A method and system for optimizing selection of petroleum molecular models based on sequential batch processing, and a medium

Abstract

This invention discloses a method, system, and medium for optimizing and selecting petroleum molecular models based on sequential batch processing, belonging to the technical field of constructing petroleum molecular models. The method includes selecting a first seed molecule from a candidate pool and adding it to a representative subset; selecting a combination of molecules from the remaining candidate molecules in the pool that contains at least a predetermined batch size; calculating the difference between the mixture formed after adding it to the representative subset and the target petroleum sample; selecting the molecular combination with the smallest difference and adding it to the representative subset; updating the current mixture attributes after each round of adding a molecular combination to the representative subset for use in the next round of molecular combination selection; repeating the molecular combination selection process until the number of molecules in the representative subset reaches a preset target value. This invention evaluates and selects candidate combinations containing multiple molecules in each round of optimization, rather than a single molecule, thereby introducing a larger search space in single-step decision-making and effectively reducing the probability of greedy algorithms getting trapped in local optima.

Description

Technical Field

[0001] This invention relates to the field of petroleum molecular modeling technology, specifically to a method, system, and medium for optimizing the selection of petroleum molecular models based on sequential batch processing. Background Technology

[0002] When constructing a petroleum molecular model, a candidate pool containing tens of thousands or even hundreds of thousands of molecules can be obtained through random generation or reconstruction. How to select a smaller subset of molecules (usually hundreds to thousands) from this huge candidate pool that can most accurately represent the overall properties of the original petroleum is an extremely challenging combinatorial optimization problem.

[0003] The existing methods have the following main shortcomings:

[0004] 1. Local optima traps in greedy algorithms: such as Figure 3 As shown, a traditional screening method employs a single-molecule greedy selection approach. Specifically, it calculates the difference between each candidate molecule added and the target sample, selects the molecule that minimizes the difference before and after addition, adds it, updates the mixture attributes, and checks if the target number of molecules has been reached. If not, it continues to calculate the difference between each remaining candidate molecule and the target sample, selecting the one that maximizes the reduction in difference. If the target number of molecules is reached, a locally optimal subset is obtained. This traditional method suffers from the problem of greedy algorithms easily falling into local optima traps. For example, the combination of molecules that best fit each attribute individually may not be the globally optimal solution.

[0005] 2. High computational complexity: Selecting M representatives from N candidate molecules is a combinatorial explosion problem. Exhaustive search is not feasible. Traditional optimization algorithms (such as genetic algorithms) need to repeatedly calculate the properties of the entire mixture when evaluating the fitness of each "population". When N is extremely large, the computational cost is high.

[0006] 3. Difficulty in balancing multi-objective and multi-scale attributes: The properties of petroleum encompass macroscopic attributes (such as density and molecular weight) and microscopic spectral attributes (such as NMR carbon type distribution), each with different dimensions, precision, and importance. Simply summing the mean squared errors can lead to high-precision but small-value spectral attributes being "submerged" by macroscopic attributes, or vice versa. Furthermore, incorporating molecular weight distribution (MWD), a crucial distributional attribute, into the optimization framework is also a challenge.

[0007] Therefore, how to provide a molecular optimization selection method that is computationally efficient, can effectively approximate the global optimum, and can intelligently balance multi-scale and multi-precision attribute objectives has become an urgent technical problem to be solved. Summary of the Invention

[0008] In view of this, the purpose of this invention is to provide a method, system, and medium for optimizing the selection of petroleum molecular models based on sequential batch processing, in order to solve the problem that existing greedy algorithms may have local optimal traps, which may lead to combinations that are not necessarily the global optimal solution.

[0009] To achieve the above objectives, this invention provides a method for optimizing the selection of petroleum molecular models based on sequential batch processing, the method comprising the following steps:

[0010] 1) Select a seed molecule from the candidate pool as the first seed molecule and add it to the representative subset;

[0011] 2) Select a combination of molecules that contains at least the set batch size from the remaining candidate molecules in the candidate pool. For each combination of molecules, calculate the difference between the mixture formed after adding it to the representative subset and the target petroleum sample, and select the combination of molecules with the smallest difference to add to the representative subset.

[0012] After each round of molecular combination is added to a representative subset, the current mixture property is updated, and the updated mixture property is used for the next round of molecular combination selection;

[0013] 3) Repeat step 2) until the number of molecules in the representative subset reaches the preset target value.

[0014] Its beneficial effects are as follows: In the process of screening molecules in the candidate pool, the method of the present invention does not select a single molecule greedily, but a batch greedy selection. That is, each evaluation selects not a single molecule, but a combination of molecules including a set number of molecules, which expands the single-step search space and avoids the situation where the combination of molecules with the best fit for each attribute may not be the global optimal solution. It solves the problem that the local optimal trap of existing greedy algorithms causes the combination to not be the global optimal solution, and enables the finding of representative molecule combinations that are closer to the global optimum.

[0015] Further, in step 1), selecting a seed molecule from the candidate pool includes: calculating the error value between the attribute vector of each individual molecule in the candidate pool and the target attribute vector, and selecting the molecule that minimizes the error value as the first seed molecule to be added to the representative subset.

[0016] This invention fully utilizes vectorized computation, performing parallel calculations on the properties of tens of thousands of molecules at once, avoiding inefficient serial computations in optimization loops. Combined with a batch processing strategy, it significantly reduces the number of iterations while ensuring optimization effectiveness, making it feasible to process large-scale candidate pools (>10,000 molecules).

[0017] Further, in step 2), selecting a molecular combination containing at least the set batch size from the remaining candidate molecules in the candidate pool includes: first, pre-screening a set number of molecules from the candidate molecules according to the set rules, and then enumerating all molecular combinations with a molecular number equal to the set batch size from all the selected molecules to form a molecular combination.

[0018] When selecting M representatives from N molecules, the method of this invention does not directly screen from all remaining molecules. Instead, it first selects a portion of molecules according to a set rule, and then iterates through this portion of molecules to obtain a combination of all molecules of a set number. Compared with directly iterating through N candidate molecules to select M representatives, the method of this invention significantly reduces computational complexity.

[0019] Further, in step 2), the setting rule is: the candidate molecule is the molecule with the highest complementarity to the current mixing property of the previously set number of molecules.

[0020] The method of the present invention does not select a set number of molecules randomly, but selects them based on the complementarity between the candidate molecules and the current mixing properties. This method can ensure that there is a representative combination of molecules that is close to the global optimum among the selected molecules.

[0021] Further, in step 2), calculating the difference between the mixture and the target petroleum sample after adding it to the representative subset includes: combining the property differences and molecular weight distribution differences between the mixture and the target petroleum sample to determine the difference between the mixture and the target petroleum sample.

[0022] In determining the differences, the method of this invention considers not only attribute differences but also molecular weight distribution differences. The molecular weight distribution fusion extension further incorporates the matching of distribution patterns into the core optimization framework, so that the final model can not only match point attributes but also key distribution attributes, thus comprehensively improving the fidelity.

[0023] Furthermore, in step 2), the attribute differences are obtained by grouping and weighting the macroscopic properties and NMR carbon types.

[0024] The method of this invention performs grouped differential weighting when determining attribute differences. By introducing intra-group average weights for precise spectral attributes such as NMR carbon type, it cleverly solves the balance problem of multi-scale and multi-precision attributes in a unified objective function.

[0025] Furthermore, in step 2), the grouped differential weighting formula is:

[0026] ;

[0027] Where F represents the property difference, P is the number of macroscopic properties, and Q is the number of NMR carbon types. as well as These are the property values ​​of the mixture. as well as The target values ​​for each attribute of the target oil sample are the experimental target values. as well as Experimental errors of experiments on various properties corresponding to the target oil sample.

[0028] This invention introduces a grouped differential weighting mechanism to divide attributes with different physical meanings into different attribute groups, and introduces an intra-group average weight for spectral attributes, so that the overall contribution of this type of attribute to the objective function matches its actual physical importance. This design does not empirically weight a single attribute, but rather performs structured weight allocation at the attribute category level, thereby achieving balanced optimization of multi-scale attributes without introducing too many free parameters.

[0029] Further, in step 2), the difference between the mixture and the target petroleum sample after adding it to the representative subset is calculated as follows: the difference between the mixture and the target petroleum sample is obtained by merging the attribute differences and molecular weight distribution differences through weighting coefficients.

[0030] Further, in step 2), the formula for calculating the difference between the mixture formed after adding it to the representative subset and the target petroleum sample is as follows:

[0031] ;

[0032] Among them, F total F represents the difference between the mixture and the target petroleum sample. props Indicates attribute differences, F mwd Indicating differences in molecular weight distribution, w mwd This represents the set molecular weight distribution weighting coefficient.

[0033] For molecular weight distribution, a key attribute with statistical distribution characteristics, relying solely on point-value attribute errors is insufficient to reflect the rationality of the molecular model in terms of the overall molecular composition. Therefore, this invention introduces a molecular weight distribution error term based on the aforementioned attribute error function, and integrates it with the point attribute error through weighting coefficients, so that the optimization objective simultaneously constrains the statistical distribution pattern and average attribute level of the mixture.

[0034] To achieve the above objectives, the present invention also provides a petroleum molecular model optimization selection system based on sequential batch processing, including a memory and a processor, and computer program instructions stored in the memory and running on the processor. The processor is used to execute the computer program instructions stored in the memory to implement the petroleum molecular model optimization selection method described above, and achieve the same beneficial effects as the method.

[0035] To achieve the above objectives, the present invention also provides a computer-readable storage medium storing a computer program internally, the computer program being executed by a processor to implement the steps of the petroleum molecular model optimization selection method described above, and to achieve the same beneficial effects as the method.

[0036] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described in detail below with reference to the accompanying drawings. Attached Figure Description

[0037] Figure 1 This is a schematic diagram of the petroleum molecular model optimization selection method of the present invention;

[0038] Figure 2 This is a schematic diagram of the structure of the grouped differential weighted objective function used in the method of this invention;

[0039] Figure 3 This is a schematic diagram of the sequential batch processing optimization selection process used in the method of this invention, and a schematic diagram of the differences between traditional single-molecule greedy selection and the batch processing selection of this invention in terms of search path and convergence effect.

[0040] Figure 4 This is a schematic diagram of an example of the final molecule generated in this invention;

[0041] Figure 5 This is a schematic diagram of 18 limiting parameters in the system optimization process of this invention. Detailed Implementation

[0042] The technical solution of the present invention will be clearly and completely described below with reference to specific embodiments. However, those skilled in the art should understand that the embodiments described below are only for illustrating the present invention and should not be regarded as limiting the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0043] Example of an Optimization Method for Petroleum Molecular Models Based on Sequential Batch Processing

[0044] The method of this invention, through vectorized computation, round-by-round batch selection, and a multi-objective weighted objective function, achieves rapid and stable screening of highly representative molecular subsets from a massive candidate pool. For example... Figure 1 As shown, the method in this embodiment includes the following steps:

[0045] S1. Load the candidate molecule pool and target attributes.

[0046] The method in this embodiment is for an established candidate molecule pool. The focus is on how to obtain representative molecule combinations that are closer to the global optimum from this pool. The specific method for establishing the candidate molecule pool is not specifically limited in this embodiment. The candidate pool is preferably generated by a petroleum molecule generation method based on module fusion and random perturbation (as described in application number 2026107453768), but the optimization selection method of this invention is not limited to this. The representative molecule combinations selected in this embodiment are for constructing a molecular model of the target petroleum. Therefore, in order to ensure that the selected representative molecule combinations can more accurately represent the target petroleum, the target attributes obtained in this embodiment are the experimental target values ​​of each attribute corresponding to the target petroleum sample and their measurement errors.

[0047] Therefore, this step in this embodiment is specifically as follows: loading a pool containing a large number of candidate molecules, pre-calculating or associating a set of key physicochemical property vectors for each molecule, wherein the property vectors include at least the properties required for rapid molecular characterization of petroleum; simultaneously, loading the experimental target values ​​of each property and their measurement errors corresponding to the target petroleum sample, specifically, pre-calculating the property vectors for each candidate molecule, wherein the property vectors include macroscopic properties, nuclear magnetic resonance carbon type distribution and molecular weight distribution characteristics.

[0048] S2. Construct the weighted multi-objective error function F total .

[0049] In this embodiment, the error function is used to determine the difference between the selected molecular combination and the target petroleum sample, and then the difference result can be used as the basis for selection to select a representative molecular combination that is closer to the global optimum.

[0050] This embodiment defines an objective function for evaluating the difference between any subset of molecules (mixture) and the target sample. This function is a weighted sum of mean squared errors, and its innovation lies in:

[0051] 1. Grouped Differential Weighting:

[0052] The attributes are divided into at least two groups, for example, the first group is macroscopic properties, and the second group is the distribution of carbon types in nuclear magnetic resonance. An intra-group average weighting coefficient is introduced for the second group of attributes to ensure that spectral attributes with smaller values ​​but higher precision have a reasonable proportion in the objective function.

[0053] ;

[0054] Where F represents the property difference, P is the number of macroscopic properties, and Q is the number of NMR carbon types. as well as These are the property values ​​of the mixture. as well as The target values ​​for each attribute of the target oil sample are the experimental target values. as well as Experimental errors for each attribute of the target petroleum sample. The first P term represents the first group of attributes (e.g., 7 macroscopic properties), and the second Q term represents the second group of attributes (e.g., 11 NMR carbon types). Figure 5 The diagram shows the 18 limiting parameters used for function convergence. The second group shares a weight coefficient of 1 / Q to achieve intra-group averaging, as shown below. Figure 2 As shown, different weighting methods are applied to macroscopic properties and NMR properties.

[0055] 2. Molecular weight distribution fusion and expansion:

[0056] To incorporate the continuous distribution property of molecular weight into the optimization, the above objective function is extended to the overall objective function F. total :

[0057] ;

[0058] Among them, F total F represents the difference between the mixture and the target petroleum sample. props This represents the attribute difference (i.e., the aforementioned attribute error function F), F mwd The difference in molecular weight distribution (i.e., a measure of the difference between the molecular weight distribution of the mixture and the target sample, such as Euclidean distance or KL divergence), w mwd This represents the set molecular weight distribution weighting coefficient (e.g., 0.3), which is used to balance the property fitting and distribution shape fitting.

[0059] To avoid interference between attributes with different dimensions, measurement precision, and physical meanings during the unified optimization process, this embodiment does not use a simple equal-weighted summation method when constructing the multi-objective error function. Instead, based on the actual physical constraints and experimental data characteristics during the construction of the petroleum molecular model, the structure of the error function is specifically designed:

[0060] First, there are significant differences in measurement accuracy, numerical scale, and impact on the representativeness of molecular models between macroscopic physicochemical properties and spectral fine-structure attributes. Macroscopic physicochemical properties typically have large numerical amplitudes and relatively high experimental errors, while spectral attributes such as NMR carbon type distribution have smaller numerical amplitudes but high measurement accuracy and a more direct constraint on the rationality of molecular structures. If the traditional method of weighted error accumulation is used, macroscopic attributes may dominate the objective function, thus "submerging" high-precision spectral attributes during the optimization process. The resulting molecular model may match the macroscopic properties, but it will have significant deviations at the molecular structure level.

[0061] To address the aforementioned issues, a differentiated weighting mechanism is introduced to group attributes with different physical meanings into different attribute groups. An average weight within each group is then applied to spectral attributes, ensuring that the overall contribution of this attribute type to the objective function matches its actual physical importance. This design does not empirically weight single attributes but rather performs structured weight allocation at the attribute category level, thereby achieving balanced optimization of multi-scale attributes without introducing excessive free parameters.

[0062] Furthermore, considering molecular weight distribution, a key attribute with statistical distribution characteristics, relying solely on point-value attribute errors is insufficient to reflect the rationality of the molecular model in terms of the overall molecular composition. Therefore, this embodiment introduces a molecular weight distribution error term based on the aforementioned attribute error function, and integrates it with the point attribute error through weighting coefficients, so that the optimization objective simultaneously constrains the statistical distribution pattern and average attribute level of the mixture.

[0063] Through the above objective function structure design, the introduction of representative molecular subsets in the subsequent optimization process can not only gradually reduce the errors in macroscopic and spectral attributes, but also guide the molecular weight distribution of the mixture to continuously converge toward the target petroleum sample, thereby significantly improving the comprehensive matching accuracy and physical rationality of the final molecular model at multiple attribute and scale levels.

[0064] S3. Using the constructed weighted multi-objective error function, the molecules in the candidate molecule pool are selected through sequential batch processing optimization to obtain a representative subset after screening.

[0065] like Figure 3 The diagram illustrates the sequential batch processing optimization selection process of this embodiment, employing an iterative sequence selection workflow to progressively construct a representative subset of molecules:

[0066] 1. First round: Initialize seed selection.

[0067] Using vectorized computation techniques, the error value F between the attribute vector of each individual molecule in the candidate pool and the target attribute vector is calculated in parallel at one time. single Choose to make Fsingle The smallest molecule is added to the representative subset as the first seed molecule. This step does not presuppose the oil type and is entirely data-driven.

[0068] 2. Round K (K>1): Batch processing greedy selection.

[0069] This is the core step of the present invention, used to avoid local optima in a single-step greedy approach.

[0070] a. Suppose that the currently selected representative subset contains M. current Several molecules, whose mixing properties are known;

[0071] b. Define the number of molecules to be selected in this round as the batch size B (e.g., B=5).

[0072] c. Among the remaining candidate molecules, evaluate all possible combinations of molecules of size B. For each candidate combination, calculate the M that, when added to the current representative subset, would form. current The total objective function value corresponding to a mixture of +B molecules ;

[0073] To reduce computational complexity in this step, this embodiment first pre-screens a predetermined number of seed molecules from the remaining seed molecules according to a set rule. Then, it enumerates all combinations of molecules with a predetermined batch size from all the selected seed molecules to form candidate combinations. A selection process is then performed on each candidate combination. Furthermore, to ensure that there are representative molecular combinations close to the global optimum among the selected molecules, the rule in this embodiment is that the candidate molecules have the highest complementarity with the current mixing properties of the top predetermined number of seed molecules.

[0074] d. Choose the option that enables Compared to the current F total The molecular combination that decreased the most;

[0075] e. Add all B molecules from the combination to the representative subset at once and update the mixture properties.

[0076] 3. Repeat step 2 for batch greedy selection until the number of molecules in the representative subset reaches the preset target value M. total .

[0077] S4. Optimize process monitoring and output.

[0078] Record the molecule index selected in each round, the corresponding objective function value, and the improvement compared to the previous round, forming an optimization history. Finally, output a representative subset list of optimized molecules and a report comparing their predicted mixing properties with the target properties.

[0079] Compared with the prior art, the present invention has the following significant advantages:

[0080] 1. Strong global optimization capability: By using "batch greedy selection", a combination of molecules is evaluated and selected each time instead of a single molecule, which greatly expands the space of single-step search and effectively overcomes the defect of traditional single-step greedy algorithms that are prone to getting trapped in local optima. It can find representative combination of molecules that is closer to the global optimum.

[0081] 2. Extremely high computational efficiency: It fully utilizes vectorized computation to perform parallel calculations on the properties of tens of thousands of molecules at once, avoiding inefficient serial computations in optimization loops. Combined with batch processing strategies, it significantly reduces the number of iterations while ensuring optimization performance, making it feasible to process large-scale candidate pools (>10,000 molecules).

[0082] 3. High accuracy in multi-objective fitting: The innovative grouped differentiated weighted objective function cleverly solves the balance problem of multi-scale and multi-precision attributes in a unified objective function by introducing intra-group average weights for precise spectral attributes such as NMR carbon type. The molecular weight distribution fusion extension further incorporates the matching of distribution morphology into the core optimization framework, enabling the final model to match not only point attributes but also key distribution attributes, thus comprehensively improving fidelity.

[0083] 4. Good practicality and interpretability: The algorithm has a clear flow, and each round of optimization has a clear record of the improvement, which is convenient for monitoring and debugging. The final output molecular subset has clear physicochemical significance and can be directly used for downstream molecular simulation or process analysis.

[0084] The method of this embodiment is further illustrated by taking the selection of 500 representative molecules from approximately 50,000 candidate molecules to construct a molecular model of a certain crude oil as an example:

[0085] S1 Preparation: Load the QMR attribute table (18-dimensional) of 50k molecules, as well as the 18 target values ​​of the crude oil and their measurement errors. Load the GPC molecular weight distribution data for this crude oil.

[0086] S2 Configuration: Set the objective function. The first 7 attributes are macroscopic properties with no additional weight; the last 11 attributes are NMR carbon types, sharing a weight of 1 / 11. Set the molecular weight distribution weight w. mwd =0.3, F mwd The root mean square error of the distribution curve is used.

[0087] S3 Optimization Execution:

[0088] 1. Seed selection:

[0089] Vectorized calculation of single-molecule error F of 50,000 molecules single(Consider only QMR attributes);

[0090] Select F single The smallest molecule (let's say ID: 12345) is added to the representative subset as a seed;

[0091] In the calculation of mixed attributes, the attribute vector of the first seed molecule is the current mixed attribute vector: X current =X 12345 .

[0092] 2. Batch processing selection (B=5):

[0093] The current blending properties are known:

[0094] The current mixture attribute vector X is calculated by weighted averaging of the selected molecular attribute vectors. current :

[0095] ;

[0096] Among them, M current X represents the number of molecules in the current subset. i The selected molecular attribute vector includes macroscopic attributes, NMR carbon type, and molecular weight distribution characteristics.

[0097] Candidate molecule complementarity determination:

[0098] For the remaining 49,999 candidate molecules, calculate the improvement in the objective function after adding them to the current representative subset:

[0099] ;

[0100] Among them, F total (X current ) represents the objective function F corresponding to the current subset. total Value, F total (X current +X candidate ) represents the objective function F after adding the candidate molecules to the current representative subset. total value. The larger the value, the higher the complementarity between the candidate molecule and the current mixing properties.

[0101] choose The top 1000 molecules are used as candidate combinations for this round of pre-screening to reduce the computational cost of searching the entire combination.

[0102] Greedy selection in batch processing:

[0103] From the pre-selected 1000 molecules, enumerate all combinations of 5 molecules and calculate the total objective function value after adding the current representative subset, denoted as . ;

[0104] Choose to The five molecules with the largest decrease (let's say IDs: 23456, 34567, 45678, 56789, 67890) are added to the representative subset at once.

[0105] Update the current blended attribute vector X current .

[0106] 3. Repeat step 2 for batch processing selection:

[0107] Each round involves a new batch selection based on the updated mixed attributes.

[0108] Records for each round: selected molecule index, objective function value, and improvement amount. This will lead to an optimized historical trajectory.

[0109] S4 output:

[0110] Selection complete. The log shows the objective function F... total The initial value was reduced from 1500 (the first molecule only) to 25 (a mixture of 500 molecules). A list of these 500 molecules was output, and the calculated mixing properties compared to the target crude oil showed errors within experimentally acceptable limits, with the molecular weight distribution curves showing a high degree of agreement. Figure 4 The image shows an example of the final generated molecule.

[0111] The method of this invention is not only applicable to the construction of petroleum molecule models, but can also be extended to the problem of representative molecule selection in complex mixture systems.

[0112] Example of an Optimization Selection System for Petroleum Molecular Models Based on Sequential Batch Processing

[0113] A petroleum molecule model optimization and selection system based on sequential batch processing includes a memory and a processor, as well as computer program instructions stored in the memory and running on the processor. The processor executes the computer program instructions stored in the memory to implement the petroleum molecule model optimization and selection method proposed in the above-described embodiment of the sequential batch processing-based petroleum molecule model optimization and selection method. Specifically, the system may include a data preprocessing module for loading candidate pools, target attributes, and errors; an objective function configuration module for setting grouping weights and molecular weight distribution weights; a vectorized calculation engine for efficiently calculating the attribute errors of individual molecules and mixtures; a sequential optimization execution module for executing seed selection and batch greedy selection processes; and a result recording and output module.

[0114] Computer-readable storage medium embodiments

[0115] One embodiment of the present invention is a computer-readable storage medium storing a computer program that, when executed by a processor, implements the petroleum molecule model optimization selection method proposed in the above-described embodiment of the petroleum molecule model optimization selection method based on sequential batch processing. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, or flash memory.

[0116] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. A method for optimizing the selection of petroleum molecular models based on sequential batch processing, characterized in that, Includes the following steps: 1) Select a seed molecule from the candidate pool as the first seed molecule and add it to the representative subset; 2) Select a combination of molecules that contains at least the set batch size from the remaining candidate molecules in the candidate pool. For each combination of molecules, calculate the difference between the mixture formed after adding it to the representative subset and the target petroleum sample, and select the combination of molecules with the smallest difference to add to the representative subset. After each round of molecular combination is added to a representative subset, the current mixture property is updated, and the updated mixture property is used for the next round of molecular combination selection; 3) Repeat step 2) until the number of molecules in the representative subset reaches the preset target value.

2. The method for optimizing the selection of petroleum molecular models according to claim 1, characterized in that, In step 1), selecting a seed molecule from the candidate pool includes: calculating the error value between the attribute vector of each individual molecule in the candidate pool and the target attribute vector, and selecting the molecule that minimizes the error value as the first seed molecule to be added to the representative subset.

3. The method for optimizing the selection of petroleum molecular models according to claim 1, characterized in that, In step 2), selecting a molecular combination containing at least a set batch size from the remaining candidate molecules in the candidate pool includes: first, pre-screening a set number of molecules from the candidate molecules according to a set rule, and then enumerating all molecular combinations with a set batch size among all the selected molecules to form a molecular combination; the set rule is: the candidate molecules are the first set number of molecules with the highest complementarity to the current mixing attribute.

4. The method for optimizing the selection of petroleum molecular models according to claim 1, characterized in that, In step 2), the difference between the mixture and the target oil sample after adding it to the representative subset is calculated, including: combining the differences in properties and molecular weight distribution between the mixture and the target oil sample to determine the difference between the mixture and the target oil sample.

5. The method for optimizing the selection of petroleum molecular models according to claim 4, characterized in that, In step 2), the attribute differences are obtained by grouping and weighting the macroscopic properties and NMR carbon types.

6. The method for optimizing the selection of petroleum molecular models according to claim 5, characterized in that, In step 2), the grouped differential weighting formula is: ； Where F represents the property difference, P is the number of macroscopic properties, and Q is the number of NMR carbon types. as well as These are the property values ​​of the mixture. as well as The target values ​​for each attribute of the target oil sample are the experimental target values. as well as Experimental errors of experiments on various properties corresponding to the target oil sample.

7. The method for optimizing the selection of petroleum molecular models according to claim 4, characterized in that, In step 2), the difference between the mixture and the target petroleum sample after adding it to the representative subset is calculated as follows: the difference between the mixture and the target petroleum sample is obtained by merging the attribute differences and molecular weight distribution differences through weighting coefficients.

8. The method for optimizing the selection of petroleum molecular models according to claim 7, characterized in that, In step 2), the formula for calculating the difference between the mixture formed after adding it to the representative subset and the target petroleum sample is as follows: ； Among them, F total F represents the difference between the mixture and the target petroleum sample. props Indicates attribute differences, F mwd Indicating differences in molecular weight distribution, w mwd This represents the set molecular weight distribution weighting coefficient.

9. A petroleum molecular model optimization selection system based on sequential batch processing, characterized in that, It includes a memory and a processor, as well as computer program instructions stored in the memory and running on the processor, the processor being used to execute the computer program instructions stored in the memory to implement the petroleum molecular model optimization selection method as described in any one of claims 1 to 8.

10. A computer-readable storage medium, wherein a computer program is stored internally, characterized in that, The computer program is executed by a processor to implement the steps of the petroleum molecular model optimization selection method as described in any one of claims 1 to 8.

Images