3C intelligent manufacturing workshop production line scheduling method and system considering line side bin capacity

By using heterogeneous graph model and reinforcement learning algorithm production line scheduling method in the 3C intelligent manufacturing workshop, the upstream blocking problem caused by insufficient line edge warehouse capacity is solved, efficient and real-time scheduling is achieved, and local optimal solutions are avoided. Improved production efficiency and resource utilization.

CN118966698APending Publication Date: 2024-11-15XI AN JIAOTONG UNIV
View PDF 0 Cites 7 Cited by

Patent Information

Application Number
CN202411086383.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-08
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

The existing 3C intelligent manufacturing workshop scheduling method cannot effectively deal with the upstream processing line blockage problem in real time when the product capacity of the line side warehouse is insufficient, resulting in the impact of production efficiency and continuity, and it is easy to fall into the local optimal solution.

Method used

Using the production line scheduling method based on heterogeneous graph model and reinforcement learning algorithm, the heterogeneous graph model for task production scheduling is constructed, and high-dimensional features are extracted using the graph attention mechanism, and combined with the deductor-criticist reinforcement learning algorithm, in real time Assign tasks, monitor and adjust production sequences to avoid blockage.

Benefits of technology

With limited online side warehouse capacity, efficient and real-time production line scheduling is achieved, local optimal solutions are avoided, production resource utilization and delivery time completion rate are improved, and scheduling accuracy and flexibility are enhanced.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118966698A_ABST
    Figure CN118966698A_ABST
Patent Text Reader

Abstract

The invention discloses a 3C intelligent manufacturing workshop production line scheduling method and system considering line side bin capacity. The production line scheduling method comprises the following steps: a workshop scheduling center firstly extracts environment information of a workshop, and constructs a heterogeneous graph model architecture of task production scheduling; according to workshop production information, utilizing a graph attention mechanism method to perform graph feature embedding, and further extracting high-dimensional features of a processing order, a processing line body and a line edge bin buffer area; performing average pooling according to the process node features, the line body node features and the connection edge features to obtain heterogeneous graph model global features, and allocating tasks to be processed to specified processing line bodies by using a deducer-commentator reinforcement learning algorithm; the processing line body starts to execute the distributed task after receiving the distributed task, interacts with an actual production environment, and uploads production information to a workshop scheduling center; the workshop dispatching center monitors the workshop machining state in time through the information of the machining line body and the machining order, and real-time dispatching of workshop tasks is achieved; the system can realize the 3C intelligent manufacturing workshop production line scheduling method. The system comprises a demand layer, a material layer, a scheduling layer, a management layer and a monitoring layer. According to the method, global scheduling of the 3C intelligent manufacturing workshop production line is realized, and the problems of parallel production of multiple processing line bodies, blockage of an upstream processing line body caused by insufficient product capacity of a line side bin and the like are solved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of workshop scheduling, and in particular relates to a production line scheduling method and system for a 3C smart manufacturing workshop that takes into account the capacity of line-side warehouses. Background Art

[0002] With the rapid development of the internet and big data, the 3C manufacturing industry is gradually transforming and upgrading toward a digital, intelligent, networked, and flexible production model. As a core component of 3C smart manufacturing workshops, the production scheduling module is a key component in improving overall factory production efficiency. In recent years, the widespread application of advanced intelligent scheduling technologies has significantly improved the efficiency and accuracy of workshop scheduling.

[0003] Common methods for solving workshop scheduling problems include heuristic algorithms and metaheuristic algorithms. Heuristic algorithms involve workshop schedulers using a single scheduling rule or a combination of multiple scheduling rules to solve problems. While they offer advantages in terms of simplicity, they target a single problem, are less capable of solving large-scale workshop scheduling problems, and are subject to complex processes and inability to achieve optimal solutions. Metaheuristic algorithms involve workshop scheduling centers using intelligent algorithms to solve the optimal order processing sequence for each processing line. Compared to heuristic algorithms, metaheuristic algorithms can generate approximate solutions within a certain timeframe, but they suffer from lower real-time and generalization capabilities.

[0004] Currently, 3C smart production workshops primarily use heuristic scheduling based on scheduling rules. Patent application number CN202311425372.4 discloses a permutation flow shop scheduling method based on the actor-critic algorithm. The method includes: generating an initial solution for permutation flow shop scheduling using a priority scheduling rule; expressing the initial solution using a complete binary tree to obtain an initial binary tree representing the initial solution, an intermediate binary tree representing the initial solution's evolutionary process, and a final binary tree representing the optimal solution; designing states, actions, rewards, and the interaction process between the agent and the environment based on the initial, intermediate, and final binary trees; designing a policy network and a value network; representing the actor in the actor-critic algorithm using the policy network and the critic in the actor-critic algorithm using the value network; and training the actor-critic algorithm to obtain a feasible solution that is superior to the initial solution. However, in most 3C smart manufacturing workshops, there are line-side warehouses for products between the processing lines. Once the upstream processing lines are blocked and stagnant due to the number of products reaching a critical value, the heuristic scheduling method based on scheduling rules lacks real-time performance and cannot handle conflicts in a timely and effective manner, which will affect the transportation efficiency and production continuity of the entire factory. Summary of the Invention

[0005] In response to the above-mentioned deficiencies in the existing technology, the present invention proposes a production line scheduling method and system for a 3C smart manufacturing workshop that takes into account the capacity of the line-side warehouse. It realizes the reasonable scheduling of processing orders and processing lines under the condition of considering the limited capacity of the work-in-progress in the line-side warehouse, and solves the problems of upstream processing line blockage caused by insufficient capacity of the work-in-progress in the line-side warehouse in real time and efficiently. It has the function of effectively overcoming the problems of poor real-time performance and easy falling into local optimal solutions existing in the heuristic scheduling method based on scheduling rules.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] A production line scheduling method for a 3C smart manufacturing workshop considering the capacity of line-side warehouses includes the following steps:

[0008] Step S1: Based on the actual production situation of the 3C smart manufacturing workshop, the scheduling center of the scheduling layer in the scheduling system extracts the initial information of the workshop environment through the demand layer and the monitoring layer, including information on processing orders, processing lines, and line warehouses, and constructs a heterogeneous graph model for task scheduling. The heterogeneous graph model includes the process nodes of the processing orders, the processing line nodes, and the connection arc constraint relationship information, which is stored in the graph model module;

[0009] Step S2: In the scheduling layer, the workshop scheduling center performs a graph embedding operation on the heterogeneous graph model obtained in step S1, and uses the graph attention mechanism to perform weighted summation on the features within the neighborhood to extract the high-dimensional features of the process nodes of the processing order, the high-dimensional features of the processing line nodes, and the high-dimensional features of the connection arcs. The above three high-dimensional features are average-pooled to obtain the global features of the heterogeneous graph model;

[0010] Step S3: In the scheduling layer, the workshop scheduling center uses the actor-critic reinforcement learning algorithm to assign the tasks to be processed to the designated processing lines based on the heterogeneous graph model obtained in step S1 and the high-dimensional features of the process nodes, line nodes, and connection arcs obtained in step S2, as well as the global features of the heterogeneous graph.

[0011] Step S4: Based on the task assigned to the designated processing line in step S3, the task execution phase begins. Meanwhile, the task execution phase interacts with the workshop work environment. The workshop scheduling center receives production information from the processing line and order processing information from the scheduling system management layer.

[0012] Step S5: In the monitoring layer and management layer, the workshop scheduling center monitors the workshop processing status in a timely manner through the processing line and order processing information in step S4, and realizes real-time scheduling of workshop tasks.

[0013] The step S1 is specifically as follows:

[0014] The workshop scheduling center obtains the initial information of the tasks that need to be processed at each scheduling time point, including the order type, the number of order processes, and the order processing quantity; at the same time, the workshop scheduling center obtains the initial information of the workshop environment before starting processing, including information on the tasks to be produced, the occupancy of the production line, and the work-in-progress capacity of the line warehouse; based on the initial information, a heterogeneous graph model for production scheduling is constructed, including the process nodes of the task, the processing line nodes, the directed arcs between the process nodes, and the undirected arcs between the machine and the process nodes.

[0015] The characteristic information of the process node in step S2 includes:

[0016] (A) Order processing status, which is 1 if the order has been completed and 0 if it has not been completed;

[0017] (B) Order processing process, the current process node corresponds to the process number of the processing order;

[0018] (C) Order completion rate, the completion rate of the order corresponding to the current process, as shown in formula (1);

[0019]

[0020] Where s i,o is the j completed processes for the i-th order;

[0021] S i is the total number of processes required for the i-th order;

[0022] (D) The number of adjacent nodes, which is the sum of the number of process nodes and processing line nodes that are one unit length away from the current process node;

[0023] (E) The start time of the current process. The start time of the current process is shown in formula (2). If there is no congestion due to the capacity of the line side warehouse, the start time is the maximum of the end time of the previous process of the current order and the end time of the current processing line. If there is congestion, the start time is determined by the time when the congestion is released, that is, the time when the downstream section generates a new processing task:

[0024]

[0025] Where: ET i,j-1 is the end time of the j-1th process of the i-th order;

[0026] ME j,p is the end time of the pth parallel processing line in the jth process;

[0027] x is whether the jth process is blocked, 0 if blocked and 1 if not;

[0028] MSj+1,a is the start time of the ath parallel processing line in the j+1th process;

[0029] (F) The end time of the current process, as shown in formula (3), the end time of the current process is the sum of the start time and processing time of the parallel processing line;

[0030]

[0031] Where: ET i,j-1 is the end time of the j-1th process of the i-th order;

[0032] MS j,a is the starting time of the ath parallel processing line in the jth process;

[0033] A i,j is the processing quantity of the jth process of the i-th order;

[0034] A i,j is the hourly output of the ath parallel processing line in the jth process.

[0035] The characteristic information of the processing line node in step S2 includes:

[0036] (1) The status of the processing line: if the processing line has started processing the order, it is 1; if it has not started, it is 0;

[0037] (2) Utilization rate of the processing line, as shown in formula (4), the utilization rate of the processing line is determined by the completion time of the processing time;

[0038]

[0039] Where: U j,a is the utilization rate of the a-th parallel processing line in the j-th process;

[0040] P j,a is the total processing time of the a-th parallel processing line in the j-th process;

[0041] C j,a is the total completion time of the a-th parallel processing line in the j-th process;

[0042] (3) The number of adjacent process nodes, which is the number of process nodes with a unit length distance from the current processing line node;

[0043] The characteristic information of the connection arcs in step S2 includes: characteristic information of directed arcs between process nodes and characteristic information of undirected arcs between process nodes and processing line nodes;

[0044] The characteristic information of the directed arcs between process nodes includes the status of the line-side warehouse. If the capacity of the line-side warehouse has not reached the warning capacity, it is 0, and if it has reached the warning capacity, it is 1; the characteristic information of the directed arcs also includes the number of work-in-progress in the line-side warehouse, which is determined by the upstream and downstream processing lines; the characteristic information of the undirected arcs of the process nodes and the processing line nodes refers to the processing time; the characteristic information of the directed arcs of the process nodes and refers to the processing quantity.

[0045] The process of step S2 is as follows: the workshop scheduling center uses a multi-linear layer perceptron with different training parameters to perform feature dimension upgrade on the feature information of the order's process nodes, processing line nodes and each connection arc, extracts high-dimensional feature information, and then uses the graph attention mechanism to embed the features of the process nodes, processing line nodes and connection arcs into local graphs; the feature graph embedding of the processing line node includes two parts, namely, the weighted sum of the feature embeddings of all process nodes in its first-order neighborhood and the product of its attention coefficient, and the product of the feature embedding of the processing line node itself and its attention coefficient; the feature graph embedding of the process node includes the weighted sum of the feature embeddings of all process nodes in its first-order neighborhood and the product of its attention coefficient. The product of the features of all process nodes, all machine node features, and the features of the edges connecting to the next process point is used; the feature graph embedding of the connecting arc contains the weighted sum of the feature attention of the line nodes and process nodes connected to it, so as to extract the high-dimensional features of the process nodes and the high-dimensional features of the processing line nodes of the processing order; the workshop scheduling center uses the average pooling method to solve the global graph embedding features, and inputs the obtained order process node features, processing line node features and the features of their connecting edges into a neural network with the same multi-layer structure but different parameters for deep feature extraction and average pooling, and finally splices them to obtain the global features of the heterogeneous graph model with a fixed dimension;

[0046] Since the heterogeneous graph model constructed based on the 3C production model obtained in step S1 will be continuously updated and changed with the order scheduling process, when the j-th process of the i-th order selects processing line a, that is, the remaining parallel processing lines in the j-th process cannot be processed, the connection arc with the corresponding process node needs to be deleted.

[0047] The step S3 is specifically as follows:

[0048] The workshop scheduling center observes the overall state of the system, including high-dimensional features of process nodes such as processing order process progress and processing order mask; high-dimensional features of line nodes such as line state and line efficiency; high-dimensional features of connection arcs such as processing time; and global feature information of heterogeneous graphs. It selects actions in the action space containing valid processing lines and processing order process connection arcs, and assigns processing orders to designated processing lines for processing. When the processing line executes the current processing process, the workshop scheduling center interacts with the production environment to calculate the reward function, including whether the delivery period is met and the utilization rate of the processing line is maximized. When selecting actions, the workshop scheduling center uses the normal distribution sampling method to select. After inputting the observed production state into the deducer algorithm in the deducer-critic reinforcement learning algorithm, the score of each valid action is obtained. Then, after simultaneously inputting the state and action into the critic algorithm in the deducer-critic reinforcement learning algorithm, the score of the current state and action is obtained. The workshop scheduling center performs normal distribution sampling based on the weighted and normalized action scores and reasonably assigns each processing order to the processing line.

[0049] The step S4 is specifically as follows:

[0050] The workshop scheduling center schedules the processes of all processing orders according to the method in step S3. The processing line starts working after receiving the order and interacts with the workshop environment. Any processing line blockage caused by processing line failure or reaching the warning capacity of the line warehouse is transmitted between the processing line and the workshop scheduling center. At the same time, the workshop scheduling center monitors the production status of the current order in real time and adjusts the processing sequence of the production order according to possible dynamic events.

[0051] The step S5 is specifically as follows:

[0052] After receiving the information sent by the processing line, the workshop scheduling center handles the emergency situation. If a processing line failure occurs, the workshop scheduling center sets the processing line node mask of the corresponding number to 0, indicating that the line is unavailable. The connection arc corresponding to the line will not be selected in the subsequent scheduling of the current process, and the task on the line will be handed over to other parallel machines of the current process for production, minimizing the impact on the scheduling target; if the line side warehouse is blocked, the workshop scheduling center suspends the production of all parallel machines in the upstream process of the line side warehouse, that is, sets the process mask of the upstream process to 0, until the capacity of the line side warehouse is lower than the warning capacity and then sets it to 1; if an emergency order is encountered, the workshop scheduling center adds the process node corresponding to the new order, the processing line node and connection arc that can produce the corresponding process in the heterogeneous graph model at that time point, updates the global graph embedding features of the heterogeneous graph, and gives priority to scheduling the order at that time point.

[0053] A production line scheduling system for 3C smart manufacturing workshops that takes into account the capacity of line-side warehouses is used to implement the above scheduling method, including the demand layer, material layer, scheduling layer, management layer, and monitoring layer:

[0054] The demand layer obtains workshop processing orders based on production order requirements and supply chain information, which is used to create original information and generate initial plans for the job scheduling module of the workshop scheduling system;

[0055] At the material layer, the workshop warehousing system provides the scheduling layer with the warehousing information of the processed raw materials and the warehousing information of the work-in-progress in each line warehouse according to the type of demand order, and obtains the quantity of raw materials and work-in-progress in real time;

[0056] The scheduling layer uses reinforcement learning algorithms to allocate production lines to multiple processes of multiple processing orders based on the original information and material information of the demand orders, combined with the current production situation in the workshop, to achieve optimal production scheduling under limited line and side warehouse capacity;

[0057] In the management layer, after the scheduling center generates and successfully issues the production schedule, the processing information of each order is managed in real time by the operation management layer, which also handles events such as emergency order insertion in the workshop.

[0058] The monitoring layer mentioned above has various dynamic events in the actual production process, including production line failures, workshop resource shortages, etc. The monitoring layer monitors the various dynamic events occurring in the workshop in real time, and feeds back to the task scheduling layer through the workshop Internet of Things for active scheduling processing.

[0059] Compared with the prior art, the present invention has the following beneficial effects and advantages:

[0060] 1. The present invention organically combines the dispatching center, warehousing system, and processing machines in the 3C smart manufacturing workshop, and utilizes the interaction between the workshop dispatching center and the actual production environment to transmit information about processing orders and processing lines in real time, so that the processing process of the processing orders can be completed efficiently within the delivery period. This production line scheduling method based on heterogeneous graph models and reinforcement learning algorithms facilitates the workshop dispatching center to integrate and extract feature information from a large amount of complex data resources in the actual production environment, thereby quickly and accurately finding the optimal scheduling solution. At the same time, it is not easy to fall into local optimal solutions, and can ensure that the result obtained is the global optimal solution, which leads to waste of workshop production resources or overdue processing orders. The production layout of 3C smart manufacturing workshops is relatively complex and there are many equipment. The production line scheduling problem cannot be accurately and comprehensively expressed using mathematical polynomials. Heuristic algorithms based on scheduling rules and meta-heuristic algorithms based on intelligent algorithms cannot accurately obtain the optimal results of production line scheduling. The production line scheduling method proposed in the present invention can well solve this problem. It utilizes the graphical expression of the workshop scheduling process by heterogeneous graph models, the ability of the graph attention mechanism method to extract feature information, and the strong generalization ability of the reinforcement learning algorithm to improve the real-time and accuracy of production line scheduling.

[0061] 2. In step S1 of the present invention, the workshop scheduling center extracts feature information from the actual production environment by constructing a heterogeneous graph model, which can avoid the data processing burden caused by the independent transmission of different types of information and reduce the difficulty of data processing; at the same time, it can clearly show whether the processing line is occupied, the production status of the processing order, and the storage volume of the work-in-progress in the line warehouse, so as to select actions and design reward functions when arranging scheduling tasks; it can describe the actual status of order processing in a relatively comprehensive and complete manner, so that the final scheduling result can be closer to the optimization goal and more in line with reality.

[0062] 3. The workshop scheduling center of the present invention presents the scheduling process in the form of a heterogeneous graph model, and stores and transmits it in a graphical manner. While efficiently and quickly extracting feature information, it avoids receiving and screening a large amount of different types of data, saves data calculation time, and realizes lightweight data storage and precise data processing, and also provides a clear data sample for the subsequent allocation of scheduling tasks.

[0063] 4. The present invention sets different task priorities and optimization goals for the global scheduling of the production line, which can further improve the flexibility of workshop production and processing, and is conducive to the workshop scheduling center to better achieve optimal scheduling according to the different actual needs of the workshop.

[0064] 5. The present invention utilizes the graph attention mechanism method to perform high-dimensional feature extraction, splicing and other operations on heterogeneous graph models based on actual production, which can ensure the independent transmission of three-dimensional information of processing orders, processing lines and line-side warehouse capacity. At the same time, the graph attention mechanism method has strong end-to-end learning capabilities and can learn key information based on the changing heterogeneous graph model. The workshop scheduling center can better obtain the characteristics of each element and the global characteristics in the heterogeneous graph model, which is conducive to reinforcement learning to be more in line with the scheduling goals when selecting actions.

[0065] 6. After the workshop scheduling center of the present invention selects the corresponding action based on the current production status, it calculates the real-time reward obtained by the action, and then interacts with the environment to obtain the next state. The current reward is related to the action and production status selected in the future. Therefore, the reinforcement learning algorithm will estimate the reward obtained by the future action, thereby accurately estimating the current action, making the obtained scheduling result more accurate; after multiple training and learning, when encountering new scheduling tasks, the reinforcement learning scheduling algorithm can quickly and accurately find the optimal solution and has good generalization.

[0066] 7. The present invention sets the processing order mask, processing line mask and process mask operation of the processing order, so as to avoid the situations such as reduced processing line utilization and long processing order retention time caused by machine failure, line warehouse blockage or resource shortage. By dynamically changing the node information of the heterogeneous graph model, the processing orders after the dynamic event time point are rescheduled; through a large amount of data training, the scheduling algorithm has good real-time and high robustness for the results of solving dynamic events; by adding the position of the process node in the original heterogeneous graph model to solve the problems such as emergency insertion of orders, the impact of emergencies on order processing can be effectively reduced.

[0067] 8. When the present invention handles emergencies such as processing line failures and emergency insertion orders, the workshop dispatching center needs to change the scheduling tasks according to the actual production situation, and thus needs the status information of the workshop production environment; the traditional collection method is that the workshop dispatching center receives the change information of the processing line or processing order, and then obtains the status data of the global workshop for rescheduling, which will cause a large delay in the issuance of the rescheduling results; the present invention uses parallel interaction between the workshop dispatching center, the processing line and the actual processing environment. When dynamic events occur, information such as actual production data can be uploaded to the workshop dispatching center in parallel, which can improve the accuracy of information transmission and reduce the capacity of a single piece of information, making it easier for the workshop dispatching center to receive it; at the same time, it can avoid the workshop dispatching center collecting scattered workshop status data, which makes it too burdened.

[0068] 9. The present invention uses the work-in-progress storage situation of the line-side warehouse as the weight between the process node connection arcs of the processing order, which makes it easier for the neural network to extract and identify the impact of the line-side warehouse capacity on the processing order, and also ensures that the obtained scheduling results can meet the conditions that the line-side warehouse does not cause blockage, thereby improving the utilization rate of the processing line.

[0069] In summary, the present invention realizes the global scheduling of the production line of the 3C smart manufacturing workshop, and solves the problems of parallel production of multiple processing lines and blockage of upstream processing lines due to insufficient work-in-progress capacity in line-side warehouses. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 This is a diagram of the production line scheduling system architecture of the present invention.

[0071] Figure 2 Schematic diagram of the heterogeneous graph model of the present invention.

[0072] Figure 3 This is the layout and production flow chart of the 3C smart manufacturing workshop of the present invention.

[0073] Figure 4 This is a flowchart of the process node feature diagram embedded in the present invention.

[0074] Figure 5 This is a flow chart for embedding a processing line node feature graph of the present invention.

[0075] Figure 6 Schematic diagram of the heterogeneous graph model updating process of the present invention.

[0076] Figure 7 This is a flowchart of the action selection of the actor-critic reinforcement learning algorithm of the present invention.

[0077] Figure 8 The flowchart is updated for the start time of the processing order process and the processing line of the present invention.

[0078] Figure 9 Flowchart of production line scheduling using the actor-critic reinforcement learning algorithm of the present invention.

[0079] Figure 10 This is a flowchart of the parameter update of the proximal optimization algorithm of the present invention.

[0080] Figure 11 This is a flow chart of the present invention for processing line faults.

[0081] Figure 12 This is a schematic diagram of processing an emergency insertion order according to the present invention.

[0082] Figure 13 A production scheduling Gantt chart implemented using a production line scheduling system based on line-side warehouse capacity.

[0083] Figure 14 This is a diagram showing the training results of the production line scheduling algorithm based on the line-side warehouse capacity of the present invention.

[0084] Figure 15 This is a test result diagram of the production line scheduling algorithm based on the line-side warehouse capacity of the present invention. DETAILED DESCRIPTION

[0085] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments.

[0086] like Figure 1 The present invention constructs a 3C smart manufacturing workshop production line scheduling system that takes into account the line-side warehouse capacity. Specifically, the scheduling system includes a demand layer, a material layer, a scheduling layer, a management layer, and a monitoring layer.

[0087] Among them, the demand layer obtains workshop processing orders based on production order requirements and supply chain information, which is used to create original information and generate initial plans for the job scheduling module of the workshop scheduling system;

[0088] At the material level, the workshop warehousing system provides the scheduling layer with warehousing information on processing raw materials and work-in-progress in each line's warehouse based on the type of demand order. It also obtains the quantity of raw materials and work-in-progress in real time and handles material shortages and other incidents during workshop production.

[0089] The scheduling layer uses reinforcement learning algorithms to allocate production lines to multiple processes for multiple processing orders based on the original information and material information of the demand orders, combined with the current production situation in the workshop, to achieve optimal production scheduling under limited line and side warehouse capacity.

[0090] Management: After the scheduling center generates and successfully issues a production plan, the operation management layer manages the processing information of each order in real time and handles emergency order insertions in the workshop.

[0091] At the monitoring layer, there are many dynamic events in the actual production process, including production line failures, workshop resource shortages, etc. The monitoring layer monitors the various dynamic events occurring in the workshop in real time, and feeds back to the task scheduling layer through the workshop Internet of Things for active scheduling and processing.

[0092] The present invention further proposes a production line scheduling method for a 3C smart manufacturing workshop that takes into account the capacity of the line-side warehouse. The method is applied to the above-mentioned production line scheduling system for a 3C smart manufacturing workshop that takes into account the capacity of the line-side warehouse, and includes the following steps:

[0093] S1. According to the actual production situation of the 3C smart manufacturing workshop, the scheduling center based on the scheduling layer in the scheduling system extracts the initial information of the processing order through the demand layer, and at the same time, the monitoring layer extracts the initial information of the workshop environment, including the information of the processing line and the line side warehouse, and constructs a heterogeneous graph model for task scheduling. The heterogeneous graph model includes the process nodes, processing line nodes and connection arc constraint relationship information of the processing order, which are stored in the graph model module, such as Figure 2 shown.

[0094] The processing order is the initial information of the processing task. It refers to the processing task obtained by the demand layer of the workshop scheduling system based on the supply chain demand at each scheduling time point. The processing order information includes the processing quantity, order type, process route, etc.

[0095] The initial information of the workshop production environment is mainly divided into the initial information of the processing line and the initial information of the workshop line warehouse.

[0096] The processing line is used to process or assemble specific procedures for each order. The production capacity of each processing line is different for different types of orders, and the required processing time is also different. When processing different types of orders continuously at the same time, the processing line needs to spend a certain amount of time to reorganize before starting processing of the next order. That is, the workshop scheduling center will obtain initial information about the workshop environment before starting processing, including the occupancy of the production line (whether the processing line is available), information on pending production tasks (hourly output of the processing line, line adjustment time for different processing orders) and the work-in-progress capacity of the line-side warehouse. The line-side warehouse is used to temporarily store the work-in-progress completed by the upstream processing line. The current work-in-progress can only be transferred to the downstream processing line after the upstream processing line completes processing.

[0097] Construct the workshop production layout and scheduling logic. The workshop production layout mainly includes the placement of the processing line, the location and capacity of the line-side warehouse, which is used to explain the positional relationship between the processing line and the line-side warehouse; the scheduling logic mainly includes the general process requirements of the processing order, workshop production constraints and other information, which is used to explain the relationship between the processing line and the processing order.

[0098] The workshop internet architecture and heterogeneous graph model are used to explain the information reception, transmission, and transmission methods between the workshop scheduling center and the processing lines, as well as the heterogeneous graph model construction process. It includes the following steps:

[0099] S101. Establishment of the workshop internet architecture, which consists of the workshop scheduling system and all processing lines, and is used to transmit scheduling-related information;

[0100] Establish a workshop internet architecture, connecting all processing lines within the workshop to the workshop dispatch center to form a local area network. Devices within the local area network transmit relevant information via the wireless network, including processing order information issued by the dispatch center to the processing lines at the dispatching layer of the dispatch system, and production information fed back by the processing lines to the dispatch center.

[0101] S102. Construction of heterogeneous graph model. Through the workshop production layout and scheduling logic, the workshop scheduling center can obtain the heterogeneous graph model applicable to the current 3C smart manufacturing workshop; obtain the specific parameters of the heterogeneous graph model through the workshop scheduling Internet.

[0102] Through the workshop production layout and scheduling logic, the workshop scheduling center can obtain information such as the current 3C smart manufacturing workshop's production sequence, processing line layout, and the number of processing lines in each process, and then build the overall architecture of the heterogeneous graph model; at the same time, through the workshop Internet, it can obtain order information and processing line information that need to be processed, including whether the workshop processing line is idle, the type of processing order, the processing line reorganization time and other specific information, and then improve and supplement the heterogeneous graph model.

[0103] S2. In the scheduling layer, the workshop scheduling center performs a graph embedding operation on the heterogeneous graph model obtained in step S1, and uses the graph attention mechanism to perform weighted summation on the features in the neighborhood to extract the high-dimensional features of the process nodes of the processing order, the high-dimensional features of the processing line nodes, and the high-dimensional features of the connection arcs. The above three high-dimensional features are averaged and pooled to obtain the global features of the heterogeneous graph model; Figure 3 As shown in the figure, the 3C smart manufacturing workshop is a hybrid assembly line workshop with parallel machining. The processing procedures for each order are the same. Due to the existence of parallel machines, orders under the same procedure can be processed on different processing lines. At the same time, due to the existence of line-side warehouses, the production time of the processing lines may be affected by their capacity.

[0104] The workshop order center integrates and extracts the information of processing orders, processing lines and line-side warehouses as feature representations of process nodes, processing line nodes and connecting arcs.

[0105] Among them, the process node features of a processing order include order processing status, order processing steps, order completion rate, etc.; the processing line node features include processing line status, processing line utilization rate, and the number of adjacent process nodes; the connection arc feature between processing line nodes and process nodes is processing time; and the connection arc feature between processing process nodes is the line side warehouse status and the number of work-in-progress. Then, the features of process nodes, processing line nodes, and connection arcs are graph embedded. The specific steps are as follows:

[0106] S201. The workshop scheduling center uses a multi-layer perceptron with different training parameters to perform feature dimension upgrade on the original feature information of the order process nodes, processing line nodes, and each connection arc, and extracts high-dimensional feature information, as shown in formula (5).

[0107] h=MLP θ (f) (5)

[0108] Where: h is the high-dimensional feature of the node or connection arc;

[0109] f is the original feature of the node or connecting arc;

[0110] MLP is a multi-linear layer perceptron;

[0111] θ is the linear layer network training parameter;

[0112] Then, the graph attention mechanism is used to perform weighted summation of features within the neighborhood of process nodes, line nodes, and connection arcs to extract corresponding high-dimensional features. The specific steps are as follows:

[0113] S202: Calculate the graph attention coefficients of the process features, processing line features, and connection arc features, and then use the neighborhood features and the corresponding attention coefficients to calculate the high-dimensional features. The specific steps are as follows:

[0114] S2021, reference Figure 4 , the graph attention coefficient of the processing line node includes two types, one is the attention coefficient of the current processing line node feature and its adjacent process node feature, and its calculation method is shown in formula (6); the other is the attention coefficient of the current processing line node feature and itself, and its calculation method is shown in formula (7).

[0115] att m,s =σ(a T (h m ||h s )) (6)

[0116] att m,m =σ(a T (h m ||h m )) (7)

[0117] Where: h m It is the high-dimensional feature of the processing line nodes;

[0118] h s It is the high-dimensional feature of the adjacent process nodes of the processing line;

[0119] a T A single-layer linear neural network

[0120] σ is the activation function

[0121] Then the two attention coefficients are normalized, as shown in formulas (8) and (9).

[0122]

[0123] Where: att m,m is the attention coefficient between the nodes of the processing line;

[0124] att m,s is the attention coefficient between the processing line node and the adjacent process;

[0125] Then the high-dimensional features of the processing line nodes are calculated, as shown in formula (10).

[0126] EB m =σ(Σ s∈N(m) τ m,s h s +τ m,m h m ) (10)

[0127] S2022, reference Figure 5 , the graph attention coefficients of the process nodes of the processing order include three types. One is the attention coefficient of the current process node feature and its first-order adjacent process node feature, the second is the attention coefficient of the adjacent processing line node feature, and the third is the attention coefficient of the connection edge feature with the next process. The three attention coefficients are normalized, and the calculation method is similar to formula (6)-(8); then the high-dimensional features of the process node are calculated, as shown in formula (11).

[0128] EB s =σ(∑ s'∈N(s) τ s,s' h s' ||∑ m∈N(s) τ s,m h m ||h s ||∑ l∈N(l) τ s,l h l ) (11)

[0129] Where: h l is the characteristic of the connection arc related to the process node;

[0130] S2023. The high-dimensional features of the connection arcs between process nodes and the high-dimensional features of the connection arcs between processing line nodes and process nodes are obtained by weighted summation using the graph attention mechanism, and their dimensions are the same as those of the process nodes and processing line nodes.

[0131] Then, the three high-dimensional features of process nodes, line nodes, and connection arcs are average-pooled to obtain the global features of the heterogeneous graph model. The specific steps are as follows:

[0132] In step S203, the workshop scheduling center uses a heterogeneous graph neural network to solve the global graph embedding feature. To improve feature extraction capabilities, as shown in Equation (12), all process node features, processing line node features, and connection arc features are input into a multi-layer neural network with the same structure but different parameters for deep feature extraction. Then, an average pooling operation is performed, and finally, a fixed-dimensional global graph embedding feature is obtained by splicing.

[0133]

[0134] Where: The high-dimensional features of the process nodes output by the K-th layer heterogeneous graph neural network;

[0135] The high-dimensional features of the processing line nodes output by the K-th layer heterogeneous graph neural network;

[0136] The high-dimensional features of the connection arcs output by the K-th layer heterogeneous graph neural network;

[0137] S3. The workshop scheduling center uses the actor-critic reinforcement learning algorithm to assign the tasks to be processed to the designated processing lines based on the heterogeneous graph model obtained in step S1 and the high-dimensional features of all process nodes, processing line nodes, connection arcs, and global features of the heterogeneous graph model obtained in step S2.

[0138] The production line scheduling optimization goals of the 3C smart manufacturing workshop include ensuring the delivery date of each processing order and maximizing the average utilization rate of the processing line. Among them, ensuring the delivery date can be evolved into two goals. If there is no unique scheduling result that meets the delivery date, the optimization goal becomes minimizing the total completion time; if there is no scheduling result that can meet the delivery date. Then the optimization goal becomes minimizing the delay time. The utilization rate of the processing line refers to the average utilization rate of all parallel processing lines in each process. To prevent the parallel machine from "overworking" in the same process, it is necessary to add the weight of the proportion of tasks executed by the current processing line when calculating the average utilization rate. If there are special requirements for the workpiece order, the workshop scheduling center will change the optimization goals of the workshop in real time.

[0139] Reference Figure 6The workshop scheduling center supplements the heterogeneous graph model based on the 3C production model obtained in step S1 with the three high-dimensional features and global features of the heterogeneous graph obtained in step S2. This model is continuously updated and changed as the order scheduling process progresses. When processing line a is selected for the jth process of the i-th order, the remaining parallel processing lines in the jth process cannot be processed, and the arcs connecting to the corresponding process nodes need to be deleted.

[0140] The specific process of allocating processing orders by the workshop scheduling center is as follows:

[0141] S301, reference Figure 7 The workshop scheduling center observes the overall status of the system, including high-dimensional features of process nodes such as processing order process progress and processing order mask, high-dimensional features of line nodes such as line status and line efficiency, high-dimensional features of connection arcs such as processing time, and global features of heterogeneous graphs. It selects actions in the action space containing valid processing lines and processing order process connection arcs, and assigns processing orders to designated processing lines for processing.

[0142] S302. By executing the currently selected action, the workshop scheduling center interacts with the production environment to calculate the reward function, including whether the delivery date is met and the utilization rate of the processing line is maximized.

[0143] S303. The workshop scheduling center executes the selected action, and the workshop production environment changes to the next state, which is then reflected in the heterogeneous graph model. The workshop scheduling center updates the parameters of the scheduling model according to the changes in the actual production environment, including the process node characteristics of the processing order, the process line node characteristics and the connection arc characteristics. Among them, due to the influence of the line warehouse WIP capacity, the process line reorganization time and the processing sequence, the update of the process order process and the process line start time is more complicated. There are many situations. Refer to Figure 8 The update is performed, where the start time and processing time are calculated as shown in formulas (2) and (3).

[0144] Reference Figure 9 The specific steps for the workshop scheduling center to update the parameters of the deducer-critic reinforcement learning algorithm are as follows:

[0145] S304, when the number of interactions with the production environment reaches the update point, the proximal optimization algorithm will use the experience in the memory library to update the batch parameters, referring to Figure 10First, the reward obtained from each action performed in experience is used to calculate the future discounted return. Next, the performer network is used to identify batch states in the memory bank and derive batch action probabilities. The KL divergence, advantage function, and upper and lower bounds for proximal policy updates are then calculated using the logarithm of the action probabilities. The parameters of the performer algorithm in the performer-critic reinforcement learning algorithm are updated using gradient ascent. The critic network parameters are updated using gradient descent using the difference between the global discounted return and the critic algorithm output score in the performer-critic reinforcement learning algorithm, resulting in scores for the current state and action. The workshop scheduling center then performs normal distribution sampling based on the weighted and normalized action and state scores to rationally assign each processing order to a processing line. Normal distribution sampling improves the algorithm's flexibility and prevents it from falling into local optimal solutions. Each time the workshop scheduling center performs an action, it receives reward feedback from the environment. The reward is consistent with the scheduling objective, so the reward function is divided into two parts: the first is the delivery deadline achievement rate, and the second is the average utilization rate of all parallel processing lines in the process.

[0146] S4, the processing line in the workshop receives the assigned task through S3 and enters the task execution stage. At the same time, it interacts with the workshop working environment and uploads the production information to the workshop scheduling center at the scheduling layer through the management layer and the monitoring layer;

[0147] The workshop scheduling center schedules all processing orders according to the method in step S3. After receiving the order, the processing line begins processing while interacting with the workshop environment. Any issues such as line failures or line blockages due to reaching the lineside warehouse's warning capacity are communicated between the processing line and the workshop scheduling center. The workshop scheduling center also monitors the production status of current orders in real time and adjusts the processing sequence of production orders based on potential dynamic events.

[0148] S5. In the monitoring and management layers, the workshop scheduling center uses the information of the processing lines and processing orders in step S4 to monitor the workshop processing status in a timely manner and realize real-time scheduling of workshop tasks.

[0149] After receiving the information sent by the processing line, the workshop dispatch center will handle the emergencies. The specific emergencies and handling strategies are as follows:

[0150] For processing line failure, refer to Figure 11 .

[0151] First, the processing line sends a fault message to the workshop scheduling center via the workshop internet. Upon receiving the message, the workshop scheduling center sets the node mask of the corresponding processing line to 0, indicating that the line is unavailable. The corresponding connection arc will not be selected for subsequent scheduling of the current process. Consequently, tasks after the faulty line are rescheduled to minimize the impact on the scheduling objectives. Once the fault is resolved, the workshop scheduling center sets the processing line mask to 1, and the corresponding connection arc is placed in the available action space for subsequent scheduling.

[0152] In the event of line-side warehouse congestion, the workshop scheduling center will suspend the production of all parallel machines in the upstream process of the line-side warehouse. That is, the process mask of the upstream process is set to 0. All connection arcs corresponding to the processing lines of the current process cannot be selected in subsequent scheduling until the capacity of the line-side warehouse falls below the warning capacity and then it is set to 1.

[0153] For line side warehouse blockage, refer to Figure 12 .

[0154] The workshop scheduling center will add the process node corresponding to the new order, the processing line node that can produce the corresponding process, and the connection arc in the heterogeneous graph model at that point in time, update the global graph embedding features of the heterogeneous graph, and prioritize scheduling the order at that point in time.

[0155] Figure 13 This is one of the simulations implemented using a production line scheduling system that considers line-side storage. In this model, the workshop is a 7*5 mixed flow workshop with a total of 5 processes. Each process has 2 processing machines. The 7 processing tasks arrive at random times, and the number of workpieces for each task is the batch size of a processing order. The scheduling algorithm based on large-scale data training is used to reasonably schedule processing orders. The training results are shown below. Figure 14 The final scheduling result of the Gantt chart is as follows: Figure 13 , the minimum delay time result is as follows Figure 15 , achieving the workshop’s goal of meeting delivery dates with minimal impact from buffer zone congestion.

[0156] If an urgent order is required, the workshop scheduling center will add the process node corresponding to the new order, the processing line node that can produce the corresponding process, and the connection arc in the heterogeneous graph model at that time point, update the global graph embedding features of the heterogeneous graph, and give priority to scheduling the order at that time point.

[0157] In summary, the present invention takes into account the limited work-in-progress capacity of the line-side warehouse, and uses the relevant parameters of the line-side warehouse capacity as part of the state of the reinforcement learning algorithm, which reduces the blocking probability of order processing in the 3C smart manufacturing workshop and ensures the continuity of production. By constructing a heterogeneous graph model and a method based on the graph attention mechanism to extract and identify feature information, the dimension of feature information is deepened, the accuracy of the scheduling algorithm's recognition state is improved, and the foundation is laid for the reinforcement learning algorithm to select appropriate actions. By establishing a state space through reinforcement learning, selecting actions that meet the reward function, and using the reward function to update the state, the calculation speed and accuracy of production line scheduling are improved. At the same time, by using different optimization objectives and learned end-to-end scheduling strategies, the workshop scheduling center can perform production line scheduling more flexibly.

[0158] The above content is only for explaining the technical idea of ​​the present invention and cannot be used to limit the protection scope of the present invention. Any changes made on the basis of the technical solution in accordance with the technical idea proposed by the present invention shall fall within the protection scope of the claims of the present invention.

Claims

1. A 3C smart manufacturing workshop production line scheduling method considering the line side warehouse capacity, characterized in that: The following steps are involved: Step S1: According to the actual production situation of the 3C smart manufacturing workshop, the scheduling center based on the scheduling layer in the scheduling system extracts the initial information of the workshop environment through the demand layer and the monitoring layer, including the information of the processing orders, processing lines and line warehouses, and constructs a heterogeneous graph model for task production scheduling. The heterogeneous graph model includes the process nodes, processing line nodes and connection arc constraint relationship information of the processing orders, which are stored in the graph model module; Step S2: In the scheduling layer, the workshop scheduling center performs a graph embedding operation on the heterogeneous graph model obtained in step S1, and uses the graph attention mechanism to perform weighted summation on the features in the neighborhood to extract the high-dimensional features of the process nodes of the processing order, the high-dimensional features of the processing line nodes, and the high-dimensional features of the connection arcs. The above three high-dimensional features are averagely pooled to obtain the global features of the heterogeneous graph model; Step S3, in the scheduling layer, the workshop scheduling center uses the actor-critic reinforcement learning algorithm to allocate the tasks to be processed to the designated processing line based on the heterogeneous graph model obtained in step S1 and the high-dimensional features of the process nodes, line nodes, connection arcs, and global features of the heterogeneous graph obtained in step S2; Step S4: Based on the task assigned to the designated processing line in step S3, the task execution phase is entered, and at the same time, the workshop working environment is interacted with. The workshop scheduling center will receive the production information from the processing line and the order processing information from the scheduling system management layer; Step S5: In the monitoring layer and the management layer, the workshop dispatching center timely monitors the workshop processing status through the processing line and order processing information in step S4, and realizes the real-time scheduling of workshop tasks.

2. According to a method for scheduling a production line in a 3C smart manufacturing workshop taking into account the capacity of the line side warehouse according to claim 1, it is characterized in that: The step S1 is specifically as follows: The workshop scheduling center obtains the initial information of the tasks that need to be processed at each scheduling time point, including the order type, the number of order processes, and the order processing quantity; at the same time, the workshop scheduling center will obtain the initial information of the workshop environment before starting processing, including the information of the tasks to be produced, the occupancy of the production line, and the work-in-process capacity of the line warehouse; based on the initial information, a heterogeneous graph model for production scheduling is constructed, including the process nodes of the task, the processing line nodes, the directed arcs between the process nodes, and the undirected arcs between the machines and the process nodes.

3. According to a method for scheduling a production line in a 3C smart manufacturing workshop taking into account the capacity of the line side warehouse according to claim 1, it is characterized in that: The characteristic information of the process node in step S2 includes: (A) Order processing status, if the order has been completed, it is 1, if it has not been completed, it is 0; (B) Order processing procedure, the current procedure node corresponds to the procedure number of the processing order; (C) Order completion rate: the completion rate of the order corresponding to the current process, as shown in formula (1); In the formula, s i,o is the j completed processes of the i-th order; S i is the total number of processes required for the i-th order; (D) The number of adjacent nodes, the sum of the number of process nodes and processing line nodes that are one unit length away from the current process node; (E) The start time of the current process. The start time of the current process is shown in formula (2). If there is no congestion due to the capacity of the line side warehouse, the start time is the maximum value of the end time of the previous process of the current order and the end time of the current processing line. If there is a congestion, the start time is determined by the time when the congestion is released, that is, the time point when the downstream section generates a new processing task: Where: ET i,j-1 is the end time of the j-1th process of the i-th order; ME j,p is the end time of the pth parallel processing line in the jth process; x is whether the jth process is blocked, 0 if blocked and 1 if not; MS j+1,a is the start time of the ath parallel processing line in the j+1th process; (F) The end time of the current process, as shown in formula (3), the end time of the current process is the sum of the start time and the processing time of the parallel processing line; Where: ET i,j-1 is the end time of the j-1th process of the i-th order; MS j,a is the start time of the ath parallel processing line in the jth process; A i,j is the processing quantity of the jth process of the i-th order; A i,j is the hourly output of the ath parallel processing line in the jth process; The characteristic information of the processing line node in step S2 includes: (1) The status of the processing line. If the processing line has started processing an order, the value is 1; if it has not started, the value is 0. (2) The utilization rate of the processing line, as shown in formula (4), the utilization rate of the processing line is determined by the completion time of the processing time; Where: U j,a is the utilization rate of the a-th parallel processing line in the j-th process; P j,a is the total processing time of the ath parallel processing line in the jth process; C j,a is the total completion time of the a-th parallel processing line in the j-th process; (3) The number of adjacent process nodes, which is the number of process nodes with a unit length distance from the current processing line node; The characteristic information of the connection arcs in step S2 includes: characteristic information of directed arcs between process nodes and characteristic information of undirected arcs between process nodes and processing line nodes; The characteristic information of the directed arcs between process nodes includes the status of the line-side warehouse. If the capacity of the line-side warehouse has not reached the warning capacity, it is 0, and if it has reached the warning capacity, it is 1; the characteristic information of the directed arcs also includes the number of work-in-progress in the line-side warehouse, which is determined by the upstream and downstream processing lines; the characteristic information of the undirected arcs of the process nodes and the processing line nodes refers to the processing time; the characteristic information of the directed arcs of the process nodes and refers to the processing quantity.

4. According to a method for scheduling production lines in a 3C smart manufacturing workshop taking into account the capacity of line-side warehouses according to claim 3, it is characterized in that: The process of step S2 is as follows: the workshop dispatching center uses a multi-linear layer perceptron with different training parameters to perform feature dimension upgrade on the feature information of the process nodes, processing line nodes and connection arcs of the order, extracts high-dimensional feature information, and then uses the graph attention mechanism to embed the features of the process nodes, processing line nodes and connection arcs into local graphs; the feature graph embedding of the processing line node includes two parts, which are the weighted sum of the feature embeddings of all process nodes in its first-order neighborhood and the product of its attention coefficient, and the feature embedding of the processing line node itself and the product of its attention coefficient; the feature graph embedding of the process node includes the features of all process nodes in its first-order neighborhood, the features of all machine nodes and the product of the edge features connecting the next process point; the feature graph embedding of the connection arc includes the weighted sum of the feature attention of the line node and the process node to which it is connected, so as to extract the high-dimensional features of the process nodes of the processing order and the high-dimensional features of the processing line node; The workshop scheduling center uses the average pooling method to solve the global graph embedding features, and inputs the obtained order process node features, processing line node features and their connecting edge features into a multi-layer neural network with the same structure but different parameters for deep feature extraction and average pooling. Finally, the global features of the heterogeneous graph model with a fixed dimension are obtained by splicing. Since the heterogeneous graph model constructed based on the 3C production model obtained in step S1 will be continuously updated and changed with the order scheduling process, when the j-th process of the i-th order selects processing line a, that is, the remaining parallel processing lines in the j-th process cannot be processed, the connection arc with the corresponding process node needs to be deleted.

5. According to a method for scheduling a production line in a 3C smart manufacturing workshop taking into account the capacity of the line side warehouse according to claim 1, it is characterized in that: The step S3 is specifically as follows: The workshop scheduling center selects actions in the action space containing valid processing lines and processing order process connection arcs by observing the overall status of the system, including high-dimensional features of process nodes such as processing order process progress and processing order mask; high-dimensional features of line nodes such as line status and line efficiency, high-dimensional features of connection arcs such as processing time, and global feature information of heterogeneous graphs, and assigns processing orders to designated processing lines for processing; the workshop scheduling center calculates reward functions by interacting with the production environment, including whether the delivery date is met and whether the utilization rate of the processing line is maximized, while executing the current processing procedure through the processing line; When selecting actions, the workshop scheduling center uses the normal distribution sampling method to make the selection. After inputting the observed production status into the actor-critic reinforcement learning algorithm, the score of each valid action is obtained. Then, the status and action are simultaneously input into the critic algorithm in the actor-critic reinforcement learning algorithm to obtain the score of the current status and action. The workshop scheduling center performs normal distribution sampling based on the weighted and normalized action scores and status scores, and reasonably allocates each processing order to the processing line.

6. According to a method for scheduling production lines in a 3C smart manufacturing workshop taking into account the capacity of line-side warehouses as described in claim 1, it is characterized in that: The step S4 is specifically as follows: The workshop dispatching center schedules the processes of all processing orders according to the method in step S3. After receiving the order, the processing line starts working and interacts with the workshop environment. The processing line is blocked due to processing line failure or reaching the warning capacity of the line warehouse. Information is transmitted between the processing line and the workshop dispatching center. At the same time, the workshop dispatching center monitors the production status of the current order in real time and adjusts the processing sequence of the production order according to possible dynamic events.

7. According to a method for scheduling a production line in a 3C smart manufacturing workshop taking into account the capacity of the line side warehouse according to claim 1, it is characterized in that: The step S5 is specifically as follows: After receiving the information sent by the processing line, the workshop scheduling center handles the emergencies. If a processing line failure occurs, the workshop scheduling center sets the processing line node mask with the corresponding number to 0, indicating that the line is unavailable. In the subsequent scheduling of the current process, the corresponding connection arc of the line will not be selected, and the task on the line will be handed over to other parallel machines of the current process for production, minimizing the impact on the scheduling target; if the line side warehouse is blocked, the workshop scheduling center suspends the production of all parallel machines in the upstream process of the line side warehouse, that is, sets the process mask of the upstream process to 0, until the capacity of the line side warehouse is lower than the warning capacity and then sets it to 1; if an emergency order is encountered, the workshop scheduling center adds the process node corresponding to the new order, the processing line node and connection arc that can produce the corresponding process in the heterogeneous graph model at that time point, updates the global graph embedding features of the heterogeneous graph, and gives priority to scheduling the order at that time point.

8. A 3C smart manufacturing workshop production line scheduling system considering the line side warehouse capacity, characterized in that: The method for scheduling a production line in a 3C smart manufacturing workshop according to any one of claims 1 to 7 comprises a demand layer, a material layer, a scheduling layer, a management layer and a monitoring layer: The demand layer obtains workshop processing orders based on production order requirements and supply chain information, which is used for original information creation and initial plan generation of the job scheduling module of the workshop scheduling system; At the material layer, the workshop storage system provides the scheduling layer with storage information of processed raw materials and storage information of work-in-progress in each line warehouse according to the type of demand order, and obtains the quantity of raw materials and work-in-progress in real time; The scheduling layer uses a reinforcement learning algorithm to allocate production lines for multiple processes of multiple processing orders based on the original information and material information of the demand orders and the current production situation in the workshop, thereby achieving optimal production scheduling under limited line side warehouse capacity; In the management layer, after the scheduling center generates the production schedule and successfully issues it, the processing information of each order is managed in real time by the operation management layer, and at the same time, events such as emergency insertion orders in the workshop are handled; The monitoring layer described above has various dynamic events in the actual production process, including production line failures, workshop resource shortages, etc. The monitoring layer monitors the various dynamic events occurring in the workshop in real time, and feeds back to the task scheduling layer through the workshop Internet of Things for active scheduling processing.

Citation Information

Cited By

  • Intelligent distribution system and method for dairy product production line

    CN119831293A

  • Discrete manufacturing workshop production execution logic modeling method

    CN120010407A

  • Task flow dynamic optimization method and system for data feedback

    CN121073185A

  • Adhesive tape production line production scheduling system based on intelligent optimization

    CN121094497A

  • An intelligent optimization-based adhesive tape production line production scheduling system

    CN121094497B