Dynamic job shop scheduling optimization method and device, equipment and medium

Abstract

The invention relates to a dynamic job-shop scheduling optimization method, device, equipment and medium, and belongs to the technical field of intelligent manufacturing and production scheduling. The dynamic job-shop scheduling optimization method comprises the steps of obtaining key parameters of a dynamic job-shop scheduling scene and a classical scheduling rule, performing offline evolution on the classical scheduling rule by adopting a large language model, and optimizing the dynamic job-shop scheduling scene; obtaining an offline optimal scheduling rule; and obtaining a machine state and a to-be-processed job list of the dynamic job shop in real time, generating a job allocation scheme based on the offline optimal scheduling rule, the machine state and the to-be-processed job list, and scheduling the to-be-processed jobs of the dynamic job shop in real time based on the job allocation scheme. And the requirements of flexible manufacturing on quick response and accurate scheduling are met.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing and production scheduling technology, and in particular to a dynamic job shop scheduling optimization method, device, equipment and medium. Background Technology

[0002] In the field of dynamic job shop scheduling, traditional methods often rely on fixed scheduling rules (such as FIFO first-in-first-out, SPT shortest processing time priority, etc.). Such rules have inherent defects such as poor adaptability and lag in dynamic response. When faced with complex production scenarios such as dynamic order arrival and sudden machine failure, fixed rules cannot balance production efficiency and resource utilization, which can easily lead to extended production cycles, increased equipment idle rate, and difficulty in adapting to the needs of flexible manufacturing.

[0003] With the development of intelligent technology, heuristic methods based on genetic algorithms and particle swarm optimization are gradually being applied to scheduling optimization, but there are still obvious limitations: On the one hand, the optimization process relies heavily on manually designed coding methods and operators, which are not adaptable to scheduling scenarios with complex process routes and changing constraints. Under multiple dynamic disturbances, it is easy to get trapped in local optima and cannot achieve global optimal scheduling. On the other hand, existing heuristic methods lack the accumulation and reuse mechanism of scheduling knowledge. Each scheduling task needs to be iterated and optimized from scratch, which has high computational cost and long iteration cycle, making it difficult to meet the real-time requirements of industrial scenarios.

[0004] Dynamic scheduling in industrial production requires processing dynamic information from multiple sources, including emergency order insertions, equipment load fluctuations, and material delays. However, existing technologies lack the ability to perceive and make adaptive decisions in real time regarding this dynamic information. They are unable to update dynamic information such as machine fault status and new order arrivals in real time, resulting in a serious disconnect between the generated scheduling schemes and actual production needs. This fails to meet the practical application requirements of flexible manufacturing for rapid response and precise scheduling. Summary of the Invention

[0005] In view of this, it is necessary to provide a dynamic job shop scheduling optimization method, device, equipment and medium to solve the technical problem that the scheduling schemes generated in complex production scenarios are difficult to achieve rapid response and accurate scheduling.

[0006] To address the aforementioned problems, in a first aspect, the present invention provides a dynamic job shop scheduling optimization method, comprising: The key parameters and classic scheduling rules of the dynamic job shop scheduling scenario are obtained, and the classic scheduling rules are evolved offline using a large language model to obtain the offline optimal scheduling rule. The machine status and the list of pending jobs in the dynamic workshop are obtained in real time. Based on the offline optimal scheduling rules, the machine status and the list of pending jobs, a job allocation scheme is generated. Based on the job allocation scheme, the pending jobs in the dynamic workshop are scheduled in real time.

[0007] In one possible implementation, the key parameters include job data, number of machines, and production constraints.

[0008] In one possible implementation, the large language model includes PAI-LLM, which includes a population generator, an evaluator, and a mutation operator; the offline evolution of the classical scheduling rule using the large language model to obtain the offline optimal scheduling rule includes: Step 1: Use the classic scheduling rule as the initial seed population; Step 2: Construct prompt words based on the key parameters. Using the initial seed population as a template, input the prompt words into PAI-LLM to generate multiple scheduling rules. Then, use the population generator to merge the multiple scheduling rules with the classic scheduling rules to obtain the initial rule population. Step 3: Determine the maximum completion time of individuals in the initial rule population, calculate the fitness value of individuals based on the maximum completion time, evaluate the performance of individuals based on the fitness value using the evaluator, and use roulette wheel selection to select individuals to obtain multiple elite individuals and the core decision characteristics of elite individuals. Step 4: Based on the core decision-making characteristics of the elite individuals, use the mutation operator of PAI-LLM to perform rule mutation on the multiple elite individuals to obtain multiple mutation rules and the core decision-making characteristics of the multiple mutation rules. The mutation rules are heuristic code segments generated autonomously by PAI-LLM. Step 5: Repeat steps 2 to 4 until the preset iteration threshold is reached, then stop the iteration and output the offline optimal scheduling rule and the core decision features of the offline optimal scheduling rule. The offline optimal scheduling rule is the scheduling rule with the minimum maximum completion time.

[0009] In one possible implementation, the classic scheduling rules include first-in-first-out (FIFO) rules, shortest processing time priority rules, longest processing time priority rules, and most remaining jobs priority rules.

[0010] In one possible implementation, the core decision features include considering remaining processing time, taking into account machine load, and prioritizing the scheduling of emergency orders.

[0011] In one possible implementation, the step of generating a job allocation scheme based on the offline optimal scheduling rules, the machine status, and the list of jobs to be processed, and then performing real-time scheduling of the jobs to be processed in the dynamic job shop based on the job allocation scheme, includes: Step 1: Construct the core parameters of the dynamic workshop at the current moment based on the machine status and the list of jobs to be processed; Step 2: Generate a job allocation scheme based on the core parameters and offline optimal scheduling rules; determine the jobs to be processed in the next time step and the key scheduling information of the jobs to be processed in the next time step based on the job allocation scheme, so as to generate a scheduling event. Step 3: Based on the scheduling event, synchronize and update the production status of the dynamic workshop, and update the list of jobs to be processed and the machine status; Step 4: Obtain the number of processed jobs based on the updated list of jobs to be processed. If the number of processed jobs has not reached the preset feedback interval, repeat steps 1 to 3. If the number of processed jobs has reached the preset feedback interval, execute step 5. Step 5: Obtain the scheduling result of the scheduling event, and after feeding back the scheduling result, repeat steps 1 to 3 until the scheduling of the pending jobs in the pending job list is completed.

[0012] In one possible implementation, the key scheduling information includes job ID, assigned machine ID, start time, and end time.

[0013] Secondly, the present invention also provides a dynamic job shop scheduling optimization device, comprising: The scheduling rule evolution module is used to obtain key parameters and classic scheduling rules for dynamic job shop scheduling scenarios, and to perform offline evolution of the classic scheduling rules using a large language model to obtain the offline optimal scheduling rule. The job scheduling module is used to obtain the machine status and the list of jobs to be processed in the dynamic job workshop in real time, generate a job allocation scheme based on the offline optimal scheduling rules, the machine status and the list of jobs to be processed, and perform real-time scheduling of the jobs to be processed in the dynamic job workshop based on the job allocation scheme.

[0014] Thirdly, the present invention also provides an electronic device, comprising: a processor and a memory; The memory stores a computer-readable program that can be executed by the processor; When the processor executes the computer-readable program, it implements the steps in the dynamic job shop scheduling optimization method described above.

[0015] Fourthly, the present invention also provides a computer-readable storage medium for storing a computer-readable program or instructions, wherein the program or instructions, when executed by a processor, are capable of implementing the steps in the dynamic job shop scheduling optimization method described in any one of the above-mentioned method items.

[0016] The beneficial effects of this invention are as follows: It acquires key parameters and classic scheduling rules for dynamic job shop scheduling scenarios, uses a large language model to perform offline evolution of the classic scheduling rules, and obtains the offline optimal scheduling rules; it acquires the machine status and the list of jobs to be processed in the dynamic job shop in real time, generates a job allocation scheme based on the offline optimal scheduling rules, machine status, and the list of jobs to be processed, and schedules the jobs to be processed in the dynamic job shop in real time based on the job allocation scheme; through the scheduling rule evolution mechanism of the large language model, it achieves autonomous evolution of scheduling rules, significantly improving the adaptability of scheduling rules to complex production scenarios; and it generates scheduling schemes through offline optimal scheduling rules, solving the problem of scheduling schemes being disconnected from actual production, thus meeting the needs of flexible manufacturing for rapid response and precise scheduling. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart illustrating an embodiment of the dynamic job shop scheduling optimization method provided by the present invention; Figure 2 A Gantt chart showing the scheduling results of the dynamic job shop scheduling optimization method provided by this invention; Figure 3 A schematic diagram of an embodiment of the dynamic job shop scheduling optimization device provided by the present invention; Figure 4 A schematic diagram of an embodiment of the electronic device provided by the present invention. Detailed Implementation

[0019] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.

[0020] In this document, the term "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0021] This invention discloses a method, apparatus, equipment, and medium for dynamic job shop scheduling optimization, which can be used in a computer. The method, apparatus, or computer-readable storage medium involved in this invention can be integrated with the aforementioned equipment or be relatively independent.

[0022] One specific embodiment of the present invention discloses a dynamic job shop scheduling optimization method, which can be executed by a computer, specifically by one or more processors of the computer. For example... Figure 1 As shown, the dynamic job shop scheduling optimization method includes: S101. Obtain key parameters and classic scheduling rules for dynamic job shop scheduling scenarios, and use a large language model to perform offline evolution of the classic scheduling rules to obtain the offline optimal scheduling rules. It should be noted that, using classic scheduling rules as seeds, PAI-LLM drives population expansion, performance evaluation, elite selection, and mutation, enabling the scheduling rules to evolve autonomously and significantly improving their adaptability to complex production scenarios.

[0023] S102. Obtain the machine status and the list of pending jobs in the dynamic work workshop in real time, generate a job allocation scheme based on the offline optimal scheduling rules, machine status and the list of pending jobs, and schedule the pending jobs in the dynamic work workshop in real time based on the job allocation scheme. It should be noted that the scheduling scheme is generated by using offline optimal scheduling rules, which meets the needs of flexible manufacturing for rapid response and precise scheduling.

[0024] In some embodiments, in step S101, key parameters of the dynamic job shop scheduling scenario and classic scheduling rules are obtained. The classic scheduling rules include First-In-First-Out (FIFO), Shortest Processing Time First (SPT), Longest Processing Time First (LPT), and Most Remaining Work First (MWKP). The code for the classic scheduling rules is written in Python functions. In the code, the FIFO rule directly returns the first processable job, and the SPT rule selects the job with the shortest processing time in the current process. The key parameters of the dynamic job shop scheduling scenario can be obtained through dynamic environment perception. The key parameters include job data, number of machines, and production constraints. The dynamic environment perception collects data from the dynamic job shop through a real-time production status monitor. The core function is to collect and update dynamic information such as job arrival time, machine availability status (normal / faulty), and process completion progress in real time. Based on key parameters and classic scheduling rules, a large language model is used to perform offline evolution of the classic scheduling rules to obtain the offline optimal scheduling rule. The large language model includes PAI-LLM (Process-Aware Intelligent Large Language Model), which comprises a population generator, an evaluator, and a mutation operator. The steps are as follows: Step 1: Use the classic scheduling rules as the initial seed population. By using the classic scheduling rules as the initial seed population, a basic template is provided for rule evolution, ensuring that the initial population has scheduling feasibility. Step 2: Construct prompt words based on key parameters. Using the initial seed population as a template, input the prompt words into PAI-LLM to generate multiple scheduling rules. The population generator then merges these rules with classic scheduling rules to obtain an initial rule population. Based on the prompt words, construct the input instructions for PAI-LLM. The prompt words contain key parameters of the current scheduling scenario. Using the initial seed population as a template, input the key parameters into PAI-LLM to generate new scheduling rules. The population generator in PAI-LLM merges the new rules with the seed population to construct an initial rule population of appropriate size, completing PAI-LLM-driven population expansion. In PAI-LLM, to drive population expansion, first, construct the prompt word template: "A dynamic job workshop scenario with 20 jobs and 5 machines exists. Jobs need to be processed on different machines according to preset procedures. Machines may suddenly malfunction, requiring the generation of a scheduling rule function." Function input parameters: available_jobs (list of available jobs), job_data (job information, including processing_times, current_step). The current process and machine_avail (list of available machine times) are used for the selection of jobs. The output is the selected job ID. The objective is to minimize the fitness value index (Maximum completion time Makespan). The rules include natural language annotations. See the seed rules: [SPT rule code]. The PAI-LLM interface is called, and after passing in the prompt words, 6 new rules are generated. The 6 new rules are merged with the 4 seed rules to build an initial rule population with 10 individuals. Step 3: Determine the maximum completion time of individuals in the initial rule population. Calculate the fitness value of each individual based on the maximum completion time. Based on the fitness value, evaluate the performance of the individuals using an evaluator, and use roulette wheel selection to select multiple elite individuals and their core decision-making characteristics. Build a simulated production environment: Set up 20 jobs, each with 5 processes, and perform assembly line scheduling. Traverse the 10 rules in the initial rule population, and use the PAI-LLM evaluator (evaluate_population function) to perform batch performance evaluation on all rules in the initial rule population. Use the maximum completion time as the core fitness index, which is the longest time all jobs take to complete processing. Quantify the scheduling effect of each rule using the fitness index, and select rules that meet the performance standards. Call the evaluate_population function to simulate the scheduling process, recording the maximum completion time (Makespan) of each rule. The results are as follows: Individual 1 / 10: Effective | Makespan = 1545 Individuals 2 / 10: Effective | Makespan = 2149 Individuals 3 / 10: Effective | Makespan = 2440 Individual 4 / 10: Effective | Makespan = 3087 Individual 5 / 10: Effective | Makespan = 1545 Individual 6 / 10: Effective | Makespan = 2149 Individual 7 / 10: Effective | Makespan = 2440 Individual 8 / 10: Effective | Makespan = 3087 Individual 9 / 10: Effective | Makespan = 1545 Individual 10 / 10: Effective | Makespan = 2149 Calculate the fitness value of an individual in the population; its fitness value is: , in, For fitness value, To determine the maximum completion time for individuals in the population, a roulette wheel selection method is used to select the top 5 high-fitness rules from 10 rules as elite individuals. Step 4: Based on the core decision-making characteristics of elite individuals, use the mutation operator of PAI-LLM to perform rule mutation on multiple elite individuals, obtaining multiple mutation rules and their core decision-making characteristics. The mutation rules are heuristic code segments generated autonomously by PAI-LLM. Taking rule A as an example, rule A has a fitness value of 0.00118. The mutation prompt is constructed as follows: the maximum completion time of rule A is 1545, which results in a long job waiting time in machine failure scenarios, requiring optimization. The core logic of "considering remaining processing time" is retained, and the logic of "prioritizing the allocation of jobs to alternative machines for failed machines" is added. The mutation prompt is then input into PAI-LLM, and the PAI-LLM mutation operator is called. The heterogeneous operator generates mutation rules; the mutation rules are evaluated using fitness metrics. After evaluation, the four effective mutation rules with the largest reduction in completion time are retained, and the core decision features of the four effective mutation rules are extracted. These core decision features include considering the remaining processing time, combining machine load, and prioritizing the scheduling of emergency orders. The core decision features are summarized as effective decision factors. The stage3_group_aggregation of PAI_LLMEvolution is called to extract the core decision features of the mutation rules. The core decision features of the mutation rules are organized into structured group experience and stored in a JSON-formatted experience repository to provide knowledge support for the next round of iterative evolution. Step 5: Repeat steps 2 to 4 until the preset iteration threshold is reached. Stop the iteration and output the offline optimal scheduling rule and its core decision features. The offline optimal scheduling rule is the scheduling rule with the minimum maximum completion time. The preset number of iterations is set. If the number of iterations does not reach the preset threshold, return to step 2 and continue to expand and evolve the population based on the updated group experience. If the iteration threshold is reached, output the optimal rule in the current population. The offline rule evolution stage ends, and the scheduling rule with the minimum maximum completion time, i.e., the offline optimal scheduling rule, is output.

[0025] PAI-LLM is used for the intelligent evolution and real-time application of scheduling rules. It comprises three main sub-components: a population generator, an evaluator, and a mutation operator. Driven by PAI-LLM, it enables the automatic generation, performance evaluation, and iterative optimization of scheduling rules without manual intervention in rule design. The evolutionary capability of the PAI-LLM method for heuristic algorithms is achieved through a unique individual encoding mechanism. The individuals generated by PAI-LLM are directly executable heuristic code segments. Its core objective is to guide the task selection process in DJSSP, that is, to determine "which job to assign to which machine under the current production state" through code logic, rather than directly outputting a complete scheduling plan (such as the process time schedule for the entire job). This design adapts to the dynamic nature of DJSSP: when changes occur in the production scenario, such as machine failures or the arrival of new jobs, the heuristic code segments can re-evaluate task selection based on real-time data without reconstructing the entire scheduling process. The plan aims to significantly improve response efficiency. Furthermore, the heuristic code segments are generated autonomously by PAI-LLM. The generation process has no predefined code length constraints or function set limitations. The code generated by PAI-LLM only needs to comply with three basic specifications: First, the function name is uniformly heuristic_function (to adapt to the calling logic of the subsequent online scheduling and execution module); second, the input parameters include job_data (basic job information, including the processing time of each process and the processing machine route), machine_avail (a list of available machine times), and current_time (current production time); third, the output parameter is the "selected job ID" (to ensure the uniqueness of the scheduling decision). For example, the heuristic code segment generated by PAI-LLM that "prioritizes the job with the shortest remaining processing time" can be directly integrated into the population through the three basic specifications without additional adaptation or modification.

[0026] In some embodiments, in step S102, the machine status and the list of jobs to be processed in the dynamic workshop are acquired in real time. The machine status and processing jobs in the dynamic workshop are collected in real time through dynamic environment perception. The processing jobs are filtered to obtain the list of jobs to be processed. The `get_available_jobs` function of the dynamic environment perception is used to filter jobs whose arrival time is less than or equal to the current time and whose processes are not yet complete, thus generating the list of jobs to be processed. Simultaneously, the `update_machine_status` function is called to update the machine status in real time (e.g., recovery of faulty machines, load changes of normal machines). The dynamic environment perception calls the `get_available_jobs` function every minute. For example, if the current time is 60 minutes, the `get_available_jobs` function filters out the list of jobs whose arrival time is less than or equal to 60 minutes. The `update_machine_status` function is called to obtain the machine status via the industrial bus: Machine 2 is faulty (fault duration 50-70 minutes), Machines 0-1 and 3-4 are normal, and the machine availability time is as follows: ; A job allocation scheme is generated based on offline optimal scheduling rules, machine status, and a list of jobs to be processed. Based on this scheme, the jobs to be processed in the dynamic job shop are scheduled in real time. The steps are as follows: Step 1: Based on the machine status and the list of jobs to be processed, construct the core parameters of the dynamic job shop at the current moment. The core parameters include job_data (job information, including the processing time of each process and the processing machine route), machine_avail (machine available time), and current_time (current production time). Step 2: Generate a job allocation scheme based on core parameters and offline optimal scheduling rules. Based on the job allocation scheme, determine the jobs to be processed in the next time step of the job list and their corresponding key scheduling information to generate a scheduling event. Load the offline optimal scheduling rules and, according to the core parameters, call the heuristic function of the offline optimal scheduling rules to generate a job allocation scheme. The job allocation scheme records the jobs to be processed in the next time step of the job list and their key scheduling information, including job ID, assigned machine ID, start time (the maximum of the current time and the machine's available time), and end time (the sum of the start time and the processing time of the current operation of the job). After determining the jobs to be processed in the next time step and the key scheduling information, generate a scheduling event. For example, if the offline optimal scheduling rules are called, select job 5 (job 5 has a short current operation processing time and the assigned machine 1 has an early available time), and calculate the scheduling information for job 5: Machine allocation: According to the processing machine route of Operation 5, the current process needs to be processed on machine 1; Start time is 60:max (current time is 60, machine 1 available time is 55); The end time is 70: the sum of the current time 60 and the current processing time of task 5, which is 10. Generate a ScheduleEvent: {"job_id": 5, "machine_id": 1, "start_time": 60, "end_time": 70}; Update status: Machine 1 has 70 hours of available time, and the processing machine route for Operation 5 is Machine 1; Step 3: Based on the scheduling event, synchronize the production status of the dynamic workshop and update the list of jobs to be processed and the machine status. After generating the scheduling event, synchronize the production status of the dynamic workshop, update the available time of the allocated machine to the job completion time, increment the number of processed jobs by 1, and synchronize the generated scheduling event to the production management system so that managers can view the scheduling progress in real time. At the same time, update the job and machine status data in the dynamic environment perception to provide the latest data for the next scheduling decision, that is, update the list of jobs to be processed and the machine status. Step 4: Obtain the number of processed jobs based on the updated list of jobs to be processed. If the number of processed jobs has not reached the preset feedback interval, repeat steps 1 to 3. If the number of processed jobs reaches the preset feedback interval, execute step 5. The preset feedback interval is 20. Count the number of processed jobs in the list of jobs to be processed. If the number of processed jobs is 15 and the preset feedback interval has not been reached, return to step 1 and continue to execute the next scheduling. If the number of processed jobs is 20 and the preset feedback interval has been reached, execute step 5. Step 5: Obtain the scheduling results of the scheduling events. After feeding back the scheduling results, repeat steps 1 to 3 until all pending jobs in the pending job list are completed. Statistically analyze the scheduling results of all scheduling events in the current scheduling cycle. These results include maximum completion time, average equipment utilization, and job delay rate. For a Gantt chart of the scheduling results, please refer to [link / reference]. Figure 2 ,like Figure 2 As shown, when the current scheduling cycle is 20, the maximum completion time (Makespan) is 1448, the average equipment utilization rate is 82%, and the job delay rate is 0. The scheduling results are compared with the group experience in the experience base to determine the adaptability of the offline optimal scheduling rule to the machine failure scenario. The scheduling results are synchronized to PAI_LLM to provide a basis for the next round of offline evolution. At the same time, return to step 1 to continue scheduling until the scheduling of the pending jobs in the pending job list is completed.

[0027] Based on dynamic environment perception, the current production status is obtained, the offline optimal scheduling rules obtained from offline evolution are invoked, and job allocation schemes are generated in real time. Successful experiences (such as decision characteristics of high-performance rules) and failure cases in the scheduling process are recorded to form a structured group experience library, which is used to guide the next round of rule evolution of PAI-LLM, realizing knowledge accumulation and reuse.

[0028] In summary, the dynamic job shop scheduling optimization method provided by this invention obtains key parameters and classic scheduling rules for the dynamic job shop scheduling scenario, and uses a large language model to perform offline evolution of the classic scheduling rules to obtain the offline optimal scheduling rules; it also obtains the machine status and the list of jobs to be processed in the dynamic job shop in real time, generates a job allocation scheme based on the offline optimal scheduling rules, machine status, and the list of jobs to be processed, and performs real-time scheduling of the jobs to be processed in the dynamic job shop based on the job allocation scheme, thus meeting the needs of flexible manufacturing for rapid response and precise scheduling.

[0029] To better implement the dynamic job shop scheduling optimization method in this embodiment of the invention, based on the dynamic job shop scheduling optimization method, correspondingly, as follows: Figure 3 As shown, this embodiment of the invention also provides a dynamic job shop scheduling optimization device, the dynamic job shop scheduling optimization device 300 including: The scheduling rule evolution module 301 is used to obtain key parameters and classic scheduling rules for dynamic job shop scheduling scenarios. It uses a large language model to perform offline evolution of the classic scheduling rules to obtain the offline optimal scheduling rules. The job scheduling module 302 is used to obtain the machine status and the list of jobs to be processed in the dynamic job workshop in real time, generate a job allocation scheme based on the offline optimal scheduling rules, machine status and the list of jobs to be processed, and perform real-time scheduling of the jobs to be processed in the dynamic job workshop based on the job allocation scheme.

[0030] like Figure 4 As shown, the present invention also provides an electronic device 400, which can be a mobile terminal, desktop computer, laptop, handheld computer, server, or other computing device. The electronic device 400 includes a processor 401, a memory 402, and a display 403. Figure 4 Only some components of the electronic device 400 are shown, but it should be understood that it is not required to implement all the components shown, and more or fewer components may be implemented instead.

[0031] In some embodiments, memory 402 may be an internal storage unit of the electronic device 400, such as a hard disk or memory of the electronic device 400. In other embodiments, memory 402 may be an external storage device of the electronic device 400, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the electronic device 400. Furthermore, memory 402 may include both internal and external storage units of the electronic device 400. Memory 402 is used to store application software and various types of data installed on the electronic device 400, such as program code installed on the electronic device 400. Memory 402 may also be used to temporarily store data that has been output or will be output. In one embodiment, memory 402 stores a dynamic job shop scheduling optimization program, which can be executed by processor 401 to implement the dynamic job shop scheduling optimization method of various embodiments of the present invention.

[0032] In some embodiments, processor 401 may be a central processing unit (CPU), microprocessor or other data processing chip, used to run program code stored in memory 402 or process data, such as dynamic job shop scheduling optimization methods.

[0033] In some embodiments, display 403 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. Display 403 is used to display identification information of the dynamic job shop scheduling optimization program and to display a visual user interface. Components 401-403 of electronic device 400 communicate with each other via a system bus.

[0034] In some embodiments, when the processor 401 executes the dynamic job shop scheduling optimization program in the memory 402, it implements each step of the dynamic job shop scheduling optimization method as described in the above embodiments. Since the dynamic job shop scheduling optimization method has been described in detail above, it will not be repeated here.

[0035] Accordingly, the present invention also provides a computer-readable storage medium for storing a computer-readable program or instruction, which, when executed by a processor, can implement the steps or functions of the dynamic job shop scheduling optimization method provided in the above-described method embodiments.

[0036] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0037] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A dynamic job shop scheduling optimization method, characterized in that, include: The key parameters and classic scheduling rules of the dynamic job shop scheduling scenario are obtained, and the classic scheduling rules are evolved offline using a large language model to obtain the offline optimal scheduling rule. The machine status and the list of pending jobs in the dynamic workshop are obtained in real time. Based on the offline optimal scheduling rules, the machine status and the list of pending jobs, a job allocation scheme is generated. Based on the job allocation scheme, the pending jobs in the dynamic workshop are scheduled in real time.

2. The dynamic job shop scheduling optimization method according to claim 1, characterized in that, The key parameters include operational data, number of machines, and production constraints.

3. The dynamic job shop scheduling optimization method according to claim 2, characterized in that, The large language model includes PAI-LLM, which includes a population generator, an evaluator, and a mutation operator; the offline evolution of the classical scheduling rule using the large language model to obtain the offline optimal scheduling rule includes: Step 1: Use the classic scheduling rule as the initial seed population; Step 2: Construct prompt words based on the key parameters. Using the initial seed population as a template, input the prompt words into PAI-LLM to generate multiple scheduling rules. Then, use the population generator to merge the multiple scheduling rules with the classic scheduling rules to obtain the initial rule population. Step 3: Determine the maximum completion time of individuals in the initial rule population, calculate the fitness value of individuals based on the maximum completion time, evaluate the performance of individuals based on the fitness value using the evaluator, and use roulette wheel selection to select individuals to obtain multiple elite individuals and the core decision characteristics of elite individuals. Step 4: Based on the core decision-making characteristics of the elite individuals, use the mutation operator of PAI-LLM to perform rule mutation on the multiple elite individuals to obtain multiple mutation rules and the core decision-making characteristics of the multiple mutation rules. The mutation rules are heuristic code segments generated autonomously by PAI-LLM. Step 5: Repeat steps 2 to 4 until the preset iteration threshold is reached, then stop the iteration and output the offline optimal scheduling rule and the core decision features of the offline optimal scheduling rule. The offline optimal scheduling rule is the scheduling rule with the minimum maximum completion time.

4. The dynamic job shop scheduling optimization method according to claim 3, characterized in that, The classic scheduling rules include the first-in-first-out rule, the shortest processing time priority rule, the longest processing time priority rule, and the most remaining work priority rule.

5. The dynamic job shop scheduling optimization method according to claim 3, characterized in that, The core decision-making features include considering the remaining processing time, combining machine load, and prioritizing the scheduling of emergency orders.

6. The dynamic job shop scheduling optimization method according to claim 3, characterized in that, The process of generating a job allocation scheme based on the offline optimal scheduling rules, the machine status, and the list of jobs to be processed, and then performing real-time scheduling of the jobs to be processed in the dynamic job workshop based on the job allocation scheme, includes: Step 1: Construct the core parameters of the dynamic workshop at the current moment based on the machine status and the list of jobs to be processed; Step 2: Generate a job allocation scheme based on the core parameters and offline optimal scheduling rules; determine the jobs to be processed in the next time step and the key scheduling information of the jobs to be processed in the next time step based on the job allocation scheme, so as to generate a scheduling event. Step 3: Based on the scheduling event, synchronize and update the production status of the dynamic workshop, and update the list of jobs to be processed and the machine status; Step 4: Obtain the number of processed jobs based on the updated list of jobs to be processed. If the number of processed jobs has not reached the preset feedback interval, repeat steps 1 to 3. If the number of processed jobs has reached the preset feedback interval, execute step 5. Step 5: Obtain the scheduling result of the scheduling event, and after feeding back the scheduling result, repeat steps 1 to 3 until the scheduling of the pending jobs in the pending job list is completed.

7. The dynamic job shop scheduling optimization method according to claim 6, characterized in that, The key scheduling information includes job ID, assigned machine ID, start time, and end time.

8. A dynamic job shop scheduling optimization device, characterized in that, include: The scheduling rule evolution module is used to obtain key parameters and classic scheduling rules for dynamic job shop scheduling scenarios. Based on the key parameters and classic scheduling rules, a large language model is used to perform offline evolution on the classic scheduling rules to obtain the offline optimal scheduling rules. The job scheduling module is used to obtain the machine status and the list of jobs to be processed in the dynamic job workshop in real time, generate a job allocation scheme based on the offline optimal scheduling rules, the machine status and the list of jobs to be processed, and perform real-time scheduling of the jobs to be processed in the dynamic job workshop based on the job allocation scheme.

9. An electronic device, characterized in that, Including memory and processor; The memory stores a computer-readable program that can be executed by the processor; When the processor executes the computer-readable program, it implements the steps in the dynamic job shop scheduling optimization method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, Used to store computer-readable programs or instructions, which, when executed by a processor, can implement the steps in the dynamic job shop scheduling optimization method according to any one of claims 1-7.

Images