A reservoir capillary pressure curve prediction method, device, equipment and storage medium

Abstract

The application discloses a reservoir capillary pressure curve prediction method, device and equipment and a storage medium, is applied to the field of oil and gas exploration, carries out feature dimension reduction processing on conventional logging data to extract core features, determines a reservoir quality factor based on physical parameters, determines the fuzzy probability of the reservoir category of each depth point based on the reservoir quality factor and a fuzzy membership function; the core features and the fuzzy probability are spliced to generate an input feature vector, and the input feature vector is input into a prediction model to obtain a capillary pressure curve prediction result; the model loss value is determined based on the capillary pressure curve of each depth point and the capillary pressure curve prediction result, and the model parameters are updated; the to-be-measured data is input into the trained prediction model to obtain a target capillary pressure curve prediction result. The rock physical prior constraint based on the reservoir quality factor is applied to the prediction model, invalid search in the parameter space is limited, and the generalization ability and prediction accuracy of the prediction model in the dense heterogeneous formation are improved.

Description

Technical Field

[0001] This invention relates to the field of oil and gas exploration, and in particular to a method for predicting reservoir capillary pressure curves, a device for predicting reservoir capillary pressure curves, electronic equipment, and a computer-readable storage medium. Background Technology

[0002] Capillary pressure curves can quantitatively reflect the size, distribution, and connectivity of pore throats within rocks, making them the most direct and effective means of evaluating the microstructure of reservoirs. However, core mercury intrusion porosimetry (CIMS) is costly, has a long measurement cycle, and the number of downhole corings is limited, failing to meet the need for continuous evaluation of reservoir pore structure across the entire well section.

[0003] For predicting capillary pressure curves using conventional well logging, existing technologies mainly rely on empirical formulas and multivariate statistical regression methods. These prediction models are highly dependent on accurate input of physical properties, but the accuracy of permeability calculations for tight sandstone reservoirs is generally low. This leads to amplified prediction errors in subsequent capillary pressure curves. Furthermore, a complex nonlinear relationship exists between conventional well logging curves and the pore structure of tight sandstone. Multivariate linear regression or simple empirical formulas are insufficient to effectively characterize the gradual changes and cross-scale variations in the pore network of highly heterogeneous reservoirs, resulting in underestimation of high-quality reservoirs or overestimation of tight layers. Summary of the Invention

[0004] The purpose of this invention is to provide a method, apparatus, equipment, and storage medium for predicting reservoir capillary pressure curves, applicable to the field of oil and gas exploration. This method applies rock physics prior constraints based on reservoir quality factors to the prediction model, thereby limiting invalid searches in the parameter space and improving the generalization ability and prediction accuracy of the prediction model in tight heterogeneous strata.

[0005] To solve the above-mentioned technical problems, the present invention provides a method for predicting reservoir capillary pressure curves, comprising: Conventional logging data and physical property parameters at various depths of the target reservoir are obtained as training samples, and capillary pressure curves at the corresponding depths are obtained as training labels. The conventional logging data is subjected to feature dimensionality reduction processing to extract core features. Based on the physical property parameters, the reservoir quality factor is determined. Based on the reservoir quality factor and the fuzzy membership function, the fuzzy probability of the reservoir category at each depth point is determined. The core features and the fuzzy probabilities are concatenated to generate an input feature vector, and the input feature vector is input into the prediction model to obtain the capillary pressure curve prediction result. The model loss value is determined based on the capillary pressure curve at each depth point and the prediction result of the capillary pressure curve. The model parameters are updated based on the model loss value until the training of the prediction model is obtained. After processing the conventional logging data and physical property parameters of the depth point to be measured, the data is input into the trained prediction model to obtain the prediction result of the target capillary pressure curve.

[0006] Optionally, determining the fuzzy probability of the reservoir category at each depth point based on the reservoir quality factor and fuzzy membership function includes: A fuzzy membership function for each reservoir category is constructed based on piecewise linear interpolation; Based on the reservoir quality factor classification standard and transition bandwidth, the reservoir quality factor at the depth point is input into the fuzzy membership function to determine the fuzzy probability of the reservoir category.

[0007] Optionally, the core features and the fuzzy probabilities are concatenated to generate an input feature vector, including: The core features, the fuzzy probability, the reservoir quality factor, and the mercury saturation sequence are concatenated to generate the input feature vector.

[0008] Optionally, the conventional well logging data can be subjected to feature dimensionality reduction processing to extract core features, including: Principal component analysis is used to perform feature dimensionality reduction on the conventional well logging data, and the core features are extracted from the conventional well logging data.

[0009] Optional, conventional logging data include natural gamma, sonic transit time, compensated neutrons, density, deep lateral resistivity, and shallow lateral resistivity.

[0010] Optionally, the physical properties include porosity and permeability.

[0011] Optionally, the prediction model is a bidirectional long short-term memory network.

[0012] To solve the above-mentioned technical problems, the present invention provides a reservoir capillary pressure curve prediction device, comprising: The first module is used to acquire conventional logging data and physical property parameters at various depths of the target reservoir as training samples, and to acquire capillary pressure curves at the corresponding depths as training labels. The second module is used to perform feature dimensionality reduction processing on the conventional logging data to extract core features, determine the reservoir quality factor based on the physical property parameters, and determine the fuzzy probability of the reservoir category at each depth point based on the reservoir quality factor and the fuzzy membership function. The third module is used to concatenate the core features and the fuzzy probabilities to generate an input feature vector, and input the input feature vector into the prediction model to obtain the capillary pressure curve prediction result. The fourth module is used to determine the model loss value based on the capillary pressure curve at each depth point and the prediction result of the capillary pressure curve, and update the model parameters based on the model loss value until the training completed prediction model is obtained. The fifth module is used to process the conventional logging data and physical property parameters of the depth point to be measured and input them into the trained prediction model to obtain the prediction result of the target capillary pressure curve.

[0013] To solve the above-mentioned technical problems, the present invention provides an electronic device, comprising: Memory, used to store computer programs; A processor is used to implement the reservoir capillary pressure curve prediction method when executing the computer program.

[0014] To address the aforementioned technical problems, the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the reservoir capillary pressure curve prediction method.

[0015] As can be seen, this invention obtains conventional logging data and physical property parameters at various depths of the target reservoir as training samples, and obtains capillary pressure curves at the corresponding depths as training labels; it performs feature dimensionality reduction processing on the conventional logging data to extract core features, determines the reservoir quality factor based on the physical property parameters, and determines the fuzzy probability of the reservoir category at each depth based on the reservoir quality factor and fuzzy membership function; it concatenates the core features and fuzzy probabilities to generate an input feature vector, and inputs the input feature vector into the prediction model to obtain the capillary pressure curve prediction result; it determines the model loss value based on the capillary pressure curves at each depth and the capillary pressure curve prediction result, and updates the model parameters based on the model loss value until a trained prediction model is obtained; it processes the conventional logging data and physical property parameters at the depth to be measured and inputs them into the trained prediction model to obtain the target capillary pressure curve prediction result.

[0016] This invention improves the generalization ability and prediction accuracy of the prediction model in tight heterogeneous strata by imposing rock physics prior constraints based on reservoir quality factors on the prediction model, thereby limiting invalid searches in the parameter space. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0018] Figure 1 A flowchart of a reservoir capillary pressure curve prediction method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the variance contribution rate and cumulative contribution rate of each principal component provided in an embodiment of the present invention. Figure 3 This is an example diagram of a model architecture provided in an embodiment of the present invention; Figure 4 This is a structural block diagram of a reservoir capillary pressure curve prediction device provided in an embodiment of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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 are within the scope of protection of the present invention.

[0020] Accurately characterizing the microstructure of pores is crucial for evaluating the quality of unconventional reservoirs such as tight sandstone. Mercury injection capillary pressure (MICP) curves obtained from high-pressure mercury intrusion porosimetry (HSI) can quantitatively reflect the size, distribution, and connectivity of pore throats within the rock, making it the most direct and effective means of evaluating reservoir microstructure. However, core mercury intrusion porosimetry is costly, time-consuming, and limited in the number of downhole corings, failing to meet the need for continuous evaluation of reservoir pore structure across the entire well section.

[0021] Currently, nuclear magnetic resonance (NMR) logging T2 distribution is commonly used in well logging evaluation to predict capillary pressure curves. Based on NMR T2 distribution or echo signals, domestic and international scholars have developed various T2-MICP transformation models, ranging from single / variable coefficient linear transformations and power function transformations to multi-parameter regression and artificial neural networks, achieving a certain degree of continuous characterization of downhole reservoir pore structure. However, NMR logging operations are costly, and in actual oil and gas exploration and development, not all well sections have NMR or other specialized logging data.

[0022] For predicting capillary pressure curves using conventional well logging, existing technologies mainly rely on empirical relationships and multivariate statistical regression methods. These primarily include two approaches: first, using empirical relationships between porosity and permeability to predict pore throat radii at different saturation levels to construct capillary pressure curves; and second, establishing multivariate regression relationships between mercury ingress saturation and reservoir physical properties at various pressure points. However, these traditional methods often exhibit extremely poor applicability when applied to tight sandstone reservoirs. Their limitations are mainly reflected in the following aspects: First, existing prediction models heavily rely on accurate input of physical properties (porosity, permeability), while the permeability calculation accuracy for tight sandstone reservoirs is generally low, which further amplifies the prediction error of subsequent capillary pressure curves. Second, there is a complex nonlinear relationship between conventional well logging curves and the pore structure of tight sandstone. Multivariate linear regression or simple empirical formulas are insufficient to effectively characterize the gradual characteristics and cross-scale changes of the pore network in strongly heterogeneous reservoirs, leading to underestimation of high-quality reservoirs or overestimation of tight layers.

[0023] The following combination Figure 1 , Figure 1 A flowchart of a reservoir capillary pressure curve prediction method provided in an embodiment of the present invention, the method may include: S101: Obtain conventional logging data and physical property parameters at each depth point of the target reservoir as training samples, and obtain the capillary pressure curves at the corresponding depth points as training labels.

[0024] This embodiment can prepare a dataset for model training. Specifically, it can collect conventional logging data and physical property parameters at various depths of the target reservoir as training samples. Conventional logging data may include natural gamma (GR), acoustic transit time (AC), compensated neutron (CNL), density (DEN), deep lateral resistivity (RT), and shallow lateral resistivity (RXO). Physical property parameters may include porosity (…). ) and penetration rate (K).

[0025] This embodiment can obtain the capillary pressure curve at the corresponding depth point as a training label. Since the capillary pressure curve is a continuous process that dynamically evolves with the injection of the non-wetting phase fluid, in order to adapt to the sequence modeling requirements of the time-series neural network, this embodiment can obtain the capillary pressure curve data at the corresponding depth, use an interpolation algorithm to discretize the original curve, and extract the capillary pressure (P) corresponding to each mercury saturation node under the mercury saturation sequence. c The value serves as a dynamic label for the temporal training of the network model.

[0026] This embodiment does not limit the number of mercury saturation nodes in the mercury saturation sequence. The number can be set based on the actual application and can generally include 23 mercury saturation nodes.

[0027] Furthermore, to avoid the normalized data from clustering around 0, this embodiment can pre-calculate K, RT, RXO, and capillary pressure (P), which are the prediction targets, if their numerical ranges are large. c The result was logarithmized to base 10.

[0028] Because the physical dimensions and numerical ranges of the various logging curves differ significantly, directly inputting them into the network would lead to unbalanced model weight updates. Therefore, a min-max normalization method is used to linearly map all input and target parameters to the dimensionless interval [0,1]. To make the prediction results more intuitively reflect the actual situation, this embodiment can inversely normalize the prediction results output by the model.

[0029] S102: Core features are extracted by performing feature dimensionality reduction processing on conventional logging data. Reservoir quality factors are determined based on physical property parameters. Fuzzy probabilities of reservoir categories at each depth point are determined based on reservoir quality factors and fuzzy membership functions.

[0030] Because conventional well logging data exhibits strong multicollinearity, directly inputting it into a neural network increases model complexity and the risk of overfitting. Principal Component Analysis (PCA) can perform orthogonal transformation and dimensionality reduction on the original well logging data, simplifying complex high-dimensional data into low-dimensional data while preserving important information from the original dataset. This achieves noise reduction and simplification of the input space, and the calculation results are only related to the data itself, requiring no human intervention. The cumulative contribution rate threshold in this embodiment can be set based on the actual application, for example, it can be set to 85%.

[0031] Therefore, this embodiment can perform feature dimensionality reduction processing on conventional well logging data through principal component analysis to extract core features from the conventional well logging data.

[0032] In one example, Figure 2 This diagram illustrates the variance contribution rate and cumulative contribution rate of each principal component. Figure 2 It can be seen that the principal components (PCs) reconstructed through orthogonal transformation are independent of each other. PC1-PC8 are the results of principal component analysis on the well logging data. Each principal component integrates multiple well logging data, with the first principal component (PC1) having the largest variance contribution rate (57.8%). When only the first four core principal components (PC1~PC4) are extracted, their cumulative variance contribution rate reaches as high as 93.1%, exceeding the 85% effective information retention threshold. This step reduces the feature dimension by half while achieving lossless preservation and decorrelation of well logging information.

[0033] By analyzing the correlation of high-dimensional spatial variables composed of well logging data, and with minimal loss of rock physical information, the high-dimensional well logging variables are transformed into low-dimensional composite variables, which are called principal components. This effectively removes redundant information between input features while preserving most of the variance of the dataset.

[0034] Purely data-driven deep learning models are prone to getting trapped in local optima when dealing with complex nonlinear mappings and lack clear geological and physical significance. The capillary pressure curve morphology of tight sandstone reservoirs is mainly controlled by the micropore throat radius. According to the Kozeny-Carman equation, the reservoir quality factor (RQI) integrates physical properties such as porosity and permeability with a direct physical relationship to the micropore throat radius. Therefore, the RQI and its derived fuzzy classification probability are introduced into the feature space of the prediction model.

[0035] This embodiment can construct fuzzy membership functions for each reservoir category based on piecewise linear interpolation; based on the reservoir quality factor classification standard and transition bandwidth, the reservoir quality factor at the depth point is input into the fuzzy membership function to determine the fuzzy probability of the reservoir category.

[0036] Because most classification modeling methods use hard classification, that is, absolutely dividing the strata into classes I, II, and III according to RQI, this can lead to a step-like abrupt change in network input characteristics in the physical property transition zones of tight sandstone (such as the boundary between class I and class II reservoirs), resulting in abrupt changes in the predicted capillary pressure curve that violate the laws of geological continuity.

[0037] Therefore, this embodiment can utilize fuzzy logic to calculate the membership probability of RQI for different reservoir categories, achieving a shift from hard classification to soft classification. Based on piecewise linear interpolation, it constructs the membership probability of each depth point to class I (P... b1 Class III (P) b3 ) and Class II (P b2 The fuzzy membership function of the reservoir is shown in the following formula: ; ; ; In the formula, x is the RQI value of the depth point. Let this be the fuzzy probability that the depth point belongs to class I. This represents the fuzzy probability that the depth point belongs to Class II. This represents the fuzzy probability that the depth point belongs to Class III. The classification criteria for Class I and Class II reservoirs The half-width of the transition zone between Class I and Class II reservoirs. The classification criteria for Class II and Class III reservoirs The transition zone half-width between Class II and Class III reservoirs is generally... 0.13 can be taken. It can be taken as 0.02. It can be taken as 0.077. You can take 0.015.

[0038] S103: The core features and fuzzy probabilities are concatenated to generate an input feature vector, which is then input into the prediction model to obtain the capillary pressure curve prediction result.

[0039] In well logging evaluation of tight sandstone, the capillary pressure curve is not a single static parameter, but a dynamic sequence that increases with the mercury injection saturation sequence. Traditional neural networks are prone to gradient vanishing when processing long sequences, making it difficult to capture the global features of the curve. This embodiment maps the set mercury injection saturation nodes to time steps in time series prediction, and uses a prediction model to perform sequence modeling of the mercury injection process.

[0040] In this embodiment, feature concatenation can generally concatenate the core features, RQI value, fuzzy probability, and mercury saturation sequence value at each time step. Specifically, the feature vector obtained from concatenation at each time step can be input into the model for training.

[0041] This embodiment does not limit the specific type of prediction model, but can generally be a bidirectional long short-term memory (BiLSTM) network.

[0042] Long Short-Term Memory (LSTM) networks, by introducing specific gating mechanisms and cell states, can effectively learn and retain long-range dependencies, controlling the flow and renewal of capillary pressure. However, unidirectional LTM networks can only predict along the increasing direction of the mercury intrusion saturation sequence. In actual mercury intrusion processes, the curve shape of the high-pressure section contains physical information about the densest part of the rock skeleton. Therefore, this model introduces a bidirectional mechanism, consisting of two parallel mercury intrusion saturation sequences, one forward and one backward.

[0043] In the evolution of mercury saturation sequence, the forward network calculates the hidden state in the direction from 0 to 100% of the mercury saturation sequence, which can simulate the displacement process of non-wetting phase fluid from large pore throat to small pore throat; the backward network calculates the hidden state in the direction from 100% to 0 of the mercury saturation sequence, and uses the high-pressure micropore characteristics to perform reverse morphological constraints on the medium and low pressure section at the front end of the curve; the hidden states in the two directions are spliced ​​and merged at each node.

[0044] In one example, the structure of the prediction model can be as follows: Figure 3As shown, the model employs a dual-channel strategy to process 8-dimensional initial conventional logging data in its input layer. The first channel utilizes Principal Component Analysis (PCA) to extract 4-dimensional core principal components, eliminating multicollinearity and environmental noise among conventional logging data. The second channel introduces the reservoir quality factor (RQI) calculated based on physical properties, a 3D fuzzy probability label constructed based on piecewise linear interpolation, and a dynamic mercury saturation sequence as prior constraints in rock physics. The features from the two channels are concatenated into a 9-dimensional input feature vector in the input layer, which then enters a hidden layer consisting of a 64-unit and a 32-unit bidirectional long short-term memory network (BiLSTM). This hidden layer, through forward and backward propagation mechanisms and Dropout regularization (to suppress overfitting), enables the model to fully capture the long-range contextual dependence and temporal characteristics of capillary pressure evolution with the mercury saturation sequence. Finally, the hidden layer state is mapped via a fully connected layer (FC) and processed by inverse normalization to achieve high-precision prediction of the complete capillary pressure curve at a single depth point.

[0045] S104: Determine the model loss value based on the capillary pressure curves and the prediction results of the capillary pressure curves at each depth point, update the model parameters based on the model loss value, until the trained prediction model is obtained.

[0046] S105: After processing the conventional logging data and physical property parameters of the depth point to be measured, input them into the trained prediction model to obtain the prediction result of the target capillary pressure curve.

[0047] This embodiment calculates the model loss value based on the predicted capillary pressure curve output by the prediction model and the capillary pressure curve used as the training label. Furthermore, the model parameters can be updated based on the model loss value, and iterative training can be performed on the training dataset until convergence or a preset number of iterations is reached.

[0048] After the model training is completed, this embodiment can obtain the conventional logging data and physical property parameters of the depth point to be measured. According to the feature processing method in the model training process, the conventional logging data and physical property parameters to be measured are used to extract features. The extracted features are then input into the trained prediction model to obtain the prediction result of the target capillary pressure curve.

[0049] Furthermore, to overcome the impact of small sample size and random fluctuations of a single neural network on the accuracy of prediction results, this embodiment can set up multiple sub-models for integrated training. For example, five sub-models with different random seeds can be trained simultaneously. After training, the average value of the sub-model parameters is taken as the model parameter value of the final prediction model to improve the accuracy of the model prediction results.

[0050] Based on the above embodiments, the present invention applies rock physics prior constraints based on reservoir quality factors to the prediction model, thereby limiting invalid searches in the parameter space and improving the generalization ability and prediction accuracy of the prediction model in tight heterogeneous strata.

[0051] The following combination Figure 4 , Figure 4 This is a structural block diagram of a reservoir capillary pressure curve prediction device provided in an embodiment of the present invention. The device may include: The first module 100 is used to acquire conventional logging data and physical property parameters at various depths of the target reservoir as training samples, and to acquire capillary pressure curves at the corresponding depths as training labels. The second module 200 is used to perform feature dimensionality reduction processing on conventional logging data to extract core features, determine reservoir quality factors based on physical property parameters, and determine the fuzzy probability of reservoir category at each depth point based on reservoir quality factors and fuzzy membership functions. The third module 300 is used to concatenate the core features and fuzzy probabilities to generate an input feature vector, and input the input feature vector into the prediction model to obtain the capillary pressure curve prediction result. The fourth module 400 is used to determine the model loss value based on the capillary pressure curve and the prediction result of the capillary pressure curve at each depth point, and update the model parameters based on the model loss value until the training completed prediction model is obtained. The fifth module, 500, is used to process the conventional logging data and physical property parameters of the depth point to be measured and input them into the trained prediction model to obtain the prediction result of the target capillary pressure curve.

[0052] Based on the above embodiments, the present invention applies rock physics prior constraints based on reservoir quality factors to the prediction model, thereby limiting invalid searches in the parameter space and improving the generalization ability and prediction accuracy of the prediction model in tight heterogeneous strata.

[0053] Based on the above embodiments, the second module 200 may include: The first unit is used to construct fuzzy membership functions for each reservoir category based on piecewise linear interpolation; The second unit is used to determine the fuzzy probability of reservoir category by inputting the reservoir quality factor of the depth point into the fuzzy membership function based on the reservoir quality factor classification standard and the transition bandwidth.

[0054] Based on the above embodiments, the third module 300 may include: The third unit is used to concatenate the core features, fuzzy probabilities, reservoir quality factors, and mercury saturation sequences to generate an input feature vector.

[0055] Based on the above embodiments, the second module 200 may include: The fourth unit is used to perform feature dimensionality reduction processing on conventional well logging data through principal component analysis, and to extract core features from the conventional well logging data.

[0056] Based on the above embodiments, conventional logging data includes natural gamma, sonic transit time, compensated neutrons, density, deep lateral resistivity, and shallow lateral resistivity.

[0057] Based on the above embodiments, the physical properties include porosity and permeability.

[0058] Based on the above embodiments, the prediction model is a bidirectional long short-term memory network.

[0059] Based on the above embodiments, the present invention also provides an electronic device, which may include a memory and a processor. The memory stores a computer program, and when the processor calls the computer program in the memory, it can implement the steps provided in the above embodiments. Of course, the device may also include various necessary network interfaces, a power supply, and other components.

[0060] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by an execution terminal or processor, can implement the method provided in the embodiments of the present invention; the storage medium may include various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0061] In this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, without necessarily requiring or implying any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

Claims

1. A method for predicting reservoir capillary pressure curves, characterized in that, include: Conventional logging data and physical property parameters at various depths of the target reservoir are obtained as training samples, and capillary pressure curves at the corresponding depths are obtained as training labels. The conventional logging data is subjected to feature dimensionality reduction processing to extract core features. Based on the physical property parameters, the reservoir quality factor is determined. Based on the reservoir quality factor and the fuzzy membership function, the fuzzy probability of the reservoir category at each depth point is determined. The core features and the fuzzy probabilities are concatenated to generate an input feature vector, and the input feature vector is input into the prediction model to obtain the capillary pressure curve prediction result. The model loss value is determined based on the capillary pressure curve at each depth point and the prediction result of the capillary pressure curve. The model parameters are updated based on the model loss value until the training of the prediction model is obtained. After processing the conventional logging data and physical property parameters of the depth point to be measured, the data is input into the trained prediction model to obtain the prediction result of the target capillary pressure curve.

2. The reservoir capillary pressure curve prediction method according to claim 1, characterized in that, The fuzzy probability of reservoir category at each depth point is determined based on the reservoir quality factor and fuzzy membership function, including: A fuzzy membership function for each reservoir category is constructed based on piecewise linear interpolation; Based on the reservoir quality factor classification standard and transition bandwidth, the reservoir quality factor at the depth point is input into the fuzzy membership function to determine the fuzzy probability of the reservoir category.

3. The method for predicting reservoir capillary pressure curves according to claim 1, characterized in that, The core features and the fuzzy probabilities are concatenated to generate an input feature vector, including: The core features, reservoir quality factor, fuzzy probability, and mercury saturation sequence are concatenated to generate the input feature vector.

4. The reservoir capillary pressure curve prediction method according to claim 1, characterized in that, The core features are extracted by performing feature dimensionality reduction processing on the conventional well logging data, including: Principal component analysis is used to perform feature dimensionality reduction on the conventional well logging data, and the core features are extracted from the conventional well logging data.

5. The reservoir capillary pressure curve prediction method according to claim 1, characterized in that, Conventional logging data includes natural gamma, sonic transit time, compensated neutrons, density, deep lateral resistivity, and shallow lateral resistivity.

6. The method for predicting reservoir capillary pressure curves according to claim 1, characterized in that, The physical properties include porosity and permeability.

7. The method for predicting reservoir capillary pressure curves according to claim 1, characterized in that, The prediction model is a bidirectional long short-term memory network.

8. A reservoir capillary pressure curve prediction device, characterized in that, include: The first module is used to acquire conventional logging data and physical property parameters at various depths of the target reservoir as training samples, and to acquire capillary pressure curves at the corresponding depths as training labels. The second module is used to perform feature dimensionality reduction processing on the conventional logging data to extract core features, determine the reservoir quality factor based on the physical property parameters, and determine the fuzzy probability of the reservoir category at each depth point based on the reservoir quality factor and the fuzzy membership function. The third module is used to concatenate the core features and the fuzzy probabilities to generate an input feature vector, and input the input feature vector into the prediction model to obtain the capillary pressure curve prediction result. The fourth module is used to determine the model loss value based on the capillary pressure curve at each depth point and the prediction result of the capillary pressure curve, and update the model parameters based on the model loss value until the training completed prediction model is obtained. The fifth module is used to process the conventional logging data and physical property parameters of the depth point to be measured and input them into the trained prediction model to obtain the prediction result of the target capillary pressure curve.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to implement the reservoir capillary pressure curve prediction method as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, implement the reservoir capillary pressure curve prediction method as described in any one of claims 1 to 7.

Images