A method for constructing an atmospheric distillation dynamic mixing model based on a crude oil molecular model

Abstract

The application belongs to the technical field of petroleum chemical industry, and discloses a construction method of an atmospheric distillation dynamic hybrid model based on a crude oil molecular model drive. In the calculation of a stable state, the distillation property parameters of the crude oil molecular model related to the molecular pure substance and the distillation process are calculated, and the distillation related property parameters of the mixture are calculated according to the mixing rule; the mechanism model is established for the whole primary distillation column, the top and bottom of the atmospheric distillation column according to the material balance equation, the phase balance equation, the mole fraction normalization equation and the heat balance equation; the agent model is established for the remaining main part of the atmospheric distillation column according to the agent model; and the atmospheric distillation process dynamic hybrid model with the structure of mechanism model-agent model-mechanism model is built in the order of the primary distillation column-atmospheric distillation column main part-atmospheric distillation column top and bottom. The model can simulate and calculate the specific molecular composition distribution of the product.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of petroleum chemical industry, and particularly relates to a construction method of an atmospheric distillation dynamic mixing model based on a crude oil molecule model. BACKGROUND

[0002] The rise of green energy has alleviated the pollution caused by traditional energy and has an impact on the cost of traditional energy. Therefore, the traditional oil refining industry is under double pressure of environmental protection and efficiency improvement, and urgently needs to upgrade and transform the process. Distillation, as one of the crude oil refining technologies, plays an important role in the crude oil processing process, and its process technology is constantly innovated and developed with the aid of computer modeling and simulation technology. Through modeling and simulation, engineers can evaluate the optimal operating parameters of the distillation process under different conditions with the correct principles, and optimize and adjust the process according to the simulated real-time situation, so as to improve product quality and save energy. The modeling of the distillation process is based on a comprehensive property library, and the mechanism model of the distillation process is mathematically modeled by material and energy balance equations, which can simulate the steady-state, multi-component and multi-stage distillation process.

[0003] However, in actual situations, the distillation process is a dynamic process affected by many factors. In order to fully understand the dynamic changes of the distillation process caused by different variables, differential equations are used to introduce time variables into the material and heat balance equations to establish a distillation dynamic model. The commercial software Aspen Dynamics can also perform dynamic simulation of the distillation process. For the atmospheric distillation process of crude oil, in order to reduce the complexity of dynamic calculation, the crude oil is usually divided into virtual components according to the actual boiling point curve for dynamic simulation calculation, but the estimation of the key properties of virtual components is controversial, and the dynamic changes of specific substances in the distillation process cannot be directly obtained. If real components are used for dynamic calculation of the crude oil distillation process, not only can the dynamic changes of all components be directly obtained, but also it is beneficial to the subsequent process simulation of the distillation products. However, crude oil is composed of thousands of independent complex mixtures including heteroatoms, and there are many variables in the entire distillation process. Therefore, its dynamic mathematical model is a complex nonlinear equation, and the entire calculation process needs to be simplified. In the face of models with strong nonlinearity, high complexity and large scale, proxy models are often used instead of mechanism models for solving, and polynomial response surface, radial basis function, Gaussian process model and artificial neural network are common proxy models. Among them, artificial neural network can excellently handle complex nonlinear relationships and multidimensional data, and has strong nonlinear mapping ability, and has been applied to solve various problems.

[0004] With the increasing complexity of the model and the requirement of the model accuracy, compared with a single model, the hybrid model makes full use of the process mechanism, data and prior knowledge, and then shows better global performance. With the continuous development of computer level and mathematical model, a number of works show that the hybrid model can effectively reduce the model complexity related to nonlinearity, uncertainty and multiscale, and plays an important role in the complex dynamic system in the field of fossil fuel energy. SUMMARY

[0005] In view of the shortcomings of the prior art, the purpose of the present application is to build the property parameters related to distillation of each molecule in the crude oil molecular model mentioned in CN118298937A, and then build a dynamic hybrid model of atmospheric distillation based on the crude oil molecular model by hybrid modeling of mechanism model and proxy model.

[0006] The technical scheme of the present application is as follows: a construction method of an atmospheric distillation dynamic hybrid model based on a crude oil molecular model driver, comprising the following steps:

[0007] Under stable state, the distillation property parameters related to the distillation process of the pure substance molecules in the crude oil molecular model are calculated, and the distillation related property parameters of the mixture are calculated according to the mixing rule;

[0008] According to the material balance equation, the phase equilibrium equation, the mole fraction normalization equation and the heat balance equation, the mechanism model is established for the whole primary distillation column, the top and bottom of the atmospheric distillation column;

[0009] According to the proxy model, the proxy model is established for the remaining main part of the atmospheric distillation column;

[0010] According to the order structure of the primary distillation column-atmospheric distillation column main part-atmospheric distillation column top and bottom, the crude oil atmospheric distillation process dynamic hybrid model with the structure of mechanism model-proxy model-mechanism model is built.

[0011] The calculation process involved in the crude oil atmospheric distillation dynamic hybrid model includes:

[0012] The calculation of the distillation related property parameters of the pure substance molecules and the mixture involved in the crude oil molecular model will obtain the data to form the basic database for the construction of the mechanism model; the distillation related property parameters include vapor phase enthalpy, liquid phase enthalpy, activity coefficient, vapor-liquid equilibrium constant and bubble point temperature;

[0013] The dynamic mechanism model of the primary distillation column, the top and bottom of the atmospheric distillation column is established, and is constructed according to the traditional material balance equation, phase equilibrium equation, mole fraction normalization equation and heat balance equation;

[0014] The construction of the proxy model is a three-layer network structure BP neural network composed of an input layer, a hidden layer and an output layer; wherein the molar flow, temperature and pressure of each component of the feed of the atmospheric distillation column are taken as the input unit of the BP neural network, and the flow and molar fraction of each molecular weight of the product stream are taken as the output unit of the BP neural network.

[0015] The calculation methods of the vapor phase enthalpy, liquid phase enthalpy, activity coefficient, vapor-liquid equilibrium constant and bubble point temperature are as follows:

[0016] The vapor phase enthalpy and liquid phase enthalpy of pure substances are calculated by obtaining the vapor phase enthalpy value and liquid phase enthalpy value of each molecule at different temperatures through commercial software, fitting them into a formula related to temperature to obtain the constant term in the formula, and calculating the vapor phase enthalpy value and liquid phase enthalpy value of the mixture through the mixing rule;

[0017] The activity coefficient of each molecule is calculated by correlating the activity coefficient with the temperature through fitting, and the constant term in the correlation formula is obtained according to the activity coefficient at different temperatures, which is used to calculate the activity coefficient at different temperatures;

[0018] The gas phase used in the dynamic mixing model of the atmospheric distillation process of crude oil is an ideal gas, and the liquid phase is a non-ideal solution system; the vapor-liquid equilibrium constant is calculated from the activity coefficient, the saturation vapor pressure and the current tray pressure, wherein the saturation vapor pressure is calculated by the Antoine equation, and the current tray pressure is a constant value;

[0019] The bubble point temperature is calculated by the Richmond iteration method.

[0020] The vapor phase product flow of the primary distillation column is calculated by multiplying the vapor phase fraction calculated by the Rachford-Rice equation using Newton iteration with the feed flow; the vapor phase flow calculation formula of the top and bottom of the atmospheric distillation column is obtained by simultaneously solving the material balance and heat balance equations.

[0021] The construction method of the proxy model is as follows:

[0022] The construction of the proxy model adopts a three-layer network structure BP neural network composed of an input layer, a hidden layer and an output layer; wherein the molar flow, temperature and pressure of each component of the feed of the atmospheric distillation column are taken as the input unit of the BP neural network, the vapor phase molar flow of each component of the second tray of the atmospheric distillation column and the liquid phase molar flow of each component of the j-1th tray in the calculation results are taken as the output unit of the BP neural network, and it is assumed that the crude oil distillation column has j trays, and each tray is numbered from top to bottom as 1, 2, 3…j.

[0023] The acquisition method of the training database in the proxy model and the setting of the training parameters are as follows:

[0024] To ensure the diversity and randomness of the training data set, 200 sample points were extracted in the range of parameter correlation by Latin hypercube sampling, and Latin hypercube sampling was performed again every time the feed composition of the primary distillation column was changed;

[0025] By random division, 70% of the data set was set as the training sample set, and 30% as the verification sample set; The activation function of the hidden layer in the BP neural network was selected as the logarithmic function in the S-type function: logsing; The transfer function from the hidden layer to the output layer was the Purelin linear function; The BP neural network selected Levenberg-Marquardt algorithm for training.

[0026] Further, the maximum number of iterations for training was set to 1000. The target value of the mean square error was 0.0003, the network learning rate was 0.002, the verification check number was 12, and the determination of the number of hidden layer nodes was made by an empirical formula.

[0027] Since the mechanism model of the atmospheric distillation column is mainly based on material balance, energy balance, and vapor-liquid phase balance equations, in order to ensure the effectiveness of the constructed mechanism model and reduce the complexity of its calculation structure, the following assumptions are made for the mechanism model of the atmospheric distillation column:

[0028] (1) The vapor-liquid two-phase on each theoretical tray is completely mixed;

[0029] (2) The heat loss inside the tower and the heat exchange with the outside are ignored;

[0030] (3) The temperature and pressure of the vapor-liquid two-phase inside the distillation column are uniform;

[0031] (4) The vapor storage on each theoretical tray is ignored;

[0032] (5) The pressure drop on each theoretical tray is the same.

[0033] The mechanism model and the agent model of the atmospheric distillation process of crude oil constructed in the application are used to simulate the process of the atmospheric distillation of crude oil and dynamically predict the product composition. The feed of the model is a mixture of 152 real components composed of the predetermined molecular model of crude oil containing saturated components (S), aromatic components (A), resin components (R) and asphaltene components (As) and water molecules constructed in the previous work. After the relevant properties of each real component and the mixture are calculated, the mechanism model of the pre-flash tower (PF), the overhead and the bottom of the atmospheric distillation tower (ADU) is established based on certain assumptions by using MATLAB / Simulink. The agent model constructed by the BP neural network is used to replace the remaining main part of the atmospheric distillation tower. The hybrid model constructed by the mechanism model and the agent model can quickly and effectively predict the dynamic changes of the molecules involved in the molecular model of crude oil in the atmospheric distillation process.

[0034] The atmospheric distillation dynamic hybrid model based on the molecular model of crude oil in the application can make predictions of the dynamic changes of molecules in the distillation process while obtaining the specific molecular composition of the products based on the original molecular model mentioned in CN118298937A. The specific molecular composition distribution of the products obtained by the simulation calculation of the atmospheric distillation dynamic hybrid model of crude oil can facilitate the subsequent modeling simulation and product property evaluation of the production process, and the dynamic change trend of each molecule in the distillation process can better guide the change of the production operation parameters. When the production raw oil variety is changed, the model can be used to predict the production results and guide the change of the operation parameters in the specific production process. BRIEF DESCRIPTION OF DRAWINGS

[0035] Figure 1 is the structural composition diagram of the atmospheric distillation dynamic hybrid model of crude oil constructed in the application;

[0036] Figure 2 is the calculation flow sequence diagram of the hybrid model in the application; (a) is the overall flow diagram; (b) is the mechanism model diagram of the preliminary distillation tower; (c) is the mechanism model diagram of the overhead of the atmospheric distillation tower; (d) is the mechanism model diagram of the bottom of the atmospheric distillation tower.

[0037] Figure 3 is the specific operation scheme adopted in the embodiment of the application;

[0038] Figure 4Absolute error of each component in the dynamic change curve of product stream composition in the embodiment of the present application and in the steady state; (a) is the change trend of each component of the vapor phase product at the top of the crude oil atmospheric distillation column under operation scheme I; (b) is the change trend of each component of the liquid phase product at the top of the crude oil atmospheric distillation column under operation scheme I; (c) is the change trend of each component of the liquid phase product at the bottom of the crude oil atmospheric distillation column under operation scheme I; (d) is the absolute error of each component of the vapor phase product at the top of the crude oil atmospheric distillation column under operation scheme I and the result of the commercial software; (e) is the absolute error of each component of the liquid phase product at the top of the crude oil atmospheric distillation column under operation scheme I and the result of the commercial software; (f) is the absolute error of each component of the liquid phase product at the bottom of the crude oil atmospheric distillation column under operation scheme I and the result of the commercial software;

[0039] Figure 5 Absolute error of each component in the dynamic change curve of product stream composition in the embodiment of the present application and in the steady state; (a) is the change trend of each component of the vapor phase product at the top of the crude oil atmospheric distillation column under operation scheme I; (b) is the change trend of each component of the liquid phase product at the top of the crude oil atmospheric distillation column under operation scheme I; (c) is the change trend of each component of the liquid phase product at the bottom of the crude oil atmospheric distillation column under operation scheme I; (d) is the absolute error of each component of the vapor phase product at the top of the crude oil atmospheric distillation column under operation scheme I and the result of the commercial software; (e) is the absolute error of each component of the liquid phase product at the top of the crude oil atmospheric distillation column under operation scheme I and the result of the commercial software; (f) is the absolute error of each component of the liquid phase product at the bottom of the crude oil atmospheric distillation column under operation scheme I and the result of the commercial software;

[0040] Figure 6 Absolute error of each component in the dynamic change curve of product stream composition in the embodiment of the present application and in the steady state; (a) is the change trend of each component of the vapor phase product at the top of the crude oil atmospheric distillation column under operation scheme I; (b) is the change trend of each component of the liquid phase product at the top of the crude oil atmospheric distillation column under operation scheme I; (c) is the change trend of each component of the liquid phase product at the bottom of the crude oil atmospheric distillation column under operation scheme I; (d) is the absolute error of each component of the vapor phase product at the top of the crude oil atmospheric distillation column under operation scheme I and the result of the commercial software; (e) is the absolute error of each component of the liquid phase product at the top of the crude oil atmospheric distillation column under operation scheme I and the result of the commercial software; (f) is the absolute error of each component of the liquid phase product at the bottom of the crude oil atmospheric distillation column under operation scheme I and the result of the commercial software;

[0041] Figure 7The following are probability histograms of the absolute errors of all molecules involved in the product streams under the three operating schemes in the embodiments of the present invention; (a) is a probability histogram of the absolute errors of the molecular content of the vapor phase product at the top of the crude oil atmospheric distillation tower under operating scheme I and the results of the commercial software; (b) is a probability histogram of the absolute errors of the molecular content of the liquid phase product at the top of the crude oil atmospheric distillation tower under operating scheme I and the results of the commercial software; (c) is a probability histogram of the absolute errors of the molecular content of the liquid phase product at the bottom of the crude oil atmospheric distillation tower under operating scheme I and the results of the commercial software; (d) is a probability histogram of the absolute errors of the molecular content of the vapor phase product at the top of the crude oil atmospheric distillation tower under the operating scheme changed from I to II and the results of the commercial software; (e) is a probability histogram of the absolute errors of the molecular content of the vapor phase product at the top of the crude oil atmospheric distillation tower under the operating scheme changed from I to II (f) is the probability histogram of the absolute error between the molecular content of the liquid phase product at the top of the crude oil atmospheric distillation tower and the commercial software results; (g) is the probability histogram of the absolute error between the molecular content of the vapor phase product at the top of the crude oil atmospheric distillation tower and the commercial software results when the operating scheme is changed from II to III; (h) is the probability histogram of the absolute error between the molecular content of the liquid phase product at the top of the crude oil atmospheric distillation tower and the commercial software results when the operating scheme is changed from II to III; (i) is the probability histogram of the absolute error between the molecular content of the liquid phase product at the bottom of the crude oil atmospheric distillation tower and the commercial software results when the operating scheme is changed from II to III. DETAILED DESCRIPTION

[0042] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0043] Figure 1 This is a structural composition diagram of the dynamic mixing model of the crude oil atmospheric distillation process constructed by the present invention. Figure 1 As shown, the feed is crude oil that has undergone electrical desalting. The pre-flash tower PF serves as the crude oil primary distillation tower. The flashed liquid product, after undergoing temperature and pressure changes in the heat exchanger HE, enters the atmospheric distillation tower ADU to complete the crude oil atmospheric distillation process. P1 and P2 are the gaseous and liquid products at the top of the ADU, respectively, P3 is the liquid product at the bottom of the ADU, and STEAM is the stripping steam. Therefore, a mechanism model is established for the PF, an agent model is established for the main part of the ADU, and mechanism models are established for the top and bottom of the ADU, resulting in a hybrid model with a mechanism-agent-mechanism model as shown in the figure.

[0044] Figure 2 This is the calculation process sequence of the hybrid model in the present invention. The liquid phase product at the bottom of the tower is calculated by the primary distillation tower mechanism model and flows to the crude oil atmospheric distillation tower proxy model. The generated vapor phase flow enters the tower top mechanism model, and the generated liquid phase flow enters the tower bottom mechanism model to complete the final calculation. Figure 2As shown in the figure, the specific crude oil molecular composition distribution is obtained by calculating the crude oil molecular model, and then enters the primary distillation column mechanism model, and the calculation is carried out under the condition of specified primary distillation temperature and pressure, and the liquid phase flow rate and specific molecular composition distribution of the primary distillation column bottom are obtained. Through the heat exchanger setting feed temperature and pressure, then the liquid phase product of the primary distillation column bottom enters the proxy model of the crude oil atmospheric distillation column main part under the condition of this temperature and pressure and carries out calculation, obtains the vapor phase flow rate and composition of the 2nd plate of the crude oil atmospheric distillation column and the liquid phase flow rate and composition of the j-1th plate, and makes them enter the mechanism model of the top and bottom of the atmospheric distillation column respectively for calculation simulation, and the final result is obtained.

[0045] Figure 3 The specific operation scheme adopted in the embodiment of the present application. Mainly includes the crude oil type, flash temperature, feed temperature, feed pressure, cooling temperature, and atmospheric distillation column top pressure. When changing from operation scheme I to scheme II, the crude oil type does not change, and other conditions increase by 20. When changing operation scheme II to scheme III, the crude oil type changes from OIL I to OIL II, and other conditions increase by 10.

[0046] Figure 4 The dynamic change curve of the product stream composition and the absolute error of each component in the steady state in the embodiment of the present application. As can be seen from figures (a)-(c), the molar fraction of the five components in the product stream reaches a stable state after a period of time. Comparing the molar fraction of the five components in the product material in the stable state of the mixed model with the simulation results of Aspen Plus, it can be seen that under the same conditions, the absolute error of each component molar fraction is small, all within ±5%, and the smallest difference is the gum in P1 figure (d), which is 7.01x10-15%, and the largest difference is the saturated fraction in P1 figure (d), which is-3.9%.

[0047] Figure 5The following is a dynamic change curve of the product stream composition after the operating parameters are disturbed in the embodiment of the present invention, as well as the absolute error of each component in the steady state. As can be seen from the figure, with the change of operating conditions, in products P1 (Figure (a)), P2 (Figure (b)) and P3 (Figure (c)), except for the molar fraction of the saturated fraction, the molar fraction of the other components increases to varying degrees, and then reaches a stable state again. Among them, the gas phase content of the components in product P1 (Figure (a)) suddenly decreases or increases when the operating scheme is changed. This is mainly because the equilibrium constant of each component changes with the change of external operating conditions. At the same time, the gas phase content of each component is determined based on the liquid phase content and the equilibrium constant. As the liquid phase content and the equilibrium constant change, the calculated initial value of the gas phase content of each component is affected, which leads to the occurrence of this phenomenon. Figures (d) to (f) show the absolute errors of the molar fractions of the five components after products P1 to P3 reach a stable state again. As can be seen from the figure, the prediction results of the mixing model still maintain good accuracy after changing the operating conditions for the same crude oil, and the absolute error of each component remains between 2.92 and -3.04%. The minimum and maximum absolute errors are asphaltene (8.7×10-25%) and saturates (-3.094%) in product P1 (Figure (d)), respectively.

[0048] Figure 6 This is the dynamic change curve of the product stream composition after the oil type is changed in the embodiment of the present invention, as well as the absolute error of each component in the steady state. When the external operating conditions change, the content of each component that has reached equilibrium in products P1 to P3 will fluctuate and then reach equilibrium again. From Figure (a), it can be seen that the content of each component in the gas phase product P1 at the top of the ADU tower changes with the change of the operating scheme. This is due to the change in the equilibrium constant caused by the change in operating conditions. The difference is that the process of reaching equilibrium is not obvious. Combined with Figure (b), it can be seen that after the operating scheme is switched to Scheme III, the content of each component in the ADU tower top product (P1, P2) changes little, resulting in the process of the content of each component in P1 reaching equilibrium is not obvious. Figure (d)

[0049] Figures (f) to (f) show the absolute errors in the content of each component in the product streams when they reach equilibrium under Scheme III. As can be seen from the figure, the absolute errors for all components fall within the range of -0.5456–0.548%. The absolute error for asphaltenes in product P1 (Figure d) is the smallest, at 7.28 × 10⁻¹⁶, while the absolute error for aromatics in product P2 (Figure (e)) is the largest, at 0.548%.

[0050] Figure 7The probability histograms of the absolute error of all molecules involved in the product streams for the three operating schemes in the embodiments of the present application are shown in Figures 1(a), (b), (c), (d), (e), (f), (g), (h) and (i). As can be seen from the histograms, the probabilities that the absolute error of all molecules in product P1 (Figures (a), (d), (g)) is within ±1% are 97.62, 97.62 and 97.6%, respectively; the probabilities that the absolute error of all molecules in product P2 (Figures (b), (e), (h)) is within ±1% are 96.62, 97.6 and 95.9%, respectively; and the probabilities that the absolute error of all molecules in product P3 (Figures (c), (f), (i)) is within ±0.1% are 88.2, 87.9 and 88%, respectively, and all the absolute errors are within ±1%. The probabilities that the absolute error of all molecules in all streams is within -0.05-1.0% are all above 90%, and the absolute error of most molecules is within ±5%. This also demonstrates the accuracy and stability of the prediction results by the mixing model.

Claims

1. A method for constructing a dynamic mixing model for atmospheric distillation driven by a crude oil molecular model, characterized in that: The steps are as follows: Calculate the distillation physical properties related to the distillation process of molecular pure substances in the crude oil molecular model under steady state, and calculate the distillation-related physical properties of the mixture according to the mixing rules; Based on the material balance equation, phase balance equation, mole fraction normalization equation and heat balance equation, a mechanism model is established for the entire primary distillation tower and the top and bottom of the atmospheric distillation tower. Based on the proxy model, a proxy model is established for the remaining main part of the atmospheric distillation tower; According to the sequential structure of primary distillation tower - main part of atmospheric distillation tower - top and bottom of atmospheric distillation tower, a dynamic hybrid model of crude oil atmospheric distillation process with the structure of mechanism model - proxy model - mechanism model was built; The construction of the agent model adopts a BP neural network with a three-layer network structure consisting of an input layer, a hidden layer, and an output layer; The molar flow rate, temperature and pressure of each component of the feed to the atmospheric distillation tower are used as the input units of the BP neural network. The molar flow rate of each component of the vapor phase on the second tray of the atmospheric distillation tower and the molar flow rate of each component of the liquid phase on the j-1th tray that have converged effectively in the calculation results are used as the output units of the BP neural network. It is assumed that the crude oil distillation tower has a total of j trays, and each tray is numbered 1, 2, 3...J from top to bottom. The calculation process of the mixed model includes: obtaining the specific molecular composition distribution of crude oil through calculation of the crude oil molecular model, and then entering the primary distillation tower mechanism model, calculating under the specified primary distillation temperature and pressure, obtaining the liquid phase flow rate and specific molecular composition distribution at the bottom of the primary distillation tower, setting the feed temperature and pressure through the heat exchanger, and then the liquid phase product at the bottom of the primary distillation tower enters the proxy model of the main part of the crude oil atmospheric distillation tower under the conditions of this temperature and pressure and calculates, obtaining the vapor phase flow rate and composition of the second plate of the crude oil atmospheric distillation tower and the liquid phase flow rate and composition of the j-1th plate, and respectively entering the mechanism models of the top and bottom of the atmospheric distillation tower for calculation and simulation to obtain the final result.

2. The method for constructing a dynamic mixing model for atmospheric distillation driven by a crude oil molecular model according to claim 1, characterized in that: The calculation process involved in the dynamic mixing model of crude oil atmospheric distillation includes: The crude oil molecular model involves the calculation of distillation-related physical property parameters of molecular pure substances and mixtures, and the obtained data constitutes a basic database for the construction of the mechanism model; the distillation-related physical property parameters include vapor phase enthalpy, liquid phase enthalpy, activity coefficient, vapor-liquid phase equilibrium constant and bubble point temperature.

3. The method for constructing a dynamic mixing model for atmospheric distillation driven by a crude oil molecular model according to claim 2, characterized in that: The calculation method of the vapor phase enthalpy, liquid phase enthalpy, activity coefficient, vapor-liquid phase equilibrium constant and bubble point temperature is as follows: The vapor and liquid enthalpies of pure substances are calculated using commercial software to obtain the vapor and liquid enthalpies of each molecule at different temperatures. These values ​​are then fitted into a temperature-dependent formula to obtain the constant term in the formula. The vapor and liquid enthalpies of the mixture are then calculated using the mixing rule. The activity coefficient of each molecule is calculated by correlating the activity coefficient with temperature through fitting. The constant term in the correlation formula is obtained based on the activity coefficient at different temperatures and is used to calculate the activity coefficient at different temperatures. The dynamic mixing model for the atmospheric distillation of crude oil assumes an ideal gas phase and a non-ideal solution system for the liquid phase. The vapor-liquid equilibrium constant is calculated from the activity coefficient, saturated vapor pressure, and the current tray pressure. The saturated vapor pressure is calculated using the Antoine equation, and the current tray pressure is a constant. The bubble point temperature is calculated using the Richmond iteration method.

4. The method for constructing a dynamic mixing model for atmospheric distillation driven by a crude oil molecular model according to claim 3, characterized in that: The vapor phase product flow rate of the primary distillation tower is obtained by multiplying the vapor phase fraction obtained by Newton iteration calculation using the Rachford-Rice equation by the feed flow rate; the vapor phase flow rate calculation formula of the top and bottom of the atmospheric distillation tower is obtained by combining the material balance equation and the heat balance equation.

5. The method for constructing a dynamic mixing model for atmospheric distillation driven by a crude oil molecular model according to claim 1, characterized in that: The method for obtaining the training database and setting the training parameters in the proxy model are as follows: To ensure the diversity and randomness of the training data set, 200 sample points were extracted within the parameter-related range through Latin hypercube sampling. Latin hypercube sampling was repeated every time the feed composition of the primary distillation tower was changed. Through random partitioning, 70% of the data set is set as the training sample set and 30% is set as the validation sample set; the activation function of the hidden layer in the BP neural network is selected as the logarithmic function of the S-type function: logsing; the transfer function from the hidden layer to the output layer is the Purelin linear function; the BP neural network uses the Levenberg-Marquardt algorithm for training.

Images