A method for predicting rock mechanical parameters in complex formations based on intelligent fusion strategy

Abstract

The present invention discloses a method for predicting rock mechanical parameters of complex formations based on an intelligent fusion strategy, including: collecting data; obtaining basic rock physical parameters; obtaining rock mechanical parameters; determining the relationship between the rock mechanical parameters of the study area and the rock longitudinal wave time difference, transverse wave time difference, and density; establishing a prediction model for rock mechanical parameters of complex formations using a machine learning algorithm; optimizing the base model of each rock mechanical parameter; based on the prediction results of the base model, constructing a fusion model of a series of intelligent prediction methods for rock mechanical parameters of complex formations based on an intelligent fusion strategy, calculating the weight coefficient of the base model using the Pearson correlation coefficient criterion, and finally predicting the rock mechanical parameters of complex formations through the fusion model. The present invention introduces the criterion of the Pearson correlation coefficient, and fuses multiple intelligent algorithms to improve the generalization of the prediction method, enhance the transfer learning ability of the intelligent algorithm, and improve the adaptability of the prediction method.

Description

Technical Field

[0001] The invention relates to a method for predicting rock mechanics parameters of complex formations based on an intelligent fusion strategy, and belongs to the field of oil and gas reservoir development. Background Art

[0002] Reservoir geomechanical parameters play a bridging role in geological-engineering integration, ensuring the seamless connection between geological knowledge and engineering technology and effectively solving engineering problems. They are key parameters in well site selection, wellbore trajectory optimization, and fracturing transformation. Among them, rock mechanics parameters are the core foundation of reservoir geomechanics parameters. Traditional evaluation methods for rock mechanics parameters mainly include prediction methods based on classical uniform isotropic elastic wave theory and a series of prediction methods based on the acoustic response mechanism of rock mechanics parameters. However, the obvious differences in complex rock mechanics parameters and the complex laws of rock physics response result in low accuracy of prediction results from traditional rock mechanics logging calculation methods, which brings great challenges to the evaluation of geomechanical parameters of complex formations. To this end, a prediction method for rock mechanics parameters of complex formations based on an intelligent fusion strategy is proposed. Summary of the invention

[0003] In order to overcome the problems in the prior art, the present invention provides a method for predicting rock mechanical parameters of complex formations based on an intelligent fusion strategy.

[0004] The present invention provides a technical solution to solve the above technical problems: a method for predicting rock mechanical parameters of complex formations based on an intelligent fusion strategy, comprising the following steps:

[0005] Step 1: Collect geological data, well logging data, and coring data in the study area;

[0006] Step 2: Collect underground or outcrop rock samples in the study area, conduct lithology description, and record the lithology of each rock sample;

[0007] Step 3: pre-process the processed rock samples and obtain basic rock physical parameters;

[0008] Step 4: Conduct mechanical tests to obtain the rock mechanical parameters of the study area;

[0009] Step 5: Based on the traditional evaluation method of rock mechanical parameters, determine the relationship between the rock mechanical parameters of the study area and the rock longitudinal wave time difference, shear wave time difference, and density;

[0010] Step 6: Comprehensively consider the indoor test data of rock lithology, rock density, P-wave time difference and S-wave time difference, and use machine learning algorithm to establish a prediction model of rock mechanical parameters in complex formations;

[0011] Step 7: Predict the correlation coefficient and average relative error of rock mechanical parameters based on the machine learning algorithm, and select the prediction model constructed by the machine learning algorithm with better prediction effect for each rock mechanical parameter as the base model for model fusion;

[0012] Step 8. Based on the prediction results of the base model, a fusion model of a series of intelligent prediction methods for rock mechanics parameters of complex formations is constructed based on the intelligent fusion strategy. The weight coefficient of the base model is calculated using the Pearson correlation coefficient criterion. Finally, the rock mechanics parameters of complex formations are predicted through the fusion model.

[0013] A further technical solution is that the geological data in step one includes logging lithology profiles and stratigraphic division data for model construction; the logging data includes natural gamma, compensated sonic, compensated density, and neutron logging data.

[0014] A further technical solution is that the specific process of step 2 is as follows: prepare 30 to 60 standard rock samples, wherein for uniaxial compression test samples, the full-diameter rock sample is processed into 30 to 60 cylindrical rock samples with a diameter of 2.5 cm and a height of 5.0 cm, and the maximum non-parallelism of the two end faces of the rock sample is less than 0.05 mm. In addition, the end face is perpendicular to the central axis of the sample, and the maximum deviation does not exceed 0.25°; for tensile strength test samples, the rock sample is processed into 30 to 60 cylinders with a diameter of 2.5 cm and a thickness of 0.5 to 1 times the diameter; for fracture toughness test samples, the rock sample is processed into 30 to 60 herringbone grooved Brazilian disc (CCNBD) samples with a diameter of 7.5 cm and a diameter to thickness ratio of 5:2.

[0015] A further technical solution is that the specific process of step three is: drying the rock sample at 60°C for 12 hours to eliminate the influence of moisture on the properties of the rock sample; measuring the geometric dimensions and weight of the sample to calculate the density of the sample; and synchronously obtaining the longitudinal wave time difference and transverse wave time difference of each rock sample during the rock mechanics test.

[0016] A further technical solution is that the rock mechanical parameters include: rock compressive strength, elastic modulus, Poisson's ratio, cohesion, internal friction angle, tensile strength, brittleness index and fracture toughness.

[0017] A further technical solution is that the specific process of step four is: according to the set experimental plan, uniaxial / triaxial compression test, tensile strength test, and fracture toughness test are carried out on rock samples prepared in different ways, and the original experimental data are processed to obtain the rock compressive strength, elastic modulus, Poisson's ratio, cohesion, internal friction angle, tensile strength, brittleness index and fracture toughness.

[0018] A further technical solution is that the machine learning algorithm includes BP neural network, XGBoost, support vector machine, random forest, convolutional neural network, decision tree, and long short-term memory neural network.

[0019] A further technical solution is that the calculation formula of the correlation coefficient is as follows:

[0020]

[0021] Where: y i is the measured value of rock mechanical parameters; is the predicted value of rock mechanical parameters; n is the number of tests.

[0022] A further technical solution is that the calculation formula of the Pearson correlation coefficient in step eight is:

[0023]

[0024] Where: ρ is the Pearson correlation coefficient; y i is the measured value of rock mechanical parameters; is the predicted value of rock mechanical parameters; n is the number of tests.

[0025] A further technical solution is that the calculation formula of the fusion model in step eight is:

[0026]

[0027]

[0028] Where: g is the fusion model; g i is the i-th intelligent prediction model; w i is the weight coefficient of the i-th basis model; ρ i is the Pearson correlation coefficient of the ith base model; N is the number of base models.

[0029] The present invention has the following beneficial effects: the present invention introduces the criterion of the Pearson correlation coefficient, integrates multiple intelligent algorithms to improve the generalization of the prediction method, enhances the transfer learning ability of the intelligent algorithm, and improves the adaptability of the prediction method. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 This is a diagram showing the predicted effect of rock mechanical parameters based on traditional evaluation methods;

[0031] Figure 2 The prediction effect of rock mechanical parameters based on multivariate nonlinear regression method;

[0032] Figure 3Correlation coefficient diagrams for rock mechanical parameters predicted by each model;

[0033] Figure 4 Plots of average relative errors of rock mechanical parameters predicted by each model;

[0034] Figure 5 This is the prediction effect diagram of rock tensile strength;

[0035] Figure 6 This is the predicted effect diagram of rock compressive strength;

[0036] Figure 7 This is the predicted effect diagram of rock cohesion;

[0037] Figure 8 This is the prediction effect diagram of rock brittleness index;

[0038] Fig. 9 This is the prediction effect diagram of rock fracture toughness. DETAILED DESCRIPTION

[0039] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0040] A method for predicting rock mechanical parameters of complex formations based on an intelligent fusion strategy of the present invention comprises the following steps:

[0041] Step 1: Collect geological data, well logging data, and coring data in the study area;

[0042] The geological data include logging lithology profiles, stratigraphic divisions and other data for model construction; the logging data include natural gamma, compensated sonic, compensated density, neutron and other conventional logging data;

[0043] Step 2: Collect underground or outcrop rock samples in the study area, conduct lithology description, and record the lithology of each rock sample;

[0044] Specifically, 30 to 60 standard rock samples are prepared, wherein for uniaxial compression test samples, the full-diameter rock samples are processed into 30 to 60 cylindrical rock samples with a diameter of 2.5 cm and a height of 5.0 cm, and the maximum non-parallelism of the two end faces of the rock sample is less than 0.05 mm. In addition, the end faces are perpendicular to the central axis of the sample, and the maximum deviation does not exceed 0.25°; for tensile strength test samples, the rock samples are processed into 30 to 60 cylinders with a diameter of 2.5 cm and a thickness of 0.5 to 1 times the diameter; for fracture toughness test samples, the rock samples are processed into 30 to 60 herringbone grooved Brazilian disc (CCNBD) samples with a diameter of 7.5 cm and a diameter-to-thickness ratio of 5:2;

[0045] Step 3: pre-process the processed rock samples and obtain basic rock physical parameters;

[0046] Specifically, the rock samples were dried at 60°C for 12 hours to eliminate the influence of moisture on the properties of the rock samples; the geometric dimensions and weight of the samples were measured to calculate the density of the samples; the longitudinal wave time difference and transverse wave time difference of each rock sample were synchronously obtained during the rock mechanics test;

[0047] Step 4: Conduct mechanical tests to obtain the rock mechanical parameters of the study area;

[0048] Specifically: according to the set experimental plan, uniaxial / triaxial compression test, tensile strength test and fracture toughness test are carried out on rock samples prepared by different sample preparation methods, and the original experimental data are processed to obtain rock mechanical parameters (rock compressive strength, elastic modulus, Poisson's ratio, cohesion, internal friction angle, tensile strength, brittleness index and fracture toughness);

[0049] Step 5: Based on the traditional evaluation method of rock mechanical parameters, determine the relationship between the rock mechanical parameters of the study area and the rock longitudinal wave time difference, shear wave time difference, and density;

[0050] σ c =0.6162e 0.5127(DTC / DEN)

[0051] Where: c is the uniaxial compressive strength, MPa; DEN is the rock density, g / cm 3 ; DTC is the longitudinal wave time difference, μs / m;

[0052] E s =-81.92ln(DTC / DEN)+396.43

[0053] Where: E s is the elastic modulus, GPa;

[0054]

[0055] Where: μ s is the static Poisson's ratio, dimensionless; μ d is the dynamic Poisson’s ratio, dimensionless;

[0056] σ t =-0.2368(DTC / DEN)+31.724

[0057] In the formula, σ t is the tensile strength, MPa;

[0058] The calculation results of a series of prediction methods for rock mechanical parameters based on traditional evaluation ideas are as follows: Figure 1 As shown in the figure, it can be seen that the prediction effect of each rock mechanical parameter is poor, and the average relative error is greater than 20%. At the same time, the prediction method of rock mechanical parameters in different complex formations was established using the least squares multivariate nonlinear regression method. The prediction effect is as follows Figure 2 As shown in the figure, it can be seen that the prediction effect of rock mechanical parameters of complex formations based on the multivariate nonlinear regression method is still not ideal;

[0059]

[0060] Where: y i is the measured value of rock mechanical parameters; is the average value of rock mechanical parameters; is the predicted value of rock mechanical parameters; n is the number of tests;

[0061] Step 6: Comprehensively consider the indoor test data of rock lithology, rock density, P-wave time difference and S-wave time difference, and use machine learning algorithms (BP neural network, XGBoost, support vector machine, random forest, convolutional neural network, decision tree, long short-term memory neural network) to establish a prediction model for rock mechanical parameters of complex formations;

[0062]

[0063] Where: y i is the measured value of rock mechanical parameters; is the predicted value of rock mechanical parameters; n is the number of tests;

[0064] The calculation results of rock mechanical parameters R 2 Correlation coefficients such as Figure 3 As shown in the figure, the average relative error is Figure 4As shown. It can be seen that the prediction effect of different machine learning algorithms on different rock mechanical parameters is significantly better than that of traditional evaluation methods, but there are obvious differences in the prediction effect of different machine learning algorithms on different rock mechanical parameters. The optimal machine learning algorithms for predicting different rock mechanical parameters are different, reflecting that different machine learning algorithms have different adaptability to different rock mechanical parameters and different prediction effects.

[0065] Step 7: Based on the correlation coefficient and average relative error of the rock mechanical parameters predicted by the machine learning algorithm, the prediction model constructed by the machine learning algorithm with better prediction effect for each rock mechanical parameter is selected as the base model for model fusion, wherein the base models shall not exceed 4;

[0066] Step 8. Based on the prediction results of the base model, the Pearson correlation coefficient of the machine learning algorithm is obtained using the calculation formula of the Pearson correlation coefficient (the results can be seen in Table 1). Based on the intelligent fusion strategy, a fusion model of a series of intelligent prediction methods for complex formation rock mechanics parameters is constructed. The weight coefficient of the base model is calculated using the Pearson correlation coefficient criterion. Finally, the fusion model is used to predict the complex formation rock mechanics parameters.

[0067]

[0068] Where: ρ is the Pearson correlation coefficient; y i is the measured value of rock mechanical parameters; is the predicted value of rock mechanical parameters; n is the number of tests.

[0069]

[0070]

[0071] Where: g is the fusion model; g i is the i-th intelligent prediction model; w i is the weight coefficient of the i-th basis model; ρ i is the Pearson correlation coefficient of the ith base model; N is the number of base models.

[0072] Table 1 Pearson correlation coefficients of various machine learning algorithms

[0073] Tensile Strength Compressive strength Elastic mold Poisson's ratio Cohesion Internal friction angle Brittleness Index Fracture toughness SVR Prediction 0.9538 0.8967 0.8889 0.7539 0.9140 0.8520 0.8856 0.9026 BP Forecast 0.6962 0.8594 0.8409 0.6732 0.9468 0.8918 0.7330 0.8929 LSTM prediction 0.9459 0.9506 0.9466 0.9434 0.9716 0.6655 0.9922 0.9570 CNN prediction 0.6837 0.9112 0.8972 0.4993 0.9155 0.7204 0.8594 0.9743 CART prediction 0.8270 0.8661 0.8208 0.4026 0.9785 0.8785 0.6410 0.7885 RF Prediction 0.8864 0.9305 0.9299 0.7743 0.9733 0.9582 0.9524 0.9082 XGBoost Prediction 0.9801 0.9519 0.9659 0.9493 0.9833 0.9863 0.9783 0.9942

[0074] Table 2 Weight coefficients of each base model

[0075] Tensile Strength Compressive strength Elastic mold Poisson's ratio Cohesion Internal friction angle Brittleness Index Fracture toughness SVR Prediction 0.3312 0.2417 / / / 0.2310 0.2325 0.2358 BP Forecast / / / 0.2624 0.2440 0.2418 / / LSTM prediction 0.3285 0.2562 0.2531 0.3677 0.2504 / 0.2605 0.2500 CNN prediction / 0.2456 0.2399 / / / / 0.2545 CART prediction / / / / 0.2522 / / / RF Prediction / / 0.2487 / / 0.2598 0.2501 / XGBoost Prediction 0.3403 0.2565 0.2583 0.3700 0.2534 0.2674 0.2569 0.2597

[0076] In the present invention, the constructed intelligent fusion prediction method is used to predict the rock mechanical parameters of complex formations. The prediction results are as follows: Figure 3 and Figure 4 As shown in Figure 2, the prediction effects of some rock mechanical parameters are as follows: Figure 5-Figure 9 As shown in the figure. It can be seen from the figure that the prediction effect of the intelligent fusion prediction method of various rock mechanical parameters is good, and the average relative error of various rock mechanical parameters is generally less than 10%. Among them, the prediction effect of the intelligent fusion prediction model of Poisson's ratio is the worst, with a maximum relative error of 9.86%, while the prediction effect of the intelligent fusion prediction model of cohesion is the best, with a minimum average relative error of 3.13%. Compared with the prediction results of a single model, the constructed series of intelligent fusion prediction methods improve the overall prediction effect of various rock mechanical parameters (uniaxial compressive strength, elastic modulus, Poisson's ratio, cohesion, internal friction angle, tensile strength, brittleness index and fracture toughness), have better prediction accuracy, and improve the adaptability of the method.

[0077] The above description is not intended to impose any form of limitation on the present invention. Although the present invention has been disclosed through the above embodiments, it is not intended to limit the present invention. Any technician familiar with the profession can make some changes or modifications to equivalent embodiments of equivalent changes using the technical contents disclosed above without departing from the scope of the technical solution of the present invention. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still falls within the scope of the technical solution of the present invention.

Claims

1. A method for predicting rock mechanical parameters of complex formations based on intelligent fusion strategy, characterized in that: The following steps are involved: Step 1: Collect geological data, well logging data, and coring data in the study area; Step 2: Collect underground or outcrop rock samples in the study area, conduct lithology description, and record the lithology of each rock sample; Step 3: pre-process the processed rock samples and obtain basic rock physical parameters; Step 4: Conduct mechanical tests to obtain the rock mechanical parameters of the study area; Step 5: Based on the traditional evaluation method of rock mechanical parameters, determine the relationship between the rock mechanical parameters of the study area and the rock longitudinal wave time difference, shear wave time difference, and density; Step 6: Comprehensively consider the indoor test data of rock lithology, rock density, P-wave time difference and S-wave time difference, and use machine learning algorithm to establish a prediction model of rock mechanical parameters in complex formations; Where: y i is the measured value of rock mechanical parameters; is the predicted value of rock mechanical parameters; n is the number of tests; Step 7: Based on the correlation coefficient and average relative error of the rock mechanical parameters predicted by the machine learning algorithm, the prediction model constructed by the machine learning algorithm with better prediction effect for each rock mechanical parameter is selected as the base model for model fusion; Step 8: Based on the prediction results of the base model, a fusion model of a series of intelligent prediction methods for rock mechanics parameters of complex formations is constructed based on the intelligent fusion strategy, the weight coefficient of the base model is calculated using the Pearson correlation coefficient criterion, and finally the rock mechanics parameters of complex formations are predicted through the fusion model; The calculation formula of Pearson correlation coefficient is: Where: ρ is the Pearson correlation coefficient; y i is the measured value of rock mechanical parameters; is the predicted value of rock mechanical parameters; n is the number of tests; The calculation formula of the fusion model is: Where: g is the fusion model; g i is the i-th intelligent prediction model; w i is the weight coefficient of the i-th basis model; ρ i is the Pearson correlation coefficient of the ith base model; N is the number of base models.

2. According to claim 1, a method for predicting rock mechanical parameters of complex formations based on intelligent fusion strategy is characterized in that: The geological data in step 1 include logging lithology profiles and stratigraphic division data, which are used for model construction; the logging data include natural gamma, compensated sonic, compensated density, and neutron logging data.

3. The method for predicting rock mechanical parameters of complex formations based on intelligent fusion strategy according to claim 1 is characterized in that: The specific process of step 2 is as follows: 30 to 60 standard rock samples are prepared, wherein for uniaxial compression test samples, the full-diameter rock samples are processed into 30 to 60 cylindrical rock samples with a diameter of 2.5 cm and a height of 5.0 cm, and the maximum non-parallelism of the two end faces of the rock sample is less than 0.05 mm. In addition, the end faces are perpendicular to the central axis of the sample, and the maximum deviation does not exceed 0.25°; for tensile strength test samples, the rock samples are processed into 30 to 60 cylinders with a diameter of 2.5 cm and a thickness of 0.5 to 1 times the diameter; for fracture toughness test samples, the rock samples are processed into 30 to 60 herringbone grooved Brazilian disc CCNBD samples with a diameter of 7.5 cm and a diameter-to-thickness ratio of 5:

2.

4. The method for predicting rock mechanical parameters of complex formations based on intelligent fusion strategy according to claim 1 is characterized in that: The specific process of step three is: drying the rock sample at 60°C for 12 hours to eliminate the influence of moisture on the properties of the rock sample; measuring the geometric dimensions and weight of the sample to calculate the density of the sample; and synchronously obtaining the longitudinal wave time difference and transverse wave time difference of each rock sample during the rock mechanics test.

5. The method for predicting rock mechanical parameters of complex formations based on intelligent fusion strategy according to claim 3 is characterized in that: The rock mechanics parameters include: rock compressive strength, elastic modulus, Poisson's ratio, cohesion, internal friction angle, tensile strength, brittleness index and fracture toughness.

6. The method for predicting rock mechanical parameters of complex formations based on intelligent fusion strategy according to claim 5 is characterized in that: The specific process of step four is: according to the set experimental plan, uniaxial / triaxial compression test, tensile strength test and fracture toughness test are carried out on rock samples prepared in different ways, and the original experimental data are processed to obtain the rock compressive strength, elastic modulus, Poisson's ratio, cohesion, internal friction angle, tensile strength, brittleness index and fracture toughness.

7. The method for predicting rock mechanical parameters of complex formations based on intelligent fusion strategy according to claim 1 is characterized in that: The machine learning algorithms include BP neural network, XGBoost, support vector machine, random forest, convolutional neural network, decision tree, and long short-term memory neural network.

Images