Production line autonomous optimization design method and system based on large model and intelligent algorithm

Abstract

The invention discloses a production line autonomous optimization design method and system based on a large model and an intelligent algorithm, and relates to the technical field of production line optimization design. Understanding and extracting input information through a large model, and analyzing production line optimization problems including production line balance optimization, production line layout optimization, resource configuration optimization and the like; judging whether corresponding models and algorithms are established in the module library or not, if yes, calling corresponding mathematical models, optimization algorithms and simulation environments for each optimization problem in a function calling mode, and performing iterative optimization; if not, understanding, learning and simulating a mathematical model, an optimization algorithm and a simulation environment on the basis of a large model, and performing iterative optimization; and carrying out production simulation verification, performance evaluation and man-machine interaction adjustment to obtain a final production line optimization design scheme. According to the method, autonomous optimization design of the production line under the driving of process data is realized, modeling and programming are reduced, the design difficulty is reduced, the design efficiency and accuracy are improved, and the planning and design period is shortened.

Description

Technical Field

[0001] This invention relates to the field of production line optimization design technology, specifically to a production line autonomous optimization design method and system based on large models and intelligent algorithms, which is particularly suitable for rapid production line planning and design scenarios with a large number of production lines, rapid product updates and iterations, and complex process and resource constraints. Background Technology

[0002] In a market demanding small-batch, customized production, characterized by diverse product types, resource constraints, complex manufacturing processes, and rapid product iteration, production lines require quick optimization. Currently, production line planning and design are largely done manually using software or algorithms, involving the design, optimization, and verification of plans—a process that is extremely time-consuming. Mainstream production line design software, such as Plant Simulation, Flexsim, Delmia, and Factory Simulation, all require extensive modeling, programming, debugging, optimization, and verification work, making modeling and optimization challenging. In scientific research, current research on production line optimization problems, such as production line balancing, layout, and resource allocation, mostly addresses specific aspects, lacking comprehensive solutions for overall production line optimization design and failing to effectively guide the implementation of production line planning schemes. Furthermore, current optimization methods require manually building mathematical models, writing optimization algorithms, and establishing simulation environments for different production line optimization problems, resulting in low efficiency and high difficulty.

[0003] With the development of generative artificial intelligence technology, the understanding, modeling, and programming capabilities of large models can be leveraged to automatically solve production line optimization problems. Through learning from and understanding similar optimization problems, large models can automatically construct mathematical models, optimization algorithms, and simulation environments, thereby enabling autonomous optimization design of production lines, improving design efficiency, and shortening production line design cycles. Summary of the Invention

[0004] To address the aforementioned technical problems, the present invention aims to provide a method and system for autonomous optimization design of production lines based on large-scale models and intelligent algorithms. This system constructs a module library for production line optimization, addressing optimization issues such as production line balancing, layout, and resource allocation. The library contains mathematical models and optimization algorithms for various sub-optimization problems. During production line optimization design, if the optimization problem is already established in the module library, the corresponding model and algorithm are directly called to solve it. If no such model exists, the large-scale model learns from similar optimization problems in the module library, then imitates and generates the model and algorithm for the problem to be optimized. This achieves autonomous optimization design of the production line, improving the efficiency and intelligence of production line optimization design.

[0005] A production line autonomous optimization design method based on large models and intelligent algorithms is used for automatic planning and design of production lines driven by process information, reducing manual modeling, programming, and iterative optimization, and realizing autonomous optimization design of production lines. The method includes the following steps:

[0006] Step S1: Based on the large model, understand the input information of the production line optimization problem, extract process information, modeling information, etc., analyze whether it is a new production line design or an adjustment to the existing production line, and analyze what optimizations need to be made, including production line balancing optimization, production line layout optimization, production resource allocation optimization, etc.

[0007] Step S2: Based on the large model for the input problem to be optimized, determine whether the corresponding mathematical model and optimization algorithm have been established in the module library. If yes, proceed to step S3; otherwise, proceed to step S4.

[0008] Step S3: For each optimization problem, the pre-established mathematical model, optimization algorithm, and production line simulation evaluation environment are called from the module library through function calling to perform iterative optimization and obtain the production line planning and design scheme.

[0009] Step S4: For each optimization problem, construct the corresponding mathematical model, optimization algorithm, and production line simulation evaluation environment based on the large model, and perform iterative optimization to obtain the production line planning and design scheme; for similar problems, the large model can first understand and learn based on the models and algorithms already built in the module library, and then imitate and generate them to improve the accuracy of the solution.

[0010] Step S5: For the optimized production line scheme, construct a production simulation environment consistent with the actual production line, perform production constraint checks, verify the feasibility of the production line design scheme, and evaluate the production line performance, such as equipment utilization and capacity; for problems such as process flow errors, production logic errors, process constraint violations, and program errors, return the runtime error information to the corresponding large model above for modification;

[0011] Step S6: Perform human-computer interaction adjustments. By changing the optimization target or weight, supplementing process and resource constraints, etc., change the obtained production line optimization design scheme. Depending on the optimization method, return to step S3 or step S4, and continue to adjust and optimize the production line design scheme through function calls or large model solving until satisfied.

[0012] Step S7: Obtain the optimal production line design scheme, including the optimal process allocation, workstation layout, production resource allocation, etc., and present and export the result chart of the production line planning and design scheme; add the optimization problem description, mathematical model, optimization algorithm, target value calculation program, and production simulation environment generated by the large model to the corresponding sub-problem in the module library, so as to provide model and algorithm support for solving similar optimization problems in the future.

[0013] Specifically, step S1 includes the following steps:

[0014] Step S1.1: Receive multimodal input information, which may include optimization problem descriptions, process data tables, process flow diagrams, CAD layouts, etc. It may also prompt the user to input or upload process tables using a unified structure, or provide the folder address where the process data is located; depending on the data requirements of the optimization problem, there may be multiple rounds of information input;

[0015] Step S1.2: Extract input information features and perform structured unification. Specifically, multimodal data such as images and CAD files are identified, and information related to the problem is extracted. The input information is divided into the following categories: feature description of the optimization problem, optimization objective, constraints, and process data input.

[0016] Step S1.3: Analyze what type of optimization problem the input problem belongs to, whether it is a new production line or an adjustment design of an existing production line (production line evolution optimization design), and what optimizations are needed;

[0017] The main categories and functional descriptions of production line optimization issues are as follows:

[0018] Production line / assembly line balancing optimization problem: The input information includes product process time, task / process sequence constraints, resource constraints, etc. The task / process is assigned to appropriate workstations. The optimization objectives are to maximize workstation load balance, achieve a high balance rate, minimize the number of workstations, and minimize equipment investment costs. The output is the optimal allocation scheme of processes on workstations.

[0019] Production line / workshop layout optimization problem: The input information includes workshop area, workstation area, logistics transfer relationship and frequency between workstations, and workstation layout constraints. The optimization objective is to minimize the total material flow in the workshop, and the output is the layout location information of workstations within the production line / workshop.

[0020] Production resource allocation optimization: The inputs are the task sequence and processing time of the workstation, production line layout, product process flow, etc. The optimization objectives are to maximize production capacity and minimize resource input costs. The outputs are the quantity of production resources in the workshop, such as the number of AGVs, the number of workers, the number of parallel devices, the capacity of the temporary storage area, etc.

[0021] If it is a newly established production line, the system will prompt you to enter the design scheme, process information, and constraints of the existing production line, as well as the process information and resource constraints of the new product. The newly established production line has the following new optimization objectives, such as minimizing equipment relocation costs, new equipment investment costs, and new personnel configuration costs.

[0022] Determine the type of optimization problem based on the problem description;

[0023] Step S1.4: Based on the complete data requirements of the optimization problem, remind the user to input the missing process and constraint information. The user can be reminded to upload the process data in the corresponding format in the form of a structured table to prevent the omission of key information and facilitate data extraction.

[0024] The large model used in all steps of this invention is described below:

[0025] The large models used include information interaction models, method selection models, mathematical modeling models, algorithm programming models, target value solving models, and production simulation verification models, etc.

[0026] These large models can be fine-tuned versions of existing large models. Based on established module libraries (problem description and classification library, mathematical model library, optimization algorithm library, target value calculation library, and production simulation environment library), supervised learning and reinforcement learning methods are used to enable the large models with corresponding functions to generate classification of optimization problems, mathematical models, optimization algorithms, target value calculation code, and simulation environment code, and to provide feedback learning to improve the accuracy of the generated models and code.

[0027] Depending on the general large model capabilities, the size of the problem to be optimized, the difficulty of solving the problem, the accuracy requirements, and the experimental conditions, these large models can also be untuned large models, such as large models deployed locally (e.g., DeepSeek, ChatGPT, etc.) or large models that are called from the network server via API (e.g., DeepSeek, ChatGPT, etc.).

[0028] These large models (mathematical modeling models, programming models, etc.) can be different large models, or the same large model can solve different tasks after passing through different prompts;

[0029] When solving these large models, we first find descriptions of similar optimization problems and corresponding mathematical models, optimization algorithms, production simulation environments, etc. in the module library to understand the modeling and programming ideas. Then, based on the characteristics of the problem to be solved, we perform imitation generation, modeling and programming to improve the accuracy of the generated content of the large model.

[0030] Specifically, step S2 includes the following:

[0031] The completed module library includes a problem description and classification library, a mathematical model library, an optimization algorithm library, an objective value calculation library, and a production simulation environment library.

[0032] The target value calculation library constructs a simplified production simulation environment based on process flow and resource constraints to calculate target values.

[0033] The production simulation environment library contains production and logistics components and production simulation engines required for production simulation verification. Components include equipment, AGVs, workers, products, conveyor lines, etc. Using discrete event simulation methods, based on process flow and production line design schemes, a production simulation model consistent with actual production is constructed.

[0034] Based on the detailed optimization problem and its characteristics, determine whether the input optimization problem has already been modeled in the module library. The specific characteristics of the problem are as follows:

[0035] Production line balancing optimization problems can be categorized as follows: single-product production line balancing optimization, multi-product production line balancing optimization; single-sided production line / assembly line optimization, double-sided production line / assembly line optimization; balancing optimization considering shared resources, balancing optimization not considering shared resources, etc.

[0036] Production line layout optimization issues include the following categories: single-row layout, multi-row layout, layout with special position constraints (such as a certain workstation being fixed in a certain position), layout without special position constraints, single-floor layout, multi-floor layout, etc.

[0037] The optimization of production resource allocation includes the following categories: transmission line flow transformation, AGV delivery type, requiring workers, not requiring workers, with temporary storage area, without temporary storage area, etc.

[0038] For each type of optimization problem, such as production line balancing, production line layout, and resource allocation optimization, they can be divided into single-objective and multi-objective optimization based on the optimization objective.

[0039] Specifically, step S4 involves modeling the optimization problem, programming the algorithm, calculating the objective value, and iteratively optimizing it, including the following steps:

[0040] Step S4.1: Establish a mixed integer linear programming (MILP) model / mathematical model for the optimization problem. Find similar optimization problems in the module library, refer to the descriptions and mathematical models of similar problems, understand the modeling ideas and the meaning of the model, and then, based on the description of the current optimization problem, imitate and construct a mathematical model, including the optimization objective and constraints.

[0041] For multiple optimization objectives, a negative sign can be used to unify the optimization direction;

[0042] Step S4.2: Based on the size of the optimization problem, let the large model determine whether to use a solver (such as Gurobi, CPLEX, etc.) or an intelligent optimization algorithm to solve it;

[0043] For small-scale optimization problems, a solver is used to solve them, and solver execution code is generated. First, similar optimization problems are found in the module library, the corresponding problem descriptions are examined, the mathematical formulas and solver code are understood, and imitation learning is performed. Then, based on the description and mathematical model of this optimization problem, the corresponding solver execution code is generated by imitation.

[0044] For large-scale optimization problems, intelligent optimization algorithms are used to solve them, generating corresponding intelligent algorithm code. For different optimization problems, the selection priority of optimization algorithms is set based on experience and test results. For each type of optimization algorithm, modular programming is performed in the algorithm library. For different functional functions, such as encoding and local search, the algorithm flow is summarized and modularized in a structured way to facilitate understanding, programming, and calling. This also makes it easier for the large programming model to understand and imitate programming for new problems. When programming the large programming model, the corresponding algorithm for similar problems is first found in the algorithm library, understood and learned, and then imitated to generate the code, thereby improving the quality of program writing.

[0045] Step S4.3: Construct a program to calculate the target value, and calculate the corresponding target value. The specific method is as follows:

[0046] To optimize the allocation of production resources, a lightweight production simulation model is constructed based on process flow, production line balancing scheme, workshop layout scheme, etc. using discrete event simulation method. Multi-factor experimental method is used to calculate information such as workshop capacity and equipment utilization rate under different production resource configurations, so as to determine the optimal production resource configuration.

[0047] During the calculation process, if an error occurs at any step, the error message is returned to the corresponding large model, such as the large model being modeled or the large model being programmed, for modification.

[0048] Step S4.4: Based on the algorithm flow, problem size, and changes in the target value, determine the number of iterations, determine the next-generation selection method, perform iterative optimization, and obtain the production line optimization design scheme;

[0049] Step S5 verifies the optimized solution through production simulation and evaluates its production performance, including the following steps:

[0050] Step S5.1: Summarize the production line optimization design schemes, such as one or all of the schemes in task allocation, workstation layout, and production resource allocation, and use the discrete event simulation method to establish a production simulation model consistent with the actual production line.

[0051] Step S5.2: Based on the input process information, such as process flow and workstation processing time, perform production simulation;

[0052] Step S5.3: If there is an error, analyze the error information and provide feedback to the corresponding large model for model or code modification. For example, provide feedback to the large modeling model if the constraints of process A and process B do not match.

[0053] Step S5.4: Run the process to obtain the production line's production performance, including capacity, equipment utilization rate, worker utilization rate, bottleneck processes, and blockage status;

[0054] Step S6 involves adjusting the optimization scheme through human-computer interaction, as follows:

[0055] Users can manually adjust production line design schemes, supplement process and constraint information; for example, restrict a certain process to have explosion-proof requirements, requiring the setting of an explosion-proof room, and specifying the direction of explosion venting; for example, specifying the location of workshop entrances and exits, and the direction of product flow; for example, adding or removing targets, changing the weight of target values, etc.

[0056] Step S7 presents the final production line design scheme and exports the results, as follows:

[0057] The system presents the adjusted production line design and production performance, obtaining the optimal production line balance, layout, and resource allocation schemes. Specifically, it obtains the allocation of processes in workstations, the tasks handled by each workstation, and the total time required; the location of each workstation within the workshop; the number of AGVs, workers, and machines configured in the production line, the capacity of the temporary storage area, etc.; and exports the result charts of the production line design.

[0058] The module library includes the problem description, mathematical model, optimization algorithm, objective value solution code, and production simulation environment for solving this new optimization problem using a large model, providing model and algorithm support for subsequent solutions.

[0059] A production line autonomous optimization design system based on a large model includes:

[0060] The front-end interaction module receives user input and presents the production line optimization design results, enabling interaction with the user.

[0061] The module library includes a problem description and classification library, a mathematical model library, an optimization algorithm library, a target value calculation library, and a production simulation environment library, which stores mathematical models, optimization algorithms, and production simulation code for related production line optimization problems.

[0062] Large models include large mathematical modeling models, large algorithm programming models, large target value solving models, large production simulation and verification models, and large information interaction models, etc., to realize functions such as automatic mathematical modeling and automatic program writing; these large models can be different large models, or the same large model can solve different tasks through different prompts;

[0063] The production simulation verification module is used for lightweight production simulations. It uses discrete event simulation methods to create a production simulation environment and verify the production line design scheme. For more refined simulations, such as those involving AGV path planning, worker operation, and vehicle operation conflicts, auxiliary tools such as Plant Simulation software and Unity 3D software can be used to perform more detailed production simulation verification and optimization.

[0064] A production line optimization design device based on a large model, used to implement the steps of the method as described in any one of claims 1-9, characterized in that, as shown in the figure, it includes a user interaction layer, a software service layer, a core computing and storage layer, etc.

[0065] A computer-readable storage medium is characterized in that a computer program is stored on the storage medium, and when the computer program is executed by a processor, a production line autonomous optimization design method based on a large model and intelligent algorithms is implemented.

[0066] The present invention has the following beneficial effects and advantages:

[0067] 1. Improve the efficiency, accuracy and intelligence of production line optimization design. For production line planning and design problems, input the optimization problem description and process data, and the module library or large model can be automatically called to generate a complete production line planning and design scheme, including production line balancing scheme, production line layout scheme, and resource configuration scheme. There is no need for manual modeling, programming, debugging, optimization and verification, which improves the efficiency of production line design, shortens the design cycle and supports just-in-time manufacturing.

[0068] 2. Improve the versatility of production line design. Through the collaboration of module library and large model, production line optimization problems under different production line types, processes, resource constraints and optimization objectives can be solved regardless of whether they have been modeled in the module library in the early stage, thus improving the versatility of production line optimization design.

[0069] 3. Enhance the interactivity of production line optimization design. The generated production line planning and design schemes can be modified and adjusted manually, realizing a comprehensive design based on module library, large model and human experience, and improving the feasibility of production line design schemes. Attached Figure Description

[0070] Figure 1 The overall process of production line autonomous optimization design method based on large model and intelligent algorithm;

[0071] Figure 2 Architecture diagram for autonomous optimization design of production lines based on large models and intelligent optimization algorithms;

[0072] Figure 3 This is a flowchart illustrating the methodology for production line optimization design based on a large model.

[0073] Figure 4 A diagram showing the components of an existing module library;

[0074] Figure 5 Design a software system architecture diagram for production line optimization based on a large model;

[0075] Figure 6 Interaction process diagram of a large model for production line optimization design (an example);

[0076] Figure 7 The workstation allocation scheme obtained after production line balancing optimization (an example);

[0077] Figure 8 A table showing the results of a workshop layout planning scheme (in an example);

[0078] Figure 9 A diagram showing the result of a workshop layout plan (an example).

[0079] Figure 10 A diagram showing the results of optimized production resource allocation (in a specific embodiment);

[0080] Figure 11 Hardware system diagram for autonomous optimization design of production lines based on large models and intelligent algorithms. Detailed Implementation

[0081] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the embodiments. However, the scope of protection of this invention is not limited to the specific embodiments described below.

[0082] A production line autonomous optimization design method based on large models and intelligent algorithms is used for automated planning and design of production lines driven by process information. This reduces manual modeling, programming, and iterative optimization, enabling autonomous optimization design of the production line. The overall flowchart and architecture are as follows: Figure 1 and Figure 2 As shown, it includes the following steps:

[0083] Step S1: Based on the large model, understand the input information of the production line optimization problem, extract process information, modeling information, etc., analyze whether it is a new production line design or an adjustment to the existing production line, and analyze what optimizations need to be made, including production line balancing optimization, production line layout optimization, production resource allocation optimization, etc.

[0084] Step S2: Based on the large model, for the input optimization problem, including optimization problems such as production line balancing, production line layout, and production resource allocation, determine whether the corresponding mathematical model and optimization algorithm have been established in the module library. If yes, proceed to step S3; otherwise, proceed to step S4.

[0085] Step S3: For each optimization problem, the pre-established mathematical model, optimization algorithm, and production line simulation evaluation environment are called from the module library through function calling to perform iterative optimization and obtain the production line planning and design scheme.

[0086] Step S4: For each optimization problem, construct the corresponding mathematical model, optimization algorithm, and production line simulation evaluation environment based on the large model, and perform iterative optimization to obtain the production line planning and design scheme; for similar problems, the large model can first understand and learn based on the models and algorithms already built in the module library, and then imitate and generate them to improve the accuracy of the solution.

[0087] Step S5: For the optimized production line scheme, construct a production simulation environment consistent with the actual production line, perform production constraint checks, verify the feasibility of the production line design scheme, and evaluate the production line performance, such as equipment utilization and capacity; for problems such as process flow errors, production logic errors, process constraint violations, and program errors, return the runtime error information to the corresponding large model above for modification;

[0088] Step S6: Perform human-computer interaction adjustments. By changing the optimization target or weight, supplementing process and resource constraints, etc., change the obtained production line optimization design scheme. Depending on the optimization method, return to step S3 or step S4, and continue to adjust and optimize the production line design scheme through function calls or large model solving until satisfied.

[0089] Step S7: Obtain the optimal production line design scheme, including the optimal process allocation, workstation layout, production resource allocation, etc., and present and export the result chart of the production line planning and design scheme; add the optimization problem description, mathematical model, optimization algorithm, target value calculation program, and production simulation environment generated by the large model to the corresponding sub-problem in the module library, so as to provide model and algorithm support for solving similar optimization problems in the future.

[0090] Specifically, step S1 includes the following steps:

[0091] Step S1.1: Receive multimodal input information, which may include optimization problem descriptions, process data tables, process flow diagrams, CAD layouts, etc. It may also prompt the user to input or upload process tables using a unified structure, or provide the folder address where the process data is located; depending on the data requirements of the optimization problem, there may be multiple rounds of information input;

[0092] Step S1.2: Extract the features of the input information and perform structured unification.

[0093] Specifically, multimodal data such as images and CAD files are identified, and information related to the problem is extracted. The input information is divided into the following categories: feature description of the optimization problem, optimization objective, constraints, and process data input.

[0094] Step S1.3: Analyze what type of optimization problem the input problem belongs to, whether it is a new production line or an adjustment design of an existing production line (production line evolution optimization design), and what optimizations are needed;

[0095] The main categories and functional descriptions of production line optimization issues are as follows:

[0096] Production line / assembly line balancing optimization problem: The input information includes product process time, task / process sequence constraints, resource constraints, etc. The task / process is assigned to appropriate workstations. The optimization objectives are to maximize workstation load balance, achieve a high balance rate, minimize the number of workstations, and minimize equipment investment costs. The output is the optimal allocation scheme of processes on workstations.

[0097] Production line / workshop layout optimization problem: The input information includes workshop area, workstation area, logistics transfer relationship and frequency between workstations, and workstation layout constraints. The optimization objective is to minimize the total material flow in the workshop, and the output is the layout location information of workstations within the production line / workshop.

[0098] Production resource allocation optimization: The inputs are the task sequence and processing time of the workstation, production line layout, product process flow, etc. The optimization objectives are to maximize production capacity and minimize resource input costs. The outputs are the quantity of production resources in the workshop, such as the number of AGVs, the number of workers, the number of parallel devices, the capacity of the temporary storage area, etc.

[0099] If it is a newly established production line, the system will prompt you to enter the design scheme, process information, and constraints of the existing production line, as well as the process information and resource constraints of the new product. The newly established production line has the following new optimization objectives, such as minimizing equipment relocation costs, new equipment investment costs, and new personnel configuration costs.

[0100] Determine the type of optimization problem based on the problem description. For example, tasks such as process / task allocation, production line balancing, and assembly line balancing fall under production line balancing optimization problems; descriptions of workstation layout, workshop layout, production line layout, and facility planning fall under layout optimization; descriptions of determining the quantity of production resources, AGVs, workers, temporary storage capacity, and parallel workstations fall under resource allocation optimization problems; and modifications to the original production line to accommodate the production of new products fall under production line evolution optimization design problems.

[0101] Step S1.4: Based on the complete data requirements of the optimization problem, remind the user to input the missing process and constraint information. The user can be reminded to upload the process data in the corresponding format using a structured table to prevent omission of key information and facilitate data extraction.

[0102] Figure 6 This is a screenshot of the interaction information of a large model in a certain embodiment.

[0103] The large model used in all steps of this invention is described in detail below:

[0104] The large models used include information interaction models, method selection models, mathematical modeling models, algorithm programming models, target value solving models, and production simulation verification models, etc.

[0105] These large models can be fine-tuned versions of the large models, based on established module libraries (problem description and classification libraries, mathematical model libraries, optimization algorithm libraries, objective value calculation libraries, and production simulation environment libraries), such as... Figure 3 As shown, supervised learning and reinforcement learning methods are used to enable the large model with the corresponding function to generate classification of optimization problems, mathematical models, optimization algorithms, target value calculation code, and simulation environment code, and to provide feedback learning to improve the accuracy of the generated models and code;

[0106] For example, extract a description and some process data about production line balancing optimization from the module library, let the information interaction model analyze which type of optimization problem it belongs to and what process information is still needed. Based on the correct problem classification and complete process information in the module library, train the information interaction model and make fine adjustments.

[0107] These large models can also be untuned large models. Depending on the capabilities of general large models, the size of the problem to be optimized, the difficulty of solving, the accuracy requirements, experimental conditions, etc., untuned large models can also be used, such as large models deployed locally (e.g., DeepSeek, ChatGPT, etc.) or large models called from the network server via API (e.g., DeepSeek, ChatGPT, etc.).

[0108] These large models (mathematical modeling models, programming models, etc.) can be different large models, or the same large model can solve different tasks after passing through different prompts;

[0109] When solving these large models, we first find descriptions of similar optimization problems and corresponding mathematical models, optimization algorithms, production simulation environments, etc. in the module library to understand the modeling and programming ideas. Then, based on the characteristics of the problem to be solved, we perform imitation generation, modeling and programming to improve the accuracy of the generated content of the large model.

[0110] Specifically, step S2 includes the following:

[0111] The completed module library includes a problem description and classification library, a mathematical model library, an optimization algorithm library, an objective value calculation library, and a production simulation environment library, such as... Figure 4 As shown;

[0112] The target value calculation library constructs a simplified production simulation environment based on process flow and resource constraints to calculate information such as production capacity;

[0113] The production simulation environment library contains production and logistics components and production simulation engines required for production simulation verification. Components include equipment, AGVs, workers, products, conveyor lines, etc. Using discrete event simulation methods, based on process flow and production line design schemes, a production simulation model consistent with actual production is constructed.

[0114] Based on the detailed optimization problem and its characteristics, determine whether the input optimization problem has already been modeled in the module library. The specific characteristics of the problem are as follows:

[0115] Production line balancing optimization problems can be categorized as follows: single-product production line balancing optimization, multi-product production line balancing optimization; single-sided production line / assembly line optimization, double-sided production line / assembly line optimization; balancing optimization considering shared resources, balancing optimization not considering shared resources, etc.

[0116] Production line layout optimization issues include the following categories: single-row layout, multi-row layout, layout with special position constraints (such as a certain workstation being fixed in a certain position), layout without special position constraints, single-floor layout, multi-floor layout, etc.

[0117] The optimization of production resource allocation includes the following categories: transmission line flow transformation, AGV delivery type, requiring workers, not requiring workers, with temporary storage area, without temporary storage area, etc.

[0118] For each type of optimization problem, such as production line balancing, production line layout, and resource allocation optimization, they can be divided into single-objective and multi-objective optimization based on the optimization objective.

[0119] Specifically, step S4 involves modeling the optimization problem, programming the algorithm, calculating the objective value, and iteratively optimizing it. The flowchart is as follows: Figure 3 As shown, it includes the following steps:

[0120] Step S4.1: Establish a mixed integer linear programming (MILP) model / mathematical model for the optimization problem. Find similar optimization problems in the module library, refer to the descriptions and mathematical models of similar problems, understand the modeling ideas and the meaning of the model, and then, based on the description of the current optimization problem, imitate and construct a mathematical model, including the optimization objective and constraints.

[0121] For multiple optimization objectives, a negative sign can be used to unify the optimization direction;

[0122] The following example illustrates the construction of the objective function for optimization:

[0123] For the production line balancing optimization problem, the formula for calculating the balance rate is as follows:

[0124]

[0125] Where T iLet C represent the operation time of the i-th workstation, K represent the total number of workstations on the production line, and C represent the operation time of the ith workstation. T Indicates the production line cycle time;

[0126] The load balancing calculation formula is as follows:

[0127]

[0128] In the formula, SI represents the workstation balance index, N is the number of workstations, N is the total number of tasks, and t i x is the operation time for assembly task i. ik This indicates that the i-th assembly task is assigned to the k-th workstation;

[0129] The formula for calculating the total material flow in the workshop for optimizing the production line layout is as follows:

[0130]

[0131] Among them W ij D represents the amount of material transported from workstation i to workstation j. ij The distance between workstation i and workstation j is represented by the following formula:

[0132]

[0133] Where, x i y i This represents the x and y coordinates of workstation i, x j y j This indicates the x and y coordinate positions of workstation j;

[0134] The formula for calculating equipment utilization rate to optimize resource allocation is as follows:

[0135]

[0136] Among them U t T represents the utilization rate of a certain device at time t. a T represents the actual operating time of the equipment within the statistical period. p Indicates the planned production time within the statistical period;

[0137] Step S4.2: Based on the size of the optimization problem, let the large model determine whether to use a solver (such as Gurobi, CPLEX, etc.) or an intelligent optimization algorithm to solve it;

[0138] For small-scale optimization problems, a solver is used to solve them, and solver execution code is generated. First, similar optimization problems are found in the module library, the corresponding problem descriptions are examined, the mathematical formulas and solver code are understood, and imitation learning is performed. Then, based on the description and mathematical model of this optimization problem, the corresponding solver execution code is generated by imitation.

[0139] For large-scale optimization problems, intelligent optimization algorithms are used to solve them, generating corresponding intelligent algorithm code. For different optimization problems, the selection priority of optimization algorithms is set based on experience and test results. For each type of optimization algorithm, modular programming is performed in the algorithm library. For different functional functions, such as encoding and local search, the algorithm flow is summarized and modularized in a structured way to facilitate understanding, programming, and calling. This also makes it easier for the large programming model to understand and imitate programming for new problems. When programming the large programming model, the corresponding algorithm for similar problems is first found in the algorithm library, understood and learned, and then imitated to generate the code, thereby improving the quality of program writing.

[0140] Step S4.3: Construct a program to calculate the target value, as follows:

[0141] To optimize the allocation of production resources, a lightweight production simulation model is constructed based on process flow, production line balancing scheme, workshop layout scheme, etc. using discrete event simulation method. Multi-factor experimental method is used to calculate information such as workshop capacity and equipment utilization rate under different production resource configurations, so as to determine the optimal production resource configuration.

[0142] During the calculation, if an error occurs at any step, the error information is returned to the corresponding large model, such as the large model being modeled or programmed, so that it can be updated and modified.

[0143] Step S4.4: Based on the algorithm flow, problem size, and changes in the target value, determine the number of iterations, determine the next-generation selection method, perform iterative optimization, and obtain the production line optimization design scheme;

[0144] Step S5 verifies the optimized solution through production simulation and evaluates its production performance, including the following steps:

[0145] Step S5.1: Summarize the production line optimization design schemes, such as one or all of the schemes in task allocation, workstation layout, and production resource allocation, and use the discrete event simulation method to establish a production simulation model consistent with the actual production line.

[0146] Step S5.2: Based on the input process information, such as process flow and workstation processing time, perform production simulation;

[0147] Step S5.3: If there is an error, analyze the error information and provide feedback to the corresponding large model for model or code modification. For example, provide feedback to the large modeling model if the constraints of process A and process B do not match.

[0148] Step S5.4: Run the process to obtain production line performance data, such as capacity, equipment utilization rate, worker utilization rate, bottleneck processes, and blockage status.

[0149] Step S6 involves adjusting the optimization scheme through human-computer interaction, and its characteristics are as follows:

[0150] Users can manually adjust production line design schemes, supplement process and constraint information; for example, restrict a certain process to have explosion-proof requirements, requiring the setting of an explosion-proof room, and specifying the direction of explosion venting; for example, specifying the location of workshop entrances and exits, and the direction of product flow; for example, adding or removing targets, changing the weight of target values, etc.

[0151] Step S7 presents the final production line design scheme and exports the results, as follows:

[0152] The system presents the adjusted production line design and production performance, obtaining the optimal production line balance, layout, and resource allocation schemes. Specifically, it obtains the allocation of processes in workstations, the tasks handled by each workstation, and the total time required; the location of each workstation within the workshop; the number of AGVs, workers, and machines configured in the production line, the capacity of the temporary storage area, etc.; and exports the result charts of the production line design.

[0153] The software system architecture diagram for production line optimization design based on a large model is as follows: Figure 5 As shown, it includes a front-end interaction module, an information interaction model, a method selection model, an optimization and solution module, and a production simulation and verification model. Based on user input information, it automatically optimizes and designs production line planning and design schemes.

[0154] Figure 6 This is an interactive process diagram of a large model for production line optimization design in a certain embodiment. When the user inputs "Help me perform production line balancing optimization", the large model prompts the user to input the required information or files.

[0155] Figure 7 In a certain embodiment, the optimized production line balancing scheme is shown. The horizontal axis represents the workstations, and the vertical axis represents the assembly tasks and time of each workstation. Based on the input process data and the original production line balancing scheme, the optimized production line balancing rate increased from the initial 56% to 83%, an improvement of 48%.

[0156] Figure 8 and Figure 9 This is a diagram of an optimized workshop layout scheme under a certain embodiment. Figure 8 This represents the row number, length, width, center coordinates in the length direction, center coordinates in the width direction, lower left corner coordinates in the length direction, and lower left corner coordinates in the width direction for each of the 7 workstations in the workshop. Figure 9 For the corresponding Figure 8 Workshop layout diagram;

[0157] Figure 10The graph shows the optimization results of production resource allocation in a certain embodiment. The horizontal axis represents the number of AGVs, the left vertical axis represents the production capacity, and the right vertical axis represents the AGV utilization rate. It can be seen that as the number of AGVs increases, the workshop production capacity shows an increasing trend, while the AGV utilization rate gradually decreases. Based on the requirements for production capacity and AGV utilization rate, an appropriate number of AGVs can be selected.

[0158] The module library includes the problem description, mathematical model, optimization algorithm, objective value solution code, and production simulation environment for solving this new optimization problem using a large model, providing model and algorithm support for subsequent solutions.

[0159] This invention also provides a production line autonomous optimization design system based on a large model, comprising:

[0160] The front-end interaction module receives user input and presents the production line optimization design results, enabling interaction with the user.

[0161] The module library includes a problem description and classification library, a mathematical model library, an optimization algorithm library, a target value calculation library, and a production simulation environment library, which stores mathematical models, optimization algorithms, and production simulation code for related production line optimization problems.

[0162] Large models include large mathematical modeling models, large algorithm programming models, large target value solving models, large production simulation and verification models, and large information interaction models, etc., to realize functions such as automatic mathematical modeling and automatic program writing; these large models can be different large models, or the same large model can solve different tasks through different prompts;

[0163] The production simulation verification module is used for lightweight production simulations. It uses discrete event simulation methods to create a production simulation environment and verify the production line design scheme. For more refined simulations, such as those involving AGV path planning, worker operation, and vehicle operation conflicts, auxiliary tools such as Plant Simulation software and Unity 3D software can be used to perform more detailed production simulation verification and optimization.

[0164] A production line optimization design device based on a large model includes a user interaction layer, a software service layer, and a core computing and storage layer.

[0165] A computer-readable storage medium storing a computer program, which, when executed by a processor, implements a production line autonomous optimization design method based on large models and intelligent algorithms.

[0166] The specific embodiments described above can be partially adjusted by those skilled in the art in different ways without departing from the principles and spirit of this invention. Modifications and alterations to the invention should also fall within the scope of protection of the claims. The scope of protection of this invention is defined by the claims and is not limited to the specific embodiments described above; all implementations within the scope of these claims are bound by this invention.

Claims

1. A large model and intelligent algorithm-based production line autonomous optimization design method, characterized in that, The method comprises the following steps: Step S1: based on the large model, understand the input information of the production line optimization problem, extract the production process, modeling information, and analyze which optimization needs to be performed; Step S2: based on the large model, judge whether the corresponding mathematical model and optimization algorithm have been established in the module library for the input optimization problem, if yes, go to step S3, if not, go to step S4; Step S3: for each optimization problem, call the established mathematical model, optimization algorithm and production line simulation evaluation environment from the module library, and perform iterative optimization to obtain a production line planning and design scheme; Step S4: for each optimization problem, based on the large model, construct the corresponding mathematical model, optimization algorithm and production line simulation evaluation environment, and perform iterative optimization to obtain a production line planning and design scheme; Step S5: for the optimized production line optimization scheme, construct a production simulation environment consistent with the actual production line, and perform production simulation verification and production performance; Step S6: perform human-computer interaction adjustment, change or supplement the target value, process information and resource constraint information, and manually adjust the production line optimization design scheme; return to step S3 or step S4 according to the optimization mode, continue to optimize the production line design scheme, and stop until the result is satisfactory; Step S7: obtain the best production line design scheme.

2. The large model and intelligent algorithm based production line autonomous optimization design method according to claim 1, characterized in that, Step S1 comprises the following steps: Step S1.1: receive multi-modal input information; Step S1.2: extract input information features and perform structured unification; Step S1.3: analyze which optimization problem the input problem belongs to, whether it is a new production line or an adjustment and design of an original production line, and which optimization needs to be performed; Step S1.4: according to the complete data requirement of the optimization problem, remind the user to input the missing process and constraint information.

3. The large model and intelligent algorithm based production line autonomous optimization design method of claim 1, wherein, The large model includes an information interaction large model, a method selection large model, a mathematical modeling large model, an algorithm programming large model, a target value solving large model, and a production simulation verification large model.

4. The large model and intelligent algorithm based production line autonomous optimization design method of claim 1, wherein, In step S2, the following features are included: The built module library includes a problem description and classification library, a mathematical model library, an optimization algorithm library, a target value calculation library, and a production simulation environment library; The target value calculation library calculates the target value according to the process flow and resource constraints; The production simulation environment library includes production and logistics components and a production simulation engine required for production simulation verification, and uses a discrete event simulation method to construct a production simulation model consistent with actual production based on the process flow and the production line design scheme; According to the characteristics of the sub-optimization problem, it is judged whether the input optimization problem has been modeled in the module library, and the problem characteristics are as follows: For production line balancing optimization problems, the following classifications are included: single variety production line balancing optimization, multi-variety production line balancing optimization; single-sided assembly line optimization, double-sided assembly line optimization; balancing optimization considering shared resources, balancing optimization not considering shared resources; For production line layout optimization problems, the following classifications are included: single-row layout, multi-row layout, special location constraint layout, no special location constraint layout, single-floor layout, multi-floor layout; For production resource configuration optimization problems, the following classifications are included: transmission line flow type, AGV distribution type, need workers, no need workers, have temporary storage area type, no temporary storage area type; According to the optimization target, each type of optimization problem can be divided into single-objective and multi-objective optimization.

5. The large model and intelligent algorithm based production line autonomous optimization design method according to claim 1, characterized in that, Step S4 is to model the optimization problem, program the algorithm, calculate the target value, and iterate the optimization, Including the following steps: Step S4.1: Establish a mixed integer linear programming model of the optimization problem, find similar optimization problems in the module library, refer to the description and mathematical model of similar problems, understand the modeling ideas and model meanings, then according to the current optimization problem description, imitate the construction of the mathematical model, including the optimization target and the constraint condition; for multiple optimization targets, unify the optimization direction through negative signs; Step S4.2: According to the size of the optimization problem, let the large model determine whether to use a solver or an intelligent optimization algorithm to solve it; For small-scale optimization problems, use a solver to solve and generate a solver running code; first find similar optimization problems in the module library, understand the problem description, mathematical formula, and solver code, and learn by imitation; then, according to the optimization problem description and mathematical model, imitate the generation of the corresponding solver running code; For large-scale optimization problems, use an intelligent optimization algorithm to solve and generate an intelligent algorithm code; For different optimization problems, set the selection priority of the optimization algorithm according to experience and test results; for each type of optimization algorithm, modularize the programming in the algorithm library, and summarize the algorithm flow and perform modular structured programming for different function functions such as coding and local search, so as to facilitate understanding, programming and calling, and also facilitate the programming of the large model to understand and imitate programming for new problems; when the large model is programmed, first find the corresponding algorithm of similar problems in the algorithm library, understand and learn, then imitate and generate, to improve the quality of program writing; Step S4.3: Build a target value calculation program to calculate the corresponding target value: For production resource configuration optimization, based on the process flow, line balancing scheme, and workshop layout scheme, a lightweight production simulation model is built using the discrete event simulation method, and the workshop capacity and equipment utilization information under different production resource configurations are calculated using the multi-factor experiment method to determine the optimal production resource configuration; When running the calculation, return the error information of each step to the corresponding large model, such as the modeling large model and the programming large model, for modification; Step S4.4: According to the algorithm flow, problem size, and target value change, determine the number of iterations, the next generation of optimization method, and perform iterative optimization to obtain the line optimization design scheme.

6. The large model and intelligent algorithm based production line autonomous optimization design method of claim 1, wherein, Step S5 verifies and evaluates the production performance of the optimization scheme through production simulation, Including the following steps: Step S5.1: Summarize the line optimization design scheme, and establish a production simulation model consistent with the actual production line using the discrete event simulation method; Step S5.2: Based on the input process information such as process flow and workstation processing time, perform production simulation; Step S5.3: If there is an error, analyze the error information and feed it back to the corresponding large model for model or code modification; Step S5.4: Run to get the production performance of the production line, including capacity, equipment utilization, worker utilization, bottleneck process, and congestion.

7. The large model and intelligent algorithm based production line autonomous optimization design method according to claim 1, characterized in that, Step S6 adjusts the optimization scheme through human-computer interaction, Its features are as follows: User manually adjusts the production line design scheme, supplements process and constraint information.

8. The large model and intelligent algorithm based production line autonomous optimization design method of claim 1, wherein, Step S7 presents the final production line design scheme and exports the results: The adjusted production line design scheme and production performance are presented to obtain the best production line balancing scheme, layout scheme, and resource allocation scheme, and the result chart of the production line design is exported. The problem description, mathematical model, optimization algorithm, target value calculation program, and production simulation environment of solving this new optimization problem by the large model are added to the module library, providing model and algorithm support for subsequent solving of similar problems.

9. A large model-based production line autonomous optimization design system for implementing the large model and intelligent algorithm-based production line autonomous optimization design method according to any one of claims 1-8, characterized in that, It includes: Front-end interaction module, receiving user information input and presenting production line optimization design results, realizing interaction with users; Module library, including problem description and classification library, mathematical model library, optimization algorithm library, target value calculation library, and production simulation environment library, storing mathematical models, optimization algorithms, and production simulation environments for related production line optimization problems; Large model, including mathematical modeling large model, algorithm programming large model, target value solving large model, production simulation verification large model, and information interaction large model, realizing automatic mathematical modeling and automatic program writing functions; Production simulation verification module, for lightweight production simulation, writing production simulation environment through discrete event simulation method to verify the production line design scheme.

Images