A Multi-Objective Optimization Solution Method for Distillation Column Groups Based on Problem Transformation and Improvement Algorithms

By constructing a collaborative optimization model for ethylene distillation column groups and introducing Logistic chaotic mapping and adaptive perturbation strategies, the Evergreen algorithm was improved, solving the problems of high computational cost and difficult decision-making in multi-objective optimization of distillation column groups, and achieving efficient and accurate optimization solutions for column groups.

CN122592893BActive Publication Date: 2026-11-17EAST CHINA UNIV OF SCI & TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611071780.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2026-07-20
Publication Date
2026-11-17
Estimated Expiration
2046-07-20

AI Technical Summary

Technical Problem

Existing multi-objective optimization methods for distillation column groups are computationally expensive, have slow convergence speeds, and are difficult to make decisions. Traditional algorithms suffer from insufficient population diversity, are prone to getting trapped in local optima, and have weak adaptive search capabilities when dealing with complex engineering problems.

Method used

A problem-transformation and algorithm-improvement approach is adopted, which includes constructing a collaborative optimization model for an ethylene distillation column group, initializing the ivy population using Logistic chaotic mapping, and transforming it into a single-objective optimization problem by combining an adaptive penalized boundary crossing method and an adaptive perturbation strategy. The improved Logistic-adaptive ivy algorithm is then used for iterative solution.

Benefits of technology

It enables efficient and low-cost collaborative optimization solutions for complex engineering systems, improves global optimization capabilities and convergence accuracy, simplifies the decision-making process, and reduces decision-making costs.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122592893B_ABST
    Figure CN122592893B_ABST
Patent Text Reader

Abstract

This application discloses a multi-objective optimization solution method for distillation column groups based on problem transformation and improved algorithms, including: constructing a collaborative optimization model for an ethylene distillation column group, dividing the ethylene distillation column group into multiple individual columns, and establishing a single-column data-driven model based on the input-output relationship of each individual column; connecting the data-driven models of each individual column according to the material flow relationship between columns to form an overall optimization model for the column group; determining the decision variables, optimization objectives, and constraints of the column group optimization model; transforming the multi-objective optimization problem into a single-objective optimization problem to obtain a comprehensive fitness function; initializing an ivy population to obtain diverse initial candidate solutions; updating the position and growth rate of the ivy population based on an adaptive perturbation strategy to obtain an improved Logistic-adaptive ivy algorithm; iteratively solving the comprehensive fitness function to output a combination of column group optimization operation parameters that satisfy the constraints.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of industrial process modeling and prediction technology, specifically to a multi-objective optimization solution method for distillation column groups based on problem transformation and improvement algorithms. Background Technology

[0002] Distillation tower groups are key equipment in chemical production, and their efficient and coordinated operation directly affects production costs, product quality, and energy consumption. Taking ethylene distillation tower groups as an example, optimization objectives often include improving product purity and reducing operating costs. These objectives are often in conflict with each other, constituting a complex multi-objective optimization problem.

[0003] Currently, there are two main solution strategies. One is to directly apply multi-objective optimization algorithms, such as NSGA-II and MOPSO, to provide decision-makers with diverse choices by searching for Pareto optimal solutions. However, for large-scale systems like distillation column groups that are highly nonlinear, strongly coupled, and have complex constraints, this approach has significant shortcomings: firstly, it is computationally expensive and has a slow convergence speed; secondly, it still requires manual selection from a large number of solutions, significantly increasing the difficulty of decision-making and the cost of practical application. The second strategy transforms the multi-objective problem into a single-objective problem and then uses a single-objective optimizer to solve it, thus avoiding the aforementioned decision-making dilemma. Common transformation methods include weighted methods, ε-constraint methods, and penalty-based boundary crossing methods. Among these, the penalty-based boundary crossing method, which guides the search through a direction vector, has certain advantages, but its performance heavily depends on the setting of the penalty factor. Traditional fixed parameters are difficult to balance the exploration and development needs of different optimization stages, and it lacks effective integration of decision-maker preferences or prior information about the problem, resulting in room for improvement in search efficiency and solution quality.

[0004] At the optimization algorithm level, metaheuristic algorithms are widely used due to their powerful global search capabilities. While the Ivy League algorithm, as an emerging algorithm, has shown potential in standard tests, it still exhibits shortcomings in handling practical engineering problems such as high-dimensional, multi-peak, and strongly constrained problems like distillation column group optimization. These shortcomings include insufficient population diversity, susceptibility to local optima, and weak adaptive search capabilities, limiting its direct application effectiveness.

[0005] To address the problems of high computational and decision-making costs, insufficient problem transformation guidance, and limited algorithm performance in existing multi-objective optimization methods for distillation column groups, this invention proposes a multi-objective optimization solution method for distillation column groups based on problem transformation and improved algorithms. This achieves efficient and low-cost collaborative optimization solutions for complex engineering systems. Summary of the Invention

[0006] The purpose of this invention is to propose a multi-objective optimization solution method for distillation column groups based on problem transformation and improvement algorithms to solve the problems mentioned in the background art.

[0007] To achieve the aforementioned objectives, the first technical solution adopted by this invention is: a multi-objective optimization solution method for distillation column groups based on problem transformation and improvement algorithms, comprising: S1. Construct a collaborative optimization model for the ethylene distillation tower group, divide the ethylene distillation tower group into multiple individual towers, and establish a single-tower data-driven model based on the input-output relationship of each individual tower. S2 connects the data-driven models of each individual tower based on the material flow relationship between towers to form an overall optimization model of the tower group. S3, determine the decision variables, optimization objectives and constraints of the tower group optimization model, wherein the decision variables include the condensate reflux flow rate and reboiler flow rate of each distillation tower, and the optimization objectives include ethylene product concentration deviation, propylene product concentration deviation and operating cost. S4. The adaptive penalty boundary crossing method based on reference points is used to transform the multi-objective optimization problem into a single-objective optimization problem, and the comprehensive fitness function is obtained. S5. The ivy population is initialized using Logistic chaotic mapping to obtain diverse initial candidate solutions. S6. Based on the adaptive perturbation strategy, the position and growth rate of the ivy population are updated to obtain the improved Logistic-adaptive ivy algorithm. S7. The improved Logistic-adaptive Ivy algorithm is used to iteratively solve the comprehensive fitness function and output the optimal combination of tower group operation parameters that satisfy the constraints.

[0008] Furthermore, the tower group contains 6 distillation towers D1-D6 and 2 reaction towers A1-A2. The data model formula for a single tower with i=1, 2, ... 8 is as follows: ; in, It is the output of the i-th tower at time t, including key performance indicators such as product purity and product temperature. This is the input to the i-th tower, including feed composition, flow rate, and temperature. These are the model's parameters, such as the weights and biases used during model training. This indicates a non-linear relationship between the output and the input; The input-output model formulas for each tower are as follows: ; ; ; ; ; ; ; in, This indicates the j-th output variable of the i-th unit at time t, where t represents the sampling time or the running time; i represents the unit number in the tower group, i=1,…,6 corresponds to distillation towers D1-D6, i=7 corresponds to the first reaction tower A1, and i=8 corresponds to the second reaction tower A2; the superscript j is used to distinguish different output indicators of the same unit. This represents the input vector of the a-th unit; Indicates the parameters of the corresponding single tower model; This represents a data-driven model for distillation columns. This represents the data-driven model of the reaction tower; T represents the vector transpose. The input variable represents the first distillation column D1. Through its data-driven model After calculation, two output variables are obtained. and ; The input variables represent the second distillation column D2. Two output variables were obtained after model calculation. and ; The input variable represents the third distillation column D3. Two output variables were obtained after model calculation. and ; The input variable represents the fourth distillation column D4. Two output variables were obtained after model calculation. and ; The input variable represents the fifth distillation column D5. Two output variables were obtained after model calculation. and ; The input variable represents the 6th distillation column D6. Two output variables were obtained after model calculation. and ; This represents the input variable for the 7th unit, namely the first reaction tower A1. The output variables were obtained after calculation using the reaction tower model. ; This represents the input variable for the 8th unit, namely the second reaction tower A2. The output variables were obtained after calculation using the reaction tower model. .

[0009] Furthermore, selecting the reflux flow rate and reboiler steam flow rate as decision variables, and using the product quality of ethylene and propylene and the operating cost of the tower group as optimization objectives, the three-objective optimization formula is minimized as follows: ; in, , This indicates the condensate reflux flow rate of devices D1 to D6. This indicates the reboiler flow rates of units D1 to D6. and Let these represent the objective functions related to the concentrations of ethylene and propylene products, respectively. The objective function related to operating costs is expressed by the following formulas: ; ; ; in, and These are the market-optimal concentration standards for polymer-grade ethylene and propylene, respectively, both with a value of 99.9%. and These represent the concentrations of ethylene and propylene products obtained from the unit operation, respectively. and The energy consumption cost and maintenance cost generated by the reflux flow rate and reboiler flow rate, respectively, are expressed by the following formulas: ; ; Where n represents the number of distillation units. and These represent the unit energy consumption cost generated by the reflux flow rate and the reboiler flow rate, respectively, and k represents the maintenance factor.

[0010] Furthermore, the formulas for the upper and lower limit constraints are as follows: ; ; in, and These represent the upper and lower limits of the return flow, respectively. and These represent the upper and lower limits of the reboiler flow rate, respectively. The physical quantity rationalization constraint formula constrains the concentrations of ethylene and propylene products during the production process. The formula is as follows: ; ; in Indicates the concentration of ethylene products. Indicates the concentration of propylene product; Secondly, constraints are applied to the intermediate calculation variables, including component concentrations, in the output variables of a single tower, using the following formula: ; in This represents the component concentration variable in the output variable Y of a single tower; For any single tower The material balance formula is: ; in, , , and Representing towers Given the feed flow rate, main product flow rate, reflux flow rate, and impurity flow rate, if the output of tower i is the input of tower j, then the relationship between the two towers is as follows: ; Combining the data model of a single tower, If we represent the product-related variables in the output variables of a single tower, then the formula is: ; The formula for constraining the material flow relationship between tower groups is: ; ; ; ; ; in, This represents the data-driven process model corresponding to the distillation column. This represents the data-driven process model corresponding to the reaction column. Since the input-output relationships differ between different distillation columns or reaction columns, different model parameters are used in the formula. Distinguish between specific units, among which , , , and These represent the data-driven model parameters corresponding to the 2nd, 3rd, 5th, 7th, and 8th units, respectively. This represents the j-th output variable of the i-th unit at time t, where the superscript i indicates the tower or reactor number and the superscript j indicates the output variable sequence number of the unit. This represents the input vector of the data-driven model for the downstream unit. In addition to the upstream output variables explicitly listed in parentheses, it also includes other operational variables, feed variables, or state variables of the downstream unit itself. The superscript T indicates vector transpose, which is used to indicate that the corresponding unit has multiple output variables. This represents the first output variable of the first unit. As one of the main inputs to the 7th unit, the output of the 7th unit was calculated using the reaction tower model. ; This indicates the output of the 7th unit. Furthermore, as one of the inputs to the second distillation column, the output variable is calculated using the second distillation column model. and ; This represents the second output variable of the first unit. As one of the inputs to the third distillation column, the output variable is calculated using the third distillation column model. and ; This represents the first output variable of the third unit. As one of the inputs to the 8th unit, the output of the 8th unit is calculated using the reaction tower model. ; This indicates the output of the 8th unit. Furthermore, as one of the inputs to the fifth distillation column, the output variables are calculated using the fifth distillation column model. and By applying the above constraints, key material transfer relationships in the ethylene and propylene product pathways can be incorporated into the optimization model. This allows individual tower models to no longer be calculated in isolation, but rather to participate in collaborative optimization based on the actual material flow direction of the tower group. For units not explicitly listed, their corresponding single-tower data-driven models, variable boundary constraints, material balance constraints, and other process constraints can all participate in the overall optimization of the tower group.

[0011] Furthermore, for an M-dimensional objective function, minimize the objective function: Define the comprehensive fitness function The formula is: ; ; ; ; in, Let represent the direction vector, whose dimension is consistent with the dimension of the objective function, and set to . This is used to measure the processing of all objectives, and the weighting coefficients can also be set according to the decision-maker's preferences. R represents the reference point, i.e. , Representing the ideal point, that is , Represents the direction vector The projection distance, Represent the objective function point Vertical distance to the reference point Indicates the penalty factor. and These represent the current iteration number and the maximum iteration number, respectively.

[0012] Furthermore, the initialization of the ivy population in step S5 includes: Let the population size be... If the dimension of the decision variables is D and the dimension of the objective function is M, then the i-th ivy individual in the population... The expression is: ; The expression for the entire ivy population is: ; To ensure the diversity of the initial population, the ivy individuals in the population are initialized randomly using the following expression: ; in, Represents a group Dimensionless random numbers, where the random numbers are taken from the uniformly distributed interval [0,1]. and Let represent the lower and upper bounds of the decision space, respectively. The operator represents element-wise multiplication between matrices. This represents the growth rate of the i-th ivy vine; The growth rate of ivy is defined by a differential equation: ; in, Indicates growth rate, Correction factor for deviation from growth; Individuals in a population growth rate The difference equation is: ; Where rand represents a random number between [0,1]. and Let represent the growth rates at discrete times t+1 and t, respectively. Let represent a D-dimensional random vector.

[0013] Furthermore, in step S7, the formula for introducing the adaptive perturbation factor P into the sun-facing growth mode is: ; in, Indicates the updated position of the i-th individual; Indicates the position before the update; This represents the supporting individual adjacent to the i-th individual; This represents the optimal individual within the neighborhood of the i-th individual; This refers to an individual randomly selected from the population; This represents the growth rate of the i-th individual; Represents a D-dimensional random vector; Represents a random number between [0, 1]; This indicates element-wise multiplication; P is the adaptive perturbation factor, calculated using the following formula: ; in, Indicates the current iteration number; Indicates the maximum number of iterations; adaptive perturbation factor. As iterations progress, the value gradually decreases, when When the value is less than 0, it is set to 0, in order to enhance exploration in the early stage and reduce disturbance in the later stage.

[0014] Furthermore, in step S7, the formula for introducing an adaptive perturbation term into the diffusion and evolution model is: ; in, It is the ratio of randomly selected after sorting. High-performing individuals That is more than Individuals who perform poorly.

[0015] Furthermore, the formula for introducing an adaptive perturbation strategy based on the t-distribution for updating the growth rate in step S7 is as follows: ; in, Indicates compliance with degrees of freedom For random numbers distributed by a t-distribution, the probability density function formula for an adaptive t-distribution is: ; in, These are the degrees of freedom, and the formula for degrees of freedom is as follows: ; in, and These are the initial degrees of freedom and the final degrees of freedom, respectively.

[0016] Due to the application of the above technical solution, the present invention has the following advantages compared with the prior art: By introducing a reference point and an adaptive penalty factor, the search direction is effectively guided and the decision-making process is simplified. The LA-IVYA algorithm significantly improves the global optimization capability and convergence accuracy in complex problems through chaotic initialization and adaptive perturbation. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the ethylene distillation tower group in this invention; Figure 2 This is a diagram summarizing the problem transformation and improved algorithm in this invention; Figure 3 This is a flowchart of the multi-objective optimization solution method for distillation column groups based on problem transformation and improvement algorithms of the present invention; Figure 4 This is a schematic diagram of the mapping from a multi-objective space to a single-objective space and the adaptive adjustment of the penalty factor in this invention; Figure 5 This is a schematic diagram illustrating the selection of the best neighboring individuals for population renewal under the sun-facing growth pattern in this invention. Figure 6 This is a schematic diagram illustrating the function of the Logistic chaotic mapping of this invention; Figure 7 This is a schematic diagram of the individual position update process after introducing an adaptive perturbation factor in this invention; Figure 8 This invention relates to the function search space and algorithm convergence curve on some test problems based on CEC2005. Figure 9 This is a box plot of fitness values ​​from 30 independent runs on some test problems of CEC2017 in this invention; Figure 10 This is the convergence curve based on the corresponding test problem of CEC2017 in this invention; Figure 11 This is a graph showing the results of Friedman rank test and Nemenyi test on the entire test problem results based on the CEC2017 benchmark in this invention. Figure 12 This is a particle search history graph of the LA-IVYA algorithm based on some test problems of CEC2017 in this invention; Figure 13 This is a curve showing the change in the proportion of exploration and development of the LA-IVYA algorithm based on some test problems in CEC2017. Figure 14 This is the iterative curve for solving the tower group optimization problem in this invention; Figure 15 This is the convergence curve of a single objective in the tower group optimization problem of this invention. Detailed Implementation

[0018] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0019] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate for the embodiments of this application described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or system that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or systems.

[0020] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0021] Figure 1 This invention illustrates a schematic diagram of the ethylene distillation tower group process. Figure 1The main function of this distillation tower group is to separate cracked gases through distillation to obtain high-purity ethylene and propylene. The cracked gas from the bottom of the demethanizer in the upstream tower group first enters the demethanizer (D1), where methane and a small amount of hydrogen are separated through cryogenic distillation. These gases are discharged from the top of the tower and are typically used as fuel gas or recycled. The bottom product stream, containing C2 and heavier components (ethylene, ethane, propylene, propane, and a small amount of heavier components), enters the low-pressure depropanizer (D3) for further separation. In the low-pressure depropanizer, the C2 components (ethylene, ethane) are mainly separated from the C3 and heavier components (propylene, propane, etc.) through distillation. The top product stream, rich in C2 components, is sent to the C2 hydrogenation reactor (A1) to remove acetylene before entering the ethylene distillation tower (D2); the bottom product stream, rich in C3 components, is sent to the high-pressure depropanizer (D4) for further separation. The top product stream of the high-pressure propane desulfurization tower is pure C3 component. After C3 desulfurization and hydrogenation, it enters the propylene distillation tower (D5) to separate high-purity propylene. The bottom product stream is C4 and heavier components, which are sent to the butane desulfurization tower (D6) for recovery.

[0022] Figure 2 This invention, based on problem transformation and an improved algorithm, is illustrated in the diagram. It mainly comprises two parts: the single-objective transformation of a multi-objective optimization problem and the improvement of the solution algorithm. First, considering the high computational complexity and decision-making costs associated with using multi-objective optimization algorithms to solve complex engineering optimization problems, a method is proposed to transform multi-objective optimization into a single-objective solution. This step proposes an adaptive PBI method based on reference points to map multi-objective optimization to a single-objective space. This method provides adjustable weight parameters, allowing decision-makers to flexibly set them according to their preferences before optimization begins, obtaining a unique optimal solution and avoiding the hassle of choosing from multiple solutions. Second, considering the complexity of the objective function, an improved optimization algorithm, LA-IVYA, is proposed to improve solution quality. To enhance population diversity, a Logistic chaotic mapping is introduced during population initialization. To enhance the algorithm's search capability, an adaptive perturbation factor, an adaptive perturbation term, and an adaptive t-distribution perturbation strategy are introduced into the search process, enabling the algorithm to explore fully in the early stages and converge quickly in the later stages, thus enhancing robustness. Finally, the proposed solution strategy was applied to the optimization model of the tower group, and the results were analyzed.

[0023] Figure 3 The flowchart illustrates the multi-objective optimization solution method for distillation column groups based on problem transformation and improvement algorithms of this invention, including the following steps: S1. Define a multi-objective collaborative optimization model for an ethylene distillation column group. The column group is divided into multiple individual columns, and the relationships between the columns can be described by setting boundary conditions and flow transfer relationships. In this way, the individual columns are effectively combined into a unified column group optimization model. The column group contains 6 distillation columns D1-D6 and 2 reaction columns A1-A2. The data model formula for the individual columns i=1, 2, ... 8 is as follows: (1); in, It is the output of the i-th tower at time t, including key performance indicators such as product purity and product temperature. It is the input of the i-th tower, including feed composition, flow rate, temperature, etc. These are the model's parameters, such as weights and biases during model training. This indicates the non-linear relationship between the output and the input.

[0024] The input-output model formulas for each tower are as follows: ; ; ; (2); ; ; ; S2. Perform multi-objective collaborative optimization on the tower group. Select reflux flow rate and reboiler steam flow rate as decision variables, and use the product quality of ethylene and propylene, as well as the operating cost of the tower group, as optimization objectives. The minimization formula for the three objectives is: (3); in, , This indicates the condensate reflux flow rate of devices D1 to D6. This indicates the reboiler flow rates of units D1 to D6. and Let represent the objective functions related to the concentrations of ethylene and propylene products, respectively. The objective function related to operating costs is expressed by the following formulas: (4); (5); (6); in, and These are the market-optimal concentration standards for polymer-grade ethylene and propylene, respectively, both with a value of 99.9%. and These represent the concentrations of ethylene and propylene products obtained from the operation of the unit. and The energy consumption cost and maintenance cost generated by the reflux flow rate and reboiler flow rate, respectively, are expressed by the following formulas: (7); (8); Where n represents the number of distillation units. and These represent the unit energy consumption cost generated by the reflux flow rate and the reboiler flow rate, respectively, and k represents the maintenance factor.

[0025] S3. Based on the operation of the tower group, the constraints in the optimization process mainly include upper and lower limit constraints, physical quantity rationalization constraints, and material flow relationship constraints between the tower group. The formulas for the upper and lower limit constraints are: (9); (10); in, and These represent the upper and lower limits of the return flow, respectively. and These represent the upper and lower limits of the reboiler flow rate, respectively.

[0026] The physical quantity rationalization constraint formula constrains the concentrations of ethylene and propylene products during the production process. The formula is as follows: (11); (12); in Indicates the concentration of ethylene products. This indicates the concentration of propylene products.

[0027] Secondly, constraints are applied to the intermediate calculation variables, including component concentrations, in the output variables of a single tower. The formula is as follows: (13); in This represents the component concentration variable in the output variable Y of a single tower.

[0028] For any single tower The material balance formula is: (14); in, , , and Representing towers The feed flow rate, main product flow rate, reflux flow rate, and impurity flow rate are given. If the output of tower i is the input of tower j, then the relationship between the two towers is as follows: (15); Combining the data model of a single tower, If we represent the product-related variables in the output variables of a single tower, then the formula is: (16); The formula for constraining the material flow relationship between tower groups is: ; ; (17); ; ; S4. The problem is transformed using the reference point adaptive penalty boundary crossing (PBI) method. A reference point R and an adaptive penalty factor are introduced into the PBI method. Mapping the multi-objective optimization problem to a single-objective space, for an M-dimensional objective function, minimize the objective function: Define the comprehensive fitness function The formula is: (18); (19); (20); (twenty one) ; in, Let represent the direction vector, whose dimension is consistent with the dimension of the objective function, and set to . This is used to measure the processing of all objectives, and the weighting coefficients can also be set according to the decision-maker's preferences. R represents the reference point, i.e. . Representing the ideal point, that is , Represents the direction vector The projection distance, Represent the objective function point Vertical distance to the reference point Indicates the penalty factor. and These represent the current iteration number and the maximum iteration number, respectively.

[0029] S5. Based on the original Ivy League (IVYA) algorithm, this paper analyzes its shortcomings in population diversity and global search capability, and establishes a framework for enhancement from two levels: initialization and search process. The Ivy League population is initialized, with a population size of [missing information]. If the dimension of the decision variables is D and the dimension of the objective function is M, then the i-th ivy individual in the population... The expression is: (twenty two) ; The expression for the entire ivy population is: (twenty three) ; To ensure the diversity of the initial population, the ivy individuals in the population are initialized randomly using the following expression: (twenty four) ; in, Represents a group Dimensional random numbers, which are taken from the uniformly distributed interval [0,1]. and Let represent the lower and upper bounds of the decision space, respectively. The operator represents element-wise multiplication between matrices. This represents the growth rate of the i-th ivy.

[0030] The growth rate of ivy is defined by a differential equation: (25); in, Indicates growth rate, Correction factor for deviation from growth.

[0031] Individuals in a population growth rate The difference equation is: (26); Where rand represents a random number between [0,1]. and Let represent the growth rates at discrete times t+1 and t, respectively. Let represent a D-dimensional random vector.

[0032] Population renewal under a sun-facing growth pattern, in this natural behavior, for individuals It is necessary to find the most important individual that is closest to it. As support. For the sorted ivy population The optimal individual is represented as: Then the best neighboring individual is defined as: (27); Then members pass The update formula for climbing towards the light source is: ; Where It is the current iteration number. Indicates the updated individual. express The absolute value, The symbol represents a division operation between corresponding elements of a matrix.

[0033] Updates under diffusion and evolutionary growth patterns, during the global search phase, for individuals The growth direction is updated based on the orientation of its nearest optimal neighbor, enabling the individual to find a better growth direction in complex environments. When an individual finds an important neighbor... Next, we need to move into the local search phase, that is, the individual search phase. It will try to follow the best individual in the entire population. The process is updated to further search for better solutions. In the It-th iteration, the individual... The update formula for this stage is: (30); Updated individuals growth rate The calculation formula is: (31); Survivor selection occurs because ivy individuals at different stages of growth within the population will be in different growth patterns. Therefore, individuals in the population will be updated according to different formulas. Here, a single parameter is used. To distinguish different growth patterns, The calculation formula is: (32); Let individuals and the best individual The fitness values ​​are respectively and Then the formula for the rule of individual renewal in the population is: ; After one iteration, individuals need to be selected to proceed to the next iteration. During this iteration, the population is updated to form a new population. First, the new population is mixed with the original population to form... The formula is: (34); Secondly, the mixed population is sorted from best to worst fitness values ​​to form the sorted mixed population. The formula is: (35); Finally, from Select before Individual ivy plants constitute the final population. Then proceed to the next iteration, with the formula as follows: (36); S6. Introducing Logistic Chaotic Mapping to Enhance Population Initialization. Logistic chaotic mapping is introduced into IVYA's population initialization to enhance randomness and improve population diversity. The expression for Logistic chaotic mapping is: (37); Where n is the discrete time step. and This represents the values ​​of the current iteration and the next iteration. This represents the system control parameters.

[0034] The population initialization formula is: (38); (39); in, Represents the i-th ivy individual. Represents a set of D-dimensional random numbers. The operator represents element-wise multiplication between matrices. and Let be the lower and upper bounds of the decision space, respectively, and let n be the discrete time step. Indicates system control parameters, It is a chaotic sequence generated by the Logistic chaotic mapping.

[0035] S7. Design and integrate multiple adaptive perturbation strategies. Improve the IVYA algorithm's search process in three aspects: introduce an adaptive perturbation factor P in the sun-facing growth mode; introduce an adaptive perturbation term in the diffusion and evolution modes; and introduce an adaptive perturbation strategy based on the t-distribution for the growth rate update. The formula for introducing the adaptive perturbation factor P in the sun-facing growth mode is: (40); in, Individuals randomly selected from the population. It is the best individual in the neighborhood, and P is the adaptive perturbation factor, calculated by the following formula: (41); in, Indicates the current iteration number. This indicates the maximum number of iterations.

[0036] The formula for introducing an adaptive perturbation term in the diffusion and evolution model is as follows: (42); in, It is the ratio of randomly selected after sorting. High-performing individuals That is more than Individuals who perform poorly.

[0037] The formula for introducing an adaptive perturbation strategy based on the t-distribution for the growth rate update is as follows: (43); in, Indicates compliance with degrees of freedom For random numbers distributed by a t-distribution, the probability density function formula for an adaptive t-distribution is: (44); in, These are the degrees of freedom, and the formula for degrees of freedom is as follows: (45); in, and These are the initial degrees of freedom and the final degrees of freedom, respectively.

[0038] The performance of the algorithm was evaluated through two comparative experiments.

[0039] a. Comparative analysis of LA-IVYA and IVYA.

[0040] IVYA underwent comparative experiments on the CEC2005 benchmark problem and demonstrated good performance. To verify the performance of the improved LA-IVYA, the CEC2005 benchmark problem was also used to conduct comparative experiments between IVYA and LA-IVYA. In the experiments, the decision variable dimension was set to 30, the population size was set to 30, the maximum number of iterations was set to 1000, and 30 independent runs were performed for each test problem. Table 1 shows the performance index results of LA-IVYA and IVYA, including the mean, standard deviation, and Wilcoxon test results of the fitness values ​​obtained from 30 independent runs. From the results in the table, LA-IVYA achieved smaller fitness values ​​on 14 complex test problems, including F5, F6, F8, and F11, which are closer to the optimal value than IVYA. Moreover, on these problems, the standard deviation of LA-IVYA is significantly smaller than that of IVYA, demonstrating higher stability. For test problems such as F1-F4 and F9-F10, the two algorithms achieve the same optimization results. This is because the test problems are relatively simple, and both algorithms can find the optimal solution well.

[0041] Table 1. Algorithm performance metrics results on the CEC2005 test problem.

[0042] Figure 8 The search space and convergence curves of some test problems are shown. To compare the convergence of the algorithms, eight test problems with the same and different optimization results for both algorithms were selected. As can be seen from the figures, LA-IVYA typically converges to a smaller value for both simple and complex test problems. Even when both algorithms converge to the same value, LA-IVYA achieves convergence with fewer iterations. Therefore, comprehensive analysis shows that LA-IVYA exhibits higher convergence accuracy and speed, better optimal solution quality, and more stable optimization performance compared to IVYA.

[0043] b. Performance comparison analysis of LA-IVYA and the new algorithm in 2024.

[0044] To further evaluate the performance of LA-IVYA, the improved algorithm was compared with six optimization algorithms proposed in 2024: Goose optimization algorithm (GO), Horned lizard optimization algorithm (HLOA), Hippopotamus optimization algorithm (HO), Love evolution algorithm (LEA), Newton-Raphson-based optimizer (NRBO), and Parrot optimizer (PO). IVYA was also included in the comparison. The CEC2017 benchmark problems were used, which included 29 test problems. In this comparison experiment, the population size was set to 30, the decision variable dimension was 30, and the maximum number of iterations was set to 1000. For each test problem, each algorithm underwent 30 independent experiments.

[0045] Table 2 shows the mean, standard deviation, and Wilcocon test results of the fitness values ​​of the eight algorithms across 29 test problems, denoted by "A", "S", and "W", respectively. As can be seen from the table, LA-IVYA achieved the lowest mean across 21 test problems, with slightly worse mean values ​​on F17 and F23. However, it achieved the lowest standard deviation on seven test problems, including F1, F12, and F13. Overall, LA-IVYA performed best among the eight algorithms.

[0046] Table 2. Algorithm performance metrics results on the CEC2017 test problem.

[0047]

[0048] To more intuitively compare the results of the algorithms, Figure 9 The image shows box plots of fitness values ​​from 30 independent runs on some test problems. Figure 10The convergence curves for the corresponding test problems are shown in the box plot. As can be seen from the box plot, LA-IVYA achieves a lower median fitness value on most problems compared to other algorithms. The convergence curves show that LA-IVYA exhibits the fastest convergence speed on most test problems, ultimately converging to the lowest fitness value. Particularly in complex test problems such as F5, F9, F25, and F26, LA-IVYA shows a significant downward trend within the first 200 iterations, far faster than IVYA and other algorithms.

[0049] To compare whether there are significant differences in the performance of these 8 algorithms on this benchmark problem, Friedman rank test and Nemenyi test were performed on the results of the entire test problem. The smaller the average rank of the calculated algorithms, the better the performance of the algorithm. Figure 11 The results are obtained through verification and calculation. (a) shows the results of the Friedman rank test, where LA-IVYA obtained the lowest average rank, ranking first in algorithm performance. (b) shows the results of the Nemenyi test. If the horizontal lines corresponding to two algorithms do not overlap, it indicates that the two algorithms have significantly different performance; otherwise, there is no significant difference. In the figure, LA-IVYA is significantly different from the other 6 algorithms except IVYA, and has only a very small, hard-to-detect overlap with IVYA. Therefore, it can be concluded that LA-IVYA is significantly better than the other algorithms.

[0050] In order to study the search process of LA-IVYA, Figure 12 and Figure 13 The graph records the particle search history and the change curves of the exploration-exploitation ratio during the optimization process of the LA-IVYA algorithm on some test problems. The black pentagrams in the particle search history distribution graph represent the global optimum. The search history shows that a few particles are distributed in sparse regions far from the optimum, while most particles are concentrated around it. This indicates that the particles can fully explore the entire space in the early stages of the search, and can quickly converge to the optimum region in the later stages. The change curves of the exploration-exploitation ratio show that LA-IVYA can dynamically adjust the proportion of exploration and exploration on different test problems, achieving a reasonable search strategy. In the early stages of the search, the exploration ratio is high, ensuring the global exploration capability of the search space and avoiding the algorithm getting trapped in local optima. In the middle stages of the search, the exploration ratio gradually decreases, while the exploration ratio increases, meaning that the algorithm begins to focus on high-quality solution regions for refined searching, improving convergence accuracy. In the later stages of the search, the exploration ratio approaches 100%, indicating that the algorithm enters the convergence phase, performing local optimization to ensure the stability of the optimum solution. This adaptive exploration-exploitation switching mechanism allows LA-IVYA to balance global search and local optimization.

[0051] A series of experimental results on the CEC2017 test problem demonstrate the powerful performance of LA-IVYA, mainly due to two improved mechanisms: Logistic chaotic mapping and adaptive perturbation strategy. LA-IVYA achieves a more reasonable balance between global search and local exploitation through Logistic chaotic mapping and adaptive perturbation, ensuring global search capabilities in the early stages while providing a more refined solution search mechanism in the later optimization process.

[0052] c. Analysis of the application of experimental solution strategies in tower group optimization.

[0053] The proposed solution strategy was applied to solve a tower group optimization model. This model contains three minimization objective functions: ethylene concentration difference, propylene concentration difference, and operating cost. First, the RAPBI method was used to transform the three objective functions into a single objective function. During the calculation, the reference point was selected as the relatively optimal point from the approximate power factor (PF) obtained from EMOBKA. The ideal point is composed of the optimal values ​​of a single objective function; that is, the optimal values ​​for ethylene and propylene concentration differences are 0, and the optimal value for operating cost is calculated by taking the lower bounds of all decision variables. For the weighting factor, 1, 10, 20, and 50 were tried, with 1 being the weighting factor that yielded the best convergence. Furthermore, in this solution, the three objectives were assumed to be of equal importance; therefore, the direction vector was set to... Subsequently, LA-IVYA was used to solve the problem, with a population size of 200 and a maximum number of iterations of 1000.

[0054] Figure 14 The iterative curves for solving the tower group optimization problem show that all eight algorithms converged rapidly within 200 iterations, while LA-IVYA converged to the lowest fitness value, verifying the high performance of LA-IVYA in engineering optimization problems. Figure 15The figures show the convergence curves of the three normalized objective function values. Firstly, the overall trend shows that as the concentration differences of ethylene and propylene decrease, the operating cost gradually increases, reflecting the characteristics of the distillation process—that is, increased product purity leads to increased energy consumption. This verifies the feasibility and rationality of the proposed solution strategy. Secondly, the three objective function values ​​converge to 0.6342 mol%, 7.7000 × 10⁻⁷ mol%, and 1869.0009 t / h, respectively. In contrast, the approximate optimal solution obtained using a multi-objective optimization algorithm corresponds to values ​​of 0.624283 mol%, 8.7463 × 10⁻⁴ mol%, and 1869.8100 t / h. The solution obtained in this study is closer to the ideal point and has higher quality. Finally, the solution obtained using this strategy is unique, eliminating the need to select from numerous solutions and reducing decision-making costs, thus demonstrating the efficiency of the strategy.

[0055] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for multi-objective optimization of distillation column networks based on problem transformation and improved algorithm, characterized in that, include: S1. Construct a collaborative optimization model for the ethylene distillation tower group, divide the ethylene distillation tower group into multiple individual towers, and establish a single-tower data-driven model based on the input-output relationship of each individual tower. S2, connect the data-driven models of each individual tower according to the material flow relationship between towers to form an overall optimization model of the tower group; S3, determine the decision variables, optimization objectives and constraints of the tower group optimization model, wherein the decision variables include the condensate reflux flow rate and reboiler flow rate of each distillation tower, and the optimization objectives include ethylene product concentration deviation, propylene product concentration deviation and operating cost. S4. The adaptive penalty boundary crossing method based on reference points is used to transform the multi-objective optimization problem into a single-objective optimization problem, and the comprehensive fitness function is obtained. S5. The ivy population is initialized using Logistic chaotic mapping to obtain diverse initial candidate solutions. S6. Based on the adaptive perturbation strategy, the position and growth rate of the ivy population are updated to obtain the improved Logistic-adaptive ivy algorithm. S7. The improved Logistic-adaptive Evergreen algorithm is used to iteratively solve the comprehensive fitness function and output the optimal combination of tower group operation parameters that meet the constraints. Step S5, which initializes the ivy population, includes: Let the population size be... If the dimension of the decision variables is D and the dimension of the objective function is M, then the i-th ivy individual in the population... The expression is: ; The entire ivy population The expression is: ; To ensure the diversity of the initial population, the ivy individuals in the population are initialized randomly using the following expression: ; in, Represents a group Dimensionless random numbers, where the random numbers are taken from the uniformly distributed interval [0,1]. and Let represent the lower and upper bounds of the decision space, respectively. The operator represents element-wise multiplication between matrices. This represents the growth rate of the i-th ivy vine; The growth rate of ivy is defined by a differential equation: ; in, Indicates growth rate, Correction factor for deviation from growth; Individuals in a population growth rate The difference equation is: ; Where rand represents a random number between [0,1]. and Let represent the growth rates at discrete times t+1 and t, respectively. Represents a D-dimensional random vector; In step S7, the formula for introducing the adaptive perturbation factor P into the sun-facing growth mode is: ; in, Indicates the updated position of the i-th individual; Indicates the position before the update; This represents the supporting individual adjacent to the i-th individual; This represents the optimal individual within the neighborhood of the i-th individual; This refers to an individual randomly selected from the population; This represents the growth rate of the i-th individual; Represents a D-dimensional random vector; Represents a random number between [0, 1]; This indicates element-wise multiplication; P is the adaptive perturbation factor, calculated using the following formula: ; in, Indicates the current iteration number; Indicates the maximum number of iterations; adaptive perturbation factor. As iterations progress, the value gradually decreases, when When the value is less than 0, it is set to 0, in order to enhance exploration in the early stage and reduce disturbance in the later stage.

2. The multi-objective optimization solution method for distillation column groups based on problem transformation and improvement algorithms as described in claim 1, characterized in that, The tower group contains 6 distillation towers D1-D6 and 2 reaction towers A1-A2. The data model formula for a single tower with i=1, 2, ... 8 is as follows: ; in, It represents the output of the i-th tower at time t, including key performance indicators such as product purity and product temperature. This is the input to the i-th tower, including feed composition, flow rate, and temperature. These are the model's parameters, including the weights and biases used during model training. This indicates a non-linear relationship between the output and the input. The input-output model formulas for each tower are as follows: ; ; ; ; ; ; ; in, This indicates the j-th output variable of the i-th unit at time t, where t represents the sampling time or the running time; i represents the unit number in the tower group, i=1,…,6 corresponds to distillation towers D1-D6, i=7 corresponds to the first reaction tower A1, and i=8 corresponds to the second reaction tower A2; the superscript j is used to distinguish different output indicators of the same unit. This represents the input vector of the a-th unit; Indicates the parameters of the corresponding single tower model; This represents a data-driven model for distillation columns. This represents the data-driven model of the reaction tower; T represents the vector transpose. The input variable represents the first distillation column D1. Data-driven model After calculation, two output variables are obtained. and ; The input variables represent the second distillation column D2. Two output variables were obtained after model calculation. and ; The input variable represents the third distillation column D3. Two output variables were obtained after model calculation. and ; The input variable represents the fourth distillation column D4. Two output variables were obtained after model calculation. and ; The input variable represents the fifth distillation column D5. Two output variables were obtained after model calculation. and ; The input variable represents the 6th distillation column D6. Two output variables were obtained after model calculation. and ; This represents the input variable for the 7th unit, namely the first reaction tower A1. The output variable was obtained after calculation using the reaction tower model. ; This represents the input variable for the 8th unit, namely the second reaction tower A2. The output variable was obtained after calculation using the reaction tower model. .

3. The multi-objective optimization solution method for distillation column groups based on problem transformation and improvement algorithms as described in claim 1, characterized in that, Selecting reflux flow rate and reboiler steam flow rate as decision variables, and using the product quality of ethylene and propylene and the operating cost of the tower group as optimization objectives, the three-objective optimization formula is minimized as follows: ; in, , This indicates the condensate reflux flow rate of devices D1 to D6. This indicates the reboiler flow rates of units D1 to D6. and Let these represent the objective functions related to the concentrations of ethylene and propylene products, respectively. The objective function related to operating costs is expressed by the following formulas: ; ; ; in, and These are the market-optimal concentration standards for polymer-grade ethylene and propylene, respectively. and These represent the concentrations of ethylene and propylene products obtained from the unit operation, respectively. and The energy consumption cost and maintenance cost generated by the reflux flow rate and reboiler flow rate, respectively, are expressed by the following formulas: ; ; Where n represents the number of distillation units. and These represent the unit energy consumption cost generated by the reflux flow rate and the reboiler flow rate, respectively, and k represents the maintenance factor.

4. The multi-objective optimization solution method for distillation column groups based on problem transformation and improvement algorithms as described in claim 3, characterized in that, The formulas for upper and lower limit constraints are: ; ; in, and These represent the upper and lower limits of the return flow, respectively. and These represent the upper and lower limits of the reboiler flow rate, respectively. The physical quantity rationalization constraint formula constrains the concentrations of ethylene and propylene products during the production process. The formula is as follows: ; ; in Indicates the concentration of ethylene products. Indicates the concentration of propylene product; Secondly, constraints are applied to the intermediate calculation variables, including component concentrations, in the output variables of a single tower, using the following formula: ; in This represents the component concentration variable in the output variable Y of a single tower; For any single tower The material balance formula is: ; in, , , and Representing towers Given the feed flow rate, main product flow rate, reflux flow rate, and impurity flow rate, if the output of tower i is the input of tower j, then the relationship between the two towers is as follows: ; Combining the data model of a single tower, If we represent the product-related variables in the output variables of a single tower, then the formula is: ; The formula for constraining the material flow relationship between tower groups is: ; ; ; ; ; in, This represents the data-driven process model corresponding to the distillation column. This represents the data-driven process model corresponding to the reaction tower. , , , and These represent the data-driven model parameters corresponding to the 2nd, 3rd, 5th, 7th, and 8th units, respectively. This represents the j-th output variable of the i-th unit at time t, where the superscript i indicates the tower or reactor number and the superscript j indicates the output variable sequence number of the unit. This represents the input vector of the data-driven model for the downstream unit; the superscript T indicates the vector transpose, used to indicate that the corresponding unit has multiple output variables; This represents the first output variable of the first unit. As one of the main inputs to the 7th unit, the output of the 7th unit was calculated using the reaction tower model. ; This indicates the output of the 7th unit. Furthermore, as one of the inputs to the second distillation column, the output variable is calculated using the second distillation column model. and ; This represents the second output variable of the first unit. As one of the inputs to the third distillation column, the output variable is calculated using the third distillation column model. and ; This represents the first output variable of the third unit. As one of the inputs to the 8th unit, the output of the 8th unit is calculated using the reaction tower model. ; This indicates the output of the 8th unit. Furthermore, as one of the inputs to the fifth distillation column, the output variables are calculated using the fifth distillation column model. and .

5. The multi-objective optimization solution method for distillation column groups based on problem transformation and improvement algorithms as described in claim 4, characterized in that, For an M-dimensional objective function, minimize the objective function: Define the comprehensive fitness function The formula is: ; ; ; ; in, Let represent the direction vector, with its dimension consistent with the dimension of the objective function, set to . Used to measure the processing of all objectives, R represents the reference point, i.e. , Representing the ideal point, that is , Represents the direction vector The projection distance, Represent the objective function point Vertical distance to the reference point Indicates the penalty factor. and These represent the current iteration number and the maximum iteration number, respectively.

6. The multi-objective optimization solution method for distillation column groups based on problem transformation and improvement algorithms as described in claim 1, characterized in that, In step S7, the formula for introducing an adaptive perturbation term into the diffusion and evolution model is as follows: ; in, It is the ratio of randomly selected after sorting. High-performing individuals That is more than Individuals who perform poorly.

7. The multi-objective optimization solution method for distillation column groups based on problem transformation and improvement algorithms as described in claim 1, characterized in that, The formula for introducing an adaptive perturbation strategy based on the t-distribution in step S7 for updating the growth rate is as follows: ; in, Indicates compliance with degrees of freedom For random numbers distributed by a t-distribution, the probability density function formula for an adaptive t-distribution is: ; in, These are the degrees of freedom, and the formula for degrees of freedom is as follows: ; in, and These are the initial degrees of freedom and the final degrees of freedom, respectively.