An intelligent prediction method for low permeability reservoir production after pressure reduction

Abstract

The present invention relates to an intelligent prediction method for the post-pressure production of low-permeability reservoirs. The method introduces a hybrid kernel function of a weighted combination of radial basis functions and polynomial functions, and adopts an improved mantis optimization algorithm combined with a hybrid kernel function support vector machine to construct an MDBO-SVM model. The main controlling factors affecting the production are used as input parameters, and the post-pressure production of the reservoir is used as output. The MDBO-SVM model is trained, and the trained MDBO-SVM model is used to predict the post-pressure production of the low-permeability reservoir. The method of the present invention can more accurately and efficiently predict the initial production capacity of horizontal wells, and the overall prediction effect of production is better. It also has higher prediction accuracy and stronger generalization ability for the prediction of production after pressure reduction of different reservoirs. When there are many reservoir features, the present invention can more accurately and reasonably screen out the main controlling factors affecting the post-pressure production of the reservoir by performing feature screening and combining multiple correlation analysis methods. The present invention effectively retains the distribution characteristics of the original data, thereby achieving effective expansion of the data.

Description

Technical Field

[0001] The present invention relates to the technical field of reservoir mining, and in particular to a method for intelligently predicting the post-pressure production of a low-permeability reservoir. Background Art

[0002] my country's low-permeability oil and gas reservoirs have considerable potential for resource extraction. The efficient development of low-permeability oil and gas reservoirs is inseparable from hydraulic fracturing. Predicting the production of reservoirs after hydraulic fracturing is crucial for formulating transformation measures. Traditional methods for predicting reservoir production after hydraulic fracturing mainly include constructing a comprehensive production prediction formula, numerical simulation, and nuclear magnetic resonance index methods. Existing rock property analysis, mercury injection data, and well logging data indicate that the reservoir in Block M is a low-permeability reservoir with poor pore-thickness sorting and strong heterogeneity. There is a complex correlation between data distribution and formation depth. Furthermore, in the oil exploration industry, original reservoir data often exhibit small sample characteristics, and the redundancy of reservoir geological characteristics will affect the production capacity prediction of key wells in Block M. This results in low prediction accuracy for traditional production prediction methods, making it difficult to accurately characterize the production of oil and gas wells. Summary of the Invention

[0003] In order to solve the above technical problems, the present invention provides a method for intelligently predicting the post-pressure production of low-permeability reservoirs.

[0004] The present invention is achieved through the following technical solutions:

[0005] A method for intelligently predicting the post-pressurization production of a low-permeability reservoir comprises the following steps: constructing a machine learning prediction model for predicting the post-pressurization production; training the machine learning prediction model with the main controlling factors affecting the production as input parameters and the post-pressurization production of the reservoir as output; and predicting the post-pressurization production of the low-permeability reservoir using the trained machine learning prediction model.

[0006] Preferably, the main controlling factors affecting the production include tensile strength, brittleness index, permeability, instantaneous pump stop pressure and sand addition intensity per meter.

[0007] Optionally, the method for determining the main controlling factor includes the following steps:

[0008] Acquiring an original data set, wherein the original data set includes reservoir geological parameters, engineering geological parameters, and engineering parameters;

[0009] Expand the original data set to obtain an expanded sample data set;

[0010] Extract and screen the characteristic parameters of the expanded sample data set to obtain key reservoir geological factors, key engineering geological factors, and key engineering parameters that affect production;

[0011] Based on the key reservoir geological factors, key engineering geological factors and key engineering parameters that affect production, the main controlling factors affecting production are further screened out.

[0012] Preferably, the improved Bayes-Bootstrap method is used to expand the original data.

[0013] Preferably, the key reservoir geological parameters affecting production are: permeability, pore structure index and reservoir quality factor; the key engineering geological parameters affecting production capacity are brittleness index, uniaxial compressive strength, tensile strength and horizontal stress difference; the key engineering parameters affecting production are pre-fluid ratio, sand addition intensity per meter and instantaneous pump stop pressure.

[0014] Optionally, the method for further screening the main controlling factors affecting yield includes the following steps:

[0015] Step 1: Using the MRMR correlation analysis method, the mutual information value I(x i ,x j ), determine the maximum correlation between each key reservoir geological parameter, engineering geological parameter, engineering parameter and target variable and the minimum redundancy between characteristics, and calculate the weight results of each key reservoir geological parameter, key engineering geological parameter and key engineering parameter;

[0016]

[0017] In the above formula, p(x,y) is the joint probability distribution function of x and y, and p(x) and p(y) are the marginal probability distribution functions of x and y;

[0018] Step 2: Use the Pearson correlation analysis method to calculate the linear correlation between key reservoir geological parameters, key engineering geological parameters, key engineering parameters and production:

[0019]

[0020] In the above formula, PCC(X,Y) represents the linear correlation coefficient between variable X and variable Y;

[0021] Step 3, calculating the correlation between key reservoir geological parameters, key engineering parameters, and key engineering parameters and production using a correlation analysis method based on the Copula function;

[0022]

[0023] In the above formula, COPULM(X,Y) represents the fused COPULA(X,Y) function. The COPULA(X,Y) function is a function that links the distribution functions of multiple random variables with their marginal distribution functions. It is used to describe the dependencies between multivariate random variables, including positive correlation, negative correlation, and no correlation. γ and λ represent the Kendall's rank correlation coefficient and Spearman's rank correlation coefficient between each key reservoir geological parameter, key engineering geological parameter, key engineering parameter, and production, respectively. The Kendall's rank correlation coefficient and Spearman's rank correlation coefficient are derived based on the COPULA(X,Y) function. The Kendall's rank correlation coefficient is used to measure the consistency of the rank changes of two variables, and the Spearman's rank correlation coefficient is used to measure the monotonicity between two variables. The calculated values of the Kendall's rank correlation coefficient and Spearman's rank correlation coefficient represented in the form of COPULA(X,Y) represent the degree of correlation between each key reservoir geological parameter, key engineering parameter, key engineering parameter, and production.

[0024] Step 4: Based on the MrMr correlation analysis method, the maximum correlation coefficient (MIC) is introduced to measure the correlation between variables:

[0025]

[0026] In the above formula, MIC(X,Y) represents the maximum correlation coefficient between variables X and Y, and I(X,Y) represents the mutual information between variables X and Y.

[0027] Step 5: Combine the MRMR correlation analysis method, the Pearson correlation analysis method, and the Copula function correlation analysis method to calculate the fusion correlation coefficient using the following formula:

[0028]

[0029] In the above formula, MPC(X,Y) represents the mean of the three correlation coefficient calculation results;

[0030] Step 6: By cross-validation, different combinations of key reservoir geological parameters, key engineering geological parameters, and key engineering parameters are selected each time, and the MDBO-SVM model is trained and validated. The effects of different parameter combinations on the determination coefficient R are observed. 2 The optimal number of reservoir geology, engineering geology, and engineering control factors corresponding to the machine learning prediction model is determined; the importance analysis results of key reservoir geology parameters, key engineering geology parameters, and key engineering parameters are compared to determine the main control factors affecting production.

[0031] Preferably, the machine learning prediction model is an MDBO-SVM model, and the method for constructing the MDBO-SVM model includes the following steps: introducing a hybrid kernel function of a weighted combination of radial basis functions and polynomial functions, and using an improved mantis optimization algorithm combined with a hybrid kernel function support vector machine to construct the MDBO-SVM model.

[0032] Optionally, the improved mantis optimization algorithm includes: using a population initialization method based on chaotic mapping Chebyshev to evenly distribute the initial population resources; using Levy flight control step length to make the mantis movement trajectory conform to the Levy distribution, ensuring that individuals can perform efficient local search while being able to jump out of local optimality and improve convergence accuracy; using Gaussian Cauchy mutation to ensure local search capability while expanding the group search range.

[0033] Compared with the prior art, this application has at least the following beneficial effects:

[0034] (1) The method of the present invention can more accurately and efficiently predict the production of reservoirs after compression, and the overall prediction effect of production is better; and it has higher prediction accuracy and stronger generalization ability for the prediction of production after compression of different reservoirs;

[0035] (2) By expanding the original data, the present invention can effectively obtain the neighborhood information of the original data when the data volume is insufficient, retain the upper and lower limits of the original sample while effectively expanding the upper and lower limits of the sample, and effectively retain the distribution characteristics of the original data, thereby achieving effective expansion of the data.

[0036] (3) When there are many reservoir characteristics, the present invention can more accurately and reasonably screen out the main controlling factors affecting the post-pressure production of the reservoir by performing characteristic screening and combining multiple correlation analysis methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0038] Figure 1 Flowchart for improving Bayes_Bootstrap in the embodiment;

[0039] Figure 2 A flowchart of generating an adversarial network in an embodiment;

[0040] Figure 3 It is an iteration diagram of the GAN discriminator loss function in the embodiment;

[0041] Figure 4 In the figure, a is the probability distribution histogram of tensile strength, b is the normal PP diagram of tensile strength; c is the probability distribution histogram of pre-fluid ratio, d is the normal PP diagram of pre-fluid ratio;

[0042] Figure 5 The cross-plot of porosity and permeability of the glutenite core in the Mahu area in the embodiment;

[0043] Figure 6 In the figure, a is the correlation diagram between the fracture pressure and the Mi production liquid index, b is the correlation diagram between the average sand ratio and the Mi production liquid index, and c is the correlation diagram between the displacement and the Mi production liquid index;

[0044] Figure 7 This is a diagram showing the importance ranking of each parameter in the embodiment;

[0045] Figure 8 Matrix diagram of Pearson correlation analysis coefficients between various factors and reservoir production after compression in the embodiment;

[0046] Figure 9 Matrix diagram of Spearman correlation coefficients between various factors and reservoir production after compression in the embodiment;

[0047] Figure 10 is a Kendall tau correlation coefficient matrix diagram of various factors and reservoir production after compression in the embodiment;

[0048] Figure 11 is the absolute coefficient R of the model for the number of main control factors in the embodiment 2 Schematic diagram of the impact;

[0049] Figure 12 Iterative curve diagram of DBO before and after improvement in the embodiment;

[0050] Figure 13 This is a comparison chart of the test set prediction results in the embodiment;

[0051] Figure 14 This is the test set Loss iteration graph in the embodiment;

[0052] Figure 15 This is a comparison chart of the prediction results in the embodiment;

[0053] Figure 16 2 is a relative error comparison diagram in the embodiment. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments.

[0055] It should be noted that, in the absence of any conflict, the embodiments and features of the embodiments of the present invention may be combined with each other. It should be noted that the various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to in conjunction with each other.

[0056] This embodiment discloses a method for intelligently predicting the post-pressure production of a low-permeability reservoir, comprising the following steps:

[0057] S1. Collect previous research results and relevant geological data of the study area, and conduct simultaneous tests of basic physical properties based on the prepared rock samples to obtain reservoir geological parameters such as porosity, permeability, mud content, and pore structure index. These parameters are recorded in the original data set for predicting post-fracture production.

[0058] It is worth mentioning that the simultaneous testing of basic physical properties includes porosity and permeability testing, nuclear magnetic resonance, rock electrical analysis, gamma ray, micro-CT analysis, high-pressure mercury injection, constant-pressure / constant-rate mercury injection testing, scanning electron microscopy, and conventional thin section analysis.

[0059] For the samples of uniaxial compression test, the full-diameter rock samples are processed into cylindrical rock samples with a diameter of 2.5 cm and a height of 5.0 cm. The maximum parallelism of the two end surfaces of the rock samples is less than 0.05 mm. For the samples used for tensile strength test, the rock samples are processed into cylinders with a diameter of 2.5 cm and a thickness of 0.5-1 times the diameter. The prepared rock samples are subjected to rock mechanics tests, such as standard plunger triaxial compression test, tensile strength test, fracture toughness test to obtain engineering geological parameters such as elastic modulus, brittleness index, Poisson's ratio, tensile strength, and compressive strength; and the experimental values of longitudinal wave, shear wave, acoustic wave time difference, density, etc. of each rock sample are analyzed and obtained in the rock mechanics test. Based on the well logging data, the effective stress method is used to calculate the effective stress according to the logging response of depth, acoustic wave, gamma, etc., as shown in formula (1):

[0060] σ e =0.00556DEPTH+0.09717AC+11.2367ln(GR)-64.7978(1)

[0061] Where, σ e is effective stress, MPa; DEPTH is formation depth, m; AC is acoustic transit time, μs / ft; GR is natural gamma, API.

[0062] Then, further considering the effective stress theory, the vertical stress value is introduced and the formation pressure of block M is calculated by formula (2). The spring combination model is used to calculate the reservoir stress profile of block M, as shown in formula (3), and the horizontal maximum principal stress σ can be calculated.H , horizontal minimum principal stress σ h .

[0063]

[0064] Where, P p is the formation pressure, MPa; σ V is the vertical stress, ρ(h) is the logging density at depth h, g / cm 3 .

[0065]

[0066] Among them, σ H is the maximum horizontal principal stress, MPa; σ h is the horizontal minimum principal stress, MPa; α is the biot coefficient; ε H represents the structural strain coefficient along the direction of maximum principal stress, ε h It represents the structural strain coefficient along the direction of minimum principal stress. The obtained reservoir geological parameters, engineering geological parameters, and engineering parameters are recorded in the original data set for predicting post-fracturing production.

[0067] Table 1 records the statistical information of the original data.

[0068] Table 1: Raw data statistics

[0069]

[0070] S2. Based on the fact that the original data sample size of the perforation section of the key well reservoir in Block M is very small, for any feature, let the total number of sample points be n, and renumber the data points in descending order, that is, X = (x1, x2, x3.....x n ). The improved Bayes_Bootstrap method is used to divide the neighborhood of the original perforation section depth point data of each feature. The parameter a value is adjusted to expand the data, as shown in formula (4), and the sample expansion multiple N is determined to obtain nN generated sample points:

[0071]

[0072] In the above formula, a is usually an integer not less than 2, x nk Generate sample points in each domain interval, merge the original sample points with the generated samples, and obtain the expanded samples of a certain feature.

[0073] like Figure 1 As shown, the improved Bayes_Bootstrap method includes the following steps:

[0074] Step 1: Import the native sample dataset;

[0075] Step 2: Determine the threshold of parameter a and expand the sample;

[0076] Step 3: Perform KS test and determine whether the KS test result is reasonable;

[0077] If the test result is reasonable, proceed to step 4, otherwise return to step 2;

[0078] Step 4, calculate the parameter point estimate and interval estimate;

[0079] Step 5: Determine whether the maximum sample expansion multiple has been reached; if so, proceed to step 6; otherwise, return to step 2;

[0080] Step 6: Calculate the MMD value, parameter point estimate, and interval estimate of the expanded sample data distribution.

[0081] like Figure 2 As shown in Figure 3, the generative adversarial network method (GAN) is applied to the perforation section data for data augmentation.

[0082] like Figure 3 As shown in Figure 2, the reservoir geological parameters and engineering geological parameters obtained from the experimental test are respectively expanded using the generative adversarial network method (GAN) and the improved Bayes-Bootstrap method. When using the GAN method to expand the data, this embodiment introduces the Wasserstein distance, that is, the minimum value of the average moving distance between the two distributions, to replace the JS divergence, thereby characterizing the similarity of the two distributions, as shown in formulas (5) to (7). Figure 3 The discriminator before and after the introduction of Wasserstein distance is difficult to measure the difference between the two distributions. It shows that for small sample data, the discriminator training effect is easy to be too good during the GAN training process, that is, the loss approaches 0, making it impossible for the generator to obtain effective information to improve its learning ability, that is, the ability to learn the distribution characteristics of the original perforation section data is poor.

[0083]

[0084] In the above formula, the expectation is used to represent the JS divergence of the two distributions P1 and P2, that is, the JS divergence is symmetric and is often used to measure the distribution difference between two probabilities. Rewriting formula (5) into a summation form, as shown in formula (6):

[0085]

[0086] In the above formula, it can be seen that when the two distributions do not overlap at all, that is, P1(x) = 0 or P2(x) = 0, the JS divergence value is always log2, that is, the loss value of the generator remains constant during the training process, and the gradient disappears.

[0087]

[0088] The above formula is the calculation process of the Wasserstein distance between two distributions. This calculation process is actually calculating an optimization problem, where γ∈Π(P1,P2) represents all the moving schemes when all values in one distribution are migrated to another distribution. The optimal migration plan for all values in one distribution to another is the plan that minimizes the relative distance between the two distributions. When using the Wasserstein distance to measure the difference between two distributions, completely aligning them before they are equal effectively prevents loss from remaining constant during training, mitigating the vanishing gradient phenomenon.

[0089] Table 2: Tensile strength distribution characteristics of expanded samples

[0090]

[0091] Table 3: Tensile strength distribution characteristics of expanded samples

[0092]

[0093] Table 4: Distribution characteristics of samples expanded by pre-fluid ratio

[0094]

[0095] Table 5: Distribution characteristics of samples with different pre-fluid ratios

[0096]

[0097] Tables 2-5 compare the distribution characteristics of the original samples and the new samples after expanding some reservoir characteristics using the two methods.

[0098] like Figure 4As shown in the figure, the probability distribution histogram and normal PP plot of the expanded data for some reservoir characteristics in Block M are consistent with the normality of the original data after the KS test. From the example of several reservoir characteristics, it can be seen that the difference between the expected value and 95% confidence interval length of the expanded sample using the improved Bayes-Bootstrap method and the corresponding indicators of the original sample is smaller than that of the Gan adversarial network method. The MMD value of the sample generated by the improved Bayes-Bootstrap method is smaller than the corresponding value of Gan. The width and kurtosis of the characteristic data generated by the improved Bayes-Bootstrap method show a small difference compared with the corresponding indicators of the original data. Combined with the above KS test plot, it can be concluded that the improved Bayes-Bootstrap method can better learn the distribution characteristics of the original sample. Therefore, the data output by the improved Bayes-Bootstrap method is used as the expanded sample data set. Feature extraction and key controlling factor screening of the various characteristic factors of the expanded sample data set are carried out to more accurately predict reservoir post-pressure production.

[0099] S3. Analyze the impact of reservoir geological parameters, engineering geological parameters and engineering parameters on the post-pressure production of low permeability reservoirs, and screen out the key reservoir geological parameters, key engineering geological parameters and key engineering parameters that affect production.

[0100] For example, matrix porosity affects the volume of saturated oil in the reservoir. Matrix permeability reflects the ability of oil and gas to flow through the reservoir. The ability to form a complex fracture network after fracturing is key to the efficient development of oil and gas wells, and the brittleness index is an important evaluation indicator for assessing the reservoir's ability to form a fracture network. The pore structure index characterizes the reservoir's permeability. The elastic modulus affects the compaction and stress distribution of the reservoir, altering the reservoir's pore structure and permeability, thereby affecting production. Changes in formation pressure can cause changes in local pressure, compressing or expanding the pore throat radius, and thus affecting oil well production. Relevant literature indicates that key engineering geological factors also affect the effectiveness of fracturing.

[0101] This example comprehensively analyzes the impact mechanism of reservoir geological factors on production and preliminarily screens out the main reservoir geological parameters affecting production, including permeability, porosity, pore structure index, mud content, and reservoir quality factor. The main engineering geological parameters include elastic modulus, Poisson's ratio, formation pressure, brittleness index, uniaxial compressive strength, tensile strength, and horizontal stress difference.

[0102] Analyze the influence of reservoir geological parameters on reservoir production. Considering that the formation pressure and density in the study area are relatively stable. This embodiment establishes a brittleness index prediction model based on elastic modulus. According to the basic geological characteristics of the study area, the permeability of the study block is mainly controlled by porosity, such as Figure 5As shown in the figure, there is a strong linear correlation between the porosity and permeability of the block, so the key reservoir geological parameters affecting production are extracted: permeability, pore structure index, and reservoir quality factor; key engineering geological parameters include brittleness index, uniaxial compressive strength, tensile strength, and horizontal stress difference.

[0103] Too little sand loading per meter can cause fractures to fail under closure pressure, while too much can lead to fracture blockage, reducing effective fracture permeability and directly impacting the average daily fluid production per well and ultimately recoverable reserves. The average sand ratio influences the proppant concentration within the fracture, which in turn affects fracture conductivity. Properly controlling the displacement rate can effectively maintain fracture width and stabilize the fracturing scale based on proppant volume. Setting an appropriate post-flowback deficit ratio ensures that more fracturing fluid is effectively utilized, improving fracturing effectiveness. The instantaneous pump-off pressure effectively reflects fracture growth during fracturing and the formation's response to the fracturing fluid. The fracture pressure gradient reflects the reservoir's fracturing effectiveness and fracture propagation. The pre-fluid ratio typically fractures the formation, creating fractures and carrying sand and fluid into the well, impacting the exploration efficiency of the oil and gas well. Based on the above analysis, the pre-fluid ratio, average sand ratio, sand loading per meter, displacement rate, instantaneous pump-off pressure, and fracture pressure are identified as the key engineering parameters influencing production.

[0104] The six selected engineering parameters were normalized and the relationship between the normalized influencing factors and the output was analyzed. Figure 6 It can be seen that the correlation between average sand ratio, displacement, fracture and production is extremely low (R 2 <0.12), indicating that these factors have no significant impact on post-pressurization production. Therefore, the key engineering parameters that affect production are screened out as follows: pre-fluid ratio, sand addition intensity per meter, and instantaneous pump stop pressure.

[0105] S4. Based on the key reservoir geological factors, key engineering geological factors and key engineering parameters that affect production, select the main controlling factors that affect production.

[0106] Optionally, the MRMR correlation analysis method is used based on the mutual information value I(x i ,x j ), as shown in formula (8); judge the maximum correlation between each key reservoir geological parameter, engineering geological parameter, engineering parameter and target variable and the minimum redundancy between characteristic parameters, and calculate the weight results of each key reservoir geological parameter, key engineering geological parameter and key engineering parameter as shown in Figure 7 .

[0107]

[0108] In the above formula, p(x,y) is the joint probability distribution function of x and y, and p(x) and p(y) are the marginal probability distribution functions of x and y.

[0109] PCCS is used to calculate the linear correlation between key reservoir geological parameters, key engineering geological parameters, key engineering parameters and production, as shown in formula (9). The COPULA (X, Y) function is a function that links the distribution functions of multiple random variables with their marginal distribution functions. It is mainly used to describe the dependence between multivariate random variables, including positive correlation, negative correlation, and no correlation. Kendall's rank correlation coefficient and Spearman's rank correlation coefficient are usually derived based on the COPULA (X, Y) function, and Kendall's rank correlation coefficient is mainly used to measure the consistency of the rank changes of two variables, and Spearman's rank correlation coefficient is mainly used to measure the monotonicity between two variables. Figures 8-10 In the equation, the absolute value of the correlation coefficient is taken, and the fusion correlation coefficient is defined by combining the three correlation analysis methods to calculate the comprehensive correlation coefficient. The correlation coefficient calculation formula is shown in formula (16):

[0110]

[0111] In the above formula, PCC(X,Y) represents the linear correlation between variables X and Y calculated using the Pearson correlation analysis method.

[0112] Assume (x1,x1), (x2,x2), x1,x2∈x, y1,y2∈y, are independent and identically distributed variables, and the Kendall's rank correlation coefficient γ is calculated as:

[0113] γ=2P[(x1-x2)(y1-y2)>0]-1 (10)

[0114] In the above formula, P represents the probability density function value.

[0115] If x0∈x, y0∈y, that is, x0 and y0 are independent of each other, then the Spearman's rank correlation coefficient λ is calculated as:

[0116] λ=3{P[(x-x0)(y-y0)>0]-P[(x-x0)(y-y0)<0]} (11)

[0117] In the above formula, P represents the probability density function value.

[0118] The range of Kendall's rank correlation coefficient and Spearman's rank correlation coefficient is [-1, 1], where -1 represents a completely negative correlation, 0 represents a completely uncorrelated correlation, and 1 represents a completely positive correlation. Equations (10) and (11) can be rewritten as COPULA functions, such as Equations (12) and (13):

[0119]

[0120] In the above formula, γ and λ represent Kendall's rank correlation coefficient and Spearman's rank correlation coefficient, respectively, and C(X,Y) represents the corresponding copula function.

[0121] Taking the mean of equations (12) and (13), we get the fused COPULA function, as shown in equation (14):

[0122]

[0123] In the above formula, COPULM(X,Y) represents the fusion COPULA(X,Y) function

[0124] This embodiment, based on the MrMr feature selection algorithm, introduces the maximum correlation coefficient MIC to more comprehensively measure the correlation between variables, as shown in formula (15):

[0125]

[0126] In the above formula, I(X,Y) represents the mutual information between variables X and Y.

[0127] Introduce the fusion correlation coefficient, as shown in formula (16):

[0128]

[0129] In the above formula, MPC(X,Y) represents the mean of the three correlation coefficient calculation results.

[0130] This embodiment predicts the post-pressure production by constructing a machine learning prediction model, using the main controlling factors affecting the production as input parameters and the post-pressure production of the reservoir as output. The machine learning prediction model is trained and the trained machine learning prediction model is used to predict the post-pressure production of low permeability reservoirs.

[0131] Too many or too few main control factors will affect the accuracy of model training. Through the cross-validation method, different key reservoir geological parameters, key engineering geological parameters, and key engineering parameter combinations are selected each time to train and verify the machine learning prediction model. By observing the effect of different parameter combinations on the determination coefficient R 2The optimal number of reservoir geology, engineering geology, and engineering control factors corresponding to the machine learning prediction model is determined by the influence of . In this implementation, the optimal number is 5, such as Figure 11 .

[0132] Comparing the importance analysis results of key reservoir geological parameters, key engineering geological parameters, and key engineering parameters, we determined that the reservoir geological parameter with the greatest impact on production is permeability. The engineering geological parameters with the greatest impact are tensile strength and brittleness index. The engineering parameters with the greatest impact are instantaneous pump stop pressure and sand addition density per meter. These five factors are determined to be the primary controlling factors affecting production.

[0133] Optionally, in some embodiments, this embodiment adopts an improved mantis optimization algorithm combined with a hybrid kernel function support vector machine to construct an MDBO hybrid kernel function SVM model (i.e., MDBO-SVM model) as a machine learning prediction model for predicting post-compression production.

[0134] Specifically, in order to improve the limitations of a single kernel function and enhance the generalization ability and accuracy of the model, this embodiment introduces a polynomial kernel function and a radial basis kernel function for weighted combination to form a hybrid kernel function, as shown in formula (17):

[0135] k mix =λk poly +(1-λ)k rbf (17)

[0136] In the above formula, k min represents the mixed kernel function, k poly represents the polynomial kernel function, k rbf Represents the radial basis kernel function, λ is the kernel function weight coefficient, and its value is between 0 and 1.

[0137] In order to solve the defects of the mantis optimization algorithm, such as low convergence accuracy and easy falling into local optimality, this embodiment introduces a population initialization method based on chaotic mapping Chebyshev to generate a series of random numerical values. These values are relatively evenly distributed in the search space, which is conducive to the uniform distribution of initial population resources, as shown in formula (18); and introduces the Levy flight method, which is a random walk method with a step size that conforms to the Levy distribution. This distribution allows the algorithm to make long-distance jumps during the search process, thereby helping individuals to conduct efficient local search while being able to jump out of the local optimal solution, as shown in formula (19); using Gaussian mutation, adding smaller disturbances that conform to the Gaussian distribution to the individual search space, so that individuals that are close to the optimal solution can obtain more accurate solutions in the search space; using Cauchy mutation, generating a larger variation based on the Cauchy distribution, so that individuals that fall into the local optimal solution can search in a larger range, which is conducive to jumping out of the local optimal solution, as shown in formula (20).

[0138] xn+1 =cos(karccosx n )(18)

[0139] In the above formula, x n represents the nth chaotic number, k is the number of iterations, and when k is greater than 2, the population initialization process is in a chaotic state.

[0140] x i ' +1 (t) = x i (t)+0.01(x i (t)-x b )×Levy(λ) (19)

[0141] In the above formula, x i (t) represents the i solutions of the t-th generation individual, x b is the current path, Levy(λ) is a random number generated by Levy distribution, which determines the direction and distance of the next step.

[0142]

[0143] In the above formula, x i (t) represents the i-th solution of the t-th generation individual; Cauchy(0,1) and Gauss(0,1) are Cauchy and Gaussian distributed random numbers respectively, β is the weight coefficient controlling the two distributions, and this scheme takes β = 0.5.

[0144] Specifically, the dataset was divided into a training set and a test set, and the MDBO-SVM model was trained. During training, the selected key controlling factors (tensile strength, brittleness index, permeability, instantaneous pump-off pressure, and sand addition intensity per meter) were used as input parameters, and the reservoir production after fracturing was used as the output parameter. The trained MDBO-SVM model was used to predict reservoir production capacity for each perforation section of key wells in Block M. The built-in parameters of the MDBO-SVM model are shown in Table 6.

[0145] Table 6: MDBO model built-in parameters

[0146]

[0147] Figure 12 This is the fitness iteration curve of the MDBO-SVM model and the improved DBO optimized hybrid kernel function SVM (i.e., DBO-SVM). Table 7 lists the average number of iterations and running time of the two methods when they are run 30 times independently.

[0148] Table 7: Run time table of different algorithms after 30 iterations

[0149] Average number of iterations Time / s MDBO-SVM 5.2 17.48 DBO-SVM 6.02 21.37

[0150] The comparison of the optimized parameters of the improved MDBO-SVM, DBO-SVM and SVM models is shown in Table 8.

[0151] Table 8: Comparison of optimization parameters of different algorithms

[0152] C σ λ MDBO-SVM 20.98 0.68 0.98 DBO-SVM 21.68 0.32 0.84 Support Vector Machine 8.5 0.8 0.87

[0153] Table 9: Comparison of evaluation indicators of various models

[0154] <![CDATA[R 2 ]]> MSE Mae MDBO-SVM 0.917 4.148 3.0421 DBO-SVM 0.882 6.755 3.9015 Support Vector Machine 0.842 7.623 4.0585 ANN 0.761 10.27 5.833

[0155] like Figure 15 As shown in FIG, the prediction results of the MDBO-SVM model are compared with those of the DBO-SVM, support vector machine (SVM), and artificial neural network (ANN). The results show that the MDBO-SVM model constructed by the method of this embodiment can more accurately and efficiently predict the initial production of horizontal wells.

[0156] like Figure 16 As shown in the figure, the trained MDBO-SVM model is used to predict production, and the results are compared with those of DBO-SVR, SVR and deep model ANN. The mean absolute error (MAE), root mean square error (RMSE) and determination coefficient (R 2 ) to evaluate the model performance, indicating that compared with several other methods, the MDBO-SVM model improves the overall prediction effect of production, indicating that the MDBO-SVM model has higher prediction accuracy and stronger generalization ability for production prediction after pressure reduction in different reservoirs.

[0157] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A method for intelligently predicting the post-pressure production of low permeability reservoirs, characterized in that: The following steps are involved: Build a machine learning prediction model for predicting post-compression yield; The machine learning prediction model is trained with the main controlling factors affecting production as input parameters and the post-pressure production of the reservoir as output, and the trained machine learning prediction model is used to predict the post-pressure production of the low permeability reservoir; The method for determining the main controlling factor comprises the following steps: Acquiring an original data set, wherein the original data set includes reservoir geological parameters, engineering geological parameters, and engineering parameters; Expand the original data set to obtain an expanded sample data set; Extract and screen the characteristic parameters of the expanded sample data set to obtain key reservoir geological factors, key engineering geological factors, and key engineering parameters that affect production; Based on the key reservoir geological factors, key engineering geological factors and key engineering parameters that affect production, further screen out the main controlling factors affecting production; The improved Bayes_Bootstrap method is used to expand the original data. The improved Bayes_Bootstrap method includes the following steps: Step 1: Import the native sample dataset; Step 2: Determine the threshold of parameter a and expand the sample; Step 3: Perform KS test and determine whether the KS test result is reasonable; If the test result is reasonable, proceed to step 4, otherwise return to step 2; Step 4, calculate the parameter point estimate and interval estimate; Step 5: Determine whether the maximum sample expansion multiple has been reached; if so, proceed to step 6; otherwise, return to step 2; Step 6, calculate the MMD value, parameter point estimate and interval estimate of the expanded sample data distribution; For the case where the original data sample size of the perforation section of the key well reservoir in a certain block is very small, for any feature, let the total number of sample points be n, and renumber the data points in descending order, that is, X = (x1, x2, x3.....x n ); the improved Bayes_Bootstrap method is used to divide the neighborhood of the original perforation depth point data of each characteristic, and the parameter a value is adjusted to expand the data. As shown in the following formula, the sample expansion multiple N is determined to obtain nN generated sample points: In the above formula, a is an integer not less than 2, x nk Generate sample points in each domain interval, merge the original sample points with the generated samples, and obtain the expanded samples of a certain feature.

2. The intelligent prediction method for post-pressure production of low permeability reservoirs according to claim 1, characterized in that: The main controlling factors affecting production include tensile strength, brittleness index, permeability, instantaneous pump stop pressure and sand addition intensity per meter.

3. The intelligent prediction method for post-pressure production of low permeability reservoirs according to claim 1, characterized in that: The key reservoir geological parameters that affect production are: permeability, pore structure index and reservoir quality factor; The key engineering geological parameters that affect productivity are brittleness index, uniaxial compressive strength, tensile strength and horizontal stress difference; The key engineering parameters affecting production are the front liquid ratio, sand addition intensity per meter and instantaneous pump stop pressure.

4. The intelligent prediction method for post-pressure production of low permeability reservoirs according to claim 1, characterized in that: The method for further screening the main controlling factors affecting yield includes the following steps: Step 1: Using the MRMR correlation analysis method, the mutual information value I(x i ,x j ), determine the maximum correlation between each key reservoir geological parameter, engineering geological parameter, engineering parameter and target variable and the minimum redundancy between characteristics, and calculate the weight results of each key reservoir geological parameter, key engineering geological parameter and key engineering parameter; In the above formula, p(x,y) is the joint probability distribution function of x and y, and p(x) and p(y) are the marginal probability distribution functions of x and y; Step 2: Use the Pearson correlation analysis method to calculate the linear correlation between key reservoir geological parameters, key engineering geological parameters, key engineering parameters and production: In the above formula, PCC(X,Y) represents the linear correlation coefficient between variable X and variable Y; Step 3: Calculate the correlation between key reservoir geological parameters, key engineering parameters, and key engineering parameters and production using a correlation analysis method based on the Copula function; In the above formula, COPULM(X,Y) is the fused COPULA(X,Y) function. The COPULA(X,Y) function is a function that links the distribution functions of multiple random variables with their marginal distribution functions. It is used to describe the dependency between multivariate random variables, including positive correlation, negative correlation, and no correlation. γ and λ represent the Kendall's rank correlation coefficient and Spearman's rank correlation coefficient between key reservoir geological parameters, key engineering geological parameters, key engineering parameters and production, respectively. Kendall's rank correlation coefficient and Spearman's rank correlation coefficient are derived based on the COPULA(X,Y) function. Kendall's rank correlation coefficient is used to measure the consistency of the rank changes of two variables, and Spearman's rank correlation coefficient is used to measure the monotonicity between two variables. Step 4: Based on the MrMr correlation analysis method, the maximum correlation coefficient (MIC) is introduced to measure the correlation between variables: In the above formula, MIC(X,Y) represents the maximum correlation coefficient between variables X and Y, I(X,Y) represents the mutual information between variables X and Y; min(|X|,|Y|) represents the minimum sample size of variables X and Y; Ib() is the grid binning function. MIC divides continuous variables into discrete grid intervals through the grid binning function. After binning, the distribution differences of variable Y under different X variable intervals can be clearly identified to evaluate the correlation and dependence between the two variables. Step 5: Combine the MRMR correlation analysis method, the Pearson correlation analysis method, and the Copula function correlation analysis method to calculate the fusion correlation coefficient using the following formula: In the above formula, MPC(X,Y) represents the mean of the three correlation coefficient calculation results; Step 6: Through the cross-validation method, different combinations of key reservoir geological parameters, key engineering geological parameters, and key engineering parameters are selected each time, and the MDBO-SVM model is trained and validated. The effects of different parameter combinations on the model determination coefficient R are observed. 2 The optimal number of reservoir geology, engineering geology, and engineering control factors corresponding to the machine learning prediction model is determined; the importance analysis results of key reservoir geology parameters, key engineering geology parameters, and key engineering parameters are compared to determine the main control factors affecting production.

5. A method for intelligently predicting the post-pressure production of a low permeability reservoir according to any one of claims 1, 2, and 4, characterized in that: The machine learning prediction model is an MDBO-SVM model. The method for constructing the MDBO-SVM model includes the following steps: introducing a hybrid kernel function of a weighted combination of a radial basis function and a polynomial function, and using an improved mantis optimization algorithm combined with a hybrid kernel function support vector machine to construct the MDBO-SVM model; The improved mantis optimization algorithm includes: using a population initialization method based on chaotic mapping Chebyshev to evenly distribute initial population resources; using Levy flight control step length to make the mantis motion trajectory conform to the Levy distribution; using Gaussian Cauchy mutation to ensure local search capability while expanding the group search range.

6. The method for intelligently predicting the post-pressure production of a low permeability reservoir according to claim 5, characterized in that: The calculation formula of the Gaussian Cauchy variation is: In the above formula, x i (t) represents the i-th solution of the t-th generation individual; Cauchy(0,1) and Gauss(0,1) are Cauchy and Gaussian distributed random numbers respectively; β is the weight coefficient controlling the two distributions, β = 0.

5.

7. The method for intelligently predicting the post-pressure production of a low permeability reservoir according to claim 5, characterized in that: The population initialization formula based on chaotic map Chebyshev is: x n+1 =cos(carccosx n ) In the above formula, x n represents the nth chaotic number, k is the number of iterations, and when k is greater than 2, the population initialization process is in a chaotic state.

Images