A production logistics integrated scheduling method and related equipment applied to island production system

Abstract

The present application relates to a production logistics integrated scheduling method and related equipment applied to an island production system. The method includes: determining the process node set, workstation node set and workstation location node set respectively according to the process, workstation and workstation location information in the island production system, and then constructing a heterogeneous graph corresponding to the production logistics scheduling task; processing the heterogeneous graph based on a heterogeneous graph neural network model to obtain an embedding vector at the node level of the heterogeneous graph; using the embedded vector as input, outputting the production logistics scheduling result through the scheduling strategy network model, and realizing integrated scheduling of production logistics based on the production logistics scheduling result. This method improves scheduling efficiency while ensuring the effectiveness of scheduling, ensures the real-time and robustness of scheduling, and expands the effect of the applicable scenarios of the method.

Description

Technical Field

[0001] The present application relates to the technical field of logistics scheduling, and in particular to a production logistics integrated scheduling method and related equipment applied to an island production system. Background Art

[0002] The island production system is an advanced manufacturing model that emerged in the context of mass customization. It includes workstations (also called production cells) with configurable multi-tasking capabilities and automated guided vehicles (AGVs) for material transportation. The variability of its production structure and logic can adapt to changing product demand and production disturbances. The core of the island production system is to decouple production and logistics in space and reconnect them through AGVs to adapt to changing product demand and production disturbances.

[0003] However, the production logistics scheduling methods applied to island production systems in related technologies are mainly concentrated at the AGV scheduling level, that is, AGV scheduling is achieved through multi-constraint and multi-objective combined optimization. It does not consider the linkage between product multi-process routes, workstation task allocation and AGV scheduling, resulting in low production efficiency and waste of resources. Moreover, the scheduling method is not dynamic and real-time, and has poor adaptability to dynamic interference such as task changes and resource failures. It also has insufficient prediction of disturbances in the operation of island production lines, making it difficult to respond quickly to dynamic changes, resulting in increased waiting time in the production process.

[0004] With respect to the above-mentioned problems in the related technologies, no effective solution has been proposed yet. Summary of the invention

[0005] The embodiments of the present invention provide a production logistics integrated scheduling method and related equipment applied to an island production system, which at least solves the problems existing in the related technology of low scheduling efficiency, weak anti-interference ability, difficulty in quickly responding to dynamic changes, resulting in poor continuity and stability of production logistics scheduling, difficulty in realizing integrated scheduling of production logistics, and narrow applicable scenarios.

[0006] In order to solve the above problem, one aspect of an embodiment of the present invention provides a production logistics integrated scheduling method applied to an island production system, including:

[0007] According to the process, workstation and workstation location information in the island production system, the process node set, the workstation node set and the workstation location node set are determined respectively, and the heterogeneous graph corresponding to the production logistics scheduling task is constructed according to the process node set, the workstation node set and the workstation location node set;

[0008] The heterogeneous graph is processed based on the heterogeneous graph neural network model to obtain the embedding vector at the node level of the heterogeneous graph;

[0009] Taking the embedded vector as input, the production logistics scheduling result is output through the scheduling strategy network model, and the integrated scheduling of production logistics is realized based on the production logistics scheduling result.

[0010] In some of the embodiments, the step of processing the heterogeneous graph based on the heterogeneous graph neural network model to obtain an embedding vector at the heterogeneous graph node level includes:

[0011] The graph attention network is used to calculate the attention coefficient between the target node and its adjacent nodes in the heterogeneous graph, and a feature vector is generated for each target node according to the attention coefficient;

[0012] The feature vectors are processed based on the heterogeneous graph neural network layer to obtain a process embedding vector set and a workstation embedding vector set respectively; wherein the heterogeneous graph neural network layer is formed by sequentially stacking multiple neural network layers with the same structure and independent trainable parameters;

[0013] The process embedding vector set and the workstation embedding vector set are respectively subjected to mean pooling to obtain the embedding vectors at the node level of the heterogeneous graph.

[0014] In some embodiments, the step of taking the embedding vector as input and outputting the production logistics scheduling result via the scheduling strategy network model includes:

[0015] Taking the embedding vector as input, the scheduling strategy network model outputs the action probability distribution of multiple process-workstation pairs;

[0016] The proximal strategy optimization algorithm is used to optimize the action probability distribution to obtain the production logistics scheduling results.

[0017] In some embodiments, the step of optimizing the action probability distribution using the proximal strategy optimization algorithm to obtain the production logistics scheduling result further includes:

[0018] Based on the policy network in the reinforcement learning algorithm framework, the action probability distribution of multiple process-workstation pairs under a given state is output;

[0019] Apply the proximal policy optimization algorithm to adjust the parameters of the policy network to optimize the action probability distribution;

[0020] The expected returns of different process-workstation pairs are evaluated through the value function network in the reinforcement learning algorithm framework, and the production logistics scheduling results are determined based on the optimized action probability distribution and expected returns.

[0021] In some embodiments, the step of implementing integrated scheduling of production logistics based on the production logistics scheduling result includes:

[0022] According to the production logistics scheduling results, the target tasks corresponding to the workstations are dynamically allocated, and the transportation paths of the automatic guided vehicles are planned based on the target tasks corresponding to each workstation to achieve integrated scheduling of production logistics.

[0023] In some of the embodiments, the step of constructing a heterogeneous graph corresponding to a production logistics scheduling task includes:

[0024] Obtain the process node set, workstation node set and workstation location node set in the island production system, extract the connection information between the process and the workstation, and extract the status information of the workstation location;

[0025] According to the process node set, workstation node set, workstation location node set, connection information and status information, a heterogeneous graph corresponding to the production logistics scheduling task is constructed.

[0026] In some of the embodiments, the heterogeneous graph neural network model is built based on computing resources, data, and models integrated on a digital base.

[0027] In order to solve the above problems, one aspect of an embodiment of the present invention provides a production logistics integrated scheduling system applied to an island production system, including:

[0028] A heterogeneous graph construction unit is used to determine the process node set, the workstation node set and the workstation location node set respectively according to the process, the workstation and the workstation location information in the island production system, and to construct a heterogeneous graph corresponding to the production logistics scheduling task according to the process node set, the workstation node set and the workstation location node set;

[0029] A processing unit, used for processing the heterogeneous graph based on the heterogeneous graph neural network model to obtain an embedding vector at the node level of the heterogeneous graph;

[0030] The scheduling unit is used to take the embedded vector as input, output the production logistics scheduling result through the scheduling strategy network model, and realize the integrated scheduling of production logistics based on the production logistics scheduling result.

[0031] In order to solve the above-mentioned problems, one aspect of an embodiment of the present invention provides an electronic device, including: a processor, and a memory for storing a program, wherein the program includes instructions, and when the instructions are executed by the processor, the processor executes any one of the above-mentioned production logistics integrated scheduling methods applied to an island production system.

[0032] In order to solve the above problems, one aspect of an embodiment of the present invention provides a non-transitory machine-readable medium storing computer instructions, where the computer instructions are used to enable a computer to execute any of the above-mentioned production logistics integrated scheduling methods applied to an island production system.

[0033] The beneficial effects of the embodiments of the present invention are as follows: by respectively determining the process node set, workstation node set and workstation location node set according to the process, workstation and workstation location information in the island production system, and constructing a heterogeneous graph corresponding to the production logistics scheduling task according to the process node set, workstation node set and workstation location node set; processing the heterogeneous graph based on the heterogeneous graph neural network model to obtain an embedding vector at the node level of the heterogeneous graph; using the embedded vector as input, outputting the production logistics scheduling result through the scheduling strategy network model, and realizing the technical means of integrated scheduling of production logistics based on the production logistics scheduling result, the production logistics scheduling method of the related technology applied to the island production system is overcome because it is mainly focused on the AGV scheduling level, and does not consider the linkage relationship between the multi-process routes of the product, the task allocation of the workstation and the AGV scheduling, resulting in low production efficiency and waste of resources. The scheduling method has weak dynamic real-time performance, poor adaptability to dynamic interferences such as task changes and resource failures, insufficient prediction of disturbances in the operation of the island production line, and difficulty in responding quickly to dynamic changes, which leads to increased waiting time in the production process. The paper constructs a heterogeneous graph through static information (process node set, workstation node set) and dynamic information (workstation location node set) in the island production system, and then uses heterogeneous graph neural network processing and integrated scheduling strategy network model to output production logistics scheduling results, so that the determined production logistics scheduling results can better adapt to the actual needs (such as task changes) and dynamic changes (such as resource failures) in the production process, thereby supporting more flexible and real-time production logistics scheduling, and achieving the technical effect of improving scheduling efficiency on the basis of ensuring scheduling effectiveness, ensuring scheduling real-time and robustness, and expanding the applicable scenarios of the method.

[0034] The details of one or more embodiments of the invention are set forth in the following drawings and description so that other features, objects, and advantages of the invention are more readily apparent. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other embodiments can be obtained based on these drawings without creative work.

[0036] Figure 1 It is a main flow chart of a production logistics integrated scheduling method applied to an island production system according to an embodiment of the present invention.

[0037] Figure 2 It is a schematic diagram of executing a scheduling scheme based on a heterogeneous graph neural network according to an embodiment of the present invention.

[0038] Figure 3 It is a schematic diagram of a heterogeneous graph showing the connection relationship among processes, workstations, and workstation locations according to an embodiment of the present invention.

[0039] Figure 4 It is a schematic diagram of the architecture of a heterogeneous graph neural network according to an embodiment of the present invention.

[0040] Figure 5 It is a main framework diagram of a production logistics integrated scheduling system applied to an island production system according to an embodiment of the present invention.

[0041] Figure 6 It is a schematic structural diagram of the electronic device of the present invention. DETAILED DESCRIPTION

[0042] Embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be construed as being limited to the embodiments described herein, which are instead provided for a more thorough and complete understanding of the present invention. It should be understood that the drawings and embodiments of the present invention are only for exemplary purposes and are not intended to limit the scope of protection of the present invention.

[0043] The core of the island production system is to decouple production and logistics in space and reconnect them through AGV to adapt to changing product demand and production disturbances. This production model can flexibly respond to personalized customized orders and meet the diverse needs of different product specifications, colors, functions, etc., so it has been widely used in modern manufacturing, especially in the fields of automobile manufacturing and electronic equipment production.

[0044] The production logistics scheduling methods used in island production systems in related technologies are mainly concentrated at the AGV scheduling level, using the following common solution models and algorithms:

[0045] (1) Exact solution algorithm: enumeration method based on disjunctive graph model or solution method based on operations research, such as integer programming, Lagrangian relaxation method, branch and bound method, etc. These methods can obtain the global optimal solution, but are only applicable to simple small-scale scheduling problems. When faced with complex large-scale AGV scheduling problems, there are problems such as large solution space and computational difficulties, making it difficult to quickly obtain an effective scheduling solution in actual production.

[0046] (2) Heuristic algorithms based on scheduling rules: This type of algorithm is based on intuition or expert experience and is composed of a combination of several priority rules and heuristics. It is easy to implement and has low time complexity, and can quickly generate feasible scheduling solutions. However, the quality of its solution is average, and it is difficult to ensure high-quality scheduling results in a complex and changing production environment, and its adaptability to dynamic interference is poor.

[0047] (3) Meta-heuristic algorithm: Adding meta-factors to the heuristic algorithm enables the algorithm to adjust the underlying logic according to the optimization progress. It has a certain degree of randomness and helps to escape the local optimum. Common meta-heuristic algorithms include genetic algorithm, taboo search algorithm, simulated annealing algorithm, particle swarm algorithm and artificial bee colony algorithm. Although these algorithms have improved the quality of scheduling solutions to a certain extent, they generally have the following problems: the setting of hyperparameters lacks a theoretical basis, the initialization of the solution has a great influence on the optimization results, and encoding and decoding are required according to specific problems. In addition, meta-heuristic algorithms are essentially built to solve static problems. Whenever dynamic events occur, a large amount of repeated calculations are required. The iterative search process is very time-consuming. Few algorithms can use historical information to learn prior knowledge and quickly adjust their search behavior. It is difficult to migrate the results to a new manufacturing environment and lacks generalization ability.

[0048] (4) Deep reinforcement learning algorithm: As an unsupervised artificial intelligence algorithm, its core is to build an intelligent agent and enable it to learn in an unknown environment, and learn strategies to maximize cumulative rewards or minimize cumulative costs in the process of interacting with the environment. When facing static problems, this algorithm can obtain high-quality solutions in a relatively short time, and it strikes a good balance between solution quality and solution time cost. When facing dynamic scheduling problems, the trained intelligent agent can select appropriate actions for efficient scheduling based on the current state of the system in a very short time, which has significant advantages over traditional optimization algorithms.

[0049] Although the above technologies have been widely used in the production logistics scheduling of island production systems, there are still the following problems to be solved:

[0050] Lack of collaborative optimization: The scheduling methods in related technologies are mainly focused on the AGV scheduling level, and do not fully consider the linkage between the multi-process routes of products, the task allocation of workstations and AGV scheduling. In actual production, the task allocation of workstations, the scheduling and path planning of AGVs are an interrelated whole. Only focusing on AGV scheduling cannot achieve the global optimization of production logistics, which may lead to problems such as low production efficiency and waste of resources.

[0051] Insufficient dynamic real-time performance: The dynamic real-time performance of existing scheduling methods is not strong, and they are poorly adaptable to dynamic disturbances such as task changes and resource failures. In the mass customization production mode, dynamic disturbances such as frequent changes in production tasks and equipment failures are the norm, and the existing scheduling methods are difficult to respond quickly to these dynamic changes and cannot adjust the scheduling plan in time, resulting in increased waiting time in the production process and reduced production efficiency.

[0052] Weak disturbance prediction capability: Insufficient prediction of disturbances during the operation of the island production line makes it impossible to take effective preventive measures in advance, which further aggravates the uncertainty and instability in the production process and affects the continuity and stability of production.

[0053] In summary, relevant technologies have problems such as insufficient collaborative optimization, poor dynamic real-time performance, and weak disturbance prediction capabilities in the production logistics scheduling of island production systems, which makes it difficult to meet the high requirements of modern manufacturing for production efficiency, flexibility, and stability.

[0054] In order to solve the above problems, an embodiment of the present invention provides a production logistics integrated scheduling method applied to an island production system, such as Figure 1 As shown, the production logistics integrated scheduling method applied to the island production system mainly includes:

[0055] Step S101, determine the process node set, workstation node set and workstation location node set respectively according to the process, workstation and workstation location information in the island production system, and construct a heterogeneous graph corresponding to the production logistics scheduling task according to the process node set, workstation node set and workstation location node set.

[0056] The process node set provided by the embodiment of the present invention refers to a set of specific processes or steps that need to be executed in the island production system. Each process node contains detailed information about the process, such as the required time, resources, and dependencies between previous and next nodes. The workstation node set provided by the embodiment of the present invention includes a set of workstation nodes such as workstations, machines, or equipment that execute the process. The workstation nodes reflect the available resources and their functions in the production system. The workstation location node set provided by the embodiment of the present invention represents a set of specific locations of workstations in physical space. This location information helps to plan material handling routes, calculate handling time, and optimize logistics processes.

[0057] Through the above settings, a heterogeneous graph is constructed based on the process node set, workstation node set and workstation location node set in the island production system, so that the heterogeneous graph not only contains static information in the production system (such as processes, workstations), but also considers dynamic information (such as location and task status changes), providing an intuitive and efficient decision-making basis for subsequent production logistics scheduling, so that the subsequent scheduling strategy can better adapt to the actual needs (such as task changes) and dynamic changes (such as resource failures) in the production process, thereby supporting more flexible and real-time production logistics scheduling and improving scheduling efficiency.

[0058] In some of the embodiments, the step of constructing a heterogeneous graph corresponding to the production logistics scheduling task includes: obtaining the process node set, workstation node set and workstation location node set in the island production system, and extracting the connection information between the process and the workstation, and extracting the status information of the workstation location; constructing a heterogeneous graph corresponding to the production logistics scheduling task based on the process node set, workstation node set, workstation location node set, connection information and status information.

[0059] After obtaining several node sets of the island production system, the connection information between the process and the workstation and the status information of the workstation location are extracted, and then a heterogeneous graph that can reflect the physical and logical structure of the production system is constructed based on several node sets, connection information and status information, so that the heterogeneous graph can represent the dependency relationship between processes and the components of resources.

[0060] The connection information between processes and workstations and the status information of workstation locations are extracted to construct a heterogeneous graph topology model that reflects the physical and logical structure of the production system. This model can accurately represent the dependencies between processes and the layout of resources.

[0061] Step S102, processing the heterogeneous graph based on the heterogeneous graph neural network model to obtain an embedding vector at the node level of the heterogeneous graph.

[0062] Among them, the embedding vector provided by the embodiment of the present invention is a vector representation that maps nodes (such as processes, workstations, etc.) in the graph structure to a continuous vector space. Through the embedding vector, the nodes in the graph structure are mapped to a low-dimensional continuous vector space. This vector not only contains the original feature information of the node, but also contains the structural information of the node in the graph, such as adjacency, etc. The embedding vector can capture the topological structure information and attribute information of the node, and is used to support machine learning tasks on the graph, such as classification, regression or action prediction.

[0063] Based on the above settings, through the processing of heterogeneous graph neural networks, it is possible to automatically extract useful features from complex production logistics scheduling tasks and generate embedding vectors that can reflect node status and relationships, providing high-quality input for subsequent scheduling decisions, thereby improving the performance and adaptability of the entire production logistics integrated scheduling method.

[0064] In some of the embodiments, the step of processing the heterogeneous graph based on the heterogeneous graph neural network model to obtain the embedding vector at the node level of the heterogeneous graph includes: using the graph attention network to calculate the attention coefficient between the target node and its adjacent nodes in the heterogeneous graph, and generating a feature vector for each target node based on the attention coefficient; processing the feature vector based on the heterogeneous graph neural network layer to obtain a process embedding vector set and a workstation embedding vector set, respectively; wherein the heterogeneous graph neural network layer is formed by sequentially stacking multiple neural network layers with the same structure and independent trainable parameters; and performing mean pooling processing on the process embedding vector set and the workstation embedding vector set, respectively, to obtain the embedding vector at the heterogeneous graph node level.

[0065] The feature vector provided in the embodiment of the present invention refers to a vector representation learned from the original features of the node (such as process type, workstation capability, etc.) through a graph attention network or other feature extraction technology. The feature vector contains not only the original feature information of the node, but also the structural information of the node in the graph, such as the neighborhood information of the node.

[0066] According to an embodiment of the present invention, by utilizing a graph attention network mechanism, it is possible to identify and emphasize the important relationship between each target node and its adjacent nodes in a heterogeneous graph, thereby enhancing the feature representation capability; by generating a feature vector for each target node, the subsequent heterogeneous graph neural network model is able to distinguish the unique attributes of different nodes and their roles in the production process; the heterogeneous graph neural network layer optimizes the learning process of the embedding vector by stacking multiple neural network layers with the same structure but independent trainable parameters, and then obtains a set of process embedding vectors and a set of workstation embedding vectors by processing the feature vectors, thereby simplifying the information that needs to be considered in the decision-making process, saving computing resources, and improving scheduling efficiency.

[0067] In the above steps, multiple embedding vectors are simplified into a single representation vector through mean pooling, which helps reduce the risk of overfitting and enhance the generalization ability of the model. The node-level embedding vector obtained in the heterogeneous graph integrates rich feature information of processes and workstations, so it can support complex production logistics scheduling decision-making and help to allocate production tasks more reasonably.

[0068] According to an embodiment of the present invention, the heterogeneous graph neural network can extract various types of relationships between nodes (such as process dependency, resource allocation, etc.), so that the embedding vector can reflect the complex interactions in the production system, and can also handle the dynamic changes of the heterogeneous graph, such as the addition or deletion of nodes and edges, so that the embedding vector can adapt to the real-time changes in the production process and provide support for dynamic scheduling. At the same time, the above embedding vector is used as the input of the scheduling strategy network model. The subsequent scheduling strategy network model can more accurately predict the effects of various actions in production logistics scheduling through the learned embedding vector, thereby improving the accuracy and efficiency of scheduling decisions.

[0069] In some of the embodiments, the heterogeneous graph neural network model is constructed based on computing resources, data and models integrated on a digital base.

[0070] The digital base mentioned above usually refers to an integrated platform or system that provides the necessary computing resources, data storage, model management, and analysis tools to support the development and operation of various applications and services. In industrial and production environments, the digital base can include cloud computing resources, big data processing capabilities, Internet of Things (IoT) connections, artificial intelligence (AI) models, and other advanced technologies to achieve digitalization and intelligence of the production process.

[0071] Based on the above settings, the computing resources provided by the digital base can be used to efficiently run complex graph neural network models, optimize the use of computing resources, and speed up the model training and inference process; the data processing capabilities of the digital base can be used to process and analyze large amounts of production data to provide high-quality input features for graph neural networks; the digital base can also centrally manage different model versions and instances to simplify the model deployment, update, and maintenance process.

[0072] The digital base provides a strong foundation for building and operating complex production systems, enabling companies to more effectively use data and models to optimize production processes, improve production efficiency and quality, while maintaining the flexibility and scalability of the system.

[0073] Step S103, using the embedded vector as input, outputting the production logistics scheduling result through the scheduling strategy network model, and realizing the integrated scheduling of production logistics based on the production logistics scheduling result.

[0074] Based on the above settings, by converting the embedded vectors into actual scheduling decisions and realizing the integrated scheduling of production logistics based on these decisions, the production process is optimized, resources are rationally allocated and production efficiency is improved. At the same time, the flexibility of the integrated scheduling system of production logistics and its ability to respond to dynamic changes are also enhanced.

[0075] In some of the embodiments, the step of taking the embedded vector as input and outputting the production logistics scheduling result via the scheduling strategy network model includes: taking the embedded vector as input and outputting the action probability distribution of multiple process-workstation pairs via the scheduling strategy network model; optimizing the action probability distribution using the proximal strategy optimization algorithm to obtain the production logistics scheduling result.

[0076] Through the action probability distribution output by the scheduling strategy network model, the optimal or near-optimal action can be accurately selected to achieve efficient scheduling of production logistics. Furthermore, the action probability distribution optimized by the proximal strategy optimization algorithm can be used to arrange the production process more reasonably, reduce waiting time and improve production efficiency. Based on the above steps, through accurate scheduling decisions (obtained production logistics scheduling results), production resources can be used more effectively, idle time can be reduced and overall resource utilization can be improved.

[0077] In some of the embodiments, the step of optimizing the action probability distribution using the proximal policy optimization algorithm to obtain the production logistics scheduling result also includes: outputting the action probability distribution of multiple process-workstation pairs under a given state based on the policy network in the reinforcement learning algorithm framework; adjusting the parameters of the policy network using the proximal policy optimization algorithm to optimize the action probability distribution; evaluating the expected returns of different process-workstation pairs through the value function network in the reinforcement learning algorithm framework, and determining the production logistics scheduling result based on the optimized action probability distribution and expected return.

[0078] Among them, the reinforcement learning algorithm framework provided by the embodiment of the present invention can be an Actor-Critic structure, which combines the Policy Gradient method and the Value Function Estimation method. In this structure, the "Actor" (i.e., the above-mentioned policy network) is responsible for selecting actions according to the current policy, while the "Critic" (i.e., the above-mentioned value function network) is responsible for evaluating the value of the current state or action (the expected value of the reward). Specifically, in reinforcement learning, the Actor is a policy network that outputs the probability distribution of taking each possible action in a given state, that is, the Actor is responsible for the decision-making process, that is, deciding which action should be taken in a specific state; the Critic is a value function network that evaluates the quality of the current state or action, usually expressed as the expected return of the state or action (i.e., the cumulative reward). By adopting the reinforcement learning algorithm framework of the Actor-Critic structure, the Actor and the Critic collaborate with each other to improve learning efficiency and performance by sharing certain parameters or information. This collaborative optimization means that the Actor's strategy update will take into account the Critic's evaluation of the value.

[0079] It should be pointed out that in addition to the above-mentioned Actor-Critic structure, other existing reinforcement learning algorithm frameworks can also be adopted, such as Deep Q-Network (DQN): DQN directly learns the action value function, that is, the expected return of each action, and then selects the action with the greatest value. It uses experience replay and target network to stabilize the learning process. Another example is Asynchronous Advantage Actor-Critic (A3C for short): A3C is an extension of Actor-Critic, which uses multiple Critic networks to reduce the variance of estimates and optimize the learning process. These algorithms and structures are all designed to solve decision-making problems in reinforcement learning, balancing exploration and exploitation in different ways, and handling the update of strategies and value functions.

[0080] Based on the above settings, the action probability distribution output by the policy network can identify the most likely action in a specific state, which helps decision makers select the most appropriate process-workstation pair for scheduling; the PPO (Proximal Policy Optimization) algorithm limits the range of policy updates to prevent the policy network from experiencing drastic fluctuations during training, thereby improving the stability of the policy. By optimizing the action probability distribution, the production process can be more effectively guided, unnecessary waiting and handling can be reduced, and production efficiency can be improved.

[0081] According to an embodiment of the present invention, the above reinforcement learning algorithm framework enables the policy network to adapt to dynamic changes in the production process, such as task changes and resource failures, thereby achieving dynamic scheduling. And evaluating the expected returns of different process-workstation pairs through the value function network (Critic network) helps to optimize resource allocation and ensure that key resources are reasonably allocated to where they are most needed. Optimized production processes and resource allocation can reduce production costs, including reducing energy consumption, reducing material waste and reducing equipment idle time.

[0082] In some of the embodiments, the above-mentioned step of realizing integrated scheduling of production logistics based on the production logistics scheduling results includes: dynamically allocating target tasks corresponding to workstations according to the production logistics scheduling results, and planning the transportation path of the automatic guided vehicle based on the target tasks corresponding to each workstation, so as to realize integrated scheduling of production logistics.

[0083] According to the real-time production logistics scheduling results, target tasks can be dynamically assigned to each workstation, so that the workstation can perform the most relevant process according to the current production needs and resource conditions. Based on the target tasks of each workstation, the AGV's handling path is automatically planned, which helps to reduce material handling time, optimize material flow, and reduce transportation costs. Through dynamic task allocation and effective path planning, a rapid response to production changes is achieved, the waiting time between processes is reduced, the production process is accelerated, and the flexibility and adaptability of the production system are improved, enabling it to adapt to changing market needs.

[0084] Based on the above settings, by realizing dynamic integrated scheduling of production logistics, not only the production efficiency and resource utilization are improved, but also the flexibility and robustness of the system are enhanced, thus providing enterprises with significant operational advantages in a highly competitive market environment.

[0085] The above-mentioned production logistics integrated scheduling method applied to the island production system provided by the embodiment of the present invention adopts the method of determining the process node set, workstation node set and workstation location node set respectively according to the process, workstation and workstation location information in the island production system, and constructing the heterogeneous graph corresponding to the production logistics scheduling task according to the process node set, workstation node set and workstation location node set; processing the heterogeneous graph based on the heterogeneous graph neural network model to obtain the node-level embedding vector; using the embedded vector as input, outputting the production logistics scheduling result through the scheduling strategy network model, and realizing the technical means of integrated scheduling of production logistics based on the production logistics scheduling result. It realizes the construction of a heterogeneous graph through static information (process node set, workstation node set) and dynamic information (workstation location node set) in the island production system, and then outputting the production logistics scheduling result by using the heterogeneous graph neural network processing and the integrated scheduling strategy network model, so that the determined production logistics scheduling result can better adapt to the actual needs (such as task changes) and dynamic changes (such as resource failures) in the production process, thereby supporting more flexible and real-time production logistics scheduling, and achieving the technical effect of improving scheduling efficiency on the basis of ensuring scheduling effectiveness, ensuring scheduling real-time and robustness, and expanding the applicable scenarios of the method.

[0086] The production logistics integrated scheduling method for an island production system provided in an embodiment of the present invention is to conceptualize the solution process of the production logistics integrated scheduling problem of the island production system as a sequential decision-making procedure. This process involves a series of scheduling processes, iteratively assigning processes to compatible workstations (also called production units) in each state, and determining workstations at suitable locations until all workstations and processes are scheduled. The workflow of the proposed method is as follows: Figure 2 As shown in Figure 1, in each iteration, the scheduling state is first converted into a heterogeneous graph structure. Subsequently, a heterogeneous graph neural network with a three-stage embedding process is used. This network is used to extract feature embeddings of processes, workstations, and workstation locations. These embedding information is provided as input to the decision network to generate a probability distribution of actions, and the scheduling actions (i.e., the above-mentioned production logistics scheduling results) are extracted from this probability distribution.

[0087] Combined with the above description, Figure 2 The execution steps of the heterogeneous graph neural network in are as follows:

[0088] (1) Representation of process, workstation and position: The left side of the figure shows the representation of process, workstation and position. Among them, the process consists of multiple steps, the workstation is the location where the process is executed, and the position is the specific coordinates of the workstation.

[0089] (2) Construction of heterogeneous graph: processes, workstations, and locations are connected through a heterogeneous graph structure. Figure 2 The connection relationship between the process (S1, S2, S3) and the workstation (W1, W2, W3, W4, W5), as well as the connection between the workstation and the position (P_C, P_E) are shown.

[0090] (3) Three stages of embedding update: Stage 1: Process embedding update, that is, first update the embedding representation of the process and convert the process information into feature embedding. Stage 2: Workstation embedding update: Then update the embedding representation of the workstation and convert the workstation information into feature embedding. Stage 3: Position embedding update: Finally, update the embedding representation of the position and convert the position information into feature embedding.

[0091] (4) Input of the scheduling strategy network model: After three stages of embedding updates, the feature embeddings of the process, workstation, and location are input into the decision network.

[0092] (5) PPO Actor-Critic Algorithm: Use the Proximal Policy Optimization (PPO) Actor-Critic algorithm for decision making. The policy network generates a probability distribution of actions and selects actions through distribution and sampling.

[0093] (6) Environment transition (or task change): Based on the selected action, the environment is transformed, including generating rewards, moving to the next location, and updating state information.

[0094] Through the above steps, the heterogeneous graph neural network architecture can effectively represent the features and make decisions on the production logistics scheduling problems in the island production system, thereby realizing dynamic integrated scheduling.

[0095] in, Figure 2The Chinese equivalents of different English words are as follows: Process: process; Work station: workstation; Positions: position; Start: start; End: end; Stage1: Process embedding update: stage 1: process embedding update; Stage2: W-Station embedding update: stage 2: workstation embedding update; Stage3: Position embedding update: stage 3: position embedding update; PPO Actor-Critic: proximal policy optimization (PPO) Actor-Critic; Policy Network: policy network; Distribution & sampling: distribution and sampling; Actions: action; Environment transition: environment transition; Generate reward: generate reward; Move to the next position: move to the next position; Update state information: update state information. Heterogeneous Graph: heterogeneous graph; Heterogeneous GNN: heterogeneous graph neural network model; Decision Making: decision process.

[0096] Figure 3 A schematic diagram of a heterogeneous graph showing the connection relationship among processes, workstations, and workstation locations according to an embodiment of the present invention is shown. Figure 3 The heterogeneous graph (H) shown in the figure has several advantages over the disjunctive graph (G) of the flexible job shop in the related art: (1) The graph density of the heterogeneous graph is significantly reduced. (2) The workstation nodes in the heterogeneous graph H provide a convenient way to inject workstation information, and can also extract useful features to distinguish workstations in different states. This type of information is very important for solving the flexible job shop scheduling problem, because the processes need to be assigned to the appropriate workstations, which is difficult to obtain in the disjunctive graph G of the related art. (3) By simply converting the processing time p ijk Attached as a feature to PW arc E ijk The processing time can be easily indicated on the

[0097] Based on the above definition of the heterogeneous graph H, the embodiment of the present invention converts each state Represented as a heterogeneous graph ; Among them, P represents the process node set (Process), that is, the individual steps or operations in the production process. W represents the workstation node set (Work station), which refers to the specific location or machine in the workstation where the process can be performed. S represents the status of the process or workstation (Status), which can be an attribute or feature related to a specific process or workstation. C represents the combination arc set (Combination Arcs), which may refer to the sequential relationship or dependency relationship between processes. represents the set of process-workstation relationships at a specific time step t. These relationships may change over time because production scheduling decisions may change which processes are assigned to which workstations. Indicates the state change of a workstation or process at a specific time step t, which may include the busy or idle state of the workstation, the completion status of the process, etc. Dynamically changes during the solution process. Specifically, the action is performed at step t. After that, only keep arc, The other PW arcs of are removed to obtain Therefore, the adjacency relationship between nodes will also change dynamically. In each step t, let For process adjacent workstations, For workstations adjacent process.

[0098] A workstation node set W is added to the heterogeneous graph provided in the embodiment of the present invention, and each node corresponds to a workstation ,This is because that a process can be processed by multiple workstations, and the ,processing time of processes on different compatible workstations is different, which makes it difficult to ,represent. Through the above setting, useful features can be better ,extracted to distinguish workstations in different states.

[0099] Specifically, product process nodes and original characteristics ( )R 4 The corresponding status information includes: whether the current product process has been assigned [0,1], the number of unassigned product processes; the number of optional workstations; the number of workstations for the process;

[0100] Workstation nodes and raw properties ( )R 3 The corresponding status information includes: whether the current workstation has been allocated [0,1]; the optional location of the workstation; the processable product steps; the allocated product steps and the corresponding processing time.

[0101] Workstation location nodes and raw properties ( )R 2The corresponding status information includes: the assigned workstation; the location (coordinates, length, width and area can also be taken into account).

[0102] Analysis arc (PWS) and original characteristics R 1 The corresponding information includes: the physical distance between workstations.

[0103] According to an embodiment of the present invention, a specific implementation method of node embedding is also provided. Workstation A neighbor is a process A collection of these processes may have different meanings. For example, processes that are expected to start earlier may be more important than those that start later. This motivates us to consider using graph attention networks, which can automatically learn the importance of different nodes by applying graph attention networks. For a homogeneous graph, given a graph with features For the target node i, the graph attention network first calculates the attention coefficient between the target node i and each adjacent node j in its first-order neighborhood N(i) (including the target node i itself). (scalar).

[0104] The above embedding process can be viewed as a HGNN layer, where the original features of each process and station are and are converted into embedding vectors and In order to enhance the feature extraction capability, L HGNN layers with the same structure but independent trainable parameters are stacked sequentially (e.g. Figure 4 The HGNN architecture diagram shown in Figure 1 stacks L neural network layers in sequence to obtain the final embedding vector and In the initial layer, only the original features of the process and workstation nodes are used. and , and the original characteristics of the PW arc Utilized in all L layers.

[0105] After L layers of HGNN, mean pooling is applied to the generated process embedding vector set and workstation embedding vector set respectively, and then the two d-dimensional vectors are concatenated to form an embedded representation of the heterogeneous graph state Ht.

[0106] According to an embodiment of the present invention, a specific implementation method for determining a production logistics scheduling result is also provided:

[0107] (1) First, the scheduling decision process: At each decision step t (time is 0 or when a process is completed), the agent observes the current system state , and make decisions (also referred to as production actions, an action represents a feasible process-workstation pair) , that is, an unscheduled process (process step) is assigned to an idle workstation and starts execution from the current time T(t). Subsequently, the environment is transferred to the next decision step t+1. This process is repeated until all processes are scheduled. The Markov decision process (MDP) can be applied in the decision process, where the applied Markov decision process is defined as follows:

[0108] State: At step t, the states of all processes and workstations constitute the state Initial state is an instance of the flexible job shop scheduling problem drawn from a distribution. It should be noted that for each , will maintain a partial schedule , which is calculated as follows: If Scheduled, with a start time is the actual value ;otherwise, is an estimate based only on precedence constraints. Already on the workstation is scheduled, then ,otherwise, .in, It can be calculated recursively for all unscheduled processes from the starting point to the end point.

[0109] Action: The embodiment of the present invention adopts a comprehensive approach to solve the job shop scheduling problem, treating process selection and workstation allocation as a composite decision. Specifically, the action at step t Defined as a feasible process-workstation pair ,in is available (i.e. its immediate predecessor has completed), and ∈ is idle. Will immediately appear on the workstation Start running on Action Set It is step-dependent and contains all feasible process-workstation pairs. Each job can have at most one process ready at any time.

[0110] State transfer: based on the current state and actions , the environment deterministically transitions to a new state , that is, the time point when a certain process is completed. In this scheme, the two different states are distinguished by the topology and characteristics of the corresponding heterogeneous graph structure.

[0111] Generate reward (expected return): The reward is defined as the part scheduled in state and The maximum completion time difference.

[0112] When formulating scheduling strategies, deep reinforcement learning is used to optimize production logistics scheduling results with the goal of maximizing the cumulative rewards of the plan.

[0113] This solution can use two strategies to solve specific problem instances and obtain production logistics scheduling results when applying deep reinforcement learning algorithms. The first strategy is to greedily select the action with the highest probability, and the second strategy is to sample actions in each state (unlike the determinism of the greedy strategy, the solutions generated by the sampling strategy may vary in different runs due to their inherent randomness. This randomness introduces additional possibilities for discovering better solutions, which is particularly important because current deep reinforcement learning algorithms often converge to suboptimal strategies). For the sampling strategy, Ns solutions can be obtained by solving Ns instances of a problem, and the best one is selected as the final production logistics scheduling result.

[0114] The present invention effectively solves the dynamic response capability and real-time decision-making problems of production logistics scheduling in island production systems by adopting heterogeneous graph neural networks and deep reinforcement learning algorithms, and has at least the following technical effects: (1) Collaborative optimization: It realizes the real-time collaborative generation and decision-making of product process route selection, workstation task allocation, AGV scheduling and path planning, thereby improving production efficiency and resource utilization. (2) Dynamic real-time performance: The deep reinforcement learning algorithm can quickly respond to dynamic interference such as task changes and resource failures, adjust the scheduling plan in time, and reduce the waiting time in the production process. (3) Disturbance prediction: By generating training data sets through discrete event simulation, the adaptability and generalization ability of the model to dynamic interference are improved, and it can predict and respond to disturbances in the production process in advance, thereby enhancing the continuity and stability of production.

[0115] Based on the above-mentioned production logistics integrated scheduling method applied to the island production system provided by the embodiment of the present invention, the embodiment of the present invention also provides a production logistics integrated scheduling system applied to the island production system, such as Figure 5 As shown, the production logistics integrated scheduling system 500 applied to the island production system includes:

[0116] The heterogeneous graph construction unit 501 is used to determine the process node set, workstation node set and workstation location node set according to the process, workstation and workstation location information in the island production system, and to construct a heterogeneous graph corresponding to the production logistics scheduling task according to the process node set, workstation node set and workstation location node set.

[0117] Through the above settings, a heterogeneous graph is constructed based on the process node set, workstation node set and workstation location node set in the island production system, so that the heterogeneous graph not only contains static information in the production system (such as processes, workstations), but also considers dynamic information (such as location and task status changes), providing an intuitive and efficient decision-making basis for subsequent production logistics scheduling, so that the subsequent scheduling strategy can better adapt to the actual needs (such as task changes) and dynamic changes (such as resource failures) in the production process, thereby supporting more flexible and real-time production logistics scheduling and improving scheduling efficiency.

[0118] In some of the embodiments, the heterogeneous graph construction unit 501 is also used to: obtain the process node set, workstation node set and workstation location node set in the island production system, and extract the connection information between the process and the workstation, and extract the status information of the workstation location; construct a heterogeneous graph corresponding to the production logistics scheduling task based on the process node set, workstation node set, workstation location node set, connection information and status information.

[0119] After obtaining several node sets of the island production system, the connection information between the process and the workstation and the status information of the workstation location are extracted, and a heterogeneous graph topology model reflecting the physical and logical structure of the production system is constructed. This model can accurately represent the dependencies between processes and the layout of resources.

[0120] The processing unit 502 is used to process the heterogeneous graph based on the heterogeneous graph neural network model to obtain an embedding vector at the node level of the heterogeneous graph.

[0121] Based on the above settings, through the processing of heterogeneous graph neural networks, it is possible to automatically extract useful features from complex production logistics scheduling tasks and generate embedding vectors that can reflect node status and relationships, providing high-quality input for subsequent scheduling decisions, thereby improving the performance and adaptability of the entire production logistics integrated scheduling method.

[0122] In some of the embodiments, the processing unit 502 is further used to: calculate the attention coefficient between the target node and its adjacent nodes in the heterogeneous graph using a graph attention network, and generate a feature vector for each target node based on the attention coefficient; process the feature vector based on the heterogeneous graph neural network layer to obtain a process embedding vector set and a workstation embedding vector set, respectively; wherein the heterogeneous graph neural network layer is formed by sequentially stacking a plurality of neural network layers with the same structure and independent trainable parameters; and perform mean pooling on the process embedding vector set and the workstation embedding vector set, respectively, to obtain heterogeneous graph node-level embedding vectors.

[0123] According to an embodiment of the present invention, a graph attention network is used to identify and emphasize the important relationship between each target node and its adjacent nodes in a heterogeneous graph, thereby enhancing the feature representation capability; by generating a feature vector for each target node, a subsequent heterogeneous graph neural network model can distinguish the unique attributes of different nodes and their roles in the production process; the heterogeneous graph neural network layer optimizes the learning process of the embedding vector by stacking multiple neural network layers with the same structure but independent trainable parameters, and then obtains a set of process embedding vectors and a set of workstation embedding vectors by processing the feature vectors, thereby simplifying the information that needs to be considered in the decision-making process, saving computing resources, and improving scheduling efficiency.

[0124] In some of the embodiments, the heterogeneous graph neural network model is constructed based on computing resources, data and models integrated on a digital base.

[0125] Based on the above settings, the computing resources provided by the digital base can be used to efficiently run complex graph neural network models, optimize the use of computing resources, and speed up the model training and inference process; the data processing capabilities of the digital base can be used to process and analyze large amounts of production data to provide high-quality input features for graph neural networks; the digital base can also centrally manage different model versions and instances to simplify the model deployment, update, and maintenance process.

[0126] The scheduling unit 503 is used to take the embedded vector as input, output the production logistics scheduling result through the scheduling strategy network model, and realize the integrated scheduling of production logistics based on the production logistics scheduling result.

[0127] Based on the above settings, by converting the embedded vectors into actual scheduling decisions and realizing the integrated scheduling of production logistics based on these decisions, the production process is optimized, resources are rationally allocated and production efficiency is improved. At the same time, the flexibility of the integrated scheduling system of production logistics and its ability to respond to dynamic changes are also enhanced.

[0128] In some of the embodiments, the scheduling unit 503 is also used to: take the embedded vector as input, output the action probability distribution of multiple process-workstation pairs through the scheduling strategy network model; and optimize the action probability distribution using the proximal strategy optimization algorithm to obtain the production logistics scheduling result.

[0129] Through the action probability distribution output by the scheduling strategy network model, the optimal or near-optimal action can be accurately selected to achieve efficient scheduling of production logistics. Furthermore, the action probability distribution optimized by the proximal strategy optimization algorithm can be used to arrange the production process more reasonably, reduce waiting time and improve production efficiency. Based on the above steps, through accurate scheduling decisions (obtained production logistics scheduling results), production resources can be used more effectively, idle time can be reduced and overall resource utilization can be improved.

[0130] In some of the embodiments, the scheduling unit 503 is also used to: output the action probability distribution of multiple process-workstation pairs under a given state based on the policy network in the reinforcement learning algorithm framework; adjust the parameters of the policy network using the proximal policy optimization algorithm to optimize the action probability distribution; evaluate the expected returns of different process-workstation pairs through the value function network in the reinforcement learning algorithm framework, and determine the production logistics scheduling results based on the optimized action probability distribution and expected returns.

[0131] Based on the above settings, the action probability distribution output by the policy network can identify the most likely action in a specific state, which helps decision makers select the most appropriate process-workstation pair for scheduling; the PPO (Proximal Policy Optimization) algorithm limits the range of policy updates to prevent the policy network from experiencing drastic fluctuations during training, thereby improving the stability of the policy. By optimizing the action probability distribution, the production process can be more effectively guided, unnecessary waiting and handling can be reduced, and production efficiency can be improved.

[0132] In some of the embodiments, the above-mentioned step of realizing integrated scheduling of production logistics based on the production logistics scheduling results includes: dynamically allocating target tasks corresponding to workstations according to the production logistics scheduling results, and planning the transportation path of the automatic guided vehicle based on the target tasks corresponding to each workstation, so as to realize integrated scheduling of production logistics.

[0133] According to the real-time production logistics scheduling results, target tasks can be dynamically assigned to each workstation, so that the workstation can perform the most relevant process according to the current production needs and resource conditions. Based on the target tasks of each workstation, the AGV's handling path is automatically planned, which helps to reduce material handling time, optimize material flow, and reduce transportation costs. Through dynamic task allocation and effective path planning, a rapid response to production changes is achieved, the waiting time between processes is reduced, the production process is accelerated, and the flexibility and adaptability of the production system are improved, enabling it to adapt to changing market needs.

[0134] Based on the above settings, by realizing dynamic integrated scheduling of production logistics, not only the production efficiency and resource utilization are improved, but also the flexibility and robustness of the system are enhanced, thus providing enterprises with significant operational advantages in a highly competitive market environment.

[0135] The above-mentioned production logistics integrated scheduling system applied to the island production system provided by the embodiment of the present invention, because the heterogeneous graph construction unit is used to determine the process node set, the workstation node set and the workstation location node set according to the process, workstation and workstation location information in the island production system, and construct a heterogeneous graph corresponding to the production logistics scheduling task according to the process node set, the workstation node set and the workstation location node set; the processing unit is used to process the heterogeneous graph based on the heterogeneous graph neural network model to obtain the node-level embedding vector; the scheduling unit is used to take the embedding vector as input, output the production logistics scheduling result through the scheduling strategy network model, and The flow scheduling results realize the integrated scheduling of production logistics, and realize the construction of a heterogeneous graph through static information (process node set, workstation node set) and dynamic information (workstation location node set) in the island production system, and then use the heterogeneous graph neural network processing and the integrated scheduling strategy network model to output the production logistics scheduling results, so that the determined production logistics scheduling results can better adapt to the actual needs (such as task changes) and dynamic changes (such as resource failures) in the production process, thereby supporting more flexible and real-time production logistics scheduling, and achieving the technical effect of improving scheduling efficiency on the basis of ensuring scheduling effectiveness, ensuring scheduling real-time and robustness, and expanding the applicable scenarios of the method.

[0136] An embodiment of the present invention further provides a non-transitory machine-readable medium storing a computer program, wherein the computer program, when executed by a processor of a computer, is used to cause the computer to execute the method of the embodiment of the present invention.

[0137] The embodiment of the present invention further provides a computer program product, including a computer program, wherein the computer program is used to make the computer perform the method of the embodiment of the present invention when executed by a processor of the computer. The computer program product should be understood as a software product that mainly implements the above method of the present application through a computer program.

[0138] An embodiment of the present invention further provides an electronic device, comprising: at least one processor; and a memory in communication with the at least one processor. The memory stores a computer program executable by the at least one processor, and when the computer program is executed by the at least one processor, the electronic device executes the method of the embodiment of the present invention.

[0139] refer to Figure 6 , a block diagram of an electronic device that can be used as a server or client of an embodiment of the present invention will now be described, which is an example of a hardware device that can be applied to various aspects of the present invention. The electronic device is intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workbenches, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0140] like Figure 6 As shown, the electronic device includes a computing unit 601, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 602 or a computer program loaded from a storage unit 608 to a random access memory (RAM) 603. In the RAM 603, various programs and data required for the operation of the electronic device can also be stored. The computing unit 601, the ROM 602, and the RAM 603 are connected to each other via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0141] Multiple components in the electronic device are connected to the I / O interface 605, including: an input unit 606, an output unit 607, a storage unit 608, and a communication unit 609. The input unit 606 can be any type of device that can input information to the electronic device, and the input unit 606 can receive input digital or character information, and generate key signal input related to user settings and / or function control of the electronic device. The output unit 607 can be any type of device that can present information, and can include but is not limited to a display, a speaker, a video / audio output terminal, a vibrator, and / or a printer. The storage unit 608 can include but is not limited to a disk, an optical disk. The communication unit 609 allows the electronic device to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks, and can include but is not limited to a modem, a network card, an infrared communication device, and / or a wireless communication transceiver, such as a Bluetooth device, a WiFi device, a WiMax device, a cellular communication device, and / or the like.

[0142] The computing unit 601 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 601 include, but are not limited to, a CPU, a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing units, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 601 performs the various methods and processes described above. For example, in some embodiments, the method embodiments of the present invention may be implemented as a computer program, which is tangibly contained in a machine-readable medium, such as a storage unit 608. In some embodiments, part or all of the computer program may be loaded and / or installed on an electronic device via a ROM 602 and / or a communication unit 609. In some embodiments, the computing unit 601 may be configured to perform the above-described method in any other appropriate manner (e.g., by means of firmware).

[0143] The computer programs for implementing the methods of the embodiments of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer programs are executed by the processor or controller, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer programs may be executed entirely on the machine, partially on the machine, partially on the machine as a stand-alone software package and partially on a remote machine, or entirely on a remote machine or server.

[0144] In the context of an embodiment of the present invention, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or apparatus. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable signal medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, or infrared systems, devices, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media may include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0145] It should be noted that the term "including" and its variations used in the embodiments of the present invention are open inclusions, that is, "including but not limited to". The term "based on" means "based at least in part on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one other embodiment"; the term "some embodiments" means "at least some embodiments". The modifications of "one" and "multiple" mentioned in the embodiments of the present invention are illustrative rather than restrictive. Those skilled in the art should understand that unless the context clearly indicates otherwise, it should be understood as "one or more".

[0146] The information / data involved in the embodiments of the present invention (including but not limited to information / data used for analysis, information / data stored, information / data displayed, etc.) are all information / data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant information / data must comply with relevant laws, regulations and standards of relevant countries and regions, and corresponding operation entrances are provided for users to choose to authorize or refuse.

[0147] The various steps described in the method implementation methods provided in the embodiments of the present invention may be performed in different orders and / or in parallel. In addition, the method implementation methods may include additional steps and / or omit the steps shown. The scope of protection of the present invention is not limited in this respect.

[0148] The term "embodiment" in this specification refers to specific features, structures or characteristics described in conjunction with the embodiment that can be included in at least one embodiment of the present invention. The appearance of this phrase in various places in the specification does not necessarily mean the same embodiment, nor does it mean that it is mutually exclusive with other embodiments and is independent or optional. The various embodiments in this specification are described in a related manner, and the same or similar parts between the various embodiments refer to each other. In particular, for the device, equipment, and system embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts refer to the partial description of the method embodiment.

[0149] The above-mentioned embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of protection. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the attached claims.

Claims

1. A production logistics integrated scheduling method applied to an island production system, characterized in that: include: According to the processes, workstations and workstation location information in the island production system, a process node set, a workstation node set and a workstation location node set are determined respectively, and a heterogeneous graph corresponding to the production logistics scheduling task is constructed according to the process node set, the workstation node set and the workstation location node set; The heterogeneous graph is processed based on a heterogeneous graph neural network model to obtain an embedding vector at the heterogeneous graph node level; wherein the step of processing the heterogeneous graph based on a heterogeneous graph neural network model to obtain an embedding vector at the heterogeneous graph node level includes: using a graph attention network to calculate an attention coefficient between a target node and its adjacent nodes in the heterogeneous graph, and generating a feature vector for each target node according to the attention coefficient; processing the feature vector based on a heterogeneous graph neural network layer to obtain a process embedding vector set and a workstation embedding vector set respectively; wherein the heterogeneous graph neural network layer is formed by sequentially stacking a plurality of neural network layers with the same structure and independent trainable parameters; performing mean pooling processing on the process embedding vector set and the workstation embedding vector set respectively to obtain an embedding vector at the heterogeneous graph node level; Taking the embedding vector as input, outputting the production logistics scheduling result through the scheduling strategy network model, and realizing the integrated scheduling of production logistics based on the production logistics scheduling result; wherein, taking the embedding vector as input, outputting the production logistics scheduling result through the scheduling strategy network model comprises: taking the embedding vector as input, outputting the action probability distribution of multiple process-workstation pairs through the scheduling strategy network model; optimizing the action probability distribution using the proximal strategy optimization algorithm to obtain the production logistics scheduling result; The step of optimizing the action probability distribution using the proximal policy optimization algorithm to obtain the production logistics scheduling result also includes: based on the policy network in the reinforcement learning algorithm framework, outputting the action probability distribution of multiple process-workstation pairs under a given state; applying the proximal policy optimization algorithm to adjust the parameters of the policy network to optimize the action probability distribution; evaluating the expected returns of different process-workstation pairs through the value function network in the reinforcement learning algorithm framework, and determining the production logistics scheduling result based on the optimized action probability distribution and the expected return.

2. The method according to claim 1, characterized in that The step of implementing integrated scheduling of production logistics based on the production logistics scheduling result comprises: The target tasks corresponding to the workstations are dynamically allocated according to the production logistics scheduling results, and the transportation paths of the automatic guided vehicles are planned based on the target tasks corresponding to each workstation, so as to realize the integrated scheduling of production logistics.

3. The method according to claim 1, characterized in that The step of constructing a heterogeneous graph corresponding to the production logistics scheduling task includes: Acquire the process node set, workstation node set and workstation location node set in the island production system, extract the connection information between the process and the workstation, and extract the status information of the workstation location; A heterogeneous graph corresponding to the production logistics scheduling task is constructed based on the process node set, the workstation node set, the workstation location node set, the connection information and the status information.

4. The method according to claim 1, characterized in that: The heterogeneous graph neural network model is constructed based on the computing resources, data and models integrated in the digital base.

5. A production logistics integrated scheduling system applied to an island production system, characterized in that: include: A heterogeneous graph construction unit is used to determine the process node set, the workstation node set and the workstation location node set respectively according to the process, the workstation and the workstation location information in the island production system, and to construct a heterogeneous graph corresponding to the production logistics scheduling task according to the process node set, the workstation node set and the workstation location node set; A processing unit is used to process the heterogeneous graph based on a heterogeneous graph neural network model to obtain an embedding vector at the heterogeneous graph node level; the processing unit is also used to: use a graph attention network to calculate the attention coefficient between a target node and its adjacent nodes in the heterogeneous graph, and generate a feature vector for each target node according to the attention coefficient; process the feature vector based on a heterogeneous graph neural network layer to obtain a process embedding vector set and a workstation embedding vector set respectively; wherein the heterogeneous graph neural network layer is formed by sequentially stacking a plurality of neural network layers with the same structure and independent trainable parameters; perform mean pooling processing on the process embedding vector set and the workstation embedding vector set respectively to obtain an embedding vector at the heterogeneous graph node level; A scheduling unit is used to take the embedded vector as input, output a production logistics scheduling result through a scheduling strategy network model, and realize integrated scheduling of production logistics based on the production logistics scheduling result; the scheduling unit is also used to: take the embedded vector as input, output the action probability distribution of multiple process-workstation pairs through the scheduling strategy network model; optimize the action probability distribution using a proximal strategy optimization algorithm to obtain the production logistics scheduling result; The scheduling unit is also used to: output the action probability distribution of multiple process-workstation pairs under a given state based on the policy network in the reinforcement learning algorithm framework; apply the proximal policy optimization algorithm to adjust the parameters of the policy network to optimize the action probability distribution; evaluate the expected returns of different process-workstation pairs through the value function network in the reinforcement learning algorithm framework, and determine the production logistics scheduling result based on the optimized action probability distribution and the expected return.

6. An electronic device comprising: A processor, and a memory storing a program, wherein the program comprises instructions, which, when executed by the processor, cause the processor to perform the method according to any one of claims 1 to 4.

7. A non-transitory machine-readable medium storing computer instructions, characterized in that: The computer instructions are used to cause the computer to execute the method according to any one of claims 1-4.

Images