A Flexible Job Shop Scheduling Method Based on Dynamic Evolutionary Disentanglement Graph and Hierarchical Collaboration Architecture

Abstract

This invention discloses a flexible job shop scheduling method based on a dynamic evolution disjunctive graph and a hierarchical collaborative architecture, belonging to the field of scheduling automation technology. It solves the problem that existing scheduling recovery mechanisms cannot simultaneously consider real-time response speed, cross-scenario generalization ability, and multi-dimensional collaborative decision-making efficiency. First, this invention achieves parallel processing of job process constraints and machine resource competition relationships through built-in operation message attention blocks and machine message attention blocks. Then, it utilizes the adaptive weight allocation mechanism of a graph attention network to accurately extract and fuse deep correlation features of "vertical job sequence" and "horizontal resource allocation." Unlike traditional methods that construct all undirected edges during the initialization phase, leading to computational redundancy, this scheme dynamically adds a necessary set of directed edges at each decision time step, based only on the available operations and machine resources at the current moment. This greatly simplifies the topological complexity of the disjunctive graph while ensuring constraint integrity.

Description

Technical Field

[0001] This invention relates to the field of scheduling automation technology, and in particular to a flexible job shop scheduling method based on dynamic evolutionary disjunctive graphs and a hierarchical collaborative architecture. Background Technology

[0002] As the global manufacturing industry accelerates its transformation towards digitalization, intelligence, and highly customized models, Flexible Job-shop Scheduling Problem (FJSP) has become a core element in achieving efficient production scheduling and control. In modern industrial environments, production workshops not only need to handle complex process constraints but also must respond in real time to highly dynamic emergencies such as machine failures, urgent order insertions, and process changes. However, existing scheduling technologies still have significant limitations when facing large-scale, highly flexible production scenarios.

[0003] First, traditional scheduling algorithms face a significant trade-off between computational efficiency and global performance. While simple heuristic rules are easy to implement, their heavy reliance on expert experience limits their global optimization capabilities, making them ill-suited for complex process combinations. Metaheuristic methods, such as genetic algorithms and particle swarm optimization, while possessing some optimization capabilities, often involve tedious parameter iterations and vast search spaces when dealing with large-scale dynamic scheduling problems, resulting in long computation times and failing to meet the millisecond-level real-time response requirements of industrial settings. Second, the lack of existing modeling strategies and collaborative decision-making mechanisms significantly increases the difficulty of solving the problem. Traditional flexible workshop modeling often employs static disjunctive graph strategies, constructing all constraints during the initialization phase, generating numerous redundant topologies and leading to low model search efficiency. Simultaneously, most existing deep reinforcement learning (DRL) frameworks still use a single-agent architecture, struggling to achieve effective logical coordination when dealing with the coupled decision dimensions of "operation selection" and "machine allocation." This results in weak generalization ability of the model in dynamic environments and excessive sensitivity to hyperparameter fluctuations.

[0004] In summary, the current field of intelligent manufacturing urgently needs a scheduling and recovery mechanism that can balance real-time response speed, cross-scenario generalization ability, and multi-dimensional collaborative decision-making efficiency to achieve efficient allocation of manufacturing resources and intelligent closed-loop control of the production process. Therefore, a flexible job shop scheduling method based on dynamic evolutionary parsing graphs and a hierarchical collaborative architecture is required. Summary of the Invention

[0005] To address the shortcomings of existing scheduling recovery mechanisms, which fail to simultaneously achieve fast real-time response, cross-scenario generalization, and efficient multi-dimensional collaborative decision-making, this invention provides a flexible job shop scheduling method based on a dynamic evolutionary parsing graph and a hierarchical collaborative architecture. This method enables dynamic closed-loop control of the production environment by constructing a highly perceptive digital model and using deep neural network inference. This invention aims to solve the technical bottlenecks of slow response, poor generalization, and difficulties in collaborative decision-making in flexible job shop scheduling (FJSP) under dynamic environments. The specific technical solution is as follows: A flexible job shop scheduling system based on a dynamic evolutionary disjunctive graph and a hierarchical collaborative architecture includes: Multi-dimensional state perception module: Real-time capture of production element data at the bottom of the workshop, extraction of global and local state features in the scheduling process, construction of scheduling state feature set, and transformation of original physical features into high-dimensional state feature vector; Dynamic graph structure modeling module: Constructs a disjunctive graph model for a flexible workshop. Based on a dynamic edge addition strategy, it updates the edge structure of the disjunctive graph. By transforming the non-corresponding edges representing machine selection into directional definite edges and simultaneously removing redundant disjunctive constraints, the original graph structure gradually evolves and solidifies into a directed acyclic graph that reflects the actual processing sequence. Dual attention feature extraction module: adopts a graph attention network architecture, and deeply decouples the scheduling state feature set through preset operation message attention blocks and machine message attention blocks; Collaborative intelligent agent decision-making module: Constructs a two-layer action space structure that includes operation selection and machine selection, and integrates job agents and machine agents to perform synchronous optimization. The job agent is responsible for selecting core processes from the set of processes to be processed, while the machine agent is responsible for matching the optimal processing resources for the selected processes. The two work together to output a comprehensive decision action set.

[0006] Preferably, it also includes a reward feedback calculation module, which generates a reward signal based on preset performance indicators such as completion time deviation or equipment utilization rate. The reward signal drives the evolution of the decision network in real time through the feedback loop of reinforcement learning.

[0007] Preferably, the feature vectors in the scheduling state feature set include operation feature vectors, machine feature vectors, and operation-machine pair feature vectors.

[0008] Preferably, the dynamic graph structure modeling module constructs the disjunctive graph model of the flexible workshop. ,in, This represents the set of all operation nodes and virtual nodes indicating the start and end of production. This represents the set of directed edges that define the internal priority constraints of a task. This represents the set of undirected edges selected by the machine.

[0009] Preferably, the dual-attention feature extraction module uses a specific adaptive attention weight calculation formula to capture the nonlinear relationship between job priority constraints and machine resource competition. The adaptive attention weight calculation formula is as follows: In the formula, .

[0010] A flexible job shop scheduling method based on dynamic evolutionary disjunctive graph and hierarchical collaborative architecture, applied to the method described above, includes the following steps: Real-time capture of production element data at the workshop floor level, by defining the current time step. The following is a set of related operations. Related machine set and compatible operation-machine pair set Extract global and local state features during the scheduling process, and transform the original physical features into high-dimensional state feature vectors by constructing a multi-dimensional feature set containing operation feature vectors, machine feature vectors, and operation-machine pair feature vectors. Constructing a disjunction graph model of a flexible workshop ,in, This represents the set of all operation nodes and virtual nodes indicating the start and end of production. This represents the set of directed edges that define the internal priority constraints of a task. This represents the set of undirected edges chosen by the machine. Based on the dynamic edge addition strategy, the edge structure of the new disjunctive graph is transformed into a directed acyclic graph that reflects the actual processing sequence by converting the non-identical edges representing machine selection into directional definite edges and simultaneously removing redundant disjunctive constraints. This allows the original complex graph structure to gradually evolve and solidify into a directed acyclic graph that reflects the actual processing sequence, thereby opening up a solution path in the constraint space. A graph attention network architecture is adopted, and the aforementioned scheduling state feature set is deeply decoupled through a preset operation message attention block and machine message attention block. By extracting the embedded features of nodes, a feature representation that integrates global and local features is generated. Implementing scheduling instructions based on collaborative Markov decision process: Construct a two-layer action space structure that includes operation selection and machine selection, and integrate job agent and machine agent for synchronous optimization. The job agent is responsible for selecting core processes from the set of processes to be processed, while the machine agent is responsible for matching the optimal processing resources for the selected processes. The two work together to output a comprehensive decision action set.

[0011] Preferably, the following steps are also included: Reward signals are generated based on preset performance indicators such as completion time deviation or equipment utilization rate. This signal drives the evolution of the decision network in real time through the feedback loop of reinforcement learning, enabling the system to achieve millisecond-level response and recovery of the scheduling scheme when faced with dynamic disturbances such as machine failure or emergency order insertion.

[0012] A computer-readable storage medium includes a stored program, wherein, when the program is executed, it controls the device on which the computer-readable storage medium is located to perform the flexible job shop scheduling system as described above.

[0013] A processor for running a program, wherein the program executes the flexible job shop scheduling method as described above during runtime.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention achieves parallel processing of job process constraints and machine resource competition relationships through built-in operation message attention blocks and machine message attention blocks. Utilizing the adaptive weight allocation mechanism of Graph Attention Networks (GAT), this scheme can accurately extract and fuse deep correlation features between "vertical job timing" and "horizontal resource allocation." This decoupled feature processing method effectively solves the problem of information redundancy or feature loss when general graph neural networks process heterogeneous scheduling data. Unlike traditional methods that construct all undirected edges during the initialization phase, leading to computational redundancy, this scheme dynamically adds a necessary set of directed edges at each decision time step, based only on the available operations and machine resources at the current moment. This modeling strategy, which transforms a static full graph into a dynamic evolving graph, greatly simplifies the topological complexity of the disjunctive graph while ensuring constraint integrity, constituting a core technical means to reduce the computational complexity of large-scale flexible job shop scheduling. Attached Figure Description

[0015] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0016] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present 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 the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0019] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0020] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0021] Example 1 In one embodiment of the present invention, a flexible job shop scheduling system based on improved graph reinforcement learning is provided. Its core logic lies in achieving dynamic closed-loop control of the production environment by constructing a highly perceptive digital model and deep neural network inference. This system aims to address the technical bottlenecks of flexible job shop scheduling (FJSP) in dynamic environments, including slow response, poor generalization, and difficulties in collaborative decision-making.

[0022] During the system initialization and data processing phases, the multi-dimensional state perception module captures real-time production element data from the workshop floor and defines the current time step. The following is a set of related operations. Related machine set and compatible operation-machine pair set The system extracts global and local state features during the scheduling process. By constructing a multi-dimensional feature set including operation feature vectors, machine feature vectors, and operation-machine pair feature vectors, the system transforms the original physical features into high-dimensional state feature vectors, providing accurate digital input for subsequent intelligent decision-making and ensuring that the scheduling system can keenly perceive dynamic fluctuations in the production site.

[0023] At the modeling and structural evolution level, the system utilizes a dynamic graph structure modeling module to construct a disjunctive graph model of the flexible work workshop. In this model, This represents the set of all operation nodes and virtual nodes indicating the start and end of production. This represents the set of directed edges that define the internal priority constraints of a task. This represents the set of undirected edges selected by the machine. During the scheduling decision-making process, this module uses a dynamic edge-adding strategy to re-evaluate the edge structure of the disjunctive graph. By transforming the undirected edges representing machine selection into directional definite edges and simultaneously removing redundant disjunctive constraints, the original complex graph structure gradually evolves and solidifies into a directed acyclic graph that reflects the actual processing sequence, thereby efficiently opening up solution paths in the complex constraint space.

[0024] The improvement in decision-making efficiency primarily relies on the dual-attention feature extraction module. This module employs a Graph Attention Network (GAT) architecture, deeply decoupling the aforementioned scheduling state feature set through pre-defined operation message attention blocks and machine message attention blocks. The system utilizes a specific adaptive attention weight calculation formula. LeakyReLU It accurately captures the nonlinear relationship between job priority constraints and machine resource competition. By extracting node embedding features, the system generates feature representations that integrate global perspective and local details, effectively overcoming the feature loss or computational redundancy problems that traditional algorithms are prone to when dealing with large-scale scheduling problems, and significantly enhancing the model's generalization ability and convergence speed.

[0025] In the final decision execution and feedback phase, the collaborative agent decision-making module implements scheduling instructions based on the Collaborative Markov Decision Process (coMDP). This module constructs a two-layer action space structure that includes operation selection and machine selection, and integrates job agents and machine agents for synchronous optimization. The job agent is responsible for selecting core processes from the set of processes to be processed, while the machine agent is responsible for matching the optimal processing resources to the selected processes. Together, they output a comprehensive decision action set. After the actions are executed, the reward feedback calculation module generates reward signals based on preset performance indicators such as completion time deviation or equipment utilization rate. This signal drives the evolution of the decision network in real time through a feedback loop of reinforcement learning, enabling the system to achieve millisecond-level response and smooth recovery of scheduling schemes when faced with dynamic disturbances such as machine failures or emergency order insertions, ultimately achieving intelligent closed-loop management of flexible workshop production.

[0026] In summary, the core logic of the above modules lies in achieving dynamic closed-loop control of the production environment by constructing highly perceptive digital models and deep neural network inference.

[0027] This solution aims to address the technical bottlenecks of Flexible Job Shop Scheduling (FJSP) in dynamic environments, including slow response, poor generalization, and difficulties in collaborative decision-making. In the data sensing phase, this solution utilizes a multi-dimensional state sensing module to capture real-time production element data at the shop floor level, defining the current time step... The following is a set of related operations. Related machine set and compatible operation-machine pair set This method extracts global and local state features during the scheduling process. By constructing a multi-dimensional feature set including operation feature vectors, machine feature vectors, and operation-machine pair feature vectors, this solution transforms the original physical features into high-dimensional state feature vectors, providing accurate digital input for subsequent intelligent decision-making and ensuring that the scheduling system can keenly perceive dynamic fluctuations in the production site.

[0028] At the modeling level, this solution utilizes a dynamic graph structure modeling module to construct a disjunctive graph model of the flexible workshop. In this model, This represents the set of all operation nodes and virtual nodes indicating the start and end of production. The set of directed edges representing the internal priority constraints of a task. The set of undirected edges representing machine selection. During the scheduling decision-making process, this scheme updates the edge structure of the disjunctive graph in real time based on a dynamic edge-adding strategy. By transforming the undirected edges representing machine selection into directional definite edges, the original complex graph structure gradually evolves and solidifies into a directed acyclic graph reflecting the actual processing sequence, thereby efficiently opening up solution paths in a complex constraint space.

[0029] The improvement in decision-making efficiency mainly relies on the dual-attention feature extraction module. This module adopts a graph attention network (GAT) architecture and uses a specific adaptive attention weight calculation formula to accurately capture the nonlinear relationship between job priority constraints and machine resource competition.

[0030] By extracting node embedding features, this scheme generates feature representations that integrate global perspective and local details, effectively overcoming the feature loss problem that traditional algorithms are prone to when dealing with large-scale scheduling problems, and significantly enhancing the model's generalization ability. In the final execution phase, this scheme implements scheduling instructions based on a collaborative agent decision-making module. This module integrates job agents and machine agents for synchronous optimization, collaboratively outputting a comprehensive set of decision actions. After action execution, the reward feedback calculation module generates reward signals based on preset performance indicators. Through a reinforcement learning feedback loop, it drives the evolution of the decision network in real time, enabling this scheme to achieve millisecond-level response and smooth recovery of the scheduling scheme when facing dynamic disturbances such as machine failures or urgent order insertions.

[0031] In summary, this architecture achieves parallel processing of job process constraints and machine resource competition relationships through built-in operation message attention blocks and machine message attention blocks. Utilizing the adaptive weight allocation mechanism of Graph Attention Networks (GAT), this scheme can accurately extract and fuse deep correlation features between "vertical job timing" and "horizontal resource allocation." This decoupled feature processing method effectively solves the problem of information redundancy or feature loss when general graph neural networks process heterogeneous scheduling data. Unlike traditional methods that construct all undirected edges during the initialization phase, leading to computational redundancy, this scheme dynamically adds a necessary set of directed edges at each decision time step, based only on the available operations and machine resources at the current moment. This modeling strategy, which transforms from a static full graph to a dynamic evolving graph, greatly simplifies the topological complexity of the disjunctive graph while ensuring constraint integrity, constituting a core technical means to reduce the computational complexity of large-scale flexible job shop scheduling.

[0032] Example 2 In one embodiment of the present invention, a flexible job shop scheduling method based on dynamic evolutionary disjunction graph and hierarchical collaborative architecture is provided, applied to the method described above, and includes the following steps: Step 1: Capture real-time production element data at the bottom of the workshop, by defining the current time step. The following is a set of related operations. Related machine set and compatible operation-machine pair set Extract global and local state features during the scheduling process, and transform the original physical features into high-dimensional state feature vectors by constructing a multi-dimensional feature set containing operation feature vectors, machine feature vectors, and operation-machine pair feature vectors. Step 2: Constructing the disjunction diagram model of the flexible workshop ,in, This represents the set of all operation nodes and virtual nodes indicating the start and end of production. This represents the set of directed edges that define the internal priority constraints of a task. This represents the set of undirected edges chosen by the machine. Step 3: Based on the dynamic edge addition strategy, the edge structure of the disjunctive graph is obtained by transforming the non-corresponding edges representing machine selection into directional definite edges, and simultaneously removing redundant disjunctive constraints. This allows the original complex graph structure to gradually evolve and solidify into a directed acyclic graph that reflects the actual processing sequence, thereby opening up a solution path in the constraint space. Step 4: Using a graph attention network architecture, the aforementioned scheduling state feature set is deeply decoupled through preset operation message attention blocks and machine message attention blocks. By extracting the embedded features of nodes, a feature representation that integrates global and local features is generated. Step 5: Implement scheduling instructions based on collaborative Markov decision process: Construct a two-layer action space structure that includes operation selection and machine selection, and integrate job agent and machine agent for synchronous optimization. The job agent is responsible for selecting core processes from the set of processes to be processed, while the machine agent is responsible for matching the optimal processing resources for the selected processes. The two work together to output a comprehensive decision action set.

[0033] Step Six: Generate reward signals based on preset performance indicators such as completion time deviation or equipment utilization rate. This signal drives the evolution of the decision network in real time through the feedback loop of reinforcement learning, enabling the system to achieve millisecond-level response and recovery of the scheduling scheme when faced with dynamic disturbances such as machine failure or emergency order insertion.

[0034] Those skilled in the art will recognize that the units of the various examples described in connection with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of the invention.

[0035] In the embodiments provided by the present invention, it should be understood that the division of units is only a logical functional division. In actual implementation, there may be other division methods, such as multiple units can be combined into one unit, one unit can be split into multiple units, or some features can be ignored.

[0036] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0037] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0038] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A flexible job shop scheduling system based on dynamic evolutionary disjunctive graph and hierarchical collaborative architecture, characterized in that, include: Multi-dimensional state perception module: Real-time capture of production element data at the bottom of the workshop, extraction of global and local state features in the scheduling process, construction of scheduling state feature set, and transformation of original physical features into high-dimensional state feature vector; Dynamic graph structure modeling module: Constructs a disjunctive graph model of a flexible workshop, updates the edge structure of the disjunctive graph in real time based on a dynamic edge addition strategy, transforms the non-identical edges representing machine selection into directional definite edges, and simultaneously removes redundant disjunctive constraints, so that the original graph structure gradually evolves and solidifies into a directed acyclic graph that reflects the actual processing sequence. Dual attention feature extraction module: adopts a graph attention network architecture, and deeply decouples the scheduling state feature set through preset operation message attention blocks and machine message attention blocks; Collaborative intelligent agent decision-making module: Constructs a two-layer action space structure that includes operation selection and machine selection, and integrates job agents and machine agents to perform synchronous optimization. The job agent is responsible for selecting core processes from the set of processes to be processed, while the machine agent is responsible for matching the optimal processing resources for the selected processes. The two work together to output a comprehensive decision action set.

2. The flexible job shop scheduling system based on dynamic evolutionary disjunctive graph and hierarchical collaborative architecture according to claim 1, characterized in that, It also includes a reward feedback calculation module, which generates reward signals based on preset performance indicators such as completion time deviation or equipment utilization rate. The reward signals drive the evolution of the decision network in real time through the feedback loop of reinforcement learning.

3. The flexible job shop scheduling system based on dynamic evolutionary disjunctive graph and hierarchical collaborative architecture according to claim 1, characterized in that, The feature vectors in the scheduling state feature set include operation feature vectors, machine feature vectors, and operation-machine pair feature vectors.

4. The flexible job shop scheduling system based on dynamic evolutionary disjunctive graph and hierarchical collaborative architecture according to claim 1, characterized in that, The dynamic graph structure modeling module constructs the disjunctive graph model of the flexible workshop. ,in, This represents the set of all operation nodes and virtual nodes indicating the start and end of production. This represents the set of directed edges that define the internal priority constraints of a task. This represents the set of undirected edges selected by the machine.

5. A flexible job shop scheduling system based on dynamic evolutionary disjunctive graph and hierarchical collaborative architecture according to claim 1, characterized in that, The dual-attention feature extraction module uses a specific adaptive attention weight calculation formula to capture the non-linear relationship between job priority constraints and machine resource competition. The specific adaptive attention weight calculation formula is as follows: In the formula, LeakyReLU is a nonlinear activation function; Let be the attention weight vector, and its transpose is: The characteristic linear transformation matrix; and Representing nodes respectively and nodes Initial eigenvectors; symbols This represents the concatenation operation of feature vectors.

6. A flexible job shop scheduling method based on dynamic evolutionary disjunctive graph and hierarchical collaborative architecture, characterized in that, The method applied to any one of claims 1 to 5 includes the following steps: Real-time capture of production element data at the workshop floor level, by defining the current time step. The following is a set of related operations. Related machine set and compatible operation-machine pair set Extract global and local state features during the scheduling process, and transform the original physical features into high-dimensional state feature vectors by constructing a multi-dimensional feature set containing operation feature vectors, machine feature vectors, and operation-machine pair feature vectors. Constructing a disjunction graph model of a flexible workshop ,in, This represents the set of all operation nodes and virtual nodes indicating the start and end of production. This represents the set of directed edges that define the internal priority constraints of a task. This represents the set of undirected edges chosen by the machine. Based on the dynamic edge addition strategy, the edge structure of the new disjunctive graph is transformed into a directed acyclic graph that reflects the actual processing sequence by converting the non-identical edges representing machine selection into directional definite edges and simultaneously removing redundant disjunctive constraints. This allows the original complex graph structure to gradually evolve and solidify into a directed acyclic graph that reflects the actual processing sequence, thereby opening up a solution path in the constraint space. A graph attention network architecture is adopted, and the aforementioned scheduling state feature set is deeply decoupled through a preset operation message attention block and machine message attention block. By extracting the embedded features of nodes, a feature representation that integrates global and local features is generated. Implementing scheduling instructions based on collaborative Markov decision process: Construct a two-layer action space structure that includes operation selection and machine selection, and integrate job agent and machine agent for synchronous optimization. The job agent is responsible for selecting core processes from the set of processes to be processed, while the machine agent is responsible for matching the optimal processing resources for the selected processes. The two work together to output a comprehensive decision action set.

7. A flexible job shop scheduling method based on dynamic evolutionary disjunctive graph and hierarchical collaborative architecture according to claim 6, characterized in that, It also includes the following steps: Reward signals are generated based on preset performance indicators such as completion time deviation or equipment utilization rate. This signal drives the evolution of the decision network in real time through the feedback loop of reinforcement learning, enabling the system to achieve millisecond-level response and recovery of the scheduling scheme when faced with dynamic disturbances such as machine failure or emergency order insertion.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein, when the program is executed, it controls the device containing the computer-readable storage medium to perform the flexible job shop scheduling system according to any one of claims 1 to 5.

9. A processor, characterized in that, The processor is used to run a program, wherein the program executes the flexible job shop scheduling method according to claim 6 or 7 when it runs.

Images