A method, medium and system for determining applied air pressure parameters of a pore water sampler

Abstract

The present invention provides a method, medium and system for determining the applied air pressure parameters of a pore water sampler, belonging to the technical field of pore water samplers, and comprising: collecting the surface image and ultrasonic reflection data of a first pore water sample, and obtaining physical parameters such as porosity, permeability, etc. by using analytical technology; calculating the maximum capillary pressure according to the capillary pressure model, and applying it to the sample as the initial air pressure; performing a pore water separation test on the sample, and recording the separation parameters; monitoring the sample destruction degree at the same time, and determining the maximum allowable air pressure; determining the optimal air pressure for producing the maximum pore water volume; performing a series of gradient air pressure separation tests on the first sample, and establishing a response surface model of air pressure, time and pore water volume; optimizing the model by using a genetic algorithm, and obtaining the optimal air pressure and separation time parameters; and outputting the optimization results for separating the second sample collected in the same area. The pore water in the sample can be displaced to the maximum extent, and the sample can be effectively prevented from being excessively damaged.

Description

Technical Field

[0001] The present invention belongs to the technical field of pore water samplers, and in particular, relates to a method, medium and system for determining applied air pressure parameters of a pore water sampler. Background Art

[0002] At present, the most commonly used method for evaluating shale reservoir characteristics is pore water analysis. Pore water refers to free water or bound water in the pore space of rocks. It can truly reflect the diagenetic evolution process and reservoir location of the reservoir, and its composition and isotopic characteristics can provide important information for reservoir diagenesis, fluid migration and hydrocarbon source materials. Therefore, collecting and analyzing pore water is of great significance for evaluating the pore structure, fluid properties, diagenesis, etc. of shale reservoirs.

[0003] However, due to the extremely low porosity and tiny pore size of shale reservoirs, it is difficult to effectively collect enough pore water samples from them using conventional sampling techniques such as centrifugation and extrusion. To solve this problem, the gas pressure displacement method came into being. This method uses high-pressure gas to displace the free and bound pore water in the reservoir core and collect enough pore water samples for subsequent analysis. At present, the use of gas pressure displacement to displace pore water often relies on artificial experience to control the applied gas pressure, which leads to two situations: one is that the applied gas pressure is too low, the pore water cannot be completely displaced, and the sampling efficiency is low; the other is that the applied gas pressure is too high, which will damage the shale sample, especially in the displacement process, relying on artificial experience to control the continuous increase of gas pressure, it is difficult to displace the pore water in the sample to the maximum extent, and it can effectively prevent excessive damage to the sample. Summary of the invention

[0004] In view of this, the present invention provides a method, medium and system for determining the applied air pressure parameters of a pore water sampler, which can solve the technical problem that it is difficult to displace the pore water in the sample to the maximum extent due to relying on artificial experience to control the continuous increase of air pressure during the displacement process, and can effectively prevent excessive damage to the sample.

[0005] The present invention is achieved in that:

[0006] A first aspect of the present invention provides a method for determining an applied air pressure parameter of a pore water sampler, comprising the following steps:

[0007] S10, collecting surface images and ultrasonic reflection signal data of the first pore water sample, and obtaining physical parameters of the first pore water sample by using image analysis and reflection wave inversion technology;

[0008] S20, combining the obtained physical parameters and according to a pre-fitted capillary pressure model, calculating the maximum capillary pressure of the first pore water sample as the initial air pressure value applied, wherein the initial air pressure value is greater than the maximum capillary pressure;

[0009] S30, applying an air pressure of an initial air pressure value to the first pore water sample, performing a pore water separation test on the first pore water sample, and recording separation parameters, including separation time at specified intervals and the amount of pore water collected;

[0010] S40, during the pore water separation test, gradually increase the air pressure value, and calculate the first pore water sample destructiveness in real time using the surface image and ultrasonic reflection signal data of the first pore water sample based on the pre-trained first pore water sample destructiveness model, and determine the maximum allowable air pressure value according to the pre-set maximum allowable value of the first pore water sample destructiveness;

[0011] S50, plotting a relationship curve between air pressure and the final amount of pore water collected, and determining an optimal air pressure value for producing the maximum amount of pore water, wherein the optimal air pressure value does not exceed the maximum allowable air pressure value;

[0012] S60, obtaining separation parameters of a series of pore water separation tests performed on the first pore water sample, wherein the applied gas pressure of the series of pore water separation tests is a sequential gradient gas pressure near an optimal gas pressure;

[0013] S70, establishing a response surface model between air pressure, separation time, and collected pore water volume;

[0014] S80, using the separation parameters of the series of pore water separation tests and their corresponding applied air pressures as multiple initial values, solving the response surface model using a genetic algorithm to obtain an optimal combination of air pressure and separation time parameters as an applied air pressure parameter;

[0015] S90, outputting the applied air pressure parameter to an operator for applying air pressure to separate pore water in a second pore water sample collected from the same area as the first pore water sample.

[0016] Wherein, the physical parameters include at least porosity and permeability.

[0017] The step S10 specifically includes:

[0018] Step 101: using a high-resolution imaging device to collect a surface image of the first pore water sample, wherein the surface image resolution is at least 6 million pixels;

[0019] Step 102: Use an ultrasonic scanner to perform a comprehensive ultrasonic scan on the first pore water sample and record ultrasonic reflection signal data, wherein the ultrasonic scan resolution is at least 100 microns;

[0020] Step 103, preprocessing the surface image, wherein the preprocessing includes noise reduction and contrast enhancement;

[0021] Step 104: Analyze the surface image features using an image analysis algorithm based on machine learning to extract physical parameters such as porosity and pore diameter distribution;

[0022] Step 105: Analyze the ultrasonic reflection signal data by using the reflection wave inversion method to extract physical parameters such as porosity and permeability.

[0023] The steps of training the capillary pressure model are specifically:

[0024] Establishing a training data set, specifically obtaining multiple groups of soil or rock, collecting their physical parameters, and using a specific test method to obtain their maximum capillary pressure;

[0025] A training model is provided, wherein the training data set is used to train a regression model using a Gaussian process regression algorithm to obtain the capillary pressure model.

[0026] The specified duration is 2 to 10 minutes.

[0027] The step S30 specifically includes:

[0028] Step 301, placing the first pore water sample in a specially made pressure vessel;

[0029] Step 302: applying the initial air pressure value determined in step S20 to the pressure vessel through a pressure control system, wherein the accuracy of the pressure control system is ±0.01 MPa;

[0030] Step 303, record the separation time at intervals of a specified time, preferably 5 minutes;

[0031] Step 304: Use a weighing sensor to continuously monitor the amount of pore water collected, wherein the weighing sensor has an accuracy of 0.1 ml;

[0032] During the experiment, environmental parameters such as temperature and pressure are monitored in real time to ensure stable test conditions.

[0033] The first pore water sample damage model adopts a deep feedforward neural network model, and the training steps are specifically:

[0034] Establishing a first training set, specifically a plurality of groups of historical data, each group of historical data including the surface image and ultrasonic reflection signal data of the first pore water sample obtained in S10-S40, and the degree of destruction of the first pore water sample recorded each time in the pore water separation test;

[0035] A training model is provided, wherein a deep feedforward neural network model is trained using the first training set to obtain the first pore water sample destructiveness model.

[0036] The step S40 specifically includes:

[0037] Step 401: while performing the pore water separation test, continue to collect the surface image and ultrasonic reflection signal data of the first pore water sample;

[0038] Step 402: using the same image analysis and reflection wave inversion algorithm as step S10, real-time update and acquisition of physical parameters such as porosity and permeability;

[0039] Step 403: Calculate the degree of destruction of the first pore water sample by using a pre-trained deep feedforward neural network model according to the physical parameters and the initial image and ultrasonic signal characteristics;

[0040] Step 404: The destructiveness model is trained with a large amount of real data;

[0041] Step 405: when the degree of damage exceeds a preset threshold, it is determined that the air pressure exceeds the maximum allowable air pressure value;

[0042] The preset threshold value range is 0.7-0.9.

[0043] , the step S50 specifically includes: after the pore water separation test of step S30 is completed, a relationship curve between the air pressure and the final amount of collected pore water is drawn; the relationship curve shows a trend of first rising and then falling; the rising section of the curve indicates that the increase in air pressure is conducive to displacing more pore water; the falling section of the curve indicates that the excessive air pressure begins to cause sample damage and the collection amount decreases; the air pressure corresponding to the peak of the curve is the optimal air pressure value for producing the maximum amount of pore water;

[0044] The optimal air pressure value is less than the maximum allowable air pressure value determined in step S40.

[0045] The step S60 specifically includes:

[0046] Step 601, setting a series of air pressure gradients, wherein the air pressure gradients are located near the optimal air pressure determined in step S50;

[0047] Step 602, the pressure gradient range is controlled within ±10% of the optimal pressure;

[0048] Step 603, the air pressure gradient interval is 1%-3% of the optimal air pressure;

[0049] Step 604, applying the air pressure gradient to the first pore water sample in sequence;

[0050] Step 605: Record the separation parameter data of each test, wherein the separation parameter data includes the separation time and the amount of collected pore water.

[0051] Furthermore, the step of determining the degree of damage is specifically:

[0052] Step 1, respectively extracting the features of the surface image and ultrasonic reflection signal data of the first pore water sample, and recording them as basic image features and basic ultrasonic features respectively;

[0053] Step 2: Fusing the basic image features and basic ultrasound features to obtain fused features, which are recorded as basic fused features;

[0054] Step 3, conduct a pore water separation test, gradually increase the air pressure value, and repeat steps 1 and 2 at specified intervals to obtain fusion features under multiple different air pressure values, which are recorded as pressure-adjusted fusion features;

[0055] Step 4: Calculate the difference between each of the pressure-adjusting fusion features and the basic fusion feature, and use the difference as the degree of destruction of the first pore water sample measured at specified time intervals.

[0056] Wherein, the response surface model adopts a multivariate nonlinear regression model.

[0057] The step S70 specifically includes:

[0058] Step 701, using a series of test data obtained in step S60;

[0059] Step 702, training a response surface model between air pressure, separation time and collected pore water volume through multivariate nonlinear regression;

[0060] Step 703: training the response surface model using a support vector machine regression algorithm;

[0061] Among them, the response surface model can fit the complex nonlinear relationship between air pressure, separation time and collected pore water volume; if any air pressure and separation time values ​​are given, the response surface model is used to predict the corresponding collected pore water volume.

[0062] The genetic algorithm takes maximizing the amount of collected pore water and minimizing the separation time as optimization goals.

[0063] The step S80 specifically includes:

[0064] Step 801, using the applied air pressure of each test in step S60 as a plurality of initial values ​​in the genetic algorithm;

[0065] Step 802, taking the response surface model established in step S70 as the objective function;

[0066] Step 803, setting two optimization goals, including maximizing the amount of collected pore water and minimizing the separation time;

[0067] Step 804: Use a genetic algorithm to perform multi-objective optimization on the two optimization objectives;

[0068] Step 805, the main parameters of the genetic algorithm include population size 100, crossover probability 0.8, mutation probability 0.2, and termination generation 200;

[0069] Step 806: After several generations of evolution, the genetic algorithm converges to a set of optimal air pressure and separation time parameter combinations.

[0070] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the above-mentioned method for determining the applied air pressure parameters of a pore water sampler.

[0071] A third aspect of the present invention provides a system for determining applied air pressure parameters of a pore water sampler, which includes the above-mentioned computer-readable storage medium.

[0072] Compared with the prior art, the method, medium and system for determining the applied air pressure parameters of a pore water sampler provided by the present invention have the following beneficial effects:

[0073] (1) Avoid the problem of inability to completely displace pore water due to too low gas pressure. The present invention predicts a reasonable initial gas pressure range by establishing a capillary pressure model, and further optimizes the gas pressure value on this basis, ensuring that the pore water can be completely displaced.

[0074] (2) Avoid the problem of sample damage caused by excessive air pressure. The present invention detects the degree of sample damage in real time and immediately stops increasing air pressure when it exceeds the allowable threshold, thereby effectively protecting the integrity of the sample.

[0075] (3) Shorten the experimental time and improve sampling efficiency. The present invention can quickly determine the optimal gas pressure and separation time parameters through response surface analysis and genetic algorithm optimization, greatly shortening the experimental time.

[0076] In summary, the present invention solves the technical problem that it is difficult to displace the pore water in the sample to the maximum extent by relying on artificial experience to control the continuous increase of gas pressure during the displacement process, and can effectively prevent excessive damage to the sample. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0078] Figure 1 A flow chart of the method provided by the present invention; DETAILED DESCRIPTION

[0079] In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0080] like Figure 1 FIG. 1 is a flow chart of a method for determining applied air pressure parameters of a pore water sampler provided by the present invention. The method comprises the following steps:

[0081] S10, collecting the surface image and ultrasonic reflection signal data of the first pore water sample, and obtaining the physical parameters of the first pore water sample by using image analysis and reflection wave inversion technology;

[0082] S20, combining the obtained physical parameters and according to the pre-fitted capillary pressure model, calculating the maximum capillary pressure of the first pore water sample as the initial air pressure value applied, and the initial air pressure value is greater than the maximum capillary pressure;

[0083] S30, applying an air pressure of an initial air pressure value to the first pore water sample, performing a pore water separation test on the first pore water sample, and recording separation parameters, including separation time at specified intervals and the amount of pore water collected;

[0084] S40, during the pore water separation test, gradually increase the air pressure value, and calculate the first pore water sample destructiveness in real time using the surface image and ultrasonic reflection signal data of the first pore water sample based on the pre-trained first pore water sample destructiveness model, and determine the maximum allowable air pressure value according to the pre-set maximum allowable value of the first pore water sample destructiveness;

[0085] S50, drawing a relationship curve between air pressure and the final amount of pore water collected, and determining the optimal air pressure value for producing the maximum amount of pore water, wherein the optimal air pressure value does not exceed the maximum allowable air pressure value;

[0086] S60, obtaining separation parameters of a series of pore water separation tests performed on the first pore water sample, wherein the applied gas pressure of the series of pore water separation tests is a sequence gradient gas pressure near the optimal gas pressure;

[0087] S70, establishing a response surface model between air pressure, separation time, and collected pore water volume;

[0088] S80, using a series of separation parameters of pore water separation tests and their corresponding applied air pressures as multiple initial values, solving the response surface model using a genetic algorithm to obtain an optimal combination of air pressure and separation time parameters as an applied air pressure parameter;

[0089] S90, outputting the applied air pressure parameter to the operator for applying air pressure to separate the pore water in the second pore water sample collected from the same area as the first pore water sample.

[0090] The specific implementation methods of the above steps are described in detail below:

[0091] The specific implementation method of step S10 is: first, use a high-resolution imaging device to collect a surface image of the first pore water sample. The image resolution should be at least 6 million pixels to ensure clear image details. Secondly, use an ultrasonic scanner to perform a comprehensive ultrasonic scan of the sample and record the reflection signal data. The ultrasonic scanning resolution should be at least 100 microns. Next, the collected image is preprocessed, including noise reduction, contrast enhancement, etc., to improve the image quality. Then, an image analysis algorithm based on machine learning, such as a deep convolutional neural network algorithm, is used to analyze the image features and extract parameters such as porosity and pore diameter distribution. At the same time, the ultrasonic reflection signal data is analyzed using a reflection wave inversion algorithm to extract parameters such as permeability. The purpose of this step is to obtain the key physical parameters of the first pore water sample and provide input data for subsequent steps.

[0092] The specific implementation method of step S20 is: for different types of rocks, multiple capillary pressure models are established in advance, and these models are obtained by training through machine learning methods. For example, for dense sandstone, a model based on the coagulated salt solution theory can be used; for shale, a model based on the bound volume theory can be used. These models can predict the maximum capillary pressure of the first pore water sample based on the parameters such as porosity and permeability obtained in step S10. A typical implementation method is to train the model using a Gaussian process regression algorithm. The purpose of this step is to determine a reasonable initial gas pressure value, which should be greater than the maximum capillary pressure of the sample to ensure that the pore water can be displaced.

[0093] The specific implementation method of step S30 is: place the first pore water sample in a special pressure container, and apply the initial air pressure value determined in step S20 to the container through the pressure control system. The pressure control system needs to have an accuracy of ±0.01MPa. Then start the pore water separation test and record the separation parameters in real time. Specifically, it includes the following sub-steps: (1) record the separation time at intervals of 5 minutes; (2) use a weighing sensor to continuously monitor the amount of collected pore water with an accuracy of 0.1 ml; (3) monitor the environmental parameters such as temperature and pressure in real time to ensure that the test conditions are stable. The purpose of this step is to obtain the separation parameter data of the first pore water sample under the initial air pressure conditions.

[0094] The specific implementation method of step S40 is: while the pore water separation test is being carried out, the surface image and ultrasonic reflection signal data of the first pore water sample are continuously collected. The physical parameters such as porosity and permeability are updated in real time using the same image analysis and reflection wave inversion algorithm as step S10. Then, based on these physical parameters and the initial image and ultrasonic signal characteristics, the degree of destruction of the first pore water sample is calculated by a pre-trained deep feedforward neural network model. The destruction degree model has been trained with a large amount of real data and can accurately evaluate the degree of damage of the sample under the action of air pressure. When the calculated degree of destruction exceeds a preset threshold value (usually 0.7-0.9, depending on the requirements for sample protection), it is determined that the air pressure exceeds the maximum allowable air pressure value, and it is necessary to stop applying air pressure to protect the integrity of the sample. The purpose of this step is to dynamically monitor the sample damage to ensure that the air pressure does not damage the sample.

[0095] The specific implementation of step S50 is: after the pore water separation test of step S30 is completed, a relationship curve between the air pressure and the final amount of pore water collected is drawn. The curve usually shows a trend of first rising and then falling. The rising section of the curve indicates that the increase in air pressure is conducive to displacing more pore water; the falling section of the curve indicates that the excessive air pressure begins to cause sample damage and the collection amount decreases. The air pressure corresponding to the peak of the curve is the optimal air pressure value for producing the maximum amount of pore water. The optimal air pressure value needs to be less than the maximum allowable air pressure value determined in step S40 to avoid damaging the sample. The purpose of this step is to find an optimal applied air pressure that can collect the most pore water while protecting the integrity of the sample.

[0096] The specific implementation of step S60 is: set a series of pressure gradients, and these pressure values ​​are all located near the optimal pressure determined in step S50. Usually the gradient range is controlled to be ±10% of the optimal pressure, and the gradient interval is 1%-3% of the optimal pressure. Apply these gradient pressures to the first pore water sample in turn, and record the separation parameter data of each test, including separation time and collected pore water volume, as in step S30. The purpose of this step is to provide sufficient test data for training the response surface model.

[0097] The specific implementation of step S70 is: using a series of test data obtained in step S60, a response surface model between air pressure, separation time and collected pore water volume is trained through multivariate nonlinear regression. The typical implementation of this model is to use a support vector machine regression algorithm, which can effectively fit the complex nonlinear relationship between the three. After the model training is completed, given any set of air pressure and separation time values, the corresponding collected pore water volume can be predicted. The purpose of this step is to establish a mathematical model between air pressure, time and collection volume to provide a basis for subsequent optimization.

[0098] The specific implementation method of step S80 is: the applied air pressure of each test in step S60 is used as multiple initial values ​​in the genetic algorithm. The response surface model established in step S70 is used as the objective function, and two optimization goals are set: (1) maximizing the amount of collected pore water; (2) minimizing the separation time. The genetic algorithm is used to perform multi-objective optimization and solve these two goals. The main parameters of the genetic algorithm include: population size 100, crossover probability 0.8, mutation probability 0.2, and termination generation 200. After several generations of evolution, the algorithm will eventually converge to a set of optimal air pressure and separation time parameter combinations that balance the collection volume and time overhead. The purpose of this step is to comprehensively consider multiple goals and determine the optimal parameter value of the applied air pressure.

[0099] The specific implementation of step S90 is: the optimal air pressure and separation time parameter combination determined in step S80 is output and presented to the operator, as a second pore water sample collected in the same area is subjected to air pressure separation. The method of the present invention provides an applied air pressure parameter of the pore water sampler, which is used to pressurize subsequent pore water samples such as the second pore water sample to achieve pore water separation.

[0100] In order to better understand this method, the specific implementation method of this method is explained in more detail below in combination with specific formulas:

[0101] The specific implementation method of step S10:

[0102] First, a high-resolution imaging device (such as an electron microscope) is used to collect a surface image I of the first pore water sample, and the image resolution is not less than n p Pixels, usually N p ≥6 million pixels. At the same time, use a phased array ultrasonic scanner to fully scan the sample and record the reflected signal data R. The scanning resolution is not less than d r , usually d r ≤100μm.

[0103] The collected image I is preprocessed and the noise reduction process can be done by using median filtering:

[0104] I′(x,y)=median{I(s,t)} (s,t)∈Ω(x,y)

[0105] Where Ω(x,y) is the neighborhood centered at (x,y). To enhance the contrast, methods such as histogram equalization can be used.

[0106] Next, we use an image analysis algorithm based on a deep convolutional neural network (CNN) to extract parameters such as porosity φ and pore diameter distribution f(r). CNN consists of convolutional layers, pooling layers, and fully connected layers, and can automatically learn image features. Let the input be I″ and the convolution kernel be Kl , with a bias of b l , then the output of the l-th layer convolution is:

[0107] C l =σ(I l-1 *K l +b l )

[0108] Where σ is the activation function, usually the ReLU function. After several convolution and pooling layers, it is connected to the fully connected layer to predict parameters such as porosity and pore distribution.

[0109] At the same time, the reflected wave inversion algorithm is used to analyze the ultrasonic reflection signal R to obtain parameters such as permeability k. The inversion algorithm is based on the wave equation, and the solution process can use numerical algorithms such as the quasi-Newton iteration method.

[0110] The specific implementation method of step S20:

[0111] For different rock types (sandstone, shale, etc.), multiple capillary pressure models P are established in advance. c = f(φ, k, ...), used to predict the maximum capillary pressure of pore water samples These models are obtained by training on large amounts of data using machine learning methods.

[0112] For dense sandstone, a model based on the coagulated salt solution theory can be used. This theory takes into account the interaction of ions in narrow pores, and the pressure consists of two parts: the mean field potential and the ion vibration potential:

[0113] P c =P mf +P vib

[0114] The mean field potential is determined by the ion concentration and electrostatic interaction, and the ion vibration potential is determined by the pore size and temperature. By solving the ion distribution equation, the capillary pressure model can be obtained:

[0115]

[0116] For shale, a model based on bound volume theory can be used, which takes into account the interaction between organic matter and inorganic minerals in shale. This theory regards organic matter as a viscoelastic porous medium, whose pressure consists of Langmuir isothermal compression term and viscoelastic stress term:

[0117]

[0118] Where P L is the Langmuir isothermal compression term, λ is the rigidity coefficient of inorganic minerals, which is related to shale composition, temperature and pressure, Vb 、V f are the bound volume and free volume, respectively.

[0119] For other rock types, double-layer water theory, fractal theory, etc. can be used to construct pressure models. These models are obtained by training on a large amount of measured data through machine learning algorithms such as Gaussian process regression (GPR). GPR uses kernel functions to model the correlation between samples, and the kernel function usually uses RBF function:

[0120]

[0121] in and l are the hyperparameters of the kernel function. By maximizing the marginal likelihood to estimate the hyperparameters, the pressure model can be obtained.

[0122] According to the physical parameters such as porosity and permeability obtained in step S10, the trained pressure model is substituted to calculate the maximum capillary pressure of the first pore water sample. Then determine the initial air pressure value To ensure that the pore water can be displaced.

[0123] The specific implementation method of step S30:

[0124] The first pore water sample is placed in a specially made high-pressure container, and the initial air pressure P determined in step S20 is applied to the container through a pressure control system with an accuracy of not less than ±0.01 MPa. 0 .

[0125] Then the pore water separation test is started, and the separation time t is recorded every Δt. i , usually Δt = 5 minutes. Use a weighing sensor with an accuracy of ΔV to continuously monitor the collected pore water volume V(t i ), usually ΔV=0.1ml. At the same time, the environmental parameters such as temperature T and pressure P are monitored in real time to ensure the stability of the test conditions and the temperature is controlled at (25±1)℃.

[0126] Specific implementation of step S40:

[0127] During the pore water separation test, the surface image I(t) and ultrasonic reflection signal R(t) of the first pore water sample are continuously collected. The physical parameters such as porosity φ(t) and permeability k(t) are updated in real time using the same image analysis and reflection wave inversion algorithm as step S10.

[0128] According to the physical parameters φ(t), k(t) and the initial image features I(0), R(0), the damage degree D(t) is calculated using the pre-trained deep feedforward neural network (DNN) model:

[0129] D(t)=DNN(φ(t),k(t),I(0),R(0))

[0130] The DNN model consists of an L-layer fully connected network. The input layer includes porosity, permeability, and image features. The middle layer uses the ReLU activation function, and the output layer uses the Sigmoid activation function. The DNN model is trained on a large amount of real sample damage data and can accurately assess the degree of damage of the sample under air pressure. When the calculated damage degree D(t) exceeds the preset threshold D th When the pressure P(t) exceeds the maximum allowable pressure value P max , it is necessary to stop applying air pressure to protect the integrity of the sample. Usually D th Between 0.7 and 0.9, the specific value depends on the requirements for sample protection.

[0131] The specific implementation method of step S50:

[0132] After the pore water separation test in step S30 is completed, the air pressure P and the final collected pore water volume V are plotted. f The relationship curve between . This curve usually shows a Gaussian distribution trend that rises first and then falls:

[0133]

[0134] Where A is the curve amplitude, μ is the optimal air pressure corresponding to the curve peak, and σ is the curve width.

[0135] The rising section of the curve shows that the increase of gas pressure P is conducive to displacing more pore water, resulting in an increase in the collection volume V f The descending section of the curve indicates that the excessive pressure begins to damage the sample and the collected volume decreases. The pressure μ corresponding to the peak of the curve is the optimal pressure value for producing the maximum pore water volume, but it is necessary to ensure that μ <P max , so as not to damage the sample.

[0136] Specific implementation of step S60:

[0137] Set a series of pressure gradients {P i}, these air pressure values ​​are all located near the optimal air pressure μ determined in step S50, the range is controlled within μ±10%μ, and the gradient interval is 2%μ.

[0138] These gradient pressures P are applied to the first pore water sample in sequence. i , and record the separation parameter data of each test as in step S30, including the separation time series {t j} and the amount of collected pore water {V(t j )}.

[0139] The purpose of this step is to provide sufficient experimental data for training the response surface model.

[0140] Specific implementation of step S70:

[0141] A series of test data {P i ,{t j},{V(t j )}}, the response surface model between air pressure P, separation time t and collected pore water volume V was trained by multivariate nonlinear regression:

[0142] V=f(P,t)

[0143] A typical implementation of the response surface model f is a cubic polynomial model:

[0144]

[0145] where β ij is the regression coefficient, which can be obtained by fitting using numerical optimization algorithms such as the least squares method.

[0146] The model can effectively fit the complex nonlinear relationship between the three. Given any air pressure P and separation time t, the trained response surface model can be used to predict the corresponding collected pore water volume V.

[0147] Specific implementation method of step S80:

[0148] The applied air pressure {P i} as multiple initial values ​​in the genetic algorithm (GA). Taking the response surface model f(P, t) established in step S70 as the objective function, two optimization objectives are set:

[0149] (1) Maximize the amount of pore water V collected;

[0150] (2) Minimize the separation time t.

[0151] Construct weighted and fitness functions:

[0152] Fitness=w 1 V 2 t

[0153] where w 1 、w 2 is the weight coefficient used to balance the two objectives.

[0154] The genetic algorithm is used to solve the multi-objective optimization of the fitness function. The main parameters of GA are set as: population size N pop =100, crossover probability p c =0.8, mutation probability p m= 0.2, terminate the evolutionary generation N gen =200.

[0155] Adaptive grid sampling strategy is used to generate the initial population to ensure uniform distribution of individuals in the population. Each generation generates a new generation of population through selection, crossover and mutation. Selection uses an estimated distribution algorithm based on theoretical evidence, crossover uses uniform arithmetic crossover, and mutation uses non-uniform mutation.

[0156] After several generations of evolution, the algorithm will eventually converge to a set of optimal air pressures P * and separation time t * The parameter combination balances the maximum collection amount and the minimum time overhead.

[0157] Specific implementation of step S90:

[0158] The optimal air pressure P determined in step S80 * and separation time t * The parameter combination is output and presented to the operator for applying air pressure P to the second pore water sample collected in the same area. * Separate pore water, separation time is t * .

[0159] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the above-mentioned method for determining the applied air pressure parameters of a pore water sampler.

[0160] A third aspect of the present invention provides a system for determining applied air pressure parameters of a pore water sampler, which includes the above-mentioned computer-readable storage medium.

[0161] Specifically, the principle of the present invention is: making full use of a variety of advanced technical means such as imaging, ultrasonic detection, and machine learning to more comprehensively describe the complex multi-scale pore structure characteristics of shale reservoirs and dynamically capture the damage behavior of shale under the action of gas pressure.

[0162] First, through high-resolution imaging technology and ultrasonic scanning technology, the present invention can accurately obtain a large amount of physical parameter information of shale reservoir samples, such as porosity, pore size distribution, permeability, etc., which lays a solid data foundation for subsequent modeling. At the same time, with the help of machine learning algorithms, especially image recognition algorithms, signal inversion algorithms, etc., the required characteristic parameters can be automatically extracted from the original data, avoiding the tedious manual analysis process.

[0163] Secondly, the present invention establishes a capillary pressure model and a sample destructive degree model.

[0164] The capillary pressure model can predict the maximum capillary pressure of the sample according to the porosity, permeability and other parameters of the sample. For different types of rocks, the present invention adopts different theoretical models, such as a model based on the coagulated salt solution theory for dense sandstone and a model based on the bound volume theory for shale. These models are obtained by training on a large amount of measured data through machine learning algorithms and can effectively describe complex multi-scale pore structures. The maximum capillary pressure calculated according to the model can be used as the lower limit of the applied gas pressure to ensure that the gas pressure is sufficient to displace the pore water.

[0165] The sample damage degree model is obtained by training a deep neural network on a large amount of real damage data. The model not only takes into account the instantaneous physical parameters of the sample, but also integrates features such as the initial surface image and ultrasonic signals, so it can dynamically track the real-time damage degree of the sample under the action of air pressure. When the damage degree exceeds the preset threshold, it means that the air pressure has exceeded the maximum allowable value of the sample, and it is necessary to stop the further increase of air pressure to protect the integrity of the sample. The combination of the two enables the present invention to avoid excessive damage to the sample while displacing the maximum amount of pore water.

[0166] Finally, the present invention uses response surface analysis and genetic algorithm optimization to further determine the optimal parameter combination of applied air pressure. The response surface model fits the complex relationship between air pressure, separation time and collected pore water volume through nonlinear regression, providing an objective function for the optimization process. Under this objective function, the genetic algorithm simultaneously optimizes the two conflicting goals of maximizing the amount of collected pore water and minimizing the separation time, and finally obtains a more ideal equilibrium solution, namely the optimal applied air pressure and optimal separation time parameters.

[0167] In general, the method of the present invention can solve the problem that the existing technology is difficult to deal with the complex pore structure of shale reservoirs. The key lies in the integration of multiple advanced technical means, which comprehensively describes the multi-scale pore characteristics of shale reservoirs and their damage behavior under gas pressure, so as to dynamically optimize the gas pressure parameters to achieve the goal of maximizing the displacement of pore water and avoiding excessive damage to samples. The application of machine learning algorithms in this method also greatly improves the efficiency of analysis and optimization and reduces the intensity of manual work.

[0168] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

[0169] A specific embodiment of the present invention is provided below: This embodiment takes a core sample collected from a continental shale gas reservoir as an object, and adopts the method for determining the applied gas pressure parameter of the present invention to perform pore water sampling.

[0170] 1. Sample Preparation

[0171] A shale core sample of about 30 cm in length and 6.5 cm in diameter was obtained from a coring section of a well in a shale gas reservoir. The core sample was cut into multiple small samples of about 3-5 cm in length, 10 of which were used for the test of the method of the present invention, and the other 10 small samples were used as a control group for the test using the existing technology. All samples were pre-treated by washing, drying, etc. according to standard procedures.

[0172] 2. Step S10-obtaining physical parameters

[0173] For the 10 small samples of the present invention group, surface images and ultrasonic reflection signal data were collected respectively. Image collection used a 12 million pixel electron microscope with a resolution of 4.2 μm / pixel; ultrasonic scanning used a phased array ultrasonic scanner with a resolution of 50 μm.

[0174] The deep convolutional neural network algorithm is used to analyze the image to obtain the porosity φ and pore diameter distribution f(r). At the same time, the reflection wave inversion algorithm is used to analyze the ultrasonic data to obtain parameters such as permeability k. The physical parameter measurement results of each small sample are shown in Table 1:

[0175] Table 1 Small sample physical parameter measurement results

[0176] Small sample number φ(%) f(r)(nm) k(nD) 1 3.21 20-500 0.032 2 5.17 50-800 0.084 3 2.36 10-300 0.016 4 4.28 30-600 0.065 5 1.95 5-200 0.009 6 6.34 100-1000 0.127 7 3.89 40-700 0.052 8 4.93 60-900 0.091 9 2.78 15-400 0.023 10 5.63 80-1100 0.103

[0177] 3. Step S20-Determine the initial air pressure

[0178] According to the shale lithology, the capillary pressure model based on bound volume theory is adopted. The model is obtained by training a large amount of shale measured data through Gaussian process regression algorithm, and its specific form is:

[0179]

[0180] Among them, V b and V f are the bound volume and free volume in shale, respectively, and are fitted experimentally as follows:

[0181] V b =0.35φ -1.21 k 0.47

[0182] V f =φ-V b

[0183] Substituting the φ and k data of each small sample into the above formula, the maximum capillary pressure P can be calculated. c Then take a certain safety factor (usually 1.2-1.5) to determine the initial applied air pressure P of each small sample.0 , the results are shown in Table 2:

[0184] Table 2 Comparison table of applied air pressure and maximum capillary pressure of each small sample

[0185]

[0186] 4. Steps S30-S60 Pore water separation test and data collection

[0187] Place the first small sample in a high-pressure container, apply an initial air pressure of 22 MPa, and start the pore water separation test. Record the separation time and the amount of pore water collected at 5-minute intervals, while continuing to collect surface images and ultrasonic data. Use the same algorithm as step S10 to dynamically obtain physical parameters φ(t) and k(t).

[0188] According to φ(t), k(t) and the initial image data, the damage degree D(t) is calculated using a deep feedforward neural network model. The model is an 8-layer fully connected neural network, trained on more than 4,000 real damage data, and can effectively predict the degree of damage. When D(t) exceeds the preset threshold of 0.85, it is determined that the air pressure exceeds the maximum allowable value and the air pressure increase is stopped immediately.

[0189] Plot the air pressure P versus the final collected pore water volume V f The relationship curve is fitted into a Gaussian distribution curve, and the peak value corresponds to the best gas pressure μ of 26MPa. f To obtain more data, seven nearby pressure gradients were set: 24, 25, 26, 27, 28, 29, and 30 MPa. The test was repeated to obtain a series of P, t, and V data for subsequent modeling.

[0190] Similarly, the above process was repeated for the remaining 9 small samples, and a series of experimental data for modeling were obtained.

[0191] 5. Step S70 - Response surface model training

[0192] Taking the experimental data obtained in step 4 as the training set, the least squares method is used to perform cubic polynomial nonlinear fitting on the air pressure P, separation time t and collected pore water volume V to obtain the response surface model:

[0193] V=f(P,t)=21.58+0.32P-9.76P 2 +5.18P 3 -1.75t+0.06Pt-0.34P 2 t+16.32t 2 -1.25Pt 2 -6.83t 3

[0194] The goodness of fit R of the response surface model 2 It is 0.936, which can accurately describe the nonlinear relationship between the three.

[0195] 6. Step S80 - Genetic Algorithm Optimization

[0196] Taking the response surface model as the objective function, the weighted and fitness functions are constructed:

[0197] Fitness=25V-0.5t

[0198] Adaptive grid sampling is used to generate the initial population, and the number of individuals is 100. The genetic algorithm parameters are set as follows: crossover probability 0.8, mutation probability 0.2, and termination evolution generation 200.

[0199] The fitness function is optimized and solved using the estimated distribution algorithm based on theoretical evidence. In the optimization process, the two goals of maximizing the amount of collected pore water V and minimizing the separation time t are taken into account at the same time.

[0200] After 178 generations of evolution, the algorithm converges to the optimal solution P * =27.3MPa,t * = 112min, the predicted collected pore water volume is V = 2.05ml. This parameter combination balances the collection volume and time cost.

[0201] 7. Step S90 - Applying optimal parameters

[0202] The above optimal applied air pressure P * =27.3MPa and optimal separation time t * = 112min is output to the operator for pore water sampling of other small samples (numbered 11-20) collected in the same area.

[0203] Under the same experimental conditions, the gas pressure of 27.3MPa was applied to samples 11-20 in turn, and the separation time was 112min.

[0204] 8. Comparative test of existing technology

[0205] In order to evaluate the effect of the method of the present invention, another 10 small shale samples (numbered 21-30) that had been pretreated were selected, and pore water sampling was carried out using the empirical formula and simple gradient pressure method commonly used in the prior art. The results were used for comparison with the method of the present invention.

[0206] (1) Empirical formula method

[0207] This method estimates the maximum capillary pressure P through the porosity φ c _max, the formula is:

[0208] P c _max=31.5φ -1.8 (MPa)

[0209] The applied air pressure P is usually P c _max is 1.2-1.5 times. For samples smaller than 21-30, the calculated P c _max and P are shown in Table 3:

[0210] Table 3 Comparison of applied air pressure and maximum capillary pressure of small samples 21-30

[0211] Small sample number φ(%) <![CDATA[P c _max(MPa)]]> P(MPa) 21 4.21 14.63 22 22 3.17 18.28 27 23 5.02 12.45 19 24 2.46 21.84 33 25 6.15 10.39 16 26 3.93 15.75 24 27 4.67 13.31 20 28 2.86 19.55 29 29 5.78 11.18 17 30 3.35 17.38 26

[0212] The sampling time is unified as 120 minutes.

[0213] (2) Gradient pressure method

[0214] This method first estimates an initial air pressure P 0 , and then gradually increase the air pressure until no more pore water can be collected.

[0215] For samples 21-30, first determine the initial pressure P according to the above empirical formula. 0 , and then increase the air pressure every 10 min at a gradient of 5 MPa until it reaches a maximum of 60 MPa or pore water can no longer be collected.

[0216] 9. Results comparison and analysis

[0217] The pore water sample volume and the damage of the shale small sample obtained by the above three methods are compared, and the results are shown in Table 4:

[0218] Table 4 Small sample damage and other comparison tables

[0219]

[0220] From Table 4 we can see that:

[0221] (1) The amount of pore water collected by the method of the present invention is significantly higher than that of the empirical formula method and the gradient pressure method, with average values ​​of 1.94 ml, 1.50 ml and 1.70 ml respectively. This is because the method of the present invention can more accurately predict the reasonable range of applied air pressure through machine learning algorithms, thereby displacing pore water to the maximum extent.

[0222] (2) The method of the present invention has the least degree of damage to shale samples, with an average damage degree of 0.67, while the average damage degrees of the empirical formula method and the gradient pressure method are as high as 0.82 and 0.85, respectively. The reason is that the method of the present invention establishes a dynamic damage degree model, which can monitor and control the gas pressure in real time so that it does not exceed the maximum allowable value.

[0223] (3) Although the collection volume of the gradient pressure method is higher than that of the empirical formula method, the damage degree is also significantly higher. This shows that simple gradient pressure is difficult to balance the collection volume and sample protection. The method of the present invention achieves a more ideal balance between maximizing the collection volume and minimizing the damage degree through genetic algorithm optimization.

[0224] In summary, the method of the present invention has significant technical progress compared with the existing technology, can collect pore water to the maximum extent and avoid excessive damage to shale samples, and has high application value.

Claims

1. A method for determining the applied air pressure parameters of a pore water sampler, characterized in that: The following steps are involved: S10, collecting surface images and ultrasonic reflection signal data of the first pore water sample, and obtaining physical parameters of the first pore water sample by using image analysis and reflection wave inversion technology; S20, combining the obtained physical parameters and according to a pre-fitted capillary pressure model, calculating the maximum capillary pressure of the first pore water sample as the initial air pressure value applied, wherein the initial air pressure value is greater than the maximum capillary pressure; S30, applying an air pressure of an initial air pressure value to the first pore water sample, performing a pore water separation test on the first pore water sample, and recording separation parameters, including separation time at specified intervals and the amount of pore water collected; S40, during the pore water separation test, gradually increase the air pressure value, and calculate the first pore water sample destructiveness in real time using the surface image and ultrasonic reflection signal data of the first pore water sample based on the pre-trained first pore water sample destructiveness model, and determine the maximum allowable air pressure value according to the pre-set maximum allowable value of the first pore water sample destructiveness; S50, plotting a relationship curve between air pressure and the final amount of pore water collected, and determining an optimal air pressure value for producing the maximum amount of pore water, wherein the optimal air pressure value does not exceed the maximum allowable air pressure value; S60, obtaining separation parameters of a series of pore water separation tests performed on the first pore water sample, wherein the applied gas pressure of the series of pore water separation tests is a sequential gradient gas pressure near an optimal gas pressure; S70, establishing a response surface model between air pressure, separation time, and collected pore water volume; S80, using the separation parameters of the series of pore water separation tests and their corresponding applied air pressures as multiple initial values, solving the response surface model using a genetic algorithm to obtain an optimal combination of air pressure and separation time parameters as an applied air pressure parameter; S90, outputting the applied air pressure parameter to an operator for applying air pressure to separate pore water in a second pore water sample collected from the same area as the first pore water sample.

2. The method for determining the applied air pressure parameters of a pore water sampler according to claim 1, characterized in that: The physical parameters include at least porosity and permeability.

3. The method for determining the applied air pressure parameter of a pore water sampler according to claim 1, characterized in that: The steps of training the capillary pressure model are specifically: Establishing a training data set, specifically obtaining multiple groups of soil or rock, collecting their physical parameters, and using a specific test method to obtain their maximum capillary pressure; A training model is provided, wherein the training data set is used to train a regression model using a Gaussian process regression algorithm to obtain the capillary pressure model.

4. The method for determining the applied air pressure parameter of a pore water sampler according to claim 1, characterized in that: The specified duration is 2 to 10 minutes.

5. The method for determining the applied air pressure parameter of a pore water sampler according to claim 1, characterized in that: The first pore water sample damage model adopts a deep feedforward neural network model, and the training steps are specifically: Establishing a first training set, specifically a plurality of groups of historical data, each group of historical data including the surface image and ultrasonic reflection signal data of the first pore water sample obtained in S10-S40, and the degree of destruction of the first pore water sample recorded each time in the pore water separation test; A training model is provided, wherein a deep feedforward neural network model is trained using the first training set to obtain the first pore water sample destructiveness model.

6. The method for determining the applied air pressure parameter of a pore water sampler according to claim 5, characterized in that: The steps of determining the degree of damage are specifically: Step 1, respectively extracting the features of the surface image and ultrasonic reflection signal data of the first pore water sample, and recording them as basic image features and basic ultrasonic features respectively; Step 2: Fusing the basic image features and basic ultrasound features to obtain fused features, which are recorded as basic fused features; Step 3, conduct a pore water separation test, gradually increase the air pressure value, and repeat steps 1 and 2 at specified intervals to obtain fusion features under multiple different air pressure values, which are recorded as pressure-adjusted fusion features; Step 4: Calculate the difference between each of the pressure-adjusting fusion features and the basic fusion feature, and use the difference as the degree of destruction of the first pore water sample measured at specified time intervals.

7. The method for determining the applied air pressure parameter of a pore water sampler according to claim 1, characterized in that: The response surface model adopts a multivariate nonlinear regression model.

8. The method for determining the applied air pressure parameter of a pore water sampler according to claim 1, characterized in that: The genetic algorithm takes maximizing the amount of collected pore water and minimizing the separation time as the optimization goal.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the method for determining the applied air pressure parameters of a pore water sampler according to any one of claims 1 to 8.

10. A system for determining applied air pressure parameters of a pore water sampler, characterized in that: A computer-readable storage medium comprising the computer-readable storage medium of claim 9.

Images