Multi-agent manufacturing process optimization method based on multi-objective particle swarm optimization algorithm
By constructing a multi-agent manufacturing process optimization model and using a multi-objective particle swarm optimization algorithm to optimize process industry production, the problem of resource sharing conflicts was solved, and low-cost, high-efficiency, and high-quality production was achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- QILU UNIVERSITY OF TECHNOLOGY (SHANDONG ACADEMY OF SCIENCES)
- Filing Date
- 2018-12-29
- Publication Date
- 2026-06-23
Smart Images
Figure CN109739087B_ABST
Abstract
Description
Technical Field
[0001] This disclosure relates to the field of process control technology, and in particular to a multi-agent manufacturing process optimization method based on a multi-objective particle swarm optimization algorithm. Background Technology
[0002] Process industries mainly include chemical, cement, and paper industries, which occupy an important position in my country's industrial production. The manufacturing process in process industries is continuous and cannot be interrupted, and the sequence of the manufacturing process is fixed. Its management characteristics are to ensure continuous material supply and to ensure that each production link must operate normally during the working period.
[0003] Because different products or multiple batches of the same product share time and resources during production, conflicts can easily arise. To resolve these issues, it is essential to rationally allocate various resources during the production process to ensure uninterrupted production flow. Therefore, system control and manufacturing process optimization in process industries are more complex, increasing the difficulty of controlling factors affecting production costs and product yield, such as raw materials and temperature. Simultaneously, it is crucial to meet various manufacturing objectives, such as ensuring product quality while minimizing operation time to achieve the lowest possible production cost.
[0004] To address these problems, heuristic multi-objective optimization techniques have seen significant development. Extensive scientific research has demonstrated that these techniques are more practical and effective than traditional methods. Commonly used techniques include: ACO (Ant Colony Optimization), NSGA (Genetic Algorithm), PAES (Pareto Archive Evolutionary Strategy), and PSO (Particle Swarm Optimization). Among these, PSO is an evolutionary technique that simulates social behavior and is based on swarm intelligence. It possesses its own unique search mechanism, is easy to implement, and has strong convergence capabilities, thus finding significant application in the manufacturing sector.
[0005] Based on the above analysis, the main technical problems to be solved in this application are how to use multiple agents to model the process industry and how to use multi-objective particle swarm optimization algorithm to optimize the model. Summary of the Invention
[0006] To address the shortcomings of existing technologies, embodiments of this disclosure provide a multi-agent manufacturing process optimization method based on the multi-objective particle swarm optimization algorithm. This method improves upon the PSO algorithm, enabling it to directly apply to the optimization of multiple objectives.
[0007] To achieve the above objectives, this application adopts the following technical solution:
[0008] Multi-agent manufacturing process optimization methods based on multi-objective particle swarm optimization include:
[0009] For the manufacturing process in the process industry, a manufacturing process optimization model based on multi-agent is constructed, which includes a two-layer structure. The upper layer consists of a central control agent and a pool agent used to store algorithm and data information. The lower layer consists of raw material agent, equipment agent, management agent and waste agent. Each intelligent agent will interact with other agents based on its own communication module.
[0010] By treating the Agent as a particle in the multi-objective particle swarm optimization algorithm, the Agent population is endowed with the ability to evolve. A corresponding data model is established for the manufacturing process optimization model based on multi-agents, and the multi-objective particle swarm optimization algorithm is used to solve it to obtain an effective solution set.
[0011] As a further technical solution of this application, the master control agent is responsible for coordinating the resources of the entire manufacturing process. It is located at the upper level of the manufacturing model and records the dynamics of the lower layer agents. It adjusts the manufacturing process in a timely manner based on the information obtained. When performing optimization analysis on multiple different production schemes, it retrieves the corresponding algorithm from the pool agent and then provides production guidance based on the optimization results of the algorithm.
[0012] The raw material agent stores the characteristics and various proportions of different materials required for the production of different products;
[0013] Equipment Agent refers to the actual machinery and equipment used in production. It records the machine's temperature, pressure, kiln speed, and airflow to ensure the utilization rate of raw materials and machinery during the production process.
[0014] The management agent is used to collect information from the interactions between the equipment agent and the raw material agent, and then feed it back to the upper-level control agent.
[0015] Waste Agent is responsible for reusing leftover materials from production, reducing environmental pollution.
[0016] As a further technical solution of this application, a mathematical model is established for one stage of cement manufacturing, namely the "two grinding and one firing" proportioning and mixing stage:
[0017] The specific functions are as follows:
[0018]
[0019]
[0020] min H(x)=(f1(x),f2(x)) T (3)
[0021] The purpose of formula (1) is to minimize the error h1(x) as much as possible, thereby improving the performance of cement. M1 represents the amount of additives added, and N k The amount of additives required to meet the standard content, n k (x) The amount of additives required to achieve the minimum range of the desired cement performance indicators, including the content of all minerals and the current degree of hardening and durability of the cement, are obtained by testing instruments during the specific manufacturing process.
[0022] Formula (2) represents the total cost required. The minimum cost h2(x) needs to be obtained through optimization, where M2 is the required amount of additives (g). t p represents the amount of additive added in the t-th step. t The unit price for each additive;
[0023] Formula (3) integrates the two objectives above.
[0024] As a further technical solution of this application, a multi-objective particle swarm optimization algorithm is used for solving the problem. First, initialization is performed by setting the size, initial position, and initial velocity of the particle swarm. The objective function of each particle is calculated, the current individual extreme value of each particle is found, the current global optimal solution of the entire particle swarm is found, the velocity and position of each particle are updated, and it is determined whether the termination condition has been met. If so, the optimal solution is output; otherwise, the objective function of each particle is recalculated until the termination condition is met.
[0025] As a further technical solution of this application, the specific process of obtaining an effective solution set by using a multi-objective particle swarm optimization algorithm is as follows:
[0026] Step 1: Set the population size Z of particles, and randomly determine the positions x of particles within the population. i and velocity v i And the number of iterations t, and set the termination condition of the algorithm;
[0027] Step 2: Add the non-dominated solutions to the external archive set, and use formulas (1) and (2) to calculate the fitness value of each particle in the population.
[0028] Step 3: If the number of iterations is less than the given number of iterations t, repeat steps 4-7; otherwise, the algorithm ends.
[0029] Step 4: Select the global optimum based on the fitness values of the particles in the archive. Update the position of each particle according to formula (3), and at the same time calculate the fitness value of the particles again.
[0030] Step 5: Repeat the process of Step 2, and remove the solutions that are relatively less suitable according to the dominance relationship, and then form the external file for the next iteration;
[0031] Step 6: If the historical best extreme value of the currently selected particle is not as good as the current position, then use the latest particle as the best extreme value; otherwise, keep the current state unchanged.
[0032] Step 7: If the position of a random particle in the population is better than the previously selected optimal value for the entire population, then proceed as in Step 6, replacing the previously selected optimal position with this position.
[0033] Step 8: After the iteration is completed, the particles in the external archive set are the effective solution set of the entire algorithm.
[0034] As a further technical solution of this application, step 4, the update formula for particle velocity can be divided into three parts:
[0035] The first is the "inertial component," which represents a degree of maintenance of one's own state of motion, described by the formula: w*v i Where w represents the weight, v i The velocity of the current particle;
[0036] The second part, the "individual perception part," originates from one's own individual experience and can be understood as the distance between the particle's current position and its best possible position. The formula is described as: c1r1(h i (t)-x i (t));
[0037] Thirdly, the "global understanding" section, derived from the experience of other excellent particles in the swarm, can be understood as the distance between particle i's current position and the best position in the swarm, described by the formula: c²v²(k i (t)-x i (t));
[0038] The particle velocity is updated using the sum of the three descriptive formulas above;
[0039] x i (t+1)=v i (t+1)+x i (t) (4)
[0040] Where w is the inertia weight, which adopts a dynamic weight setting rule, with the maximum weight set to 0.9 and the minimum weight set to 0.4. During the experiment, a value between 0.4 and 0.9 will be dynamically and randomly selected for calculation. c1c2 is a non-negative constant, which is also given to 2 during the experiment. r1r2 is a random number between 0 and 1.
[0041] The embodiments of this disclosure also disclose a multi-agent manufacturing process optimization system based on a multi-objective particle swarm optimization algorithm, including:
[0042] The manufacturing process optimization model building unit based on multi-agent is designed for the manufacturing process of the process industry. It constructs a manufacturing process optimization model based on multi-agent, which includes a two-layer structure. The upper layer consists of a central control agent and a pool agent used to store algorithm and data information. The lower layer consists of raw material agent, equipment agent, management agent and waste agent. Each intelligent agent will interact with other agents based on its own communication module.
[0043] The multi-objective particle swarm optimization (MPS) solution unit treats the Agent as a particle in the multi-objective PMS algorithm, endowing the Agent population with the ability to evolve. It establishes a corresponding data model for the multi-agent-based manufacturing process optimization model, and uses the multi-objective PMS algorithm to solve it, obtaining an effective solution set.
[0044] As a further technical solution of this application, the multi-objective particle swarm optimization (MPS) algorithm solving unit uses the MPS algorithm for solving the problem. First, initialization is performed by setting the size, initial position, and initial velocity of the particle swarm. The objective function of each particle is calculated, the current individual extreme value of each particle is found, the current global optimal solution of the entire particle swarm is found, the velocity and position of each particle are updated, and it is determined whether the termination condition has been met. If so, the optimal solution is output; otherwise, the objective function of each particle is recalculated until the termination condition is met.
[0045] The embodiments of this disclosure also disclose a computer-readable storage medium storing a plurality of instructions adapted for loading and execution by a processor of a terminal device of the multi-agent manufacturing process optimization method based on a multi-objective particle swarm optimization algorithm.
[0046] The embodiments of this disclosure also disclose a terminal device, including a processor and a computer-readable storage medium, wherein the processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, which are adapted to be loaded and executed by the processor to optimize the multi-agent manufacturing process based on the multi-objective particle swarm algorithm.
[0047] Compared with the prior art, the beneficial effects of this disclosure are:
[0048] The multi-objective particle swarm optimization algorithm involved in the embodiments of this disclosure is an improvement on the PSO algorithm, which directly applies to the optimization of multiple objectives. The overall cost after optimization is lower than the actual cost. Therefore, the optimized non-dominated solution set can be used as a reference in actual production to improve the low consumption, high output, and high quality of process operation manufacturing and reduce the required cost. Attached Figure Description
[0049] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments of this application and their descriptions are used to explain this application and do not constitute an undue limitation of this application.
[0050] Figure 1 A schematic diagram of the manufacturing process optimization model structure of a multi-agent system constructed for one or more embodiments of this disclosure;
[0051] Figure 2 This is a schematic diagram of the solution process of one or more embodiments of the present disclosure based on the multi-agent particle swarm optimization algorithm;
[0052] Figure 3 This diagram illustrates the optimization and verification of particles in the MOPSO algorithm using one or more embodiments of this disclosure.
[0053] Figure 4 Pareto solution set curves for the optimization results of one or more embodiments of this disclosure;
[0054] Figure 5 This diagram illustrates a comparison between the actual cost and the optimized cost of one or more embodiments of this disclosure. Detailed Implementation
[0055] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used in this application have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.
[0056] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.
[0057] Terminology Explanation Section:
[0058] Particle Swarm Optimization (PSO): PSO is a population-based stochastic optimization technique proposed by Eberhart and Kennedy in 1995. It mimics the swarming behavior of insects, herds of mammals, flocks of birds, and schools of fish, which cooperate to find food. Each member of the group continuously changes its search pattern by learning from its own experience and that of other members.
[0059] Multi-Agent Systems (MAS): A multi-agent system (MAS) is a network structure composed of several loosely coupled agents. These entities are logically or physically loosely coupled, but their behavior is autonomous. They work together to control complex tasks and solve multi-objective problems through information exchange and negotiation.
[0060] Process industries, also known as industrial manufacturing, refer to production processes that involve physical and chemical changes. Each stage is continuous and uninterrupted. Examples include the chemical, oil refining, paper, and cement industries.
[0061] In a typical embodiment of this application, a multi-agent manufacturing process optimization method based on multi-objective particle swarm optimization algorithm is provided, including: constructing a multi-agent manufacturing process optimization model for the manufacturing process of the process industry, including a two-layer structure, wherein the upper layer structure is a general control agent and a pool agent used to store algorithm and data information, and the lower layer structure is a raw material agent, an equipment agent, a management agent and a waste agent, and each intelligent agent will interact with other agents based on its own communication module;
[0062] By treating the Agent as a particle in the multi-objective particle swarm optimization algorithm, the Agent population is endowed with the ability to evolve. A corresponding data model is established for the manufacturing process optimization model based on multi-agents, and the multi-objective particle swarm optimization algorithm is used to solve it to obtain an effective solution set.
[0063] In specific implementation examples, such as Figure 1 As shown, the multi-agent system model introduces the concept of a pool to store the required optimization algorithms and process information during production. Each lower-level agent and the central control agent obtains algorithms and corresponding information from the pool agents during production and data optimization. After resource utilization is complete, the information is returned to the pool for use by other agents.
[0064] Depend on Figure 1As can be seen, the multi-agent manufacturing process optimization model mainly consists of two layers: the upper layer is the central control agent and the pool agent, which stores algorithms and data information. The lower layer consists of raw material agents, equipment agents, management agents, and waste agents. Each intelligent agent interacts with other agents based on its own communication module. The central control agent is responsible for coordinating resources throughout the manufacturing process. It resides at the upper layer of the manufacturing model and records the dynamics of the lower-layer agents, adjusting the manufacturing process promptly based on the information received. When optimizing various production schemes, it retrieves the corresponding algorithms from the pool agent and provides production guidance based on the optimization results. The raw material agent stores the characteristics and various proportions of different materials required for producing different products. The equipment agent represents the actual machinery in production, recording machine temperature, pressure, kiln speed, and airflow to ensure the utilization rate of raw materials and machinery during production. The management agent collects information from the interactions between the equipment agent and the raw material agent and feeds it back to the central control agent at the upper layer. Waste Agent is responsible for reusing leftover materials from production, reducing environmental pollution.
[0065] like Figure 2 As shown, in the multi-agent particle swarm optimization (PSO) algorithm, an Agent is an abstract entity capable of perceiving the external environment and communicating with other Agents to collaboratively solve problems. It also possesses intelligent thinking and behavior. A multi-Agent system is a loosely coupled cooperative network composed of several Agents. The PSO algorithm, on the other hand, is an abstraction and simulation of the bird flock's foraging process, a stochastic search iterative algorithm. Essentially, both are loosely coupled "swarm intelligence" structures. Therefore, in this application, Agents are treated as particles in a multi-objective PSO algorithm, endowing them with the ability to evolve as a population. This is used to address multi-objective problems involved in the manufacturing process.
[0066] To verify the effectiveness of the algorithm, the Binh and Korn functions were selected:
[0067]
[0068] During the calculation, the number of iterations was 5000, the number of particles was 30, the maximum number of runs was 1000, and the acceleration factors c1 = c2 = 2. To better demonstrate the convergence effect, the result was calculated using log2, and then the result was displayed graphically. The results are as follows: Figure 3 As shown.
[0069] from Figure 3As can be seen, using Agent as the particle in the MOPSO algorithm for optimization yields better results than the usual MOPSO.
[0070] This example will use the cement production process as an example to elaborate on the technical solution of the invention and provide a general analysis and description of its manufacturing process. Those familiar with cement production processes will mention "two grindings and one firing," namely, raw meal preparation (first grinding), clinker calcination (first firing), and cement grinding (second grinding). In a cement plant, cement production mainly involves the following stages: raw meal preparation, raw meal grinding, clinker firing, cement grinding, storage, and shipping. Cement grinding is the final and most energy-intensive step in cement manufacturing. Before this step, clinker is mixed with gypsum and additives to ensure that the minerals in the clinker and the finished product meet the required performance indicators. This experiment, based on a multi-agent technology model of the manufacturing process, will establish and optimize a mathematical model for one stage of cement manufacturing—the aforementioned mixing stage.
[0071] The specific functions are as follows:
[0072]
[0073]
[0074] min H(x)=(f1(x),f2(x)) T (3)
[0075] The purpose of formula (1) is to minimize the error h1(x) as much as possible, thereby improving the performance of cement. M1 represents the amount of additives added, N k The amount of additives required to meet the standard content, n k (x) The amount of additives required to achieve the minimum range of desired cement performance indicators. The content of all minerals, as well as the current hardening degree and durability of the cement, are obtained from testing instruments used during the specific manufacturing process.
[0076] Formula (2) represents the total cost required. The minimum cost h2(x) needs to be obtained through optimization. M2 represents the required amount of additives, and g... t p represents the amount of additive added in the t-th step. t This represents the unit price of each additive.
[0077] Formula (3) integrates the two objectives above.
[0078] Specific implementation process based on instances
[0079] Task 1: Set the population size Z of particles, and randomly select the positions x of particles in the population. i and velocity v i The algorithm also includes the iteration count t and the termination condition.
[0080] Task 2: Add non-dominated solutions to the external archive set, and use formulas (1) and (2) to calculate the fitness value of each particle in the population. That is, calculate the fitness value of each particle according to the set objective function and archive it.
[0081] Task 3: If the number of iterations is less than the given number of iterations t, repeat steps 4-7; otherwise, the algorithm ends.
[0082] Task 4: Select the individual extreme value based on the fitness values of the particles saved in Task 2. and global extrema Update the position of each particle according to formula (4), and at the same time calculate the fitness value of the particles again. That is, after updating the position and velocity of the particles, calculate based on the updated particles at this time.
[0083] x i (t+1)=v i (t+1)+x i (t) (4)
[0084] Meanwhile, the formula for updating particle velocity can be divided into three parts:
[0085] The first is the "inertial component," which represents a degree of maintenance of one's own state of motion, described by the formula: w*v i Where w represents the weight, v i The velocity of the current particle;
[0086] The second part, the "individual perception part," originates from one's own individual experience and can be understood as the distance between the particle's current position and its best possible position. The formula is described as: c1r1(h i (t)-x i (t));
[0087] Thirdly, the "global understanding" section, derived from the experience of other excellent particles in the swarm, can be understood as the distance between particle i's current position and the best position in the swarm, described by the formula: c²v²(k i (t)-x i (t)).
[0088] The velocity of the particle is updated using the sum of the three descriptive formulas above.
[0089] Where w is the inertia weight, and in this experiment we adopted a dynamic weight setting rule, specifying a maximum weight of 0.9 and a minimum weight of 0.4. During the experiment, a value between 0.4 and 0.9 will be dynamically and randomly selected for calculation. c1c2 is a non-negative constant, and we also gave it a value of 2 during the experiment. r1r2 is a random number between 0 and 1.
[0090] Task 5: Repeat the process of Task 2, and remove the solutions that are relatively less suitable based on the dominance relationship, and then form the external archive for the next iteration.
[0091] Task 6: If the historical best extreme value of the currently selected particle is not as good as the current position, then use the latest particle as the best extreme value; otherwise, keep the current state unchanged.
[0092] Task 7: If the position of a random particle in the population is better than the previously selected best position for the entire population, then proceed in the same manner as in Task 6, replacing the previously selected best position with this new position.
[0093] Task 8: After the iteration is completed, the particles in the external archive set are the effective solution set of the entire algorithm.
[0094] After establishing the objective function to be optimized, the MOPSO algorithm was used to optimize it. In the simulation experiment, we used the production data of a specific type (A1) cement from a cement manufacturing plant to verify the results.
[0095] The required contents of various minerals and oxides needed in the manufacturing process of A1 grade cement are shown in Table 1:
[0096] Table 1: Content and range of minerals and oxides in cement
[0097]
[0098]
[0099] Table 2 shows the unit price of various additives:
[0100] Table 2: Unit Price of Various Additives
[0101] Mineral types Unit price (RMB / ton) Early strength agent 25000 antifreeze 5000 quick-drying agent 1800 Retarder (Accelerator) 33150
[0102] When optimizing the data, we chose 200 iterations, and both the population size and the size of the external archive set were 100. The experimental results are as follows. Figure 4 As shown, this is the Pareto front obtained from the selected portion of the data:
[0103] from Figure 4 The results show that when the comprehensive error decreased from 0.42 to 0.03, the manufacturing cost increased from 11,300 yuan to 42,000 yuan, indicating a clear inverse relationship between minimizing mineral content error and production cost. In actual manufacturing, we often focus more on the cement's compliance with standards and whether it can fully meet requirements for strength, compressive and flexural strength, hydration and setting properties during large-scale projects. Therefore, in actual production, we generally choose the extreme solution among the non-inferior solutions, i.e., the minimum comprehensive error, for cement production.
[0104] superior Figure 5 The paper presents 50 sets of manufacturing costs optimized by the MOPSO algorithm and the costs consumed in actual cement production. It can be clearly concluded that the overall optimized cost is lower than the actual cost. Therefore, the optimized non-dominated solution set can be used as a reference in actual production to improve the low consumption, high output, and high quality of cement manufacturing and reduce the required costs.
[0105] The model constructed using the manufacturing process of a specific type of cement in this application, along with the two objective functions of minimizing the comprehensive error and the minimum required processing cost of various mineral contents in cement based on the MOPSO algorithm, verify that applying the multi-objective particle swarm optimization algorithm to the process industry is feasible and effective.
[0106] This invention takes cement manufacturing as an example. It addresses the problems of complex and difficult-to-control manufacturing processes, which lead to serious resource waste and increased costs. It utilizes a multi-objective particle swarm optimization algorithm to optimize the manufacturing process of the process industry, thereby finding a better set of control factors for the cement production process, achieving the goal of rational use of resources and reducing production costs.
[0107] Specifically, this study aims to optimize production in the process industry by applying the MOPSO algorithm to the manufacturing process optimization. Based on the analysis of the manufacturing process in the process industry, and taking a specific manufacturing process of a process industry enterprise as an example, a mathematical model is established with the optimization objectives of minimizing total processing cost and overall error. The specific implementation process of the algorithm is also presented. Experimental simulations were conducted using existing control system models and the MOPSO algorithm. The results show that the MOPSO algorithm is feasible for optimizing the manufacturing process.
[0108] The embodiments of this disclosure also disclose a multi-agent manufacturing process optimization system based on a multi-objective particle swarm optimization algorithm, including:
[0109] The manufacturing process optimization model building unit based on multi-agent is designed for the manufacturing process of the process industry. It constructs a manufacturing process optimization model based on multi-agent, which includes a two-layer structure. The upper layer consists of a central control agent and a pool agent used to store algorithm and data information. The lower layer consists of raw material agent, equipment agent, management agent and waste agent. Each intelligent agent will interact with other agents based on its own communication module.
[0110] The multi-objective particle swarm optimization (MPS) solution unit treats the Agent as a particle in the multi-objective PMS algorithm, endowing the Agent population with the ability to evolve. It establishes a corresponding data model for the multi-agent-based manufacturing process optimization model, and uses the multi-objective PMS algorithm to solve it, obtaining an effective solution set.
[0111] In the multi-objective particle swarm optimization (MPS) algorithm solution unit, the multi-objective particle swarm optimization algorithm is used for solution. First, initialization is performed by setting the size, initial position, and initial velocity of the particle swarm. The objective function of each particle is calculated, the current individual extreme value of each particle is found, the current global optimal solution of the entire particle swarm is found, the velocity and position of each particle are updated, and it is determined whether the termination condition has been met. If so, the optimal solution is output; otherwise, the objective function of each particle is recalculated until the termination condition is met.
[0112] The embodiments of this disclosure also disclose a computer-readable storage medium storing a plurality of instructions adapted for loading and execution by a processor of a terminal device of the multi-agent manufacturing process optimization method based on a multi-objective particle swarm optimization algorithm.
[0113] The embodiments of this disclosure also disclose a terminal device, including a processor and a computer-readable storage medium, wherein the processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, which are adapted to be loaded and executed by the processor to optimize the multi-agent manufacturing process based on the multi-objective particle swarm algorithm.
[0114] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.
Claims
1. A multi-agent manufacturing process optimization method based on multi-objective particle swarm optimization algorithm, characterized by: include: For the manufacturing process in the process industry, a manufacturing process optimization model based on multi-agent is constructed, which includes a two-layer structure. The upper layer consists of a central control agent and a pool agent used to store algorithm and data information. The lower layer consists of raw material agent, equipment agent, management agent and waste agent. Each intelligent agent will interact with other agents based on its own communication module. By treating Agent as a particle in the multi-objective particle swarm optimization algorithm, the Agent population is given the ability to evolve. A corresponding data model is established for the manufacturing process optimization model based on multi-agents, and the multi-objective particle swarm optimization algorithm is used to solve the problem and obtain an effective solution set. Among them, a mathematical model was established for one stage of cement manufacturing, namely the "two grinding and one firing" mixing stage: The specific functions are as follows: (1) （2） （3） The purpose of formula (1) is to minimize the error h1(x) as much as possible, thereby improving the performance of cement. M1 represents the amount of additives added, and N k The amount of additives required to meet the standard content, n k (x) The amount of additives required to achieve the minimum range of the desired cement performance indicators, wherein the content of all minerals and the current hardening degree and durability of the cement are obtained by the testing instruments in the specific manufacturing process. Formula (2) represents the total cost required. The minimum cost h2(x) needs to be obtained through optimization. M2 represents the required amount of additives, and g... t p represents the amount of additive added in the t-th step. t The unit price for each additive; Formula (3) integrates the two objectives above; The specific process of obtaining an effective solution set using the multi-objective particle swarm optimization algorithm is as follows: Task 1: Set the population size Z of particles, and randomly select the positions x of particles in the population. i and velocity v i And the number of iterations t, and set the termination condition of the algorithm; Task 2: Add non-dominated solutions to the external archive set, and use formulas (1) and (2) to calculate the fitness value of each particle in the population. That is, calculate the fitness value of each particle according to the set objective function and archive it. Task 3: If the number of iterations is less than the given number of iterations t, repeat steps 4-7; otherwise, the algorithm ends. Task 4: Select the individual extreme value based on the fitness values of the particles saved in Task 2. and global extrema The position of each particle is updated according to formula (4), and the fitness value of the particles is calculated again. That is, after updating the position and velocity of the particles, the particle is calculated based on the updated particle. (4)； Meanwhile, the formula for updating particle velocity can be divided into three parts: The first is the "inertial component," which represents the maintenance of one's own state of motion. The formula is described as: w v i Where w represents the weight, v i The velocity of the current particle; The second part, "individual cognition," originates from one's own individual experience and can be understood as the distance between the particle's current position and its best possible position. The formula is described as follows: ; The third part, "Global Understanding," draws on the experience of other excellent particles in the swarm. It can be understood as the distance between particle i's current position and the best position in the swarm, and is described by the formula: ; The particle velocity is updated using the sum of the three descriptive formulas above; Where w is the inertia weight. In this experiment, we adopted a dynamic weight setting rule, specifying that the maximum weight is 0.9 and the minimum weight is 0.
4. During the experiment, a value between 0.4 and 0.9 will be dynamically and randomly selected for calculation. c1c2 is a non-negative constant, which we also set to 2 during the experiment. r1r2 is a random number between 0 and 1. Task 5: Repeat the process of Task 2, and remove the solutions that are relatively less suitable based on the dominance relationship, and then form the external file for the next iteration; Task 6: If the historical best extreme value of the currently selected particle is not as good as the current position, then use the latest particle as the best extreme value; otherwise, keep the current state unchanged. Task 7: If the position of a random particle in the population is better than the previously selected best position for the entire population, then proceed in the same manner as in Task 6, replacing the previously selected best position with this new position. ; Task 8: After the iteration is completed, the particles in the external archive set are the effective solution set of the entire algorithm.
2. The multi-agent manufacturing process optimization method based on multi-objective particle swarm optimization algorithm as described in claim 1, characterized in that, The master control agent is responsible for coordinating resources throughout the entire manufacturing process. It is located at the upper level of the manufacturing model and records the dynamics of the lower-level agents. It adjusts the manufacturing process in a timely manner based on the information it receives. When performing optimization analysis on various production plans, it retrieves the corresponding algorithm from the pool agents and then provides production guidance based on the optimization results of the algorithm. The raw material agent stores the characteristics and various proportions of different materials required for the production of different products; Equipment Agent refers to the actual machinery and equipment used in production. It records the machine's temperature, pressure, kiln speed, and airflow to ensure the utilization rate of raw materials and machinery during the production process. The management agent is used to collect information from the interactions between the equipment agent and the raw material agent, and then feed it back to the upper-level control agent. Waste Agent is responsible for reusing leftover materials from production, reducing environmental pollution.
3. The multi-agent manufacturing process optimization method based on multi-objective particle swarm optimization algorithm as described in claim 1, characterized in that, The multi-objective particle swarm optimization algorithm is used to solve the problem. First, initialization is performed by setting the size, initial position, and initial velocity of the particle swarm. The objective function of each particle is calculated, the current individual extreme value of each particle is found, the current global optimal solution of the entire particle swarm is found, the velocity and position of each particle are updated, and it is determined whether the termination condition has been met. If so, the optimal solution is output; otherwise, the objective function of each particle is recalculated until the termination condition is met.
4. A multi-agent manufacturing process optimization system based on multi-objective particle swarm optimization algorithm, characterized in that: include: The manufacturing process optimization model building unit based on multi-agent is designed for the manufacturing process of the process industry. It constructs a manufacturing process optimization model based on multi-agent, which includes a two-layer structure. The upper layer consists of a central control agent and a pool agent used to store algorithm and data information. The lower layer consists of raw material agent, equipment agent, management agent and waste agent. Each intelligent agent will interact with other agents based on its own communication module. The multi-objective particle swarm optimization algorithm solution unit treats the Agent as a particle in the multi-objective particle swarm optimization algorithm, endows the Agent with the ability of population evolution, establishes a corresponding data model for the manufacturing process optimization model based on multi-agents, and uses the multi-objective particle swarm optimization algorithm to solve it and obtain an effective solution set. Among them, a mathematical model was established for one stage of cement manufacturing, namely the "two grinding and one firing" mixing stage: The specific functions are as follows: (1) （2） （3） The purpose of formula (1) is to minimize the error h1(x) as much as possible, thereby improving the performance of cement. M1 represents the amount of additives added, and N k The amount of additives required to meet the standard content, n k (x) The amount of additives required to achieve the minimum range of the desired cement performance indicators, including the content of all minerals and the current degree of hardening and durability of the cement, are obtained by testing instruments during the specific manufacturing process. Formula (2) represents the total cost required. The minimum cost h2(x) needs to be obtained through optimization. M2 represents the required amount of additives, and g... t p represents the amount of additive added in the t-th step. t The unit price for each additive; Formula (3) integrates the two objectives above; The specific process of obtaining an effective solution set using the multi-objective particle swarm optimization algorithm is as follows: Task 1: Set the population size Z of particles, and randomly select the positions x of particles in the population. i and velocity v i And the number of iterations t, and set the termination condition of the algorithm; Task 2: Add non-dominated solutions to the external archive set, and use formulas (1) and (2) to calculate the fitness value of each particle in the population. That is, calculate the fitness value of each particle according to the set objective function and archive it. Task 3: If the number of iterations is less than the given number of iterations t, repeat steps 4-7; otherwise, the algorithm ends. Task 4: Select the individual extreme value based on the fitness values of the particles saved in Task 2. and global extrema The position of each particle is updated according to formula (4), and the fitness value of the particles is calculated again. That is, after updating the position and velocity of the particles, the particle is calculated based on the updated particle. (4)； Meanwhile, the formula for updating particle velocity can be divided into three parts: The first is the "inertial component," which represents the maintenance of one's own state of motion. The formula is described as: w v i Where w represents the weight, v i The velocity of the current particle; The second part, "individual cognition," originates from one's own individual experience and can be understood as the distance between the particle's current position and its best possible position; the formula is described as follows: ; The third part, "Global Understanding," draws on the experience of other excellent particles in the swarm. It can be understood as the distance between particle i's current position and the best position in the swarm, and is described by the formula: ; The particle velocity is updated using the sum of the three descriptive formulas above; Where w is the inertia weight. In this experiment, we adopted a dynamic weight setting rule, specifying that the maximum weight is 0.9 and the minimum weight is 0.
4. During the experiment, a value between 0.4 and 0.9 will be dynamically and randomly selected for calculation. c1c2 is a non-negative constant, which we also set to 2 during the experiment. r1r2 is a random number between 0 and 1. Task 5: Repeat the process of Task 2, and remove the solutions that are relatively less suitable based on the dominance relationship, and then form the external file for the next iteration; Task 6: If the historical best extreme value of the currently selected particle is not as good as the current position, then use the latest particle as the best extreme value; otherwise, keep the current state unchanged. Task 7: If the position of a random particle in the population is better than the previously selected best position for the entire population, then proceed in the same manner as in Task 6, replacing the previously selected best position with this new position. ; Task 8: After the iteration is completed, the particles in the external archive set are the effective solution set of the entire algorithm.
5. The multi-agent manufacturing process optimization system based on multi-objective particle swarm optimization algorithm as described in claim 4, characterized in that, In the multi-objective particle swarm optimization (MPS) algorithm solution unit, the multi-objective particle swarm optimization algorithm is used for solution. First, initialization is performed by setting the size, initial position, and initial velocity of the particle swarm. The objective function of each particle is calculated, the current individual extreme value of each particle is found, the current global optimal solution of the entire particle swarm is found, the velocity and position of each particle are updated, and it is determined whether the termination condition has been met. If so, the optimal solution is output; otherwise, the objective function of each particle is recalculated until the termination condition is met.
6. A computer-readable storage medium storing a plurality of instructions adapted for loading and execution by a processor of a terminal device of the multi-agent manufacturing process optimization method based on a multi-objective particle swarm optimization algorithm as described in any one of claims 1-3.
7. A terminal device comprising a processor and a computer-readable storage medium, the processor being configured to implement various instructions; the computer-readable storage medium being configured to store multiple instructions, the instructions being adapted to be loaded by the processor and executed by the processor to optimize the multi-agent manufacturing process based on the multi-objective particle swarm algorithm as described in any one of claims 1-3.