A reservoir pore structure evaluation method, device, equipment and storage medium

Abstract

The application discloses a reservoir pore structure evaluation method, device and equipment and a storage medium, and is applied to the field of geological exploration. The method trains a model by using a capillary pressure curve measured by a sample reservoir high-pressure mercury injection experiment and logging data corresponding to depth points, classifies the reservoir based on a reservoir classification index, splices a classification probability vector and conventional logging data features, designs a physical position coding based on logarithmic transformation, explicitly embeds the inherent order of capillary pressure nodes into the spliced features, and couples a physical constraint term in a loss function of the model, so that the prediction result meets the rock physical law. The application learns the complex nonlinear mapping relationship between the conventional logging data and the capillary pressure curve, improves the prediction accuracy by combining the position coding and the physical constraint term in the loss function, and thus predicts the capillary pressure curve of the whole well section at low cost, quickly and accurately, and further realizes accurate reservoir pore structure evaluation.

Description

Technical Field

[0001] This invention relates to the field of geological exploration, and in particular to a method for evaluating reservoir pore structure, a device for evaluating reservoir pore structure, electronic equipment, and a computer-readable storage medium. Background Technology

[0002] Low-permeability sandstone reservoirs exhibit poor physical properties, generally characterized by low porosity and extremely low permeability, resulting in low mobile fluid saturation. Their microscopic pore structure is complex, dominated by nano- to micrometer-scale micropores and fine throats, with diverse pore types and complex pore-throat configurations, leading to poor connectivity. The quality of low-permeability sandstone reservoirs is closely related to their pore structure characteristics. Capillary pressure curves are crucial data for characterizing rock pore structure and evaluating permeability. While typically obtained through experiments such as high-pressure mercury intrusion porosimetry (HPMI), these conventional methods are difficult to cor, costly, and discontinuous. Nuclear magnetic resonance (NMR) logging is also an effective method for evaluating reservoir pore structure, but its high cost and limited coverage of the entire study area make it difficult to perform well logging. Summary of the Invention

[0003] The purpose of this invention is to provide a method, apparatus, equipment, and storage medium for evaluating reservoir pore structure, which can be applied in the field of geological exploration. This method improves prediction accuracy by learning the complex nonlinear mapping relationship between conventional logging data and capillary pressure curves, and by combining location coding and physical constraint terms in the loss function. Thus, it can predict the capillary pressure curve of the entire well section at low cost, quickly and accurately using conventional logging data, thereby achieving accurate evaluation of reservoir pore structure.

[0004] To address the aforementioned technical problems, this invention provides a method for evaluating reservoir pore structure, comprising: Obtain conventional logging data and capillary pressure curves of the sample reservoir, determine the reservoir classification index based on the conventional logging data, and perform fuzzy classification on the sample reservoir based on the reservoir classification index to obtain the classification probability vector. Extract the logging feature vector from the conventional logging data, concatenate the logging feature vector, the reservoir classification index, and the classification probability vector to obtain a concatenated feature vector, construct a physical location code based on the pressure value of the capillary pressure node in the capillary pressure curve, and combine the physical location code with the concatenated feature vector to form an input sequence; The input sequence is input into the prediction model to obtain the predicted mercury saturation of each capillary pressure node. The model loss value is determined based on the predicted mercury saturation and the loss function containing physical constraint terms. The model parameters of the prediction model are updated based on the model loss value until the prediction model is trained. The processed conventional logging data of the reservoir to be tested is input into the trained prediction model to obtain the target predicted mercury saturation of the capillary pressure node. The target capillary pressure curve is determined based on the target predicted mercury saturation of each capillary pressure node. The pore structure of the reservoir to be tested is evaluated based on the target capillary pressure curve.

[0005] Optionally, a physical location code is constructed based on the pressure values ​​of the capillary pressure nodes in the capillary pressure curve. The physical location code is then combined with the concatenated feature vector to form an input sequence, including: The maximum and minimum pressure values ​​are determined from the pressure values ​​at each capillary pressure node in the capillary pressure curve. The physical location encoding scalar of each capillary pressure node is determined based on the pressure value after logarithmic transformation, the maximum pressure value, and the minimum pressure value. The physical location encoding scalar is expanded into a physical location encoding vector with the same dimension as the concatenated feature vector; A physical position encoding matrix is ​​constructed based on the physical position encoding vector of each capillary pressure node; The physical location code is combined with the concatenated feature vector to form the input sequence.

[0006] Optionally, combining the physical location code with the concatenated feature vector to form an input sequence includes: The spliced ​​feature vector is obtained by performing feature dimensionality reduction on the spliced ​​feature vector based on principal component analysis. The input sequence is formed by combining the dimensionality-reduced spliced ​​feature vector with the physical location code.

[0007] Optionally, the loss function includes a numerical fitting error term and a physical constraint term; the numerical fitting error term includes a mean squared error term.

[0008] Optionally, the physical constraint terms include monotonic physical constraint terms and rank-form constraint terms.

[0009] Optionally, the prediction model is a Transformer regression network model.

[0010] Optionally, the Transformer regression network model includes: two Transformer encoders and a fully connected regression layer connected in sequence; the Transformer encoder includes a multi-head self-attention mechanism, residual connection, layer normalization, feedforward neural network, residual connection and layer normalization connected in sequence.

[0011] To solve the above-mentioned technical problems, the present invention provides a reservoir pore structure evaluation device, comprising: The first module is used to acquire conventional logging data and capillary pressure curves of the sample reservoir, determine the reservoir classification index based on the conventional logging data, and perform fuzzy classification on the sample reservoir based on the reservoir classification index to obtain a classification probability vector. The second module is used to extract the logging feature vector of the conventional logging data, concatenate the logging feature vector, the reservoir classification index and the classification probability vector to obtain the concatenated feature vector, construct the physical location code based on the pressure value of the capillary pressure node in the capillary pressure curve, and combine the physical location code with the concatenated feature vector to form an input sequence. The third module is used to input the input sequence into the prediction model to obtain the predicted mercury saturation of each capillary pressure node, determine the model loss value based on the predicted mercury saturation and the loss function containing physical constraint terms, and update the model parameters of the prediction model based on the model loss value until the trained prediction model is obtained. The fourth module is used to input the processed conventional logging data of the reservoir to be tested into the trained prediction model to obtain the target predicted mercury saturation of the capillary pressure node, determine the target capillary pressure curve based on the target predicted mercury saturation of each capillary pressure node, and evaluate the pore structure of the reservoir to be tested based on the target capillary pressure curve.

[0012] 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 porosity structure evaluation method when executing the computer program.

[0013] 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 porosity structure evaluation method.

[0014] As can be seen, this invention acquires conventional logging data and capillary pressure curves of sample reservoirs, determines reservoir classification indices based on conventional logging data, performs fuzzy classification of sample reservoirs based on reservoir classification indices to obtain classification probability vectors, extracts logging feature vectors from conventional logging data, concatenates logging feature vectors, reservoir classification indices, and classification probability vectors to obtain concatenated feature vectors, constructs physical location codes based on pressure values ​​of capillary pressure nodes in capillary pressure curves, combines physical location codes with concatenated feature vectors to form input sequences, inputs the input sequence into a prediction model to obtain predicted mercury saturation for each capillary pressure node, determines model loss values ​​based on predicted mercury saturation and a loss function including physical constraint terms, updates model parameters based on model loss values ​​until a trained prediction model is obtained, inputs the conventional logging data of the reservoir to be tested into the trained prediction model to obtain target predicted mercury saturation for capillary pressure nodes, determines target capillary pressure curves based on target predicted mercury saturation for each capillary pressure node, and evaluates the pore structure of the reservoir to be tested based on the target capillary pressure curves.

[0015] This invention improves prediction accuracy by learning the complex nonlinear mapping relationship between conventional logging data and capillary pressure curves, and by combining location encoding and physical constraint terms in the loss function. Thus, it can predict the capillary pressure curve of the entire well section at low cost, quickly and accurately using conventional logging data, thereby achieving accurate evaluation of reservoir pore structure. Attached Figure Description

[0016] 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.

[0017] Figure 1 A flowchart of a reservoir pore structure evaluation method provided in an embodiment of the present invention; Figure 2 This is a structural block diagram of a reservoir pore structure evaluation device provided in an embodiment of the present invention. Detailed Implementation

[0018] 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.

[0019] Low-permeability sandstone reservoirs exhibit poor physical properties, generally characterized by low porosity and extremely low permeability, resulting in low mobile fluid saturation. Their microscopic pore structure is complex, dominated by nano- to micrometer-scale micropores and fine throats, with diverse pore types and complex pore-throat configurations, leading to poor connectivity. The quality of low-permeability sandstone reservoirs is closely related to their pore structure characteristics. Capillary pressure curves are obtained through mercury intrusion porosimetry (MIP) or centrifugation experiments, and pore structure parameters can be extracted from these curves. MIP involves measuring the mercury saturation in the core sample under different pressures and creating a capillary pressure curve, from which characteristic pore structure parameters can be extracted. Pore structure evaluation methods based on nuclear magnetic resonance (NMR) logging can be broadly categorized into two types. The first type directly extracts various characteristic parameters from the NMR relaxation time spectrum to evaluate pore structure. The second type primarily converts the T2 spectrum obtained from NMR into a pseudo-capillary pressure curve for analysis; the T2 spectrum is a parameter in NMR technology characterizing the lateral relaxation time distribution. The basic idea is to establish the correspondence between relaxation time and pore throat distribution through experimental calibration, and to introduce a conversion coefficient or conversion factor to transform the T2 spectrum into a capillary pressure curve that can reflect the pore throat distribution.

[0020] Conventional logging methods such as high-pressure mercury intrusion porosimetry (HSI) are difficult to cor, costly, and discontinuous. Nuclear magnetic resonance (NMR) logging is an effective logging method for evaluating reservoir porosity, but it is expensive and difficult to cover the entire study area. Conventional logging methods such as natural gamma ray, three-porosity, and resistivity logging are widely used and have lower costs. Therefore, developing methods to predict reservoir capillary pressure curves using conventional logging data and then continuously evaluate reservoir porosity is of great significance.

[0021] The purpose of this invention is to utilize the Transformer regression network model to learn the complex nonlinear mapping relationship between conventional logging data and capillary pressure curves. By combining location encoding, coupling mean square error in the loss function, monotonicity constraints, and Spearman rank correlation coefficient constraints, the prediction accuracy can be improved. Thus, the capillary pressure curve of the entire well section can be predicted in a low-cost, fast, and accurate manner using conventional logging data.

[0022] The following combination Figure 1 , Figure 1 A flowchart of a reservoir pore structure evaluation method provided in an embodiment of the present invention, the method may include: S101: Obtain conventional logging data and capillary pressure curves of the sample reservoir, determine the reservoir classification index based on the conventional logging data, and perform fuzzy classification of the sample reservoir based on the reservoir classification index to obtain the classification probability vector.

[0023] This embodiment can acquire conventional logging data and capillary pressure curves of the sample reservoir for training the prediction model, so that the prediction model can learn the mapping relationship between conventional logging data and capillary pressure curves.

[0024] This embodiment first normalizes conventional logging data to eliminate dimensional differences.

[0025] Furthermore, based on the normalized conventional logging data, the reservoir classification index (RCI) is determined. Then, the reservoir classification index is used to perform fuzzy classification on the sample reservoirs to obtain the classification probability vector, as shown in the following formula: ; In the formula, RCI represents a function for calculating the reservoir classification index. For depth The reservoir classification index at the location. The classification function represents the classification function, which will Transform into a probability distribution express Mapped onto a 3-dimensional probability simplex. The final classification output for the i-th sample (i.e., the i-th depth point) is a vector of length 3, and the sum of its three elements is 1.

[0026] S102: Extract the logging feature vector from conventional logging data, concatenate the logging feature vector, reservoir classification index, and classification probability vector to obtain the concatenated feature vector, construct the physical location code based on the pressure value of the capillary pressure node in the capillary pressure curve, and combine the physical location code with the concatenated feature vector to form the input sequence.

[0027] This embodiment can extract the logging feature vector of the conventional logging data of the i-th sample. As one of the inputs to the prediction model, the well logging feature vector, reservoir classification index, and classification probability vector are concatenated to obtain the concatenated feature vector, as shown in the following formula: ; In the formula, Let be the concatenated feature vector of the i-th sample. Let be the normalized feature vector of the i-th sample. Represents the normalized , Let R represent the classification probability vector for the i-th sample, where R is the set of real numbers. This represents the dimension of the concatenated feature vectors.

[0028] Furthermore, to avoid collinearity of input features affecting the model and reduce feature redundancy, this embodiment can use principal component analysis (PCA) for feature dimensionality reduction. Specifically, PCA is used to perform feature dimensionality reduction on the concatenated feature vector to obtain the dimensionality-reduced concatenated feature vector; the dimensionality-reduced concatenated feature vector is then combined with the physical location code to form the input sequence.

[0029] The concatenated feature vector after dimensionality reduction can be: ; It is the concatenated feature vector after dimensionality reduction of the i-th sample, where W is the weight matrix and T is the matrix transpose. This represents the dimension of the concatenated feature vectors after dimensionality reduction.

[0030] The order of capillary pressure points has a clear physical meaning; therefore, a physical position encoding based on logarithmic transformation is constructed to enable the model to clearly identify which capillary pressure node (P) it is currently predicting. c The mercury saturation corresponding to the pressure value of the capillary pressure node in the capillary pressure curve is used to construct a physical location code. The physical location code is then combined with the concatenated feature vector to form the input sequence.

[0031] Specifically, the maximum and minimum pressure values ​​are determined from the pressure values ​​of each capillary pressure node in the capillary pressure curve; based on the logarithmically transformed pressure values, the maximum pressure value, and the minimum pressure value, the physical location encoding scalar of each capillary pressure node is determined, as shown in the following formula: ; In the formula, Let J be the pressure value at the j-th capillary pressure node, in MPa. The minimum pressure value among all k capillary pressure nodes; The maximum pressure value among all k capillary pressure nodes; It is a normalized physical location encoding scalar with a value range between [0,1], representing the position of the j-th capillary pressure node in the entire pressure range.

[0032] Furthermore, in this embodiment, the physical position encoding scalar can be expanded into a physical position encoding vector with the same dimension as the concatenated feature vector. A physical position encoding matrix is ​​constructed based on the physical position encoding vectors of each capillary pressure node, so that when the model processes each capillary pressure node, it can obtain its relative position in the entire capillary pressure sequence, as shown in the following formula: ; In the formula, E is the physical location encoding matrix with dimension 1. M is the length of the input sequence. Input dimensions to the model, This is the physical location encoding vector corresponding to the j-th capillary pressure node. For a length of A column vector of all 1s.

[0033] This embodiment combines physical location encoding with concatenated feature vectors to form the input sequence of the prediction model, as shown in the following formula: ; In the formula Let be the input sequence for the i-th sample, with dimension . M is the length of the input sequence. For feature dimensions; for The j-th element is composed of two parts: The concatenated feature vector after dimensionality reduction of the i-th sample; Encode the physical location scalar for the j-th position.

[0034] S103: Input the input sequence into the prediction model to obtain the predicted mercury saturation of each capillary pressure node. Determine the model loss value based on the predicted mercury saturation and the loss function containing physical constraint terms. Update the model parameters of the prediction model based on the model loss value until the trained prediction model is obtained.

[0035] This embodiment inputs the input sequence into a prediction model to obtain the predicted mercury saturation at each capillary pressure node. This embodiment does not limit the specific type of prediction model; it can be set based on the actual application, and is generally a Transformer regression network model. This embodiment does not limit the specific architecture of the Transformer regression network model; it generally includes two Transformer encoders connected in sequence and a fully connected regression layer. The Transformer encoder includes a multi-head self-attention mechanism, residual connections, layer normalization, a feedforward neural network, residual connections, and layer normalization, connected in sequence. Dropout layers are interspersed to suppress overfitting, and the ReLU (Rectified Linear Unit) activation function is used to enhance the model's nonlinear expressive power. After global dependency modeling and feature transformation by the two encoder layers, the prediction result is finally output through the fully connected regression layer, thereby achieving a nonlinear mapping from input variables to target values.

[0036] This embodiment can determine the model loss value based on the predicted mercury saturation and the loss function including physical constraints, and update the model parameters of the prediction model based on the model loss value until the trained prediction model is obtained.

[0037] This embodiment does not limit the specific type of loss function, which can generally include a numerical fitting error term and a physical constraint term. To ensure that the prediction results conform to the basic laws of rock physics, the loss function of this model can be composed of both numerical fitting error and physical constraints.

[0038] The numerical fitting error term in this embodiment may include a mean squared error term. The physical constraint term may include a monotonic physical constraint term and a rank-type constraint term.

[0039] During mercury impregnation, as capillary pressure increases, mercury gradually enters a smaller pore throat space, macroscopically manifested as a continuous increase in mercury saturation. Therefore, the predicted capillary pressure curve should satisfy a monotonically increasing constraint, i.e., a monotonic physical constraint term, as shown in the following equation: ; In the formula, For monotonic physical constraint terms, To calculate the average expected value of the loss for the i-th sample, M is the sequence length of the predicted capillary pressure curve, which is also the total number of capillary pressure nodes. Let i be the predicted mercury saturation of the i-th sample at the j-th capillary pressure node. Let ReLU be the predicted value of the i-th sample at position j+1, and let ReLU be the activation function. When the value is greater than 0, the difference is output as a penalty for non-physical increments; otherwise, 0 is output, meaning no penalty is applied when the prediction result is monotonic.

[0040] In addition to monotonicity, the model also needs to maintain the consistency of the rank structure of the curve shape. In this embodiment, Spearman's rank correlation coefficient can be used to measure the morphological similarity between the predicted curve and the measured curve. The calculation process is as follows: ; In the formula, Let be the Spearman rank correlation coefficient of the i-th sample, which measures the morphological similarity between the measured curve and the predicted curve. The closer the correlation coefficient is to 1, the smaller the loss; the greater the morphological difference, the greater the loss. Let represent the measured mercury saturation of the i-th sample at the j-th capillary pressure node. Let represent the predicted mercury saturation of the i-th sample at the j-th capillary pressure node. The rank is the rank of the elements in the sequence after sorting them by size. The rank-form constraint term penalizes the difference in shape between the predicted curve and the measured curve. The average expected value of the loss for the i-th sample is taken.

[0041] In this embodiment, the mean square error term is shown in the following formula: ; In the formula, This is the mean square error loss term. Let i be the measured mercury saturation sequence of the i-th sample. Let i be the predicted mercury saturation sequence for the i-th sample. It is the square of the L2 norm.

[0042] In this embodiment, the loss function can be obtained by weighting the mean squared error term, the monotonic physical constraint term, and the rank-shaped constraint term, as shown in the following formula: ; In the formula, Let α be the loss function, and β and γ be the weighting coefficients. In this embodiment, α = 0.2, β = 0.2, and γ = 0.6 can be set to balance the contributions of the three losses.

[0043] In this embodiment, the prediction model can be iteratively trained using data obtained from sampled reservoirs until the model parameters converge or a preset number of iterations are reached, thus obtaining a trained prediction model.

[0044] S104: Input the processed conventional logging data of the reservoir to be tested into the trained prediction model to obtain the target predicted mercury saturation of the capillary pressure node. Based on the target predicted mercury saturation of each capillary pressure node, determine the target capillary pressure curve. Based on the target capillary pressure curve, evaluate the pore structure of the reservoir to be tested.

[0045] This embodiment can predict pressure values ​​based on a trained prediction model. The processed conventional logging data of the target depth point of the reservoir to be measured is input into the trained prediction model to obtain the target predicted mercury saturation of the capillary pressure node. The processing flow refers to the construction flow of the input sequence during the model training process.

[0046] Furthermore, by connecting the target predicted mercury saturation corresponding to each capillary pressure node, the target capillary pressure curve of the target depth point of the reservoir to be tested can be obtained. Finally, the pore structure of the reservoir to be tested can be evaluated based on the target capillary pressure curve.

[0047] Based on the above embodiments, this invention improves prediction accuracy by learning the complex nonlinear mapping relationship between conventional logging data and capillary pressure curves, and by combining location encoding and physical constraint terms in the loss function. Thus, it can use conventional logging data to predict the capillary pressure curve of the entire well section at low cost, quickly and accurately, thereby achieving accurate evaluation of reservoir pore structure.

[0048] The following combination Figure 2 , Figure 2 This is a structural block diagram of a reservoir pore structure evaluation 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 capillary pressure curves of the sample reservoir, determine the reservoir classification index based on the conventional logging data, and perform fuzzy classification of the sample reservoir based on the reservoir classification index to obtain the classification probability vector. The second module 200 is used to extract the logging feature vector of conventional logging data. The logging feature vector, reservoir classification index and classification probability vector are concatenated to obtain the concatenated feature vector. The physical location code is constructed based on the pressure value of the capillary pressure node in the capillary pressure curve. The physical location code is combined with the concatenated feature vector to form the input sequence. The third module 300 is used to input the input sequence into the prediction model to obtain the predicted mercury saturation of each capillary pressure node, determine the model loss value based on the predicted mercury saturation and the loss function containing physical constraint terms, update the model parameters of the prediction model based on the model loss value, until the trained prediction model is obtained. The fourth module 400 is used to input the processed conventional logging data of the reservoir to be tested into the trained prediction model to obtain the target predicted mercury saturation of the capillary pressure node, determine the target capillary pressure curve based on the target predicted mercury saturation of each capillary pressure node, and evaluate the pore structure of the reservoir to be tested based on the target capillary pressure curve.

[0049] Based on the above embodiments, this invention improves prediction accuracy by learning the complex nonlinear mapping relationship between conventional logging data and capillary pressure curves, and by combining location encoding and physical constraint terms in the loss function. Thus, it can use conventional logging data to predict the capillary pressure curve of the entire well section at low cost, quickly and accurately, thereby achieving accurate evaluation of reservoir pore structure.

[0050] Based on the above embodiments, the second module 200 may include: The first unit is used to determine the maximum and minimum pressure values ​​from the pressure values ​​of each capillary pressure node in the capillary pressure curve. The second unit is used to determine the physical location encoding scalar of each capillary pressure node based on the pressure value after logarithmic transformation, the maximum pressure value, and the minimum pressure value; The third unit is used to expand the physical location encoding scalar into a physical location encoding vector with the same dimension as the concatenated feature vector; The fourth unit is used to construct a physical position encoding matrix based on the physical position encoding vector of each capillary pressure node; The fifth unit is used to combine the physical location code with the concatenated feature vectors to form the input sequence.

[0051] Based on the above embodiments, the second module 200 may include: Unit 6 is used to perform feature dimensionality reduction on the spliced ​​feature vectors based on principal component analysis to obtain the dimensionality-reduced spliced ​​feature vectors. The seventh unit is used to form an input sequence based on the dimensionality-reduced concatenated feature vector and the physical location code.

[0052] Based on the above embodiments, the loss function includes a numerical fitting error term and a physical constraint term; the numerical fitting error term includes a mean squared error term.

[0053] Based on the above embodiments, the physical constraint terms include monotonic physical constraint terms and rank-form constraint terms.

[0054] Based on the above embodiments, the prediction model is a Transformer regression network model.

[0055] Based on the above embodiments, the Transformer regression network model includes: two Transformer encoders and a fully connected regression layer connected in sequence; the Transformer encoder includes a multi-head self-attention mechanism, residual connection, layer normalization, feedforward neural network, residual connection and layer normalization connected in sequence.

[0056] 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.

[0057] 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.

[0058] 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 evaluating reservoir pore structure, characterized in that, include: Obtain conventional logging data and capillary pressure curves of the sample reservoir, determine the reservoir classification index based on the conventional logging data, and perform fuzzy classification on the sample reservoir based on the reservoir classification index to obtain the classification probability vector. Extract the logging feature vector from the conventional logging data, concatenate the logging feature vector, the reservoir classification index, and the classification probability vector to obtain a concatenated feature vector, construct a physical location code based on the pressure value of the capillary pressure node in the capillary pressure curve, and combine the physical location code with the concatenated feature vector to form an input sequence; The input sequence is input into the prediction model to obtain the predicted mercury saturation of each capillary pressure node. The model loss value is determined based on the predicted mercury saturation and the loss function containing physical constraint terms. The model parameters of the prediction model are updated based on the model loss value until the prediction model is trained. The processed conventional logging data of the reservoir to be tested is input into the trained prediction model to obtain the target predicted mercury saturation of the capillary pressure node. The target capillary pressure curve is determined based on the target predicted mercury saturation of each capillary pressure node. The pore structure of the reservoir to be tested is evaluated based on the target capillary pressure curve.

2. The reservoir pore structure evaluation method according to claim 1, characterized in that, A physical location code is constructed based on the pressure values ​​of the capillary pressure nodes in the capillary pressure curve. The physical location code is then combined with the concatenated feature vector to form an input sequence, including: The maximum and minimum pressure values ​​are determined from the pressure values ​​at each capillary pressure node in the capillary pressure curve. The physical location encoding scalar of each capillary pressure node is determined based on the pressure value after logarithmic transformation, the maximum pressure value, and the minimum pressure value. The physical location encoding scalar is expanded into a physical location encoding vector with the same dimension as the concatenated feature vector; A physical position encoding matrix is ​​constructed based on the physical position encoding vector of each capillary pressure node; The physical location code is combined with the concatenated feature vector to form the input sequence.

3. The reservoir pore structure evaluation method according to claim 1, characterized in that, The physical location code is combined with the concatenated feature vector to form an input sequence, including: The spliced ​​feature vector is obtained by performing feature dimensionality reduction on the spliced ​​feature vector based on principal component analysis. The input sequence is formed by combining the dimensionality-reduced spliced ​​feature vector with the physical location code.

4. The reservoir pore structure evaluation method according to claim 1, characterized in that, The loss function includes a numerical fitting error term and a physical constraint term; the numerical fitting error term includes a mean squared error term.

5. The reservoir pore structure evaluation method according to claim 1, characterized in that, The physical constraints include monotonic physical constraints and rank-form constraints.

6. The reservoir pore structure evaluation method according to claim 1, characterized in that, The prediction model is a Transformer regression network model.

7. The reservoir pore structure evaluation method according to claim 6, characterized in that, The Transformer regression network model includes: two Transformer encoders and a fully connected regression layer connected in sequence; the Transformer encoder includes a multi-head self-attention mechanism, residual connection, layer normalization, feedforward neural network, residual connection and layer normalization connected in sequence.

8. A reservoir pore structure evaluation device, characterized in that, include: The first module is used to acquire conventional logging data and capillary pressure curves of the sample reservoir, determine the reservoir classification index based on the conventional logging data, and perform fuzzy classification on the sample reservoir based on the reservoir classification index to obtain a classification probability vector. The second module is used to extract the logging feature vector of the conventional logging data, concatenate the logging feature vector, the reservoir classification index and the classification probability vector to obtain the concatenated feature vector, construct the physical location code based on the pressure value of the capillary pressure node in the capillary pressure curve, and combine the physical location code with the concatenated feature vector to form an input sequence. The third module is used to input the input sequence into the prediction model to obtain the predicted mercury saturation of each capillary pressure node, determine the model loss value based on the predicted mercury saturation and the loss function containing physical constraint terms, and update the model parameters of the prediction model based on the model loss value until the trained prediction model is obtained. The fourth module is used to input the processed conventional logging data of the reservoir to be tested into the trained prediction model to obtain the target predicted mercury saturation of the capillary pressure node, determine the target capillary pressure curve based on the target predicted mercury saturation of each capillary pressure node, and evaluate the pore structure of the reservoir to be tested based on the target capillary pressure curve.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the reservoir porosity structure evaluation method as described in any one of claims 1 to 7.

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 porosity structure evaluation method as described in any one of claims 1 to 7.

Images