Method for solving design problem of partition plate tower based on proxy model

By using the agent model assisted optimization method of individual punishment in partition tower design and the alternative model of fully thermally coupled tower, the problems of limited prediction accuracy and noise data in the prior art are solved, and efficiently finding partitions under limited resources are achieved The ability to optimally design towers.

CN119049572APending Publication Date: 2024-11-29TIANJIN UNIV
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202411071556.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-06
Publication Date
2024-11-29

AI Technical Summary

Technical Problem

The existing process optimization-oriented agent model assisted optimization method handles constraint functions and noise data with limited prediction accuracy in partition tower design, making it difficult to find the optimal process structure and operating parameters within a limited number of simulations, affecting the Optimization effect of partition tower design.

Method used

Using the agent model assisted optimization method (IP-SALS) based on individual punishment, the constraint function is processed by generating individual punishment coefficients, and the local proxy model of the objective function and constraint function is constructed using the nearest neighbor individual to predict the optimal candidate solution, and the overall solution is Independent material flow strands and energy flow strands are used in the thermal coupling tower to replace the coupled flow strands, improving the convergence success rate and prediction accuracy of the simulation.

Benefits of technology

With limited computing resources, the convergence speed and quality of the optimization process of partition tower design is significantly improved, and the optimal solution can be found within a limited number of simulations, ensuring the prediction accuracy of the built proxy model, reducing noise data, and finding the best Good tower structure and operating parameters.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119049572A_ABST
    Figure CN119049572A_ABST
Patent Text Reader

Abstract

The invention relates to a method for solving a partition plate tower design problem based on an agent model, belongs to the field of chemical design, and is used for solving the problem of key parameter determination during partition plate tower design. Specifically, an agent model aided optimization method (IP-SALS) based on individual penalty is adopted; an individual penalty coefficient and an agent model are introduced to assist local search and are combined with a partition plate tower model in simulation software, independent material streams and energy streams are used for replacing coupling streams, a proper initial value is provided for strict simulation of a partition plate tower, and an optimal solution can be found in limited computing resources. According to the method, the convergence speed and quality of the optimization process can be greatly improved, and the optimization result can be used as a basis for designing the partition plate tower.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the field of chemical engineering design, and in particular relates to a method for solving a partition tower design problem based on an agent model. Background Art

[0002] The baffle tower is an operating unit and equipment widely used for separation and purification in the petroleum and chemical production process. Compared with conventional distillation towers, the baffle tower significantly improves the energy efficiency of the distillation system by suppressing the back-mixing phenomenon. In addition, the baffle tower integrates the two originally independent tower processes through thermal coupling technology, effectively reducing the volume of the tower body and reducing the investment cost of the equipment.

[0003] The baffle tower and the fully thermally coupled tower are thermodynamically equivalent, and the baffle tower is actually another form of the fully thermally coupled tower. Therefore, the fully thermally coupled tower model is usually used to simulate the separation process of the baffle tower. Since the fully thermally coupled tower has more decision variables and the variables are coupled with each other, the design of the baffle tower is more complicated than that of the traditional distillation tower. In addition, the fully thermally coupled tower contains two nested loops, and the simulation is difficult to converge. If the simulation cannot converge, the result is noisy data, which will weaken the performance of the optimization method and cannot guarantee the optimal solution for the baffle tower design.

[0004] In order to guide the optimization method to find the optimal solution of the baffle tower separation process under limited computing resources, a surrogate model can be constructed to predict promising candidate solutions and improve the optimization efficiency. However, the existing surrogate model-assisted optimization methods for process optimization have some shortcomings, such as limited prediction accuracy, empirical treatment of constraint functions and noisy data, which restrict the ability of the optimization method to find the best process structure and operating parameters within a limited number of simulations. Summary of the invention

[0005] The present invention proposes an individual penalty-based proxy model assisted optimization method (IP-SALS) to solve the problem of determining key parameters in the design of partition towers. During the operation of the method, individual penalty coefficients are first generated to process the constraint function, and then the local proxy models of the objective function and the constraint function are respectively constructed using the nearest neighbor individuals. The best candidate solution is predicted by the proxy model, and the optimal solution can be found within a limited number of simulations. In the separation process of the simulated partition tower, independent material streams and energy streams are used to replace the coupled streams in the full thermal coupling tower to provide suitable initial values ​​for the simulation of the full thermal coupling tower, improve the convergence success rate of the simulation, reduce noise data, and ensure the prediction accuracy of the constructed proxy model. When using this method to solve the optimization problem of three groups of partition towers, the optimal tower structure and operating parameters can be found as the design result under limited computing resources.

[0006] A method for solving the design problem of a baffle tower based on an agent model, the main process of the method is to connect the process simulation software and the IP-SALS algorithm program, and use IP-SALS to continuously adjust the decision variables to find the optimal solution of the design. The present invention is characterized in that the method further includes the following steps in sequence:

[0007] Step 1: Determine the information of the partition tower separation process to be solved, including the number of components, feed information, and product requirement information; determine the decision variables to be solved; determine the objective function Its function value is used to evaluate the quality of the design; determine the constraint function Its function value is used to determine whether the product purity requirements are met; the design problem of the baffle tower is converted into an optimization problem; the maximum number of function evaluations is MaxFEs;

[0008] Step 2: Establish the actual model and substitute model of the baffle tower simulation in the process simulation software, and connect the process simulation software and the IP-SALS self-written algorithm program;

[0009] Step 3: Obtain the initial population through sampling. The population size is NP, and the individuals in the population are vectors composed of D decision variables. Iteration number t = 1;

[0010] Step 4: Input the individual decision variables into the process simulation software and simulate the alternative model of the baffle tower;

[0011] Step 5: Input the simulation results of the substitute model into the actual model as the initial simulation value of the actual model;

[0012] Step 6: Simulate the actual model of the baffle tower and bring the simulation results into the objective function and constraint function to calculate the individual objective function value and constraint function value Set the number of function evaluations FEs = NP;

[0013] Step 7: Generate individuals The penalty coefficient r i ; Calculate the penalty function value of the individual

[0014] Step 8: Find the individual The K nearest neighbor individuals of the target function are constructed by using the K nearest neighbor individuals.

[0015] Step 9: Get the best candidate solution predicted by the surrogate model

[0016] Step 10: Input the decision variables of the best candidate solution into the process simulation software and simulate the alternative model of the baffle tower;

[0017] Step 11: Input the simulation results of the substitute model into the actual model as the simulation initial values ​​of the actual model;

[0018] Step 12: Simulate the actual model of the baffle tower and bring the simulation results into the objective function and constraint function to calculate the objective function value of the best candidate solution and constraint function value Calculate the penalty function value of the best candidate solution

[0019] Step 13: Execute the selection operation: if the penalty function value of the best candidate solution is less than the penalty function value of the current individual, replace the current individual with the best candidate solution; otherwise, keep the current individual; set FEs = FEs + 1;

[0020] Step 14: Determine whether FEs>MaxFEs holds true: If so, output the optimal tower structure and operating parameters as the design result; otherwise, set t=t+1 and return to step 7.

[0021] Preferably, in step 1, the optimization problem transformed from the design problem of the partition wall tower is as follows:

[0022] Min

[0023] stMESH equations

[0024]

[0025] in, is a vector consisting of D decision variables, where the decision variables are selected from the number of plates, operating pressure, gas phase split ratio, liquid phase split ratio, reflux ratio, side product flow rate and bottom product flow rate; is the objective function, and its function value is used to describe the quality of the design, including the annualized total cost or energy consumption. The smaller the objective function value, the better the design. MESH equations are the material balance equation, phase balance equation, component fraction normalization equation and heat balance equation of the baffle tower. is the kth constraint function, and its function value is used to describe the purity requirement of the kth product; m represents the number of constraint functions.

[0026] Preferably, in step 2, an actual model and a substitute model for baffle tower simulation are established in the process simulation software, which is implemented as follows:

[0027] The actual model of the baffle tower is a fully thermally coupled tower model; the two side streams connecting the distillation section of the first tower and the second tower in the actual model are equivalently replaced by a saturated steam material flow and an energy flow, and then the two side streams connecting the distillation section of the first tower and the second tower in the actual model are equivalently replaced by a saturated liquid material flow and an energy flow, so as to obtain an alternative model of the baffle tower.

[0028] Preferably, obtaining the initial population by sampling in step 3 is achieved by random sampling or Latin hypercube sampling.

[0029] Preferably, the step of generating individual The penalty coefficient r i , achieved by adopting any penalty coefficient generation strategy.

[0030] Preferably, in step 7, the penalty function value of the individual The calculation formula is as follows:

[0031]

[0032] in, Represents an individual The objective function value, r i Represents an individual The penalty coefficient, Represents an individual The constraint violation value of Represents an individual The kth constraint function value of , where m represents the number of constraint functions.

[0033] Preferably, in step eight, find the individual The K nearest neighbor individuals are implemented as follows:

[0034] Calculate individual The distance between all evaluated individuals is found The nearest K individuals, where the distance between individuals is the Euclidean distance

[0035] Preferably, in step eight, a proxy model of the objective function and the constraint function is constructed using K nearest neighbor individuals, which is implemented as follows:

[0036] According to the K nearest neighbor individuals and their objective function values, the objective function is obtained through training. Radial Basis Function Surrogate Model According to the K nearest neighbor individuals and their constraint function values, the constraint function is trained Radial Basis Function Surrogate Model

[0037] Preferably, in step nine, the best candidate solution predicted by the proxy model is obtained The implementation is as follows:

[0038] Use differential evolution operator to generate M experimental offspring according to Evaluate all experimental offspring and get is the best candidate solution, where Represents the surrogate model of the objective function, r i Represents an individual The penalty coefficient, a proxy model representing constraint violations, represents the proxy model of the kth constraint function, and m represents the number of constraint functions.

[0039] Technical effects and advantages of the present invention: In the application test in the chemical field, the present invention shows good performance in sensitivity analysis, performance analysis and result analysis. For the optimization problem of the baffle tower, the present invention can greatly improve the convergence speed and quality of the optimization process, and can find the optimal solution within limited computing resources. The optimization result can be used as the basis for the design of the baffle tower. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a flow chart of the core part of the invention;

[0041] Figure 2 It is a schematic diagram of the structure of a three-group partition plate tower;

[0042] Figure 3 It is a schematic diagram of the actual model of the baffle tower simulation;

[0043] Figure 4 It is a schematic diagram of an alternative model for baffle tower simulation;

[0044] Figure 5 This is the optimal design solution obtained by the present invention. DETAILED DESCRIPTION

[0045] This embodiment uses simulation software to model the baffle tower, and then connects the IP-SALS algorithm program to automatically solve the problem of determining key parameters when designing the baffle tower. The key to the design of the baffle tower lies in the determination of 11 decision variables such as different number of plates, operating pressure, gas phase split ratio, liquid phase split ratio, reflux ratio, side product flow rate and tower bottom product flow rate.

[0046] In order to make the description of the present invention more detailed, this section takes three groups of partition plate towers as an example to illustrate the specific implementation of the present invention. Here, the schematic implementation of the present invention and its description are used to explain the present invention, but are not intended to limit the present invention.

[0047] For the structure of three-group partition plate tower, see Figure 2 , the actual model of the baffle tower simulation can be found in Figure 3 , an alternative model for baffle tower simulation can be found in Figure 4 The three-component mixture feed to be separated is input into the tower, and high-purity light component products and heavy component products are produced through the tower top and the tower bottom. High-purity intermediate component products can be obtained at a reasonable side line extraction position. The purpose of this embodiment is to quickly solve the optimal number of plates, operating pressure, gas phase split ratio, liquid phase split ratio, reflux ratio, side line product flow rate and tower bottom product flow rate of each tower section.

[0048] like Figure 1 As shown, in this embodiment, the method for solving the baffle tower design problem based on the agent model specifically includes the following contents:

[0049] Step 1: Determine the information of the partition tower separation process to be solved, including the number of components, feed information, and product requirement information: In this case, the number of components is 3, the feed is an equal molar mixture of benzene, toluene and xylene, the feed flow rate is 300kmol / h, the feed pressure is 500kPa, the feed temperature is 323.15K, and the molar fractions of the corresponding benzene, toluene and xylene in the product streams at the top, side line and bottom of the tower are all ≥99.9%.

[0050] Determine the decision variables that need to be solved. The decision variables are the number of plates in each tower section (N i , i=1,2,3,4,5), operating pressure P, liquid flow rate F on the upper right edge of the partition L , gas phase flow rate F at the lower right edge of the partition V , reflux ratio R, side product flow rate F S and the bottom product flow rate F B .

[0051] Determine the objective function The function value is used to evaluate the design. In this example, the objective function is the annualized total cost (TAC) of the baffle tower, and its calculation formula is as follows:

[0052]

[0053] Among them, the total operating cost (TAOC) includes the cost of cold and hot utilities; the total equipment cost (TEC) includes the cost of equipment such as pumps, heat exchangers, tower shells and tower internals; the cost payback period is 3 years; the total operating cost and total equipment cost are determined by the values ​​of the decision variables.

[0054] The optimization problem transformed from the design problem of the partition wall tower is as follows:

[0055] MinTAC=f(N1,N2,N3,N4,N5,P,F L ,F V ,R,F S ,F B )

[0056] stMESH equations

[0057] g1(N1,N2,N3,N4,N5,P,F L ,F V ,R,F S ,F B )=0.999-x D_BEN ≤0

[0058] g2(N1,N2,N3,N4,N5,P,F L ,F V ,R,F S ,F B )=0.999-x S_TOL ≤0

[0059] g3(N1,N2,N3,N4,N5,P,F L ,F V ,R,F S ,F B )=0.999-x B_XYL ≤0

[0060] Among them, N i , i=1,2,3,4,5 are the number of plates above the partition, the number of plates from the upper edge of the partition to the feed plate, the number of plates from the feed plate to the lower edge of the partition, the number of plates below the partition and the position of the side line extraction plate respectively; P is the operating pressure; F L is the liquid phase flow rate at the upper right edge of the partition (instead of the liquid phase split ratio in this embodiment); F V is the gas phase flow rate at the lower right edge of the partition (instead of the gas phase split ratio in this embodiment); R is the reflux ratio; F S is the side product flow rate; F B is the product flow rate of the tower bottom; MESHequations is the material balance equation, phase balance equation, component fraction normalization equation and heat balance equation of the baffle tower; x D_BEN 、x S_TOL and x B_XYL The mole fractions of benzene, toluene and xylene in the top, side and bottom product streams, respectively.

[0061] Step 2: Create a practical model for the baffle tower simulation in the process simulation software AspenPlus (see Figure 3 ) and alternative models (see Figure 4 ), connect AspenPlus and IP-SALS algorithm program;

[0062] Step 3: Obtain the initial population through sampling. The population size is NP = 30. The individuals in the population are vectors composed of 11 decision variables. Iteration number t = 1; Maximum function evaluation number MaxFEs = 20000;

[0063] Step 4: Input the individual decision variables into Aspen Plus and simulate the alternative model of the baffle tower;

[0064] Step 5: Input the simulation results of the substitute model into the actual model as the initial simulation value of the actual model;

[0065] Step 6: Simulate the actual model of the baffle tower and bring the simulation results into the objective function and constraint function to calculate the individual objective function value and constraint function value Set the function evaluation times FEs = 30;

[0066] Step 7: Generate individuals The penalty coefficient r i : Set the lower bound r of the penalty coefficient min =0, the upper bound of the penalty coefficient Among them, t represents the number of iterations, fr represents the proportion of feasible individuals in the current population, and the penalty coefficient r i In [r min ,r max ] Randomly select a value within ; calculate the penalty function value of the individual

[0067] Step 8: Find the individual The K nearest neighbor individuals of the training set are: Radial Basis Function Surrogate Model According to the K nearest neighbor individuals and their constraint function values, the constraint function is trained and Radial Basis Function Surrogate Model and

[0068] Step 9: Get the best candidate solution predicted by the surrogate model Use differential evolution operator to generate 2000 experimental offspring according to Evaluate all experimental offspring and get is the best candidate solution, where Represents the surrogate model of the objective function, r i Represents an individual The penalty coefficient, a proxy model representing constraint violations, Represents the surrogate model for the k-th constraint function.

[0069] Step 10: Input the decision variables of the best candidate solution into Aspen Plus and simulate the alternative model of the baffle tower;

[0070] Step 11: Input the simulation results of the substitute model into the actual model as the simulation initial values ​​of the actual model;

[0071] Step 12: Simulate the actual model of the baffle tower and use the simulation results to enter the objective function and constraint function to calculate the objective function value of the best candidate solution. and constraint function value Calculate the penalty function value of the best candidate solution

[0072] Step 13: Execute the selection operation: if the penalty function value of the best candidate solution is less than the penalty function value of the current individual, replace the current individual with the best candidate solution; otherwise, keep the current individual; set FEs = FEs + 1;

[0073] Step 14: Determine whether FEs>MaxFEs holds true: if so, output the optimal tower structure and operating parameters as the design result; otherwise, set t=t+1 and return to step 7; Figure 5 The optimal design solution obtained in this example is shown.

[0074] For the optimization problem of the baffle tower, the present invention can greatly improve the convergence speed and quality of the optimization process, can find the optimal solution within limited computing resources, and its optimization result can be used as the basis for the design of the baffle tower.

Claims

1. A method for solving the baffle tower design problem based on surrogate model, It is characterized in that The following steps are involved: Step 1: Determine the information of the partition tower separation process to be solved, including the number of components, feed information, and product requirement information; Determine the decision variables that need to be solved; Determine the objective function Its function value is used to evaluate the quality of the design; determine the constraint function Its function value is used to determine whether the product purity requirements are met; the design problem of the baffle tower is converted into an optimization problem; the maximum number of function evaluations is MaxFEs; Step 2: Establish the actual model and substitute model of the baffle tower simulation in the process simulation software, and connect the process simulation software and the IP-SALS self-written algorithm program; Step 3: Obtain the initial population through sampling. The population size is NP, and the individuals in the population are vectors composed of D decision variables. Iteration number t = 1; Step 4: Input the individual decision variables into the process simulation software and simulate the alternative model of the baffle tower; Step 5: Input the simulation results of the substitute model into the actual model as the initial simulation value of the actual model; Step 6: Simulate the actual model of the baffle tower and bring the simulation results into the objective function and constraint function to calculate the individual objective function value and constraint function value Set the number of function evaluations FEs = NP; Step 7: Generate individuals The penalty coefficient r i ; Calculate the penalty function value of the individual Step 8: Find the individual The K nearest neighbor individuals of the target function are constructed by using the K nearest neighbor individuals. Step 9: Get the best candidate solution predicted by the surrogate model Step 10: Input the decision variables of the best candidate solution into the process simulation software and simulate the alternative model of the baffle tower; Step 11: Input the simulation results of the substitute model into the actual model as the simulation initial values ​​of the actual model; Step 12: Simulate the actual model of the baffle tower and use the simulation results to enter the objective function and constraint function to calculate the objective function value of the best candidate solution. and constraint function value Calculate the penalty function value of the best candidate solution Step 13: Execute the selection operation: if the penalty function value of the best candidate solution is less than the penalty function value of the current individual, replace the current individual with the best candidate solution; otherwise, keep the current individual; set FEs = FEs + 1; Step 14: Determine whether FEs>MaxFEs holds true: If so, output the optimal tower structure and operating parameters as the design result; otherwise, set t=t+1 and return to step 7.

2. A method for solving the design problem of a baffle tower based on an agent model according to claim 1, Features: In step 1, the optimization problem transformed from the design problem of the partition wall column is as follows: stMESH equations in, is a vector of D decision variables, where the decision variables are selected from the number of plates, operating pressure, gas phase split ratio, liquid phase split ratio, reflux ratio, side product flow rate, and bottom product flow rate; is the objective function, and its function value is used to describe the quality of the design, including the annualized total cost or energy consumption. The smaller the objective function value, the better the design. MESHequations is the material balance equation, phase balance equation, component fraction normalization equation and heat balance equation of the baffle tower. is the kth constraint function, and its function value is used to describe the purity requirement of the kth product; m represents the number of constraint functions.

3. A method for solving the baffle tower design problem based on an agent model according to claim 1, Features: In step 2, the actual model and the alternative model for the baffle tower simulation are established in the process simulation software, which is implemented as follows: The actual model of the baffle tower is a fully thermally coupled tower model; the two side streams connecting the distillation section of the first tower and the second tower in the actual model are equivalently replaced by a saturated steam material flow and an energy flow, and then the two side streams connecting the distillation section of the first tower and the second tower in the actual model are equivalently replaced by a saturated liquid material flow and an energy flow, so as to obtain an alternative model of the baffle tower.

4. A method for solving the design problem of a baffle tower based on an agent model according to claim 1, Features: The initial population is obtained by sampling as described in step 3, and is implemented by random sampling or Latin hypercube sampling.

5. A method for solving the baffle tower design problem based on an agent model according to claim 1, Features: Generate individuals as described in step 7 The penalty coefficient r i , achieved by adopting any penalty coefficient generation strategy.

6. A method for solving the design problem of a baffle tower based on an agent model according to claim 5, Features: In step 7, the penalty function value of the individual The calculation formula is as follows: in, Represents an individual The objective function value, r i Represents an individual The penalty coefficient, Represents an individual The constraint violation value of Represents an individual The kth constraint function value of , where m represents the number of constraint functions.

7. A method for solving the design problem of a baffle tower based on an agent model according to claim 6, Features: In step eight, find the individual The K nearest neighbor individuals are implemented as follows: Calculate individual The distance between all evaluated individuals is found The nearest K individuals, where the distance between individuals is the Euclidean distance 8. A method for solving the design problem of a baffle tower based on an agent model according to claim 7, Features: In step eight, the K nearest neighbor individuals are used to construct a proxy model of the objective function and constraint function, which is implemented as follows: According to the K nearest neighbor individuals and their objective function values, the objective function is obtained through training. Radial Basis Function Surrogate Model According to the K nearest neighbor individuals and their constraint function values, the constraint function is trained Radial Basis Function Surrogate Model 9. A method for solving the baffle tower design problem based on an agent model according to claim 8, Features: In step nine, the best candidate solution predicted by the surrogate model is obtained The implementation is as follows: Use differential evolution operator to generate M experimental offspring according to Evaluate all experimental offspring and get is the best candidate solution, where Represents the surrogate model of the objective function, r i Represents an individual The penalty coefficient, a proxy model representing constraint violations, represents the proxy model of the kth constraint function, and m represents the number of constraint functions.

Citation Information

Patent Citations

  • Partition plate tower gas phase flow control scheme

    CN105498268A

  • Absorption-stabilization technique capable of enhancing absorption by using complementary energy

    CN107338068A

  • Method for solving isolation wall rectifying tower design problem based on heuristic algorithm

    CN114741772A

  • Multicomponent Dividing Wall Columns

    US20150119612A1