Method for predicting composition and production rate of supercritical multi-component thermal fluid for heavy oil thermal recovery
Through simulated reactions and machine learning algorithms, a model for predicting the composition and yield of supercritical multivariate thermal fluids was established, which solved the problem of lack of effective prediction methods in the existing technology, and improved the efficiency of development solution design and optimization.
Patent Information
- Application Number
- CN202411313593.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-20
- Publication Date
- 2025-06-20
- Estimated Expiration
- 2044-09-20
AI Technical Summary
The existing technology lacks effective means to analyze and predict the composition and yield of supercritical multivariate thermal fluids, resulting in the inability to effectively carry out the design and engineering parameter optimization of heavy oil injection thermal production development schemes.
By establishing multiple reaction molecular systems with different organic concentrations, setting reaction force fields and boundary conditions, using simulation platforms to perform simulation reactions, recording product molecules number and complete reaction time, calculating product composition and generation rate, and establishing a prediction model through machine learning algorithms.
Accurate prediction of the composition and yield of supercritical multivariate thermal fluids is achieved, the efficiency of development plan design and engineering parameter optimization is improved, and economic and time costs are reduced.
Smart Images
Figure CN119252359B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of heavy oil reservoir development, and particularly relates to a method for predicting the composition and production rate of supercritical multi-component thermal fluid for heavy oil thermal recovery. Background Art
[0002] Thermal fluid injection for production is an important development method for heavy oil, accounting for more than 70% of the total heavy oil development. The traditional heat carrier for heavy oil thermal injection is steam. In the past decade or so, composite heat carriers of steam mixed with non-condensable gas have been widely studied and applied due to their excellent oil displacement performance.
[0003] Supercritical multi-component thermal fluid is a new type of composite heat carrier proposed in recent years. It is composed of supercritical water, supercritical carbon dioxide, and supercritical nitrogen. Research shows that its oil displacement efficiency can be increased by 10% - 20% compared with traditional steam heat carriers or conventional composite heat carriers. At the same time, crude oil, oily sewage, oily sludge, etc. produced from oil wells can all be used as raw materials for generating supercritical multi-component thermal fluid, greatly saving raw material costs and waste treatment costs. During the process of injecting supercritical multi-component thermal fluid for heavy oil thermal recovery, a large amount of carbon dioxide will be injected into the formation and a part of it will be retained in the reservoir, synergistically achieving geological carbon sequestration. Due to its advantages in enhancing production efficiency, controlling costs, and green development, the development of injecting supercritical multi-component thermal fluid is widely regarded as a highly potential heavy oil development technology.
[0004] The reaction conditions (temperature, pressure, and the concentration of organic matter in the reactants) of supercritical multi-component thermal fluid will affect the composition and production rate of supercritical multi-component thermal fluid. The composition of the thermal fluid affects the physical properties of the thermal fluid, and the production rate of the thermal fluid affects the injection capacity of the thermal fluid generation end (supercritical multi-component thermal fluid generator) into the reservoir. Both the physical properties of the fluid and the injection capacity greatly affect the heavy oil thermal recovery development effect. Therefore, the composition and production rate of the thermal fluid are the key points for the design of heavy oil thermal recovery development plans by injecting supercritical multi-component thermal fluid and the optimization of engineering parameters. However, due to the lack of means to effectively analyze and predict the composition and production rate of supercritical multi-component thermal fluid based on reaction conditions, the design of development plans and the optimization of engineering parameters cannot be effectively carried out. Summary of the Invention
[0005] The purpose of the present invention is to provide a method for predicting the composition and production rate of supercritical multi-component thermal fluid for heavy oil thermal recovery, so as to solve the problem that in the prior art, due to the lack of means to effectively analyze and predict the composition and production rate of supercritical multi-component thermal fluid based on reaction conditions, the design of development plans and the optimization of engineering parameters cannot be effectively carried out.
[0006] To achieve the above object, the present invention provides a method for predicting the composition and production rate of supercritical multi-component thermal fluids for heavy oil thermal recovery. The method for predicting the composition and production rate of supercritical multi-component thermal fluids for heavy oil thermal recovery includes the following steps:
[0007] S1: Establish reaction molecular systems with different organic matter concentrations;
[0008] S2: Set the reaction force field and boundary conditions, determine the ensemble, and relax the reaction molecular systems to minimize the internal energy of the reaction molecular systems, so that the reaction molecular systems reach a stable state, and set up the simulation platform;
[0009] S3: Set a series of different temperatures and pressures, use the simulation platform to simulate the reaction of the reaction molecular systems with different organic matter concentrations, and record the number of product molecules and the complete reaction time of the reaction molecular systems with different organic matter concentrations at different temperatures and pressures;
[0010] S4: Calculate the product composition and the product formation rate according to the recorded number of product molecules and the complete reaction time;
[0011] S5: Perform data preprocessing on the organic matter concentration, temperature, pressure, product composition, and product formation rate;
[0012] S6: Divide the preprocessed data into a training group and a validation group. Based on the training group data, use a machine learning algorithm to establish a model for predicting the product composition and the product formation rate through the organic matter concentration, temperature, and pressure, and verify the accuracy and reliability of the model based on the validation group data.
[0013] Among them, in step S1, the reaction molecular systems are established based on the Materials Studio molecular construction / optimization and simulation platform, and the reaction molecular systems include organic molecules and water molecules.
[0014] Among them, in step S2, the set reaction force field is the ReaxFF reaction force field containing C / H / O atoms, and the set ensemble is the NPT constant temperature and constant pressure ensemble.
[0015] Among them, in step S2, the simulation platform is a large-scale atomic / molecular parallel simulation operation platform based on Lammps. The settings of the simulation platform include balancing charges by the QEQ method; minimizing the system energy by the conjugate gradient method; randomly generating the initial velocities of atoms according to the Maxwell-Boltzmann distribution at a set temperature; controlling the system temperature and pressure by the Nose-Hoover method based on the NPT ensemble, relaxing the system at a specific temperature and pressure to make the system reach an equilibrium state; canceling the ensemble setting and resetting the time step; simulating the reaction of supercritical multi-component thermal fluid under the NPT ensemble, with the reaction temperature and pressure being the simulated set temperature and pressure, and the temperature-pressure damping coefficients being 100 times and 1000 times the time step respectively and remaining unchanged during the simulation process; when each set time step is reached, outputting the system parameters and saving the product information.
[0016] Among them, in step S3, the set temperature and pressure are set with reference to the occurrence temperature and environment of supercritical multi-component thermal fluid in the oil field. The number of product molecules can be summarized and statistically analyzed through the built-in product statistics keyword in the Lammps software. The complete reaction time refers to the time when the organic reaction raw materials set before the reaction are completely converted into carbon dioxide through chemical reactions. The characteristic shown during the simulation process is that the number of product carbon dioxide molecules changes from an upward trend to stabilizing at a specific value. The time point corresponding to the inflection point of the trend change is the complete reaction time.
[0017] Among them, in step S4, the calculation method of the product composition is as follows: for the produced supercritical water, supercritical carbon dioxide, and supercritical nitrogen, based on Avogadro's constant, calculate the amount of substance of each component according to the number of molecules of each component in the system statistically described above, calculate the mass of each component according to the molar mass of each component, and calculate the mass fraction of each component in the system according to the mass of each component; the calculation method of the product generation rate is to divide the total mass of carbon dioxide generated by the reaction by the complete reaction time, that is, the generation rate of supercritical carbon dioxide per unit time during the reaction process.
[0018] Among them, in step S5, the data preprocessing methods include two data normalization methods: the Max-Min method and decimal point shifting.
[0019] Among them, in step S6, the machine learning algorithms adopted include the random forest algorithm and the particle swarm optimization algorithm. The indicators for evaluating the accuracy and reliability of the model include the mean square error MSE, the mean absolute error MAE, and the coefficient of determination R 2 。
[0020] A method for predicting the composition and yield of supercritical multi-component thermal fluids for heavy oil thermal recovery according to the present invention uses a simulation platform to simulate the reaction of reaction molecular systems with different organic matter concentrations. The experimental results obtained are approximate to the physical experimental results, indicating that using molecular simulation to study the composition and yield of supercritical multi-component thermal fluids is accurate and reliable. Moreover, the molecular simulation method has very low economic and time costs. By means of molecular simulation reactions, the composition and yield of supercritical multi-component thermal fluids under a large number of different reaction condition factors can be obtained, and through machine learning modeling, a non-linear relationship between the composition and yield and multiple reaction condition factors can be established, facilitating the study of the relationship between the composition and yield of supercritical multi-component thermal fluids and numerous experimental condition factors and establishing a prediction model. At the same time, due to the strong plasticity of molecular simulation and machine learning models, more factors can be added according to the research process and requirements to further improve the accuracy and wide applicability of the prediction model. Description of the Drawings
[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following-described drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.
[0022] Figure 1 It is a flowchart of the steps of the method for predicting the composition and yield of supercritical multi-component thermal fluids for heavy oil thermal recovery provided by the present invention.
[0023] Figure 2 It is a schematic diagram of the optimized diesel-water reaction system provided by the present invention.
[0024] Figure 3 It is a schematic diagram of the change of reaction products under a specific combination of reaction conditions during the simulation provided by the present invention.
[0025] Figure 4 It is a statistical schematic diagram of the product composition provided by the present invention.
[0026] Figure 5 It is a schematic diagram of the change in the number of carbon dioxide molecules under different reaction conditions (taking temperature as an example) provided by the present invention.
[0027] Figure 6 It is a schematic diagram of the distribution of the complete reaction time under different reaction conditions (taking temperature as an example) provided by the present invention.
[0028] Figure 7 It is a comparison chart of the molecular simulation reaction results and the actual physical test results under similar reaction conditions.
[0029] Figure 8 It is a schematic diagram of the predicted value and the simulated value of the supercritical water component content provided by the present invention.
[0030] Figure 9 It is a distribution diagram of the error between the predicted value and the simulated value of the supercritical water component content provided by the present invention.
[0031] Figure 10 It is a comparison diagram of the predicted value and the simulated value of the supercritical carbon dioxide component content provided by the present invention.
[0032] Figure 11 It is a distribution diagram of the error between the predicted value and the simulated value of the supercritical carbon dioxide component content provided by the present invention.
[0033] Figure 12 It is a comparison diagram of the predicted value and the simulated value of the yield provided by the present invention.
[0034] Figure 13 It is a distribution diagram of the error between the predicted value and the simulated value of the yield provided by the present invention. Detailed Embodiment
[0035] The embodiments of the present invention will be described in detail below. The examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the accompanying drawings are exemplary and are intended to explain the present invention, and should not be construed as limiting the present invention.
[0036] Please refer to Figure 1 , the present invention provides a method for predicting the composition and yield of supercritical multi-component thermal fluid for heavy oil thermal recovery. The method for predicting the composition and yield of supercritical multi-component thermal fluid for heavy oil thermal recovery includes the following steps:
[0037] S1: Establish a plurality of reaction molecular systems with different organic matter concentrations;
[0038] S2: Set the reaction force field and boundary conditions, determine the ensemble and relax the reaction molecular system to minimize the internal energy of the reaction molecular system, so that the reaction molecular system reaches a stable state, and set the simulation platform;
[0039] S3: Set a series of different temperatures and pressures, use the simulation platform to simulate the reaction of the reaction molecular system with different organic matter concentrations, and record the number of product molecules and the complete reaction time of the reaction molecular system with different organic matter concentrations at different temperatures and different pressures;
[0040] S4: Calculate the product composition and the product formation rate according to the recorded number of product molecules and the complete reaction time;
[0041] S5: Perform data preprocessing on the organic matter concentration, temperature, pressure, product composition, and product formation rate;
[0042] S6: Divide the preprocessed data into a training group and a validation group. Based on the training group data, use a machine learning algorithm to establish a model for predicting the product composition and product formation rate through the organic matter concentration, temperature, and pressure, and verify the accuracy and reliability of the model based on the validation group data.
[0043] In this embodiment, by using the simulation platform to perform simulation reactions on reaction molecular systems with different organic matter concentrations, the experimental results obtained are approximate to the physical experimental results, indicating that using molecular simulation to study the composition and yield of supercritical multi-component thermal fluids is accurate and reliable. Moreover, the molecular simulation method has very low economic and time costs. By means of molecular simulation reactions, the compositions and yields of supercritical multi-component thermal fluids under a large number of different reaction condition factors can be obtained, and through machine learning modeling, a non-linear relationship between the composition and yield and multiple reaction condition factors can be established, which is convenient for studying the relationship between the composition and yield of supercritical multi-component thermal fluids and numerous experimental condition factors, and establishing a prediction model. At the same time, due to the strong plasticity of molecular simulation and machine learning models, more factors can be added according to the research process and requirements to further improve the accuracy and wide applicability of the prediction model.
[0044] Among them, in step S1, the method for establishing the reaction molecular system is as follows:
[0045] Based on the Materials Studio atomic / molecular system construction, optimization, and simulation platform, using paraffin, naphthenes, monocyclic aromatics, and bicyclic aromatics as model compounds, construct a multi-component diesel molecular model, and then add water molecules. Use the Amorphous Cell module in the Materials Studio platform to construct a p-p-p boundary periodic simulation box, establish a reaction system of diesel molecules and water molecules, and perform geometric structure optimization. The modeling components of the reaction system are shown in Table 1, and the optimized reaction system is as Figure 2 shown, and establish multiple reaction molecular simulation systems with different organic matter mass concentrations, as shown in Table 2.
[0046] Table 1 Composition of diesel molecular model
[0047]
[0048] Table 2 Molecular model systems with different organic matter mass concentrations
[0049]
[0050] Among them, in step S2, the reaction force field is the ReaxFF reaction force field containing C / H / O atoms. The ensemble set is the NPT isothermal and isobaric ensemble. The simulation platform is based on the Lammps large-scale atomic / molecular parallel simulation operation platform. The simulation time step of the simulation platform is 0.1 fs. The settings of the simulation platform include the QEQ method to balance charges; the conjugate gradient method to minimize the system energy; randomly generate the initial velocities of atoms according to the Maxwell-Boltzmann distribution at 200 °C; based on the NPT ensemble, use the Nose-Hoover method to control the system temperature and pressure, relax the system at a specific temperature and pressure, and the temperature and pressure damping coefficients are 100 times and 1000 times of the simulation time step respectively, and the relaxation time is 10 ps to make the system reach an equilibrium state; cancel the ensemble setting and reset the time step; simulate the reaction of supercritical multi-component thermal fluid under the NPT ensemble, and the reaction temperature and pressure are the simulated set temperature and pressure, and the temperature-pressure damping coefficient remains unchanged; output parameters such as the system temperature, pressure, system volume, and atomic spatial positions every 100 time steps and save the product information.
[0051] Among them, in step S3, the set temperature and pressure are set with reference to the occurrence temperature and environment of supercritical multi-component thermal fluid in the mine. The number of product molecules can be summarized and counted through the built-in product statistics keyword in the Lammps software. The complete reaction time refers to the time when the organic reaction raw materials set before the reaction are completely converted into carbon dioxide through chemical reactions. The characteristic shown in the simulation process is that the number of product carbon dioxide molecules changes from an upward trend to a stable value at a certain specific value, and the time point corresponding to the inflection point of the trend change is the complete reaction time. Here, the change of the product during the simulation process is as Figure 3 shown.
[0052] Among them, in step S4, the calculation method of the product composition is as follows: for the produced supercritical water, supercritical carbon dioxide, and supercritical nitrogen, based on Avogadro's constant, calculate the amount of substance of each component according to the number of molecules of each component in the system statistically described above, calculate the mass of each component according to the molar mass of each component, and calculate the mass fraction of each component in the system according to the mass of each component; the calculation method of the product generation rate is to divide the total mass of carbon dioxide generated by the reaction by the complete reaction time, that is, the carbon dioxide generation rate per unit time during the reaction process. Here, taking the temperature variable as an example, Figure 4 shows the mass ratios of supercritical water, supercritical carbon dioxide, and supercritical nitrogen in the supercritical multi-component thermal fluid system under different temperature conditions, Figure 5 and Figure 6 shows the complete reaction time of the system under different temperature conditions.
[0053] Figure 7The comparison between the simulation results and the experimental results is shown. The density of the simulated reaction system is similar to that of the measured reaction system. At the same time, the mass concentration of the products generated by the simulated reaction under different reaction organic matter concentrations, different temperatures, and different pressures is also similar to that generated by the physical experiment, and the change trends are highly consistent. Therefore, it can be considered that the reactive molecular dynamics simulation can better reflect the actual situation of supercritical multi-component thermal fluids.
[0054] Among them, in step S5, to solve problems such as poor model fitting effect and slow training speed caused by inconsistent data dimensions, two data normalization methods, namely the Max-Min method and decimal point shifting, are used to normalize the eigenvalue (C, P, T) and target value data (H2O, CO2, v) respectively, so that the eigenvalues are distributed between 0 and 1, and the target values are distributed between 0 and 10. Then, 80% of the experimental data is randomly selected for model training and hyperparameter adjustment, and the remaining 20% of the data is used for comprehensive performance evaluation of the model. The Max-Min normalization formula is as follows:
[0055]
[0056] In the formula: x' is the normalized data; x is the original data; x max is the maximum value in the data; x min is the minimum value in the data.
[0057] Among them, in step S6, the machine learning algorithms used include the random forest algorithm and the particle swarm optimization algorithm. The indicators for evaluating the accuracy and reliability of the model include the mean square error MSE, the mean absolute error MAE, and the determination coefficient R 2 .
[0058] Among them, in step S6, based on the training set data, with the mean square error as the evaluation index, the 5-fold cross-validation method and the PSO intelligent optimization algorithm are used to optimize the four hyperparameters of the combined random forest model, namely n_estimators, max_depth, min_samples_split, and min_samples_leaf. When min_samples_split = 2, min_samples_leaf = 1, n_estimators are 78 and 34 respectively, and max_depth are 11 and 29 respectively, the combined random forest model performs best in validation. Subsequently, the random forest combined model is retrained using the best values of this set of hyperparameters and the training set data, and the comprehensive performance of the trained random forest combined model is comprehensively evaluated using the test set data and various evaluation indicators. The evaluation results are shown in Table 3 and Figure 5 as shown.
[0059] Table 3 Evaluation Table of the Prediction Performance of the Random Forest Combined Model
[0060]
[0061] Here, from the perspective of the comprehensive performance of the combined random forest model, the correlation between the predicted values and the simulated values of the model is relatively strong (R 2 > 0.9), and both the mean square error and the mean absolute error are maintained at a low level. From the comprehensive performance evaluation Figures 8 to 13 viewpoint, the predicted values and the simulated values of the combined random forest model on the training set and the test set are both concentrated on the unit diagonal line. At the same time, the errors between the predicted values and the simulated values of the combined random forest model for the supercritical water component content and the supercritical carbon dioxide component content on the training set and the test set are both less than 0.02, and most of the reaction yields are less than 0.05×10 -22 g / ps. Therefore, the combined random forest model has a good fitting degree with the vast majority of sample data, the accuracy of the prediction results is relatively high, and the model is reliable. In addition, since the sum of the three components of supercritical water, supercritical carbon dioxide, and supercritical nitrogen is 1, after obtaining the supercritical water component and the supercritical carbon dioxide component content, the supercritical nitrogen component can be obtained through simple calculation.
[0062] The above-disclosed is only a preferred embodiment of the present invention. Of course, it cannot be used to limit the scope of the rights of the present invention. Those of ordinary skill in the art can understand the entire or partial processes of implementing the above embodiments, and the equivalent changes made according to the claims of the present invention still fall within the scope covered by the invention.
Claims
1. A method for predicting the composition and yield of supercritical multi-component thermal fluid in heavy oil thermal recovery, characterized in that: The steps include: S1: Establish multiple reaction molecular systems with different organic matter concentrations; S2: Set the reaction force field and boundary conditions, determine the ensemble and relax the reaction molecular system to minimize the internal energy of the reaction molecular system, so that the reaction molecular system reaches a stable state, and set up the simulation program; S3: Set a series of different simulation temperatures and simulation pressures, use the simulation platform to simulate the reaction of reaction molecular systems with different organic concentrations, and record the number of product molecules and complete reaction time of reaction molecular systems with different organic concentrations at different temperatures and pressures; S4: Calculate the product composition and product generation rate based on the recorded number of product molecules and complete reaction time; S5: data preprocessing of organic matter concentration, temperature and pressure, product composition and product generation rate; S6: The preprocessed data were divided into a training group and a validation group. Based on the training group data, a machine learning algorithm was used to establish a model for predicting product composition and product generation rate through organic matter concentration, temperature and pressure. The accuracy and reliability of the model were verified based on the validation group data.
2. The method for predicting the composition and yield of supercritical multi-component thermal fluid for heavy oil thermal recovery according to claim 1, characterized in that: In step S1, the reaction molecular system is established based on the Materials Studio molecular construction / optimization and simulation platform, and the reaction molecular system includes organic molecules and water molecules.
3. The method for predicting the composition and yield of supercritical multi-component thermal fluid for thermal recovery of heavy oil according to claim 2, characterized in that: In step S2, the reaction force field is set to be a ReaxFF reaction force field containing C / H / O atoms, and the ensemble is set to be an NPT constant temperature and constant pressure ensemble.
4. The method for predicting the composition and yield of supercritical multi-component thermal fluid for heavy oil thermal recovery according to claim 3, characterized in that: In step S2, the simulation platform is based on the Lammps large-scale atomic / molecular parallel simulation computing platform, and the settings of the simulation program include the QEQ method to balance the charge; the conjugate gradient method to minimize the system energy; and the random generation of the initial atomic velocity according to the Maxwell-Boltzmann distribution at a set temperature. Based on the NPT ensemble, the Nose-Hoover method is used to control the system temperature and pressure, and the system is relaxed at a specific temperature and pressure to achieve equilibrium. Cancel the ensemble setting and reset the time step; simulate the reaction of supercritical multi-component thermal fluid under the NPT ensemble, the reaction temperature and pressure are the simulation setting temperature and pressure, the temperature-pressure damping coefficient is 100 times and 1000 times the time step respectively, and remains unchanged during the simulation process; each time the set time step is reached, the system parameters are output and the product information is saved.
5. The method for predicting the composition and yield of supercritical multi-component thermal fluid for heavy oil thermal recovery according to claim 4, characterized in that: In step S3, the set temperature and pressure are set with reference to the temperature and environment of supercritical multi-component thermal fluid generation in the mine. The number of product molecules can be summarized and counted by the product statistics keywords provided in the Lammps software. The complete reaction time refers to the time for the organic reaction raw materials set before the reaction to be completely converted into carbon dioxide through chemical reaction. The characteristic shown in the simulation process is that the number of molecules of the product carbon dioxide changes from an upward trend to a stable value at a certain value, and the time point corresponding to the inflection point of the trend change is the complete reaction time.
6. The method for predicting the composition and yield of supercritical multi-component thermal fluid for heavy oil thermal recovery according to claim 5, characterized in that: In step S4, the method for calculating the product composition is to calculate the amount of substance of each component of the produced supercritical water, supercritical carbon dioxide and supercritical nitrogen based on the Avogadro constant and the number of molecules of each component in the system as mentioned above, calculate the mass of each component according to the molar mass of each component, and calculate the mass fraction of each component in the system according to the mass of each component; the method for calculating the product generation rate is to divide the total mass of carbon dioxide generated by the reaction by the complete reaction time, that is, the carbon dioxide generation rate per unit time during the reaction.
7. The method for predicting the composition and yield of supercritical multi-component thermal fluid for heavy oil thermal recovery according to claim 6, characterized in that: In step S5, the data preprocessing methods include two data normalization methods: the Max-Min method and the decimal point shifting method.
8. The method for predicting the composition and yield of supercritical multi-component thermal fluid for heavy oil thermal recovery according to claim 7, characterized in that: In step S6, the machine learning algorithms used include random forest algorithm and particle swarm optimization algorithm, and the indicators for evaluating the accuracy and reliability of the model include mean square error MSE, mean absolute error MAE and determination coefficient R 2 .