Polyurethane high-pressure foaming process parameter optimization method and system

Abstract

The invention discloses a polyurethane high-pressure foaming process parameter optimization method and system, and the method comprises the steps: building a polyurethane high-pressure foaming process model through Aspen Plus software, and calibrating the polyurethane high-pressure foaming process model through actual operation data; simulation operation data are obtained through the calibrated polyurethane high-pressure foaming process model; optimizing variables and constraint conditions are set, the total public engineering consumption minQ of the whole process is used as a target function, a BP neural network model is constructed, and a data set is constructed by using actual operation data and simulated operation data to train the BP neural network model; introducing the trained BP neural network model into a queue competition algorithm LCA to form a BP-LCA hybrid algorithm; and optimizing the parameters of the polyurethane high-pressure foaming process by using a BP-LCA hybrid algorithm. The method has the advantages of being high in optimization result search speed and good in optimization effect.

Description

Technical Field

[0001] The invention relates to chemical process parameter optimization, and in particular to a polyurethane high-pressure foaming process parameter optimization method and system. Background Art

[0002] Polyurethane materials have excellent thermal insulation, unique water permeability resistance and chemical stability, and are lightweight and easy to mold. They are gradually becoming the sixth largest synthetic material in the world and are widely used in various fields of the national economy. Polyurethane high-pressure foaming is simple to operate, has a short process flow, and high production efficiency, making it suitable for mass production. However, when operating high-pressure foaming machines, workers often rely on experience to adjust equipment parameters and process data (such as feed components, temperature, pressure, etc.), which may produce too many by-products and unreacted monomers, reduce product quality and production efficiency, and cause waste of resources and increase costs.

[0003] Aspen Plus is a commonly used process simulation software in the chemical industry. It can build a process model and then simulate and determine the production process data to achieve the purpose of improving production efficiency and resource utilization. At present, Aspen Plus software has not been introduced for simulation of polyurethane high-pressure foaming. In addition, most chemical process flows are composed of multiple independent process units, and process convergence must be achieved through the convergence of each process unit. Moreover, the process model may have non-convex nonlinear characteristics. When solving complex system processes, the process simulator involves many optimization variables, the solution effect is poor and the time is long. There may be multiple local optima during the optimization process. Therefore, even if Aspen Plus software is used for the optimization of polyurethane high-pressure foaming process parameters, there are still problems of long time consumption and poor optimization effect. Summary of the invention

[0004] Purpose of the invention: The purpose of the present invention is to provide a polyurethane high-pressure foaming process parameter optimization method that can quickly determine the optimization results and has good optimization effects; the second purpose of the present invention is to provide a polyurethane high-pressure foaming process parameter optimization system.

[0005] Technical solution: The polyurethane high-pressure foaming process parameter optimization method of the present invention comprises:

[0006] (1) Using Aspen Plus software to establish a polyurethane high-pressure foaming process model, and using actual operation data to calibrate the polyurethane high-pressure foaming process model; obtaining simulated operation data through the calibrated polyurethane high-pressure foaming process model;

[0007] (2) Set optimization variables and constraints, take the total utility consumption minQ of the whole process as the objective function, build a BP neural network model, and use the actual operation data and simulated operation data to build a data set to train the BP neural network model;

[0008] The optimization variables include the flow rate F1 of MDI, the flow rate F2 of PPG, the consumption F3 of N2, the initial temperature T1 of the raw material, the temperature T2 of MDI entering RCSTR2, the temperature T3 of PPG entering RCSTR1, the reaction temperature T4 of the mixture entering RCSTR3, the pipeline pressure P of stream 9, and the bottom extraction and feed ratio S of RCSTR3; stream 9 enters RCSTR3;

[0009] The constraints include the mole fraction C1 of MDI in RCSTR2, the mole fraction C2 of cyclopentane in RCSTR1, the mole fraction C3 of PPG in RCSTR1, the mole fraction C4 of PU in the product of RCSTR3, and the mole fraction C5 of water in the product of RCSTR3;

[0010] (3) The trained BP neural network model is introduced into the queue competition algorithm LCA to form a BP-LCA hybrid algorithm; the BP-LCA hybrid algorithm is used to optimize the polyurethane high-pressure foaming process parameters.

[0011] Furthermore, step (1) includes: selecting a mixing module Mixer, a heat exchange module Heater, a reaction module RCSTR and a pressure transmitter module Pump in Aspen Plus software to establish a polyurethane high-pressure foaming process model, inputting component data, physical property methods, and initial logistics data, running the polyurethane high-pressure foaming process model, and comparing the output result of the polyurethane high-pressure foaming process model with the actual operation data to determine the error. If the error is within the allowable range, the calibration is completed; otherwise, continue to modify the physical property method and adjust the model data.

[0012] Furthermore, the physical property method initially selects POLYSL.

[0013] Furthermore, in step (2), the total utility consumption Q is calculated using a problem table method.

[0014] Furthermore, the total utility consumption Q is calculated using the problem table method, including:

[0015] In the process flow, streams 1, 2, 3 → stream 4, stream 5 → stream 6, stream 9 → stream 10 are all hot logistics, which require external heat transfer, and all heat consumption is aggregated as the minimum heating utility consumption Q1; a refrigerator is required to quickly reduce the material temperature of stream 11 to room temperature, so stream 11 → stream 12 is cold logistics, and the energy consumed is the minimum cooling utility consumption Q2; Q1 and Q2 data can be retrieved from the software simulation data set; Q = Q1 + Q2.

[0016] Furthermore, in step (2), the optimization variable range is:

[0017]

[0018] Furthermore, the constraints are as follows:

[0019]

[0020] Among them, P is the process equation group described by the BP network; x = [F1, F2, F3, T1, T2, T3, T4, P, S] is the input variable vector of the BP neural network model; y = [C1, C2, C3, C4, C5, Q] is the output variable vector of the BP neural network model.

[0021] Furthermore, the BP neural network model includes an input layer, three hidden layers and an output layer. The number of input layer nodes is 9, the number of output layer nodes is 6, the number of hidden layer nodes is set to 8, the hidden layer transfer function selects logsig, and the output layer transfer function selects linear purelin.

[0022] Further, step (3) comprises:

[0023] (3-1) Randomly generate m individuals in a given search space to form an initial population, where the individuals are a set of optimization variables;

[0024] (3-2) Using the BP neural network model to calculate the objective function value of each individual;

[0025] (3-3) Sort the m individuals in the initial population according to the value of the objective function;

[0026] (3-4) According to the position of each individual in the queue, the search space of each family is allocated in a certain proportion. The search space of the first family is the smallest, while the search space of the last family is the largest.

[0027] (3-5) Each individual reproduces asexually within the assigned search range, producing n offspring individuals;

[0028] (3-6) Calculate the objective function values ​​of n offspring individuals through the BP neural network model, and let these offspring individuals compete for survival with their parent individuals. The best individuals are retained to participate in the next round of competition;

[0029] (3-7) Determine whether the termination condition is met: the objective function value is regionally stable after several iterations and no longer changes significantly; if the termination condition is met, output the result; otherwise, use the shrinkage factor to shrink the search space and return to step (3-3).

[0030] The polyurethane high-pressure foaming process parameter optimization system of the present invention comprises a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the steps of the polyurethane high-pressure foaming process parameter optimization method are implemented.

[0031] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: the present invention introduces a BP neural network algorithm combined with an LCA algorithm, which can avoid the problem of the LCA algorithm optimization parameters falling into local search, and can greatly shorten the calculation time, thereby achieving the effect of improving rapid search capabilities and global optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 This is the process flow chart of polyurethane high pressure foaming;

[0033] Figure 2 It is a polyurethane high-pressure foaming process model established by Aspen Plus software in the embodiment of the present invention;

[0034] Figure 3 is a prediction error graph of the BP neural network model in an embodiment of the present invention;

[0035] Figure 4 It is a block diagram of optimization calculation based on BP neural network and queue competition algorithm in an embodiment of the present invention. DETAILED DESCRIPTION

[0036] The present invention will be further described below in conjunction with the accompanying drawings.

[0037] The embodiment of the present invention provides a method for optimizing polyurethane high-pressure foaming process parameters, which specifically includes the following steps:

[0038] (1) Using Aspen Plus software to establish a polyurethane high-pressure foaming process model, and using actual operation data to calibrate the polyurethane high-pressure foaming process model; obtaining simulated operation data through the calibrated polyurethane high-pressure foaming process model;

[0039] like Figure 1 As shown, the polyurethane high-pressure foaming process mainly includes premixing, pressurization, mixing, injection, heating reaction and other sections, and the process conditions of each section are as follows.

[0040] Process conditions of premixing and stirring section: the temperature of tanks A and B is maintained at room temperature (20-30°C); PPG, catalyst, foaming agent, flame retardant and other additives are mixed and stirred in tank A; MDI is stirred in tank B to maintain activity; stirring time is 30 minutes.

[0041] Process conditions of the pressurization section: The metering pump accurately controls the flow rate and pressurizes the liquid with a pressure of 8 to 9 MPa.

[0042] Process conditions of the mixing section: The original liquids in tanks A and B are quickly transported to the mixing gun head under high pressure, and collide at high speed in the mixing chamber to be fully mixed.

[0043] Injection process conditions: At the high-pressure gun head, the two streams are combined into one, the pressure reaches 12-13 MPa, and the mixed liquid is injected into the mold with high pressure.

[0044] Process conditions of the heating reaction section: the mold into which the raw materials are injected is transported to the laminator via a crawler belt, and multiple reactions occur in the reaction chamber, and the materials are ripened and solidified, and finally cooled and demolded at room temperature. The temperature of the reaction chamber is 45-90°C.

[0045] The reaction equation involved in the heating reaction section is as follows:

[0046] Reaction of isocyanate and hydroxyl groups: Isocyanate groups (-NCO) react with hydroxyl groups (-OH) to form polyurethane chains. This is the core step in polyurethane synthesis and is usually accompanied by an exothermic reaction. The products generated include carbamate (-NHCOO-) and polyurethane chains.

[0047] R~NCO+OH~R'→R~NHCOO~R'

[0048] Reaction of isocyanate and water: Polyisocyanate reacts with water to form carbamic acid. Carbamic acid has poor stability and is easily further decomposed into amine and carbon dioxide.

[0049] R~NCO+H2O→R~NHCOOH→NH2~R+CO2↑

[0050] On this basis, the amine group reacts with isocyanate to form a polymer containing urea functional groups.

[0051] R~NCO+NH2~R'→R~NHCONH~R'

[0052] Allophanate reaction: The hydrogen on the nitrogen atom of the carbamate group reacts with isocyanate to form allophanate. This reaction involves the interaction of the carbamate group with isocyanate to form a compound with a urea structure.

[0053] R~NCO+R~NHCOO~R'→R~NHCONCOO~R'

[0054] Biuret reaction: Biuret is formed by the reaction of isocyanate with the hydrogen on the nitrogen atom in the urea group.

[0055] R~NCO+R~NHCONH~R'→R~NHCONCONH~R'

[0056] Using the process simulation software Aspen Plus, the corresponding unit modules (mixing module Mixer, heat exchange module Heater, reaction module RCSTR and pressure transmitter module Pump) are selected to establish the polyurethane high-pressure foaming process model, such as Figure 2 As shown, the physical property method selects POLYSL. After the model is established, it needs to be verified and calibrated through actual operation data. It can only be applied when the accuracy meets the error requirements. The specific calibration steps are as follows:

[0057] Step 1: Input component data, physical property methods, starting logistics data and all unit operation model data;

[0058] Step 2: Make sure the data input is complete, run the model, and output the simulation results;

[0059] Step 3: Determine the error range and compare it with the actual operation data to determine the error;

[0060] Step 4: If the error requirements are met, the model can be applied; otherwise, return to Step 1, check the data input, modify the physical property method, and adjust the model data.

[0061] (2) Set optimization variables and constraints, take the total utility consumption minQ of the whole process as the objective function, build a BP neural network model, and use the actual operation data and simulated operation data to build a data set to train the BP neural network model;

[0062] The present invention mainly studies the influence of the change of operating parameters in the polyurethane high-pressure foaming process on the total utility consumption under the condition of meeting the process requirements. Therefore, the main operating parameters will be used as the input value of the BP neural network model, and the main operating units involved in the process, with the content of each substance in the corresponding stream as the measurement standard, and the total utility consumption of the process as the output value of the network. After analysis, the input and output parameters of the BP neural network model are determined as follows:

[0063] Input parameters, namely the main 9 operating variables:

[0064] MDI flow rate F1, kmol·h -1 ;

[0065] PPG flow rate F2, kmol·h -1 ;

[0066] N2 consumption F3, kmol·h -1 ;

[0067] Initial temperature of raw materials (MDI, PPG, etc.) T1, °C;

[0068] The temperature of MDI entering RCSTR2 is T2, °C;

[0069] PPG temperature entering RCSTR1 is T3, °C;

[0070] The reaction temperature of the mixture entering RCSTR3 is T4, °C;

[0071] Pipeline pressure of stream 9, P, MPa;

[0072] The bottom extraction and feed ratio S of RCSTR3.

[0073] The output parameters are the six main indicators of the low-temperature methanol washing process:

[0074] The mole fraction of diphenylmethane diisocyanate (MDI) in RCSTR2 is C1;

[0075] Cyclopentane mole fraction C2 in RCSTR1;

[0076] The mole fraction of polypropylene glycol (PPG) in RCSTR1 is C3;

[0077] The mole fraction of polyurethane (PU) in RCSTR3 product is C4;

[0078] Water mole fraction C5 in RCSTR3 product;

[0079] Utility consumption Q, kW.

[0080] The BP algorithm is a feedforward multi-layer network algorithm, and its main feature is that the error will be transmitted backward. In addition, in the learning process, the neural network will continuously correct the threshold and weight of the network, and even the structure of the network, according to the input sample data, so that the output value of the network is as close as possible to the expected value. The present invention adopts a 5-layer BP neural network model, including an input layer, three hidden layers and an output layer, the number of input layer nodes is 9, the number of output layer nodes is 6, the number of hidden layer nodes is set to 8, the hidden layer transfer function selects logsig, and the output layer transfer function selects linear purelin.

[0081] 350 groups of actual operation data and simulated operation data were randomly selected for training the network model, and the remaining 33 groups of data were used for prediction. Figure 3 The comparison between the training error and the test error is shown. The results show that the test sample error is close to the training sample error, which indicates that the BP neural network model can not only fit the training data well, but also reflect the inherent laws of the process.

[0082] The total utility consumption of the whole process of polyurethane high-pressure foaming is taken as the target, the 9 input values ​​of the BP network model are taken as the optimization variables, and the first 5 output values ​​of the BP network model are taken as the constraints.

[0083] Objective function:

[0084] Q

[0085] Constraints:

[0086]

[0087] Among them, P is the process equation group described by the BP network, which includes the forward propagation equation, the error calculation equation, and the back propagation equation; x = [F1, F2, F3, T1, T2, T3, T4, P, S], which is the input variable vector of the BP neural network model; y = [C1, C2, C3, C4, C5, Q], which is the output variable vector of the BP neural network model.

[0088] Optimization variables:

[0089]

[0090] Total utility consumption calculation:

[0091] The total utility consumption Q is calculated by the problem table method, which can conveniently and quickly calculate the minimum utility consumption required by the heat exchange network. The specific calculation steps include:

[0092] In the process flow, streams 1, 2, 3 → stream 4, stream 5 → stream 6, stream 9 → stream 10 are all hot logistics, which require external heat transfer, and all heat consumption is aggregated as the minimum heating utility consumption Q1; a refrigerator is required to quickly reduce the material temperature of stream 11 to room temperature, so stream 11 → stream 12 is cold logistics, and the energy consumed is the minimum cooling utility consumption Q2; Q1 and Q2 data can be retrieved from the software simulation data set, Q = Q1 + Q2.

[0093] (3) The trained BP neural network model is introduced into the queue competition algorithm LCA to form a BP-LCA hybrid algorithm; the BP-LCA hybrid algorithm is used to optimize the polyurethane high-pressure foaming process parameters.

[0094] The queue competition algorithm is a global optimization algorithm with parallel search and multi-layer competition. In essence, the queue competition algorithm is also an evolutionary algorithm, and it also has operators such as reproduction, mutation, competition and selection. However, its main feature that is different from other algorithms is that the algorithm has always maintained independent parallel evolutionary families, and the evolution mechanism is different. The queue competition algorithm has both internal competition within the same family and competition between different families. Through the above two levels of competition, the family ranked first will be constantly replaced, so as to approach the optimal point at the fastest speed. The characteristic of the queue competition algorithm is that it not only ensures the quality of the global optimization solution, but also takes into account the convergence speed, and can also prevent the group from falling into the global optimal solution.

[0095] Combination Figure 4 The specific steps of optimizing polyurethane high-pressure foaming process parameters using the BP-LCA hybrid algorithm are as follows:

[0096] (3-1) Randomly generate m individuals in a given search space to form an initial population, where the individuals are a set of optimization variables;

[0097] (3-2) Using the BP neural network model to calculate the objective function value of each individual;

[0098] (3-3) Sort the m individuals in the initial population according to the value of the objective function;

[0099] (3-4) According to the position of each individual in the queue, the search space of each family is allocated in a certain proportion. The search space of the first family is the smallest, while the search space of the last family is the largest.

[0100] (3-5) Each individual reproduces asexually within the assigned search range, producing n offspring individuals;

[0101] (3-6) Calculate the objective function values ​​of n offspring individuals through the BP neural network model, and let these offspring individuals compete for survival with their parent individuals. The best individuals are retained to participate in the next round of competition;

[0102] (3-7) Determine whether the termination condition is met: the objective function value tends to be stable after several iterations and no longer changes significantly. If the termination condition is met, the result is output; otherwise, the search space is shrunk using the shrinkage factor and the process returns to step (3-3).

[0103] The optimization 1 results obtained by the BP-LCA hybrid algorithm, the optimization 2 results obtained by directly calling Aspen Plus for optimization using the queue competition algorithm, and the operating parameters of the original process are compared and listed in Table 1. The optimization results were input into the process software Aspen Plus for verification, and the total utility consumption under the operating conditions was calculated. Finally, the test values ​​of Aspen Plus, the predicted values ​​of the neural network model, and the output values ​​of the original operating plan were listed in Table 2.

[0104] Table 1 Comparison of operating parameters before and after optimization

[0105]

[0106] Table 2 Comparison of output values ​​before and after optimization

[0107]

[0108] After analyzing Table 1, it can be found that the optimized operating conditions reduce the amount of raw materials and nitrogen compared with the original process, and the nitrogen consumption is saved by 8.1%. As shown in Table 2, the optimized several main outlet concentrations are within the range of process index requirements, and the total utility consumption is saved by 13.1% compared with before optimization. The total utility consumption in the calculation scheme using the queue competition algorithm for direct optimization calculation is saved by 10.5% compared with the original process. Moreover, compared with the method of directly calling Aspen Plus to optimize the process by the queue competition algorithm, it takes 4 hours of calculation time to directly search and select 20 populations, 20 offspring, and 20 iterations. It only takes more than ten minutes to obtain the optimization calculation results by using the BP-LCA hybrid algorithm to replace the process for optimization, which greatly shortens the calculation time and makes it more convenient for operators to adjust the operating parameters in industrial production to reduce production costs and obtain greater economic benefits.

[0109] An embodiment of the present invention also provides a polyurethane high-pressure foaming process parameter optimization system, including a memory and a processor, wherein a computer program is stored in the memory. When the computer program is executed by the processor, the steps of the polyurethane high-pressure foaming process parameter optimization method described in the embodiment of the present invention are implemented.

Claims

1. A method for optimizing polyurethane high-pressure foaming process parameters, characterized in that: include: (1) Use Aspen Plus software to establish a polyurethane high-pressure foaming process model, and use actual operation data to calibrate the polyurethane high-pressure foaming process model; Acquire simulation operation data through a calibrated polyurethane high-pressure foaming process model; (2) Set optimization variables and constraints, take the total utility consumption minQ of the whole process as the objective function, build a BP neural network model, and use the actual operation data and simulated operation data to build a data set to train the BP neural network model; The optimization variables include the flow rate F1 of MDI, the flow rate F2 of PPG, the consumption F3 of N2, the initial temperature T1 of the raw material, the temperature T2 of MDI entering RCSTR2, the temperature T3 of PPG entering RCSTR1, the reaction temperature T4 of the mixture entering RCSTR3, the pipeline pressure P of stream 9, and the bottom extraction and feed ratio S of RCSTR3; stream 9 enters RCSTR3; The constraints include the mole fraction C1 of MDI in RCSTR2, the mole fraction C2 of cyclopentane in RCSTR1, the mole fraction C3 of PPG in RCSTR1, the mole fraction C4 of PU in the product of RCSTR3, and the mole fraction C5 of water in the product of RCSTR3; (3) The trained BP neural network model is introduced into the queue competition algorithm LCA to form a BP-LCA hybrid algorithm; the BP-LCA hybrid algorithm is used to optimize the polyurethane high-pressure foaming process parameters.

2. The polyurethane high pressure foaming process parameter optimization method according to claim 1, characterized in that: Step (1) includes: selecting a mixing module Mixer, a heat exchange module Heater, a reaction module RCSTR and a pressure transmitter module Pump in Aspen Plus software to establish a polyurethane high-pressure foaming process model, inputting component data, physical property methods, and initial logistics data, running the polyurethane high-pressure foaming process model, and comparing the output result of the polyurethane high-pressure foaming process model with the actual operation data to determine the error. If the error is within the allowable range, the calibration is completed; otherwise, continue to modify the physical property method and adjust the model data.

3. The polyurethane high pressure foaming process parameter optimization method according to claim 2, characterized in that: The physical property method initially selects POLYSL.

4. The polyurethane high-pressure foaming process parameter optimization method according to claim 1, characterized in that: In step (2), the total utility consumption Q is calculated using the problem table method.

5. The polyurethane high pressure foaming process parameter optimization method according to claim 4, characterized in that: The total utility consumption Q is calculated using the table method, including: In the process flow, streams 1, 2, 3 → stream 4, stream 5 → stream 6, stream 9 → stream 10 are all hot logistics, which require external heat transfer, and all heat consumption is aggregated as the minimum heating utility consumption Q1; a refrigerator is required to quickly reduce the material temperature of stream 11 to room temperature, so stream 11 → stream 12 is cold logistics, and the energy consumed is the minimum cooling utility consumption Q2; Q1 and Q2 data can be retrieved from the software simulation data set; Q = Q1 + Q2.

6. The polyurethane high pressure foaming process parameter optimization method according to claim 1, characterized in that: In step (2), the optimization variable range is:

7. The polyurethane high pressure foaming process parameter optimization method according to claim 6, characterized in that: The specific constraints are: Among them, P is the process equation group described by the BP network; x = [F1, F2, F3, T1, T2, T3, T4, P, S] is the input variable vector of the BP neural network model; y = [C1, C2, C3, C4, C5, Q] is the output variable vector of the BP neural network model.

8. The polyurethane high-pressure foaming process parameter optimization method according to claim 7, characterized in that: The BP neural network model includes an input layer, three hidden layers and an output layer. The number of nodes in the input layer is 9, the number of nodes in the output layer is 6, and the number of nodes in the hidden layer is 8. The transfer function of the hidden layer is logsig, and the transfer function of the output layer is linear purelin.

9. The method for optimizing polyurethane high pressure foaming process parameters according to any one of claims 1 to 8, characterized in that: Step (3) includes: (3-1) Randomly generate m individuals in a given search space to form an initial population, where the individuals are a set of optimization variables; (3-2) Using the BP neural network model to calculate the objective function value of each individual; (3-3) Sort the m individuals in the initial population according to the value of the objective function; (3-4) According to the position of each individual in the queue, the search space of each family is allocated in a certain proportion. The search space of the first family is the smallest, while the search space of the last family is the largest. (3-5) Each individual reproduces asexually within the assigned search range, producing n offspring individuals; (3-6) Calculate the objective function values ​​of n offspring individuals through the BP neural network model, and let these offspring individuals compete for survival with their parent individuals. The best individuals are retained to participate in the next round of competition; (3-7) Determine whether the termination condition is met: the objective function value is regionally stable after several iterations and no longer changes significantly; if the termination condition is met, output the result; otherwise, use the shrinkage factor to shrink the search space and return to step (3-3).

10. A polyurethane high-pressure foaming process parameter optimization system, comprising a memory and a processor, wherein a computer program is stored in the memory, characterized in that: When the computer program is executed by a processor, the steps of the polyurethane high-pressure foaming process parameter optimization method according to any one of claims 1 to 9 are implemented.

Images