections to maintain a smooth and reasonable transition even in the abrupt change zone of physical properties, thus solving the problem of low prediction accuracy of traditional single models or hard classification models in heterogeneous reservoirs.

[0074] Furthermore, the above process can be used to calculate the permeability of the entire target well section, obtain the permeability curve of the entire target well section, and ensure that the permeability curve of the subsequent predicted entire well section can maintain a smooth and reasonable transition in the abrupt change zone of physical properties.

[0075] In practical applications, the predicted probability distribution of pore structure categories can be presented on the well logging composite chart with different color filling methods, which can intuitively show the vertical distribution of high-quality reservoirs (such as pore structure categories I and II) and the development of interlayers (such as pore structure categories IV and V), thus improving the visualization effect.

[0076] In this embodiment, the final target permeability is obtained by weighting and fusing the corresponding candidate permeabilities with the probability of each pore structure category as a weight. This fully takes into account the actual geological characteristics of the gradual transition of reservoir pore structure, effectively weakens the value deviation caused by a single rigid classification, and comprehensively integrates the seepage parameter response laws under different pore structure development states. It not only makes full use of the computational advantages of the multi-category permeability calculation model, but also uses probability weights to reflect the actual stratum attribution tendency, which greatly improves the rationality and geological fit of the permeability calculation results. This makes the obtained target permeability more consistent with the heterogeneous development characteristics of underground reservoirs, and further ensures the accuracy and reliability of the quantitative evaluation results of reservoir physical parameters.

[0077] The technical solution of this invention constructs a permeability calculation model for different pore structure categories in the study area, uses a trained target multi-task learning neural network model to determine the probability of the target well section belonging to each pore structure category and the target average pore radius, determines the candidate permeability under each pore structure category based on the target average pore radius and each permeability calculation model, and finally determines the target permeability of the target well section based on the weighted sum of the candidate permeability under the pore structure category and the probability of the pore structure category. This avoids the problem of low permeability prediction accuracy of traditional single models or hard classification models in heterogeneous reservoirs, and improves the accuracy and geological consistency of the quantitative calculation results of reservoir permeability.

[0078] This embodiment takes the deep, low-permeability gravelly sandstone reservoir in the XX slope zone of the Xihu Depression in a certain sea area in eastern China as an example to illustrate the implementation process of the present invention in detail.

[0079] Specifically, the reservoir pore structure was first classified and a dedicated permeability calculation model was constructed based on core experiments.

[0080] This embodiment collected experimental data from 50 core samples from two cored wells (Well XX-1-1 and Well XX-1-2) in the study area, including high-pressure mercury intrusion porosimetry curves, nuclear magnetic resonance T2 spectra, thin-section images of the cast bodies, and conventional physical property (porosity, permeability) analysis data. All core depths were precisely mapped to correspond one-to-one with well logging depths.

[0081] Next, analysis of 50 high-pressure mercury intrusion curves revealed a wide distribution of their morphologies. Based on key characteristic parameters of the curves, particularly the displacement pressure (Pd) and maximum mercury intrusion saturation (SHg), and according to preset threshold values ​​for these parameters, the reservoir pore structure was classified into five categories, representing an evolution trend from favorable to unfavorable: When Pd < 0.1 MPa and SHg > 90%, the reservoir pore structure is classified as Class I pore structure, characterized by low displacement pressure, high mercury saturation, and a broad plateau in the curve, indicating large pore throats and good sorting.

[0082] When Pd is between 0.1 and 0.4 MPa and SHg is between 85 and 90%, the reservoir pore structure is classified as Class II pore structure, characterized by moderate displacement pressure and moderate mercury saturation.

[0083] When Pd is between 0.3 and 0.6 MPa and SHg is between 80 and 85%, the reservoir pore structure is classified as Class III pore structure, characterized by moderate displacement pressure and moderate mercury saturation.

[0084] When Pd is between 0.6 and 1.1 MPa and SHg is between 75 and 80%, the reservoir pore structure is classified as Class IV, characterized by high displacement pressure and moderate mercury saturation.

[0085] When the pressure is greater than 1.1 MPa and SHg is less than 75%, the reservoir pore structure is classified as Class V, characterized by high displacement pressure, low mercury saturation, steep curves, and no plateaus, indicating small pore throats and poor sorting.

[0086] Furthermore, the standard high-pressure mercury intrusion curve morphology for each type of pore structure is summarized and determined, the displacement pressure range and maximum mercury saturation value distribution range for each type of pore structure are clarified, and a category-specific mercury intrusion parameter judgment standard is established to achieve the characterization of mercury intrusion experimental features.

[0087] The NMR characteristics from Type I to Type V are as follows: the NMR T2 spectrum morphology gradually evolves from "bimodal long relaxation high amplitude" (main peak > 200 ms) to "bimodal short relaxation low amplitude" (main peak about 1 ms), which directly reflects the reduction of the pore volume of the mobile fluid and the increase of micropores.

[0088] The thin-film characteristics from Class I to Class V are as follows: Class I reservoirs are dominated by primary intergranular pores and intragranular dissolution pores, with chlorite films visible on the particle surface; Class II reservoirs are dominated by primary intergranular pores with good connectivity; Class III reservoirs have reduced porosity and enlarged quartz is visible; Class IV and Class V reservoirs are dominated by micropores, with enlarged quartz being common and clay filling being obvious.

[0089] See Table 1 for a statistical table of reservoir characteristics for different pore structure categories.

[0090] Table 1. Statistical table of reservoir characteristics for different pore structure categories. The physical properties from Class I to Class V are as follows: the average porosity decreases from 15.8% to 4.2%, the average permeability decreases from 139.6 mD to 0.17 mD, the bound water saturation increases from 34.7% to 93.0%, and the average pore radius decreases from 6.8 μm to 0.1 μm. All parameters show regular changes, proving that the classification is effective.

[0091] Furthermore, a specific permeability calculation model was constructed for each type of pore structure. For core data of each pore structure type, porosity (φ) and average pore radius (Rm) were used as independent variables, and permeability (K) was used as the dependent variable. Nonlinear regression fitting was performed to obtain five sets of specific permeability calculation models. See Table 2 for the specific permeability calculation models for reservoirs with different pore structure types.

[0092] Table 2. Specific permeability calculation models for reservoirs with different pore structure types. For example, for reservoirs with a porosity type I, the permeability calculation model is K = 924.286 * φ 1.078 * Rm 0.622 Compared to the unclassified single penetration rate calculation model, the classified penetration rate calculation model has a better fit to the data (R²). 2 The significant improvement indicates that it more accurately describes the seepage mechanism under different pore structure categories.

[0093] Furthermore, a pre-defined multi-task learning neural network model was constructed and trained. First, seven characteristic parameters of the logging curves corresponding to core samples at 50 core depths were extracted from the logging curves, including vegetative shale content (VSH), deep lateral resistivity (RT), compensated density (DEN), compensated neutron (CNL), P-wave transit time (DTC), S-wave transit time (DTS), and effective porosity (PHIE).

[0094] Next, Z-score standardization was performed on all seven logging curve features, calculated using the formula X_std = (X - μ) / σ, where μ and σ are the mean and standard deviation of the feature on the training set, respectively. The standardization parameters need to be saved for later application to the full well data.

[0095] Next, the classification results of the pore structure categories of the core samples (Class I-V) were transformed into 5-dimensional One-Hot vectors as classification labels, and the measured average pore radius (Rm) was used as the regression label.

[0096] A stratified sampling method was used to randomly divide 50 core samples into a training set (40 core samples) and a test set (10 core samples) at a ratio of 8:2, ensuring that the proportions of the five pore structure categories in the training set and the test set were consistent with the overall population.

[0097] Furthermore, a pre-defined multi-task learning neural network model is constructed, see [link to relevant documentation]. Figure 3 The diagram illustrates the architecture of a pre-defined multi-task learning neural network model. The input layer has 7 neurons, and the shared feature extraction layer consists of two fully connected layers: the first layer has 128 neurons, and the second layer has 64 neurons, both using the ReLU activation function. The classification branch follows the 64-dimensional shared features with a 5-neuron fully connected layer, using the Softmax activation function, outputting a 5-dimensional probability vector for pore structure categories. The regression branch follows the 64-dimensional shared features with a 1-neuron fully connected layer, using a linear activation function, outputting the predicted average pore radius.

[0098] The total loss function is set as L_total = λ * L_class + (1-λ) * L_reg. Here, the classification loss L_class is the cross-entropy, the regression loss L_reg is the mean squared error, and the weighting coefficient λ is set to 0.7 to emphasize the dominant role of the classification task while also taking into account the auxiliary role of the average pore radius regression.

[0099] The Adam optimizer was used with an initial learning rate of 0.01, weight decay of 0.01, a batch size of 8, and a preset training epoch count of 64. An early stopping strategy was employed: training was stopped and the target multi-task learning neural network model was saved when the validation set loss no longer decreased for 10 consecutive epochs. As the number of training epochs increased, the losses on both the training and validation sets steadily decreased and eventually leveled off, indicating stable model training without significant overfitting or underfitting.

[0100] Furthermore, the preprocessed seven logging curves of the entire target well (e.g., well XX-1-2) are input point by point into the trained target multi-task learning neural network model. The target multi-task learning neural network outputs the probabilities P1 to P5 of each depth point in the target well section belonging to pore structure categories I to V. For example, for a depth point in a certain target well section, the target multi-task learning neural network model may output P=[0.05,0.60,0.30,0.04,0.01], indicating that the depth point in the target well section is most likely to belong to pore structure category II, but there is a 30% probability that it belongs to category III.

[0101] Table 3 shows the probability distribution of the actual pore structure classification and predicted pore structure category of the test set core samples. The category with the highest predicted probability for most samples is consistent with the original label. The few inconsistent samples (such as core sample No. 3) are basically located in the transition zone of the category boundary. The target multi-task learning neural network model retains this geological uncertainty through probabilistic output.

[0102] Table 3. Classification of actual pore structure and probability distribution of predicted pore structure categories in core samples from the test set. Furthermore, for the depth point of the target well section, based on the obtained pore structure category probability vector P and the average pore radius Rm_pred predicted by the regression branch, combined with the five specific permeability calculation models, the final target permeability K_final of the target well section is calculated using a probability weighted method: K_final = P1*K1 + P2*K2 + P3*K 3+ P4*K4+ P5*K5; Repeat this process to obtain the permeability at each depth point throughout the target well, and compare the calculated permeability curve with the measured permeability from the core sample.

[0103] See Figure 4 The diagram shows a comparison between the predicted and measured permeability of the entire target well section. The data points are closely distributed around the 45° diagonal, indicating that the prediction accuracy is high and the error is within an acceptable range.

[0104] Applying the method of this invention to the P8 high-quality reservoir, the resistivity response of the P8 layer is influenced by lithology, physical properties, gas content, and pore structure, making identification difficult. The pore structure categories predicted by this method are mainly Type I and Type II, with interlayers of Type IV and V interlayers approximately 10m in length, which is highly consistent with the morphological changes in the T2 NMR spectrum. The method predicts a permeability of 60mD in the sweet spot region. Subsequent casing perforation DST testing in this region yielded high production, validating the reliability of this method.

[0105] Example 3 Figure 5 This is a schematic diagram of a reservoir permeability determination device provided in Embodiment 3 of the present invention. This embodiment is applicable to reservoir permeability determination. The reservoir permeability determination device can be implemented in hardware and / or software, and can be configured in any electronic device with network communication and computing capabilities. Figure 5 As shown, the device includes: The data determination module 310 is used to determine the target logging curve characteristics, target porosity, and permeability calculation model represented by porosity and average pore radius for each type of pore structure in the target area of ​​the study area. The pore radius and probability determination module 320 is used to utilize a target multi-task learning neural network model, input the target logging curve features of the target well section, and output the target average pore radius of the target well section and the probability of the target well section belonging to each pore structure category; the target multi-task learning neural network model includes a pore structure classification branch and an average pore radius regression branch, the pore structure classification branch is used to output the probability vector of the input sample belonging to each pore structure category, and the average pore radius regression branch is used to output the average pore radius of the input sample; The candidate permeability determination module 330 is used to calculate the candidate permeability corresponding to each pore structure category by substituting the target average pore radius and the target porosity into the permeability calculation model corresponding to each pore structure category. The target permeability determination module 340 is used to determine the target permeability of the target well section based on the probability of the pore structure category and the candidate permeability corresponding to each pore structure category.

[0106] Optionally, before determining the permeability calculation model represented by porosity and average pore radius for each pore structure category in the study area, the following steps are included: Determine the high-pressure mercury intrusion curve morphology, displacement pressure, maximum mercury saturation, average pore radius, and pore throat sorting coefficient of core samples from each candidate well section in the study area; Based on at least two parameters among the high-pressure mercury intrusion curve morphology, displacement pressure, maximum mercury saturation, average pore radius, and pore throat sorting coefficient, and a preset parameter threshold, the reservoirs in the study area are classified into at least five pore structure categories, from high-quality reservoirs to non-reservoirs.

[0107] Optionally, determine the permeability calculation model, represented by porosity and average pore radius, for each pore structure category in the study area, including: Porosity, average pore radius, and permeability of core samples corresponding to each pore structure type in the study area were obtained. Based on the porosity, average pore radius, and permeability, a permeability calculation model, represented by porosity and average pore radius, is determined for each pore structure category through nonlinear regression fitting.

[0108] Optionally, the permeability calculation model is represented by the following formula: ; in, For penetration rate, Porosity The average pore radius, , , is the fitting coefficient corresponding to the i-th pore structure category.

[0109] Optionally, before using a target multi-task learning neural network model, inputting the target logging curve features of the target well section, and determining the target average pore radius of the target well section and the probability of the target well section belonging to each pore structure category, the following steps are included: Obtain a core sample dataset, which includes the pore structure category, logging curve characteristics, and average pore radius for each core sample. The logging curve characteristics include at least the clay content, deep lateral resistivity, compensated density, compensated neutron, P-wave transit time, S-wave transit time, and effective porosity. The target multi-task learning neural network model is obtained by training a preset multi-task learning neural network model based on the core sample dataset.

[0110] Optionally, a preset multi-task learning neural network model is trained based on the core sample dataset to obtain a target multi-task learning neural network model, including: The core sample dataset is divided into a training set and a test set according to a preset ratio; The well logging curve features are used as input features, the pore structure category is used as a classification label, and the average pore radius is used as a regression label. The preset multi-task learning neural network model is trained based on the training set. The classification loss and regression loss of the pre-trained multi-task learning neural network model are determined based on the test set, and the parameters of the pre-trained multi-task learning neural network model are updated according to the classification loss and regression loss until the classification loss and regression loss remain unchanged or the preset training rounds are reached, thus obtaining the target multi-task learning neural network model.

[0111] Optionally, the target permeability of the target well section is determined based on the probability of the pore structure category and the candidate permeability corresponding to each pore structure category, including: The target permeability of the target well section is obtained by weighting the probability of the pore structure category and the candidate permeability for each pore structure category.

[0112] The technical solution of this invention constructs a permeability calculation model for different pore structure categories in the study area, uses a trained target multi-task learning neural network model to determine the probability of the target well section belonging to each pore structure category and the target average pore radius, determines candidate permeabilities under different pore structure categories based on the target average pore radius and each permeability calculation model, and finally determines the target permeability of the target well section based on the candidate permeabilities under the pore structure category and the probability of the pore structure category. This avoids the problem of low permeability prediction accuracy of traditional single models or hard classification models in heterogeneous reservoirs, and improves the prediction accuracy and geological consistency of reservoir permeability.

[0113] The reservoir permeability determination device provided in the embodiments of the present invention can execute the reservoir permeability determination method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the method.

[0114] Example 4 Figure 6 A schematic diagram of an electronic device 10, which can be used to implement embodiments of the present invention, is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0115] like Figure 6As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0116] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0117] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as reservoir permeability determination methods.

[0118] In some embodiments, the reservoir permeability determination method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or mounted on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the reservoir permeability determination method described above may be performed. Alternatively, in other embodiments, processor 11 may be configured to perform the reservoir permeability determination method by any other suitable means (e.g., by means of firmware).

[0119] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication unit 19, or installed from storage unit 18, or installed from ROM 12. When the computer program is executed by processor 11, it performs the functions defined in the methods of the embodiments of the present invention.

[0120] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0121] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0122] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0123] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0124] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0125] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0126] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0127] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for determining reservoir permeability, characterized in that, include: Determine the target logging curve characteristics, target porosity, and permeability calculation model (represented by porosity and average pore radius) for each pore structure category in the study area; Using a target multi-task learning neural network model, the target logging curve features of the target well section are input to obtain the target average pore radius of the target well section and the probability of the target well section belonging to each pore structure category. The target multi-task learning neural network model includes a pore structure classification branch and an average pore radius regression branch. The pore structure classification branch is used to output the probability vector of the input sample belonging to each pore structure category, and the average pore radius regression branch is used to output the average pore radius of the input sample. Based on the permeability calculation model corresponding to each pore structure category, the target average pore radius and target porosity are substituted into the calculation model to obtain the candidate permeability corresponding to each pore structure category; The target permeability of the target well section is determined based on the probability of the pore structure category and the candidate permeability corresponding to each pore structure category.

2. The method according to claim 1, characterized in that, Before determining the permeability calculation model, represented by porosity and average pore radius, for each pore structure category in the study area, the following steps are included: The high-pressure mercury intrusion curve morphology, displacement pressure, maximum mercury saturation, average pore radius, and pore throat sorting coefficient of core samples from each candidate well section in the study area were determined. Based on at least two parameters among the high-pressure mercury intrusion curve morphology, displacement pressure, maximum mercury saturation, average pore radius, and pore throat sorting coefficient, and a preset parameter threshold, the reservoirs in the study area are classified into at least five pore structure categories, from high-quality reservoirs to non-reservoirs.

3. The method according to claim 1, characterized in that, Determine the permeability calculation model, represented by porosity and average pore radius, for each pore structure category in the study area, including: Porosity, average pore radius, and permeability of core samples corresponding to each pore structure type in the study area were obtained. Based on the porosity, average pore radius, and permeability, a permeability calculation model, represented by porosity and average pore radius, is determined for each pore structure category through nonlinear regression fitting.

4. The method according to claim 3, characterized in that, The permeability calculation model is expressed by the following formula: ； in, For penetration rate, Porosity The average pore radius, , , is the fitting coefficient corresponding to the i-th pore structure category.

5. The method according to claim 1, characterized in that, Before using a target multi-task learning neural network model, inputting the target well logging curve features of the target well section, and determining the target average pore radius and the probability of the target well section belonging to each pore structure category, the following steps are included: Obtain a core sample dataset, which includes the pore structure category, logging curve characteristics, and average pore radius for each core sample. The logging curve characteristics include at least the clay content, deep lateral resistivity, compensated density, compensated neutron, P-wave transit time, S-wave transit time, and effective porosity. The target multi-task learning neural network model is obtained by training a preset multi-task learning neural network model based on the core sample dataset.

6. The method according to claim 5, characterized in that, The target multi-task learning neural network model is obtained by training a preset multi-task learning neural network model based on the core sample dataset, including: The core sample dataset is divided into a training set and a test set according to a preset ratio; The well logging curve features are used as input features, the pore structure category is used as a classification label, and the average pore radius is used as a regression label. The preset multi-task learning neural network model is trained based on the training set. The classification loss and regression loss of the pre-trained multi-task learning neural network model are determined based on the test set, and the parameters of the pre-trained multi-task learning neural network model are updated according to the classification loss and regression loss until the classification loss and regression loss remain unchanged or the preset training rounds are reached, thus obtaining the target multi-task learning neural network model.

7. The method according to claim 1, characterized in that, Based on the probability of the pore structure category and the candidate permeability corresponding to each pore structure category, the target permeability of the target well section is determined, including: The target permeability of the target well section is obtained by weighting the probability of the pore structure category and the candidate permeability for each pore structure category.

8. A reservoir permeability determination device, characterized in that, include: The data determination module is used to determine the target logging curve characteristics, target porosity, and permeability calculation model (represented by porosity and average pore radius) for each pore structure category in the study area. The pore radius and probability determination module is used to utilize a target multi-task learning neural network model, inputting the target logging curve features of the target well section, and outputting the target average pore radius of the target well section and the probability of the target well section belonging to each pore structure category; the target multi-task learning neural network model includes a pore structure classification branch and an average pore radius regression branch, the pore structure classification branch is used to output the probability vector of the input sample belonging to each pore structure category, and the average pore radius regression branch is used to output the average pore radius of the input sample; The candidate permeability determination module is used to calculate the candidate permeability corresponding to each pore structure category by substituting the target average pore radius and target porosity into the permeability calculation model corresponding to each pore structure category. The target permeability determination module is used to determine the target permeability of the target well section based on the probability of the pore structure category and the candidate permeability corresponding to each pore structure category.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the reservoir permeability determination method as described in any one of claims 1-7.

10. A storage medium for storing computer-executable instructions, characterized in that, The computer-executable instructions, when executed by a computer processor, are used to perform the reservoir permeability determination method as described in any one of claims 1-7.

Images