A method for correcting sample porosity under high pressure conditions

Abstract

The application discloses a sample porosity correction method under high pressure conditions and relates to the technical field of rock physics. The method comprises the following steps: collecting the confining pressure value, strain signal and ultrasonic waveform signal of a sample to form a basic data set; superimposing a preset multi-frequency dynamic pressure disturbance signal on the static water confining pressure background of a pressure step and exciting and collecting acoustic signals; processing the dynamic pressure disturbance signal and the acoustic signal to extract a complex frequency response function; constructing a characteristic vector of the high pressure response of the sample based on the complex frequency response function, the strain signal and the confining pressure value; and inputting the characteristic vector of the high pressure response of the sample into a graph neural network constructed based on a pore network abstraction and a message passing mechanism to simulate the propagation of wave motion in the abstract pore network. The application realizes effective separation of pressure effects and structural effects in internal representation, and the network outputs intrinsic porosity estimation values and generates intrinsic porosity-confining pressure relationship curves through a decoder based on the separated structural information.

Description

Technical Field

[0001] This invention relates to the field of rock physics, and in particular to a method for correcting sample porosity under high pressure conditions. Background Technology

[0002] In the field of oil and gas exploration, accurately obtaining the true pore structure of porous materials such as rocks under high pressure is crucial. Conventional porosity measurements are mostly conducted under normal pressure, which cannot reflect high-pressure conditions. Therefore, in-situ high-pressure porosity assessment technology based on ultrasonic measurement has been developed. By measuring the sound wave velocity under high pressure and calculating porosity using a rock physics model, in-situ, non-destructive measurement is achieved. However, this relies on the accuracy of the model, which is usually based on idealized assumptions and cannot accurately describe the real behavior of complex porous media under high pressure, thus having inherent limitations.

[0003] The fundamental challenge of existing technologies lies in the difficulty of distinguishing between the two different mechanisms by which high pressure affects porosity: one is the reversible pressure effect caused by elastic compression of the material; the other is the structural effect caused by irreversible changes such as plastic deformation of the material's own structure. These two effects are nonlinearly coupled together in the ultrasonic wave propagation signal. Existing model-based inversion methods cannot effectively separate these two effects. The inversion results are essentially the apparent porosity under the combined effect of the two, rather than the intrinsic porosity reflecting the inherent properties of the material. How to effectively decouple the pressure effect and the structural effect from the high-pressure composite signal is a prominent problem in the current technical field. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a method for correcting sample porosity under high pressure conditions. Existing technologies cannot effectively decouple pressure effects and structural effects from high-pressure composite signals to obtain the intrinsic porosity of the sample.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a method for correcting the porosity of a sample under high pressure conditions, which includes collecting the confining pressure value, strain signal and ultrasonic waveform signal of the sample to form a basic dataset;

[0008] On the background of static water confining pressure of the pressure step, a preset multi-frequency dynamic pressure disturbance signal is superimposed and applied, and acoustic signals are excited and collected. The complex frequency response function is extracted by processing the dynamic pressure disturbance signal and the acoustic signal.

[0009] Based on the complex frequency response function, strain signal and confining pressure value, a feature vector of the high-pressure response of the sample is constructed.

[0010] The feature vector of the high-pressure response of the sample is input into a graph neural network constructed based on pore network abstraction and message passing mechanism. By simulating the propagation of waves in the abstract pore network, the propagation process is modulated using the information in the feature vector of the high-pressure response of the sample, and the pressure effect and structural effect are separated.

[0011] The graph neural network executes a physics-inspired message passing mechanism to obtain an estimate of the intrinsic porosity of the sample. By correlating the estimated intrinsic porosity of the sample under different pressures with the corresponding confining pressure, a curve describing the relationship between intrinsic porosity and pressure is generated.

[0012] As a preferred embodiment of the high-pressure sample porosity correction method of the present invention, the formation of the basic dataset includes,

[0013] The confining pressure control unit executes stepped static pressure loading, starting from the initial confining pressure, increasing the pressure of the hydraulic medium in the high-pressure chamber to the preset pressure step for acquisition and recording, the confining pressure value, the strain signal measured by the fiber optic grating sensor, and the ultrasonic wave signal acquired by the broadband ultrasonic transducer.

[0014] The confining pressure values, strain signals, and ultrasonic waveform signals recorded synchronously at the pressure steps during the stepped pressure loading process are collected and structured to form a basic dataset.

[0015] As a preferred embodiment of the sample porosity correction method under high pressure conditions described in this invention, the extraction of the complex frequency response function includes,

[0016] On top of the static confining pressure background established by the pressure step recorded in the basic dataset, a multi-frequency composite dynamic pressure disturbance signal with controlled amplitude is additionally applied by the confining pressure control unit.

[0017] When a multi-frequency composite dynamic pressure disturbance signal is superimposed, the broadband ultrasonic transducer emits a continuous acoustic signal that matches the spectrum of the multi-frequency composite dynamic pressure disturbance signal.

[0018] A broadband ultrasonic receiving transducer synchronously acquires a continuous acoustic signal that matches the spectrum of a multi-frequency composite dynamic pressure disturbance signal after penetrating the sample.

[0019] Frequency domain comparison and relationship analysis were performed on the continuous acoustic signal and the multi-frequency composite dynamic pressure disturbance signal to extract the complex frequency response function characterizing the dynamic mechanical response of the sample under static pressure.

[0020] As a preferred embodiment of the high-pressure sample porosity correction method of the present invention, wherein: the feature vector for constructing the high-pressure response of the sample includes,

[0021] The complex frequency response function is analyzed to extract the amplitude and phase information of the complex frequency response function at the characteristic frequency bands that reflect the main frequency response of the rock skeleton, and to identify the frequency positions and bandwidths of the resonance peaks that characterize structural resonance and the anti-resonance valleys that characterize the pressure densification effect in the complex frequency response function.

[0022] By combining the amplitude information, the phase information, the frequency positions and bandwidths of the resonance peaks and anti-resonance valleys of the complex frequency response function with the strain signal and the confining pressure value, a feature vector of the high-pressure response of the sample is constructed.

[0023] As a preferred embodiment of the high-pressure sample porosity correction method of the present invention, wherein: the simulated wave propagation in the abstract pore network includes,

[0024] By analyzing historical pore structure images of samples, the topological connection relationship between pore bodies and throats is extracted, and a graph neural network based on pore network abstraction and message passing mechanism is constructed.

[0025] In this network, the nodes correspond to the pores, and the edges correspond to the throats.

[0026] In the constructed graph neural network based on the abstraction of porous network and message passing mechanism, a message passing function inspired by the stress wave propagation mechanism is defined. The message passing function specifies the update amount of information transmitted from the source node to the target node through the connecting edge, and is a function of the current state of the source node, the current attribute of the connecting edge, and the current state of the target node.

[0027] From the high-dimensional feature vector describing the comprehensive response of the sample under high pressure, parameters related to the compressibility of the pore body are extracted. These parameters are then assigned to the initial state of the corresponding node in the graph neural network of pore network abstraction and message passing mechanism to form the node attribute feature vector.

[0028] From the high-dimensional feature vector describing the comprehensive response of the sample under a specific high-pressure environment, parameters related to throat conductivity and equivalent size are extracted. These parameters are then assigned to the initial attributes of the corresponding edges in the graph neural network of pore network abstraction and message passing mechanism, forming edge attribute feature vectors.

[0029] As a preferred embodiment of the high-pressure sample porosity correction method of the present invention, wherein: the separation of pressure effect and structural effect includes,

[0030] In the message passing process of graph neural networks with pore network abstraction and message passing mechanism, the confining pressure value is read from the high-dimensional feature vector describing the comprehensive response of the sample under high pressure.

[0031] In the message passing function, the confining pressure value, the throat conductivity and equivalent size parameters encoded by the edge attribute feature vector, and the pore compressibility parameters encoded by the node attribute feature vectors of adjacent nodes are jointly input into the learnable attention mechanism. The pressure sensitivity of the throat is evaluated based on the confining pressure value and the throat conductivity and equivalent size parameters, and the structural stability of the pore is evaluated based on the pore compressibility parameters.

[0032] Based on the throat pressure sensitivity and pore structure stability, a gating signal is generated to control the on / off state and strength of the message transmission path. The message transmission path connected to a high-pressure-sensitive throat and a low-structure-stability pore is suppressed, while the message transmission path connected to a low-pressure-sensitive throat and a high-structure-stability pore is enhanced.

[0033] Through multi-layer iterative message passing, node state update, and edge attribute update of the graph neural network using the porous network abstraction and message passing mechanism, the finally updated node state vector and edge state vector mainly encode information related to structural effects and information related to pressure effects, respectively, thus separating pressure effects and structural effects.

[0034] As a preferred embodiment of the high-pressure sample porosity correction method of the present invention, wherein: the estimated value of the intrinsic porosity of the sample includes,

[0035] In a graph neural network based on the abstraction of pore networks that separates pressure effects and structural effects and the message passing mechanism, the node state vectors that encode information related to structural effects are read.

[0036] Global average pooling is performed on the node state vectors that encode information related to structural effects to generate graph-level feature vectors;

[0037] The graph-level feature vectors are input into the feedforward neural network decoder to obtain an estimate of the intrinsic porosity.

[0038] As a preferred embodiment of the high-pressure sample porosity correction method of the present invention, wherein: the curve describing the relationship between intrinsic porosity and pressure includes,

[0039] The estimated values ​​of intrinsic porosity obtained at all pressure steps during the stepped static loading process are collected to form a sequence of estimated intrinsic porosity values.

[0040] Collect the confining pressure values ​​of all pressure steps recorded during the stepped static pressure loading process to form a confining pressure value sequence corresponding to the estimated value sequence of intrinsic porosity;

[0041] In a two-dimensional coordinate system, the confining pressure value sequence is used as the horizontal axis and the estimated intrinsic porosity value sequence is used as the vertical axis. Data points are plotted and connected to generate a curve describing the relationship between intrinsic porosity and pressure.

[0042] In a second aspect, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the sample porosity correction method under high pressure conditions as described in the first aspect of the present invention.

[0043] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the sample porosity correction method under high pressure conditions as described in the first aspect of the present invention.

[0044] The beneficial effects of this invention are as follows: By using stepped static pressure loading and synchronous sensing to acquire basic datasets, multi-frequency dynamic pressure disturbances are superimposed on the static pressure background and matched acoustic signals are excited. Complex frequency response functions containing rich dynamic information are extracted through frequency domain analysis, and high-dimensional feature vectors are constructed. The feature vectors are input into a graph neural network based on the abstract construction of a pore network. This network simulates the propagation of waves in the pore network through its physically inspired message passing mechanism, and dynamically modulates the propagation path using information such as confining pressure values. Thus, pressure effects and structural effects are effectively separated in the internal representation. Based on the separated structural information, the network outputs an intrinsic porosity estimate through a decoder and generates an intrinsic porosity-confining pressure relationship curve. Attached Figure Description

[0045] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the 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.

[0046] Figure 1 This is a flowchart of a method for correcting sample porosity under high pressure conditions.

[0047] Figure 2 A schematic diagram illustrating the process of constructing the high-pressure response feature vector of a sample. Detailed Implementation

[0048] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0049] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0050] Secondly, the term "one embodiment" or "example" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the invention. The appearance of an embodiment in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that mutually excludes other embodiments.

[0051] Reference Figures 1-2 This is one embodiment of the present invention, which provides a method for correcting sample porosity under high pressure conditions, including the following steps:

[0052] S1. Collect the confining pressure, strain signal and ultrasonic waveform signal of the sample to form a basic dataset.

[0053] S1.1. Stepped static pressure loading is performed through the confining pressure control unit. Starting from the initial confining pressure, the pressure of the hydraulic medium in the high-pressure chamber is increased to the preset pressure step for acquisition and recording. The confining pressure value, the strain signal measured by the fiber optic grating sensor, and the ultrasonic wave signal acquired by the broadband ultrasonic transducer are recorded.

[0054] Furthermore, a stepped static pressure loading process is performed on the hydraulic medium within the high-pressure chamber via a confining pressure control unit. The loading process begins with a pre-set initial confining pressure, and the control unit gradually increases the pressure of the hydraulic medium within the high-pressure chamber to a sequential target pressure step according to a preset step size and rate. After each target pressure step is reached, the pressure of the hydraulic medium within the high-pressure chamber remains stable until a preset mechanical equilibrium criterion is met, such as the strain signal change rate output by the fiber Bragg grating sensor being lower than a threshold. At the stable state of each pressure step, the data acquisition system is synchronously triggered to record the precise confining pressure value at the current moment, the axial and circumferential strain signals measured by the fiber Bragg grating sensors distributed on the sample surface, and the full-waveform digital ultrasonic signal excited by the broadband ultrasonic transmitting transducer, propagated through the sample, and acquired by the broadband ultrasonic receiving transducer.

[0055] S1.2. Collect and structure the confining pressure values, strain signals and ultrasonic waveform signals recorded synchronously under the pressure steps during the stepped pressure loading process to form a basic dataset.

[0056] Furthermore, the confining pressure values, strain signals, and ultrasonic waveform signals recorded synchronously at each pressure step during the stepped pressure loading process are aggregated. The aggregation process aligns the confining pressure values, strain signals, and ultrasonic waveform signals according to the order of the pressure steps, performing time-series alignment and index association. The structured integration operation organizes the aggregated data into a machine-readable and processable format, for example, using the pressure step as the index key, with each key-value pair associated with a data structure containing a specific confining pressure value, a corresponding array of strain signals, and a corresponding array of ultrasonic waveform signals. Through aggregation and structured integration, a basic dataset is formed.

[0057] S2. On the background of static water confining pressure of the pressure step, a preset multi-frequency dynamic pressure disturbance signal is superimposed and applied, and acoustic signals are excited and collected. The complex frequency response function is extracted by processing the dynamic pressure disturbance signal and the acoustic signal.

[0058] S2.1 On the static confining pressure background established by the pressure step recorded in the basic dataset, an amplitude-controlled multi-frequency composite dynamic pressure disturbance signal is additionally applied by the confining pressure control unit.

[0059] Furthermore, by monitoring the sample strain signal change rate through a fiber optic grating sensor to be lower than a preset threshold and maintained for a specified duration, it is confirmed that the hydrostatic confining pressure background has reached a stable state. On the hydrostatic confining pressure background established by each pressure step that has reached a stable state, recorded in the basic dataset, the confining pressure control unit, while maintaining the stability of the hydrostatic confining pressure background, additionally superimposes a preset multi-frequency composite dynamic pressure disturbance signal with a precisely calibrated amplitude to ensure it is within the linear response range in the control command. This composite command is output to drive the pressure servo mechanism to actually generate and apply a multi-frequency composite dynamic pressure disturbance signal with a controlled amplitude superimposed on the hydrostatic confining pressure background in the hydraulic medium in the high-pressure chamber.

[0060] S2.2 When a multi-frequency composite dynamic pressure disturbance signal is superimposed, a broadband ultrasonic transducer emits a continuous acoustic signal that matches the spectrum of the multi-frequency composite dynamic pressure disturbance signal.

[0061] Furthermore, at the same moment that the confining pressure control unit applies the multi-frequency composite dynamic pressure disturbance signal, the broadband ultrasonic transducer is triggered and emits a continuous acoustic signal that matches or covers the main spectral components of the multi-frequency composite dynamic pressure disturbance signal in the frequency domain, according to the preset waveform and parameters. This continuous acoustic signal passes through the sample under high pressure.

[0062] S2.3, A broadband ultrasonic receiving transducer synchronously acquires a continuous acoustic signal that matches the spectrum of a multi-frequency composite dynamic pressure disturbance signal after penetrating the sample.

[0063] Furthermore, the broadband ultrasonic receiving transducer synchronously acquires, at a sufficiently high sampling rate, the continuous acoustic signal that has penetrated the sample under high pressure and matches the spectrum of the multi-frequency composite dynamic pressure disturbance signal throughout the entire time period during which the broadband ultrasonic transmitting transducer emits the continuous acoustic signal, and converts the acquired analog signal into a high-fidelity digital waveform signal for recording.

[0064] S2.4 Perform frequency domain comparison and relationship analysis on the continuous acoustic signal and the multi-frequency composite dynamic pressure disturbance signal, and extract the complex frequency response function that characterizes the dynamic mechanical response of the sample under static pressure.

[0065] Furthermore, the continuous acoustic signal and the multi-frequency composite dynamic pressure disturbance signal applied by the confining pressure control unit, which are synchronously acquired by the broadband ultrasonic receiving transducer after penetrating the sample, are matched with the spectrum of the multi-frequency composite dynamic pressure disturbance signal. Digital signal preprocessing, such as denoising and filtering, is then performed on both signals. The two signals are then converted to the frequency domain. By obtaining the complex ratio of the spectrum of the continuous acoustic signal and the spectrum of the multi-frequency composite dynamic pressure disturbance signal at the corresponding frequency points, or by using the frequency response function based on the cross power spectral density estimation, frequency domain comparison and relationship analysis are performed to extract a complex array. This array is the complex frequency response function characterizing the dynamic mechanical response of the sample under a specific static water confining pressure background, which contains the amplitude response and phase response information at each analysis frequency point.

[0066] S3. Based on the complex frequency response function, strain signal and confining pressure value, construct the feature vector of the high pressure response of the sample.

[0067] S3.1 Analyze the complex frequency response function, extract the amplitude and phase information of the complex frequency response function at the characteristic frequency bands that reflect the main frequency response of the rock skeleton, and identify the frequency positions and bandwidths of the resonance peaks that characterize structural resonance and the anti-resonance valleys that characterize the pressure densification effect in the complex frequency response function.

[0068] Furthermore, characteristic frequency bands reflecting the dominant frequency response of the rock skeleton are determined on the frequency axis. This can be achieved, for example, by identifying the main energy concentration areas of the complex frequency response function amplitude spectrum within a specific frequency range or by referencing the known characteristic frequency ranges of similar rocks. Within these determined characteristic frequency bands, the amplitude and phase values ​​of the complex frequency response function at each discrete frequency point or the center frequency point of the frequency band are read as extracted amplitude and phase information. On the amplitude spectrum curve of the complex frequency response function, peak detection algorithms are used to identify extreme points with amplitudes significantly higher than the local background. The frequency positions corresponding to these extreme points and the bandwidths obtained through methods such as half-peak width are recorded, and these extreme points are identified as resonance peaks characterizing structural resonance. Similarly, valley detection algorithms are used to identify extreme points with amplitudes significantly lower than the local background on the amplitude spectrum curve. The frequency positions and bandwidths corresponding to these extreme points are recorded, and these extreme points are identified as anti-resonance valleys characterizing the pressure densification effect, thus completing the analysis and extraction of key dynamic features in the complex frequency response function.

[0069] S3.2. Combine the amplitude information, the phase information, the frequency positions and bandwidths of the resonance peaks and anti-resonance valleys of the complex frequency response function with the strain signal and the confining pressure value to construct a feature vector of the high-pressure response of the sample.

[0070] Furthermore, the amplitude information, phase information, resonance peak frequency position and bandwidth, and anti-resonance valley frequency position and bandwidth of the complex frequency response function extracted analytically from the complex frequency response function, along with the scalar features such as average strain or volumetric strain obtained after processing the strain signal measured by the fiber optic grating sensor under the same pressure step, and the confining pressure value corresponding to the pressure step, are spliced, combined, and normalized according to a predetermined order and format to form a one-dimensional numerical array. This numerical array is the constructed high-dimensional feature vector used to describe the comprehensive response of the sample under a specific high-pressure environment.

[0071] S4. Input the feature vector of the sample's high-pressure response into a graph neural network constructed based on pore network abstraction and message passing mechanism. By simulating the propagation of waves in the abstract pore network, the propagation process is modulated using the information in the feature vector of the sample's high-pressure response, thus separating the pressure effect from the structural effect.

[0072] S4.1 By analyzing the historical pore structure images of the samples, the topological connection relationship between the pore body and the throat is extracted, and a graph neural network based on pore network abstraction and message passing mechanism is constructed.

[0073] In this network, nodes correspond to pores, and edges correspond to throats.

[0074] Furthermore, based on historical pore structure images of the samples, such as three-dimensional images obtained through microcomputed tomography or scanning electron microscopy, image segmentation and pore network extraction algorithms are applied to identify connected pore spaces and narrow throat regions connecting adjacent pore spaces in the images, and to establish a list of topological adjacency relationships between pore spaces connected by throats. This topological adjacency relationship serves as the basic skeleton of the graph structure, where each pore space is mapped to a graph node, and each throat connecting two pore spaces is mapped to an undirected or directed edge connecting the two corresponding nodes. In this way, the node set and edge set of the graph structure are constructed, completing the basic construction of the graph structure of the graph neural network based on pore network abstraction and message passing mechanism.

[0075] Specifically, by extracting the topological connections between pores and throats from historical images and directly defining the graph structure of the graph neural network, the network architecture itself becomes a mathematical abstraction of the microscopic pore structure of the sample. Nodes represent pores, and edges represent the mapping relationships of throats. The graph structure, built upon real topology, allows the network to naturally process relational data in non-Euclidean space. This establishes a physically interpretable computational foundation for simulating wave propagation along the pore network and quantifying the impact of throats on information transmission. This is a structural prerequisite for achieving subsequent physical heuristic message passing and effect separation.

[0076] S4.2 In the constructed graph neural network based on the abstraction of porous network and message passing mechanism, a message passing function inspired by the stress wave propagation mechanism is defined. The message passing function specifies the update amount of information transmitted from the source node to the target node through the connecting edge, and is a function of the current state of the source node, the current attribute of the connecting edge, and the current state of the target node.

[0077] Furthermore, within the framework of a graph neural network based on porous network abstraction and message passing mechanism, which has already constructed a set of nodes and edges, a specific mathematical form of the message passing function is defined. This function receives three input parameters: the current state vector of the source node, the current attribute vector of the edge connecting the source node and the target node, and the current state vector of the target node. Based on these three input parameters, the message passing function calculates a message vector through a learnable mapping. This message vector represents the amount of information update transmitted from the source node to the target node via the connecting edge. This functional form aims to simulate the changes in information (such as energy and phase) carried by a wavefront as it propagates in a porous medium from one pore body through a connecting throat to the next pore body. These changes depend on the state of the triggering pore body, the characteristics of the throat, and the state of the pore body upon arrival.

[0078] Specifically, the physical process of wave propagation is abstracted as the transmission and interaction of information along edges on a graph from the source node to the target node. The function specifies that the update quantity is a function of the source, edge, and target, which physically corresponds to the fact that during wave propagation, the amplitude and phase are jointly affected by the properties of the emission point, the characteristics of the propagation path, and the impedance of the receiving point. By embedding physical constraints into the basic form of the message passing function, the parameter changes learned by the network during training are restricted to a solution space that conforms to the laws of physical propagation, greatly enhancing the interpretability and generalization ability of the model. The definition of the message passing function embedded in the physical mechanism is key to the graph neural network's ability to understand and simulate wave behavior in porous media.

[0079] S4.3 From the high-dimensional feature vector describing the comprehensive response of the sample under high pressure, the parameters related to the compressibility of the pore body are extracted. The parameters related to the compressibility of the pore body are assigned to the initial state of the corresponding node in the graph neural network of pore network abstraction and message passing mechanism to form the node attribute feature vector.

[0080] Furthermore, from the high-dimensional feature vector describing the comprehensive response of the sample under high pressure, based on the pre-established mapping table of correspondence between feature components and physical parameters, the parameter components related to the compressibility of the pore body are located and read, such as derived parameters that may be related to the low-frequency amplitude or specific phase characteristics of the complex frequency response function. These read parameter components related to the compressibility of the pore body are assigned one by one to the initial state vector of the corresponding node according to the one-to-one correspondence between nodes in the graph neural network of the pore network abstraction and message passing mechanism and pore bodies in the historical pore structure image, thereby generating a node attribute feature vector containing its initial compressibility attribute for each node in the network.

[0081] Specifically, through an allocation mechanism (such as based on empirical or theoretical relationships between features and physical parameters), this macroscopic information is decomposed and assigned as initial values ​​to each node representing the microscopic pore body. Parameters related to the compressibility of the pore body are assigned to the nodes, and each node is given an initial state reflecting its softness or hardness under the current high-pressure environment at the beginning of the computation. This microscopic initialization based on macroscopic features allows the graph neural network to iteratively update from a physically reasonable starting point rather than learning all microscopic properties from scratch, thus accelerating training convergence and ensuring the physical consistency between the network's internal state and external observations.

[0082] S4.4 From the high-dimensional feature vector describing the comprehensive response of the sample under a specific high-pressure environment, the parameters related to throat conductivity and equivalent size are extracted, and the parameters related to throat conductivity and equivalent size are assigned to the initial attributes of the corresponding edges in the graph neural network of pore network abstraction and message passing mechanism to form edge attribute feature vectors.

[0083] Furthermore, from the high-dimensional feature vector describing the comprehensive response of the sample under a specific high-pressure environment, based on the pre-established mapping table of correspondence between feature components and physical parameters, the parameter components related to throat conductivity and equivalent size are located and read, such as derived parameters that may be related to the high-frequency attenuation or resonance peak characteristics of the complex frequency response function. These read parameter components related to throat conductivity and equivalent size are assigned one by one to the initial attribute vector of the corresponding edge according to the correspondence between the edges in the graph neural network of pore network abstraction and message passing mechanism and the throat in the historical pore structure image, so as to generate an edge attribute feature vector containing its initial conductivity and size attributes for each edge in the network.

[0084] Specifically, parameters related to throat function in the macroscopic features are mapped to each edge. The conductivity and equivalent size of the throat are key to controlling fluid flow or stress transmission between pores and are extremely sensitive to pressure. Assigning parameters reflecting the throat state under current high pressure as edge attributes gives the edge attribute inputs in the message passing function a real physical meaning; it represents the throat's ability to function as an information transmission channel under the current confining pressure. Initialization transforms the edges in the graph neural network from static connections into variable channels that bear the influence of environmental pressure. Combined with node initialization, the attributes of nodes and edges together constitute an initial profile of the micromechanical state of the porous medium under current high pressure.

[0085] S4.5 In the message passing process of the graph neural network with pore network abstraction and message passing mechanism, the confining pressure value is read from the high-dimensional feature vector describing the comprehensive response of the sample under high pressure.

[0086] Furthermore, before the graph neural network with pore network abstraction and message passing mechanism performs message passing calculations at each layer, the confining pressure scalar value corresponding to the current pressure state is read from the specified index position of the high-dimensional feature vector describing the comprehensive response of the sample under high pressure, or through a specific mapping relationship.

[0087] Specifically, the confining pressure value is read from the high-dimensional feature vector and used as an independent input parameter in the message passing process, enabling each network update to be aware of the current specific pressure level. This differs from implicitly embedding pressure information in the initialization attributes of nodes or edges; instead, it elevates it to an independent control variable. The global explicit injection of the pressure parameter provides a direct basis for the message passing function to dynamically adjust its behavior to adapt to different pressure conditions, and is a key prerequisite for realizing pressure modulation of the message passing path, thereby simulating the impact of high pressure on wave propagation.

[0088] S4.6 In the message passing function, the confining pressure value, the throat conductivity and equivalent size parameters encoded by the edge attribute feature vector, and the pore compressibility parameters encoded by the node attribute feature vectors of adjacent nodes are jointly input into the learnable attention mechanism. The pressure sensitivity of the throat is evaluated based on the confining pressure value, throat conductivity and equivalent size parameters, and the structural stability of the pore is evaluated based on the pore compressibility parameters.

[0089] Furthermore, within the defined message passing function, the read confining pressure value, the throat conductivity and equivalent size parameters encoded in the edge attribute feature vector corresponding to the currently processed edge, and the pore volume compressibility parameters encoded in the node attribute feature vectors of the two adjacent nodes connected by the edge are concatenated or fused into a joint input vector. This joint input vector is then input into a learnable, parameter-sharing attention mechanism calculation module, such as a small feedforward neural network. The attention mechanism calculation module calculates the joint input vector and outputs two scalar evaluation values. One evaluation value represents the sensitivity of the current throat to pressure changes, determined based on the confining pressure value, throat conductivity, and equivalent size parameters, i.e., throat pressure sensitivity. The other evaluation value represents the average or combined structural stability of the pore volume pair connected by the edge, determined based on the pore volume compressibility parameters of the two adjacent nodes, i.e., pore volume structural stability.

[0090] Specifically, the competitive relationship between pressure and structural effects is transformed into a joint evaluation of path attributes (throat pressure sensitivity) and node attributes (pore structure stability). The attention mechanism computation module, through learning, can automatically discover and quantify the following patterns from combinations of confining pressure, throat attributes, and pore properties: under high confining pressure, throats with variable conductivity may be more sensitive to pressure; the edge connecting two highly compressible pores may have poor overall structural stability. This automatic evaluation based on joint parameters avoids the difficulty of manually formulating complex evaluation rules, enabling the model to learn from data the complex patterns of how pressure and structure jointly influence the behavior of microscopic conductive units.

[0091] S4.7 Based on the throat pressure sensitivity and pore structure stability, generate gating signals to control the on / off state and strength of the message transmission path. Suppress the message transmission path that is connected to a pore with low structural stability via a throat with high pressure sensitivity, and enhance the message transmission path that is connected to a pore with high structural stability via a throat with low pressure sensitivity.

[0092] Furthermore, based on the throat pressure sensitivity assessment value and the pore structure stability assessment value, a gating function, such as the sigmoid function, is used to combine the two assessment values ​​into a scalar gating signal between zero and one. The generation logic of the gating signal is set as follows: when the throat pressure sensitivity assessment value is high and the pore structure stability assessment value is low, a weak gating signal close to zero is generated, which corresponds to suppressing the message transmission path connected via a high-pressure-sensitivity throat and a low-structure-stability pore; when the throat pressure sensitivity assessment value is low and the pore structure stability assessment value is high, a strong gating signal close to one is generated, which corresponds to enhancing the message transmission path connected via a low-pressure-sensitivity throat and a high-structure-stability pore. The generated gating signal will be used as a multiplicative coefficient and directly applied to the message vector transmitted from the source node to the target node via this edge, realizing differentiated control of the information flow intensity on paths with different physical characteristics.

[0093] Specifically, under high pressure, the pressure-dominated path (high-pressure-sensitive throat connecting deformable pores) should contribute less information related to the intrinsic structure and therefore needs to be suppressed; while the structure-dominated path (stable throat connecting rigid pores) should retain and enhance its information reflecting the intrinsic structure and therefore needs to be enhanced. By converting the evaluation value into a gating signal and applying it to the message vector, the information flow is redistributed within the neural network. Gating modulation based on physical evaluation allows the network to automatically and gradually attenuate pressure-sensitive information components while highlighting information components reflecting the stable structure. The dynamic filtering process at the feature level is the core manifestation of the separation action.

[0094] S4.8. Through multi-layer iterative message passing, node state update and edge attribute update of graph neural network with porous network abstraction and message passing mechanism, the finally updated node state vector and edge state vector mainly encode information related to structural effect and information related to pressure effect, respectively, thus separating pressure effect and structural effect.

[0095] Furthermore, the message passing process of pressure information injection, local evaluation, and gating modulation is iteratively executed in multiple layers within a graph neural network based on pore network abstraction and message passing mechanism. In each layer, all nodes update their own node state vectors based on aggregated, gating-modulated neighbor messages, and all edges can also update their edge attribute feature vectors based on the messages they pass through and the node states. After a sufficient number of iterations, the information between nodes is fully exchanged and fused through the modulated paths. When the network reaches stability or a preset number of layers, the final node state vectors of all nodes mainly aggregate information from structurally stable paths, thus primarily encoding structural effect information related to the inherent pore structure of the sample. Meanwhile, the changes in edge attribute feature vectors or the cumulative gating signals related to the edges during the update process mainly reflect the modulation history of pressure on the conduction path, thus primarily encoding pressure effect information related to pressure action. This achieves the separation of pressure effects and structural effects at the internal state representation level of the network.

[0096] Specifically, this is a natural outcome of multiple rounds of propagation, competition, and sedimentation on physically constrained networks. Multi-level iteration allows information to propagate from local to global, enabling the accumulation and diffusion of the effects of local evaluation and modulation. The node states eventually converge to primarily reflect structural effects because, during the iteration process, information from pressure-sensitive paths is continuously suppressed, while information from stable structural paths is continuously enhanced and becomes dominant. The evolution of edge attributes records traces of pressure modulation. Based on the decoupling of effects from iterative propagation, the physical process of different frequency components (corresponding to different effects) attenuating and accumulating differently due to path attribute differences when waves propagate in real porous media is simulated. This makes the separation results more physically and dynamically based; pressure effects and structural effects are no longer coupled in a single signal but are sedimented in the node states representing entities and the edge states representing connections, respectively.

[0097] S5. The intrinsic porosity of the sample is estimated by executing a physically inspired message passing mechanism through the graph neural network.

[0098] S5.1 In a graph neural network based on the abstraction of pore networks that separates pressure effects and structural effects and the message passing mechanism, the node state vector that encodes information related to structural effects is read.

[0099] Furthermore, in the graph neural network that has undergone multi-layer iterative message passing, node state updates, and edge attribute updates, and has completed the separation of pressure effects and structural effects, the final state storage area of ​​all nodes in the network is accessed. The node state vector of each node, which has finally converged after information propagation and modulation within the network, is read in the order of node index. Since these node state vectors are mainly formed through information enhancement of structurally stable paths and information suppression of pressure-sensitive paths during the message passing process, they are identified as encoding information related to structural effects, thus completing the node-level representation acquisition of the separated structural effect information.

[0100] S5.2 Perform global average pooling on the node state vectors that encode information related to structural effects to generate graph-level feature vectors.

[0101] Furthermore, a global average pooling operation is performed on all node state vectors that encode information related to structural effects, read from the graph neural network with its pore network abstraction and message passing mechanism. This operation sums the components of all node state vectors at corresponding positions in the feature dimension and divides them by the total number of nodes to obtain a new vector with the same dimension as the state vector of a single node. This new vector is the generated graph-level feature vector, which represents the global statistical features of the entire sample pore network structure obtained by aggregating the local structural effect information of all nodes.

[0102] S5.3 Input the graph-level feature vector into the feedforward neural network decoder to obtain the estimated value of intrinsic porosity.

[0103] Furthermore, the graph-level feature vectors are used as input data and fed into a predefined architecture and pre-trained feedforward neural network decoder. The feedforward neural network decoder consists of multiple fully connected layers and nonlinear activation functions. After receiving the graph-level feature vectors, it performs linear transformations and nonlinear mappings layer by layer according to their internal parameters and computational graph. The last layer of the feedforward neural network decoder usually uses an appropriate activation function to ensure that the output value is within a reasonable physical range. A scalar value is generated from the output layer of the feedforward neural network decoder. This scalar value is the estimate of the intrinsic porosity of the sample under the current high-pressure conditions.

[0104] The expression for estimating intrinsic porosity is:

[0105] ;

[0106] in, This is an estimate of the intrinsic porosity. This is the decoder function for the feedforward neural network. For the graph-level feature vectors, It is a feedforward neural network decoder.

[0107] S6. By correlating the estimated intrinsic porosity of samples under different pressures with the corresponding confining pressure, a curve describing the relationship between intrinsic porosity and pressure is generated.

[0108] S6.1. Collect the estimated intrinsic porosity values ​​obtained under all pressure steps during the stepped static pressure loading process to form a sequence of intrinsic porosity estimates.

[0109] Furthermore, following the loading sequence of the pressure steps during the stepped static pressure loading process, from low confining pressure to high confining pressure, the estimated intrinsic porosity obtained by the feedforward neural network decoder at each pressure step is accessed and read sequentially. These estimated intrinsic porosity values ​​are then stored in a one-dimensional array according to the order of the same pressure steps read. This one-dimensional array is the sequence of estimated intrinsic porosity values ​​arranged according to the pressure loading process.

[0110] S6.2. Collect the confining pressure values ​​of all pressure steps recorded during the stepped static pressure loading process to form a confining pressure value sequence corresponding to the estimated value sequence of intrinsic porosity.

[0111] Furthermore, following the loading sequence of the pressure steps during the stepped static pressure loading process, from low confining pressure to high confining pressure, the confining pressure values ​​recorded in the basic dataset and synchronously recorded under each corresponding pressure step are accessed and read in sequence. These confining pressure values ​​are then stored in a one-dimensional array according to the order of the same pressure steps read. This one-dimensional array is the sequence of confining pressure values ​​that corresponds to the estimated sequence of intrinsic porosity.

[0112] S6.3 In a two-dimensional coordinate system, plot the data points using the confining pressure value sequence as the abscissa and the estimated intrinsic porosity value sequence as the ordinate, and connect the data points to generate a curve describing the relationship between intrinsic porosity and pressure.

[0113] Furthermore, a two-dimensional Cartesian coordinate system is established. Each value in the confining pressure value sequence is used as the abscissa value of each data point in the two-dimensional coordinate system. The values ​​at the same index positions in the intrinsic porosity estimation value sequence as the confining pressure value sequence are used as the ordinate values ​​of the corresponding data points, thereby determining a series of discrete data points in the two-dimensional coordinate system. Line segments are used to connect adjacent data points in ascending order of their abscissa values, or an interpolation algorithm is used to generate a smooth curve. Through these data points, a curve describing the relationship between intrinsic porosity and pressure is generated in the two-dimensional coordinate system.

[0114] This embodiment also provides a computer device applicable to the sample porosity correction method under high pressure conditions, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the sample porosity correction method under high pressure conditions as proposed in the above embodiment.

[0115] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0116] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the sample porosity correction method under high pressure as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0117] In summary, this invention uses stepped static pressure loading and synchronous sensing to acquire a basic dataset. Multi-frequency dynamic pressure disturbances are superimposed on the static pressure background to excite matched acoustic signals. A complex frequency response function containing rich dynamic information is extracted through frequency domain analysis, and a high-dimensional feature vector is constructed. This feature vector is then input into a graph neural network based on the abstraction of a pore network. This network, through its physically inspired message passing mechanism, simulates the propagation of waves in the pore network and dynamically modulates the propagation path using information such as confining pressure values. This achieves effective separation of pressure effects and structural effects in the internal representation. Based on the separated structural information, the network outputs an intrinsic porosity estimate through a decoder and generates an intrinsic porosity-confining pressure relationship curve.

[0118] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for correcting sample porosity under high pressure conditions, characterized in that: include, Collect the confining pressure, strain signal and ultrasonic waveform signal of the sample to form a basic dataset, including the following steps; At each pressure step, the data acquisition system is synchronously triggered to record the precise confining pressure value at the current moment, the axial and circumferential strain signals measured by fiber optic grating sensors distributed on the sample surface, and the full-waveform digital ultrasonic signal excited by a broadband ultrasonic transmitting transducer, propagated through the sample, and acquired by a broadband ultrasonic receiving transducer. The confining pressure values, strain signals, and ultrasonic waveform signals recorded synchronously at the pressure steps during the stepped pressure loading process are collected and structured to form a basic dataset. On the background of static water confining pressure of the pressure step, a preset multi-frequency dynamic pressure disturbance signal is superimposed and applied, and acoustic signals are excited and collected. The complex frequency response function is extracted by processing the dynamic pressure disturbance signal and the acoustic signal. Based on the complex frequency response function, strain signal and confining pressure value, a feature vector of the high-pressure response of the sample is constructed. The feature vector of the high-pressure response of the sample is input into a graph neural network constructed based on pore network abstraction and message passing mechanism. By simulating the propagation of waves in the abstract pore network, the propagation process is modulated using the information in the feature vector of the high-pressure response of the sample, and the pressure effect and structural effect are separated. The graph neural network executes a physics-inspired message passing mechanism to obtain an estimate of the intrinsic porosity of the sample. By correlating the estimated intrinsic porosity of the sample under different pressures with the corresponding confining pressure, a curve describing the relationship between intrinsic porosity and pressure is generated.

2. The sample porosity correction method under high pressure conditions as described in claim 1, characterized in that: The formation of the basic dataset includes, The confining pressure control unit executes stepped static pressure loading, starting from the initial confining pressure, increasing the pressure of the hydraulic medium in the high-pressure chamber to a preset pressure step for acquisition and recording. The confining pressure value, the strain signal measured by the fiber optic grating sensor, and the ultrasonic wave signal acquired by the broadband ultrasonic transducer are recorded.

3. The sample porosity correction method under high pressure conditions as described in claim 2, characterized in that: The extraction of the complex frequency response function includes, On top of the static confining pressure background established by the pressure step recorded in the basic dataset, a multi-frequency composite dynamic pressure disturbance signal with controlled amplitude is additionally applied by the confining pressure control unit. When a multi-frequency composite dynamic pressure disturbance signal is superimposed, the broadband ultrasonic transducer emits a continuous acoustic signal that matches the spectrum of the multi-frequency composite dynamic pressure disturbance signal. A broadband ultrasonic receiving transducer synchronously acquires a continuous acoustic signal that matches the spectrum of a multi-frequency composite dynamic pressure disturbance signal after penetrating the sample. Frequency domain comparison and relationship analysis were performed on the continuous acoustic signal and the multi-frequency composite dynamic pressure disturbance signal to extract the complex frequency response function characterizing the dynamic mechanical response of the sample under static pressure.

4. The sample porosity correction method under high pressure conditions as described in claim 3, characterized in that: The feature vector of the constructed sample high-pressure response includes The complex frequency response function is analyzed to extract the amplitude and phase information of the complex frequency response function at the characteristic frequency bands that reflect the main frequency response of the rock skeleton, and to identify the frequency positions and bandwidths of the resonance peaks that characterize structural resonance and the anti-resonance valleys that characterize the pressure densification effect in the complex frequency response function. By combining the amplitude information, the phase information, the frequency positions and bandwidths of the resonance peaks and anti-resonance valleys of the complex frequency response function with the strain signal and the confining pressure value, a feature vector of the high-pressure response of the sample is constructed.

5. The sample porosity correction method under high pressure conditions as described in claim 4, characterized in that: The simulated wave propagation in the abstract porous network includes, Three-dimensional images obtained by microcomputed tomography or scanning electron microscopy are analyzed to extract the topological connection relationship between pore bodies and throats by analyzing the historical pore structure three-dimensional images of the sample, and a graph neural network based on pore network abstraction and message passing mechanism is constructed. In this network, the nodes correspond to the pores, and the edges correspond to the throats. In the constructed graph neural network based on the abstraction of porous network and message passing mechanism, a message passing function inspired by the stress wave propagation mechanism is defined. The message passing function specifies the update amount of information transmitted from the source node to the target node through the connecting edge, and is a function of the current state of the source node, the current attribute of the connecting edge, and the current state of the target node. From the high-dimensional feature vector describing the comprehensive response of the sample under high pressure, parameters related to the compressibility of the pore body are extracted. These parameters are then assigned to the initial state of the corresponding node in the graph neural network of pore network abstraction and message passing mechanism to form the node attribute feature vector. From the high-dimensional feature vector describing the comprehensive response of the sample under a specific high-pressure environment, parameters related to throat conductivity and equivalent size are extracted. These parameters are then assigned to the initial attributes of the corresponding edges in the graph neural network of pore network abstraction and message passing mechanism, forming edge attribute feature vectors.

6. The sample porosity correction method under high pressure conditions as described in claim 5, characterized in that: The separation pressure effect and structural effect include, In the message passing process of graph neural networks with pore network abstraction and message passing mechanism, the confining pressure value is read from the high-dimensional feature vector describing the comprehensive response of the sample under high pressure. In the message passing function, the confining pressure value, the throat conductivity and equivalent size parameters encoded by the edge attribute feature vector, and the pore compressibility parameters encoded by the node attribute feature vectors of adjacent nodes are jointly input into the learnable attention mechanism. The pressure sensitivity of the throat is evaluated based on the confining pressure value and the throat conductivity and equivalent size parameters, and the structural stability of the pore is evaluated based on the pore compressibility parameters. Based on the throat pressure sensitivity and pore structure stability, a gating signal is generated to control the on / off state and strength of the message transmission path. The message transmission path connected to a high-pressure-sensitive throat and a low-structure-stability pore is suppressed, while the message transmission path connected to a low-pressure-sensitive throat and a high-structure-stability pore is enhanced. Through multi-layer iterative message passing, node state update, and edge attribute update of the graph neural network using the porous network abstraction and message passing mechanism, the finally updated node state vector and edge state vector mainly encode information related to structural effects and information related to pressure effects, respectively, thus separating pressure effects and structural effects.

7. The sample porosity correction method under high pressure conditions as described in claim 6, characterized in that: The estimated intrinsic porosity of the sample includes, In a graph neural network based on the abstraction of pore networks that separates pressure effects and structural effects and the message passing mechanism, the node state vectors that encode information related to structural effects are read. Global average pooling is performed on the node state vectors that encode information related to structural effects to generate graph-level feature vectors; The graph-level feature vectors are input into the feedforward neural network decoder to obtain an estimate of the intrinsic porosity.

8. The sample porosity correction method under high pressure conditions as described in claim 7, characterized in that, The curves describing the relationship between intrinsic porosity and pressure include... The estimated values ​​of intrinsic porosity obtained at all pressure steps during the stepped static loading process are collected to form a sequence of estimated intrinsic porosity values. Collect the confining pressure values ​​of all pressure steps recorded during the stepped static pressure loading process to form a confining pressure value sequence corresponding to the estimated value sequence of intrinsic porosity; In a two-dimensional coordinate system, the confining pressure value sequence is used as the horizontal axis and the estimated intrinsic porosity value sequence is used as the vertical axis. Data points are plotted and connected to generate a curve describing the relationship between intrinsic porosity and pressure.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the sample porosity correction method under high pressure conditions as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the sample porosity correction method under high pressure conditions as described in any one of claims 1 to 8.

Images