Petrochemical production stream molecular-level composition prediction method and system based on multiple models

Abstract

The invention relates to a multi-model-based petrochemical production stream molecular-level composition prediction method and system, and relates to the technical field of oil refining chemical molecule management, and the method comprises the steps: measuring the molecular composition of a raw material, splicing structural units based on a structural unit-bond electric matrix frame, and constructing a molecular digital structure topology; constructing a molecular composition model by combining a preset molecular database, a probability density function and a molecular property predictor; establishing a molecular-level reaction network based on a catalytic reforming reaction mechanism, and constructing a molecular-level reaction kinetic model in combination with a reaction rate expression; constructing a reactor model according to mass, energy and momentum transfer equations; inactivation and separation unit models are constructed, multiple models are coupled in sequence, a whole-process reaction-separation molecular level prediction model is generated, and any stream molecular level composition is predicted. The method solves the problem that traditional petrochemical production depends on measurement results to adjust process parameters and has operation lag, and can eliminate lag and improve production control precision by predicting stream molecular composition.

Description

Technical Field

[0001] This application relates to the field of molecular management in petrochemical refining and chemical engineering, and in particular to a method and system for predicting the molecular-level composition of petrochemical production streams based on multiple models. Background Technology

[0002] As the petrochemical industry upgrades towards molecular-level management and intelligentization, accurately grasping the molecular composition of any stream in production has become a key technological requirement for equipment control and optimization.

[0003] Currently, traditional petrochemical production operation methods rely solely on measurement results to adjust process parameters, resulting in operational lag. This not only leads to excessively long production adjustment cycles but also easily causes the generation of substandard products. Consequently, production efficiency and product qualification rates are reduced, and the difficulty of process optimization is increased, making it difficult to meet the refined and real-time control requirements of petrochemical production. Summary of the Invention

[0004] This application provides a method and system for predicting the molecular-level composition of petrochemical production streams based on multiple models. This improves upon the shortcomings of traditional petrochemical production operations, which rely solely on measurement results to adjust process parameters, resulting in operational lag and consequently, excessively long production adjustment cycles or the production of substandard products.

[0005] The embodiments of this application disclose the following technical solutions: In a first aspect, embodiments of this application provide a method for predicting the molecular-level composition of petrochemical production streams based on multiple models, the method comprising: According to the preset detection scheme, the raw materials in oil refining and chemical production are measured to obtain the molecular composition of the raw materials. Based on the structural unit-bond electric matrix framework, the molecular composition of the raw materials is spliced ​​into structural units to construct a molecular digital structure topology. Based on the pre-set molecular database and the digital molecular structure topology, molecular properties are predicted by combining the probability density function and the molecular property predictor, and a molecular composition model is constructed. A molecular-level reaction network was established based on the catalytic reforming reaction mechanism, and a molecular-level reaction kinetic model was constructed by combining the reaction rate expression of the reaction system. Based on the mass transfer equation, energy transfer equation, and momentum transfer equation, a reactor model is constructed for simulation modeling of reactors in oil refining and chemical production. A deactivation model and a separation unit model are constructed, and the molecular composition model, molecular-level reaction kinetics model, reactor model, deactivation model and separation unit model are connected and coupled in sequence to generate a full-process reaction-separation molecular-level prediction model, which can predict the molecular-level composition of any stream in oil refining and chemical production.

[0006] Secondly, embodiments of this application provide a multi-model-based prediction system for the molecular-level composition of petrochemical production streams, the system comprising: The raw material molecule digitization module is used to determine the raw material molecular composition in oil refining and chemical production according to a preset detection scheme. Based on the structural unit-bond electric matrix framework, the raw material molecular composition is spliced ​​into structural units to construct a molecular digital structure topology. The molecular composition modeling module is used to predict molecular properties and construct a molecular composition model based on a pre-set molecular database and the digital structure topology of the molecules, combined with a probability density function and a molecular property predictor. The reaction kinetics modeling module is used to establish molecular-level reaction networks based on the catalytic reforming reaction mechanism and to construct molecular-level reaction kinetic models by combining the reaction rate expressions of the reaction system. The reactor simulation modeling module is used to simulate and model reactors in oil refining and chemical production based on the mass transfer equation, energy transfer equation, and momentum transfer equation, and to build reactor models. The full-process coupled prediction module is used to construct the deactivation model and the separation unit model, and connect and couple the molecular composition model, molecular-level reaction kinetics model, reactor model, deactivation model and separation unit model in sequence to generate a full-process reaction-separation molecular-level prediction model, which can predict the molecular-level composition of any stream in oil refining and chemical production.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages: This application proposes a multi-model-based method and system for predicting the molecular-level composition of petrochemical production streams. It achieves accurate prediction of the molecular-level composition of any stream in refining and chemical processes by constructing a molecular composition model, a molecular-level reaction kinetics model, a reactor model, a deactivation model, and a separation unit model in a step-by-step manner and coupling them sequentially. First, the molecular composition of the feedstock is detected, and a digital molecular structure topology is constructed using structural unit-bond electrical matrix. This is combined with a pre-set molecular database, probability density function, and molecular property predictor to build the molecular composition model. Second, a molecular-level reaction network is constructed based on the catalytic reforming mechanism, and a molecular-level reaction kinetics model is built using the reaction rate expression. A reactor model is constructed by collecting structural and operational parameters according to mass, energy, and momentum transfer equations and matching equations to reactor type. Simultaneously, a deactivation model is constructed based on catalyst carbon deposition patterns, and a separation unit model is built using the SRK / PR equation and flash-distillation column algorithm. Finally, a full-process reaction-separation molecular-level prediction model is generated by sequentially coupling the molecular composition model, molecular-level reaction kinetics model, reactor model coupled with the deactivation model, and separation unit model, enabling the prediction of the molecular-level composition of any stream in refining and chemical production.

[0008] The technical solution of this application integrates steps such as digital analysis of raw material molecular composition, molecular-level reaction quantification, reactor simulation modeling, catalyst deactivation correction, molecular prediction of separation units, and multi-model coupling. It solves the problem that traditional petrochemical production stream prediction only focuses on macroscopic components, ignores molecular-level transformation laws, and relies on lagging measurement data to adjust the process. It realizes accurate tracking and prediction of molecular-level composition throughout the entire process, eliminates the lag in process adjustment, and improves the accuracy of production control. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 A flowchart illustrating the method for predicting the molecular-level composition of petrochemical production streams based on a multi-model approach provided in this application embodiment; Figure 2 A schematic diagram illustrating the process of constructing a molecular composition model provided in the embodiments of this application; Figure 3 A schematic diagram of the process for constructing the full-process reaction-separation molecular-level prediction model provided in the embodiments of this application; Figure 4 This is a schematic diagram of the structure of a multi-model-based prediction system for the molecular-level composition of petrochemical production streams provided in an embodiment of this application.

[0011] The components represented by each number in the attached diagram are explained below: Raw material molecule digitization module 01, molecular composition modeling module 02, reaction kinetics modeling module 03, reactor simulation modeling module 04, and whole-process coupled prediction module 05. Detailed Implementation

[0012] This application provides a method and system for predicting the molecular-level composition of petrochemical production streams based on multiple models. This method addresses the technical problems in existing petrochemical production operations that rely solely on measurement results to adjust process parameters, resulting in operational lag and excessively long production adjustment cycles or the production of substandard products.

[0013] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0014] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0015] In the description of this application, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0016] Example 1, as shown in the appendix Figure 1 As shown, this application provides a method for predicting the molecular-level composition of petrochemical production streams based on multiple models. The method includes the following steps: S110: The raw materials in oil refining and chemical production are measured according to the preset detection scheme to obtain the molecular composition of the raw materials. The molecular composition of the raw materials is spliced ​​with structural units based on the structural unit-bond electric matrix framework to construct a molecular digital structure topology. In this embodiment of the application, in the scenario of molecular-level simulation of oil refining and chemical processes, in order to accurately obtain basic information on raw material molecules and achieve standardized characterization, it is necessary to measure the molecular composition of raw materials through multiple instruments and perform structured splicing to provide reliable data support for subsequent molecular composition model construction and full-process prediction.

[0017] Specifically, based on the characteristics of refining and chemical raw materials, a preset detection scheme is determined, with gas chromatography and infrared spectroscopy as the core. The raw materials are then repeatedly measured to ensure that the obtained data covers the true distribution of the raw material's molecular composition.

[0018] Furthermore, a comprehensive evaluation of several measurement results was conducted, and abnormal data was eliminated before integrating the data to obtain comprehensive and accurate information on the molecular composition of the raw materials.

[0019] Furthermore, according to the structural unit theory, the various molecules in the raw material molecular composition are broken down into basic structural units such as carbon chains and ring structures. Then, combined with the molecular chemical properties and bond-electric matrix connection rules, the basic structural units are orderly spliced ​​together to form multiple complete molecular structures that conform to the actual chemical structure, ensuring the authenticity and rationality of the molecular structure.

[0020] Finally, based on the structural unit-bond-electric matrix framework, the obtained complete molecular structures are digitally converted, and information such as the structural features and bonding relationships of the molecules are transformed into a standardized digital matrix form to generate the digital molecular structural topology.

[0021] This step, through a technical solution of "precise measurement - structured assembly - digital characterization," realizes the transformation of raw material molecules from physical analysis to digital carriers, providing unified and calculable basic data for subsequent steps such as molecular property prediction and reaction network construction.

[0022] Step S110 in the method provided in this application embodiment includes: The raw materials in oil refining and chemical production were measured several times using a gas chromatograph and an infrared spectrometer. The molecular composition of the raw materials was obtained by comprehensively evaluating the results of the several measurements. According to the structural unit theory, the structural units in the molecular composition of the raw materials are spliced ​​together, and combined with chemical properties and bond-electric matrix connection rules to form complete molecules, thereby obtaining multiple complete molecular structures. Based on the structural unit-bond electric matrix framework, the multiple complete molecular structures are digitally represented to generate a digital molecular structural topology.

[0023] In this embodiment of the application, in order to improve the accuracy of predicting the molecular-level composition of petrochemical production streams, it is necessary to first obtain reliable basic data of raw material molecules and complete standardized digital characterization, so as to provide core data support for subsequent molecular composition model construction, reaction network construction and full-process simulation.

[0024] First, the molecular composition of raw materials used in oil refining and chemical production was determined. Considering the complex molecular structure and diverse components of oil refining and chemical raw materials, and the susceptibility of single detection to instrument errors and fluctuations in the operating environment, a detection scheme combining gas chromatography-FID (GC-FID) and infrared spectroscopy was adopted to perform three repeated measurements on the raw materials in order to avoid random errors from a single detection.

[0025] Among them, GC-FID can accurately quantify the molar percentage of different hydrocarbon components in raw materials, while infrared spectroscopy can help identify functional groups in molecules, and the measurement data of the two complement each other.

[0026] Furthermore, after the determination of the molecular composition of the raw materials is completed, outliers are removed from the three test results, and then the arithmetic mean is calculated for comprehensive evaluation, finally obtaining the molecular composition of the raw materials covering multiple components such as alkanes, cycloalkanes, and aromatics.

[0027] For example, taking the testing of straight-run naphtha feedstock as an example, in three GC-FID tests, the first test showed a molar fraction of 7.35% for n-pentane, 4.98% for methylcyclohexane, and 2.36% for toluene; the second test showed 7.40% for n-pentane, 4.95% for methylcyclohexane, and 2.40% for toluene; and the third test showed 7.36% for n-pentane, 4.95% for methylcyclohexane, and 2.38% for toluene. The deviation of the three data was less than 1%, and there were no outliers.

[0028] Furthermore, after calculating the arithmetic mean, the molecular composition of the three components in the raw material was finally determined to be 7.37% for n-pentane ((7.35%+7.40%+7.36%) / 3=7.37%), 4.96% for methylcyclohexane ((4.98%+4.95%+4.95%) / 3=4.96%), and 2.38% for toluene ((2.36%+2.40%+2.38%) / 3=2.38%).

[0029] At the same time, combined with the results of infrared spectroscopy, additional cycloalkanes such as 2,6-dimethyloctane and aromatic hydrocarbons such as m-xylene were identified in the raw materials, forming complete raw material molecular composition data covering alkanes, cycloalkanes, and aromatic hydrocarbons.

[0030] Furthermore, based on the structural unit theory, the various molecular components in the above-mentioned raw material molecular composition are broken down into basic structural units, such as the alkane n-pentane (C5H) 12 ) is broken down into 5 -CH2- carbon chain units and 2 -CH3 terminal units, and n-decane (C 10 H 22 It is broken down into 10 consecutive -CH2- carbon chain units.

[0031] Furthermore, by combining molecular chemical properties with bond-electric matrix connection rules (i.e., characterizing the bond energy and charge distribution between different atoms through matrix elements), the basic structural units are assembled in an orderly manner according to the rationality of chemical bonds.

[0032] For example, six -CH2- units are spliced ​​together to form a six-membered ring, and one -CH3 unit is attached to any carbon atom of the six-membered ring to form the complete molecular structure of methylcyclohexane; eight -CH2- units and two -CH3 units are spliced ​​together to form the straight-chain structure of n-decane.

[0033] Furthermore, the splicing method described above is repeated to obtain multiple complete molecular structures corresponding to all components in the raw materials, while ensuring that each molecular structure conforms to the actual chemical bonding rules and has no unreasonable bonding situations.

[0034] Finally, a hybrid framework of structural unit-bond electrical matrix is ​​introduced to realize the transformation of the complete molecular structure into a standardized digital form, providing a computable digital carrier for subsequent model calculations.

[0035] Specifically, each complete molecular structure is first assigned a unique structural unit code, and then a bond-electron matrix is ​​constructed, that is, the atoms in the molecule are the rows / columns of the matrix, and the matrix element values ​​represent the bond type (1 represents a single bond, 2 represents a double bond) and electronegativity differences between atoms.

[0036] Finally, by using the method of "structural unit encoding + bond-electric matrix construction", the structural features, bonding relationships, charge distribution and other information of each complete molecule are transformed into standardized digital codes and matrix forms, generating the molecular digital structure topology.

[0037] For example, regarding methylcyclohexane (C7H) in the feedstock 14 First, assign structural unit codes to them: the six-membered ring structural unit is coded as "R6-01", and the methyl substituent unit is coded as "M1-03".

[0038] Furthermore, a bond-electric matrix is ​​constructed, consisting of 7 C atoms (numbered C1-C7) and 14 H atoms (numbered H1-H2) in the molecule. 14 The matrix is ​​divided into rows and columns. Due to the six-membered ring structure, the corresponding matrix element values ​​between C1-C6 atoms are all 1 (carbon-carbon single bond). The corresponding element values ​​between C1 and C7 (methyl C atoms) are 1 (carbon-carbon single bond). The corresponding element values ​​between each C atom and the connected H atom are all 1 (carbon-hydrogen single bond). The electronegativity difference between C and C is also marked (0) and the electronegativity difference between CH is (0.35).

[0039] Through the above steps, the ring structure characteristics, substituent positions, bonding types, and electronegativity information of methylcyclohexane are all converted into digital codes and matrices, generating its unique molecular digital structure topology.

[0040] Similarly, the same operation is performed on all component molecules in raw materials such as n-pentane, toluene, and 2,6-dimethyloctane, ultimately forming a set of molecular digital structure topologies covering all raw material components, providing accurate digital foundational data for subsequent steps such as molecular composition model construction and reaction network matching.

[0041] S120: Based on the pre-set molecular database and the digital structure topology of the molecules, molecular properties are predicted by combining the probability density function and the molecular property predictor, and a molecular composition model is constructed. In this embodiment of the application, in the scenario of predicting the molecular-level composition of petrochemical production streams, in order to transform the digital molecular structure topology into accurate molecular composition data, it is necessary to rely on a pre-set molecular database to carry out initial matching, introduce a probability density function to optimize the initial composition, and use a molecular property predictor to correct and optimize, so as to solve the problems of initial matching deviation and insufficient composition prediction accuracy.

[0042] Specifically, the initial matching work is carried out first based on a pre-built molecular database. This pre-built molecular database covers digital topological templates of various types of molecules commonly found in the oil refining and chemical industry, such as alkanes, cycloalkanes, and aromatics. Each template is associated with corresponding standard molecular composition information.

[0043] Furthermore, the generated digital structure topology of the raw material molecules is used as a search condition. The most suitable digital topology template is searched in the pre-set molecular database using the topological feature cosine similarity algorithm. When the feature similarity is ≥95%, the standard composition interval is obtained by association and integration to obtain the initial molecular composition.

[0044] Furthermore, to address the matching deviation of the initial molecular composition, a Gaussian distribution function is introduced as the probability density function for optimization. The mole fraction of the initial molecular composition is used as the mean, and the standard deviation is calculated by combining the test data of multiple batches of raw materials in the past. The component distribution probability is simulated, and high probability fraction combinations with a probability ≥99% are screened to reduce the range of data fluctuation.

[0045] Meanwhile, during the optimization process, the probability density function is corrected using a molecular property predictor. The candidate composition is input into the predictor to obtain the predicted properties, which are then compared with the measured properties of the raw materials to calculate the error. If the error is ≥1%, the function parameters are adjusted in reverse, and the operation is repeated until the prediction error is <1%, thus obtaining the optimized molecular composition and constructing a molecular composition model.

[0046] This step, through a technical solution of "database matching - function optimization - predictor correction," achieves the transformation of molecular composition from initial matching to precise optimization, providing basic data support for subsequent molecular-level reaction network construction and full-process simulation.

[0047] As attached Figure 2 As shown, step S120 in the method provided in this application embodiment includes: Based on a pre-built molecular database, the initial molecular composition is obtained by topological matching of the digital molecular structure. The initial molecular composition is optimized using a probability density function. During the optimization process, the probability density function is corrected using a molecular property predictor until the prediction error is less than 1%, thus obtaining the optimized molecular composition and constructing a molecular composition model. The molecular property predictor is built based on deep learning and trained to convergence using sample data.

[0048] In this embodiment of the application, in order to transform the digital molecular structure topology into accurate molecular composition data that can support subsequent modeling, it is necessary to use a hierarchical process of database matching, function optimization, and predictor correction to solve the problems of deviation between the initial matching data and the actual distribution of raw material molecules, and the difficulty of simply optimizing the function to take into account the consistency of macroscopic molecular properties, and finally construct a molecular composition model that conforms to actual production.

[0049] First, we conduct initial molecular composition matching based on a pre-built molecular database to provide a basic data framework for subsequent optimization.

[0050] Among them, the pre-built molecular database is a professional database built based on molecular research results and production measurement data in the oil refining and chemical industry. It covers digital topological templates of various molecules such as alkanes, cycloalkanes, and aromatics commonly found in typical feedstocks such as straight-run naphtha and catalytic cracking gasoline. Each template is associated with corresponding standard molecular composition information, including key parameters such as the conventional molar fraction range of the molecule in the same type of feedstock and the proportion of characteristic functional groups.

[0051] Specifically, when performing molecular composition matching, the obtained digital structural topology of the raw material molecules is used as the core search condition. The topological feature cosine similarity algorithm is used to match the topology to be matched with each template in the database. That is, the feature similarity between the topology to be matched and each template in the database is calculated. When the feature similarity is ≥95%, it is determined to be a suitable template, and the standard molecular composition information associated with the template is extracted.

[0052] If multiple templates with similar features exist (e.g., feature similarity of 96% and 95.5% respectively), further screening is conducted based on the basic properties of the raw materials to ultimately determine the unique suitable template.

[0053] Furthermore, by integrating the information of all successfully matched molecules and their composition, an initial molecular composition is obtained to clarify the basic distribution range of various molecules in the raw materials, providing an initial data boundary for subsequent probability density function optimization.

[0054] For example, taking straight-run naphtha feedstock as an example, after matching, the initial molecular composition includes alkane (7.2%-7.5% molar fraction), n-hexane (5.1%-5.4% molar fraction), and n-heptane (3.8%-4.1% molar fraction), cycloalkanes (3.5%-3.8% molar fraction) and methylcyclohexane (4.8%-5.0% molar fraction), and aromatics (1.2%-1.5% molar fraction) and toluene (2.3%-2.5% molar fraction). The composition range of all molecules corresponds to the standard information of the straight-run naphtha template in the pre-set molecular database, forming a complete initial molecular composition list of the feedstock.

[0055] Furthermore, the initial molecular composition is optimized using a probability density function to reduce the fluctuation range of the initial data and improve the rationality of the composition distribution.

[0056] In particular, considering that the initial molecular composition is a conventional range value obtained by matching a pre-set molecular database template, while the molecular distribution of different batches of raw materials in actual production has slight differences, for example, the n-pentane fraction of straight-run naphtha purchased by the same refinery at different times may fluctuate by 0.2%-0.3% due to different crude oil origins, it is necessary to simulate the distribution law of the actual molecular composition through probability density function.

[0057] Specifically, the Gaussian distribution function is selected as the probability density function, and the average mole fraction of each molecule in the initial molecular composition is taken as the distribution center. The standard deviation is calculated by combining the test data of multiple batches of the raw material in the past, and the probability density curve of the mole fraction of each molecule is constructed.

[0058] Furthermore, by analyzing the probability density curve of the mole fraction, a fraction interval with a probability ≥ 99% was selected. The molecular composition within this fraction interval not only conforms to the conventional rules of the database but also closely matches the fluctuation range of actual production, thus obtaining a preliminary set of optimized candidate molecular compositions.

[0059] For example, taking n-pentane and methylcyclohexane in the initial molecular composition of straight-run naphtha as examples, the average initial mole fraction of n-pentane is 7.35%, and the standard deviation calculated based on the data from 120 previous batches is 0.12%. In the constructed mole fraction probability density curve, the fraction range with a probability ≥99% is 7.35% ± 0.15% (i.e., 7.20%-7.50%). The average initial mole fraction of methylcyclohexane is 4.9%, with a standard deviation of 0.08%, corresponding to a fraction range with a probability ≥99% of 4.9% ± 0.1% (i.e., 4.80%-5.00%). After integrating the fraction ranges of all molecules, a preliminary optimized candidate molecular composition set is obtained, such as n-pentane 7.22%-7.48%, methylcyclohexane 4.81%-4.99%, and toluene 2.32%-2.48%, etc.

[0060] Furthermore, the probability density function is iteratively corrected using a molecular property predictor to ensure that the optimized molecular composition accurately reflects the macroscopic properties of the raw materials, thus avoiding the problem of reasonable composition but deviating properties.

[0061] Among them, the molecular property predictor is a prediction model built based on deep learning. By analyzing the input molecular composition data, it explores the nonlinear correlation between molecular composition and macroscopic properties, so as to output macroscopic property prediction values ​​that match the actual characteristics of the raw materials, and must meet the accuracy requirement of prediction error (|predicted value - measured value| / measured value × 100%) < 1%.

[0062] Specifically, the molecular property predictor is based on a fully connected neural network framework, which includes a three-level structure of "input layer-hidden layer-output layer".

[0063] The number of neurons in the input layer is consistent with the number of key molecular types in the raw material molecular composition, and is used to receive the standardized molecular mole fraction data. The hidden layer has 3 layers, each with 128 neurons, and uses the ReLU activation function to enhance the model's ability to extract complex correlation features through nonlinear transformation, so as to avoid the gradient vanishing problem during training.

[0064] Meanwhile, a Dropout layer (with a dropout probability of 0.2) is added after each hidden layer to prevent the model from overfitting; the number of neurons in the output layer corresponds to the types of macroscopic properties to be predicted, and a linear activation function is used to directly output the normalized macroscopic property prediction values, which are then restored to the actual physical quantity units through an inverse normalization operation.

[0065] During the model training phase, the first step is to construct a training sample set, which involves collecting sample data from multiple typical raw materials in the refining and chemical industry. Each sample set includes detailed molecular composition and measured values ​​of macroscopic properties.

[0066] Furthermore, the sample data is preprocessed to remove outlier samples, and then the sample set is divided into a training set and a validation set in an 8:2 ratio to ensure that the two sets of data are consistent in terms of raw material type, molecular composition range, and macroscopic property distribution, thus ensuring the model's generalization ability.

[0067] Meanwhile, during training, the root mean square error (RMSE) between predicted and measured properties is used as the loss function, and the Adam optimizer is used to update the network parameters. The initial learning rate is set to 0.001, and the learning rate is reduced to half of the previous rate every 100 iterations to balance training speed and convergence accuracy.

[0068] In addition, the model performance is evaluated using a validation set every 10 iterations, and the RMSE trend is recorded. When the RMSE of the validation set decreases by less than 0.005 for 20 consecutive iterations, and the macroscopic property prediction error of all samples is stable at less than 1.5%, the model is considered to have converged, training is stopped, and a usable molecular property predictor is obtained.

[0069] For example, the measured density of a group of straight-run naphtha samples was 0.655 g / cm³. 3 The predicted values ​​output by the molecular property predictor stabilized at 0.654-0.656 g / cm³. 3 The error is less than 0.3%, which meets the prediction accuracy standard.

[0070] Furthermore, after the molecular property predictor is constructed, the probability density function is iteratively corrected based on the molecular property predictor to ensure that the optimized molecular composition not only conforms to the probability density function screening range with a probability ≥ 99%, but also accurately matches the macroscopic properties of the raw materials.

[0071] Specifically, firstly, a representative set of compositional data is randomly selected from the candidate molecular composition set obtained by probability density function optimization. For example, in straight-run naphtha feedstock, a combination of n-pentane 7.36%, methylcyclohexane 4.92%, toluene 2.38%, and octane 3.25% is used. After standardization, this data is input into the molecular property predictor to obtain the macroscopic property prediction results corresponding to this composition, such as a predicted density of 0.650 g / cm³. 3 Boiling point at normal pressure: 82.2℃; Refractive index: 1.423.

[0072] Furthermore, laboratory-measured macroscopic property data of this batch of straight-run naphtha feedstock were retrieved, for example, the density at 20°C was measured to be 0.655 g / cm³ using a densitometer. 3 The boiling point at normal pressure was measured to be 83.0-83.8℃ using a distillation apparatus (average 83.4℃), and the refractive index was measured to be 1.426 using a refractometer.

[0073] Meanwhile, the prediction errors for each type of macroscopic property were calculated: density error was |0.650-0.655| / 0.655×100%≈0.76%, boiling point error was |82.2-83.4| / 83.4×100%≈1.44%, and refractive index error was |1.423-1.426| / 1.426×100%≈0.21%.

[0074] Among them, since the boiling point error (1.44%) is ≥1%, the probability density function parameters need to be adjusted in reverse to correct the deviation.

[0075] Specifically, based on molecular characteristic analysis, the boiling point of the raw material is mainly positively correlated with the molar fraction of high carbon number alkanes. The predicted boiling point value is lower than expected, which is presumably because the proportion of high carbon number alkanes in the candidate composition is lower than the actual value.

[0076] Therefore, the mean of the probability density function distribution of octane was increased from 3.2% to 3.4%, and the standard deviation was increased from 0.1% to 0.12%, while the mean of the distribution of nonane was finely adjusted (from 2.1% to 2.2%), and the optimized candidate molecule composition set was regenerated.

[0077] Furthermore, new compositional data (7.37% n-pentane, 4.95% methylcyclohexane, 2.37% toluene, 3.38% octane) were selected from the adjusted candidate molecule composition set and input again into the molecular property predictor, yielding a new prediction result: density 0.653 g / cm³. 3Boiling point 83.1℃, refractive index 1.425.

[0078] Similarly, the errors are calculated by comparing the predicted results with the measured values: density error 0.31% (|0.653-0.655| / 0.655×100%≈0.31%), boiling point error 0.36% (|83.1-83.4| / 83.4×100%≈0.36%), refractive index error 0.07% (|1.425-1.426| / 1.426×100%≈0.07%). Since the errors of all molecular properties are <1%, the probability density function correction is considered complete.

[0079] Finally, the molecular composition that meets the error requirements after this correction (7.37% n-pentane, 4.95% methylcyclohexane, 2.37% toluene, 3.38% octane, etc.) is determined as the final optimized molecular composition. The optimized fractions of all molecules are integrated to complete the construction of the molecular composition model. This model not only meets the matching requirements of the molecular digital structure topology, but also has a high degree of consistency with the actual macroscopic properties of the raw materials. It can be used for subsequent molecular-level reaction network construction and whole-process flow prediction.

[0080] S130: A molecular-level reaction network is established based on the catalytic reforming reaction mechanism, and a molecular-level reaction kinetics model is constructed by combining the reaction rate expression of the reaction system. In this embodiment of the application, in order to accurately describe the reaction path and transformation law of raw material molecules in the catalytic reforming process, it is necessary to transform the reaction mechanism into a computable model framework through steps such as constructing a reaction rule library, building a molecular-level reaction network, and quantifying the kinetic model, so as to provide reaction simulation support for subsequent prediction of the molecular composition of the entire process stream.

[0081] Specifically, based on the catalytic reforming reaction mechanism in oil refining and chemical production, the reaction rules of the process are sorted out, and a reaction rule library containing several molecular-level reaction rules is constructed.

[0082] Among them, the catalytic reforming reaction mechanism includes typical reaction types such as dehydrogenation cyclization, isomerization, and hydrocracking. Each reaction rule must be precisely matched with the molecular structure characteristics and reaction conditions to ensure that the reaction rule can cover the core related elements of the molecular reaction.

[0083] Furthermore, a molecular-level reaction network is established based on the constructed reaction rule library.

[0084] Specifically, firstly, based on the obtained molecular composition model, the optimized molecular composition information in the model is extracted to determine the initial reactant set. Then, each molecule in the set is matched according to the reaction rules in the reaction rule library, and the corresponding product is identified when the reaction rules are met.

[0085] Furthermore, reactants and products are used as nodes, and reaction paths corresponding to the reaction rules are used as edges for association and integration to form a complete molecular-level reaction network.

[0086] Simultaneously, the molecular composition model and the molecular-level reaction network are coupled. After obtaining molecular composition information through the molecular composition model, each molecule in the information will be re-matched to the reaction according to the rules in the reaction rule library. If the composition of the raw material molecules is updated in the future, the molecular-level reaction network can adjust the initial node parameters synchronously and dynamically optimize the path allocation.

[0087] Finally, based on the structural unit-bond electrical matrix framework, and under computer-aided conditions, a molecular-level reaction kinetic model is constructed based on the reaction rate expression of the reaction system and the constructed molecular-level reaction network. The transformation relationship between molecular reactants and products in this molecular-level reaction network provides quantitative support for the prediction of the entire process flow.

[0088] Step S130 in the method provided in this application embodiment includes: A reaction rule library is constructed based on the catalytic reforming reaction mechanism in oil refining and chemical production. The reaction rule library includes several reaction rules established based on molecular-level reaction mechanisms. A molecular-level reaction network is established based on the reaction rule base. The molecular composition model and the molecular-level reaction network are coupled. After obtaining molecular composition information through the molecular composition model, each molecule in the molecular composition information will be matched according to the reaction rules in the reaction rule base, and react according to the reaction rules when the reaction rules are met. Based on the structural unit-bond electrical matrix framework, a molecular-level reaction kinetic model is constructed under computer-aided conditions, based on the reaction rate expression of the reaction system and the molecular-level reaction network. The molecular-level reaction kinetic model is used to quantify the transformation relationship between molecular reactants and products in the molecular-level reaction network.

[0089] In this embodiment of the application, in order to accurately simulate the reaction path and transformation law of feedstock molecules in the catalytic reforming process, it is necessary to construct a molecular-level reaction kinetic model based on the catalytic reforming reaction mechanism in oil refining and chemical production, and then transform the abstract reaction mechanism into a computable model system.

[0090] First, based on the catalytic reforming reaction mechanism in oil refining and chemical production, a reaction rule library is constructed to provide clear rule basis for molecular reaction matching.

[0091] Among them, the catalytic reforming reaction mechanism in oil refining and chemical production is based on the principle that molecular structure determines reaction activity. It covers typical reaction types such as dehydrogenation cyclization, isomerization, hydrocracking, and dehydrogenation aromatization. It is necessary to extract reaction rules based on the molecular level for each type of reaction and form a set of rules covering the main reaction pathways.

[0092] For example, the rules for dehydrogenation cyclization clearly state that straight-chain alkanes must have a carbon chain length of ≥6 carbon atoms, no quaternary carbon atoms, and can be converted into monocyclic aromatic hydrocarbons with the same number of carbon atoms under the conditions of 380-420℃, 1.0-1.5MPa, and platinum-rhenium bimetallic catalyst. The rules for isomerization stipulate that straight-chain alkanes (carbon chain ≥5 C) can be isomerized into isomers under the conditions of 350-390℃ and 1.2-1.6MPa, changing only the degree of branching of the molecular structure without changing the number of carbon atoms.

[0093] Furthermore, by systematically analyzing the molecular interaction rules of typical reactions such as dehydrogenation cyclization, isomerization, hydrocracking, and dehydrogenation aromatization, the molecular structure characteristics, reaction condition thresholds, and product correlations of each type of reaction are extracted, and a reaction rule library is constructed.

[0094] Furthermore, a molecular-level reaction network is established based on the constructed reaction rule library and coupled with a molecular composition model to construct a dynamic response molecular transformation framework.

[0095] Specifically, the optimized molecular composition information output by the molecular composition model is first used as the initial set of reactant nodes in the reaction network to clarify the molecular identifier and initial mole fraction of each node.

[0096] Furthermore, following the "molecule-reaction rule matching" approach, each initial reactant molecule is compared with all reaction rules in the reaction rule library for feature matching.

[0097] For example, taking n-heptane molecule as an example, first match the dehydrogenation cyclization reaction rules to check whether it meets the structural requirements of "carbon chain length ≥ 6, no quaternary carbon" and whether it is suitable for the preset reaction conditions (temperature 380-420℃, etc.). If it meets the requirements, it is determined that the molecule can generate toluene (C7-AR-001) through the dehydrogenation cyclization reaction, and the reaction path is recorded.

[0098] Furthermore, by matching the isomerization reaction rules, it was confirmed that it can be isomerized into branched alkanes such as 2-methylhexane and 3-methylhexane, forming two isomerization reaction pathways.

[0099] Finally, the hydrocracking reaction rules were matched. Since the carbon chain length of n-heptane is less than 10 C, it does not meet the structural requirements of the hydrocracking reaction. Therefore, this type of reaction pathway was excluded. In the end, n-heptane molecules formed a total of 3 potential reaction pathways.

[0100] Furthermore, after repeating the above matching process for all initial reactant molecules, the correlation between "reactant molecule-reaction pathway-product molecule" is integrated.

[0101] Specifically, using molecules as nodes (initial reactant nodes, product molecule nodes, and intermediate reactant nodes if the product molecule can further match the reaction rules), and reaction paths as directed edges (the direction of the edges is "reactant → product"), a complete molecular-level reaction network containing initial reactants, intermediate products, and final products is constructed.

[0102] At the same time, the molecular composition model is deeply coupled with the molecular-level reaction network to ensure that changes in molecular composition information can drive the adjustment of the reaction network in real time.

[0103] Specifically, when the molecular composition model updates the molecular composition information due to changes in raw material batches, the molecular-level reaction network will automatically update the mole fraction parameters of the initial reactant nodes synchronously and re-execute the "molecular-reaction rule matching" process, thereby dynamically adjusting the predicted trend of mole fraction of each intermediate product and final product node.

[0104] Furthermore, when new raw material molecular composition information is obtained through the molecular composition model, the molecular-level reaction network will automatically add the new molecules as initial reactant nodes, match the corresponding reaction rules, expand the reaction path, and realize the dynamic linkage between molecular composition and molecular-level reaction network to ensure the real-time performance and accuracy of reaction simulation.

[0105] For example, when the molecular composition model outputs a decrease in the molar fraction of n-heptane from 5.2% to 4.8% in a certain batch of straight-run naphtha feedstock, and a new n-octane (C8 straight-chain alkane) molecule (molar fraction 1.3%) is added, the molecular-level reaction network first synchronously updates the molar fraction of the n-heptane node to 4.8%. After rematching the rules, the flow rates of its dehydrogenation cyclization to toluene and isomerization to 2-methylhexane decrease as the fraction decreases, and the initial predicted value of the toluene node is reduced.

[0106] Simultaneously, by adding n-octane as a new initial reactant node, the dehydrogenation cyclization rule is matched to generate ethylbenzene (C8 monocyclic aromatic hydrocarbon), and the isomerization rule is matched to generate 2-methylheptane (C8 branched alkane), thus adding two new reaction pathways: "n-octane → ethylbenzene" and "n-octane → 2-methylheptane", achieving dynamic adaptation of the reaction network to the new molecular composition.

[0107] Furthermore, based on the structural unit-bond electrical matrix framework, and under computer-aided conditions, a molecular-level reaction kinetic model for quantifying molecular transformation relationships is constructed by combining the reaction rate expression of the reaction system.

[0108] Specifically, the first step is to digitally convert molecular information based on the structural unit-bond-electric matrix framework. This involves transforming the structural features of all nodes in the molecular-level reaction network into numerical parameters that the model can recognize: for alkane molecules, the number and connection methods of structural units such as methyl and methylene are extracted to generate structural unit codes; for cycloalkanes and aromatics, the ring structure type (such as six-membered rings and five-membered rings) and the position and number of side chains are additionally labeled.

[0109] Simultaneously, the bond-electric matrix of each molecule is constructed, and the electronegativity difference of each chemical bond in the molecule is recorded by the matrix elements. Then, the structural unit code and bond-electric matrix parameters are imported into the computer simulation platform to realize the numerical characterization of molecular structure.

[0110] Furthermore, reaction rate expressions derived from experimental data and reaction mechanisms are introduced. These reaction rate expressions are then integrated with the molecular information and molecular-level reaction networks that have been numerically characterized in a computer simulation platform to perform rate calculations.

[0111] Specifically, based on the molecular parameters and process conditions (temperature, pressure) corresponding to the reaction path, the reaction rate of each path is calculated in real time. Then, combined with the reaction time, differential equations are solved to quantify the conversion of reactants to products at different time points, ultimately forming a molecular-level reaction kinetic model.

[0112] For example, for the dehydrogenation cyclization reaction pathway of "n-octane → ethylbenzene", the rate expression corresponding to the reaction is first called, and then the numerical parameters and process conditions of n-octane are imported to calculate the reaction rate of the pathway as 0.0028 mol / (L·min).

[0113] Meanwhile, by solving the differential equation and combining the reaction time of 30 min, the amount of n-octane converted into ethylbenzene through this pathway within 30 minutes was quantified to be 0.084 mol / L, corresponding to a 1.8% decrease in the molar fraction of n-octane and a 1.8% increase in the molar fraction of ethylbenzene. The resulting molecular-level reaction kinetic model can accurately output such quantitative conversion results, realizing the quantitative simulation of conversion relationships in the molecular-level reaction network.

[0114] S140: Based on the mass transfer equation, energy transfer equation, and momentum transfer equation, simulate and model reactors in oil refining and chemical production, and construct reactor models. In this embodiment of the application, in order to accurately simulate the mass, energy and momentum transfer process in the oil refining and chemical reactor and to restore the real transformation environment of molecules in the reactor, it is necessary to adapt the transfer equation according to the reactor type and construct a reactor model that fits the actual industrial situation, so as to provide core reaction unit support for subsequent full-process molecular-level prediction.

[0115] Specifically, when constructing a reactor model, it is first necessary to clarify the type of reactor used in oil refining and chemical production.

[0116] In the method provided in this application embodiment, the reactor in the oil refining and chemical production is one of an axial fixed bed reactor, a radial fixed bed reactor, and a radial moving bed reactor.

[0117] Among them, different types of reactors have significantly different material flow characteristics and transfer laws. It is necessary to select appropriate mass transfer equations, energy transfer equations and momentum transfer equations to ensure that the equations can accurately describe the transfer behavior in the reactor.

[0118] For example, in a radial fixed-bed reactor, the material diffuses radially, and axial diffusion can be ignored; in an axial fixed-bed reactor, the material flows axially, and axial mass and heat transfer are the focus; in a radial moving-bed reactor, the influence of catalyst movement on the transfer process must also be considered.

[0119] Furthermore, for the selected reactor type, simulation modeling is performed based on the mass transfer equation, energy transfer equation, and momentum transfer equation.

[0120] The mass transfer equation describes the diffusion, reaction consumption, and generation processes of molecules within the reactor. For a radially fixed-bed reactor, neglecting axial diffusion, the mass transfer equation can be expressed as: ; In the formula, Components The molar flow rate (mol / s) reflects the macroscopic rate of material flow; The radial coordinates (m) of the reactor represent the spatial location of the material flow. The effective bed length (m) of the reactor is indicated at different radial positions; Catalyst packing density (mol / m³) 3 This quantifies the packing of catalyst within a unit volume of reactor, reflecting the tightness of the catalyst packing. Components reaction rate (mol / (m 3 The ·s) is provided by a molecular-level reaction kinetics model, reflecting the microscopic rate of molecular transformation.

[0121] This equation is expressed through radial coordinates. The derivative is used to calculate the composition. The variation of molar flow rate with radial position accurately characterizes the differences in concentration distribution of materials within a radial fixed-bed reactor caused by reaction and diffusion.

[0122] Furthermore, the energy transfer equation is used to describe heat generation, transfer, and temperature distribution within the reactor. For a radial fixed-bed reactor, the energy transfer equation takes the form: ; In the formula, The temperature (K) inside the reactor reflects the energy state of the reaction system. Components The molar heat of reaction (kJ / mol) involved in the reaction is positive for endothermic reactions and negative for exothermic reactions, and needs to be determined by calorimetry experiments or thermodynamic databases. Components The molar heat capacity at constant pressure (kJ / (mol·K)) reflects the heat storage capacity of the material itself.

[0123] The equation simulates the radial temperature field distribution of the reactor by relating the heat release of the reaction (the sum of the enthalpy changes of molecular reactions) to the heat capacity of the material (the sum of the products of the isobaric specific heat capacity of the components and the flow rate) through the radial temperature change rate.

[0124] Furthermore, the momentum transfer equation is used to describe pressure changes and fluid flow resistance within the reactor. For a radially fixed-bed reactor, the momentum transfer equation can be expressed as: ; In the formula, The pressure inside the reactor (Pa) reflects the driving force of fluid flow. The catalyst bed porosity (dimensionless) represents the proportion of voids between catalyst particles. The fluid dynamic viscosity (Pa·s) reflects the fluid viscous resistance. The fluid velocity in the empty tower (m / s) is calculated from the material volumetric flow rate and the reactor cross-sectional area, reflecting the macroscopic rate of fluid flow. is the catalyst particle shape factor (dimensionless). The equivalent diameter (m) of the catalyst particles is determined by sieve analysis or laser particle size analyzer. Fluid density (kg / m³) 3 The properties of the material vary significantly with temperature and pressure, requiring calculation using equations of state.

[0125] This equation calculates the pressure loss caused by bed resistance during fluid flow using the radial pressure change rate, reflecting the influence of pressure distribution within the reactor on material flow and reaction contact time.

[0126] Furthermore, after completing the construction of the transfer equation, the reactor model needs to be solved and verified in combination with the actual structure and operating parameters of the reactor to ensure that the reactor model can accurately reproduce the operating state of the industrial reactor.

[0127] First, regarding structural parameters, for radial fixed-bed reactors, it is necessary to obtain the reactor inner diameter, outer diameter, effective bed radial thickness, and catalyst loading height; for axial fixed-bed reactors, the focus is on reactor length, inner diameter, and catalyst bed height; for radial moving-bed reactors, it is necessary to obtain catalyst movement rate, bed radial thickness, catalyst circulation path size, and particle circulation flow rate.

[0128] In addition, operating parameters include feed flow rate, feed temperature, feed pressure, catalyst activity parameters, and material composition information (molecular composition and content). These structural and operating parameters are used as inputs to the transfer equation, thus determining the accuracy of the reactor model output. For example, the feed flow rate affects the empty tower velocity. This, in turn, alters the momentum and mass transfer processes.

[0129] Furthermore, after completing the construction of the transfer equation and the acquisition of parameters, the reactor model needs to be solved and verified to ensure that the operating state of the industrial reactor can be accurately reproduced.

[0130] Specifically, the differential equations describing the mass, energy, and momentum transfer within the reactor are first transformed into a computable set of algebraic equations using the finite difference method. Then, iterative solutions are obtained using existing technologies such as MATLAB to program and obtain data on the molar flow rate distribution, temperature field distribution, and pressure distribution of components at different locations within the reactor, as output by the reactor model.

[0131] Furthermore, actual industrial operating data of the corresponding reactors in oil refining and chemical production are collected, including the molecular composition of materials at the reactor inlet and outlet, inlet and outlet temperatures, inlet and outlet pressures, as well as the measured values ​​of temperature and pressure at key monitoring points within the bed.

[0132] Meanwhile, the parameters calculated and output by the reactor model are compared with the industrial measured data. If the deviation of the core indicators exceeds the preset threshold, the key parameters in the reactor model are adjusted in reverse, and the discrete solution and data comparison process is re-executed until the deviation between the reactor model calculation results and the actual industrial data is controlled within an acceptable range, ensuring that the reactor model can accurately reproduce the real operating state of the industrial reactor.

[0133] Through the above steps, the constructed reactor model can accurately simulate the mass, energy, and momentum transfer processes in different types of reactors, coupling the microscopic rate of molecular-level reaction kinetics with the macroscopic transfer process of the reactor, and outputting key information such as the molecular composition, temperature, and pressure of the post-reaction stream, providing support for the accurate calculation of subsequent deactivation models and separation unit models.

[0134] S150: Construct the deactivation model and the separation unit model, and connect and couple the molecular composition model, molecular-level reaction kinetics model, reactor model, deactivation model and separation unit model in sequence to generate a full-process reaction-separation molecular-level prediction model, and predict the molecular-level composition of any stream in oil refining and chemical production.

[0135] In this embodiment of the application, in order to accurately reflect the impact of catalyst activity decay on the reaction in petrochemical production, as well as the molecular distribution law in the separation process, it is necessary to first construct a targeted model, and then form a full-process prediction capability through the orderly coupling of multiple models, so as to provide complete support for the molecular-level composition analysis of any stream.

[0136] Specifically, a deactivation model and a separation unit model were constructed separately. When constructing the deactivation model, based on the correlation expression between the catalyst bed position and the coke content, the correlation between the amount of coke deposited and the catalyst activity was used to accurately describe the decay law of catalyst activity over time or feed rate.

[0137] Secondly, when constructing the separation unit model, the structural characteristics of the oil refining and chemical separation unit are combined with the introduction of rigorous molecular thermodynamics and flash distillation tower algorithms. By calculating the gas-liquid phase equilibrium, the molecular orientation in each separation unit can be predicted.

[0138] Furthermore, after the two sub-models are constructed, the molecular composition model, molecular-level reaction kinetics model, reactor model, deactivation model, and separation unit model are connected and coupled according to the actual process sequence of "raw material-reaction-separation" in oil refining and chemical engineering.

[0139] Specifically, the optimized molecular composition output by the molecular composition model is used as the initial input, which is then passed to the molecular-level reaction kinetics model to quantify the molecular transformation relationship, the reactor model to simulate the reaction process, the deactivation model to correct the reaction deviation caused by the decay of catalyst activity, and finally the separation unit model to predict the distribution results of molecules in the separation process.

[0140] Finally, by integrating the coupling and data transfer relationships of the above multiple models, a molecular-level prediction model for the entire reaction-separation process is generated. This model can cover all stages of production and can predict the molecular-level composition of any stream in oil refining and chemical production, providing a data foundation for process optimization and product quality control.

[0141] As attached Figure 3 As shown, step S150 in the method provided in this application embodiment includes: Based on the correlation expression between catalyst bed position and coke content, a deactivation model is constructed and coupled with the reactor model. The deactivation model is used to describe the decay law of catalyst activity with time or feed rate. Based on the structural characteristics of separation units in oil refining and chemical production, a separation unit model is constructed by combining rigorous molecular thermodynamics and flash distillation column algorithms. The separation unit model is used to predict the molecular orientation in each separation unit by calculating the gas-liquid phase equilibrium.

[0142] In this embodiment of the application, in order to accurately simulate the dynamic impact of catalyst activity decay on the reaction process in petrochemical production, as well as the fine control of molecular distribution by the separation unit, it is necessary to construct specialized sub-models step by step and deeply couple them to correlate the molecular-level behavior of each link, so as to provide core support for the prediction of the molecular composition of the entire process stream.

[0143] In the catalytic reactions of petrochemical production, catalysts gradually deactivate due to carbon buildup (coke deposition), which directly affects reaction efficiency and product distribution. Therefore, it is necessary to construct a deactivation model to reflect the law of activity decay and coordinate it with the reactor model to restore the real reaction scenario.

[0144] Specifically, firstly, through industrial side-line experiments and laboratory fixed-bed evaluations, data on coke content at different locations in the catalyst bed were collected under different operating times and feed loads. Combined with catalyst activity test results, a correlation expression between catalyst bed location and coke content was fitted to clarify the distribution gradient of carbon deposits in the bed.

[0145] Furthermore, based on this correlation expression, a deactivation model is constructed with time or feed rate as the independent variable. That is, an activity decay factor is introduced to correlate the quantitative relationship between carbon deposition and catalyst activity, so that the model can output the activity coefficient of each region of the catalyst bed at any time and with any feed rate.

[0146] Furthermore, the deactivation model is coupled with the reactor model to achieve real-time linkage between the dynamic decay of catalyst activity and molecular transformation within the reactor during the reaction process, making the output results of the reactor model more consistent with the actual working conditions of industrial production.

[0147] Specifically, when calculating reaction kinetics, the reactor model calls the bed activity distribution data output by the deactivation model in real time to correct the reaction rate constants in different regions.

[0148] In petroleum refining and chemical catalytic reactions, the reaction rate constant is the core parameter that determines molecular conversion efficiency and product distribution, and its value is directly affected by the catalyst activity.

[0149] The higher the catalyst activity, the larger the reaction rate constant and the faster the molecular transformation; conversely, the conversion efficiency decreases. By coupling the deactivation model with the reactor model, the reactor model can be divided into bed regions, and the current catalyst activity data of each region can be obtained and used as a correction coefficient in the calculation of the reaction rate constant, ensuring that the reaction rate calculation of each region matches the actual activity state of the catalyst in that region.

[0150] Furthermore, when constructing the separation unit model, it is necessary to base it on the actual structural characteristics of the separation unit in oil refining and chemical production to ensure that the model is highly matched with the industrial equipment.

[0151] Specifically, the core structural parameters of the separation unit are first obtained, including the number of trays in the distillation column, the material and specifications of the packing, the number and location of the feed inlet layers, the heat exchange area of ​​the top condenser and the bottom reboiler, or the effective volume of the flash tank, the structure of the feed distributor, etc. These parameters are used as the basic inputs for the separation unit model to build a model framework that fits the actual equipment.

[0152] Furthermore, a rigorous molecular thermodynamics and flash-distillation column algorithm is introduced to accurately calculate the gas-liquid phase equilibrium relationship within the separation unit, ensuring that the separation unit model can truly reflect the material separation effect of the industrial separation device.

[0153] In the method provided in this application embodiment, the rigorous molecular thermodynamics and flash-distillation column calculation algorithm adopts one or both of the SRK equation and the PR equation.

[0154] Specifically, for separation systems with a high content of light hydrocarbons, the SRK equation is preferred because it has higher accuracy in calculating the gas-liquid phase equilibrium parameters of low-carbon alkanes and alkenes. For separation systems containing heavy hydrocarbons and polar components, the PR equation is selected because it can more accurately calculate the fugacity coefficient and phase equilibrium relationship of heavy components. If the separation system contains both light and heavy hydrocarbon components, the two equations can be combined to calculate the gas-liquid phase equilibrium data of different component ranges in segments.

[0155] Furthermore, in the actual operation of the flash-distillation column algorithm, calculations need to be carried out in stages in conjunction with the process logic of the separation device to achieve full-process simulation from preliminary phase separation to fine molecular distribution.

[0156] For the flash evaporation process, the mixed stream output from the reactor is introduced into the flash evaporation module, and the rapid phase separation of the material under set temperature and pressure is simulated based on the gas-liquid phase equilibrium parameters calculated by the selected SRK equation or PR equation.

[0157] For example, when the mixed stream enters the flash tank, the light molecules, due to their low boiling point and easy vaporization, are enriched in the gas phase, while the heavy molecules, due to their high boiling point and difficulty in vaporization, remain in the liquid phase. The flash module outputs the composition of the gas and liquid streams after preliminary separation, providing pre-processed material data for the subsequent distillation process.

[0158] For the distillation column stage, the gas or liquid phase stream after flash evaporation is used as input, and combined with the structural parameters of the distillation column, the gas-liquid mass and heat transfer process is simulated plate by plate through the distillation column algorithm.

[0159] The distillation column algorithm is based on plate-by-plate phase equilibrium data calculated using rigorous molecular thermodynamics. Starting from the feed plate, it iteratively calculates the gas phase composition of each plate in the direction of the top of the column (the content of light components gradually increases with the height of the plate) and iteratively calculates the liquid phase composition of each plate in the direction of the bottom of the column (the content of heavy components gradually increases with the height of the plate) until the temperature, pressure and molecular composition of each plate in the column reach a stable state.

[0160] For example, if the light hydrocarbon gas stream (containing propane, butane, and pentane) after flash evaporation is processed, the gas-liquid phase equilibrium is calculated using the SRK equation, the distillation column is set with 20 trays, the feed inlet is located on the 10th tray, and the reflux ratio is controlled at 2.5.

[0161] Meanwhile, the distillation column algorithm simulates the enrichment of propane molecules in the gas phase at the top of the column (molar fraction of over 98%), the extraction of butane molecules in the side stream in the middle of the column (molar fraction of 92%), and the retention of pentane molecules in the liquid phase at the bottom of the column (molar fraction of 95%), accurately reproducing the separation law of "light components moving upward and heavy components moving downward" in the distillation column.

[0162] Furthermore, if the separation system contains both light and heavy hydrocarbons, the SRK equation is applied for the light hydrocarbon range (C5-C8) and for the heavy hydrocarbon range (C9-C8). 12 The PR equation is invoked to calculate phase equilibrium data in segments to ensure the accuracy of molecular orientation prediction across the entire tower.

[0163] Finally, through the deep integration of the above-mentioned rigorous molecular thermodynamics and flash-distillation column algorithm, the separation unit model can accurately output the molecular-level composition of each separated stream, providing reliable separation data support for the coupling of the whole process model.

[0164] Furthermore, after the deactivation model and separation unit model are constructed, the actual process sequence of refining and chemical production, namely "raw material molecular analysis → reaction path quantification → reactor in-process conversion → catalyst deactivation correction → product separation and purification", must be strictly followed. The molecular composition model, molecular-level reaction kinetics model, reactor model, deactivation model and separation unit model should be connected and coupled sequentially to form a logically closed-loop full-process reaction-separation molecular-level prediction model.

[0165] Specifically, the output of the molecular composition model is first used as the initial input for the entire coupling process.

[0166] Among them, the molecular composition model is directly transferred to the molecular-level reaction dynamics model by matching with a pre-set molecular database, optimizing the probability density function, and correcting with a deep learning molecular property predictor (error less than 1%). The resulting optimized molecular composition and corresponding digital molecular structure topology are then directly transferred to the molecular-level reaction dynamics model.

[0167] Furthermore, the molecular-level reaction kinetics model is based on a reaction rule library constructed from the catalytic reforming reaction mechanism. It calls upon the input molecular composition information to allow each molecule to match and trigger the reaction according to the reaction rules. At the same time, it combines the reaction rate expression of the reaction system to quantify the transformation relationship between molecular reactants and products, and outputs molecular transformation pathways and preliminary reaction rate data.

[0168] Furthermore, the preliminary reaction rate data is imported into the reactor model, which simulates the mass transfer, heat transfer, and flow processes within the reactor based on the mass transfer equation, energy transfer equation, and momentum transfer equation (adapted to reactor types such as axially fixed bed, radially fixed bed, or radially moving bed).

[0169] At this point, the deactivation model and the reactor model are coupled. The deactivation model outputs the activity coefficients of each region of the catalyst bed in real time and dynamically corrects the reaction rate constants of different regions in the reactor, ensuring that the composition of the post-reaction mixed stream output by the reactor model can truly reflect the actual reaction results under catalyst activity decay.

[0170] Finally, the post-reaction mixed stream data is fed into the separation unit model, which is based on the previously constructed device structure framework, rigorous molecular thermodynamics (SRK / PR equations), and flash-distillation column algorithm. The separation unit model calculates the gas-liquid phase equilibrium and predicts molecular orientation, outputting the molecular-level composition of each separated stream, such as the top, side stream, and bottom of the column.

[0171] Ultimately, through the orderly coupling and data linkage of the above multiple models, the full-process reaction-separation molecular-level prediction model can cover the entire process of oil refining and chemical production. It can not only trace the molecular composition characteristics of the feed stream, but also accurately predict the molecular transformation dynamics of the intermediate reaction stream, and further determine the molecular-level composition of each final separated product stream.

[0172] The embodiments of this application, through the specific implementation methods described above, achieve the following technical effects: This application proposes a multi-model-based method for predicting the molecular composition of petrochemical production streams. First, following a pre-defined detection scheme, gas chromatography and infrared spectroscopy are used to perform multiple measurements on refining and chemical feedstocks to comprehensively evaluate and obtain their molecular composition. Then, structural units are assembled based on a structural unit-bond-electric matrix framework to generate a digital molecular structure topology. Next, the initial molecular composition is matched against a pre-set molecular database, and a molecular composition model is constructed through probability density function optimization and molecular property predictor correction. Then, a reaction rule base and molecular-level reaction network are constructed based on the catalytic reforming reaction mechanism. Combined with the reaction rate expression, a molecular-level reaction kinetic model is constructed based on the structural unit-bond-electric matrix framework. Next, based on mass, energy, and momentum transfer equations, and adapting equations for different reactors, structural and operational parameters are collected and verified through solving to construct a reactor model. Simultaneously, a deactivation model is constructed based on the catalyst bed carbon deposition law, and a separation unit model is constructed by combining the separation unit structure with the SRK / PR equation and flash-distillation column algorithm. Finally, the above models are coupled sequentially to generate a full-process reaction-separation molecular-level prediction model, which predicts the molecular composition of any stream.

[0173] The method provided in this application, through the technical solution of "digital analysis of raw material molecules - precise optimization of molecular composition - quantification of reaction network and kinetics - reactor transfer simulation - construction of deactivation and septum model - coupling of multiple models throughout the process", solves the problems of traditional petrochemical stream prediction that only focuses on macroscopic components and ignores molecular-level transformation laws, and relies on lag measurement data to adjust process parameters, thus failing to dynamically reflect the impact of catalyst deactivation on reaction efficiency. It realizes the advance prediction of molecular-level composition of the entire process from raw materials to products, eliminates the lag in process adjustment, and helps petrochemical production to transform towards molecular-level fine control.

[0174] Example 2, as shown in the appendix Figure 4 As shown, based on the inventive concept of the multi-model-based method for predicting the molecular-level composition of petrochemical production streams provided in Embodiment 1, this application also provides a multi-model-based system for predicting the molecular-level composition of petrochemical production streams, specifically including: The raw material molecule digitization module 01 is used to determine the raw material molecular composition in oil refining and chemical production according to a preset detection scheme, and to construct the molecular digital structure topology by splicing the structural units of the raw material molecular composition based on the structural unit-bond electric matrix framework. Molecular composition modeling module 02 is used to predict molecular properties and construct a molecular composition model based on a pre-set molecular database and the digital structure topology of the molecules, combined with a probability density function and a molecular property predictor. The reaction kinetics modeling module 03 is used to establish a molecular-level reaction network based on the catalytic reforming reaction mechanism and to construct a molecular-level reaction kinetics model by combining the reaction rate expression of the reaction system. The reactor simulation modeling module 04 is used to simulate and model reactors in oil refining and chemical production based on the mass transfer equation, energy transfer equation, and momentum transfer equation, and to build reactor models. The full-process coupling prediction module 05 is used to construct the deactivation model and the separation unit model, and connect and couple the molecular composition model, molecular-level reaction kinetics model, reactor model, deactivation model and separation unit model in sequence to generate a full-process reaction-separation molecular-level prediction model, which can predict the molecular-level composition of any stream in oil refining and chemical production.

[0175] In one embodiment, the raw material molecule digitization module 01 is also used for: Gas chromatography and infrared spectroscopy were used to perform several measurements on the raw materials used in oil refining and chemical production. The molecular composition of the raw materials was obtained by comprehensively evaluating the results of these measurements. The structural units in the molecular composition of the raw materials were spliced ​​together according to the structural unit theory, and complete molecules were formed by combining chemical properties and bond-electric matrix connection rules, thus obtaining multiple complete molecular structures. Based on the structural unit-bond-electric matrix framework, the multiple complete molecular structures were digitally represented to generate a digital molecular structure topology.

[0176] In one embodiment, the molecular composition modeling module 02 is also used for: Based on a pre-built molecular database, an initial molecular composition is obtained by topological matching of the digitized molecular structure. The initial molecular composition is then optimized using a probability density function. During the optimization process, a molecular property predictor is used to correct the probability density function until the prediction error is less than 1%, thus obtaining the optimized molecular composition. A molecular composition model is then constructed. The molecular property predictor is built based on deep learning and trained to convergence using sample data.

[0177] In one embodiment, the reaction kinetics modeling module 03 is also used for: A reaction rule library is constructed based on the catalytic reforming reaction mechanism in oil refining and chemical production. This library includes several reaction rules established based on molecular-level reaction mechanisms. A molecular-level reaction network is then established based on this rule library. The molecular composition model and the molecular-level reaction network are coupled. After obtaining molecular composition information through the molecular composition model, each molecule in the molecular composition information is matched according to the reaction rules in the rule library, and reacts according to the rules when they are satisfied. Using the structural unit-bond-electric matrix framework as a reference, a molecular-level reaction kinetic model is constructed under computer-aided conditions based on the reaction rate expression of the reaction system and the molecular-level reaction network. This molecular-level reaction kinetic model is used to quantify the transformation relationship between reactants and products in the molecular-level reaction network.

[0178] In one embodiment, the reactor simulation modeling module 04 further includes: The reactor used in the oil refining and chemical production is one of the following: axial fixed bed reactor, radial fixed bed reactor, or radial moving bed reactor.

[0179] In one embodiment, the end-to-end coupling prediction module 05 is further configured to: Based on the correlation expression between catalyst bed position and coke content, a deactivation model is constructed and coupled with the reactor model. The deactivation model is used to describe the decay law of catalyst activity with time or feed rate. According to the structural characteristics of separation units in oil refining and chemical production, a separation unit model is constructed by combining rigorous molecular thermodynamics and flash distillation column algorithm. The separation unit model is used to predict the molecular orientation in each separation unit by calculating gas-liquid phase equilibrium.

[0180] Furthermore, the end-to-end coupled prediction module 05 also includes: The rigorous molecular thermodynamics and flash-distillation column calculation algorithm adopts one or both of the SRK equation and PR equation.

[0181] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0182] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0183] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application intends to include such modifications and variations.

Claims

1. A method for predicting the molecular-level composition of petrochemical production streams based on multiple models, characterized in that, The methods include: According to the preset detection scheme, the raw materials in oil refining and chemical production are measured to obtain the molecular composition of the raw materials. Based on the structural unit-bond electric matrix framework, the molecular composition of the raw materials is spliced ​​into structural units to construct a molecular digital structure topology. Based on the pre-set molecular database and the digital molecular structure topology, molecular properties are predicted by combining the probability density function and the molecular property predictor, and a molecular composition model is constructed. A molecular-level reaction network was established based on the catalytic reforming reaction mechanism, and a molecular-level reaction kinetic model was constructed by combining the reaction rate expression of the reaction system. Based on the mass transfer equation, energy transfer equation, and momentum transfer equation, a reactor model is constructed for simulation modeling of reactors in oil refining and chemical production. A deactivation model and a separation unit model are constructed, and the molecular composition model, molecular-level reaction kinetics model, reactor model, deactivation model and separation unit model are connected and coupled in sequence to generate a full-process reaction-separation molecular-level prediction model, which can predict the molecular-level composition of any stream in oil refining and chemical production.

2. The method for predicting the molecular-level composition of petrochemical production streams based on multiple models according to claim 1, characterized in that, According to a preset detection scheme, the molecular composition of raw materials in oil refining and chemical production is obtained by measuring the raw material molecular composition. Based on the structural unit-bond electrical matrix framework, the molecular composition of the raw material is assembled into structural units to construct a digital molecular structure topology, including: The raw materials in oil refining and chemical production were measured several times using a gas chromatograph and an infrared spectrometer. The molecular composition of the raw materials was obtained by comprehensively evaluating the results of the several measurements. According to the structural unit theory, the structural units in the molecular composition of the raw materials are spliced ​​together, and combined with chemical properties and bond-electric matrix connection rules to form complete molecules, thereby obtaining multiple complete molecular structures. Based on the structural unit-bond electric matrix framework, the multiple complete molecular structures are digitally represented to generate a digital molecular structural topology.

3. The method for predicting the molecular-level composition of petrochemical production streams based on multiple models according to claim 1, characterized in that, Based on a pre-set molecular database and the aforementioned digital molecular structure topology, molecular properties are predicted using a probability density function and a molecular property predictor, and a molecular composition model is constructed, including: Based on a pre-built molecular database, the initial molecular composition is obtained by topological matching of the digital molecular structure. The initial molecular composition is optimized using a probability density function. During the optimization process, the probability density function is corrected using a molecular property predictor until the prediction error is less than 1%, thus obtaining the optimized molecular composition and constructing a molecular composition model. The molecular property predictor is built based on deep learning and trained to convergence using sample data.

4. The method for predicting the molecular-level composition of petrochemical production streams based on multiple models according to claim 1, characterized in that, A molecular-level reaction network was established based on the catalytic reforming reaction mechanism, and a molecular-level reaction kinetic model was constructed by combining the reaction rate expression of the reaction system, including: A reaction rule library is constructed based on the catalytic reforming reaction mechanism in oil refining and chemical production. The reaction rule library includes several reaction rules established based on molecular-level reaction mechanisms. A molecular-level reaction network is established based on the reaction rule base. The molecular composition model and the molecular-level reaction network are coupled. After obtaining molecular composition information through the molecular composition model, each molecule in the molecular composition information will be matched according to the reaction rules in the reaction rule base, and react according to the reaction rules when the reaction rules are met. Based on the structural unit-bond electrical matrix framework, a molecular-level reaction kinetic model is constructed under computer-aided conditions, based on the reaction rate expression of the reaction system and the molecular-level reaction network. The molecular-level reaction kinetic model is used to quantify the transformation relationship between molecular reactants and products in the molecular-level reaction network.

5. The method for predicting the molecular-level composition of petrochemical production streams based on multiple models according to claim 1, characterized in that, The reactor used in the oil refining and chemical production is one of the following: axial fixed bed reactor, radial fixed bed reactor, or radial moving bed reactor.

6. The method for predicting the molecular-level composition of petrochemical production streams based on multiple models according to claim 1, characterized in that, Constructing the inactivation model and the separation unit model includes: Based on the correlation expression between catalyst bed position and coke content, a deactivation model is constructed and coupled with the reactor model. The deactivation model is used to describe the decay law of catalyst activity with time or feed rate. Based on the structural characteristics of separation units in oil refining and chemical production, a separation unit model is constructed by combining rigorous molecular thermodynamics and flash distillation column algorithms. The separation unit model is used to predict the molecular orientation in each separation unit by calculating the gas-liquid phase equilibrium.

7. The method for predicting the molecular-level composition of petrochemical production streams based on multiple models according to claim 6, characterized in that, The rigorous molecular thermodynamics and flash-distillation column calculation algorithm adopts one or both of the SRK equation and PR equation.

8. A multi-model-based prediction system for the molecular-level composition of petrochemical production streams, characterized in that, The system is used to execute the multi-model-based method for predicting the molecular-level composition of petrochemical production streams as described in any one of claims 1-7, and the system comprises: The raw material molecule digitization module is used to determine the raw material molecular composition in oil refining and chemical production according to a preset detection scheme. Based on the structural unit-bond electric matrix framework, the raw material molecular composition is spliced ​​into structural units to construct a molecular digital structure topology. The molecular composition modeling module is used to predict molecular properties and construct a molecular composition model based on a pre-set molecular database and the digital structure topology of the molecules, combined with a probability density function and a molecular property predictor. The reaction kinetics modeling module is used to establish molecular-level reaction networks based on the catalytic reforming reaction mechanism and to construct molecular-level reaction kinetic models by combining the reaction rate expressions of the reaction system. The reactor simulation modeling module is used to simulate and model reactors in oil refining and chemical production based on the mass transfer equation, energy transfer equation, and momentum transfer equation, and to build reactor models. The full-process coupled prediction module is used to construct the deactivation model and the separation unit model, and connect and couple the molecular composition model, molecular-level reaction kinetics model, reactor model, deactivation model and separation unit model in sequence to generate a full-process reaction-separation molecular-level prediction model, which can predict the molecular-level composition of any stream in oil refining and chemical production.

Images