A multi-agent multi-objective optimization workshop scheduling method based on human-computer cooperation

Abstract

The application discloses a kind of multi-agent multi-objective optimization workshop scheduling method based on man-machine cooperation.First, three objective functions including process sequencing, machine selection, speed selection and worker selection are constructed to optimize maximum completion time, total energy consumption and machine load balancing.Second, a distributed constraint processing mechanism based on local resource agent (LRA) is designed, and each LRA autonomously decides speed and worker allocation based on local resource state to ensure the feasibility of the scheduling scheme.Finally, a hierarchical multi-agent evolutionary algorithm is used to evolve collaboratively between global planning layer and field execution layer, and efficiently search for the Pareto optimal solution set through distributed evaluation, environment selection and adaptive policy adjustment.The application overcomes the problems of single resource constraint, rigid processing model and limited optimization objectives in traditional scheduling, achieving multi-objective collaborative optimization under double resource and multi-speed constraints, and significantly improving the greenness, balance and efficiency of production scheduling.

Description

Technical Field

[0001] This invention relates to a production scheduling optimization method for flexible workshops, specifically a multi-agent, multi-objective optimization workshop scheduling method based on human-machine collaboration, belonging to the field of intelligent manufacturing and industrial automation technology. Background Technology

[0002] In modern manufacturing, the manufacturing workshops for core automotive components (such as engine blocks and transmissions) are typical flexible job workshops (FJSPs), and their production scheduling is key to ensuring efficiency and reducing costs.

[0003] However, existing workshop scheduling methods suffer from several major drawbacks when applied to such scenarios: First, resource constraints are incomplete. Traditional scheduling models often only consider machine resources, but in actual production, many high-precision or special processes (such as finishing and special assembly) must be completed jointly by "machines" and "workers," constituting a "machine-worker" dual-resource constraint (DRC). Furthermore, workers often possess different skills, and specific machines can only be operated by specific workers; this constraint is often overlooked in existing research. Second, the machining model is rigid. Existing research often assumes that the machining time of a process is fixed, while modern CNC machine tools generally have multi-speed or multi-speed machining capabilities. Choosing a high speed can shorten machining time but will drastically increase machining power; choosing a low speed has the opposite effect. This inherent trade-off between "time-energy consumption-machine load balancing" is the core decision point for scheduling but has not been fully studied. Finally, the optimization objective is singular. Most scheduling methods only aim to minimize the maximum completion time (Makespan). Enterprises urgently need to optimize production efficiency (completion time), green metrics (total energy consumption), and machine load balancing at the same time. This is a multi-objective optimization problem.

[0004] Therefore, a novel scheduling method is needed that can accurately model the above-mentioned complex constraints and achieve the optimization of the three objectives of "completion time - total energy consumption - machine load balancing" through a layered distributed architecture. Summary of the Invention

[0005] To address the shortcomings of existing flexible shop floor scheduling methods in accurately modeling the dual resource constraints of "machine-worker" and the "multi-speed" characteristics of machines, and the difficulty of centralized scheduling in simultaneously considering global multi-objective optimization, this invention proposes a multi-agent, multi-objective optimization shop floor scheduling method based on human-machine collaboration, comprising: Obtain the process sequence of the workpiece to be processed, the set of optional processing machines for each process, the multiple optional processing speed gears supported by each machine for each process and their corresponding processing time and processing power data, as well as the set of optional operator options for each machine. Construct a three-objective function, which is the maximum completion time, total energy consumption, and machine load balance. A hierarchical distributed multi-agent evolutionary algorithm is adopted, aiming to minimize a three-objective function under constraints. The three-objective function is solved, and the hierarchical distributed multi-agent algorithm includes a global planning agent (GPA) and multiple field resource agents (LRA). The hierarchical distributed multi-agent evolutionary algorithm includes the following steps: An initial population of size N is formed, where each individual is encoded by four gene segments: The process sequence, arranged in ascending order, represents the first process of workpiece i. Steps The global temporal sorting sequence; The machine selects a sequence, indicating the process. Assigned to one of its optional processing machines ; Speed ​​selection sequence indicates the speed selection sequence for the selected machine. The above are the corresponding processes. Specify an optional machining speed level ; Worker selection sequence, represented as process. In the assigned machine Specify a worker w from the set of optional operators; In each generation of evolution: the Global Planning Agent (GPA) generates a macro-level solution, which includes a process sequencing sequence and a machine selection sequence associated with the workpiece; each Field Resource Agent (LRA) is bound to at least one machine and a corresponding set of selectable operators, which is used to receive the macro-level solution from the GPA, and based on the resource schedule of the machine bound to the LRA, to generate a speed selection sequence and an operator selection sequence for the scheduled processes to be assigned to that machine, and calculate the three-objective function value of each generation of individuals, which is then fed back to the GPA. After the global planning agent GPA collects feedback, it performs fast non-dominated sorting on the mixed population after merging the parent and offspring generations, and calculates the crowding distance of individuals within each non-dominated front. According to the principle of Pareto priority and priority of individuals with large crowding distance at the same level, N high-quality individuals are selected to form the new generation population. The process continues iterating until the preset maximum number of evolutions is reached. All non-dominated solutions with a Pareto front level of 1 are then extracted and output as the final optimized solution. The resource schedule records the start and completion times of the scheduled processes of the machines bound to the field resource agent (LRA). The scheduled processes are the associated processes of the workpieces to be processed.

[0006] Furthermore, when the field resource agent (LRA) generates the speed selection sequence and worker selection sequence for the scheduled process, it does so based on at least one of the following local strategies: (1) Greedy energy-saving strategy: Based on the energy consumption characteristic curve of the process on the optional processing machine, select the optional processing speed range that minimizes the energy consumption product of a single process; (2) Efficiency balancing strategy: Guided by the reduction of processing cycle, prioritize matching the most selectable processing speed range and the earliest available selectable operator option.

[0007] Furthermore, in the hierarchical distributed multi-agent evolutionary algorithm, the global planning agent (GPA) generates a population through tournament selection, crossover, and mutation operations. After receiving the process ordering sequence and the machine selection sequence associated with the workpiece from the global planning agent (GPA), each field resource agent (LRA) dynamically adjusts its local strategy based on the Metropolis criterion to accept the probability of the newly generated combination of speed selection sequence and worker selection sequence. In the early stage of evolution, the field resource agents (LRA) are allowed to boldly try diverse combinations of speed selection sequence and worker selection sequence. In the later stage of evolution, only non-inferior or better solutions are accepted, which promotes the convergence of the algorithm.

[0008] Furthermore, when multiple field resource agents (LRAs) compete for the same machine or worker, an arbitration is performed using a timing locking protocol based on global priority. The global priority of a process is determined by the process sequencing sequence issued by the Global Planning Agent (GPA). Processes with higher priority have priority to obtain the right to lock the time window of the machine or worker, and update the next available timestamp of the worker or machine. (), When a low-priority field resource agent (LRA) performs a simulation, it reads the updated next available timestamp for the worker or machine and determines its availability based on the formula. Automatically calculates the start time of its processes; In the formula, Indicates the completion time of the preceding process for the same workpiece; and They are machines and workers The next available time.

[0009] Furthermore, the formula for calculating the total energy consumption is as follows: In the formula, It is a process Select machine k and set the time required for speed level v. It is a process Select machine k and set the power required for the speed level v; It is the idle power of machine k when it is turned on but not processing. Let k be the idle time of machine k.

[0010] Furthermore, the formula for calculating the machine load balancing degree is as follows: In the formula, K represents the total number of machines. It is the total busy time of machine k. It is the average busy time of all machines.

[0011] Furthermore, the constraints include: (1) Each process of the workpiece must be processed in sequence according to its preset process order; (2) At any given time, a machine can process at most one operation; (3) At any given time, a worker may operate at most one machine; (4) The workers assigned to the process must be from the set of available operator options for the selected machine.

[0012] Furthermore, the initial population is obtained by independently and randomly generating the above four gene segments within their respective legal value ranges; if the generated rotation speed selection sequence or worker selection sequence exceeds the actual configuration of the machine specified by the current machine selection sequence, it will be corrected through a local repair mechanism in the subsequent field resource agent LRA processing stage.

[0013] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the methods described herein.

[0014] Compared with the prior art, the beneficial effects of the present invention are: This invention establishes a more realistic dual-resource collaborative scheduling model by introducing worker selection variables and skill constraints, thereby improving the feasibility and practicality of the scheduling scheme.

[0015] This invention incorporates the processing time-energy consumption trade-off into scheduling decisions by selecting speed as a variable, and uses total energy consumption and machine load balancing as optimization objectives to support enterprises in multi-objective collaborative optimization between efficiency and green manufacturing.

[0016] This invention proposes an adaptive evolutionary algorithm that integrates distributed evaluation, non-dominated sorting, and the Metropolis criterion, enabling efficient searching of the Pareto front in complex solution spaces. The algorithm dynamically adjusts local strategies using temperature parameters, balancing search diversity and convergence speed, and significantly improving multi-objective optimization performance. Attached Figure Description

[0017] Figure 1 Flowchart of this invention; Figure 2 Initialization scheme Gantt chart; Figure 3 Gantt chart of the optimized algorithm. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0019] Example This embodiment uses a core automotive component manufacturing workshop as an application scenario. First, it acquires information about the workpieces, machines, and engineers in the workshop, which serves as the basis for subsequent scheduling modeling and algorithm execution. Specifically: Machine information: Includes the set of available operator options for each machine (to implement worker skill constraints) and idle power (to calculate idle energy consumption).

[0020] Workpiece information: Each workpiece corresponds to multiple processes, and each process is marked as "whether it is a special process" (determining whether workers need to be assigned) and associated with multiple optional processing machines.

[0021] Process parameters: Each selectable machining machine is further associated with multiple selectable machining speed ranges. Each selectable machining speed range is associated with the machining time and machining power of the process when it is performed on that machine at that speed, thereby supporting multi-speed scheduling and energy consumption calculation. Specific data are shown in Tables 1 to 3.

[0022] Table 1. List of machines available to workers Table 2. List of machines available for each process. Table 3 Machine-Speed ​​Selection Table for Processes The scheduling method of the present invention includes the following steps: Step 1: Establish a three-objective scheduling model.

[0023] This invention models the scheduling problem as a three-objective optimization problem. A complete scheduling scheme X consists of four sets of decision sequences: the operation ordering sequence (OS), which is a set of real-number priority codes used to determine the global scheduling priority of all operations for all jobs; the machine selection sequence (MS), where each element is... This indicates that the process j of workpiece i (denoted as O) will be performed. ij Assign the rotational speed to machine k; select the rotational speed sequence (VS), where each element in the sequence is... , indicating process O ij When processing on machine k, the rotational speed level v is selected; and the worker selection sequence (WS) is used, where each element of the sequence is... , indicating when When it is a special process, it is assigned to worker w.

[0024] In this invention, the only constraint in the actual scenario is the decision space, which has a huge solution space: In the formula, Number of processes; Let i be the number of machines that can be selected for process i; The number of worker options for special process i; The speed option for process i is designed to simultaneously optimize the following conflicting objective functions F(X): It is the factorial symbol; The first objective is to minimize the maximum completion time. ): This refers to the time taken to complete the final step in a production batch. Suppose there are j workpieces to be produced. Workpieces 1 through j-1 are completed within one hour, but workpiece j requires 10 hours. Therefore, the entire batch of work in the workshop is not finished within one hour, but rather delayed until 10 hours. Definition: That is, the maximum value among the completion times of all processes for all workpieces is taken, representing the final completion time of one process for all workpieces; In the formula, For workpiece process The completion time.

[0025] The goal is to pinpoint the final step in the entire production batch and bring it forward as early as possible. Minimizing this value means shortening the overall production cycle and improving production efficiency.

[0026] The second objective is to minimize total energy consumption. ): in, Total processing energy consumption is the sum of the products of processing time and processing power for all processes. It is a process The time taken when machine k is selected and speed level v is set. It is a process Select machine k and set the power required for speed level v; = The total idle power consumption is the sum of the products of the cumulative idle time and standby idle power of all machines. It is the idle power of machine k when it is turned on but not processing. This refers to free time.

[0027] The third objective is to minimize machine load balancing. ) In the formula, K represents the total number of machines. It is the total busy time of machine k. It is the average busy time of all machines. The goal is to minimize the standard deviation of the busy time of all machines. The smaller the standard deviation, the more evenly the load is distributed among the machines.

[0028] All scheduling schemes must satisfy the core constraints (inequality constraints) including: process constraints (the internal processes of a workpiece must be processed sequentially). ; This indicates the completion time of the preceding operations for the same workpiece, and machine resource constraints (a machine can only process one operation at a time). ; Indicates machine The completion time of the previous process; resource constraints (one worker can only operate one machine at a time). ; Indicates workers The completion time of the previous operation; skill constraints (the assigned worker w must be in the set of available operators for the selected machine k); and (equality constraints) the definition of completion time: the completion time of the j-th operation of workpiece i is equal to its start time plus the actual processing time. ,in Let be the start time of the j-th process for workpiece i. The time taken to process the material at revolutions v on machine k.

[0029] Step 2: Design a distributed constraint processing mechanism based on Local Intelligent Agents (LRAs).

[0030] This invention does not employ the traditional penalty function method, but instead decodes decision variables through the local computational logic of the field resource agent (LRA) and automatically satisfies all constraints.

[0031] Each machine is configured with an LRA to maintain its own "Resource Schedule" (which records the start and finish times of the j-th operation for workpiece i, representing the machine's...). and workers The next available time), the upper-level Global Planning Agent (GPA) generates a macroscopic process sequencing sequence (OS) and machine selection sequence (MS). The LRA receives the scheduleable processes (determined by the machine selection sequence MS) and priorities (determined by the process sequencing sequence OS) allocated to its machine by the GPA (Global Planning Agent), and autonomously generates its own speed selection sequence (VS) and worker selection sequence (WS) based on its local policy. Scheduled processes are associated processes of the workpiece to be processed. The LRA uses autonomous deduction logic, i.e., it iterates through the scheduleable processes allocated to its machine. For each process... The machine is only "unlocked" and allowed to be used after the current process is completed. The LRA selects the speed according to local policies. and workers And strictly calculate the actual start time of the j-th process of workpiece i. : in, and They are machines and workers The next available time, the completion time of the j-th operation of workpiece i. It is calculated and updated in the local machine's resource schedule.

[0032] This distributed mechanism ensures that any global instruction can eventually be decoded at the field level into a feasible scheduling scheme that fully satisfies all constraints.

[0033] The aforementioned local strategies specifically include: (1) a greedy energy-saving strategy, in which the LRA selects the low-energy-consumption optional processing speed range that minimizes the single-process energy consumption product (power * time) based on the machine energy consumption characteristic curve when deciding on the speed selection sequence (VS); and (2) an efficiency balancing strategy, in which the LRA prioritizes matching high-speed speed ranges and the earliest available optional operator options to eliminate equipment idle time windows, guided by the goal of compressing the processing cycle. During the algorithm's operation, the LRA does not consistently execute a particular strategy, but rather adaptively activates one of the above two strategies at different process nodes through the genetic encoding of the evolutionary algorithm.

[0034] Step 3: Construct a hierarchical adaptive evolution algorithm.

[0035] To efficiently search complex three-objective solution spaces, this invention proposes a hierarchical distributed multi-agent evolutionary algorithm. This algorithm adaptively balances "macro-planning" and "micro-execution" through the interaction between the global layer and the field layer.

[0036] Specifically, it includes the following sub-steps: The GPA is initialized with a population of size N (N=100 in this example), where N is a positive integer; each individual consists of four gene segments, namely: The process order sequence (OS) has a length equal to the total number of processes, and each element is a real number in the range [0,1]. The process order is arranged in ascending order to represent the global priority order of all processes. The machine selection sequence (MS) is an integer of length equal to the total number of operations, ranging from [1, the number of available machines]. The index in the sequence indicates which machine in the set of available machines will be assigned to the corresponding operation. Speed ​​selection sequence (VS), with a length equal to the total number of processes, in the range of [1, number of selectable speed levels], indicates that a selectable machining speed level is specified for the corresponding process on the selected machine; Worker Selection Sequence (WS), an integer of length equal to the number of special operations, in the range of [1, number of available workers], represents the selection of an operator from the set of available operator options for the assigned machine for a special operation.

[0037] The initial population is obtained by independently and randomly generating the above four gene segments within their respective legal value ranges; if the generated VS or WS exceeds the actual configuration of the machine specified by the current MS (such as the selected speed gear or the worker is not within the range supported by the machine), it will be corrected by the local repair mechanism in the subsequent LRA processing stage; (for example, "speed gear 3" is randomly selected, but this process only has 2 gears on the currently selected "machine A").

[0038] In each generation of evolution: (1) Global planning: GPA generates N macro-plans through tournament selection, crossover and mutation operations. Each macro-plan is a process ordering sequence (OS) and a machine selection sequence (MS).

[0039] (2) Distributed Collaborative Evaluation: Each OS / MS scheme is distributed to the machine-level LRA. If multiple LRAs simultaneously apply for the same worker or machine, a time-series locking protocol based on global priority is adopted. When multiple LRAs compete for the same shared resource (such as a worker with a specific skill), the arbitration right is determined by the process ordering variable (OS) issued by the GPA. The LRA corresponding to the process with higher priority has priority to obtain the time window locking right for the worker or machine and updates the 'next available timestamp' of that worker or machine. When calculating the LRA for lower priority processes, the updated next available timestamp for that worker or machine is read, and the calculation is performed according to the formula. The necessary waiting time is automatically calculated. This mechanism ensures that physical resource constraints are strictly satisfied at any time during distributed execution, and deadlocks will not occur. Based on the current local strategy (such as greedy energy saving or efficiency balancing), LRA autonomously generates the speed selection sequence VS and the worker selection sequence WS, and calculates the actual values ​​through simulation deduction using the logic described in step 2. , , This feedback is given to GPA.

[0040] (3) Environmental selection: After GPA collects feedback, rapid non-dominated sorting and crowding distance calculation are performed on the mixed population. According to the principle of Pareto priority and priority of individuals with large crowding distance under the same level, N high-quality individuals are selected to form the new generation population.

[0041] Domination means that if all the objective solutions of a feasible individual are not inferior to all the objective solutions of another individual, and the feasible individual is superior to the other individual in at least one objective, then the individual dominates the other individual. Non-dominance means that if there is no situation where one individual dominates another individual, and one individual is inferior to another individual in at least one objective and the other individual is inferior to one individual in at least one objective, then the relationship between the two individuals is said to be non-dominant. (4) Adaptive adjustment: The Metropolis criterion mechanism is introduced, and the current temperature T is used as the control parameter to dynamically adjust the LRA’s acceptance probability of the newly generated VS / WS allocation scheme (i.e. its local strategy).

[0042] Specifically, when the LRA generates a new VS / WS combination based on the current strategy (such as trying an aggressive high-speed strategy VS or a random worker matching strategy WS), and generates a specific scheduling scheme (individual), i.e., the new scheme (Intruder), the system compares it with a randomly selected existing better scheme (called Victim) from the current population. That is, with the introduction of the Metropolis criterion: the system will tolerate this bold attempt made by the LRA (even if it currently seems poor) with a certain probability, keeping it in the population. This is equivalent to accepting the LRA's local exploration strategy to prevent getting trapped in local optima.

[0043] When a new scheme generated by an LRA (denoted as Intruder) is determined to be inferior to a better scheme in the current population (denoted as Victim) in a non-dominated ranking, the difference in their Pareto ranks is calculated: , The system uses the current annealing temperature T as the control parameter and proceeds according to probability. : Decide whether to accept this new proposal and replace the Victim proposal.

[0044] Among them, the annealing temperature T decreases with the number of generations: in the early stage of evolution, T is larger and P is higher, allowing LRA to boldly try diverse VS / WS combinations; in the later stage of evolution, T approaches 0 and P approaches 0, only accepting non-inferior or better solutions, thus promoting the convergence of the algorithm.

[0045] Step 4: Algorithm termination and result output.

[0046] The algorithm continues to run until it reaches the preset maximum number of evaluations (set to 50,000 in this embodiment), at which point it automatically terminates. Finally, the algorithm extracts all non-dominated solutions with a Pareto front level of 1 as the final optimized solution output. Each solution in this set is non-dominated across the three objectives of "maximum completion time," "total energy consumption," and "machine load balancing," thus providing production decision-makers with a series of feasible scheduling options that balance different objectives.

[0047] The algorithm output includes: Pareto front data file, detailed scheduling scheme (including complete decisions for GPA and LRA), and optimal scheduling Gantt chart to support visualization analysis and final decision-making.

[0048] To verify the effectiveness of the method of the present invention, based on the above workshop example data, an initial scheduling scheme and a scheduling scheme optimized by the algorithm were generated and compared and analyzed.

[0049] The execution order of the processes obtained from the initial decoding is shown in Table 4. The corresponding decoding rule is as follows: The obtained numerical encoding is decoded into readable text according to the decoding format list. The real values ​​in the OS vector of the process selection variable are sorted in non-decreasing order; the sorted index sequence is the global priority order for process deployment. Machine selection sequence MS, for process Read the corresponding integer code ,Will As an index, the first machine is directly selected from the set of available machines for that process. A machine is used as a processing device. If If the value exceeds the set limit, it will be automatically corrected to a valid boundary value. For example... corresponding The value is 2. Decoding process: Take... Machine options list: Option 1 is Option 2 is This means selecting the second machine in the list; Speed ​​Selection Sequence (VS), read the corresponding integer code. First, identify the specific machine determined by the MS and obtain the "RPM / Power Range List" supported by that machine. Then, Select the first item from the list as an index. Each RPM gear.

[0050] Worker selection sequence (WS) only when the process Only when marked as a "special process" is the next integer code read sequentially from the WS vector. First, locate the machine determined by MS, query the list of workers qualified to operate that machine (i.e., the skill matrix), and then... Select the first item from the list as an index. One worker.

[0051] Table 4 Initialization Individual Plan Table The three target values ​​corresponding to the initial plan are: Target 1, Maximum completion time. Target 2, Total Energy Consumption Objective 3: Machine load balancing is 28.58.

[0052] The worker allocation for special processes in the initial plan is shown in Table 5, and the scheduling Gantt chart is shown below. Figure 2 As shown.

[0053] Table 5 Initialization Special Process Arrangement Table After optimization using the hierarchical multi-agent evolutionary algorithm described in this invention, an optimized scheduling scheme is obtained, as shown in Table 6. The target value of the optimized scheme is significantly improved: Table 6 Optimized Solution Table The three optimized target values ​​are: Target 1, Maximum Completion Time Target 2, Total Energy Consumption Objective 3: Machine load balancing is 4.91.

[0054] The optimized worker allocation for special processes is shown in Table 7, and the scheduling Gantt chart is as follows. Figure 3 As shown.

[0055] Table 7 Optimized Special Process Arrangement Table Comparing the data before and after optimization, it can be seen that after optimization by the method of the present invention, the maximum completion time is shortened by about 67.9%; the total energy consumption is reduced by about 43.7%; and the machine load balancing index (standard deviation) decreases by about 82.8%. The results show that the method of the present invention achieves significant synergistic optimization effects on the three objectives of "completion time-energy consumption-load balancing".

[0056] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-agent, multi-objective optimization workshop scheduling method based on human-machine collaboration, characterized in that, The method includes: Obtain the process sequence of the workpiece to be processed, the set of optional processing machines for each process, the multiple optional processing speed gears supported by each machine for each process and their corresponding processing time and processing power data, as well as the set of optional operator options for each machine. Construct a three-objective function, which is the maximum completion time, total energy consumption, and machine load balance. A hierarchical distributed multi-agent evolutionary algorithm is adopted, aiming to minimize a three-objective function under constraints. The three-objective function is solved, and the hierarchical distributed multi-agent algorithm includes a global planning agent (GPA) and multiple field resource agents (LRA). The hierarchical distributed multi-agent evolutionary algorithm includes the following steps: An initial population of size N is formed, where each individual is encoded by four gene segments: The process sequence, arranged in ascending order, represents the first process of workpiece i. Steps A global temporal sorted sequence; The machine selects a sequence, indicating the process. Assigned to one of its optional processing machines ; Speed ​​selection sequence indicates the speed selection sequence for the selected machine. The above are the corresponding processes. Specify an optional machining speed level ; Worker selection sequence, represented as process. In the assigned machine Specify a worker w from the set of optional operators; In each generation of evolution: the Global Planning Agent (GPA) generates a macro-level solution, which includes a process sequencing sequence and a machine selection sequence associated with the workpiece; each Field Resource Agent (LRA) is bound to at least one machine and a corresponding set of selectable operators, which is used to receive the macro-level solution from the GPA, and based on the resource schedule of the machine bound to the LRA, to generate a speed selection sequence and an operator selection sequence for the scheduled processes to be assigned to that machine, and calculate the three-objective function value of each generation of individuals, which is then fed back to the GPA. After the global planning agent GPA collects feedback, it performs fast non-dominated sorting on the mixed population after merging the parent and offspring generations, and calculates the crowding distance of individuals within each non-dominated front. According to the principle of Pareto priority and priority of individuals with large crowding distance at the same level, N high-quality individuals are selected to form the new generation population. The process continues iterating until the preset maximum number of evolutions is reached. All non-dominated solutions with a Pareto front level of 1 are then extracted and output as the final optimized solution. The resource schedule records the start and completion times of the scheduled processes of the machines bound to the field resource agent (LRA). The scheduled processes are the associated processes of the workpieces to be processed.

2. The method according to claim 1, characterized in that, When generating speed selection sequences and worker selection sequences for the scheduled processes, the field resource agent (LRA) follows at least one of the following local strategies: (1) Greedy energy-saving strategy: Based on the energy consumption characteristic curve of the process on the optional processing machine, select the optional processing speed range that minimizes the energy consumption product of a single process; (2) Efficiency balancing strategy: Guided by the reduction of processing cycle, prioritize matching the most selectable processing speed range and the earliest available selectable operator option.

3. The method according to claim 1, characterized in that, In the hierarchical distributed multi-agent evolutionary algorithm, the global planning agent (GPA) generates a population through tournament selection, crossover, and mutation operations. After receiving the process ordering sequence and the machine selection sequence associated with the workpiece from the global planning agent (GPA), each field resource agent (LRA) dynamically adjusts its local strategy based on the Metropolis criterion to accept the probability of the newly generated combination of speed selection sequence and worker selection sequence. In the early stage of evolution, the field resource agents (LRA) are allowed to boldly try diverse combinations of speed selection sequence and worker selection sequence. In the later stage of evolution, only non-inferior or better solutions are accepted, which promotes the convergence of the algorithm.

4. The method according to claim 1, characterized in that, When multiple field resource agents (LRAs) compete for the same machine or worker, an arbitration is performed using a timing locking protocol based on global priority. The global priority of a process is determined by the process sequencing sequence issued by the Global Planning Agent (GPA). Processes with higher priority have priority to obtain the right to lock the time window of the machine or worker, and update the next available timestamp of the worker or machine. (), When a low-priority field resource agent (LRA) performs a simulation, it reads the updated next available timestamp for the worker or machine and determines its availability based on the formula. Automatically calculates the start time of its processes; In the formula, Indicates the completion time of the preceding process for the same workpiece; and They are machines and workers The next available time.

5. The method according to claim 1, characterized in that, The formula for calculating the total energy consumption is as follows: In the formula, It is a process Select machine k and set the time required for speed level v. It is a process Select machine k and set the power required for the speed level v; It is the idle power of machine k when it is turned on but not processing. Let k be the idle time of machine k.

6. The method according to claim 1, characterized in that, The formula for calculating the machine load balancing degree is: In the formula, K represents the total number of machines. It is the total busy time of machine k. It is the average busy time of all machines.

7. The method according to claim 1, characterized in that, The constraints include: (1) Each process of the workpiece must be processed in sequence according to its preset process order; (2) At any given time, a machine can process at most one operation; (3) At any given time, a worker may operate at most one machine; (4) The workers assigned to the process must be from the set of available operator options for the selected machine.

8. The method according to claim 1, characterized in that, The initial population is obtained by independently and randomly generating the above four gene segments within their respective legal value ranges; if the generated rotation speed selection sequence or worker selection sequence exceeds the actual configuration of the machine specified by the current machine selection sequence, it will be corrected through a local repair mechanism in the subsequent field resource agent LRA processing stage.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 8.

Images