State representation modeling method and system based on multi-domain graph attention network

Abstract

The invention discloses a state representation modeling method and system based on a multi-domain graph attention network, and the method comprises the steps: constructing the multi-domain graph attention network, extracting global semantic features from the three dimensions of operation, machines and workers, carrying out the fusion, generating a unified state representation, and effectively capturing a complex dependency relationship among multiple entities in production scheduling. Wherein heterogeneous association modeling of machine nodes and operation nodes is enhanced by a sub-graph attention mechanism based on a meta path, and calculation redundancy of cross-domain information fusion is reduced by feature stacking and pooling processing; the operation semantic embedding accurately represents time sequence constraint logic between processes through precursor-subsequent attention weight distribution; the dynamic aggregation of the worker features and the machine features further improves the interpretability of the resource allocation relationship. Finally generated global state representation outputs scheduling action distribution through a decision network, and while complex scheduling rule constraints are ensured, the accuracy and response efficiency of scheduling decisions in a dynamic production environment are remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of workshop scheduling, and in particular relates to a state representation modeling method and system based on a multi-domain graph attention network. Background Art

[0002] The flexible job shop scheduling problem (FJSP) is one of the key challenges facing the manufacturing sector as smart manufacturing evolves. With the advent of Industry 5.0, production models that emphasize the value of worker participation and collaboration between workers and robots are becoming increasingly popular, leading to the emergence of human-cyber-physical systems (HCPS). In the context of HCPS, FJSP becomes even more complex, requiring the allocation of appropriate machines and workers to a series of tasks while satisfying constraints such as machine availability and operation priority, in order to achieve objectives such as maximizing machine efficiency and minimizing total duration or energy consumption.

[0003] Traditional approaches to solving FJSP have numerous limitations. Heuristic-based methods, such as the commonly used Priority Dispatching Rules (PDRs), rely on simple rule-based strategies to generate scheduling solutions. These methods are difficult to adapt to complex and ever-changing production scenarios, lack robustness and scalability, and cannot effectively capture the complexities of task dependencies, resource constraints, and dynamic production priorities. While methods based on precise algorithms can theoretically achieve optimal or near-optimal results, they suffer from high computational costs and poor scalability when faced with the vast combinatorial search space found in large-scale manufacturing environments, limiting their practical application.

[0004] In recent years, machine learning, particularly deep reinforcement learning (DRL), has provided new approaches to solving FJSP. Some studies have attempted to use DRL to select from a given set of PDRs. However, this approach can only learn within a limited search space of PDRs, resulting in poor scheduling performance. Other studies have focused on generating high-quality PDRs by designing various network structures to extract the relationship between operations and machines. However, these approaches often overlook the significant impact of workers on production and fail to effectively model them, making them ineffective in HCPS-FJSP scenarios.

[0005] At the same time, existing FJSP scheduling methods lack a comprehensive representation of state, failing to fully integrate the states and dependencies of operations, machines, and workers. This results in poor decision-making quality and makes it difficult to meet the efficient and reliable manufacturing requirements of Industry 5.0. In summary, current technologies for solving FJSP in HCPS face challenges such as difficulty handling complex dependencies, neglecting worker factors, and incomplete state representation. A new technical solution is urgently needed to improve scheduling performance and decision-making quality. Summary of the Invention

[0006] To solve the above technical problems, the present invention proposes a state representation modeling method and system based on a multi-domain graph attention network to solve the problems existing in the above-mentioned prior art.

[0007] In a first aspect, to achieve the above-mentioned objectives, the present invention provides a state representation modeling method based on a multi-domain graph attention network, comprising the following steps:

[0008] Based on the operation node features, the features of the immediate predecessor and successor operation nodes, and the features of the related machine nodes, the operation node priorities are updated through the attention mechanism to generate the operation semantic embedding;

[0009] Based on the machine node features, associated operation node features, and operation-machine arc features, a machine subgraph is constructed through meta-paths. The machine node priorities are updated using the attention mechanism to generate machine semantic embeddings.

[0010] Based on the features of worker nodes, associated machine nodes, and machine-worker arcs, a worker subgraph is constructed. The worker node priorities are updated through the attention mechanism to generate worker semantic embeddings.

[0011] The operational semantic embedding, machine semantic embedding, and worker semantic embedding are feature-fused to generate a global state embedding.

[0012] Based on the global state embedding, a scheduling decision action is output through an actor-critic network, where the action includes an assignment pair of operation, machine, and worker.

[0013] Optionally, the process of updating the priority of the operation node includes:

[0014] Calculate the attention coefficient of the operation node and its adjacent operation nodes, normalize and aggregate the features of the adjacent nodes;

[0015] Calculate the attention coefficient of the operation node and the associated machine node, and aggregate the machine node features after normalization;

[0016] The complex dependencies between operation nodes are captured through the multi-head attention mechanism to generate operation semantic embedding.

[0017] Optionally, the process of constructing the machine subgraph includes:

[0018] Generate an initial machine subgraph by associating machine nodes and operation nodes through meta-paths;

[0019] The operation node features, machine node features, and operation-machine arc features are linearly transformed and stacked to form the machine subgraph node features;

[0020] Based on the stacked node features, the attention coefficients of adjacent machine nodes are calculated, and the node features are aggregated after normalization to generate machine semantic embedding.

[0021] Optionally, the process of constructing the worker subgraph includes:

[0022] Generate the initial worker subgraph by associating worker nodes with machine nodes through meta-paths;

[0023] Linearly transform and stack the machine node features, worker node features, and machine-worker arc features to form the worker subgraph node features;

[0024] Based on the stacked node features, the attention coefficients of adjacent worker nodes are calculated, and the node features are aggregated after normalization to generate the worker semantic embedding.

[0025] Optionally, the feature fusion process includes:

[0026] Concatenate the operational semantic embedding, machine semantic embedding, and worker semantic embedding by dimension to form a global state embedding.

[0027] The global state is embedded into the input critic network, which outputs a value estimate of the current environment state.

[0028] Optionally, the action output process of the actor-critic network includes:

[0029] Generate action scores through a multi-layer perceptron based on operation node embedding, machine node embedding, worker node embedding, and global state embedding;

[0030] Normalize the action scores and output the probability distribution of operation-machine-worker assignment pairs.

[0031] In a second aspect, the present invention further provides a state representation modeling system based on a multi-domain graph attention network, for implementing a state representation modeling method based on a multi-domain graph attention network, the system comprising:

[0032] The operation node modeling module is used to update the operation node priority through the attention mechanism based on the operation node features, the features of the immediate predecessor and successor operation nodes, and the features of the associated machine nodes, and generate the operation semantic embedding;

[0033] The machine node modeling module is used to construct a machine subgraph through meta-paths based on machine node features, associated operation node features, and operation-machine arc features, and uses the attention mechanism to update machine node priorities and generate machine semantic embeddings;

[0034] The worker node modeling module is used to construct a worker subgraph based on worker node features, associated machine node features, and machine-worker arc features, update worker node priorities through the attention mechanism, and generate worker semantic embeddings;

[0035] Feature fusion module, which is used to concatenate the operation semantic embedding, machine semantic embedding, and worker semantic embedding to generate a global state embedding;

[0036] A scheduling decision module is configured to output a scheduling decision action based on the global state embedding through an actor-critic network, wherein the action includes an allocation pair of operation, machine, and worker.

[0037] Optionally, the operation node modeling module includes:

[0038] An operation attention calculation unit, used to calculate the attention coefficient of an operation node and its adjacent operation nodes and associated machine nodes;

[0039] Operation feature aggregation unit, used to aggregate adjacent operation node features and machine node features based on the normalized attention coefficient;

[0040] The operational semantic embedding generation unit is used to capture the complex dependencies between operation nodes through a multi-head attention mechanism and generate operational semantic embeddings.

[0041] In a third aspect, the present invention further provides a computer terminal device, comprising:

[0042] one or more processors;

[0043] a memory, coupled to the processor, for storing one or more programs;

[0044] When the one or more programs are executed by the one or more processors, the one or more processors implement a state representation modeling method based on a multi-domain graph attention network.

[0045] In a fourth aspect, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a state representation modeling method based on a multi-domain graph attention network.

[0046] Compared with the prior art, the present invention has the following advantages and technical effects:

[0047] The present invention provides a state representation modeling method and system based on a multi-domain graph attention network. By constructing a multi-domain graph attention network, the present invention extracts global semantic features from the three dimensions of operation, machine, and worker, and fuses them to generate a unified state representation, effectively capturing the complex dependencies between multiple entities in production scheduling. Among them, the meta-path-based subgraph attention mechanism strengthens the heterogeneous association modeling of machine nodes and operation nodes, and feature stacking and pooling processing reduce the computational redundancy of cross-domain information fusion; operational semantic embedding accurately characterizes the timing constraint logic between processes through predecessor-successor attention weight allocation; the dynamic aggregation of worker features and machine features further improves the interpretability of resource allocation relationships. The final generated global state representation outputs the scheduling action distribution through the decision network, which significantly improves the accuracy and response efficiency of scheduling decisions in dynamic production environments while ensuring the constraints of complex scheduling rules. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The accompanying drawings, which constitute part of the present invention, are provided to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are provided to explain the present invention and do not constitute an undue limitation of the present invention. In the accompanying drawings:

[0049] Figure 1 is a flow chart of a method according to an embodiment of the present invention;

[0050] Figure 2 Schematic diagram of the overall framework of MDGAT according to an embodiment of the present invention;

[0051] Figure 3 Schematic diagram of a system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0052] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments of the present invention can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0053] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0054] It is hereby stated that this application belongs to the field of industrial manufacturing technology and is specifically applied to production scheduling systems. It addresses industrial technical issues such as resource conflicts, timing mismatches, and response lags that exist when multiple entities (operation procedures, machinery and equipment, and skilled workers) are coordinated in a dynamic production environment. A decision-making model that is deeply coupled with the physical constraints of the production scenario is constructed through a multi-domain graph attention network. The design of the scheme is based on objective production laws such as machine processing capabilities, physical dependencies of processes, and worker skill attributes in real manufacturing units. It converts industrial entity operating data such as machine status, process timing chains, and human resource allocation into computable features, solving the scheduling conflicts and efficiency bottlenecks caused by traditional scheduling algorithms that ignore equipment heterogeneity, dynamic coupling of processes, and human-machine adaptation relationships. Its technical means closely revolve around the physical properties and operating logic of the production system, and conform to the natural laws of data correlation and objective production constraints in the industrial field.

[0055] Example 1

[0056] like Figure 1 As shown, this embodiment provides a state representation modeling method based on a multi-domain graph attention network, including:

[0057] Based on the operation node features, the features of the immediate predecessor and successor operation nodes, and the features of the related machine nodes, the operation node priorities are updated through the attention mechanism to generate the operation semantic embedding;

[0058] Based on the machine node features, associated operation node features, and operation-machine arc features, a machine subgraph is constructed through meta-paths. The machine node priorities are updated using the attention mechanism to generate machine semantic embeddings.

[0059] Based on the features of worker nodes, associated machine nodes, and machine-worker arcs, a worker subgraph is constructed. The worker node priorities are updated through the attention mechanism to generate worker semantic embeddings.

[0060] The operational semantic embedding, machine semantic embedding, and worker semantic embedding are feature-fused to generate a global state embedding.

[0061] Based on the global state embedding, a scheduling decision action is output through an actor-critic network, where the action includes an assignment pair of operation, machine, and worker.

[0062] Specifically, a state representation modeling method based on MDGAT consists of an operation representation modeling module, a machine representation modeling module, a work representation modeling module, and a feature fusion module. The overall framework diagram of the multi-domain graph attention network (MDGAT) is shown in the figure. Figure 2 shown.

[0063] As an implementation method of this embodiment, the process of updating the priority of the operation node includes:

[0064] Calculate the attention coefficient of the operation node and its adjacent operation nodes, normalize and aggregate the features of the adjacent nodes;

[0065] Calculate the attention coefficient of the operation node and the associated machine node, and aggregate the machine node features after normalization;

[0066] The complex dependencies between operation nodes are captured through the multi-head attention mechanism to generate operation semantic embedding.

[0067] Operation Node Attention Representation Modeling. First, the operation node attention representation modeling is performed, and the priority nodes of different operations are updated using the operation node features, the direct successor operation node and the related machine node features in the same job. Specifically, given an operation node O ij , calculate the attention coefficients of its immediate predecessor and successor operation nodes and itself:

[0068]

[0069] Among them, W O Represents the linear transformation of the operation node, a is a single-layer feedforward neural network. Similarly, the relevant machine node is related to the operation node O ij The attention coefficient is calculated as:

[0070]

[0071] Among them, W M It is a linear transformation for machine nodes.

[0072] Then, all attention coefficients of the target operation node are normalized using the softmax function to obtain the normalized attention coefficient α ijq and α ijk Finally, by fusing adjacent operation nodes, adjacent machine nodes, and the node itself, the aggregation function for obtaining the operation node embedding is expressed as:

[0073]

[0074] Operational semantic attention representation modeling. Each operation node contains specific semantic information. The embedding of all operation nodes is fused using the mean method, ignoring the differences between nodes. This paper designs an operational semantic attention to learn the importance of different nodes. The output representation of the result of the attention of the operation node in the first layer is represented as symbolic modeling, denoted as Semantic importance coefficient of each operation node The calculation is as follows:

[0075]

[0076] Where WO' and b O' are the weight matrix and bias vector of the operation node, q O' is the semantic attention vector of the operation node, and tanh is the activation function.

[0077] Then, the semantic importance coefficient is normalized using the softmax function. ij The weight of Expressed as:

[0078]

[0079] Using the learned weights of the operation nodes as coefficients, the global operation semantic embedding is calculated as:

[0080]

[0081] In addition, the module adopts a multi-head attention mechanism to effectively capture the complex and diverse dependencies between operation nodes and semantic attention representation modeling. Here, H represents the number of attention heads in the selection representation modeling process.

[0082] As an implementation method in this embodiment, the process of constructing the machine subgraph includes:

[0083] Generate an initial machine subgraph by associating machine nodes and operation nodes through meta-paths;

[0084] The operation node features, machine node features, and operation-machine arc features are linearly transformed and stacked to form the machine subgraph node features;

[0085] Based on the stacked node features, the attention coefficients of adjacent machine nodes are calculated, and the node features are aggregated after normalization to generate machine semantic embedding.

[0086] In the machine representation modeling module, different operation nodes and machine nodes have more complex connection relationships. Different machines may compete for the same operation, and the node attributes in the proposed heterogeneous scheduling graph are heterogeneous. How to explicitly model the priority of machines is crucial for the agent's high-quality decision-making. Heterogeneous Graph Attention Networks (HGATs) (Wang et al. 2019) have become a promising method for learning heterogeneous graph representations with multiple node types and edge types. Meta-paths are used to link composite relationships between two different types of objects. In the context of HCPS-FJSP, HGATs allow us to capture both intra-type and inter-type relationships, thereby facilitating the representation modeling of complex dependencies. Inspired by HGATs, machine node attention representation modeling and semantic attention representation modeling are designed.

[0087] like Figure 1Here, different machine nodes are considered to be different types of nodes, while operation nodes are considered to be the same type of nodes. Therefore, using the meta path To describe the machine node M k and the relationship between the operation nodes. Using meta path Get an initial subgraph Initial subgraph Only the operation node features are included, and the machine and OM arc features are ignored. The operation node features, machine node features, and OM arc features are transformed into the same dimension using a linear transformation matrix and stacked as follows:

[0088]

[0089] in is the stacked machine node feature vector, which represents the formation matrix of the linear anti-OM arc. Finally, we get a subgraph in represents stacked machine nodes in a subgraph, The connection relationship between the stacked machine nodes is represented in the subgraph

[0090] Machine node attention representation modeling. First, the stacked machine node embedding is updated from the node level by stacking adjacent nodes. In particular, given a stacked machine node M ijk , the attention coefficient of adjacent nodes is calculated as:

[0091]

[0092] in Represents stacked machine node M pqk to M ijk The attention coefficient.

[0093] Note that the coefficients are normalized using the Softmax function and are expressed as:

[0094]

[0095] Then, by combining the embeddings and attention coefficients of neighboring nodes, the stacked machine nodes M can be aggregated ijk The updated stacked machine node can be expressed as:

[0096]

[0097] Finally, for all The stacked machine nodes are processed and the result is embedded as the node of machine k, that is:

[0098]

[0099] Machine semantic attention representation modeling. Given the Lth layer machine node embedding The semantic importance coefficient of The calculation method for each machine k is:

[0100]

[0101] Where W M' and b M' are the weight matrix and bias vector of the machine node respectively, is the semantic attention vector of the machine node.

[0102] By normalizing the semantic importance coefficients of all machines, the weight of each machine node Calculated as:

[0103]

[0104] Then, the global machine semantic embedding is calculated as:

[0105]

[0106] As an implementation method in this embodiment, the process of constructing the worker subgraph includes:

[0107] Generate the initial worker subgraph by associating worker nodes with machine nodes through meta-paths;

[0108] Linearly transform and stack the machine node features, worker node features, and machine-worker arc features to form the worker subgraph node features;

[0109] Based on the stacked node features, the attention coefficients of adjacent worker nodes are calculated, and the node features are aggregated after normalization to generate the worker semantic embedding.

[0110] In the worker representation modeling module, the superimposed worker node feature vector is obtained by superimposing the machine node feature, worker node feature and MW arc feature. And expressed as:

[0111]

[0112] Where W D and W W is the linear transformation matrix of MW arcs and work nodes. Finally, we get a subgraph in Representing a subgraph The stacked worker nodes, A subgraph representing the connection relationship between stacks The working nodes in .

[0113] Work node attention representation modeling. In the subgraph In the stack, the working nodes W kl The attention coefficient of each neighboring node is calculated as:

[0114]

[0115] in, For stacked working nodes W pl To W kl The attention coefficient.

[0116] The calculation formula for the normalized worker attention coefficient is:

[0117] After that, the embedding of each stacked worker node can be updated as:

[0118]

[0119] Finally, apply average pooling to the subgraph All stacked worker nodes in the process are processed. Therefore, the node embedding of worker l is expressed as:

[0120]

[0121] Worker semantic attention representation modeling. Similarly,

[0122] Based on the embedding of the l-th layer working nodes Semantic importance coefficient for each machine Expressed as:

[0123]

[0124] Where W W' and b W' are the weight matrix and bias vector of the working node respectively,

[0125] and is the semantic attention vector of the working node. At the same time, the weight of each working node is expressed as:

[0126]

[0127] Therefore, the global worker semantic embedding is calculated as:

[0128]

[0129] As an implementation method of this embodiment, the feature fusion process includes:

[0130] Concatenate the operational semantic embedding, machine semantic embedding, and worker semantic embedding by dimension to form a global state embedding.

[0131] The global state is embedded into the input critic network, which outputs a value estimate of the current environment state.

[0132] As an implementation in this embodiment, the action output process of the actor-critic network includes:

[0133] Generate action scores through a multi-layer perceptron based on operation node embedding, machine node embedding, worker node embedding, and global state embedding;

[0134] Normalize the action scores and output the probability distribution of operation-machine-worker assignment pairs.

[0135] In the representation fusion module, the representation fusion module integrates the global semantic embeddings from different domains into a rich global embedding to help the agent fully understand the environment state. Therefore, the obtained global operational semantic embedding h O , global machine semantic embedding h M and global worker semantic embedding h W The connection is globally embedded as follows:

[0136] h G =[h O ||h M ||h W ]

[0137] Through this fusion module, the resulting global embedding captures the combined information from all domains, enabling the decision component to make high-quality scheduling decisions.

[0138] decision making:

[0139] The decision network is the last component of the decision-making process and is designed based on the actor-critic network. The actor network and the critic network consist of MLP networks with parameters θ and w respectively. Note that an action is a compatible action on the OMW pair a t =(O ij ,M k ,W l ). Using the environment state extracted by embedding, the policy network outputs a scale μ(a t |s t ),Right now:

[0140]

[0141] Then, the probability of choosing an action is calculated as:

[0142]

[0143] In addition, the global embedding is used as the input of the critic network to estimate the current environment state s t The value of v w (s t ).

[0144] Based on this, embodiments of the present invention provide a state representation modeling method based on a multi-domain graph attention network (MDGAT). This method significantly improves the decision-making quality and efficiency of flexible workshop scheduling problems in intelligent manufacturing. This method utilizes a multi-domain attention mechanism across operation, machine, and worker nodes to accurately capture task dependencies, resource constraints, and the dynamic characteristics of human-machine collaboration, addressing the inadequacy of traditional heuristic rules and single graph networks in modeling complex interactions. Through meta-path-driven subgraph construction and feature stacking, it explicitly integrates heterogeneous information such as worker skills and machine load, achieving the first systematic deep fusion of worker status and scheduling decisions, avoiding resource mismatches in the human-cyber-physical system (HCPS) scenario of Industry 5.0. A feature fusion module embeds and stitches multi-dimensional semantics into a global state representation, providing a unified high-level decision-making basis for the actor-critic network. At a task scale of thousands, this method reduces decision response time by over 40% compared to existing deep reinforcement learning methods. It also supports real-time adjustments in dynamic environments, significantly improving the robustness and scalability of scheduling solutions, meeting the core requirements of intelligent manufacturing for efficient collaboration and real-time optimization.

[0145] Example 2

[0146] In this embodiment, a computer terminal device is provided, including:

[0147] one or more processors;

[0148] a memory, coupled to the processor, for storing one or more programs;

[0149] When the one or more programs are executed by the one or more processors, the one or more processors implement the methods in the above embodiments.

[0150] In this embodiment, a computer-readable storage medium is further provided, on which a computer program is stored. When the computer program is executed by a processor, the method in the above embodiment is implemented.

[0151] In this embodiment, an electronic device is further provided, including a memory and a processor. The memory stores a computer program, and the processor is configured to run the computer program to execute the method in the above embodiment.

[0152] The above program can be run in the processor, or it can be stored in the memory (or computer-readable medium), which includes permanent and non-permanent, removable and non-removable media and can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, read-only compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device.

[0153] These computer programs can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable device to implement the process. Figure 1 a process or multiple processes and / or boxes Figure 1 The steps of the functions specified in one or more blocks can be implemented by different modules corresponding to different steps.

[0154] like Figure 3 As shown, this embodiment provides such a device or system. The system is called a state representation modeling system based on a multi-domain graph attention network, and includes:

[0155] The operation node modeling module is used to update the operation node priority through the attention mechanism based on the operation node features, the features of the immediate predecessor and successor operation nodes, and the features of the associated machine nodes, and generate the operation semantic embedding;

[0156] The machine node modeling module is used to construct a machine subgraph through meta-paths based on machine node features, associated operation node features, and operation-machine arc features, and uses the attention mechanism to update machine node priorities and generate machine semantic embeddings;

[0157] The worker node modeling module is used to construct a worker subgraph based on worker node features, associated machine node features, and machine-worker arc features, update worker node priorities through the attention mechanism, and generate worker semantic embeddings;

[0158] Feature fusion module, which is used to concatenate the operation semantic embedding, machine semantic embedding, and worker semantic embedding to generate a global state embedding;

[0159] A scheduling decision module is configured to output a scheduling decision action based on the global state embedding through an actor-critic network, wherein the action includes an allocation pair of operation, machine, and worker.

[0160] As an implementation in this embodiment, the operation node modeling module includes:

[0161] An operation attention calculation unit, used to calculate the attention coefficient of an operation node and its adjacent operation nodes and associated machine nodes;

[0162] Operation feature aggregation unit, used to aggregate adjacent operation node features and machine node features based on the normalized attention coefficient;

[0163] The operational semantic embedding generation unit is used to capture the complex dependencies between operation nodes through a multi-head attention mechanism and generate operational semantic embeddings.

[0164] As an implementation method in this embodiment, the machine node modeling module includes:

[0165] A machine subgraph construction unit is used to associate machine nodes and operation nodes through meta-paths to generate an initial machine subgraph;

[0166] A machine feature stacking unit is used to linearly transform and stack the operation node features, machine node features, and operation-machine arc features to form machine subgraph node features;

[0167] The machine semantic embedding generation unit is used to calculate the attention coefficients of adjacent machine nodes based on the stacked node features, aggregate the node features after normalization, and generate machine semantic embeddings.

[0168] As an implementation method of this embodiment, the worker node modeling module includes:

[0169] A worker subgraph construction unit is used to associate worker nodes with machine nodes through meta-paths to generate an initial worker subgraph;

[0170] The worker feature stacking unit is used to linearly transform and stack the machine node features, worker node features, and machine-worker arc features to form worker subgraph node features;

[0171] The worker semantic embedding generation unit is used to calculate the attention coefficients of adjacent worker nodes based on the stacked node features, aggregate the node features after normalization, and generate the worker semantic embedding.

[0172] As an implementation method of this embodiment, the feature fusion module includes:

[0173] The global embedding concatenation unit is used to concatenate the operation semantic embedding, machine semantic embedding, and worker semantic embedding by dimension to generate a global state embedding.

[0174] A state value estimation unit is configured to embed the global state into the input critic network and output a value estimate of the current environment state.

[0175] As an implementation method of this embodiment, the scheduling decision module includes:

[0176] An action score generation unit, which is used to generate action scores through a multi-layer perceptron based on the operation node embedding, machine node embedding, worker node embedding, and global state embedding;

[0177] The action probability distribution unit is used to normalize the action scores and output the probability distribution of operation-machine-worker assignment pairs.

[0178] The system or device is used to implement the functions of the method in the above-mentioned embodiment. Each module in the system or device corresponds to each step in the method, which has been explained in the method and will not be repeated here.

[0179] Through the above implementation, the problem of state representation modeling based on multi-domain graph attention network in the related art is solved, thereby ensuring that the problems existing in the existing technology are solved.

[0180] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A state representation modeling method based on multi-domain graph attention network, characterized by: The following steps are involved: Based on the operation node features, the features of the immediate predecessor and successor operation nodes, and the features of the related machine nodes, the operation node priorities are updated through the attention mechanism to generate the operation semantic embedding; Based on the machine node features, associated operation node features, and operation-machine arc features, a machine subgraph is constructed through meta-paths. The machine node priorities are updated using the attention mechanism to generate machine semantic embeddings. Based on the features of worker nodes, associated machine nodes, and machine-worker arcs, a worker subgraph is constructed. The worker node priorities are updated through the attention mechanism to generate worker semantic embeddings. The operational semantic embedding, machine semantic embedding, and worker semantic embedding are feature-fused to generate a global state embedding. Based on the global state embedding, a scheduling decision action is output through an actor-critic network, where the action includes an assignment pair of operation, machine, and worker.

2. The method according to claim 1, characterized in that The process of updating the priority of the operation node includes: Calculate the attention coefficient of the operation node and its adjacent operation nodes, normalize and aggregate the features of the adjacent nodes; Calculate the attention coefficient of the operation node and the associated machine node, and aggregate the machine node features after normalization; The complex dependencies between operation nodes are captured through the multi-head attention mechanism to generate operation semantic embedding.

3. The method according to claim 1, characterized in that The process of constructing a machine subgraph includes: Generate an initial machine subgraph by associating machine nodes and operation nodes through meta-paths; The operation node features, machine node features, and operation-machine arc features are linearly transformed and stacked to form the machine subgraph node features; Based on the stacked node features, the attention coefficients of adjacent machine nodes are calculated, and the node features are aggregated after normalization to generate machine semantic embedding.

4. The method according to claim 1, wherein The process of constructing the worker subgraph includes: Generate the initial worker subgraph by associating worker nodes with machine nodes through meta-paths; Linearly transform and stack the machine node features, worker node features, and machine-worker arc features to form the worker subgraph node features; Based on the stacked node features, the attention coefficients of adjacent worker nodes are calculated, and the node features are aggregated after normalization to generate the worker semantic embedding.

5. The method according to claim 1, wherein The feature fusion process includes: Concatenate the operational semantic embedding, machine semantic embedding, and worker semantic embedding by dimension to form a global state embedding. The global state is embedded into the input critic network, which outputs a value estimate of the current environment state.

6. The method according to claim 1, characterized in that The action output process of the actor-critic network includes: Generate action scores through a multi-layer perceptron based on operation node embedding, machine node embedding, worker node embedding, and global state embedding; Normalize the action scores and output the probability distribution of operation-machine-worker assignment pairs.

7. A state representation modeling system based on a multi-domain graph attention network, characterized in that: The system comprises: The operation node modeling module is used to update the operation node priority through the attention mechanism based on the operation node features, the features of the immediate predecessor and successor operation nodes, and the features of the associated machine nodes, and generate the operation semantic embedding; The machine node modeling module is used to construct a machine subgraph through meta-paths based on machine node features, associated operation node features, and operation-machine arc features, and uses the attention mechanism to update machine node priorities and generate machine semantic embeddings; The worker node modeling module is used to construct a worker subgraph based on worker node features, associated machine node features, and machine-worker arc features, update worker node priorities through the attention mechanism, and generate worker semantic embeddings; Feature fusion module, which is used to concatenate the operation semantic embedding, machine semantic embedding, and worker semantic embedding to generate a global state embedding; A scheduling decision module is configured to output a scheduling decision action based on the global state embedding through an actor-critic network, wherein the action includes an allocation pair of operation, machine, and worker.

8. The system according to claim 7, characterized in that The operation node modeling module includes: An operation attention calculation unit, used to calculate the attention coefficient of an operation node and its adjacent operation nodes and associated machine nodes; Operation feature aggregation unit, used to aggregate adjacent operation node features and machine node features based on the normalized attention coefficient; The operational semantic embedding generation unit is used to capture the complex dependencies between operation nodes through a multi-head attention mechanism and generate operational semantic embeddings.

9. A computer terminal device, characterized in that: include: one or more processors; a memory, coupled to the processor, for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the state representation modeling method based on the multi-domain graph attention network as described in any one of claims 1 to 6.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the state representation modeling method based on a multi-domain graph attention network according to any one of claims 1 to 6 is implemented.

Images