A method and system for constructing a proxy model for constant-rate mercury injection experiment in low-permeability reservoirs

Abstract

The present invention discloses a method and system for constructing a constant-rate mercury injection experiment proxy model for a low-permeability oil reservoir, and relates to the technical field of constant-rate mercury injection experiment simulation. The present invention constructs a meta-learning model based on an extreme learning machine (ELM) and a whale group optimization algorithm (WOA). During the training process, the encirclement and predation behavior of the WOA is utilized to simulate multiple hidden nodes in a hidden layer of the ELM as a population, and the fitness value of the hyperparameter configuration population is obtained according to the fitness function of the WOA, so as to update the parameter values ​​of each hyperparameter configuration of the constant-rate mercury injection experiment proxy model. In the process, the meta-learning model updates the parameter values ​​of each hyperparameter configuration of the constant-rate mercury injection experiment proxy model in a double-interactive form by learning the constant-rate mercury injection experiment data of a low-permeability oil reservoir with a small sample size. This construction process only requires a small sample size of data, and does not require pre-setting of parameters, so as to improve the prediction accuracy of the model, thereby constructing a constant-rate mercury injection experiment proxy model that meets the parameter characteristics of the constant-rate mercury injection experiment in a low-permeability oil reservoir.

Description

Technical Field

[0001] The present invention relates to the technical field of constant-rate mercury injection experiment simulation, and in particular to a method, system, equipment and medium for constructing a constant-rate mercury injection experiment proxy model for a low-permeability oil reservoir. Background Art

[0002] The pore structure parameters of low permeability reservoirs are crucial for reservoir evaluation and development. Due to their complex pore structure, small pore throat radius, strong capillary force and other pore parameters, it is difficult to identify oil and gas accumulation areas and develop in low permeability reservoirs. Traditional methods for obtaining permeability include numerical simulation and laboratory experiments, but for low permeability reservoirs, numerical simulation is not accurate enough in obtaining pore structure parameters. In addition, laboratory experiments require a lot of time and financial resources. Constant rate mercury injection experiment is one of the laboratory methods for measuring the pore structure of low permeability reservoirs. By injecting mercury at a constant pressure and recording the injected mercury volume and the corresponding capillary pressure, reservoir information about pore radius, throat radius, pore throat ratio, sortability, connectivity and permeability can be provided. However, the process of obtaining pore structure parameters of low permeability reservoirs through constant rate mercury injection experiments takes several days and requires expensive experimental equipment. Therefore, developing a fast and accurate proxy model to predict the pore structure parameters of low permeability reservoirs is crucial for optimizing development strategies and designing technical solutions, and has a significant impact on evaluating the production capacity and profitability of reservoirs.

[0003] In recent years, a large number of studies have shown that artificial intelligence is applicable in proxy models that process multiple data sources, including research in the fields of well logging and rock physics. The current proxy models for predicting pore structure parameters often use complex deep learning models based on specific assumptions and conditions such as specific pore shapes, distribution patterns, or fluid properties, and then use a large amount of training data for training. However, the complex geometric characteristics of the pore space in ultra-low permeability reservoirs often make the actual situation of ultra-low permeability reservoirs differ greatly from these assumptions and conditions, making the constructed proxy models less accurate when making predictions.

[0004] Therefore, it is difficult to construct a constant-rate mercury injection experimental proxy model that meets the constant-rate mercury injection experimental parameter characteristics of low permeability reservoirs using the current prediction proxy model approach. Summary of the invention

[0005] The embodiment of the present invention provides a method and system for constructing a constant-rate mercury injection experiment proxy model for a low-permeability oil reservoir, which can solve the problem in the prior art that it is difficult to construct a constant-rate mercury injection experiment proxy model that meets the constant-rate mercury injection experiment parameter characteristics of a low-permeability oil reservoir using the current prediction proxy model method.

[0006] The embodiment of the present invention provides a method for constructing a proxy model of a constant-rate mercury injection experiment in a low-permeability reservoir, comprising the following steps:

[0007] Carry out correlation analysis on constant rate mercury injection experimental data of low permeability reservoirs to obtain correlation data;

[0008] A constant-rate mercury injection experiment proxy model based on the extreme learning machine (ELM) was constructed, and the constant-rate mercury injection experiment proxy model was trained using correlation data. During the training process, the multiple hidden nodes in the hidden layer of the extreme learning machine (ELM) were simulated as a population using the siege-predation behavior of the whale swarm optimization algorithm (WOA). The fitness value of the hyperparameter configuration population was obtained according to the fitness function of the whale swarm optimization algorithm (WOA), so as to update the parameter values ​​of each hyperparameter configuration of the constant-rate mercury injection experiment proxy model.

[0009] By utilizing the shrinking and surrounding mechanism and spiral updating position of the whale optimization algorithm WOA, the parameter values ​​of each hyperparameter configuration of the constant-rate mercury injection experiment proxy model are updated again to obtain the constant-rate mercury injection experiment proxy model for low permeability reservoirs.

[0010] Preferably, the obtaining of correlation data includes:

[0011] The correlation analysis of the constant rate mercury injection test parameters of low permeability reservoirs is carried out. The parameters of the correlation analysis include: rock density RD, depth DEP, porosity POR, movable fluid percentage PMF, movable fluid porosity MFP and permeability PE;

[0012] The Pearson correlation coefficients of various parameters were calculated to obtain the Pearson correlation coefficient results. According to the Pearson correlation coefficient results, the samples with the absolute values ​​of the average Pearson coefficients of the throat radius, maximum throat radius, sorting coefficient and uniformity coefficient lower than 0.2 were removed.

[0013] Preferably, the step of simulating a plurality of hidden nodes in a hidden layer of an extreme learning machine ELM as a population comprises:

[0014] The Whale Swarm Optimization Algorithm (WOA) is used to generate a random population of extreme learning machine (ELM) hyperparameter configurations, including the number of hidden layer nodes, input weight range, and activation function selection, and a value is randomly generated for each hyperparameter.

[0015] The number of hidden layer nodes is an integer in the range of [10, 100]; the input weight range is a continuous value in the range of [0.01, 1.0]; the activation function is one of Sigmoid, ReLU or Tanh;

[0016] Assume the population size is P , each hyperparameter configuration contains d dimensions, the population matrix is ​​expressed as:

[0017] ;

[0018] in: Indicates i The individual in j Values ​​along the hyperparameter dimensions;

[0019] The specific method of population generation is: initialize the hyperparameter search space, each hyperparameter j The value range is set to [ LB j , UB j ],in LB j and UB j are the lower and upper bounds respectively; the initial value of each hyperparameter Generated by:

[0020] ;

[0021] in: r is a random number uniformly distributed in the interval [0,1]; i Indicates the first i individual; j Represents the dimension index of the hyperparameter;

[0022] For parameters that require discrete values, after generating a random number 𝑟, the value range of 𝑟 is divided into multiple intervals. The division method is expressed as:

[0023] r∈[0, 1) → interval mapping to {Sigmoid, ReLU, Tanh};

[0024] According to random number r The interval it falls into is assigned the corresponding discrete value.

[0025] Preferably, obtaining the fitness value of the hyperparameter configuration population includes:

[0026] For the constant rate mercury injection test proxy model, the fitness value is defined as the model error, and its equation is:

[0027] ;

[0028] in: X Represents the hyperparameter configuration; represents the true value; Representation model usage X Configure the predicted value; represents the average of the true values; n Indicates the number of individuals in a population;

[0029] For each hyperparameter configuration X in the populationi Repeat the above process to get the fitness value of all individuals , , , …, .

[0030] Preferably, the updating of the parameter values ​​of each hyperparameter configuration of the constant rate mercury injection experiment proxy model includes:

[0031] The population is initialized and the fitness value of each individual is calculated. The individuals perform prey encirclement or foraging behaviors, and the current optimal solution is updated until the number of iterations or the required accuracy is reached;

[0032] In order to satisfy the behavioral asynchrony between individuals’ encircling prey and foraging actions, a random probability Satisfy the following formula:

[0033] ;

[0034] Among them, the mathematical model of encirclement and hunting is:

[0035] ;

[0036] ;

[0037] ;

[0038] ;

[0039] in: Indicates t The best solution in the iteration; represents the randomly obtained reference whale position vector; represents the convergence vector whose magnitude decreases linearly from 2 to 0; , Respectively represent random vectors with a modulus ranging from 0 to 1;

[0040] when When When the individual chooses the direction of the best individual to move, it represents surrounding the prey. When , the individual chooses a random direction to move, which represents a global search for other feasible solutions.

[0041] Preferably, the step of re-updating the parameter values ​​of each hyperparameter configuration of the constant rate mercury injection experiment proxy model includes:

[0042] Using the spiral upward mechanism behavior trend of the whale optimization algorithm WOA, the parameter values ​​of each hyperparameter configuration of the constant rate mercury injection experiment proxy model are updated again. Its mathematical model is:

[0043] ;

[0044] in: Indicates t The best solution in iterations.

[0045] The embodiment of the present invention further provides a system for constructing a proxy model of a constant-rate mercury injection experiment in a low-permeability reservoir, comprising:

[0046] A data module is used to perform correlation analysis on the constant-rate mercury injection experimental data of low-permeability reservoirs and obtain correlation data;

[0047] The model parameter updating module is used to construct a constant-rate mercury injection experiment proxy model based on the extreme learning machine ELM, and use correlation data to train the constant-rate mercury injection experiment proxy model; during the training process, the siege predation behavior of the whale swarm optimization algorithm WOA is used to simulate multiple hidden nodes in the hidden layer of the extreme learning machine ELM as a population, and the fitness value of the hyperparameter configuration population is obtained according to the fitness function of the whale swarm optimization algorithm WOA, so as to update the parameter values ​​of each hyperparameter configuration of the constant-rate mercury injection experiment proxy model;

[0048] The module is determined, and the shrinkage and encirclement mechanism and spiral update position of the whale group optimization algorithm WOA are used to update the parameter values ​​of each hyperparameter configuration of the constant-rate mercury injection experiment proxy model again to obtain the constant-rate mercury injection experiment proxy model for low permeability oil reservoirs.

[0049] An embodiment of the present invention further provides an electronic device, including a memory and a processor;

[0050] The memory is used to store computer programs;

[0051] The processor is used to implement the steps of the method for constructing a proxy model of a constant-rate mercury injection experiment for a low-permeability reservoir as described above when executing the computer program stored in the memory.

[0052] An embodiment of the present invention further provides a computer-readable storage medium for storing a computer program, which, when executed by a processor, implements the steps of the method for constructing a proxy model for a constant-rate mercury injection experiment of a low-permeability reservoir as described above.

[0053] The embodiment of the present invention provides a method and system for constructing a proxy model for a constant-rate mercury injection experiment in a low-permeability reservoir. Compared with the prior art, the method and system have the following beneficial effects:

[0054] The present invention constructs a meta-learning model based on an extreme learning machine ELM and a whale swarm optimization algorithm WOA. During the training process, the encirclement and predation behavior of the whale swarm optimization algorithm WOA is used to simulate multiple hidden nodes in the hidden layer of the extreme learning machine ELM as a population, and the fitness value of the hyperparameter configuration population is obtained according to the fitness function of the whale swarm optimization algorithm WOA to update the parameter values ​​of each hyperparameter configuration of the constant-rate mercury injection experiment proxy model. In this process, the meta-learning model learns the constant-rate mercury injection experiment data of a low-permeability oil reservoir with a small sample size, based on the extreme learning machine ELM, and uses the whale swarm optimization algorithm WOA. The encirclement and predation behavior and the contraction and encirclement mechanism in OA continuously explore the solution space. At the same time, the contraction and encirclement mechanism and the spiral update position in the whale group optimization algorithm WOA are used to continuously iterate, and the parameter values ​​of each hyperparameter configuration of the constant-rate mercury injection experiment proxy model are updated in a double-layer form of double interaction. This construction process only requires a small sample of constant-rate mercury injection experiment data of low-permeability reservoirs, and there is no need to pre-set the pore shape, distribution pattern or fluid properties of the low-permeability reservoirs, thereby improving the accuracy of the constructed proxy model in prediction, so as to construct a constant-rate mercury injection experiment proxy model that meets the characteristics of the constant-rate mercury injection experiment parameters in low-permeability reservoirs.

[0055] In addition, the present invention can also realize self-optimization of machine learning parameters and adaptively adjust hyperparameters to adapt to different types of low permeability reservoir data, thereby solving the problem that the model is prone to overfitting learning for small sample size data. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 A schematic diagram of the overall process of a method for constructing a proxy model for a constant-rate mercury injection experiment in a low-permeability reservoir provided by an embodiment of the present invention;

[0057] Figure 2 A schematic diagram of the hyperparameter optimization process of the whale swarm optimization algorithm in a meta-learning model of a method for constructing a low-permeability reservoir constant-rate mercury injection experiment proxy model provided by an embodiment of the present invention;

[0058] Figure 3 A schematic diagram of the prediction results of the average throat radius of a method for constructing a proxy model of a constant-rate mercury injection experiment for a low-permeability reservoir provided by an embodiment of the present invention;

[0059] Figure 4 A schematic diagram of the prediction results of the maximum throat radius of a method for constructing a proxy model of a constant-rate mercury injection experiment for a low-permeability reservoir provided by an embodiment of the present invention;

[0060] Figure 5 A schematic diagram of prediction results of ranking coefficients of a method for constructing a proxy model for a constant-rate mercury injection experiment in a low-permeability reservoir provided by an embodiment of the present invention;

[0061] Figure 6A schematic diagram of the prediction results of the relative ranking coefficient of a method for constructing a proxy model of a constant-rate mercury injection experiment for a low-permeability reservoir provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0062] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention, so the present invention is not limited by the specific embodiments disclosed below.

[0063] See also Figure 1 The embodiment of the present invention provides a method for constructing a proxy model of a constant-rate mercury injection experiment in a low-permeability reservoir, specifically a proxy method for a constant-rate mercury injection experiment for a low-permeability reservoir. The method realizes the self-optimization of parameters of a machine learning model through an optimization algorithm, and can realize the proxy of a constant-rate mercury injection experiment model of a low-permeability reservoir by learning a small sample of constant-rate mercury injection experiment data of a low-permeability reservoir. Considering the high difficulty in obtaining pore structure parameters of a low-permeability reservoir and the high experimental cost of a constant-rate mercury injection experiment, the method can predict the constant-rate mercury injection experiment data of a low-permeability reservoir only through conventional experimental parameters, and has the characteristics of high timeliness, low cost, and the ability to proxy constant-rate mercury injection experiments in low-permeability reservoirs in batches, and is suitable for on-site measurement and analysis.

[0064] The specific steps include:

[0065] Step 1: Correlation analysis of constant rate mercury injection test parameters. Parameter correlation analysis includes: RD, rock density (g / cm³); DEP, depth (m); POR, porosity (%); PMF, movable fluid percentage (%); MFP, movable fluid porosity (%); PE, permeability (mD), etc. Among them, the Pearson correlation coefficient is calculated for each parameter to quantify the linear correlation strength.

[0066] Step 2: Based on the Pearson correlation coefficient results of the correlation analysis, samples with an average Pearson coefficient absolute value of less than 0.2 between the input features and the throat radius (μm), maximum throat radius (μm), sorting coefficient, and uniformity coefficient in the training set were taken as outliers for data cleaning.

[0067] Step 3: Standardize the data.

[0068] Step 4: Divide the data into training set and test set; randomly divide them in proportion, for example, 80% of the data is the training set and 20% is the test set.

[0069] Step 5: Use the whale optimization algorithm WOA to generate a random population of extreme learning machine ELM hyperparameter configurations, including the number of hidden layer nodes, input weight range, and activation function selection; randomly generate a value for each hyperparameter, and the range is specified by the user. Among them, the number of hidden layer nodes: integer, range is [10, 100]. Input weight range: continuous value, range [0.01,1.0]. Activation function: randomly selected from Sigmoid, ReLU or Tanh. Let the population size be P, each hyperparameter configuration contains D dimensions (that is, the number of hyperparameters that need to be optimized, such as the number of hidden layer nodes, weight range, etc.), and the population matrix can be expressed as:

[0070] .

[0071] in: Indicates i The individual in j The value of the hyperparameter dimension.

[0072] The specific generation method is: initialize the hyperparameter search space, each hyperparameter j The value range is set to [ LB j , UB j ],in LB j and UB j are the lower and upper bounds respectively; the initial value of each hyperparameter Generated by:

[0073] .

[0074] in: r is a random number uniformly distributed in the interval [0,1]. i Indicates the first i Individual. j Dimension index representing the hyperparameter.

[0075] For parameters that require discrete values ​​(such as activation function selection), after generating a random number 𝑟, the value range of 𝑟 is divided into several intervals as follows:

[0076] r ∈[0, 1) → the interval is mapped to {Sigmoid, ReLU, Tanh}.

[0077] Then, according to the random number r The interval it falls into is assigned the corresponding discrete value.

[0078] Step 6: Input the training set into the extreme learning machine ELM, and calculate the fitness value of the hyperparameter configuration population according to the fitness function of the whale swarm optimization algorithm WOA. The details are as follows:

[0079] For the constant rate mercury injection experiment proxy model, the fitness value is defined as the model error, as shown in the following formula:

[0080] .

[0081] in, X For hyperparameter configuration (such as the number of hidden layer nodes, weight range, etc.), is the true value, For model use X Configure the predicted value, is the average of the true values, n is the number of individuals in the population. For each hyperparameter configuration X in the population i Repeat the above process to get the fitness value of all individuals , , , …, .

[0082] Step 7: According to the encirclement and predation behavior of the whale group optimization algorithm WOA, update the parameter values ​​of each hyperparameter configuration in the population.

[0083] The specific WOA process is as follows:

[0084] ① Initialize the population and calculate the fitness value of each individual.

[0085] ② Individuals surround prey or forage for food.

[0086] ③Update the current optimal solution.

[0087] ④ Repeat ② and ③ until the number of iterations or the required accuracy is reached.

[0088] In order to satisfy the behavioral asynchrony between individuals’ encircling prey and foraging actions, a random probability Satisfy the following formula:

[0089] .

[0090] Among them, the mathematical model of encirclement hunting is as follows:

[0091] .

[0092] .

[0093] .

[0094] .

[0095] in: For the t The best solution in the iteration, is the randomly obtained reference whale position vector, is a convergence vector whose modulus decreases linearly from 2 to 0, , is a random vector with a modulus between 0 and 1. When When the individual chooses the direction of the best individual to move to represent the surrounding prey (close to the local optimal solution), When , the individual chooses a random direction to move, which represents a global search for other feasible solutions.

[0096] Step 8: Update the parameter values ​​of each hyperparameter configuration in the population according to the shrinking and encircling mechanism and spiral update position of the whale group optimization algorithm WOA. Specifically:

[0097] The behavior trend of the spiral upward mechanism is shown below, and its mathematical model is as follows:

[0098] .

[0099] Step 9: During the iteration process, the optimal hyperparameter configuration is output as a proxy model for the constant-rate mercury injection experiment. When the optimized extreme learning machine (ELM) model is used as a proxy for the constant-rate mercury injection experiment, the core task of the model is to learn the mapping relationship between experimental parameters based on the experimental history data, thereby replacing the actual experiment for rapid prediction. Specifically, the proxy model can be used to input parameters such as rock density (g / cm³), depth (m), porosity (%), and calculate the average throat radius (μm), maximum throat radius (μm) sorting coefficient, relative sorting coefficient, and uniformity coefficient.

[0100] Among them, the present invention utilizes the shrinking and encircling mechanism and spiral update position of the whale optimization algorithm (WOA), simulates multiple hidden nodes in the hidden layer of the extreme learning machine (ELM) as a population, and updates the parameter values ​​of each hyperparameter configuration of the constant-rate mercury injection experiment proxy model again to obtain the trained constant-rate mercury injection experiment proxy model. Through the random initialization weights and biases of the extreme learning machine (ELM) model, combined with the optimized hyperparameter configuration, the training obtains a proxy model that can efficiently predict the average throat radius, maximum throat radius, ranking coefficient, and relative ranking coefficient.

[0101] The present invention can predict pore structure parameters through a meta-learning model with only a small amount of sample data, which has the advantages of high timeliness and low cost; combining the WOA algorithm with the ELM model, the machine learning algorithm can adaptively adjust hyperparameters and has good generalization ability; this method has the advantages of high timeliness, low cost, and batch operation, which can further reduce the cost and improve the efficiency of low permeability oil reservoir development.

[0102] The evaluation method of the present invention is based on a meta-learning model to proxy the constant-rate mercury injection experiment of a low-permeability oil reservoir. It only needs a small sample of constant-rate mercury injection experiment data of a low-permeability oil reservoir, and predicts pore structure parameters through a meta-learning method, which has the advantages of high timeliness, low cost, and batch operation; the constant-rate mercury injection data is used to characterize the pore structure of the low-permeability reservoir, and a low-permeability oil reservoir pore structure parameter prediction method suitable for small sample amounts, low time cost, and batch processing is established through a meta-learning model. The constant-rate mercury injection experiment data of the low-permeability oil reservoir can be predicted only through conventional experimental parameters, providing guiding suggestions for the development of low-permeability oil and gas reservoirs; the WOA algorithm and the ELM machine learning model are combined, thereby realizing the ability of the machine learning algorithm to adaptively adjust the hyperparameter configuration according to the low-permeability oil reservoir, and can be applicable to various types of low-permeability oil reservoir pore structure data, and has the characteristics of strong generalization.

[0103] The present invention uses a meta-learning method and combines it with an optimization algorithm to construct a meta-learning-based proxy model, thereby realizing the self-optimization of machine learning parameters during the training process, and realizing the constant-rate mercury injection experiment proxy through a meta-learning proxy model suitable for a small sample size. This method uses constant-rate mercury injection data to characterize the pore structure of low-permeability reservoirs, and establishes a low-permeability reservoir pore structure parameter prediction method suitable for small sample sizes, low time cost, and batch processing through a meta-learning model, providing guiding suggestions for the development of low-permeability oil and gas reservoirs.

[0104] The present invention realizes the parameter self-optimization of the machine learning model through the optimization algorithm, and can realize the proxy of the constant-rate mercury injection test model of the low-permeability reservoir by learning the constant-rate mercury injection test data of the low-permeability reservoir with a small sample size. Considering the high difficulty of obtaining the pore structure parameters of the low-permeability reservoir and the high experimental cost of the constant-rate mercury injection test, the method can predict the constant-rate mercury injection test data of the low-permeability reservoir only through conventional experimental parameters, has the characteristics of high timeliness, low cost, and can be used as a proxy for the constant-rate mercury injection test of the low-permeability reservoir in batches, and is suitable for on-site measurement and analysis.

[0105] The present invention only requires a small sample of constant-rate mercury injection experimental data of low-permeability reservoirs, and predicts pore structure parameters through a meta-learning method, which has the advantages of high timeliness, low cost, and batch operation; in the model construction of the present invention, the WOA algorithm and the ELM machine learning model are combined, thereby realizing the ability of the machine learning algorithm to adaptively adjust the hyperparameter configuration according to the low-permeability reservoir, and can be applied to various types of low-permeability reservoir pore structure data, with the characteristics of strong generalization; the entire evaluation process does not require complex calculations, manual processing, or huge computing resources, and is suitable for promotion to oil field sites.

[0106] Specific experiments:

[0107] Step 1: Taking a low permeability oil reservoir as an example, a correlation analysis is performed on its constant-rate mercury injection test data. The parameters of the correlation analysis include: RD, rock density (g / cm³); DEP, depth (m); POR, porosity (%); PMF, movable fluid percentage (%); MFP, movable fluid porosity (%); PE, permeability (mD); ATR, average throat radius (μm); MTR, maximum throat radius (μm); SC, ranking coefficient; RSC, relative ranking coefficient; UC, uniformity coefficient; RD, DEP, POR, PMF, MFP, PE, UC are input parameters, and ATR, SC, RSC, UC are predicted parameters of the constant-rate mercury injection test.

[0108] Step 2: Based on the results of correlation analysis, the constant rate mercury injection test data is cleaned. Specifically, samples with an absolute value of the Pearson coefficient between the input feature and the target feature in the training set lower than 0.2 are considered as outliers and removed. Through this process, the quality and effectiveness of the model training data are ensured, while the negative impact of low-correlation samples on model performance is reduced.

[0109] Step 3: After data cleaning, standardize the data. By calculating the mean and standard deviation of each feature, convert it into a standard normal distribution with a mean of 0 and a standard deviation of 1, thereby eliminating the dimensional differences between different parameters and improving the efficiency and stability of model training.

[0110] Step 4: Divide the standardized data into training set and test set. Use random division to use 80% of the data as training set and the remaining 20% ​​as test set to evaluate the model.

[0111] Step 5: In the model construction stage, the Whale Swarm Optimization Algorithm (WOA) and the Extreme Learning Machine (ELM) are combined to establish a meta-learning agent model for the constant rate mercury injection experiment. First, a random ELM hyperparameter configuration population is generated through WOA, including the number of hidden layer nodes (integer, range [10, 100]), input weight range (continuous value, range [0.01, 1.0]), activation function, etc. In order to eliminate the impact of differences in WOA parameter settings on the experimental results, the swarm size of WOA is set to 20, 50, 100, and 200, respectively, and optimization experiments are carried out in parallel. When the population is initialized, the initial value of each hyperparameter is According to the formula Randomly generated, where r is a random number uniformly distributed in the interval [0, 1]. The optimization results are as follows Figure 2 As shown in the figure, the fitness value of the hyperparameter configuration population is calculated through the fitness function of WOA. After multiple iterations, the best individual in the last iteration is selected and translated back to the hyperparameter configuration.

[0112] Step 6: Translate the best individual in the last iteration back to the hyperparameter configuration and save the model.

[0113] Step 7: Substitute the test set into the optimized constant rate mercury injection experiment proxy model to predict the experimental results. The prediction results include key parameters such as ATR, MTR, SC, RSC, UC, and generate corresponding charts (such as Figure 3 , Figure 4 , Figure 5 and Figure 6 To verify the effectiveness of the model, a comparative experiment was conducted with the conventional ELM model, and the prediction error indicators were statistically analyzed (as shown in Table 1). The results show that the ELM model based on WOA optimization is significantly better than the conventional machine learning model in prediction accuracy, providing an efficient and reliable method for the rapid prediction of pore characteristics of low permeability reservoirs.

[0114] Table 1 Error comparison results of the proxy model of constant-rate mercury injection experiment in low permeability reservoirs

[0115]

[0116] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A method for constructing a proxy model for a constant-rate mercury injection experiment in a low-permeability reservoir, characterized in that: The following steps are involved: Carry out correlation analysis on constant rate mercury injection experimental data of low permeability reservoirs to obtain correlation data; A constant-rate mercury injection experiment proxy model based on the extreme learning machine (ELM) was constructed, and the constant-rate mercury injection experiment proxy model was trained using correlation data. During the training process, the multiple hidden nodes in the hidden layer of the extreme learning machine (ELM) were simulated as a population using the siege-predation behavior of the whale swarm optimization algorithm (WOA). The fitness value of the hyperparameter configuration population was obtained according to the fitness function of the whale swarm optimization algorithm (WOA), so as to update the parameter values ​​of each hyperparameter configuration of the constant-rate mercury injection experiment proxy model. By utilizing the shrinking and surrounding mechanism and spiral updating position of the whale optimization algorithm WOA, the parameter values ​​of each hyperparameter configuration of the constant-rate mercury injection experiment proxy model are updated again to obtain the constant-rate mercury injection experiment proxy model for low permeability reservoirs.

2. The method for constructing a proxy model for a constant-rate mercury injection experiment in a low-permeability reservoir according to claim 1, characterized in that: The obtaining of correlation data comprises: The correlation analysis of the constant rate mercury injection test parameters of low permeability reservoirs is carried out. The parameters of the correlation analysis include: rock density RD, depth DEP, porosity POR, movable fluid percentage PMF, movable fluid porosity MFP and permeability PE; The Pearson correlation coefficients of various parameters were calculated to obtain the Pearson correlation coefficient results. According to the Pearson correlation coefficient results, the samples with the absolute values ​​of the average Pearson coefficients of the throat radius, maximum throat radius, sorting coefficient and uniformity coefficient lower than 0.2 were removed.

3. The method for constructing a proxy model for a constant-rate mercury injection experiment in a low-permeability reservoir according to claim 1, characterized in that: The method of simulating multiple hidden nodes in the hidden layer of the extreme learning machine ELM as a population includes: The Whale Swarm Optimization Algorithm (WOA) is used to generate a random population of extreme learning machine (ELM) hyperparameter configurations, including the number of hidden layer nodes, input weight range, and activation function selection, and a value is randomly generated for each hyperparameter. The number of hidden layer nodes is an integer in the range of [10, 100]; the input weight range is a continuous value in the range of [0.01, 1.0]; the activation function is one of Sigmoid, ReLU or Tanh; Assume the population size is P , each hyperparameter configuration contains d dimensions, the population matrix is ​​expressed as: ； in: Indicates i The individual in j Values ​​along the hyperparameter dimensions; The specific method of population generation is: initialize the hyperparameter search space, each hyperparameter j The value range is set to [ LB j , UB j ],in LB j and UB j are the lower and upper bounds respectively; the initial value of each hyperparameter Generated by: ； in: r is a random number uniformly distributed in the interval [0,1]; i Indicates the first i individual; j Represents the dimension index of the hyperparameter; For parameters that require discrete values, after generating a random number 𝑟, the value range of 𝑟 is divided into multiple intervals. The division method is expressed as: r ∈[0, 1) → interval mapping to {Sigmoid, ReLU, Tanh}; According to random number r The interval it falls into is assigned the corresponding discrete value.

4. The method for constructing a proxy model for a constant-rate mercury injection experiment in a low-permeability reservoir according to claim 1, characterized in that: The step of obtaining the fitness value of the hyperparameter configuration population includes: For the constant rate mercury injection test proxy model, the fitness value is defined as the model error, and its equation is: ； in: X Represents the hyperparameter configuration; represents the true value; Representation model usage X Configure the predicted value; represents the average of the true values; n Indicates the number of individuals in a population; For each hyperparameter configuration X in the population i Repeat the above process to get the fitness value of all individuals , , , …, .

5. The method for constructing a proxy model for a constant-rate mercury injection experiment in a low-permeability reservoir according to claim 1, characterized in that: The parameter values ​​of each hyperparameter configuration of the constant rate mercury injection experiment proxy model are updated, including: The population is initialized and the fitness value of each individual is calculated. The individuals perform prey encirclement or foraging behaviors, and the current optimal solution is updated until the number of iterations or the required accuracy is reached; In order to satisfy the behavioral asynchrony between individuals’ encircling prey and foraging actions, a random probability Satisfy the following formula: ； Among them, the mathematical model of encirclement and hunting is: ； ； ； ； in: Indicates t The best solution in the iteration; represents the randomly obtained reference whale position vector; represents the convergence vector whose magnitude decreases linearly from 2 to 0; , Respectively represent random vectors with a modulus ranging from 0 to 1; when When When the individual chooses the direction of the best individual to move, it represents surrounding the prey. When , the individual chooses a random direction to move, which represents a global search for other feasible solutions.

6. The method for constructing a proxy model for a constant-rate mercury injection experiment in a low-permeability reservoir according to claim 5, characterized in that: The method of re-updating the parameter values ​​of each hyperparameter configuration of the constant rate mercury injection experiment proxy model includes: Using the spiral upward mechanism behavior trend of the whale optimization algorithm WOA, the parameter values ​​of each hyperparameter configuration of the constant rate mercury injection experiment proxy model are updated again. Its mathematical model is: ； in: Indicates t The best solution in iterations.

7. A system for constructing a proxy model for a constant-rate mercury injection experiment in a low-permeability reservoir, characterized in that: include: A data module is used to perform correlation analysis on the constant-rate mercury injection experimental data of low-permeability reservoirs and obtain correlation data; The model parameter updating module is used to construct a constant-rate mercury injection experiment proxy model based on the extreme learning machine ELM, and use correlation data to train the constant-rate mercury injection experiment proxy model; during the training process, the siege predation behavior of the whale swarm optimization algorithm WOA is used to simulate multiple hidden nodes in the hidden layer of the extreme learning machine ELM as a population, and the fitness value of the hyperparameter configuration population is obtained according to the fitness function of the whale swarm optimization algorithm WOA, so as to update the parameter values ​​of each hyperparameter configuration of the constant-rate mercury injection experiment proxy model; The module is determined, and the shrinkage and encirclement mechanism and spiral update position of the whale group optimization algorithm WOA are used to update the parameter values ​​of each hyperparameter configuration of the constant-rate mercury injection experiment proxy model again to obtain the constant-rate mercury injection experiment proxy model for low permeability oil reservoirs.

8. An electronic device, characterized in that: include: Memory and processor; The memory is used to store computer programs; The processor is used to implement the steps of the method for constructing a proxy model of a constant-rate mercury injection experiment for a low-permeability reservoir as described in any one of claims 1 to 6 when executing the computer program stored in the memory.

9. A computer-readable storage medium, characterized in that: Used to store a computer program, which, when executed by a processor, implements the steps of a method for constructing a low-permeability reservoir constant-rate mercury injection experiment proxy model as described in any one of claims 1 to 6.

Images