Method and apparatus for predicting physical property data for process simulation

Abstract

Disclosed in the present invention are a method and apparatus for predicting physical property data for process simulation. The method for predicting physical property data comprises: acquiring the types and number of groups of crude oil molecules in a crude oil molecular library; on the basis of the types and number of groups of the crude oil molecules, estimating physical properties of the crude oil molecules, and obtaining a first physical property result; acquiring several categories of physical properties of isomers and modifying the first physical property result to obtain a second physical property result; acquiring feature data of several first molecules, and on the basis of the feature data of the several first molecules, constructing a physical property prediction model; acquiring feature data of the crude oil molecules, and using the feature data of the crude oil molecules as input data of the physical property prediction model, and by means of the physical property prediction model, obtaining a third physical property result of the crude oil molecules; and on the basis of the second physical property result and the third physical property result, obtaining a physical property prediction result of the crude oil molecules. The present invention integrates three approaches to predict physical properties of crude oil molecules, thereby realizing the accurate prediction of the physical properties of molecules.

Description

A method and apparatus for predicting physical property data in process simulation. Technical Field

[0001] This invention relates to the field of molecular property prediction, and more particularly to a method and apparatus for predicting property data for process simulation. Background Technology

[0002] Hydrocarbon macromolecules have complex structures and numerous isomers, making it impossible to collect sufficient pure components from nature for property analysis. Many fundamental properties, such as boiling point and enthalpy of formation, are lacking, hindering process simulation calculations. Therefore, the application of crude oil molecular analysis technology is currently limited to upstream areas such as crude oil blending and property prediction. Downstream applications based on process simulation calculations, such as the refinement of crude oil and the prediction of product properties after atmospheric and vacuum distillation, remain unrealized. Accurately estimating the fundamental macromolecular properties of crude oil will facilitate its application in downstream process simulation calculations, further guiding the refinement of crude oil production.

[0003] Therefore, there is an urgent need for a predictive method for physical property data that can accurately estimate the basic physical properties of crude oil macromolecules. Summary of the Invention

[0004] This invention provides a method and apparatus for predicting physical property data for process simulation, which effectively improves the accuracy of estimating the basic physical properties of crude oil macromolecules.

[0005] To address the aforementioned technical problems, embodiments of the present invention provide a method for predicting physical property data for process simulation, comprising:

[0006] Obtain the group types and numbers of crude oil molecules in the crude oil molecule library;

[0007] The physical properties of the crude oil molecules are estimated based on the group type and number of groups, and a first physical property result is obtained.

[0008] Obtain several types of physical properties of the isomers, and modify the first physical property result based on the several types of physical properties of the isomers to obtain the second physical property result;

[0009] Acquire feature data of several first molecules, and construct a property prediction model based on the feature data of the several first molecules;

[0010] The characteristic data of the crude oil molecules are obtained and used as input data for the physical property prediction model. The third physical property result of the crude oil molecules is obtained through the physical property prediction model.

[0011] Based on the second and third property results, the predicted properties of the crude oil molecules are obtained.

[0012] The beneficial effects of the embodiments of the present invention are as follows:

[0013] This invention estimates the physical properties of crude oil molecules based on the group types and numbers in the obtained crude oil molecule library. It then corrects the estimated properties based on several types of isomers, constructs a property prediction model to obtain another property estimate for the crude oil molecules, and finally obtains the final property prediction result for the crude oil molecules based on the corrected estimated properties and the other property estimate structure. This invention combines three methods to predict the physical properties of crude oil molecules, enabling accurate prediction. The accurate prediction results facilitate the application of crude oil molecule analysis technology to downstream process simulation calculations, further guiding the refined production of crude oil.

[0014] As a preferred embodiment, the method for predicting physical property data for process simulation further includes:

[0015] Define the group types of several groups, and statistically obtain the physical properties of several second molecules, as well as the group types and number of groups in each molecule;

[0016] Based on the physical properties of the aforementioned second molecules and the group type and number of groups in each molecule, the contribution value of each group type to the physical properties is obtained;

[0017] The contribution value of each group type to the physical properties is stored in the crude oil physical property database; wherein, the largest molecule in the crude oil physical property database is C100.

[0018] This preferred scheme sets the group types of several groups, statistically obtains the physical properties of several second molecules as well as the group types and number of groups in each molecule, and stores the contribution value of each group type to the physical properties in the crude oil physical property database, which can regress the group contribution value with a wider range of applications.

[0019] As a preferred embodiment, obtaining the group types and numbers of crude oil molecules in the crude oil molecular library includes:

[0020] Obtain the SOL format string of crude oil molecules in the crude oil molecule library, and perform standardization processing on the SOL format string to obtain the SMILES format string of crude oil molecules;

[0021] Convert the SMILES format string of the crude oil molecule into an MOL file of the crude oil molecule;

[0022] The SMARTS method is used to describe the several groups, and the SMARTS strings of the several groups are obtained;

[0023] Convert the SMARTS strings of the aforementioned groups into MOL files for the groups;

[0024] By comparing the MOL files of the crude oil molecules with the MOL files of the functional groups, the functional group types and numbers of crude oil molecules in the crude oil molecule library can be obtained.

[0025] This preferred method obtains and compares the MOL files of crude oil molecules and groups to determine the group types and numbers of crude oil molecules, so as to subsequently estimate the physical properties of crude oil molecules based on the group types and numbers of crude oil molecules.

[0026] As a preferred embodiment, the step of estimating the physical properties of the crude oil molecules based on the group type and number of groups, and obtaining a first physical property result, includes:

[0027] A preliminary physical property estimate of the crude oil molecules is performed using a physical property calculation formula, which is as follows:

[0028] f=∑ni*Gi+a

[0029] In the above formula, ni represents the number of functional groups, Gi represents the contribution value of functional group properties, and a is a preset parameter.

[0030] This preferred scheme estimates the physical properties of crude oil molecules using the group contribution value method.

[0031] As a preferred embodiment, the step of acquiring feature data of several first molecules and constructing a property prediction model based on the feature data of the several first molecules includes:

[0032] Acquire characteristic data of several first molecules, the true values ​​of preset physical properties of several first molecules, and basic physical properties; wherein, the characteristic data includes the number of carbons, degree of unsaturation, number of rings, number of heteroatoms, and number of branches of each molecule; the basic physical properties include density and boiling point;

[0033] Obtain a preset neural network model and initialize the weights of the preset neural network model;

[0034] The feature data and basic physical properties of the aforementioned first molecules are used as input data for the preset neural network model, and output prediction data is obtained through the preset neural network model.

[0035] The error is obtained based on the true values ​​of the preset physical properties of the plurality of first molecules and the output prediction data, and the weights of the preset neural network model are adjusted according to the error; this step is repeated until the error is less than the preset value.

[0036] By cross-validation, the optimal parameters of the preset neural network model are determined, and the physical property prediction model is constructed.

[0037] This preferred solution obtains a weighted neural network model by using the feature data of several first molecules, the true values ​​of the preset physical properties of several first molecules, and the basic physical properties. The optimal parameters of the preset neural network model are determined by cross-validation, thus completing the construction of the physical property prediction model.

[0038] Accordingly, in order to solve the above-mentioned technical problems, embodiments of the present invention also provide a predictive device for physical property data, including: an acquisition module, an estimation module, a correction module, a construction module, a model prediction module, and a physical property acquisition module;

[0039] The acquisition module is used to acquire the group type and number of groups of crude oil molecules in the crude oil molecule library;

[0040] The estimation module is used to estimate the physical properties of the crude oil molecules based on the group type and number of groups, and obtain a first physical property result.

[0041] The correction module is used to obtain several types of physical properties of the isomers, and to correct the first physical property result based on the several types of physical properties of the isomers to obtain the second physical property result;

[0042] The construction module is used to acquire feature data of several first molecules and construct a property prediction model based on the feature data of the several first molecules;

[0043] The model prediction module is used to acquire the feature data of the crude oil molecules and use the feature data of the crude oil molecules as input data of the physical property prediction model to obtain the third physical property result of the crude oil molecules through the physical property prediction model.

[0044] The property acquisition module is used to obtain the property prediction results of the crude oil molecules based on the second property result and the third property result.

[0045] As a preferred embodiment, the device for predicting physical property data further includes: a contribution value acquisition module; the contribution value acquisition module is used for:

[0046] Define the group types of several groups, and statistically obtain the physical properties of several second molecules, as well as the group types and number of groups in each molecule;

[0047] Based on the physical properties of the aforementioned second molecules and the group type and number of groups in each molecule, the contribution value of each group type to the physical properties is obtained;

[0048] The contribution value of each group type to the physical properties is stored in the crude oil physical property database; wherein, the largest molecule in the crude oil physical property database is C100.

[0049] As a preferred embodiment, the acquisition module includes: a processing unit, a first conversion unit, a description unit, a second conversion unit, and a comparison unit;

[0050] The processing unit is used to obtain the SOL format string of crude oil molecules in the crude oil molecule library, and to perform standardization processing on the SOL format string to obtain the SMILES format string of crude oil molecules.

[0051] The first conversion unit is used to convert the SMILES format string of the crude oil molecule into an MOL file of the crude oil molecule;

[0052] The description unit is used to describe the plurality of groups using SMARTS and obtain the SMARTS strings of the plurality of groups;

[0053] The second conversion unit is used to convert the SMARTS strings of the plurality of groups into MOL files of the groups;

[0054] The comparison unit is used to obtain the group type and number of crude oil molecules in the crude oil molecule library by comparing the MOL file of the crude oil molecule and the MOL file of the group.

[0055] As a preferred embodiment, the estimation module is used to estimate the physical properties of the crude oil molecules based on the group type and number of groups, and obtain a first physical property result, including:

[0056] A preliminary physical property estimate of the crude oil molecules is performed using a physical property calculation formula, which is as follows:

[0057] f=∑ni*Gi+a

[0058] In the above formula, ni represents the number of functional groups, Gi represents the contribution value of functional group properties, and a is a preset parameter.

[0059] As a preferred embodiment, the construction module includes: an acquisition unit, an initialization unit, an output unit, an adjustment unit, and a construction unit;

[0060] The acquisition unit is used to acquire characteristic data of several first molecules, the true values ​​of preset physical properties of several first molecules, and basic physical properties; wherein, the characteristic data includes the number of carbons, degree of unsaturation, number of rings, number of heteroatoms, and number of branches of each molecule; the basic physical properties include density and boiling point;

[0061] The initialization unit is used to obtain a preset neural network model and initialize the weights of the preset neural network model;

[0062] The output unit is used to take the feature data and basic physical properties of the plurality of first molecules as input data of the preset neural network model, and obtain output prediction data through the preset neural network model.

[0063] The adjustment unit is used to obtain the error based on the true values ​​of the preset physical properties of the plurality of first molecules and the output prediction data, and adjust the weights of the preset neural network model according to the error; repeat this step until the error is less than the preset value;

[0064] The construction unit is used to determine the optimal parameters of the preset neural network model through cross-validation, thereby completing the construction of the physical property prediction model. Attached Figure Description

[0065] Figure 1: A schematic flowchart of an embodiment of a method for predicting physical property data for process simulation provided by the present invention;

[0066] Figure 2: A schematic diagram of an embodiment of a physical property data prediction device provided by the present invention;

[0067] Figure 3: A schematic diagram of the structure of an embodiment of the acquisition module of a predictive device for physical property data provided by the present invention;

[0068] Figure 4: A schematic diagram of the structure of a construction module of a predictive device for physical property data provided by the present invention;

[0069] Figure 5: A comparison diagram of molecular group types provided in the embodiments of the present invention;

[0070] Figure 6: A schematic diagram of converting a SOL structure string to a SMILES structure string according to an embodiment of the present invention;

[0071] Figure 7: A schematic diagram of the prediction effect of the property prediction model provided in the embodiment of the present invention. Detailed Implementation

[0072] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. 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.

[0073] Hydrocarbon macromolecules have complex structures and numerous isomers, making it impossible to collect sufficient pure components from nature for property analysis. Many fundamental properties, such as boiling point and enthalpy of formation, are lacking, hindering process simulation calculations. Therefore, the application of crude oil molecular analysis technology is currently limited to upstream areas such as crude oil blending and property prediction. Downstream applications based on process simulation calculations, such as the refinement of crude oil and the prediction of product properties after atmospheric and vacuum distillation, remain unrealized. Accurately estimating the fundamental properties of crude oil macromolecules will facilitate the application of crude oil molecular analysis technology in downstream process simulation calculations, further guiding refined crude oil production. Therefore, a predictive method for property data in process simulation that can accurately estimate the fundamental properties of crude oil macromolecules is urgently needed.

[0074] Example 1

[0075] Please refer to Figure 1, which is a schematic flowchart of an embodiment of a method for predicting physical property data for process simulation provided by this invention. As shown in Figure 1, the method for predicting physical property data for process simulation includes steps 101 to 106, specifically:

[0076] Step 101: Obtain the group types and number of groups of crude oil molecules in the crude oil molecule library;

[0077] Step 102: Estimate the physical properties of the crude oil molecules based on the group type and number of groups, and obtain the first physical property result;

[0078] Step 103: Obtain several types of physical properties of the isomers, and correct the first physical property result based on the several types of physical properties of the isomers to obtain the second physical property result;

[0079] Step 104: Obtain feature data of several first molecules, and construct a property prediction model based on the feature data of the several first molecules;

[0080] Step 105: Obtain the characteristic data of the crude oil molecules, and use the characteristic data of the crude oil molecules as input data for the physical property prediction model, and obtain the third physical property result of the crude oil molecules through the physical property prediction model;

[0081] Step 106: Based on the second property result and the third property result, obtain the property prediction result of the crude oil molecule.

[0082] In this embodiment, the method for predicting physical property data for process simulation further includes:

[0083] Define the group types of several groups, and statistically obtain the physical properties of several second molecules, as well as the group types and number of groups in each molecule;

[0084] Based on the physical properties of the aforementioned second molecules and the group type and number of groups in each molecule, the contribution value of each group type to the physical properties is obtained;

[0085] The contribution value of each group type to the physical properties is stored in the crude oil physical property database; wherein, the largest molecule in the crude oil physical property database is C100.

[0086] In this embodiment, approximately 500 group types were defined, and the physical properties of approximately 3000 molecules, as well as the group type and number of groups in each molecule, were statistically obtained. Using a regression formula, based on the defined approximately 500 group types and the physical properties of the 3000 molecules, as well as the group type and number of groups in each molecule, data regression was performed with the minimum error to calculate the contribution value of group type to physical properties. The physical properties involved in the calculation process are those required in the process simulation, and the required physical properties are shown in the table below.

[0087] Table 1. Physical properties required for process simulation calculations

[0088] In this embodiment, the regression formula may be, but is not limited to, the following:

[0089] Ax = b

[0090] In the above formula, x is the number of functional groups, and A is the functional group contribution matrix.

[0091] In this embodiment, based on existing molecular property data and the number of molecular group structures, the regression formula or other regression methods in this embodiment can also be used to expand the types of properties.

[0092] Group contribution methods, such as those used by Gani and Benson, were initially based on property regression of approximately 1000 molecules, considering molecular sizes up to C30. With the availability of more real molecular property data, and referring to existing formats and group types, group contribution values ​​with a wider range of applicability were regressed. In this embodiment of the invention, the largest molecule in the crude oil property database is C100. This embodiment of the invention further refines molecular properties based on existing literature.

[0093] In this embodiment of the invention, several group types are defined, and the physical properties of several molecules, as well as the group type and number of groups in each molecule, are statistically obtained. The contribution value of each group type to the physical properties is then stored in the crude oil physical property database, which can regress group contribution values ​​with a wider range of applications.

[0094] Currently, there are three structured molecular description languages: SOL, SMILES, and SMARTS. Among them, SMILES can describe molecules relatively simply and is suitable for processing the conversion of SOL format information, but it cannot express the specific bond positions of molecules and is not suitable for describing group structures. On the other hand, the SMARTS language can describe the bonding of atoms and their surroundings in detail and is suitable for describing group structures. The following table shows how to describe groups using the three structured molecular description languages.

[0095] Table 2. Description of functional groups for SOL, SMILES, and SMARTS

[0096] In step 101, obtaining the group types and numbers of crude oil molecules in the crude oil molecule library includes:

[0097] Obtain the SOL format string of crude oil molecules in the crude oil molecule library, and perform standardization processing on the SOL format string to obtain the SMILES format string of crude oil molecules;

[0098] Convert the SMILES format string of the crude oil molecule into an MOL file of the crude oil molecule;

[0099] The SMARTS method is used to describe the several groups, and the SMARTS strings of the several groups are obtained;

[0100] Convert the SMARTS strings of the aforementioned groups into MOL files for the groups;

[0101] By comparing the MOL files of the crude oil molecules with the MOL files of the functional groups, the functional group types and numbers of crude oil molecules in the crude oil molecule library can be obtained.

[0102] In this embodiment, an open-source algorithm package, such as rdkit, can be used to convert the SMILES format string of the molecule into an MOL file.

[0103] MOL (Molecular Structure Data File) is a common chemical information file format used to store data related to molecular structure. MOL files typically contain information about the atoms, bonds, stereochemistry, and other related properties of a molecule, and can preserve bond information.

[0104] In this embodiment, an open-source algorithm package can be used to convert the SMARTS string describing the molecule or group into an MOL file.

[0105] In this embodiment, the order of comparing the MOL file of the crude oil molecule and the MOL file of the group is from the larger group with more atoms to the smaller group, and the groups do not overlap. That is, after some atoms in the molecular structure are occupied by a certain group, the subsequent matching groups cannot use these atoms.

[0106] Please refer to Figure 5, which is a molecular group type comparison diagram provided in an embodiment of the present invention. As shown in Figure 5, the molecular group is compared according to the comparison rules to determine whether it has atomic and bond information of a group. If it does, the corresponding group number is incremented by one.

[0107] Crude oil molecular detection technology combines four-component analysis, elemental analysis, high-temperature simulated distillation, gas chromatography, high-resolution mass spectrometry, and various derivatization methods to analyze the molecular composition of crude oil at each distillation range separately, and then synthesizes the results into a molecular analysis of the overall crude oil. The analysis results of crude oil molecules are often represented as SOL format strings.

[0108] In this embodiment, there are several methods for standardizing the SOL format string, including direct conversion, core filling, and manual filling. For simple linear alkane molecules, direct conversion is used, such as "2R H" in SOL format corresponding to "CC" in SMILES format. For more complex molecules, such as "A62R", core pre-filling is used, and the side chains are converted using direct conversion to form "c1c(CC)cccc1". For some complex macromolecules, when the first two methods are not satisfactory, manual filling is used.

[0109] In this embodiment, several processing methods can be used to process existing values ​​in the crude oil molecular library, and can also be used to generate SOL values ​​that are not in the crude oil molecular library.

[0110] In this embodiment, the crude oil molecule library contains approximately 20,000 molecules. The direct conversion method can process approximately 1,000 molecules, the core value filling method can process approximately 18,000 molecules, and the core value filling method can process approximately 300 molecules.

[0111] In step 102, the step of estimating the physical properties of the crude oil molecules based on the group type and number of groups, and obtaining a first physical property result, includes:

[0112] A preliminary physical property estimate of the crude oil molecules is performed using a physical property calculation formula, which is as follows:

[0113] f=∑ni*Gi+a

[0114] In the above formula, ni represents the number of functional groups, Gi represents the contribution value of functional group properties, and a is a preset parameter.

[0115] When performing different physical property calculations, the corresponding calculation formula is selected based on the physical property in order to achieve linearization.

[0116] Figure 6 is a schematic diagram of converting the SOL structure string to the SMILES structure string according to an embodiment of the present invention.

[0117] In step 103, several types of physical properties of isomers are obtained, and the first physical property result is corrected based on these isomer properties. This is because, as shown in Figure 6, one SOL structure may correspond to many SMILES, and different SMILES will have different numbers of groups, thus resulting in different physical properties. That is, crude oil molecules are parsed and expressed as SOL structure strings, but the SOL structured language cannot express the isomers of some hydrocarbon molecules. One SOL structure string may correspond to several or even dozens of isomers. Therefore, for small molecule isomers, the crude oil molecule library adopts a traversal method to record all isomer molecules when building the library. For large hydrocarbon molecules in the crude oil molecule library, since there are many isomers, and in order to control the data size of the crude oil molecule database, it is necessary to use different physical properties of isomers to correct the physical property data corresponding to one SOL structure.

[0118] When constructing functional groups, there may be insufficient consideration of some molecular structures and heteroatoms. Therefore, to avoid large cumulative errors in the calculation of macromolecular properties, this invention establishes a neural network property prediction model for each molecule based on molecular composition, structure, and other characteristic data.

[0119] In step 104, acquiring feature data of several first molecules and constructing a property prediction model based on the feature data of the several first molecules includes:

[0120] Acquire characteristic data of several first molecules, the true values ​​of preset physical properties of several first molecules, and basic physical properties; wherein, the characteristic data includes the number of carbons, degree of unsaturation, number of rings, number of heteroatoms, and number of branches of each molecule; the basic physical properties include density and boiling point;

[0121] Obtain a preset neural network model and initialize the weights of the preset neural network model;

[0122] The feature data and basic physical properties of the aforementioned first molecules are used as input data for the preset neural network model, and output prediction data is obtained through the preset neural network model.

[0123] The error is obtained based on the true values ​​of the preset physical properties of the plurality of first molecules and the output prediction data, and the weights of the preset neural network model are adjusted according to the error; this step is repeated until the error is less than the preset value.

[0124] By cross-validation, the optimal parameters of the preset neural network model are determined, and the physical property prediction model is constructed.

[0125] In this embodiment, the optimal parameters of the neural network model may be, but are not limited to, the number of hidden layer neurons and the algorithm learning rate.

[0126] In step 105, after the neural network model is established, the model is input with data such as the number of carbon atoms, degree of unsaturation, number of rings, number of heteroatoms, number of branches, density, and boiling point of the unknown molecule. Multi-layer weighted calculation is then performed in combination with the multi-layer weights that have been trained within the model to finally obtain the model's prediction results of the physical properties of the unknown molecule.

[0127] In step 106, the third physical property result of crude oil molecules predicted by the neural network model, the first physical property result obtained by physical property estimation in the aforementioned steps, and the second physical property result obtained by correcting the first physical property result based on several types of physical properties of isomers are combined to obtain the final physical property prediction result of crude oil molecules.

[0128] Accordingly, this embodiment of the invention also provides a predictive device 20 for physical property data. Please refer to Figures 2 to 4. Figure 2 is a structural schematic diagram of an embodiment of the predictive device for physical property data provided by the present invention. Figure 3 is a structural schematic diagram of an embodiment of the acquisition module of the predictive device for physical property data provided by the present invention. Figure 4 is a structural schematic diagram of an embodiment of the construction module of the predictive device for physical property data provided by the present invention.

[0129] As shown in Figure 2, the prediction device 20 for the physical property data includes: an acquisition module 201, an estimation module 202, a correction module 203, a construction module 204, a model prediction module 205, and a physical property acquisition module 206.

[0130] The acquisition module 201 is used to acquire the group type and number of groups of crude oil molecules in the crude oil molecule library;

[0131] The estimation module 202 is used to estimate the physical properties of the crude oil molecules based on the group type and number of groups, and obtain a first physical property result;

[0132] The correction module 203 is used to obtain several types of physical properties of the isomers, and to correct the first physical property result according to the several types of physical properties of the isomers to obtain the second physical property result;

[0133] The construction module 204 is used to acquire feature data of several first molecules and construct a property prediction model based on the feature data of the several first molecules;

[0134] The model prediction module 205 is used to acquire the feature data of the crude oil molecules and use the feature data of the crude oil molecules as input data of the physical property prediction model to obtain the third physical property result of the crude oil molecules through the physical property prediction model.

[0135] The property acquisition module 206 is used to obtain the property prediction results of the crude oil molecules based on the second property result and the third property result.

[0136] In this embodiment, the prediction device 20 for the physical property data further includes: a contribution value acquisition module 207; the contribution value acquisition module 207 is used for:

[0137] Define the group types of several groups, and statistically obtain the physical properties of several second molecules, as well as the group types and number of groups in each molecule;

[0138] Based on the physical properties of the aforementioned second molecules and the group type and number of groups in each molecule, the contribution value of each group type to the physical properties is obtained;

[0139] The contribution value of each group type to the physical properties is stored in the crude oil physical property database; wherein, the largest molecule in the crude oil physical property database is C100.

[0140] In this embodiment, as shown in FIG3, the acquisition module 201 includes: a processing unit 2011, a first conversion unit 2012, a description unit 2013, a second conversion unit 2014, and a comparison unit 2015;

[0141] The processing unit 2011 is used to obtain the SOL format string of crude oil molecules in the crude oil molecule library, and to perform standardization processing on the SOL format string to obtain the SMILES format string of crude oil molecules.

[0142] The first conversion unit 2012 is used to convert the SMILES format string of the crude oil molecule into an MOL file of the crude oil molecule;

[0143] The description unit 2013 is used to describe the plurality of groups using SMARTS and obtain the SMARTS strings of the plurality of groups;

[0144] The second conversion unit 2014 is used to convert the SMARTS strings of the plurality of groups into MOL files of the groups;

[0145] The comparison unit 2015 is used to obtain the group type and number of crude oil molecules in the crude oil molecule library by comparing the MOL file of the crude oil molecule and the MOL file of the group.

[0146] In this embodiment, the estimation module 202 is used to estimate the physical properties of the crude oil molecules based on the group type and number of groups, and obtain a first physical property result, including:

[0147] A preliminary physical property estimate of the crude oil molecules is performed using a physical property calculation formula, which is as follows:

[0148] f=∑ni*Gi+a

[0149] In the above formula, ni represents the number of functional groups, Gi represents the contribution value of functional group properties, and a is a preset parameter.

[0150] In this embodiment, as shown in FIG4, the construction module 204 includes: an acquisition unit 2041, an initialization unit 2042, an output unit 2043, an adjustment unit 2044, and a construction unit 2045;

[0151] The acquisition unit 2041 is used to acquire characteristic data of several first molecules, the true values ​​of preset physical properties of several first molecules, and basic physical properties; wherein, the characteristic data includes the number of carbons, degree of unsaturation, number of rings, number of heteroatoms, and number of branches of each molecule; the basic physical properties include density and boiling point;

[0152] The initialization unit 2042 is used to obtain a preset neural network model and initialize the weights of the preset neural network model;

[0153] The output unit 2043 is used to take the feature data and basic physical properties of the plurality of first molecules as input data of the preset neural network model, and obtain output prediction data through the preset neural network model.

[0154] The adjustment unit 2044 is used to obtain the error based on the true values ​​of the preset physical properties of the plurality of first molecules and the output prediction data, and adjust the weights of the preset neural network model according to the error; repeat this step until the error is less than the preset value;

[0155] The construction unit 2045 is used to determine the optimal parameters of the preset neural network model through cross-validation, thereby completing the construction of the physical property prediction model.

[0156] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0157] 1. In this embodiment of the invention, the physical properties of crude oil molecules are estimated based on the group types and numbers of crude oil molecules in the obtained crude oil molecule library. The estimated physical properties are then corrected based on several types of physical properties of isomers. A physical property prediction model is then constructed to obtain another physical property estimation result for the crude oil molecules. Finally, the final physical property prediction result of the crude oil molecules is obtained based on the corrected estimated physical properties and the other physical property estimation structure. This invention combines three methods to predict the physical properties of crude oil molecules, which can achieve accurate prediction of the physical properties of crude oil molecules. The accurate prediction results obtained help to apply crude oil molecule analysis technology to downstream process simulation calculations, and can further guide the refined production of crude oil.

[0158] 2. In this embodiment of the invention, several group types are defined, and the physical properties of several second molecules, as well as the group type and number of groups in each molecule, are statistically obtained. The contribution value of each group type to the physical properties is stored in the crude oil physical property database, which can regress the group contribution value with a wider range of applications.

[0159] 3. In addition to predicting the physical properties of crude oil molecules in the crude oil molecular library, the embodiments of the present invention can also predict the physical properties of molecules with any known structure that are not in the crude oil molecular library.

[0160] Example 2

[0161] This invention provides another embodiment of the method for predicting physical property data, and the specific embodiment is as follows:

[0162] In this embodiment, the first step is to convert the SOL structure string of the molecule into the SMILES structure string, and the method used is shown in the table below.

[0163] In this embodiment, Method 1 is a direct conversion method used to process simple straight-chain alkane molecules. When branches are present, the branches are distributed to different positions of the carbon chain molecules according to specific rules. Method 2 is a core pre-filling and side chain direct conversion method.

[0164] Table 3 Methods for converting SOL to SMILES

[0165] As shown in Table 3, molecule number 9 is the core of molecules numbered 10-15. For molecules numbered 10-15, only the side chains need to be added to the known SMILES structure string of molecule number 9. When the SMILES structure string of the core molecule is fixed, the side chain lengths and structures of its derived molecules are different. The conversion from SOL structure string to SMILES structure string can be completed by simply concatenating the side chain SMILES structure string to the core SMILES structure string. If some molecules are too rare and have no derived types, the conversion is completed by manually filling in the values.

[0166] In this embodiment, the second step is to define the group type and calculate the contribution value of the group property.

[0167] In this embodiment, approximately 500 group types are defined and the basic properties of approximately 3,000 molecules are collected for the calculation of group property contribution values.

[0168] As shown in Table 1, the calculated physical property types are DHFORM, DGFORM, CPIG, CPL, Tb, Tc, Pc, Vc, Hv, Psat, Surface, and Viscosity.

[0169] Based on hydrocarbon molecule data, such as the physical properties of straight-chain alkanes and isoalkanes from C1 to C10, statistical regression is used to obtain group property values. The partial regression results are shown in the table below. The regression formula is similar to the regression formula in Example 1. Since the group contribution method is only a summation of values ​​and the regression equation is a linear equation, it has a faster convergence speed.

[0170] In this embodiment, some regression value examples are provided, as shown in the table below.

[0171] Table 4 Regression Value Cases

[0172] In this embodiment, the third step is to generate the MOL file and matching molecular groups.

[0173] After molecules are expressed using SMILES structure strings and functional groups are expressed using SMARTS structure strings, the rdkit open-source package can be used to convert the SMILES and SMARTS structure strings into MOL format files; by parsing the MOL file content, the matching of molecules and functional groups can be completed.

[0174] In this embodiment, the fourth step is the calculation of physical properties and the display of results.

[0175] In the third step of this embodiment, the types and number of functional groups can be obtained. Then, based on the defined formula for the contribution of different functional groups, the basic physical properties of a SMILES structure molecule can be calculated, as shown in the table below.

[0176] Table 5 Results of Group Contribution Method

[0177] In this embodiment, the fifth step is the correction of the physical properties of isomers.

[0178] In this embodiment, for a small hydrocarbon molecule with an SOL structure, its possible isomers are given by an ergonomic approach; for a large hydrocarbon molecule, its possible isomers are given as much as possible.

[0179] As shown in molecules 1 to 3 in Table 5, small molecule isomers also have different physical properties; for macromolecules, the core SMILES structural string remains unchanged, but the side chains will undergo isomerism. Therefore, it is necessary to calculate the physical properties of different isomers for side chain isomers, and then average the physical properties of all isomers to convert them into the physical properties of a certain SOL structural string.

[0180] In this embodiment, the sixth step is to directly establish a physical property prediction model based on big data algorithms using feature data such as molecular composition and structure.

[0181] By parsing the MOL file of a molecule, basic information such as the number of carbon atoms, degree of unsaturation, number of rings, number of heteroatoms, and number of branches can be obtained. At the same time, basic physical properties such as boiling point and density, which are helpful for predicting other physical properties, are obtained from authoritative data sources to assist in building a physical property prediction model. The real data of the physical properties are obtained from authoritative experimental data sources to verify the accuracy of the model.

[0182] In this embodiment, after collecting relevant data from approximately 3,000 molecules, the data was divided into a training set and a test set in a 7:3 ratio, ensuring the sufficiency and representativeness of the data. The training set was then input into a neural network model for training. During the training process, the network structure of the model was adjusted, and the optimal parameters of the model were obtained through a 10-fold cross-validation algorithm, resulting in the final property prediction model.

[0183] The molecular properties obtained through the property prediction model are combined with the property results calculated based on the group contribution method and obtained through isomer property correction to correct the molecular properties and finally obtain the molecular property prediction results.

[0184] Please refer to Figure 7, which is a schematic diagram of the prediction effect of the property prediction model provided in the embodiment of the present invention.

[0185] In this embodiment, taking the standard Gibbs free energy as an example, the prediction effect of the property prediction model provided by this embodiment is shown in Figure 7. In the figure, the horizontal axis is the true value, the vertical axis is the predicted value, and the 45° red line represents the case where the true value and the predicted value are equal.

[0186] This embodiment utilizes the coefficient of determination R. 2 To quantify the model's predictive performance, the closer the data points are to the red line, the higher the R-value. 2 The closer the value is to 1, the better the model's predictive performance; as shown in Figure 7, the coefficient of determination R of the property prediction model in this embodiment... 2 =0.9869, indicating that the property prediction model in this embodiment has a good prediction effect.

[0187] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0188] This invention estimates the physical properties of crude oil molecules based on the group types and numbers in the obtained crude oil molecule library. The estimated properties are then corrected based on several types of isomer properties. A property prediction model is then constructed to obtain another property estimate for the crude oil molecules. Finally, the corrected estimated properties and the other property estimate structure are used to obtain the final property prediction result for the crude oil molecules. This invention combines three methods to predict the physical properties of crude oil molecules, enabling accurate prediction. The accurate prediction results facilitate the application of crude oil molecule analysis technology to downstream process simulation calculations, further guiding the refined production of crude oil.

[0189] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for predicting physical property data for process simulation, characterized in that, include: Obtain the group types and numbers of crude oil molecules in the crude oil molecule library; The physical properties of the crude oil molecules are estimated based on the group type and number of groups, and a first physical property result is obtained. Obtain several types of physical properties of the isomers, and modify the first physical property result based on the several types of physical properties of the isomers to obtain the second physical property result; Acquire feature data of several first molecules, and construct a property prediction model based on the feature data of the several first molecules; The characteristic data of the crude oil molecules are obtained and used as input data for the physical property prediction model. The third physical property result of the crude oil molecules is obtained through the physical property prediction model. Based on the second and third property results, the predicted properties of the crude oil molecules are obtained.

2. The method for predicting physical property data for process simulation as described in claim 1, characterized in that, Also includes: Define the group types of several groups, and statistically obtain the physical properties of several second molecules, as well as the group types and number of groups in each molecule; Based on the physical properties of the aforementioned second molecules and the group type and number of groups in each molecule, the contribution value of each group type to the physical properties is obtained; The contribution value of each group type to the physical properties is stored in the crude oil physical property database; wherein, the largest molecule in the crude oil physical property database is C100.

3. The method for predicting physical property data for process simulation as described in claim 2, characterized in that, The acquisition of the group types and numbers of crude oil molecules in the crude oil molecular library includes: Obtain the SOL format string of crude oil molecules in the crude oil molecule library, and perform standardization processing on the SOL format string to obtain the SMILES format string of crude oil molecules; Convert the SMILES format string of the crude oil molecule into an MOL file of the crude oil molecule; The SMARTS method is used to describe the several groups, and the SMARTS strings of the several groups are obtained; Convert the SMARTS strings of the aforementioned groups into MOL files for the groups; By comparing the MOL files of the crude oil molecules with the MOL files of the functional groups, the functional group types and numbers of crude oil molecules in the crude oil molecule library can be obtained.

4. The method for predicting physical property data for process simulation as described in claim 1, characterized in that, The step of estimating the physical properties of the crude oil molecules based on the group type and number of groups, and obtaining a first physical property result, includes: A preliminary physical property estimate of the crude oil molecules is performed using a physical property calculation formula, which is as follows: f=∑ni*Gi+a In the above formula, ni represents the number of functional groups, Gi represents the contribution value of functional group properties, and a is a preset parameter.

5. The method for predicting physical property data for process simulation as described in claim 1, characterized in that, The step of acquiring feature data of several first molecules and constructing a property prediction model based on the feature data of the several first molecules includes: Acquire characteristic data of several first molecules, the true values ​​of preset physical properties of several first molecules, and basic physical properties; wherein, the characteristic data includes the number of carbons, degree of unsaturation, number of rings, number of heteroatoms, and number of branches of each molecule; the basic physical properties include density and boiling point; Obtain a preset neural network model and initialize the weights of the preset neural network model; The feature data and basic physical properties of the aforementioned first molecules are used as input data for the preset neural network model, and output prediction data is obtained through the preset neural network model. The error is obtained based on the true values ​​of the preset physical properties of the plurality of first molecules and the output prediction data, and the weights of the preset neural network model are adjusted according to the error; this step is repeated until the error is less than the preset value. By cross-validation, the optimal parameters of the preset neural network model are determined, and the physical property prediction model is constructed.

6. A device for predicting physical property data for process simulation, characterized in that, include: The module includes an acquisition module, an estimation module, a correction module, a construction module, a model prediction module, and a property acquisition module. The acquisition module is used to acquire the group type and number of groups of crude oil molecules in the crude oil molecule library; The estimation module is used to estimate the physical properties of the crude oil molecules based on the group type and number of groups, and obtain a first physical property result. The correction module is used to obtain several types of physical properties of the isomers, and to correct the first physical property result based on the several types of physical properties of the isomers to obtain the second physical property result; The construction module is used to acquire feature data of several first molecules and construct a property prediction model based on the feature data of the several first molecules; The model prediction module is used to acquire the feature data of the crude oil molecules and use the feature data of the crude oil molecules as input data of the physical property prediction model to obtain the third physical property result of the crude oil molecules through the physical property prediction model. The property acquisition module is used to obtain the property prediction results of the crude oil molecules based on the second property result and the third property result.

7. The predictive device for physical property data in process simulation as described in claim 6, characterized in that, Also includes: Contribution value acquisition module; the contribution value acquisition module is used for: Define the group types of several groups, and statistically obtain the physical properties of several second molecules, as well as the group types and number of groups in each molecule; Based on the physical properties of the aforementioned second molecules and the group type and number of groups in each molecule, the contribution value of each group type to the physical properties is obtained; The contribution value of each group type to the physical properties is stored in the crude oil physical property database; wherein, the largest molecule in the crude oil physical property database is C100.

8. The predictive device for physical property data in process simulation as described in claim 7, characterized in that, The acquisition module includes: a processing unit, a first conversion unit, a description unit, a second conversion unit, and a comparison unit; The processing unit is used to obtain the SOL format string of crude oil molecules in the crude oil molecule library, and to perform standardization processing on the SOL format string to obtain the SMILES format string of crude oil molecules. The first conversion unit is used to convert the SMILES format string of the crude oil molecule into an MOL file of the crude oil molecule; The description unit is used to describe the plurality of groups using SMARTS and obtain the SMARTS strings of the plurality of groups; The second conversion unit is used to convert the SMARTS strings of the plurality of groups into MOL files of the groups; The comparison unit is used to obtain the group type and number of crude oil molecules in the crude oil molecule library by comparing the MOL file of the crude oil molecule and the MOL file of the group.

9. The predictive device for physical property data in process simulation as described in claim 6, characterized in that, The estimation module is used to estimate the physical properties of the crude oil molecules based on the group type and number of groups, and obtain a first physical property result, including: A preliminary physical property estimate of the crude oil molecules is performed using a physical property calculation formula, which is as follows: f=∑ni*Gi+a In the above formula, ni represents the number of functional groups, Gi represents the contribution value of functional group properties, and a is a preset parameter.

10. The predictive device for physical property data in process simulation as described in claim 6, characterized in that, The construction module includes: an acquisition unit, an initialization unit, an output unit, an adjustment unit, and a construction unit; The acquisition unit is used to acquire characteristic data of several first molecules, the true values ​​of preset physical properties of several first molecules, and basic physical properties; wherein, the characteristic data includes the number of carbons, degree of unsaturation, number of rings, number of heteroatoms, and number of branches of each molecule; the basic physical properties include density and boiling point; The initialization unit is used to obtain a preset neural network model and initialize the weights of the preset neural network model; The output unit is used to take the feature data and basic physical properties of the plurality of first molecules as input data of the preset neural network model, and obtain output prediction data through the preset neural network model. The adjustment unit is used to obtain the error based on the true values ​​of the preset physical properties of the plurality of first molecules and the output prediction data, and adjust the weights of the preset neural network model according to the error; repeat this step until the error is less than the preset value; The construction unit is used to determine the optimal parameters of the preset neural network model through cross-validation, thereby completing the construction of the physical property prediction model.

Images