A multi-agent collaborative production scheduling method and system based on an industrial large model

Abstract

The application relates to the technical field of production scheduling, in particular to a multi-agent collaborative production scheduling method and system based on an industrial large model, which comprises the following steps: acquiring production state data and constructing a production semantic state graph; performing semantic identification on production disturbance information based on the industrial large model to obtain effective disturbance events; performing influence propagation analysis on the effective disturbance events to determine affected task groups; initiating an ability inquiry to a resource agent to obtain ability commitment data, then generating a candidate collaborative strategy, and performing collaborative scheduling to obtain a candidate scheduling scheme; performing local execution rehearsal on the candidate scheduling scheme to determine a target scheduling scheme; acquiring execution feedback of the target scheduling scheme, performing state deviation identification and local scheduling repair, and obtaining a production scheduling result. The application can improve the accuracy of disturbance influence range identification and the executability of a scheduling scheme, and reduce the collateral adjustment on unaffected tasks.

Description

Technical Field

[0001] This invention relates to the field of production scheduling, and in particular to a multi-agent collaborative production scheduling method, system, electronic device, and storage medium based on a large industrial model. Background Technology

[0002] Flexible manufacturing, as an important production method for modern manufacturing to achieve multi-variety, small-batch, and rapid delivery, is increasingly being applied in flexible manufacturing workshops due to the increasing types of resources such as production equipment, logistics and transportation, and quality inspection. However, the flexible manufacturing site is characterized by complex task associations, so local disturbances are prone to spread along the process sequence and resource dependencies.

[0003] Existing production scheduling methods typically obtain order, process, and resource status data through the manufacturing execution system. After detecting disturbances such as equipment failure or order changes, they rearrange the resource allocation, execution order, and execution time of production tasks according to preset scheduling rules, heuristic algorithms, mathematical programming models, or multi-agent negotiation mechanisms.

[0004] However, existing technologies still have significant limitations in practical applications: On the one hand, existing methods typically determine the adjustment targets based on a single disturbance event, making it difficult to combine the relationships between orders, processes, and multiple types of resources to conduct continuous propagation analysis of disturbances, which can easily lead to incomplete identification of affected tasks or excessive adjustment scope; on the other hand, when generating scheduling schemes, existing methods usually only arrange resources based on their current idle status, lacking pre-running verification of resource preparation and execution processes before the scheme is issued, which may result in the generated scheduling scheme having problems such as resources not being ready, making it difficult to actually execute the scheduling scheme. Summary of the Invention

[0005] To at least partially address the disconnect between production disturbance identification and scheduling adjustment in existing technologies, which typically rely solely on local rule matching based on single events such as equipment failures or order changes, leading to inaccurate identification of the disturbance's impact range and over-adjustment of irrelevant production tasks, this invention proposes a multi-agent collaborative production scheduling method, system, electronic device, and storage medium based on a large industrial model.

[0006] To achieve the above objectives, in a first aspect, this invention proposes a multi-agent collaborative production scheduling method based on a large industrial model, comprising:

[0007] Acquire production status data and construct a production semantic state graph;

[0008] Semantic recognition of production disturbance information is performed based on a large industrial model to obtain valid disturbance events;

[0009] In the production semantic state diagram, the impact propagation analysis of effective disturbance events is performed to obtain the affected task groups;

[0010] Based on the affected task group, a capability query is initiated to the resource agent to obtain capability commitment data;

[0011] Candidate coordination strategies are generated based on the effective disturbance events, affected task groups, and capability commitment data, and candidate scheduling schemes are formed based on the candidate coordination strategies.

[0012] Perform partial execution rehearsals on the candidate scheduling schemes to determine the target scheduling scheme;

[0013] Obtain the execution feedback of the target scheduling scheme, identify the state deviation of the execution feedback, and obtain the production scheduling result.

[0014] Preferably, an impact propagation analysis is performed on effective disturbance events in the production semantic state graph to obtain the affected task groups, including:

[0015] Based on the event type of the effective disturbance event, select the production relationship that matches the event type from the production semantic state diagram;

[0016] The process is propagated hierarchically along the production relations to obtain candidate tasks;

[0017] The degree of influence of each candidate task is calculated, and the candidate tasks whose degree of influence reaches the preset influence threshold are identified as affected tasks, and the affected task group is obtained by summarizing them.

[0018] Preferably, candidate tasks are obtained by hierarchical propagation along the production relationship, including:

[0019] Based on the event object in the effective disturbance event, determine the corresponding production object node in the production semantic state graph, and use the production object node as the propagation start node;

[0020] Search along the production relationship for production object nodes directly associated with the propagation starting node to obtain the first layer of associated nodes;

[0021] Use the current layer's associated node as the new propagation node, continue searching for the next layer's associated node, and record the propagation level and propagation path corresponding to each associated node;

[0022] The process nodes corresponding to the production tasks to be executed are selected from the associated nodes at each layer. The corresponding production tasks are determined according to each process node, and the production tasks are deduplicated to obtain candidate tasks.

[0023] Preferably, after obtaining the affected task groups, locking the unaffected task groups includes:

[0024] Remove the production tasks corresponding to the affected task groups from the current production tasks to obtain the unaffected task groups;

[0025] Obtain the original scheduling information of each production task in the unaffected task group;

[0026] Based on the original scheduling information, the resource allocation and planned execution time of each production task are locked, and the locked unaffected task group is used as the local scheduling boundary.

[0027] Preferably, candidate coordination strategies are generated based on the effective disturbance events, affected task groups, and capability commitment data, and candidate scheduling schemes are formed based on the candidate coordination strategies, including:

[0028] Input the effective disturbance events, affected task groups, and capacity commitment data into the industrial big data model to determine the candidate adjustment actions and candidate resources corresponding to each production task in the affected task group, and generate candidate coordination strategies.

[0029] Based on the candidate collaborative strategy, the corresponding resource agent is controlled to negotiate and obtain negotiation data;

[0030] Based on the negotiated data, a local scheduling solution is performed to obtain a candidate scheduling scheme.

[0031] Preferably, the resource agent corresponding to the candidate collaborative strategy is controlled to negotiate and obtain negotiation data, including:

[0032] Send the candidate adjustment actions corresponding to each production task to the resource agent corresponding to the candidate resource;

[0033] Each resource agent verifies the execution conditions of the candidate adjustment actions based on the corresponding capability commitment data, and feeds back candidate resource information and constraint information.

[0034] For candidate adjustment actions that pass the execution condition verification, determine the resource coordination cost;

[0035] The candidate resource information, constraint information, and corresponding resource coordination costs are summarized to obtain the negotiation data.

[0036] Preferably, a local scheduling solution is performed based on the negotiated data to obtain a candidate scheduling scheme, including:

[0037] The negotiated data is analyzed to determine the candidate resources, scheduling constraints, and resource coordination costs corresponding to each production task.

[0038] Using the affected task group as the object to be adjusted, the original scheduling information corresponding to the unaffected task group as the fixed scheduling boundary, the candidate resources corresponding to each production task as the resource allocation range, the scheduling constraints as the solution constraints, and the resource coordination cost as the resource selection evaluation parameter, a local scheduling model is constructed.

[0039] Based on the local scheduling model, the resource allocation, execution order, and execution time of each production task in the affected task group are jointly solved to obtain multiple candidate scheduling schemes.

[0040] Preferably, a partial execution rehearsal is performed on the candidate scheduling schemes to determine the target scheduling scheme, including:

[0041] A local execution environment is constructed based on the affected task group, and each of the candidate scheduling schemes is written into the local execution environment.

[0042] The local execution environment is advanced according to each of the candidate scheduling schemes, and the production tasks are rehearsed to obtain the rehearsal results corresponding to each candidate scheduling scheme.

[0043] Based on the pre-simulation results, candidate scheduling schemes that have execution conflicts or do not meet scheduling constraints are eliminated, and the remaining candidate scheduling schemes are comprehensively evaluated to obtain the target scheduling scheme.

[0044] Preferably, the local execution environment is advanced according to each of the candidate scheduling schemes to perform a pre-simulation of the production tasks, obtaining the pre-simulation results corresponding to each candidate scheduling scheme, including:

[0045] Based on the resource requirements, planned execution time, and simulated status of associated resources for each production task in the current candidate scheduling scheme, a resource readiness analysis is performed to determine the readiness level of the current candidate scheduling scheme.

[0046] Calculate the total delay and completion span based on the pre-execution time of each production task in the current candidate scheduling scheme, compare the current candidate scheduling scheme with the original scheduling information, and determine the number of changes to the process.

[0047] The total delay, completion span, number of changed procedures, and readiness level are summarized to obtain the pre-rehearsal results corresponding to the current candidate scheduling scheme.

[0048] Secondly, the present invention provides a multi-agent collaborative production scheduling system based on an industrial large-scale model, comprising:

[0049] The state diagram construction module is used to acquire production state data and construct a production semantic state diagram;

[0050] The disturbance analysis module is used to perform semantic recognition on production disturbance information based on the industrial big data model to obtain valid disturbance events;

[0051] The impact identification module is used to perform impact propagation analysis on valid disturbance events in the production semantic state graph to obtain the affected task groups.

[0052] The multi-agent negotiation module is used to initiate capability queries to the resource agent based on the affected task group and obtain capability commitment data;

[0053] The collaborative scheduling module is used to generate candidate collaborative strategies based on the effective disturbance events, affected task groups, and capability commitment data, and to form candidate scheduling schemes based on the candidate collaborative strategies.

[0054] The pre-execution verification module is used to perform partial execution pre-execution of the candidate scheduling scheme to determine the target scheduling scheme;

[0055] The feedback repair module is used to obtain the execution feedback of the target scheduling scheme, identify the state deviation of the execution feedback, and obtain the production scheduling result.

[0056] Thirdly, the present invention provides an electronic device including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method in the first aspect.

[0057] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method in the first aspect. Compared with the prior art, the present invention has the following beneficial effects:

[0058] I. This invention constructs a production semantic state diagram to express the process sequence, resource occupation, and task connection relationships between orders, processes, and various production resources. It also combines an industrial big data model to perform semantic recognition of production disturbance information and conducts impact propagation analysis along production relationships that match the type of disturbance event. This can improve the completeness and accuracy of identifying affected tasks. At the same time, by locking the original resource allocation and planned execution time of unaffected tasks, the scope of local scheduling is limited, thereby reducing unnecessary adjustments to normal production tasks.

[0059] Second, this invention generates capability commitment data with state versions and validity periods through resource intelligence agents, and performs resource negotiation and local scheduling based on the capability commitment data, so that the candidate scheduling scheme matches the actual capabilities and preparation conditions of the on-site resources, thereby reducing the risk that the scheduling scheme cannot be executed due to inconsistent resource states, lack of supporting resources, or insufficient task connection conditions.

[0060] Third, by performing execution rehearsals of candidate scheduling schemes in a local execution environment, this invention improves the accuracy of conflict identification and executability verification of candidate scheduling schemes. Furthermore, by combining the preparation confirmation of the target scheduling version and execution feedback to identify state deviations and repair local scheduling, it further reduces the probability of scheme execution rollback and duplicate scheduling, thereby improving the executability, stability and disturbance recovery efficiency of multi-agent collaborative production scheduling. Attached Figure Description

[0061] Figure 1 A flowchart of a multi-agent collaborative production scheduling method based on a large industrial model is provided for an embodiment of the present invention.

[0062] Figure 2 This is a block diagram of a multi-agent collaborative production scheduling system based on an industrial large model, provided for an embodiment of the present invention. Detailed Implementation

[0063] To make the technical means, creative features, objectives, and effects of this invention easier to understand, the invention is further described below with reference to specific embodiments. However, the following embodiments are merely preferred embodiments of this invention and not all of them. Other embodiments obtained by those skilled in the art based on the embodiments described herein without creative effort are all within the protection scope of this invention.

[0064] Example 1:

[0065] To achieve the above objectives, please refer to Figure 1 This invention provides a multi-agent collaborative production scheduling method based on a large industrial model, comprising:

[0066] S1. Obtain production status data and construct a production semantic state diagram.

[0067] In this embodiment, it should be specifically explained that the production status data is used to characterize the status of production tasks and production resources at the current scheduling time, including order data, process data, equipment status data, material data, logistics data, quality data, and current scheduling data; among them, each type of data includes at least the identifier, status, time, capacity, resource requirements, and scheduling information of the corresponding production object.

[0068] Specifically, corresponding data is obtained from the production management system, equipment control system, warehouse management system, logistics scheduling system, and quality management system. The object identifiers, time formats, status codes, and resource numbers in different data sources are uniformly processed and aligned with a unified production clock to obtain production status data. The latest status of each production object is summarized according to a preset status cycle to form a production status snapshot, and a status version number is configured for each production status snapshot. For example, the preset status cycle can be set to 5 seconds, and the system forms a production status snapshot every 5 seconds.

[0069] Furthermore, a production semantic state diagram is constructed based on the production status data, specifically including:

[0070] Identify production objects based on production status snapshots and establish corresponding nodes for each production object. The nodes in the production semantic state diagram include, but are not limited to, order nodes, process nodes, equipment nodes, tool nodes, fixture nodes, material nodes, transportation task nodes, and inspection resource nodes.

[0071] Production relationships are established between various production object nodes. These relationships include process sequence relationships, equipment executability relationships, tooling compatibility relationships, fixture occupancy relationships, material supply relationships, transportation arrival relationships, quality inspection relationships, and shared resource competition relationships. These production relationships are used to characterize the process sequence between production tasks, as well as the compatibility, occupancy, supply, connection, and competition relationships between production tasks and various production resources.

[0072] By associating each production object node with the production relationship according to a unified state version, a production semantic state diagram is obtained.

[0073] When the status of a production object or a production relationship changes, only the corresponding node, production relationship and its status attributes are updated, and a new status version is generated. The status version can distinguish the production status corresponding to different collection times, preventing the subsequent industrial large model, resource intelligence and local scheduling process from using production data of different time versions.

[0074] S2. Based on the industrial large-scale model, semantic recognition is performed on production disturbance information to obtain valid disturbance events. The specific processing steps include:

[0075] Acquire production disturbance information generated within a preset time range. Production disturbance information refers to information that may cause changes in production tasks, production resources, or the current scheduling status.

[0076] Extract the current state, preceding and following processes, resource occupancy and current scheduling information associated with the corresponding production object node from the production semantic state diagram to form a disturbance state fragment. The disturbance state fragment is used to provide the industrial big model with the current workshop state basis, so as to avoid the industrial big model judging disturbance events based solely on a single alarm text or operation record.

[0077] The production disturbance information and disturbance state fragments are input into the industrial big model, and the industrial big model is constrained to perform semantic recognition according to preset event fields, and output structured disturbance events. The structured disturbance events include at least the event object, event type, start time, duration, capacity change, evidence identifier, and possible influencing relationships.

[0078] It is important to know that the industrial big data model is only used to convert scattered and differently expressed production disturbance information into disturbance events with a unified field structure, and does not directly generate or issue production scheduling instructions based on the structured disturbance events.

[0079] Furthermore, the structured disturbance events are validated for validity. The validity validation includes field integrity validation, time consistency validation, production relationship validation, and capability boundary validation, which are used to determine whether the event fields are complete, whether the event time is reasonable, whether the event association exists in the production semantic state diagram, and whether the capability change is within the actual capability range of the corresponding production object.

[0080] The credibility of structured disturbance events is calculated based on the results of multiple verification methods. Event credibility is obtained by summing the products of data source credibility, temporal consistency, relational consistency, and cross-verification with their respective weights. When the event credibility is not lower than a preset event confirmation threshold, the structured disturbance event is determined as a valid disturbance event. When the event credibility of a structured disturbance event is lower than the preset event confirmation threshold, it is marked as an event awaiting confirmation and manually reviewed. Specifically, data source credibility, temporal consistency, relational consistency, and cross-verification are all normalized to between 0 and 1, and each corresponding weight is non-negative with a sum of 1, ensuring that the obtained event credibility is between 0 and 1.

[0081] S3. Perform impact propagation analysis on effective disturbance events in the production semantic state diagram to obtain the affected task groups.

[0082] In this embodiment, it should be noted that the impact propagation analysis is used to determine the actual impact range of the effective disturbance event in the current production plan. The affected task group includes production tasks directly affected by the effective disturbance event, as well as production tasks indirectly affected through process sequence, resource occupation, material supply, logistics connection, or quality inspection. The specific processing procedure is as follows:

[0083] S31. Based on the event type of the effective disturbance event, select the production relationship that matches the event type from the production semantic state diagram.

[0084] S32. Propagate hierarchically along the production relationship to obtain candidate tasks.

[0085] S321. Based on the event object in the effective disturbance event, determine the corresponding production object node in the production semantic state graph, and use the production object node as the propagation start node.

[0086] When a valid disturbance event is associated with multiple production objects, the node corresponding to each production object is used as the propagation starting node. For example, the equipment capacity reduction event corresponds to the equipment node, the emergency order insertion event corresponds to the order node, and the material delay event corresponds to the material node.

[0087] S322. Search along the production relationship for production object nodes that are directly associated with the propagation starting node to obtain the first layer of associated nodes.

[0088] S323. Using the current layer's associated node as a new propagation node, continue searching for the next layer's associated node, and record the propagation level and propagation path corresponding to each associated node; terminate the corresponding propagation path when any of the following propagation termination conditions are met. The propagation termination conditions include: the current propagation level reaches the preset propagation level upper limit;

[0089] The current propagation node does not have any associated nodes that have not yet been searched;

[0090] The production relationship between the current propagation node and the next associated node is not one of the production relationships selected in step S31;

[0091] The production task corresponding to the next associated node has been completed, and its completion result will not be affected by the current disturbance;

[0092] The propagation path has returned to the previously searched nodes, and continued propagation will form a loop path.

[0093] S324. Select process nodes in the pending execution state from the associated nodes of each layer, determine the production task according to the order identifier and process identifier corresponding to each process node, and merge duplicate production tasks to obtain candidate tasks.

[0094] Specifically, process nodes in the pending execution state are selected from the associated nodes at each layer. The order identifier and process identifier corresponding to each process node are read to determine the production task. The order identifier is used to distinguish different orders, and the process identifier is used to distinguish different production processes in the same order. Production tasks that are repeatedly searched through different propagation paths and have the same order identifier and process identifier are merged, and the corresponding propagation level and propagation path information are retained to obtain candidate tasks. Candidate tasks are used to calculate the degree of task impact in the subsequent process, and affected tasks that need to be adjusted in scheduling are selected from them.

[0095] S33. Calculate the degree of influence of each candidate task, and determine the candidate tasks whose degree of influence reaches the preset influence threshold as the affected tasks, and summarize them to obtain the affected task group.

[0096] In one implementation, the node influence level of the propagation initiation node is set to 1. For candidate task j, its task influence level can be determined based on the influence level of preceding propagation nodes, the strength of production relationships, the task time buffer state, and the task delivery sensitivity, calculated using the following formula:

[0097]

[0098] in Indicate candidate tasks The degree of impact on the task; Indicates the propagation to the candidate task. The degree of influence of each preceding propagation node; the degree of influence is used to characterize the impact of an effective disturbance event propagating along the preceding propagation path to the corresponding node. Indicates the first One preceding propagation node and candidate task The intensity of production relations between them; Indicate candidate tasks The proportion of absorbable time buffer; Indicate candidate tasks Delivery sensitivity; This represents the number of preceding propagation nodes propagating to candidate task j. The node influence degree of the propagation starting node is set to 1. For preceding propagation nodes other than the propagation starting node, the node influence degree of the preceding propagation node is determined based on the node influence degree of the previous propagation node and the strength of the production relationship between adjacent nodes. When the current node has only one previous propagation node, the node influence degree of the previous propagation node is determined by multiplying its node influence degree by the strength of the production relationship between them. When the current node has multiple previous propagation nodes, the influence degrees formed by each previous propagation path are aggregated to obtain the node influence degree of the current node. Specifically, when the current node has multiple previous propagation nodes, the path influence degree corresponding to each propagation path is determined based on the node influence degree of each previous propagation node and the strength of the production relationship between it and the current node, and then aggregated according to the following formula: ,in, Indicates the current node The degree of influence of the nodes; Indicates the first The degree of influence of each propagating node in the next layer; Indicates the first Each propagation node at the previous level and the current node The intensity of production relations between them; This indicates the number of propagation nodes in the previous layer corresponding to the current node. This indicates a series of multiplication operations. Indicates will From 1 to Corresponding items Multiply sequentially. For example, when the current node has two parent nodes, the multiplication term is: .because and All values ​​are normalized to between 0 and 1, thus ensuring that the current node's influence level, obtained through aggregation, remains between 0 and 1. This aggregation method allows the impact of disturbances arriving at the same node from different propagation paths to be merged, preventing direct accumulation that could cause the node's influence level to exceed its normalized range. The obtained node influence level is then used for propagation at the next layer; when propagating to candidate tasks, its task influence level is determined by combining the candidate task's absorbable time buffer ratio and delivery sensitivity.

[0099] It should be explained that the strength of the production relationship is used to characterize the degree to which a disturbance propagates along the corresponding production relationship; the absorbable time buffer ratio is used to characterize the ability of a candidate task to absorb the impact of a disturbance by utilizing the remaining time buffer; and the delivery sensitivity is used to characterize the degree to which a candidate task is affected by order priority, remaining delivery time, and critical process path.

[0100] The degree of node impact, strength of production relations, proportion of time buffer that can be absorbed, and delivery sensitivity are all normalized to between 0 and 1. The greater the strength of production relations, the higher the delivery sensitivity, and the smaller the proportion of time buffer that can be absorbed, the higher the possibility that the candidate task will be actually affected by the disturbance. With the above value limit, the calculated degree of task impact is between 0 and 1.

[0101] The task impact level of each candidate task is compared with the preset impact threshold. When the task impact level of a candidate task reaches the preset impact threshold, the candidate task is identified as an affected task, and all affected tasks are summarized to obtain the affected task group.

[0102] Furthermore, after obtaining the affected task group, the production tasks corresponding to the affected task group are removed from the production tasks included in the current production plan, and the remaining production tasks are determined as the unaffected task group.

[0103] Obtain the original scheduling information of each production task in the unaffected task group. The original scheduling information includes at least the execution resources corresponding to the production task, the planned start time, the planned end time, and the task execution order.

[0104] Based on the original scheduling information, lock flags are set for the execution resources and planned execution times of each production task in the unaffected task group, ensuring they remain unchanged during subsequent local scheduling. The resource time slots and task time slots occupied by the locked unaffected task group are written into the local scheduling model, serving as local scheduling boundaries that cannot be occupied or changed when the affected task group performs resource allocation and execution timing adjustments.

[0105] As one implementation method:

[0106] In a precision valve body flexible manufacturing workshop, the workshop is equipped with six CNC machining centers M1 to M6, one heat treatment equipment H1, two coordinate measuring machines C1 and C2, three automated guided vehicles A1 to A3, as well as a shared tool magazine and a shared fixture magazine; the current production plan includes 32 orders and 146 production tasks to be executed or being executed, and the current rolling scheduling time domain is from 08:00 of the current day to 08:00 of the next day;

[0107] Among them, CNC machining centers M1 to M4 can perform valve body finishing tasks, while CNC machining centers M5 and M6 mainly perform roughing tasks; CNC machining center M4 can perform high-precision hole machining, but fixture F is required when performing this machining task. 27 and cutting tools T 08 Fixture F 27 In the current schedule, CNC machining center M2 is occupied until 11:50, tool T 08 The remaining processing time is 95 minutes; the coordinate measuring machine C1 is equipped with the inspection program QP-17, which can perform the final inspection task of the corresponding valve body;

[0108] At 10:20, equipment M3 showed increased spindle vibration for three consecutive sampling windows, with the vibration value increasing from 2.1 mm / s to 4.8 mm / s. Simultaneously, the maintenance log stated, "High-precision hole machining is restricted; current low-load operations are permitted." At 10:23, the system received an urgent order O requiring completion before 17:00 that day. 33 ;

[0109] The industrial big data model combines the status data of equipment M3, maintenance records, equipment capacity information, and order O. 33 Semantic recognition is performed on the process route to generate events indicating decreased equipment processing capacity and emergency order insertion. These events are then validated; the confidence level for the event corresponding to equipment M3 is 0.93, and for order O... 33 The confidence level of the corresponding events is 1.00, both reaching the event confirmation threshold of 0.80. Therefore, both are identified as valid disturbance events.

[0110] In response to the reduced processing capacity of equipment M3 and urgent orders O 33Insert events, the system uses device M3 node and order O respectively. 33 The node is used as a production object node, and production relationships such as equipment executable relationships, process sequence relationships, shared resource competition relationships, and quality inspection relationships are selected from the production semantic state diagram according to the event type.

[0111] The search proceeds along a limited hierarchical path of production relations, examining the high-precision processes originally planned for execution by equipment M3, the corresponding subsequent heat treatment and testing processes, and the connections to order O. 33 Production tasks that compete for fixtures, cutting tools, and inspection resources are used to generate candidate tasks.

[0112] The impact degree of each candidate task is calculated. In this embodiment, the preset impact threshold is set to 0.55. After screening, the impact degree of 11 processes reaches the impact threshold. Among them, the impact of 2 processes is propagated through the shared resource competition relationship corresponding to the shared fixture F27, resulting in an affected task group containing 11 processes. The production tasks corresponding to the affected task group are removed from the current production tasks, and the remaining production tasks are determined as unaffected task groups, while maintaining their original resource allocation and planned execution time.

[0113] S4. Based on the affected task group, initiate a capability query to the resource agent to obtain capability commitment data.

[0114] In this embodiment, it should be noted that the resource intelligence agent includes equipment intelligence agent, process intelligence agent, material intelligence agent, logistics intelligence agent and quality intelligence agent. Among them, the equipment intelligence agent is also used to uniformly manage the tool status and fixture status of the corresponding equipment. The capability commitment data is used to characterize the ability and preparation conditions of each resource to execute the corresponding production task within a specified time range, and serves as the resource basis for subsequently generating candidate collaborative strategies and candidate scheduling schemes.

[0115] Specifically, the task information of each production task in the affected task group is analyzed to obtain the process requirements, resource requirements, quality requirements, expected execution time and the connection requirements between previous and subsequent tasks for each production task, and the corresponding resource agent is determined according to the resource requirements of each production task.

[0116] Send capability query information to the resource agent. The capability query information includes at least the production task identifier, the process to be executed, the range of process parameters, the expected execution time, the resource preparation requirements, and the current production status version, so that different resource agents can make resource status judgments based on the same production status version.

[0117] After receiving the capability query information, each resource agent verifies the execution conditions of the production task based on the locally managed resource status and the production status version. Specifically, the equipment agent verifies the equipment processing capacity and tool and fixture status, the process agent verifies the process route and parameter range, the material agent verifies the material availability and supply time, the logistics agent verifies transportation resources, arrival time and route conflicts, and the quality agent verifies the testing capability, testing procedures and testing time window.

[0118] When the resource agent confirms that the resources it manages meet the execution conditions of the production task, it generates a corresponding capability commitment record, which includes at least the resource identifier, capability scope, available time window, preparation time, commitment validity period, and commitment version.

[0119] When a resource agent generates a capability commitment record, the production state version on which the record is based is recorded as the commitment version. Further, the capability commitment records returned by each resource agent are aggregated, and the commitment version and validity period of each record are verified. Capability commitment records whose commitment version matches the current production state version and are within their validity period are associated to obtain capability commitment data. Capability commitment records whose commitment version does not match the current production state version or whose validity period has expired are not used for subsequent candidate collaborative strategy generation.

[0120] As one implementation method:

[0121] For the finishing task of urgent order O33, the equipment agent confirms that equipment M4 has the corresponding high-precision hole machining capability, and the available time window for equipment M4 is from 12:00 to 15:20. Fixture F27 is released at 11:50, and the automated guided vehicle A2 can retrieve fixture F27 at 11:52 and deliver it to equipment M4 at 12:02. Fixture installation is expected to take 8 minutes; therefore, the overall preparation completion time for fixture F27 is 12:10. The quality agent confirms that inspection equipment C1 has a reserved inspection window. Each resource agent returns its corresponding capability range, available time, preparation time, and commitment validity period, which are then summarized to obtain the capability commitment data corresponding to order O33.

[0122] This step generates capability commitment data through each agent, so that the results returned by the resource agent not only reflect whether the resource has basic execution capabilities, but also reflect the resource's available time window, preparation time, and commitment validity status, thereby providing real and timely resource conditions for subsequent candidate cooperative strategy generation and local scheduling solution.

[0123] S5. Generate candidate coordination strategies based on the effective disturbance events, affected task groups, and capability commitment data, and form candidate scheduling schemes based on the candidate coordination strategies.

[0124] In this embodiment, it should be noted that the candidate coordination strategy is used to limit the candidate adjustment direction and the range of callable resources for the affected tasks, and the candidate scheduling scheme is used to clarify the execution resources, execution order and execution time corresponding to each affected task.

[0125] Specifically, step S5 includes the following processing procedures:

[0126] S51. Input the effective disturbance events, affected task groups and capacity commitment data into the industrial big data model, determine the candidate adjustment actions and candidate resources corresponding to each production task in the affected task group, and generate candidate coordination strategies.

[0127] By combining the industrial big data model with the process relationships, resource relationships and historical similar disturbance processing records in the production semantic state diagram, candidate adjustment actions are generated for each production task. The candidate adjustment actions include at least one of task migration, execution time adjustment, task sequence adjustment and related resource reservation.

[0128] Based on the processing capacity, process conditions, available time, and associated resources required for each candidate adjustment action, corresponding candidate resources are selected from the capacity commitment data. Candidate resources may include equipment resources, material resources, logistics resources, and quality inspection resources. Then, candidate coordination strategies are generated. The candidate coordination strategies adopt a structured expression of action objects, preconditions, target states, and prohibition conditions.

[0129] S52. Control the corresponding resource agent to negotiate according to the candidate collaborative strategy to obtain negotiation data.

[0130] S521. Send the candidate adjustment actions corresponding to each production task to the resource agent corresponding to the candidate resource.

[0131] S522. Each resource agent verifies the execution conditions of the candidate adjustment actions based on the corresponding capability commitment data, and feeds back candidate resource information and constraint information.

[0132] The equipment intelligence agent verifies the equipment's processing capacity, available time, remaining tool life, fixture arrival time, and processing program status; the material intelligence agent verifies the material completeness status and estimated arrival time; the logistics intelligence agent verifies the transportation equipment's available status, material picking time, arrival time, and path conflict status; and the quality intelligence agent verifies the testing capabilities, testing procedures, and available testing windows.

[0133] When a candidate resource meets the execution conditions of a candidate adjustment action, the corresponding resource agent feeds back candidate resource information and constraint information. The candidate resource information includes at least the resource identifier and available time window; the constraint information is used to characterize the resource preparation, task connection and commitment validity period conditions that the candidate adjustment action needs to meet; when a candidate resource does not meet the necessary execution conditions, the corresponding resource agent feeds back rejection information and will no longer use the candidate resource for subsequent local scheduling.

[0134] S523. For candidate adjustment actions that pass the execution condition verification, determine the resource coordination cost. The resource coordination cost can be determined based on resource preparation time, logistics connection time, and capacity commitment risk.

[0135] S524. The candidate resource information, constraint information and corresponding resource coordination costs are summarized to obtain negotiation data. The resource coordination costs can participate in the evaluation of different resource allocation combinations in the subsequent local scheduling solution.

[0136] S53. Perform local scheduling solution based on the negotiated data to obtain candidate scheduling schemes.

[0137] S531. Analyze the negotiated data to determine the candidate resources, scheduling constraints, and resource coordination costs corresponding to each production task. The scheduling constraints are those that can be directly determined in the local scheduling solution stage, including at least equipment processing capacity, process sequence relationship, fixed scheduling boundary formed by unaffected tasks, resource availability time window, and the inability of the same resource to be occupied repeatedly. For conditions that need to be further verified in conjunction with the task progress process, such as fixture transfer and installation completion time, actual material arrival status, CNC program verification status, and dynamic connection status of detection window, these are determined as execution verification conditions and associated with and saved with the corresponding candidate scheduling scheme.

[0138] S532. Taking the affected task groups as the objects to be adjusted, the original scheduling information corresponding to the unaffected task groups as the fixed scheduling boundary, the candidate resources corresponding to each production task as the resource allocation range, the scheduling constraints as the solution constraints, and the resource coordination cost as the resource selection evaluation parameter, a local scheduling model is constructed.

[0139] S533. Based on the local scheduling model, the resource allocation, execution order and execution time of each production task in the affected task group are jointly solved to obtain multiple candidate scheduling schemes.

[0140] It should be explained that the local scheduling model can be solved using mixed integer programming, constraint programming, heuristic search, or reinforcement learning to determine the execution resources, execution order, and execution time of each production task. By retaining multiple solution results that satisfy the scheduling constraints but have different resource allocations or execution times, multiple candidate scheduling schemes can be obtained.

[0141] For each resource allocation and execution sequence combination obtained from the solution, it is verified whether it meets the requirements of process sequence, resource occupancy status, resource availability time window, material arrival time, and task connection. In the local scheduling solution stage, the resource allocation and execution sequence combinations that satisfy the scheduling constraints are retained, and the execution verification conditions are written into the candidate scheduling scheme for event-by-event verification by subsequent local execution pre-runs.

[0142] As one implementation method:

[0143] The industrial big data model inputs the event of decreased high-precision machining capability of equipment M3, the event of insertion of emergency order O33, the affected task groups, and the capability commitment data returned by each resource agent. The industrial big data model generates three types of candidate collaborative strategies based on the disturbance type, task delivery requirements, and capability commitment data:

[0144] The first type of candidate collaborative strategy focuses on reducing order delays, allowing the O33 finishing process to be migrated to equipment M4, and allowing for a wide range of execution time adjustments for the affected processes.

[0145] The second type of candidate collaborative strategy takes the balance between resource readiness status and delivery requirements as the main adjustment direction, and limits equipment M4, fixture F27, cutting tool T08, automated guided vehicle A2 and testing equipment C1 as candidate resources.

[0146] The third type of candidate coordination strategy focuses on reducing changes to the original schedule and prioritizes maintaining the original resource allocation and execution order of other affected processes.

[0147] Based on the aforementioned candidate collaborative strategies, the relevant resource agents are controlled to perform execution condition verification, obtaining candidate resource information, constraint information, and resource collaboration costs. Subsequently, local scheduling is solved based on the negotiated data, generating multiple candidate scheduling schemes. Among them, candidate scheduling scheme A will O 33 The finishing process is assigned to equipment M4, with a planned start time of 12:00; candidate scheduling scheme B will... 33 The finishing process is assigned to equipment M4, with a planned start time of 12:10. The intended usage times for fixtures, cutting tools, transportation, and inspection resources are included in the scheduling plan. Candidate scheduling plan C is postponed to O. 33 The finishing process should be carried out, and the original resource allocation and execution order of other affected processes should be maintained as much as possible.

[0148] This step generates candidate coordination strategies by jointly analyzing effective disturbance events, affected task groups, and capacity commitment data. This limits candidate adjustment actions and available resources to a reasonable range, avoiding indiscriminate searching of all production tasks and all production resources.

[0149] S6. Perform a partial execution rehearsal of the candidate scheduling schemes to determine the target scheduling scheme.

[0150] Specifically, S6 includes the following processing steps:

[0151] S61. Construct a local execution environment based on the affected task group, and write each of the candidate scheduling schemes into the local execution environment.

[0152] Based on the affected task groups, extract the preceding and following processes, equipment, tools and fixtures, materials, logistics and testing resources corresponding to each production task from the production semantic state diagram, and obtain the status data of each associated resource under the current production state version to construct a local execution environment; write the resource time periods occupied by unaffected tasks as fixed occupancy status into the local execution environment.

[0153] It should be explained that the local execution environment only includes production tasks and resources that are directly related to the execution process of the affected task group, and therefore does not repeat the simulation of all production processes corresponding to the unaffected task group.

[0154] S62. Advance the local execution environment according to each of the candidate scheduling schemes, perform a pre-run of the production tasks, and obtain the pre-run results corresponding to each candidate scheduling scheme.

[0155] S621. Based on the resource requirements, planned execution time, and simulated status of associated resources for each production task in the current candidate scheduling scheme, perform resource readiness analysis to determine the readiness level of the current candidate scheduling scheme.

[0156] It is important to know that related resources include equipment resources, tooling and fixture resources, material resources, logistics resources, and quality inspection resources; resource demand conditions include at least one of the following: capacity matching conditions, quantity satisfaction conditions, time satisfaction conditions, and occupancy status conditions.

[0157] The simulated resource status of each associated resource during the production task plan execution period is extracted from the local execution environment. The simulated resource status is compared with the corresponding resource demand conditions one by one to obtain the condition satisfaction value corresponding to each resource demand condition. Then, the lowest value among the condition satisfaction values ​​corresponding to the same associated resource is determined as the resource satisfaction degree of the associated resource for the corresponding production task.

[0158] It should be added that, for the same candidate scheduling scheme, if the same associated resource corresponds to multiple production tasks, the lowest value among the resource satisfaction levels of the associated resource for each production task is determined as the resource readiness level of the associated resource. Furthermore, the lowest value among the resource readiness levels of all associated resources involved in the current candidate scheduling scheme is determined as the readiness level of the current candidate scheduling scheme. By using the lowest value for aggregation, it is possible to avoid the high satisfaction level of one resource compensating for the unreadiness of other key resources.

[0159] As one implementation method:

[0160] For the quantity to meet the condition, the smaller of the ratio of the actual available resources to the production task demand and 1 is determined as the condition-met value;

[0161] For time-based conditions, the condition satisfaction value is determined based on the degree of overlap between the available resource time window and the planned usage period of the production task, as well as whether the resource preparation completion time is earlier than the planned start time.

[0162] For the occupancy status condition, the condition value is 1 if the planned usage period does not overlap with the resource occupancy period of other production tasks, and 0 otherwise.

[0163] S622. Calculate the total delay and completion span based on the rehearsal execution time of each production task in the current candidate scheduling scheme. Compare the candidate scheduling scheme with the original scheduling information to determine the number of changed processes. The total delay is the sum of the delay durations of each order's rehearsal completion time exceeding the corresponding delivery deadline. For orders that do not exceed the delivery deadline, the delay duration is 0. The completion span is the time difference between the earliest rehearsal start time and the latest rehearsal completion time in the affected task group. The number of changed processes is the number of processes whose execution resources, execution order, or planned execution time have changed compared to the original scheduling information. If there are multiple changes to the same process, it is counted only once.

[0164] S623. Summarize the total delay, completion span, number of changed procedures, and readiness level to obtain the pre-run results corresponding to the current candidate scheduling scheme.

[0165] S63. Based on the pre-simulation results, candidate scheduling schemes that have execution conflicts or do not meet scheduling constraints are eliminated, and the remaining candidate scheduling schemes are comprehensively evaluated to obtain the target scheduling scheme.

[0166] Based on the results of the rehearsal, candidate scheduling schemes that have resource conflicts, untimely resource preparation, abnormal task connection, or do not meet delivery constraints will be eliminated.

[0167] The total delay, completion span, and number of changed processes of the remaining candidate scheduling schemes are used as negative evaluation indicators, and the readiness level is used as a positive evaluation indicator. The evaluation indicators are standardized, and the comprehensive evaluation value of each remaining candidate scheduling scheme is calculated according to the preset evaluation weight. The smaller the total delay, completion span, and number of changed processes, the less the production task is delayed, the shorter the local scheduling cycle, and the smaller the change to the original schedule. The higher the readiness level, the more likely the associated resources are to complete the execution preparation according to the candidate scheduling scheme.

[0168] The candidate scheduling scheme with the best comprehensive evaluation is determined as the target scheduling scheme; when only one candidate scheduling scheme passes the pre-run, it is directly determined as the target scheduling scheme.

[0169] As one implementation method:

[0170] Based on the production semantic state diagram and current production state data, 11 affected processes and their related local resource states are copied, including equipment M4, fixture F27, tool T08, automated guided vehicle A2 and testing equipment C1, etc. The equipment time interval and resource time window occupied by unaffected tasks are written into the local execution environment as immutable occupancy, and the local execution environment is constructed.

[0171] When the rehearsal time reached 12:00, equipment M4 was ready to execute, but fixture F27 had not yet been transferred and installed, and the local execution environment would be O. 33 The finishing process remains in a waiting state, and resource insecurity conflicts are recorded. To determine the impact of these conflicts on subsequent tasks, the rehearsal clock continues to advance to 12:10. After fixture F27 is installed, the execution process of the finishing process and subsequent tasks is simulated, thereby obtaining the total delay, completion span, number of changed processes, and readiness level corresponding to scheme A. Although scheme A can complete the rehearsal after the task is postponed, its originally planned start time is inconsistent with the actual resource readiness time. Therefore, it is determined that there is an execution conflict and is eliminated.

[0172] After completing the rehearsal of Plan A, the local execution environment is restored to its initial state before the rehearsal, and Plan B is written into the local execution environment. During the rehearsal according to Plan B, the automated guided vehicle A2 retrieves fixture F27 at 11:52 and delivers it to equipment M4 at 12:02. Before the scheduled start time of 12:10, fixture installation, CNC program verification, and tool status confirmation are all completed. Equipment M4 is able to perform the O33 finishing process within the promised time window, the remaining life of tool T08 can cover the machining process, and subsequent processes can use the inspection window reserved by C1.

[0173] Restore the initial state of the local execution environment and conduct an independent pre-run of scheme C.

[0174] The preliminary results of the candidate scheduling schemes are shown in the table below, where the completion span is the time difference between the earliest start time and the latest completion time of the preliminary rehearsal in the affected task group:

[0175] Option A 72 1035 11 0.82 The F-27 had not completed its transfer and installation by 12:00, and the rehearsal was rejected. Option B 88 1048 8 0.96 The rehearsal verification passed, and O33 is expected to complete at 16:28. Option C 105 1060 5 0.91 O33 was scheduled to complete at 17:22, violating delivery constraints.

[0176] According to the table above, Option A was eliminated because fixture F27 was not ready as planned, Option C was eliminated because it violated order delivery constraints, and Option B met the resource, process and delivery constraints. Therefore, Option B was determined as the target scheduling option.

[0177] S7. Obtain the execution feedback of the target scheduling scheme, identify the state deviation of the execution feedback, and obtain the production scheduling result.

[0178] S71. Prepare and confirm the versioning of the target scheduling scheme and submit it for execution.

[0179] Before obtaining execution feedback of the target scheduling scheme, a target scheduling version is generated for the target scheduling scheme, and a preparation instruction is sent to the relevant resource agents. Each resource agent completes resource reservation and pre-execution status confirmation, and returns a preparation certificate carrying the target scheduling version and validity period.

[0180] When all necessary preparation credentials are of the same version, within their validity period, and the corresponding resource status has not changed, the target scheduling scheme will be submitted for execution; otherwise, resources that have not completed preparation or whose preparation credentials have expired will be rescheduled locally.

[0181] S72. After the target scheduling plan is executed, the local repair task group is determined.

[0182] Continuously acquire the actual execution status of each production task and related resources, compare the actual start time, material arrival time and inspection completion time with the corresponding promised time windows, and compare the actual status of the equipment with the corresponding promised capacity range;

[0183] When any actual state exceeds the allowable deviation range, exceeds the promised time window, or falls below the promised capability requirements, the corresponding capability commitment is marked as failed. The resource node corresponding to the failed resource in the production semantic state diagram is used as the propagation starting node, and the impact propagation analysis is re-performed to determine the local repair task group.

[0184] S73. Perform local scheduling and repair on the local repair task group to obtain the production scheduling result.

[0185] For the local repair task group, a new capability query is initiated with the relevant resource agents to obtain the update capability commitment data corresponding to the current production state version. Then, a local repair scheduling scheme is generated according to the aforementioned candidate collaboration strategy generation, resource agent negotiation, local scheduling solution and local execution rehearsal process.

[0186] When no state deviation is identified, the target scheduling plan continues to be executed, and the target scheduling plan and its current execution status are determined as the production scheduling result. When a state deviation is identified, the local repair scheduling plan is re-prepared and confirmed as a version, and submitted for execution when all necessary preparation documents are consistent and valid. The submitted local repair scheduling plan and the updated execution status are determined as the production scheduling result.

[0187] This invention also provides a multi-agent collaborative production scheduling system based on a large industrial model, such as... Figure 2 As shown, it includes:

[0188] The state diagram construction module is used to acquire production state data and construct a production semantic state diagram;

[0189] The disturbance analysis module is used to perform semantic recognition on production disturbance information based on the industrial big data model to obtain valid disturbance events;

[0190] The impact identification module is used to perform impact propagation analysis on valid disturbance events in the production semantic state graph to obtain the affected task groups.

[0191] The multi-agent negotiation module is used to initiate capability queries to the resource agent based on the affected task group and obtain capability commitment data;

[0192] The collaborative scheduling module is used to generate candidate collaborative strategies based on the effective disturbance events, affected task groups, and capability commitment data, and to form candidate scheduling schemes based on the candidate collaborative strategies.

[0193] The pre-execution verification module is used to perform partial execution pre-execution of the candidate scheduling scheme to determine the target scheduling scheme;

[0194] The feedback repair module is used to obtain the execution feedback of the target scheduling scheme, identify the state deviation of the execution feedback, and obtain the production scheduling result.

[0195] Based on the foregoing description in conjunction with the accompanying drawings, those skilled in the art will understand that the embodiments of this application can also be implemented by software programs. Therefore, this application also provides a computer-readable storage medium. This computer-readable storage medium stores computer-readable instructions thereon, which, when executed by one or more processors, implement the method described above in conjunction with the accompanying drawings.

[0196] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the application. As used in this specification and claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this specification and claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes such combinations.

[0197] Although the embodiments of this application are as described above, the content is merely an example adopted for the purpose of facilitating understanding of this application and is not intended to limit the scope and application scenarios of this application. Any person skilled in the art described in this application may make any modifications and changes in the form and details of the implementation without departing from the spirit and scope disclosed in this application, but the scope of patent protection of this application shall still be determined by the scope defined in the appended claims.

Claims

1. A multi-agent collaborative production scheduling method based on a large industrial model, characterized in that, include: Acquire production status data and construct a production semantic state graph; Semantic recognition of production disturbance information is performed based on a large industrial model to obtain valid disturbance events; In the production semantic state diagram, the impact propagation analysis of effective disturbance events is performed to obtain the affected task groups; Based on the affected task group, a capability query is initiated to the resource agent to obtain capability commitment data; Candidate coordination strategies are generated based on the effective disturbance events, affected task groups, and capability commitment data, and candidate scheduling schemes are formed based on the candidate coordination strategies. Perform partial execution rehearsals on the candidate scheduling schemes to determine the target scheduling scheme; Obtain the execution feedback of the target scheduling scheme, identify the state deviation of the execution feedback, and obtain the production scheduling result.

2. The multi-agent collaborative production scheduling method based on a large industrial model according to claim 1, characterized in that, An impact propagation analysis of effective disturbance events is performed on the production semantic state diagram to identify the affected task groups, including: Based on the event type of the effective disturbance event, select the production relationship that matches the event type from the production semantic state diagram; The process is propagated hierarchically along the production relations to obtain candidate tasks; The degree of influence of each candidate task is calculated, and the candidate tasks whose degree of influence reaches the preset influence threshold are identified as affected tasks, and the affected task group is obtained by summarizing them.

3. The multi-agent collaborative production scheduling method based on a large industrial model according to claim 2, characterized in that, The process is propagated hierarchically along the production relations to obtain candidate tasks, including: Based on the event object in the effective disturbance event, determine the corresponding production object node in the production semantic state graph, and use the production object node as the propagation start node; Search along the production relationship for production object nodes directly associated with the propagation starting node to obtain the first layer of associated nodes; Use the current layer's associated node as the new propagation node, continue searching for the next layer's associated node, and record the propagation level and propagation path corresponding to each associated node; The process nodes corresponding to the production tasks to be executed are selected from the associated nodes at each layer. The corresponding production tasks are determined according to each process node, and the production tasks are deduplicated to obtain candidate tasks.

4. The multi-agent collaborative production scheduling method based on a large industrial model according to claim 2, characterized in that, After obtaining the affected task groups, lock the unaffected task groups, including: Remove the production tasks corresponding to the affected task groups from the current production tasks to obtain the unaffected task groups; Obtain the original scheduling information of each production task in the unaffected task group; Based on the original scheduling information, the resource allocation and planned execution time of each production task are locked, and the locked unaffected task group is used as the local scheduling boundary.

5. The multi-agent collaborative production scheduling method based on a large industrial model according to claim 4, characterized in that, Based on the effective disturbance events, affected task groups, and capability commitment data, candidate coordination strategies are generated, and candidate scheduling schemes are formed based on the candidate coordination strategies, including: Input the effective disturbance events, affected task groups, and capacity commitment data into the industrial big data model to determine the candidate adjustment actions and candidate resources corresponding to each production task in the affected task group, and generate candidate coordination strategies. Based on the candidate collaborative strategy, the corresponding resource agent is controlled to negotiate and obtain negotiation data; Based on the negotiated data, a local scheduling solution is performed to obtain a candidate scheduling scheme.

6. The multi-agent collaborative production scheduling method based on a large industrial model according to claim 5, characterized in that, Based on the candidate collaborative strategy, the corresponding resource agent is controlled to negotiate and obtain negotiation data, including: Send the candidate adjustment actions corresponding to each production task to the resource agent corresponding to the candidate resource; Each resource agent verifies the execution conditions of the candidate adjustment actions based on the corresponding capability commitment data, and feeds back candidate resource information and constraint information. For candidate adjustment actions that pass the execution condition verification, determine the resource coordination cost; The candidate resource information, constraint information, and corresponding resource coordination costs are summarized to obtain the negotiation data.

7. The multi-agent collaborative production scheduling method based on a large industrial model according to claim 6, characterized in that, Based on the negotiated data, a local scheduling solution is performed to obtain candidate scheduling schemes, including: The negotiated data is analyzed to determine the candidate resources, scheduling constraints, and resource coordination costs corresponding to each production task. Using the affected task group as the object to be adjusted, the original scheduling information corresponding to the unaffected task group as the fixed scheduling boundary, the candidate resources corresponding to each production task as the resource allocation range, the scheduling constraints as the solution constraints, and the resource coordination cost as the resource selection evaluation parameter, a local scheduling model is constructed. Based on the local scheduling model, the resource allocation, execution order, and execution time of each production task in the affected task group are jointly solved to obtain multiple candidate scheduling schemes.

8. The multi-agent collaborative production scheduling method based on a large industrial model according to claim 7, characterized in that, Perform partial execution rehearsals of the candidate scheduling schemes to determine the target scheduling scheme, including: A local execution environment is constructed based on the affected task group, and each of the candidate scheduling schemes is written into the local execution environment. The local execution environment is advanced according to each of the candidate scheduling schemes, and the production tasks are rehearsed to obtain the rehearsal results corresponding to each candidate scheduling scheme. Based on the pre-simulation results, candidate scheduling schemes that have execution conflicts or do not meet scheduling constraints are eliminated, and the remaining candidate scheduling schemes are comprehensively evaluated to obtain the target scheduling scheme.

9. The multi-agent collaborative production scheduling method based on a large industrial model according to claim 8, characterized in that, The local execution environment is advanced according to each candidate scheduling scheme, and a pre-simulation of the production tasks is performed to obtain the pre-simulation results corresponding to each candidate scheduling scheme, including: Based on the resource requirements, planned execution time, and simulated status of associated resources for each production task in the current candidate scheduling scheme, a resource readiness analysis is performed to determine the readiness level of the current candidate scheduling scheme. Calculate the total delay and completion span based on the pre-execution time of each production task in the current candidate scheduling scheme, compare the current candidate scheduling scheme with the original scheduling information, and determine the number of changes to the process. The total delay, completion span, number of changed procedures, and readiness level are summarized to obtain the pre-rehearsal results corresponding to the current candidate scheduling scheme.

10. A multi-agent collaborative production scheduling system based on an industrial large-scale model, used to execute the multi-agent collaborative production scheduling method as described in any one of claims 1-9, characterized in that, include: The state diagram construction module is used to acquire production state data and construct a production semantic state diagram; The disturbance analysis module is used to perform semantic recognition on production disturbance information based on the industrial big data model to obtain valid disturbance events; The impact identification module is used to perform impact propagation analysis on valid disturbance events in the production semantic state graph to obtain the affected task groups. The multi-agent negotiation module is used to initiate capability queries to the resource agent based on the affected task group and obtain capability commitment data; The collaborative scheduling module is used to generate candidate collaborative strategies based on the effective disturbance events, affected task groups, and capability commitment data, and to form candidate scheduling schemes based on the candidate collaborative strategies. The pre-execution verification module is used to perform partial execution pre-execution of the candidate scheduling scheme to determine the target scheduling scheme; The feedback repair module is used to obtain the execution feedback of the target scheduling scheme, identify the state deviation of the execution feedback, and obtain the production scheduling result.

Images