Method for predicting minimum miscible pressure of CO2 and crude oil by using machine learning

Abstract

The invention discloses a method for predicting the minimum miscible pressure of CO2 and crude oil by using machine learning, and the method comprises the steps: carrying out the high-precision modeling based on a linear SVM algorithm, building a correlation prediction formula, and constructing a correlation prediction model; and training and optimizing the correlation prediction model by using the collected data, and establishing a stacked integrated model in combination with a random forest regression model RF and the correlation prediction model to realize accurate prediction of the CO2-crude oil minimum miscible pressure MMP. According to the method, the time and the cost for determining the CO2-crude oil MMP value can be reduced to the maximum extent. Therefore, a reliable CO2-crude oil MMP value can be provided through prediction, and the miscible phase type of CO2 oil displacement is determined, so that the design of a CO2 oil displacement scheme is improved, the optimal design of an injection-production scheme is guided, and the crude oil recovery rate is increased.

Description

Technical Field

[0001] The present invention belongs to the technical field of reservoir engineering in oilfield development, and particularly relates to a method for predicting the minimum miscibility pressure of CO2 and crude oil by using machine learning. Background Art

[0002] In the process of oil extraction, for most low-permeability reservoirs, gas flooding is a commonly used method in tertiary oil recovery, and CO 2 flooding accounts for a relatively large proportion. CO 2 flooding technology can not only improve the oil recovery rate, but also realize the permanent storage of CO 2 in the earth's crust to mitigate the impact of the greenhouse effect. In recent years, with the continuous increase in the pressure of crude oil production, the CCUS technology has been paid more and more attention. Under the background of sequestering CO 2 to reduce the CO 2 level in the atmosphere, the development momentum of CO 2 flooding enhanced oil recovery technology is also getting better and better. Injecting CO 2 is one of the most practical and effective methods for improving the oil recovery rate (EOR) because it significantly reduces the viscosity of crude oil and improves the sweep efficiency, which are the key factors affecting EOR. This technology is not only applicable to conventional reservoirs, but also to low-permeability and extra-low-permeability reservoirs, and can effectively improve the oil recovery rate, with broad development and application prospects.

[0003] During the injection of CO 2 process, the minimum miscibility pressure MMP is used to determine the miscibility of crude oil and CO 2A very important parameter for achieving complete miscibility. Currently, the methods used to determine the MMP include the slim tube test method, the analytical model method, and the empirical correlation method. However, experimental measurements are both costly and time-consuming, and the mathematical models used may lead to significant estimation errors. The current mainstream four measurement methods are used to determine the MMP of specific fluids, and these are laboratory measurement techniques such as the counteracting interfacial tension experiment, the bubble rise experiment, the slim tube displacement experiment, and the mixing cell experiment (also known as the multiple contact experiment), compositional simulation, the mixing cell model, and the analytical method. The slim tube displacement experiment is the most reliable experimental method for determining the petroleum MMP because it is consistent with the definition of MMP, but its cost is usually expensive, time-consuming, requires representative reservoir fluid samples to conduct, may give inaccurate results according to the degree of physical dispersion present, and using the slim tube limits the amount of available data. While the other methods are subject to various limitations because they violate the inherent assumptions behind the MMP definition and are also affected by numerical dispersion. For several developed analytical correlation methods, some do not consider the petroleum composition, some consider the petroleum composition, and these correlations can lead to significant deviations, where the absolute error can be as high as 25%. The empirical correlation method can be based on reservoir conditions and fluid composition. Usually, the accuracy of the empirical correlation increases with the increase in the mathematical complexity of the equation. Most empirical correlations are mainly used for rapid screening applications. At the same time, methods developed using a limited experimental data set or a data set of MMP from a miscible flood, sometimes combined with inappropriate operating conditions, all bring great unreliability to calculating the MMP value. Summary of the Invention

[0004] The object of the present invention is to provide a method for predicting the minimum miscibility pressure of CO 2 with crude oil, which solves the problems of long time-consuming and large error in determining the minimum miscibility pressure during the injection of CO 2 process.

[0005] The technical solution adopted by the present invention is: a method for predicting the minimum miscibility pressure of CO 2 with crude oil, based on the linear SVM algorithm for high-precision modeling, establishing a correlation prediction formula, constructing a correlation prediction model; and using the collected data to train and optimize the correlation prediction model, combining the random forest regression model RF with the correlation prediction model to establish a stacked ensemble model, realizing the accurate prediction of the minimum miscibility pressure MMP of CO 2 - crude oil.

[0006] The characteristics of the present invention also lie in that

[0007] a method for predicting the minimum miscibility pressure of CO 2 with crude oil, the specific operation steps are as follows:

[0008] Step 1: First, collect the dataset including the oil composition, injection gas components, reservoir temperature, and the MMP value of crude oil obtained from the slim tube displacement experiment; 2 - The MMP value of crude oil;

[0009] Step 2: Use the linear SVM algorithm to efficiently optimize the linear SVM model, and establish a correlation prediction formula for CO 2 - The MMP of crude oil using the correlation coefficient in the optimized linear SVM model, thereby constructing a correlation prediction model;

[0010] The specific correlation prediction formula is as follows:

[0011]

[0012] Where:

[0013] A 0 ……A 17 is the correlation coefficient of the optimized linear SVM model;

[0014] MMP Eq.1 is the predicted minimum miscibility pressure;

[0015] MW C7 + is the molecular weight of the C 7 + component;

[0016] MW oil is the molecular weight of the crude oil;

[0017] x i is the mole percentage of component i in the injection gas or oil;

[0018] T is the temperature;

[0019] The method for obtaining the correlation coefficient is as follows:

[0020] Use the linear SVM with the maximum margin algorithm to ensure the existence of the global minimum through the optimization of the linear SVM model, and find the correlation coefficient from the optimized linear SVM model to establish the correlation formula prediction formula;

[0021] Step 3: Use the dataset collected in Step 1 to train and learn the correlation prediction model established in Step 2;

[0022] Step 4: Use the random forest regression integrated supervised machine learning method to train the correlation prediction model with different proportions of the training dataset, and compare the cross-contrast graphs and statistical parameters of the MMP predicted values and measured values before and after training and under different training datasets to determine the optimal model parameters;

[0023] The training dataset selects any one of 30%, 60%, and 100% of the dataset described in the entire Step 1.

[0024] Select datasets with different proportions as the training set to train the model. The higher the proportion of the training set, the higher the accuracy of the model after training. When building the stacking model below, the higher the accuracy of the prediction result obtained after inputting the prediction result of the correlation prediction formula.

[0025] Step 5: Combine the correlation prediction model and the random forest regression model trained in Step 4 to establish a stacking ensemble model;

[0026] The process of establishing the stacking ensemble model is as follows:

[0027] Use the optimal model parameters obtained in Step 4 as the input parameters of the random forest regression model, and the remaining input parameters and configurations are the same as those of the correlation prediction model

[0028] Step 6: Use the stacking ensemble model established in Step 5 to carry out minimum miscibility pressure prediction.

[0029] The beneficial effects of the present invention are:

[0030] (1) Different from the traditional slim tube displacement experiment that requires representative reservoir fluid sample experiments to be carried out, this method only needs to clarify the basic conditions of formation crude oil, injection gas components and reservoir temperature. At the same time, when establishing the model, various oils that meet the CO 2 oil displacement screening criteria are fully considered, and there will be no inherent bias towards specific types of oil components.

[0031] (2) The present invention uses machine learning to predict the minimum miscibility pressure of CO 2 with crude oil, which can minimize the time and cost of determining the CO 2 - crude oil MMP value. Therefore, this invention will provide a reliable CO 2 - crude oil MMP value through prediction, clarify the miscible type of CO 2 oil displacement, improve the design of the CO 2 oil displacement scheme, guide the optimization design of the injection-production scheme, and improve the crude oil recovery rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 is a flow chart of the method of the present invention;

[0033] Figure 2 is a cross-comparison chart of MMP predicted by the linear SVM model and the measured MMP;

[0034] Figure 3 is a cross-comparison chart of MMP predicted by the random forest regression model and the measured MMP (trained with 30% data);

[0035] Figure 4 It is a cross - comparison graph of the MMP predicted by the random forest regression model and the measured MMP (training with 60% of the data);

[0036] Figure 5 It is a cross - comparison graph of the MMP predicted by the random forest regression model and the measured MMP (training with 100% of the data);

[0037] Figure 6 It is a cross - comparison graph of the MMP predicted by the stacked ensemble model of the present invention and the measured MMP. Detailed implementation manners

[0038] In order to more fully explain the purpose and technical solution of the present invention, the present invention will be further described below in conjunction with specific embodiments.

[0039] Embodiment 1

[0040] The method for predicting the minimum miscibility pressure (MMP) of CO 2 with crude oil by the present invention is based on the linear SVM algorithm to perform high - precision modeling, establish a correlation prediction formula, and construct a correlation prediction model; and use the collected data to train and optimize the correlation prediction model, and combine the random forest regression model RF and the correlation prediction model to establish a stacked ensemble model to achieve accurate prediction of the minimum miscibility pressure (MMP) of CO 2 - crude oil.

[0041] Embodiment 2

[0042] The method for predicting the minimum miscibility pressure (MMP) of CO 2 with crude oil by using machine learning is as follows:

[0043] Step 1: First, collect data on oil composition, injection gas composition, reservoir temperature, and the MMP values of CO 2 - crude oil obtained from the slim - tube displacement experiment as a data set;

[0044] Step 2: Use the linear SVM algorithm to efficiently optimize the linear SVM model, and use the correlation coefficient in the optimized linear SVM model to establish a correlation prediction formula for CO 2 - crude oil MMP, thereby constructing a correlation prediction model;

[0045] Step 3: Use the data set collected in Step 1 to train and learn the correlation prediction model established in Step 2;

[0046] Step 4: Use the same data set as in Step 3, and use random forest regression integrated supervised machine learning to train the correlation prediction model to determine the optimal model parameters;

[0047] Step 5: Combine the correlation prediction model and the random forest regression model trained in Step 4 to establish a stacked ensemble model;

[0048] Step 6: Use the stacked ensemble model established in Step 5 to conduct minimum miscibility pressure prediction.

[0049] Example 3

[0050] The difference from Example 2 is that

[0051] The correlation prediction formula in Step 2 is specifically as follows:

[0052]

[0053] Where:

[0054] A 0 ……A 17 Is the correlation coefficient of the optimized linear SVM model;

[0055] MMP Eq.1 Is the predicted minimum miscibility pressure;

[0056] MW C7 + Is for C 7 + The molecular weight of the component;

[0057] MW oil Is the molecular weight of the crude oil;

[0058] x i Is the molar percentage of component i in the injected gas or oil;

[0059] T is the temperature.

[0060] The method for obtaining the correlation coefficient in Step 2 is as follows:

[0061] Use linear SVM with the maximum margin algorithm to ensure the existence of the global minimum through the optimization of the linear SVM model, and find the correlation coefficient from the optimized linear SVM model, thereby establishing a correlation formula prediction formula.

[0062] Step 4 is specifically as follows:

[0063] Use the random forest regression integrated supervised machine learning method to train the correlation prediction model with the training dataset in different proportions, compare the cross-contrast graphs and statistical parameters of the MMP predicted values and the measured values before and after training and under different training datasets, and clarify the optimal model parameters.

[0064] The training dataset in Step 4 selects any one of 30%, 60%, and 100% of the entire dataset described in Step 1.

[0065] Step 5: The process of establishing the stacked integration model is as follows:

[0066] Use the optimal model parameters in Step 4 as the input parameters of the random forest regression model. The remaining input parameters and configurations are the same as those of the correlation prediction model.

[0067] Example 4

[0068] The method for predicting the minimum miscibility pressure of CO 2 with crude oil by using machine learning is implemented according to the following steps:

[0069] Step 1: First, collect the data set of oil composition, injection gas composition, reservoir temperature, and the MMP value of CO 2 - crude oil obtained from the slim tube displacement experiment (as shown in Table 1);

[0070] Table 1 Basic parameters

[0071]

[0072] Step 2: Use the linear SVM algorithm to efficiently optimize the linear SVM model, and establish a correlation prediction formula for the MMP of CO 2 - crude oil by using the correlation coefficient in the optimized linear SVM model (as shown in Table 2), so as to construct a correlation prediction model. The correlation prediction formula is as follows:

[0073]

[0074] Where:

[0075] MMP Eq.1 is the predicted minimum miscibility pressure;

[0076] A 0 ... A 17 is the correlation coefficient of the optimized linear SVM model;

[0077] MW C7 + is the molecular weight of component C 7 + ;

[0078] MW oil is the molecular weight of crude oil;

[0079] x i is the mole percentage of component i in the injection gas or oil;

[0080] T is the temperature;

[0081] Table 2 Simulated MMP correlation coefficients (Formula 1)

[0082]

[0083] Step 3: By using the data set collected in Step 1, train and learn the correlation prediction model established in Step 2 (such as Figure 2 );

[0084] Step 4: Use the same data set as in Step 3 and train the data set in different proportions by using the "random forest regression" integrated supervised machine learning method (such as Figures 3 - 5 ). The higher the number of training sets, the more accurate the model. Compare the cross-comparison charts and statistical parameters of the MMP prediction results and the measured results before and after training and learning and under different training data sets to determine the optimal model parameters;

[0085] Step 5: Combine the correlation prediction model trained in Step 4 and the random forest regression model to establish a stacked ensemble model;

[0086] Step 6: Use the stacked ensemble model established in Step 5 to carry out minimum miscibility pressure prediction, and further compare the cross-comparison charts and statistical parameters of the prediction results and the measured results (such as Figure 6 ). It can be seen from the figure that the error between the minimum miscibility pressure predicted by the method of the present invention and the measured value is the smallest, and it is operable and executable.

[0087] The present invention combines a linear SVM model and a random forest regression algorithm to establish a stacked ensemble model to predict CO 2 -crude oil MMP. By comparing the statistical parameters of the prediction errors of different models and the cross-comparison charts, it is clear that the statistical parameters of the stacked ensemble model are significantly lower than those of the single prediction model, and each prediction point is also concentrated towards the measured value. It fully shows that the prediction errors of each model are continuously decreasing and the prediction accuracy is continuously improving. Therefore, the stacked ensemble model established by the present invention greatly improves the prediction accuracy of CO 2 -crude oil MMP, providing reliable parameters for the design of CO 2 flooding scheme.

[0088] Comparing the errors between the MMP prediction values and the measured values of the stacked ensemble model of the present invention, the linear SVM algorithm, and the random forest regression under three different prediction models, it can be clearly seen that the stacked ensemble model of the present invention has higher accuracy.

[0089] Table 3 Error comparison between MMP prediction values and measured values under different prediction models

[0090]

Claims

1. Method for predicting the minimum miscibility pressure of CO with crude oil using machine learning 2 ​ It is characterized in that Based on the linear SVM algorithm, high-precision modeling is carried out, a correlation prediction formula is established, and a correlation prediction model is constructed; and the collected data is used to train and optimize the correlation prediction model. Combining the random forest regression model RF and the correlation prediction model, a stacked ensemble model is established to achieve the prediction of the minimum miscibility pressure MMP of CO 2 - crude oil with high precision.

2. Method for predicting the minimum miscibility pressure of CO 2 with crude oil according to claim 1 It is characterized in that The specific operation steps are as follows: Step 1: First, collect the data set of oil composition, injection gas components, reservoir temperature, and the MMP value of crude oil obtained from the slim tube displacement experiment for CO 2 - as the data set; Step 2: Use the linear SVM algorithm to efficiently optimize the linear SVM model, and establish a CO 2 - crude oil MMP correlation prediction formula using the correlation coefficient in the optimized linear SVM model, thereby constructing a correlation prediction model; Step 3: By using the data set collected in Step 1, train and learn the correlation prediction model established in Step 2; Step 4: Use the same data set as in Step 3, and use the random forest regression integrated supervised machine learning method to train the correlation prediction model to determine the optimal model parameters; Step 5: Use the optimal model parameters obtained in Step 4 as the input parameters of the random forest regression model to construct a stacked integrated model; Step 6: Use the stacked integrated model established in Step 5 to carry out minimum miscibility pressure prediction.

3. The method for predicting the minimum miscibility pressure of CO 2 with crude oil according to claim 2 It is characterized in that The specific correlation prediction formula in Step 2 is as follows: Where: A 0 ……A 17 is the correlation coefficient of the optimized linear SVM model; MMP Eq.1 is the predicted minimum miscibility pressure; MW C7 + is C 7 + the molecular weight of the component; MW oil is the molecular weight of crude oil; x i is the mole percentage of component i in the injected gas or oil; T is the temperature.

4. The method for predicting the minimum miscibility pressure of CO 2 with crude oil according to claim 2 It is characterized in that The method for obtaining the correlation coefficient in Step 2 is as follows: Use linear SVM with the maximum margin algorithm, ensure the existence of the global minimum through the optimization of the linear SVM model, find the correlation coefficient from the optimized linear SVM model, and thus establish the correlation formula prediction formula.

5. The method for predicting the minimum miscibility pressure of CO 2 with crude oil according to claim 2 It is characterized in that Step 4 is specifically as follows: Use the random forest regression integrated supervised machine learning method to train the correlation prediction model with the training data set in different proportions, compare the cross-comparison charts and statistical parameters of the MMP predicted values and the measured values before and after training and under different training data sets, and determine the optimal model parameters.

6. The method for predicting the minimum miscibility pressure of CO 2 with crude oil according to claim 2 It is characterized in that The training data set in Step 4 selects any one of 30%, 60%, and 100% of the data set described in the whole Step 1.

7. The method for predicting the minimum miscibility pressure of CO 2 with crude oil according to claim 2 It is characterized in that The process of establishing the stacked integrated model in Step 5 is specifically as follows: Use the optimal model parameters obtained in Step 4 as the input parameters of the random forest regression model, and the other parameters and configurations are the same as those of the correlation prediction model, that is, the stacked integrated model is constructed.

Images