A digital twin system based on Grignard reaction

Abstract

The present invention relates to a Grignard reaction modeling technology, and discloses a digital twin system based on Grignard reaction, wherein the method comprises the following steps: obtaining sample data; filtering the sample data, filtering the obtained sample data; preprocessing the sample data, calculating the reactant concentration of the filtered sample data through a thermodynamic model; and establishing a Grignard reaction model, establishing a reaction model using an RBF neural network model based on the calculated reactant concentration, thereby determining the Grignard reaction model. The present invention uses a hybrid modeling method, i.e., a combination of a mechanism model and a machine learning method, to construct a hybrid model using industrial production data, thereby being able to simulate the Grignard reaction well.

Description

Technical Field

[0001] The invention relates to a Grignard reaction modeling technology, and in particular to a digital twin system based on Grignard reaction. Background Art

[0002] Grignard reagents have significant chemical activity and can extract protons from water and other Lewis acids. The reactions in which Grignard reagents can be used to extend carbon chains are called Grignard reactions, which mainly include alkylation reactions, carbonyl addition reactions, conjugate addition reactions, and halogenated hydrocarbon reduction reactions.

[0003] RMgX + R'COR'' ---> RR'R''COMgX (carbonyl addition reaction).

[0004] The simulation of Grignard reaction usually requires obtaining reaction kinetic data and selecting a suitable kinetic model to calculate the reaction degree. However, the obtained reaction kinetic simulation is still biased towards theory. In the actual reaction process, the reactor equipment size, structure, stirring speed, temperature disturbance, etc. will affect the reaction. Therefore, simply using the reaction kinetic model cannot simulate the reaction process well. Summary of the invention

[0005] In view of the problem in the prior art that the Grignard reaction cannot be realistically simulated by only using a reaction kinetic model, the present invention provides a digital twin system based on the Grignard reaction.

[0006] In order to solve the above technical problems, the present invention is solved by the following technical solutions:

[0007] A digital twin system based on Grignard reaction, comprising:

[0008] A module for acquiring sample data, wherein the sample data includes the temperature of the Grignard reactor, the pressure of the Grignard reactor, the raw material flow rate of the reactants, the reaction time of the reactants, the power of the stirring paddle motor of the Grignard reactor, the flow rate of the output materials and the composition of the outlet reactants;

[0009] A sample data filtering module, the sample data filtering module filters the acquired sample data;

[0010] A reactant concentration calculation module, which is used to calculate the reactant concentration of the filtered sample data through a thermodynamic model;

[0011] The module for establishing the Grignard reaction model establishes a reaction kinetic model based on the calculated reactant concentrations using the RBF neural network model, thereby determining the Grignard reaction model.

[0012] Preferably, the calculation of reactant concentrations includes:

[0013] The enthalpy of each reactant is calculated before mixing, and the enthalpy of each reactant before mixing is obtained by performing PT flash calculation on the reactants through a thermodynamic model;

[0014] Calculation of reactants after mixing, and then perform pH flash calculation on the mixed reactants to obtain the mole fraction and temperature of each component;

[0015] Calculate the concentration of the mixed reactants and obtain the concentration of the mixed reactants. ;

[0016]

[0017] In the formula, is the molar flow rate of the total logistics, is the liquid phase mole fraction of the component, is the volume of the reactor.

[0018] Preferably, the method for calculating PT flash includes:

[0019] Step 1: Determine the flow rate F of the reactants and the composition of the reactants , pressure is P, temperature is ;

[0020] Step 2: Estimate the liquid phase composition and gas phase composition ,

[0021] ; ;

[0022] in, is the estimated phase equilibrium constant,

[0023] ;

[0024] ;

[0025] ;

[0026] Step 3: Calculate the liquid phase composition and gas phase composition ;

[0027] ; ;

[0028] in, is the calculated phase equilibrium constant

[0029] ;

[0030] ;

[0031] , ;

[0032] ;

[0033] ;

[0034] in, , , , , , are the binary interaction parameters of the substances, T is the temperature, x is the mole fraction of the component, , , , Respectively represent , , , components, is the fugacity coefficient;

[0035] Step 4:

[0036] Calculate the gas phase fraction ;

[0037] ;

[0038] in, is the number of components, is the gas phase fraction;

[0039] Step 5: Convergence judgment, estimated liquid phase composition and gas phase composition And the calculated liquid composition and gas phase composition Determine whether it converges, if it converges, output the liquid phase composition , gas phase composition , phase equilibrium constant and gas phase fraction ; Otherwise, proceed to step 2.

[0040] Preferably, the method for pH flash calculation comprises:

[0041] S1: known enthalpy H and pressure P;

[0042] S2: Assume that the temperature is ;

[0043] S3: perform PT flash calculation;

[0044] S4: Calculate enthalpy ;

[0045] ;

[0046] in, is the molar coefficient of the component, is a constant, is the fugacity coefficient of the component, It is the heat of vaporization;

[0047] S5: Determine enthalpy value and enthalpy value H error. When the error is within the tol range, the output temperature T and liquid phase composition , gas phase composition and gas phase fraction e, otherwise re-estimate T and return to S3 to perform PT flash calculation.

[0048] As a preferred method, the Grignard reaction model is established by taking the reaction mechanism mathematical model as prior knowledge and embedding it into the feedforward neural network. For the all-liquid Grignard reaction, its kinetic model is:

[0049]

[0050] in, is the reaction rate, is the reaction rate constant for the forward reaction, is the concentration of RMgX, is the concentration of R'COR'', represents the reaction rate constant for the reverse reaction, is the concentration of RR'R''COMgX.

[0051] In order to solve the above technical problems, the present invention also provides a digital twin system based on Grignard reaction, which includes:

[0052] A module for acquiring sample data, wherein the sample data includes the temperature of the Grignard reactor, the pressure of the Grignard reactor, the raw material flow rate of the reactants, the reaction time of the reactants, the power of the stirring paddle motor of the Grignard reactor, the flow rate of the output materials and the composition of the outlet reactants;

[0053] A sample data filtering module, the sample data filtering module filters the acquired sample data;

[0054] A reactant concentration calculation module, which is used to calculate the reactant concentration of the filtered sample data through a thermodynamic model;

[0055] The module for establishing the Grignard reaction model establishes a reaction kinetic model based on the calculated reactant concentrations using the RBF neural network model, thereby determining the Grignard reaction model.

[0056] In order to solve the above technical problems, the present invention also provides a storage medium, which includes a storage medium implemented by a digital twin system based on the Grignard reaction.

[0057] In order to solve the above technical problems, the present invention also provides an electronic device, which includes an electronic device implemented by a digital twin system based on Grignard reaction.

[0058] The present invention has significant technical effects due to the adoption of the above technical solution:

[0059] The reaction kinetics of the Grignard reaction was modeled based on the RBF (radial basis function) neural network model. By analyzing the operating characteristics of the reactor, it was found that the factors affecting the reaction include reaction temperature, reaction pressure, reactant concentration, feeding method, stirring speed, reaction residence time and other factors. These factors will affect the conversion rate of the reaction. Therefore, the established Grignard reaction neural network model selects these six main factors as the input variables of the neural network, and takes the content of the Grignard reagent at the reactor outlet as the output of the neural network to construct a training data set with 6 inputs and one output.

[0060] The training of RBF neural network requires a large number of data samples. In order to reduce the amount of data samples and improve the accuracy of the neural network model, the neural network model needs to be improved.

[0061] The present invention takes the reaction mechanism mathematical model and the thermodynamic model as prior knowledge and embeds them into a feedforward neural network. The introduction of prior knowledge can make up for the inherent defects of the neural network, reduce the dimension of the neural network, and thus reduce the sample size of the neural network training.

[0062] The factors input in the training data set, including reactant concentration, feeding method and stirring speed, cannot be measured directly. The feeding method directly affects the concentration of the reactants. Therefore, only the reactant concentration needs to be obtained to know the effect of the feeding method on the reaction conversion rate. The reactant concentration can be calculated by combining the feed raw material flow rate of each reaction with the thermodynamic calculation model. Therefore, the input of the reactant concentration can be changed to the raw material flow rate of each reaction. The stirring speed cannot be measured directly, but the stirring speed is proportional to the stirring paddle motor power, so the stirring motor power can be used to replace the stirring rate input. The sample data source is obtained from the process history data of the production device. The data includes Grignard reactor temperature, pressure, raw material flow rate of each reactant, reaction time, stirring paddle motor power, output material flow rate and outlet reactant composition. All data of the entire production cycle are obtained to train the RBF neural network model. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 It is a flow chart of the present invention.

[0064] Figure 2 It is the PT flash distillation flow chart of the present invention.

[0065] Figure 3 It is a pH flash distillation flow chart of the present invention. DETAILED DESCRIPTION

[0066] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments.

[0067] Example 1

[0068] A digital twin system based on Grignard reaction, the method comprising:

[0069] Acquisition of sample data; the sample data includes the temperature of the Grignard reactor, the pressure of the Grignard reactor, the raw material flow rate of the reactants, the reaction time of the reactants, the power of the stirring paddle motor of the Grignard reactor, the flow rate of the output material and the composition of the outlet reactants;

[0070] Sample data filtering: Filter the acquired sample data; To make the sample data accurately reflect the actual production situation, the sample data must be filtered to eliminate the data that obviously deviates from the normal value. The normal value is generally about 30% of the design value. These data may be caused by factors such as instrument abnormality.

[0071] Calculation of reactant concentration: for filtered sample data, the reactant concentration is calculated using a thermodynamic model;

[0072] The establishment of the Grignard reaction model is based on the calculated reactant concentrations and the reaction kinetic model is established using the RBF neural network model to determine the Grignard reaction model.

[0073] Factors such as reaction temperature, reaction pressure, reactant concentration, feeding method, stirring speed, and reaction residence time will affect the conversion rate of the reaction. Therefore, the established Grignard reaction neural network model selects these six main factors as the input variables of the neural network, and takes the concentration of each substance at the reactor outlet as the output of the neural network to construct a training data set with 6 inputs and one output; the RBF neural network model is used to establish a reaction kinetic model to determine the Grignard reaction model.

[0074] The calculation of reactant concentrations includes:

[0075] The enthalpy of each reactant is calculated before mixing, and the enthalpy of each reactant before mixing is obtained by performing PT flash calculation on the reactants through a thermodynamic model;

[0076] Calculation of reactants after mixing, and then perform pH flash calculation on the mixed reactants to obtain the mole fraction and temperature of each component;

[0077] Calculate the concentration of the mixed reactants and obtain the concentration of the mixed reactants. ;

[0078]

[0079] In the formula, is the molar flow rate of the total logistics, is the liquid phase mole fraction of the component, is the volume of the reactor.

[0080] Methods for PT flash calculation include:

[0081] Step 1: Determine the flow rate F of the reactants and the composition of the reactants , pressure is P, temperature is ;

[0082] Step 2: Estimate the liquid phase composition and gas phase composition ,

[0083] ; ;

[0084] in, is the estimated phase equilibrium constant,

[0085] ;

[0086] ;

[0087] ;

[0088] Step 3: Calculate the liquid phase composition and gas phase composition ;

[0089] ; ;

[0090] in, is the calculated phase equilibrium constant

[0091] ;

[0092] ;

[0093] , ;

[0094] ;

[0095] ;

[0096] in, , , , , , are the binary interaction parameters of the substances, T is the temperature, x is the mole fraction of the component, , , , Respectively represent , , , components, is the fugacity coefficient;

[0097] Step 4:

[0098] Calculate the gas phase fraction ;

[0099] ;

[0100] in, is the number of components, is the gas phase fraction;

[0101] Step 5: Convergence judgment, estimated liquid phase composition and gas phase composition And the calculated liquid composition and gas phase composition Determine whether it converges, if it converges, output the liquid phase composition , gas phase composition , phase equilibrium constant and gas phase fraction ; and the output liquid phase composition The calculated liquid phase composition ; Output gas phase composition The calculated gas phase composition ; Output phase equilibrium constant is the calculated phase equilibrium constant ; Output gas phase fraction ; Otherwise, proceed to step 2.

[0102] Methods for pH flash calculation include:

[0103] S1: known enthalpy H and pressure P;

[0104] S2: Assume that the temperature is ;

[0105] S3: perform PT flash calculation;

[0106] S4: Calculate enthalpy ;

[0107] ;

[0108] in, is the molar coefficient of the component, is a constant, is the fugacity coefficient of the component, It is the heat of vaporization;

[0109] S5: Determine enthalpy value and enthalpy value H error. When the error is within the tol range, the output temperature T and liquid phase composition , gas phase composition and gas phase fraction e, otherwise re-estimate T and return to S3 to perform PT flash calculation.

[0110] The establishment of the Grignard reaction model takes the mathematical model of the reaction mechanism as prior knowledge and embeds it into the feedforward neural network. For the Grignard reaction in the full liquid phase, its kinetic model is:

[0111]

[0112] in, is the reaction rate, is the reaction rate constant for the forward reaction, is the concentration of RMgX, is the concentration of R'COR'', represents the reaction rate constant for the reverse reaction, is the concentration of RR'R''COMgX.

[0113] Example 2

[0114] Based on Example 1, the present invention further provides a digital twin system based on Grignard reaction, which includes:

[0115] A module for acquiring sample data, wherein the sample data includes the temperature of the Grignard reactor, the pressure of the Grignard reactor, the raw material flow rate of the reactants, the reaction time of the reactants, the power of the stirring paddle motor of the Grignard reactor, the flow rate of the output materials and the composition of the outlet reactants;

[0116] A sample data filtering module, the sample data filtering module filters the acquired sample data;

[0117] A reactant concentration calculation module, which is used to calculate the reactant concentration of the filtered sample data through a thermodynamic model;

[0118] The module for establishing the Grignard reaction model establishes a reaction kinetic model based on the calculated reactant concentrations using the RBF neural network model, thereby determining the Grignard reaction model.

[0119] Example 3

[0120] Based on the above embodiment, a storage medium includes a storage medium implemented by a digital twin system based on a Grignard reaction.

[0121] Example 4

[0122] Based on the above embodiments, this embodiment further provides an electronic device, which includes an electronic device implemented by a digital twin system based on Grignard reaction.

Claims

1. A digital twin system based on Grignard reaction, comprising: A module for acquiring sample data, wherein the sample data includes the temperature of the Grignard reactor, the pressure of the Grignard reactor, the raw material flow rate of the reactants, the reaction time of the reactants, the power of the stirring paddle motor of the Grignard reactor, the flow rate of the output materials and the composition of the outlet reactants; A sample data filtering module, the sample data filtering module filters the acquired sample data; A reactant concentration calculation module, which is used to calculate the reactant concentration of the filtered sample data through a thermodynamic model; The module for establishing the Grignard reaction model uses the RBF neural network model to establish a reaction kinetic model based on the calculated reactant concentrations, thereby determining the Grignard reaction model; the calculation of reactant concentrations includes: The enthalpy of each reactant is calculated before mixing, and the enthalpy of each reactant before mixing is obtained by performing PT flash calculation on the reactants through a thermodynamic model; Calculation of reactants after mixing, and then perform pH flash calculation on the mixed reactants to obtain the mole fraction and temperature of each component; Calculate the concentration of the mixed reactants and calculate the concentration of the mixed reactants to obtain the mixed reactant component concentration C i ; C i =F*x i / V In the formula, F is the molar flow rate of the total logistics, x i is the liquid phase composition, V is the volume of the reactor, and i represents the i-th component; Methods for PT flash calculation include: Step 1: Determine the flow rate F of the reactants and the composition z of the reactants i , pressure is P, temperature is T; Step 2: Estimate liquid phase composition x1 i and gas phase composition y1 i , x1 i =from i / k1 i ;y1 i =from i *k1 i ; Among them, k1 i is the estimated phase equilibrium constant, k1 i =Exp(5.37*(1+ω)*(1-1 / T r )P r ); T r =T / T c ; P r =P / P c ; Step 3: Calculate liquid phase composition x2 i and gas phase composition y2 i ; <h2 style=";text-align:left;direction:ltr">x2<h2 style=";text-align:left;direction:ltr"> i <h2 style=";text-align:left;direction:ltr"> =z<h2 style=";text-align:left;direction:ltr"> i <h2 style=";text-align:left;direction:ltr"> / k2<h2 style=";text-align:left;direction:ltr"> i <h2 style=";text-align:left;direction:ltr"> ;y2<h2 style=";text-align:left;direction:ltr"> i <h2 style=";text-align:left;direction:ltr"> =z<h2 style=";text-align:left;direction:ltr"> i <h2 style=";text-align:left;direction:ltr"> *k2<h2 style=";text-align:left;direction:ltr"> i <h2 style=";text-align:left;direction:ltr"> ; Among them, k2 i is the calculated phase equilibrium constant k2 i =Exp(lnγ i ); G ji =exp(-a ji τ ji ),a ji =a ij ; a ij =c ij +d ij T; Among them, a ij , b ij , c ij , d ij , e ij , f ij are the binary interaction parameters of the substances, T is the temperature, x is the mole fraction of the component, i, j, l, r represent the i-th, j-th, l-th, r-th components respectively, and γ is the fugacity coefficient; Step 4: Calculate the gas phase fraction e; Where c is the number of components and e is the gas phase fraction; Step 5: Convergence judgment, estimated liquid phase composition x1 i and gas phase composition y1 i and calculated liquid phase composition x2 i and gas phase composition y2 i Determine whether it converges. If it converges, output the liquid phase composition x i , gas phase composition i , phase equilibrium constant k i and gas phase fraction e; otherwise, return to step 2; the method for calculating pH flash evaporation includes: S1: known enthalpy H and pressure P; S2: Assume the temperature is T; S3: perform PT flash calculation; S4: Calculate enthalpy H 计算值 ; Among them, x i is the molar coefficient of the component, R is a constant, γ i is the fugacity coefficient of the component, deltaVap is the heat of vaporization; S5: Determine enthalpy value H 计算值 The error with the enthalpy value H, when the error is within the tol range, the output temperature T, liquid phase composition x i , gas phase composition i and gas phase fraction e, otherwise re-estimate T and return to S3 to perform PT flash calculation.

2. According to claim 1, a digital twin system based on Grignard reaction is characterized in that: The establishment of the Grignard reaction model takes the reaction kinetic model as prior knowledge and embeds it into the feedforward neural network. For the Grignard reaction, its kinetic model is: R=k c C A C B -k ′ c C L Where R is the reaction rate, k c is the reaction rate constant of the forward reaction, C A is the concentration of RMgX, C B is the concentration of R'COR", k ′ c represents the reaction rate constant of the reverse reaction, C L is the concentration of RR'R"COMgX.

3. A storage medium, characterized in that: A storage medium implemented by a digital twin system based on a Grignard reaction as described in any one of claims 1-2.

4. An electronic device, characterized in that: It includes an electronic device implemented by a digital twin system based on the Grignard reaction as described in any one of claims 1-2.

Images