A resource-aware joint scheduling method and system for flexible work shops and automated guided vehicles

Abstract

This invention belongs to the field of intelligent manufacturing and production scheduling technology, and discloses a resource-aware joint scheduling method and system for flexible workshops and automated guided vehicles (AGVs). The method first constructs a scheduling environment including multiple operations, processing machines, and AGVs. At scheduling time, it acquires real-time status information of the operations, processing machines, and AGVs, and performs resource-aware modeling to form a comprehensive state representation describing the current environment. Based on this, it jointly decides on the operations, processing machines, and transport vehicles to be scheduled, generates joint scheduling instructions and executes them, while simultaneously updating the system state and advancing the simulation clock. By analyzing the state changes before and after scheduling execution, the subsequent decision-making process is continuously optimized. This invention achieves collaborative scheduling of process processing and material handling, effectively improving equipment utilization and production efficiency.

Description

Technical Field

[0001] This invention belongs to the field of intelligent manufacturing and production scheduling technology, specifically a resource perception joint scheduling method and system for flexible workshops and automated guided vehicles, which is used to solve the scheduling optimization problem under the condition of high coupling between processing resources and transportation resources. Background Technology

[0002] With the continuous development of Industry 4.0 and intelligent manufacturing technologies, manufacturing systems are evolving towards greater flexibility, automation, and intelligence. The Flexible Job Shop Scheduling Problem (FJSP), as a typical problem in discrete manufacturing systems, has long been a research focus in the field of production scheduling due to its diverse processing paths, flexible resource allocation, and complex constraints.

[0003] In traditional flexible workshop models, scheduling objects mainly include jobs, processes, and processing machines, with research focusing on process sequencing and machine allocation. However, in actual manufacturing systems, the transfer of workpieces between different processing equipment also significantly impacts production efficiency. With the widespread application of Automated Guided Vehicles (AGVs) in workshop logistics systems, transportation resources have gradually become one of the key factors affecting production cycle time and system performance.

[0004] With the introduction of automated guided vehicles (AGVs), the flexible workshop scheduling problem evolves into a complex scheduling problem where processing resources and transportation resources are highly coupled, such as... Figure 1 As shown in the diagram. In this type of problem, the start time of a process is affected not only by the availability of the processing machine, but also by the availability of the automated guided vehicle (AGV), its current location, and transportation time. A close temporal correlation and resource competition relationship exists between the processing and transportation processes. The Gantt chart description of FJSPAGV is as follows: Figure 2 As shown, in Figure 2 In this context, machines require automated guided vehicles (AGVs) for transporting different workpieces during processing, and these AGVs have limited load capacities. This significantly increases the scale and complexity of the scheduling problem.

[0005] For the scheduling problem of flexible workshops integrating automated guided vehicles (AGVs), existing research mainly adopts the following methods: Separate modeling methods: Processing scheduling and transportation scheduling are modeled and solved independently, or transportation resources are allocated based on the processing scheduling results. This type of method is simple to model, but often ignores the dynamic interaction between processing and transportation, easily leading to conflicts or efficiency reductions in actual execution. Phased or sequential decision-making methods: These employ a phased decision-making strategy of first selecting processes, then allocating machines, and finally assigning AGVs. Although this method reduces the dimensionality of the decision space, information transfer between different stages is limited, easily causing loss of cross-resource dependency information and affecting the overall scheduling quality. Rule-based or heuristic scheduling methods: These generate scheduling schemes through priority rules or metaheuristic algorithms, offering advantages such as simple implementation and fast computation speed. However, in complex scenarios or with varying scales, they often rely on human experience, resulting in insufficient generalization ability and stability.

[0006] In recent years, some studies have begun to introduce data-driven scheduling decision-making methods. By modeling the workshop operating status, these methods achieve adaptive optimization of scheduling rules to improve scheduling performance in complex scenarios. However, existing methods mostly focus on modeling single resources or local state information, and are insufficient in characterizing the relationships between multiple types of resources such as process resources, processing resources, and automated guided vehicles. Under conditions of strong coupling of multiple resources, it is still difficult to achieve coordinated optimization of processing and transportation processes.

[0007] Therefore, how to model and jointly decide on various heterogeneous resources such as processes, processing machines, and automated guided vehicles under a unified framework in order to achieve collaborative optimization of the processing and transportation process remains an urgent technical problem to be solved. Summary of the Invention

[0008] To address the problems of fragmented processing and transportation scheduling, insufficient utilization of cross-resource information, and low stability of scheduling results in existing technologies, this invention provides a resource-aware joint scheduling method and system for flexible work workshops and automated guided vehicles.

[0009] To achieve the above objectives, the technical solution of the present invention is as follows:

[0010] A resource-aware joint scheduling method for flexible work shops and automated guided vehicles includes the following steps:

[0011] Step 1: Construct a flexible workshop scheduling environment, which includes multiple jobs, multiple processing machines, and multiple automated guided vehicles (AGVs). Each job consists of several processes, and each process is processed sequentially according to a predetermined process order. Transfer between adjacent processes is completed by AGVs.

[0012] Step 2: At the scheduling decision moment, obtain the process status information of the currently unfinished process, the operating status information of each processing machine, and the transportation status information of each automated guided vehicle, and construct the corresponding process resource status, processing resource status, and transportation resource status respectively;

[0013] Step 3: Based on the aforementioned multi-type resource status information, perform resource-aware modeling on the relationships between process resources, processing resources, and transportation resources to form a comprehensive status representation that describes the current scheduling environment;

[0014] Step 4: At the same scheduling moment, based on the comprehensive state representation, jointly determine the process to be scheduled, the corresponding processing machine, and the automated guided vehicle used to perform the transportation task, and form a joint scheduling decision result;

[0015] Step 5: Execute the corresponding processing and transportation operations based on the joint scheduling decision results, and update the process completion status, processing machine occupancy status, and automated guided vehicle transportation status, while advancing the system running time;

[0016] Step 6: Update the scheduling decision process based on the changes in system state before and after scheduling execution in order to continuously optimize the effect of subsequent scheduling decisions.

[0017] Preferably, the resource-aware modeling in step 3 further includes:

[0018] (1) Attention modeling is performed on process resources, and the process representation is updated by aggregating local neighborhood information:

[0019]

[0020] in, For process The preorder-self-postorder neighborhood set, Number the attention layer. Embedding the original features of the process, Indicates the first Layer attention update post-process Feature representation, Represents the local neighborhood set Any neighborhood process index in, Indicates the first In-layer neighbor process For the target process Attention weights Indicates the first The linear transformation matrix corresponding to the layer-by-layer attention network. Used to map the features of the previous process to the current feature space, representing the first... In-layer neighboring processes Feature representation;

[0021] (2) Attention modeling is performed on the processing machine resources, and the competitive relationship between machines is introduced for feature updating:

[0022]

[0023] in, Indicates the first Machine after layer attention update Feature representation, Indicates the target machine index. Indicates the target machine Another machine's index is in competition with it; Indicates the first Machine in layer For the machine Attention weights are used to characterize competing machines. For the target machine Indicates the importance of the update; Indicates the first The linear transformation matrix corresponding to the layered machine attention network; Indicates the first Machine in layer Feature representation;

[0024] (3) Perform attention modeling on transportation resources, perform global modeling on the competitive relationship between automated guided vehicles, and update the transportation resource representation:

[0025]

[0026] in, Indicates the first After layer attention update, the automated guided vehicle Feature representation, Indicates the target automated guided vehicle index. Indicates the target automated guided vehicle Another automated guided vehicle index that is in competition with it. Indicates the first Automated Guided Vehicle (AGV) For automated guided vehicles The attention weights are used to characterize the importance of other automated guided vehicles (AGVs) to the target AGV's representation update. Indicates the first The linear transformation matrix corresponding to the layered transport resource attention network; Indicates the first Automated Guided Vehicle (AGV) Feature representation;

[0027] (4) Based on the representation of the above three types of resources, construct a comprehensive global state representation. This is used for subsequent joint scheduling decisions; the comprehensive global state representation is constructed as follows:

[0028]

[0029] in, This represents the overall global state. This indicates the number of the last attention network layer. Indicates that all processes are in the first stage. The feature set output by the layer Indicates that all machines are in the first... The feature set output by the layer This indicates that all automated guided vehicles are in the first... The feature set output by the layer This indicates a pooling operation, used to aggregate multiple feature vectors of similar resources into a fixed-dimensional resource-level representation. The pooling operation is average pooling. The fusion function is used to combine the process resource representation, machine resource representation, and transportation resource representation to generate a unified global state representation.

[0030] Preferably, the joint scheduling decision in step 4 further includes:

[0031] (1) At each decision step Define scheduling actions ,in Indicate homework The One process, Indicates processing machinery. Indicates an automated guided vehicle;

[0032] (2) Filter action combinations that do not meet process constraints, machine availability constraints and transportation feasibility constraints through a dynamic masking mechanism;

[0033] (3) Based on the comprehensive state representation, candidate actions are scored, and the scheduling probability distribution is output using masked Softmax:

[0034]

[0035] in, Candidate Triples The joint feature vector, where f is the scoring function;

[0036] (4) Based on the aforementioned scheduling probability distribution Select or sample actions from the set of available actions. And execute.

[0037] Preferably, the scheduling policy update in step 6 employs a reinforcement learning-based training method, including:

[0038] (1) Constructing a policy network and value network The scheduling strategy is optimized using the PPO algorithm;

[0039] (2) At each decision step Sampling action Calculate value estimation ;

[0040] (3) Accepting constraint violation vector By introducing dual variables and a penalty mechanism, the constraint violation is transformed into a moderating reward:

[0041]

[0042] in, For dual updater, This is a penalty item;

[0043] (4) Collecting and transferring samples Calculate generalized dominance estimation ;

[0044] (5) Calculate the total loss of PPO And update all parameters ;

[0045] (6) Output the policy parameters after training through multiple rounds of training and validation set evaluation. Used for online scheduling decisions.

[0046] A resource-aware joint scheduling system for flexible workshops and automated guided vehicles (AGVs) includes: an environment construction module for constructing a flexible workshop scheduling environment containing multiple jobs, processing machines, and AGVs; a state acquisition module for acquiring various resource state information of processes, processing machines, and AGVs at the time of scheduling decision-making; a resource-aware modeling module for modeling the relationships between process resources, processing resources, and transportation resources to form a comprehensive state representation; a joint decision-making module for jointly determining the processes, processing machines, and AGVs to be scheduled based on the comprehensive state representation at the same scheduling time; a scheduling execution and state update module for executing scheduling decisions and updating the system state; and a strategy update module for optimizing and updating the scheduling strategy based on the scheduling execution results.

[0047] Preferably, the resource awareness modeling module further includes: a process attention unit, used to aggregate process neighborhood information through an attention mechanism and update the process representation; a machine attention unit, used to characterize the competitive relationship between machines through an attention mechanism and update the machine representation; a transportation attention unit, used to characterize the competitive relationship between automated guided vehicles through an attention mechanism and update the transportation representation; and a fusion unit, used to fuse the updated three types of resource representations into a comprehensive global representation.

[0048] Preferably, the joint decision-making module further includes: an action definition unit, used to define joint scheduling action triples. The feasibility masking unit is used to filter out combinations of unfeasible actions through a dynamic masking mechanism; the scoring and selection unit is used to score feasible actions based on the comprehensive state representation and according to the scheduling probability distribution. Select the action to perform.

[0049] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method.

[0050] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described above.

[0051] Compared with the prior art, the present invention has the following advantages:

[0052] (1) This invention reduces information loss caused by phased scheduling by jointly considering process resources, processing resources and transportation resources under a unified decision-making framework;

[0053] (2) By using resource-aware modeling to depict the relationships between multiple types of resources, this approach is conducive to improving the adaptability of scheduling decisions to complex scenarios;

[0054] (3) This invention improves the coordination between the processing and transportation processes by jointly determining the scheduling schemes of multiple types of resources at the same scheduling time;

[0055] (4) This invention is applicable to flexible workshops of different sizes and configurations. It can maintain the stability of scheduling performance when the size of the workshop changes or the resource allocation is adjusted, and it has good scalability. The specific algorithm framework is as follows: Figure 4 As shown. Attached Figure Description

[0056] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0057] Figure 1 This is a comparison diagram of the scheduling of flexible work workshops and integrated automated guided vehicles (AGVs).

[0058] Figure 2 This is a Gantt chart of a flexible workshop scheduling problem integrating automated guided vehicles (AGVs);

[0059] Figure 3 It is an algorithm flowchart;

[0060] Figure 4 This is an algorithm framework diagram;

[0061] Figure 5 This is a diagram illustrating multiple attention. Detailed Implementation

[0062] 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 embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0063] Example 1

[0064] A resource-aware joint scheduling method for flexible workshops and automated guided vehicles, such as Figure 3 As shown, the method of this invention generally includes steps such as scheduling environment construction, state acquisition, resource-aware modeling, joint scheduling decision-making, scheduling execution and state update, and strategy optimization and update. Resource-aware modeling is used to extract the correlation features between multiple types of resources in the current scheduling environment, and joint scheduling decision-making is used to simultaneously determine the matching relationship between processes, machines, and automated guided vehicles. Specifically, it includes the following steps:

[0065] Step 1: Construct a flexible workshop scheduling environment. This environment includes multiple jobs, multiple processing machines, and multiple automated guided vehicles (AGVs). Each job consists of several processes, which are processed sequentially according to a predetermined process order. Transfer between adjacent processes is completed by the AGVs. For example... Figure 1 As shown, compared to the traditional flexible workshop scheduling model that only considers the relationship between processes and machines, the introduction of automated guided vehicles (AGVs) requires workpieces to be transported between different machines. Therefore, the scheduling object expands from processes and machines to three types of resources: processes, machines, and AGVs. At this point, whether a process can start on time depends not only on the availability of the target machine but also on whether the AGV can arrive in time and complete the material transfer. Thus, a significant coupling relationship is formed between processing resources and transportation resources.

[0066] Step 2: At the scheduling decision moment, obtain the process status information of the currently unfinished process, the operating status information of each processing machine, and the transportation status information of each automated guided vehicle, and construct the corresponding process resource status, processing resource status, and transportation resource status respectively;

[0067] Step 3: Based on the aforementioned multi-type resource status information, perform resource-aware modeling of the relationships between process resources, processing resources, and transportation resources to form a comprehensive status representation describing the current scheduling environment; such as... Figure 4 As shown, the algorithm framework of this invention includes a process resource representation learning branch, a machine resource representation learning branch, and a transportation resource representation learning branch. After feature updates are completed by the corresponding attention modules for each of the three types of resources, a comprehensive global state representation is generated by the fusion module and input into the joint decision-making module to output scheduling actions. The resource-aware modeling further includes:

[0068] (1) Attention modeling is performed on process resources, and the process representation is updated by aggregating local neighborhood information:

[0069]

[0070] in, For process The preorder-self-postorder neighborhood set, Number the attention layer. Embedding the original features of the process, Indicates the first Layer attention update post-process Feature representation, Represents the local neighborhood set Any neighborhood process index in, Indicates the first In-layer neighboring processes For the target process Attention weights Indicates the first The linear transformation matrix corresponding to the layer-by-layer attention network. Used to map the features of the previous process to the current feature space, representing the first... In-layer neighboring processes Feature representation;

[0071] (2) Attention modeling is performed on the processing machine resources, and the competitive relationship between machines is introduced for feature updating:

[0072]

[0073] in, Indicates the first Machine after layer attention update Feature representation, Indicates the target machine index. Indicates the target machine Another machine's index is in competition with it; Indicates the first Machine in layer For the machine Attention weights are used to characterize competing machines. For the target machine Indicates the importance of the update; Indicates the first The linear transformation matrix corresponding to the layered machine attention network; Indicates the first Machine in layer Feature representation;

[0074] (3) Perform attention modeling on transportation resources, perform global modeling on the competitive relationship between automated guided vehicles, and update the transportation resource representation:

[0075]

[0076] in, Indicates the first After layer attention update, the automated guided vehicle Feature representation, Indicates the target automated guided vehicle index. Indicates the target automated guided vehicle Another automated guided vehicle index that is in competition with it. Indicates the first Automated Guided Vehicle (AGV) For automated guided vehicles The attention weights are used to characterize the importance of other automated guided vehicles (AGVs) to the target AGV's representation update. Indicates the first The linear transformation matrix corresponding to the layered transport resource attention network; Indicates the first Automated Guided Vehicle (AGV) Feature representation;

[0077] (4) Based on the representation of the above three types of resources, construct a comprehensive global state representation. This is used for subsequent joint scheduling decisions; the comprehensive global state representation is constructed as follows:

[0078]

[0079] in, This represents the overall global state. This indicates the number of the last attention network layer. Indicates that all processes are in the first stage. The feature set output by the layer Indicates that all machines are in the first... The feature set output by the layer This indicates that all automated guided vehicles are in the first... The feature set output by the layer This indicates a pooling operation, used to aggregate multiple feature vectors of similar resources into a fixed-dimensional resource-level representation. The pooling operation is average pooling. The fusion function is used to combine the process resource representation, machine resource representation, and transportation resource representation to generate a unified global state representation.

[0080] Step 4: At the same scheduling moment, based on the comprehensive state representation, jointly determine the process to be scheduled, the corresponding processing machine, and the automated guided vehicle (AGV) used to perform the transportation task, forming a joint scheduling decision result; the joint scheduling decision further includes:

[0081] (1) At each decision step Define scheduling actions ,in Indicate homework The One process, Indicates processing machinery. Indicates an automated guided vehicle;

[0082] (2) Filter action combinations that do not meet process constraints, machine availability constraints and transportation feasibility constraints through a dynamic masking mechanism;

[0083] (3) Based on the comprehensive state representation, candidate actions are scored, and the scheduling probability distribution is output using masked Softmax:

[0084]

[0085] in, Candidate Triples The joint feature vector, where f is the scoring function;

[0086] (4) Based on the aforementioned scheduling probability distribution Select or sample actions from the set of available actions. And execute.

[0087] Step 5: Execute the corresponding processing and transportation operations based on the joint scheduling decision results, and update the process completion status, processing machine occupancy status, and automated guided vehicle transportation status, while advancing the system running time;

[0088] Step 6: Update the scheduling decision-making process based on the changes in system state before and after scheduling execution to continuously optimize the effectiveness of subsequent scheduling decisions; the scheduling strategy update adopts a reinforcement learning-based training method, including:

[0089] (1) Constructing a policy network and value network The scheduling strategy is optimized using the PPO algorithm;

[0090] (2) At each decision step Sampling action Calculate value estimation ;

[0091] (3) Accepting constraint violation vector By introducing dual variables and a penalty mechanism, the constraint violation is transformed into a moderating reward:

[0092]

[0093] in, For dual updater, This is a penalty item;

[0094] (4) Collecting and transferring samples Calculate generalized dominance estimation ;

[0095] (5) Calculate the total loss of PPO And update all parameters ;

[0096] (6) Output the policy parameters after training through multiple rounds of training and validation set evaluation. Used for online scheduling decisions.

[0097] Example 2

[0098] A resource-aware joint scheduling system for flexible workshops and automated guided vehicles includes:

[0099] The environment building module is used to build a flexible workshop scheduling environment that includes multiple jobs, processing machines, and automated guided vehicles;

[0100] The status acquisition module is used to acquire various resource status information of processes, processing machines, and automated guided vehicles at the time of scheduling decision-making.

[0101] The resource awareness modeling module is used to model the relationships between process resources, processing resources, and transportation resources to form a comprehensive state representation;

[0102] The joint decision-making module is used to jointly determine the scheduling process, processing machine, and automated guided vehicle based on the comprehensive state representation at the same scheduling time.

[0103] The scheduling execution and status update module is used to execute scheduling decisions and update the system status;

[0104] The strategy update module is used to optimize and update the scheduling strategy based on the scheduling execution results.

[0105] Preferably, the resource-aware modeling module further includes:

[0106] The process attention unit is used to aggregate process neighborhood information and update the process representation through an attention mechanism.

[0107] The machine attention unit is used to characterize the competitive relationship between machines and update the machine representation through an attention mechanism.

[0108] The transportation attention unit is used to characterize the competitive relationship between automated guided vehicles through an attention mechanism and update the transportation representation.

[0109] The fusion unit is used to merge the updated three types of resource representations into a comprehensive global representation.

[0110] Preferably, the joint decision-making module further includes:

[0111] Action definition unit, used to define joint scheduling action triples. ;

[0112] Feasibility masking unit, used to filter out combinations of infeasible actions through a dynamic masking mechanism;

[0113] The scoring and selection unit is used to score feasible actions based on the comprehensive state representation and according to the scheduling probability distribution. Select the action to perform.

[0114] Example 3

[0115] I. Symbol Conventions and Object Definition

[0116] Let the set of jobs be The processing machines are assembled as The automated guided vehicles are grouped into Each assignment It consists of several processes and satisfies process sequence constraints. The scheduling process follows discrete decision steps. Progress, System Status Record for.

[0117] II. Joint Scheduling Actions and Feasibility Constraints

[0118] At each decision step The scheduling action simultaneously determines the process to be scheduled, the processing machine, and the automated guided vehicle, defined as follows: ,in Indicate homework The One process, Indicates processing machinery. This refers to an automated guided vehicle (AGV). Unsuitable combinations of actions are filtered out using dynamic masks to ensure that process constraints, machine availability constraints, and transportation feasibility constraints are all satisfied.

[0119] III. Resource-Aware Attention Modeling

[0120] To depict the relationships between process resources, handling resources, and transportation resources, three types of resource representations are constructed and resource-aware attention aggregation is performed to form a comprehensive state representation. Detailed design is as follows: Figure 5 As shown, the multi-attention modeling structure extracts features for three types of resources: local neighborhood attention is used for process resources to characterize the relationship between preceding, current, and subsequent processes; competitive attention is used for machine resources to characterize the competitive relationship between different machines due to the overlap of their process sets; and global attention is used for automated guided vehicles (AGVs) to characterize the competitive relationship among multiple AGVs around a limited transportation task. These three attention mechanisms enhance the model's ability to express the coupling relationships of multiple resources.

[0121] Process resource attention: For each incomplete process Take its local neighborhood as the preorder-self-postorder set. The process of aggregating and updating neighborhood information according to attention weights is represented as follows: ,in Number the attention layer. Embed the original features of the process.

[0122] Machine resource attention: Model the competition relationship between the processing machine representation and update it. Furthermore, machine-competitive edge features are introduced into the attention score. To depict the competitive relationship caused by the overlap of processable steps.

[0123] Transportation resource attention: Globally model the competitive relationships between automated guided vehicles and update them. .

[0124] Through the above three types of representations, a comprehensive global representation is obtained. , used for joint scheduling scoring.

[0125] IV. Joint Scheduling Decision Scoring and Output

[0126] For candidate triples Construct joint features and score them, then use masked softmax to output the scheduling probability distribution. ,in Let f be the joint feature vector, and f be the scoring function. According to... Select or sample actions from the feasible set. And execute.

[0127] VI. System Flow

[0128]

[0129]

[0130] Each round of training first extracts instances from the training instance set and resets the parallel scheduling environment; then, at each decision step... In feasibility mask Under constraints, the strategy Sampling joint scheduling action And by value function Computational value estimation; environmental return constraint violation vector The dual updater U converts it into a penalty term. Thus, a regulatory reward is obtained. and transfer samples The data is recorded for subsequent updates. After trajectory acquisition is complete, the advantage estimate is calculated, and K rounds of PPO updates are performed to simultaneously update the policy and value parameters. The training data is used as the training set, and the performance of the current policy is evaluated on the validation set. Finally, the trained policy parameters are output. Used for online joint scheduling decisions. The algorithm pseudocode is shown in Algorithm 1.

[0131] Example 4

[0132] This embodiment applies the resource-aware joint scheduling method for flexible workshops and automated guided vehicles (AGVs) described in this invention to a flexible workshop production scheduling scenario to illustrate the application and effects of this invention in actual production organization. The production scenario includes 10 jobs, 5 processing machines, and 3 AGVs. Each job contains several processes, which are processed sequentially according to a predetermined process order. Material transfer between adjacent processes is completed by the AGVs. The scenario corresponds to... A flexible work workshop and automated guided vehicle collaborative scheduling environment.

[0133] In this embodiment, a scheduling environment is first constructed based on production task information, machine configuration information, and automated guided vehicle (AGV) configuration information. Then, at each scheduling decision moment, the process status information of currently incomplete processes, the operating status information of each processing machine, and the transportation status information of each AGV are acquired, and process resource status, processing resource status, and transportation resource status are constructed respectively. Next, based on the resource-aware modeling method described in this invention, the relationships between process resources, processing resources, and transportation resources are modeled to form a comprehensive global state representation describing the current scheduling environment. Then, based on the comprehensive global state representation, the processes to be scheduled, the corresponding processing machines, and the AGVs used to perform transportation tasks are jointly determined at the same scheduling moment, forming a joint scheduling action and executing it. After the action is executed, the process completion status, processing machine occupancy status, and AGV transportation status are updated, while the system running time is advanced until all operations are completed.

[0134] like Figure 2 As shown, in the scheduling process of a flexible workshop integrating automated guided vehicles (AGVs), the machine processing section and the AGV transportation section alternate on the time axis. If the AGV fails to complete transportation in time, subsequent processes cannot begin processing immediately even if the corresponding machines are idle; conversely, if the machines are not released in time, the AGVs may also wait. Therefore, the Gantt chart reflects the close temporal coupling between the processing and transportation processes, and also illustrates that in this type of production scheduling scenario, collaborative organization and joint scheduling of processes, processing machines, and AGVs are necessary to reduce waiting and conflicts and improve overall scheduling efficiency.

[0135] To illustrate the effectiveness of the method of this invention, it is compared with a conventional phased scheduling method. The conventional phased scheduling method first determines the processing steps and corresponding processing machines, and then determines the automated guided vehicle (AGV) transportation task based on the processing allocation results. Compared to this method, this invention jointly determines the processing steps, machines, and AGVs at the same scheduling moment, which can reduce information fragmentation caused by separate decision-making for multiple types of resources and improve the connection between the processing and transportation processes.

[0136] In the above In a production scheduling scenario, the total completion time was 471.09 seconds after using the method of this invention. This result demonstrates that the method of this invention can effectively complete joint scheduling in a production environment with a strong coupling relationship between processing and transportation resources, reducing unnecessary waiting and conflicts, and improving scheduling efficiency.

[0137] Furthermore, when the scale of production tasks expands and workshop resource allocation increases, the method of the present invention can still be applied to larger-scale flexible work workshops and automated guided vehicles collaborative scheduling scenarios to achieve joint scheduling and collaborative optimization under multi-resource coupling conditions.

[0138] Therefore, the method of the present invention can be applied to the production scheduling process of flexible operation workshops to realize the coordinated organization of process processing and material transfer, and has good application value.

[0139] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A resource-aware joint scheduling method for flexible workshops and automated guided vehicles, characterized in that, Includes the following steps: Step 1: Construct a flexible workshop scheduling environment, which includes multiple jobs, multiple processing machines, and multiple automated guided vehicles (AGVs). Each job consists of several processes, and each process is processed sequentially according to a predetermined process order. Transfer between adjacent processes is completed by AGVs. Step 2: At the scheduling decision moment, obtain the process status information of the currently unfinished process, the operating status information of each processing machine, and the transportation status information of each automated guided vehicle, and construct the corresponding process resource status, processing resource status, and transportation resource status respectively; Step 3: Based on the aforementioned multi-type resource status information, perform resource-aware modeling on the relationships between process resources, processing resources, and transportation resources to form a comprehensive status representation that describes the current scheduling environment; Step 4: At the same scheduling moment, based on the comprehensive state representation, jointly determine the process to be scheduled, the corresponding processing machine, and the automated guided vehicle used to perform the transportation task, and form a joint scheduling decision result; Step 5: Execute the corresponding processing and transportation operations based on the joint scheduling decision results, and update the process completion status, processing machine occupancy status, and automated guided vehicle transportation status, while advancing the system running time; Step 6: Update the scheduling decision process based on the changes in system state before and after scheduling execution in order to continuously optimize the effect of subsequent scheduling decisions.

2. The resource-aware joint scheduling method according to claim 1, characterized in that, Step 3, resource-aware modeling, further includes: (1) Attention modeling is performed on process resources, and the process representation is updated by aggregating local neighborhood information: ， in, For process The preorder-self-postorder neighborhood set, Number the attention layer. Embedding the original features of the process, Indicates the first Layer attention update post-process Feature representation, Represents the local neighborhood set Any neighborhood process index in, Indicates the first In-layer neighbor process For the target process Attention weights Indicates the first The linear transformation matrix corresponding to the layer-by-layer attention network. Used to map the features of the previous process to the current feature space, representing the first... In-layer neighboring processes Feature representation; (2) Attention modeling is performed on the processing machine resources, and the competitive relationship between machines is introduced for feature updating: ， in, Indicates the first Machine after layer attention update Feature representation, Indicates the target machine index. Indicates the target machine Another machine's index is in competition with it; Indicates the first Machine in layer For the machine Attention weights are used to characterize competing machines. For the target machine Indicates the importance of the update; Indicates the first The linear transformation matrix corresponding to the layered machine attention network; Indicates the first Machine in layer Feature representation; (3) Perform attention modeling on transportation resources, perform global modeling on the competitive relationship between automated guided vehicles, and update the transportation resource representation: ， in, Indicates the first After layer attention update, the automated guided vehicle Feature representation, Indicates the target automated guided vehicle index. Indicates the target automated guided vehicle Another automated guided vehicle index that is in competition with it. Indicates the first Automated Guided Vehicle (AGV) For automated guided vehicles The attention weights are used to characterize the importance of other automated guided vehicles (AGVs) to the target AGV's representation update. Indicates the first The linear transformation matrix corresponding to the layered transport resource attention network; Indicates the first Automated Guided Vehicle (AGV) Feature representation; (4) Based on the representation of the above three types of resources, construct a comprehensive global state representation. This is used for subsequent joint scheduling decisions; the comprehensive global state representation is constructed as follows: ， in, This represents the overall global state. This indicates the number of the last attention network layer. This indicates that all processes are in the [number]th stage. The feature set output by the layer Indicates that all machines are in the first... The feature set output by the layer This indicates that all automated guided vehicles are in the first... The feature set output by the layer This indicates a pooling operation, used to aggregate multiple feature vectors of similar resources into a fixed-dimensional resource-level representation. The pooling operation is average pooling. The fusion function is used to combine the process resource representation, machine resource representation, and transportation resource representation to generate a unified global state representation.

3. The method according to claim 1, characterized in that, The joint scheduling decision in step 4 further includes: (1) At each decision step Define scheduling actions ,in Indicate homework The One process, Indicates processing machinery. Indicates an automated guided vehicle; (2) Filter action combinations that do not meet process constraints, machine availability constraints and transportation feasibility constraints through a dynamic masking mechanism; (3) Based on the comprehensive state representation, candidate actions are scored, and the scheduling probability distribution is output using masked Softmax: ， in, Candidate Triples The joint feature vector, where f is the scoring function; (4) Based on the aforementioned scheduling probability distribution Select or sample actions from the set of available actions. And execute.

4. The resource-aware joint scheduling method according to claim 1, characterized in that, The scheduling policy update in step 6 employs a reinforcement learning-based training method, including: (1) Constructing a policy network and value network The scheduling strategy is optimized using the PPO algorithm; (2) At each decision step Sampling action Calculate value estimation ; (3) Accepting constraint violation vector By introducing dual variables and a penalty mechanism, the constraint violation is transformed into a moderating reward: ， in, For dual updater, This is a penalty item; (4) Collecting and transferring samples Calculate generalized dominance estimation ; (5) Calculate the total loss of PPO And update all parameters ; (6) Output the policy parameters after training through multiple rounds of training and validation set evaluation. Used for online scheduling decisions.

5. A resource perception and joint scheduling system for flexible workshops and automated guided vehicles, characterized in that, include: The environment building module is used to build a flexible workshop scheduling environment that includes multiple jobs, processing machines, and automated guided vehicles; The status acquisition module is used to acquire various resource status information of processes, processing machines, and automated guided vehicles at the time of scheduling decision-making. The resource awareness modeling module is used to model the relationships between process resources, processing resources, and transportation resources to form a comprehensive state representation; The joint decision-making module is used to jointly determine the scheduling process, processing machine, and automated guided vehicle based on the comprehensive state representation at the same scheduling time. The scheduling execution and status update module is used to execute scheduling decisions and update the system status; The strategy update module is used to optimize and update the scheduling strategy based on the scheduling execution results.

6. The system according to claim 5, characterized in that, The resource-aware modeling module further includes: The process attention unit is used to aggregate process neighborhood information and update the process representation through an attention mechanism. The machine attention unit is used to characterize the competitive relationship between machines and update the machine representation through an attention mechanism. The transportation attention unit is used to characterize the competitive relationship between automated guided vehicles through an attention mechanism and update the transportation representation. The fusion unit is used to merge the updated three types of resource representations into a comprehensive global representation.

7. The system according to claim 5, characterized in that, The joint decision-making module further includes: Action definition unit, used to define joint scheduling action triples. ; Feasibility masking unit, used to filter out combinations of infeasible actions through a dynamic masking mechanism; The scoring and selection unit is used to score feasible actions based on the comprehensive state representation and according to the scheduling probability distribution. Select the action to perform.

8. A computer-readable storage medium having a computer program stored thereon, the computer program, when executed by a processor, implementing the method as claimed in any one of claims 1 to 4.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1 to 4.

Images