A surgical set distribution scheduling method based on a multi-agent system

Abstract

The present application relates to computer data processing, resource scheduling and multi-agent collaborative decision-making technical field, specifically to a kind of surgical set package distribution scheduling method based on multi-agent system. Get surgery scheduling, set package inventory, sorting load, distribution robot state and passage occupation information, establish surgery task, set package, sorting resource, distribution resource and operation room association, and according to the latest delivery time reverse generation time limit commitment chain covering set package, sorting, pick-up, distribution and receiving link;Distributed quotation is formed by candidate resource in combination with local state, task new occupation time, existing task adjustable time and switching occupation, and resource allocation is determined according to time redundancy occupation relationship;When state changes, update residual redundancy and propagate along time limit commitment chain, implement local renegotiation and commitment version update for affected tasks, reduce time accumulation compression caused by local resource adjustment of task chain, maintain time coordination of execution link, and improve set package punctual delivery stability.

Description

Technical Field

[0001] This invention relates to the fields of computer data processing, resource scheduling, and multi-agent collaborative decision-making, specifically a method for scheduling the delivery of surgical kits based on a multi-agent system. Background Technology

[0002] Existing surgical kit delivery scheduling technologies typically employ task allocation methods based on task priority and fixed process rules. This involves reading surgical time, surgical procedure type, and kit demand information from the surgical scheduling system, determining task priorities based on the surgical start time and kit preparation requirements, and then assigning tasks according to inventory availability, sorting equipment idleness, and the current location of delivery robots. In this approach, the system usually treats kit preparation, sorting, and delivery as interconnected but relatively independent execution stages, determining the processing order of each stage through preset scheduling rules. For example, a kit preparation time window is pre-set based on the surgical start time; when inventory conditions are met, nearby or currently idle sorting equipment and delivery robots are prioritized to complete the task. When multiple surgical tasks exist simultaneously, time sorting, distance sorting, or task priority sorting are typically used to select the execution target, and resource allocation is achieved by adjusting the task queue. This scheduling method can complete basic task management and resource allocation, but its scheduling process mainly relies on pre-set rules and lacks dynamic analysis of the time constraints between kit processing, sorting, delivery, and in-operative reception, making it difficult to describe the impact of different resource occupancy behaviors on the overall task chain.

[0003] Some existing improvement schemes introduce the concept of multi-entity collaborative scheduling, dividing inventory management, equipment scheduling, and logistics distribution processes into multiple independent execution entities. Each entity selects tasks and competes for resources based on its own status information. For example, the inventory management entity selects available kits based on the quantity and expiration date of consumables, the sorting entity determines the processing order based on the current equipment load, and the distribution entity selects transportation tasks based on robot location and path status, achieving local negotiation through task requests and resource feedback. Compared to fixed-rule scheduling, this method increases the ability to handle changes in resource status. However, each entity typically makes decisions based on its own local status, and resource pricing or task selection focuses more on the current task completion cost, lacking a unified description of time redundancy changes in the entire surgical kit distribution chain. When a link is completed ahead of schedule or delayed, existing schemes usually only adjust the single link that has changed, such as reassigning delivery robots or adjusting the sorting order, without passing the time change to other related tasks. This results in subsequent tasks still executing according to the original plan, leading to a discrepancy between local adjustments and overall performance constraints.

[0004] The lack of a collaborative scheduling mechanism based on overall time constraints in the various execution stages of the surgical kit delivery process leads to individual resource decisions failing to accurately reflect their impact on the overall task chain's time status. In actual scheduling, there are continuous time dependencies between kit assembly, sorting, robot delivery, and operating room reception. Any additional time spent in any stage can compress the execution time of subsequent stages. Existing scheduling methods cannot establish time commitment relationships between stages, nor can they dynamically assess changes in time redundancy within the task chain based on resource competition behavior. When sorting equipment processes a task ahead of schedule but delivery resources are insufficient, the pre-released time cannot be utilized by other tasks. When the delivery robot is delayed, related tasks other than the affected task continue to compete for resources according to the original plan, easily causing the time window for subsequent tasks to be continuously compressed. Therefore, existing solutions struggle to maintain time coordination between execution stages under dynamic conditions of multiple tasks and resources, resulting in a mismatch between local resource scheduling and overall on-time delivery constraints during the surgical kit delivery process. Summary of the Invention

[0005] The purpose of this invention is to provide a surgical kit delivery scheduling method based on a multi-agent system, which can solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] A method for scheduling the delivery of surgical kits based on a multi-agent system includes: acquiring surgical schedules, surgical procedures, kit inventory, sorting load, delivery robot task status and channel occupancy information, and establishing task associations between surgical tasks, kits, inventory, sorting resources, delivery resources and operating rooms;

[0008] Based on the latest delivery time of the surgical task, a time limit commitment chain covering the complete set, sorting, picking up, delivery and receiving stages is generated in reverse.

[0009] The participating resources form a distributed bid based on their local execution status, and the resource allocation result is determined by combining the task time redundancy occupancy relationship corresponding to the bid.

[0010] When the execution status changes, the remaining redundancy of each stage is updated, and the updated redundancy constraints are propagated along the time commitment chain. The subsequent resource allocation is then adjusted on a rolling basis according to the updated redundancy constraints.

[0011] Preferably, establishing the task association includes: taking each surgical task as a task node, taking kit assembly, sorting, robot pickup, delivery and transfer and operating room reception as sequential execution nodes, and taking available inventory batches, available sorting time periods, available delivery robot execution time periods and available passageway time periods as resource nodes.

[0012] The task nodes are connected to the sequential execution nodes according to their execution order, and the sequential execution nodes are connected to the resource nodes according to their resource availability. Competitive relationships are established for tasks that share the same resource and have overlapping execution times, forming a task-resource association graph that is updated synchronously with the scheduling and resource status.

[0013] Preferably, generating the time commitment chain includes: taking the latest delivery time of the surgical task as the final commitment boundary, reading the current queue occupancy, committed tasks and executable time windows of each link in the reverse order of receiving, delivering, picking up, sorting and assembling in the operating room, and determining the latest delivery boundary between adjacent links level by level.

[0014] For adjacent delivery boundaries of the same task, set up before-after dependencies, limit the time interval allowed for upstream links to before the downstream delivery boundary, and write the delivery boundaries of each link and the adjustable time between the boundaries into the commitment status record of the corresponding task.

[0015] Preferably, forming the distributed quotation includes: reading the commitment status record of the task to be assigned and its own set of committed tasks by the candidate sorting resource or the candidate delivery resource; performing a local rearrangement of the execution sequence after inserting the task to be assigned; and determining the additional time occupied by the task to be assigned, the adjustable time of the compressed existing task, the connection time occupied when the task transfer occurs, and the change in the delivery boundary of each task.

[0016] The newly added occupied time, adjustable time, connection occupied time and delivery boundary change are encoded into the same quotation vector, and the quotation vector is bound to the corresponding executable time period and submitted to the task node corresponding to the surgical task.

[0017] Preferably, updating the task resource association graph includes: when the inventory batch status, sorting queuing order, robot task sequence, or channel occupancy interval changes, only the sequential execution nodes directly connected to the changed resource nodes are recalculated, and the process is extended upstream and downstream along the task execution sequence to the first sequential execution node whose resource occupancy relationship has not changed;

[0018] The original task competition relationships outside the extended scope are retained. Resource competition relationships that are added, removed, or have changed time overlap relationships within the extended scope are re-marked, and the updated version number of the association graph is written into the commitment status record of the corresponding task. When submitting a bid, the version number carried by the bid is verified to be consistent with the updated version number of the association graph in the commitment status record.

[0019] Preferably, propagating the redundancy constraint includes: comparing the actual completion time, the current estimated completion time, and the delivery boundary of each sequential execution node to obtain the remaining adjustable time for the corresponding sequential execution node;

[0020] When an upstream sequential execution node releases its time ahead of schedule, the release amount is limited to the committed interval that has not yet been occupied by the adjacent downstream sequential execution nodes and transferred level by level. When the upstream sequential execution node occupies its time late, the amount is deducted sequentially according to the existing adjustable time of the adjacent sequential execution nodes.

[0021] Based on the surgical risk level, a protection period is allocated within the adjustable time of the corresponding task that does not participate in bidding for other tasks. After each propagation, the effective time interval for each sequential execution node to participate in the next round of resource competition is synchronously adjusted.

[0022] Preferably, determining the resource allocation result based on the quotation vector includes: verifying the delivery boundary of the task to be allocated, the delivery boundary of the existing task, and the protection time period for the execution sequence corresponding to the candidate quotation, and eliminating the candidate quotation that causes any delivery boundary mismatch;

[0023] The remaining bids are sorted into a bid sequence based on the additional time occupied by the tasks to be assigned, the adjustable time consumption of existing tasks, the time occupied by task transfer and connection, and the delivery boundary compression. When multiple bids share resources, a mutual exclusion selection relationship is established, and a task allocation set corresponding to the executable time periods of each resource is determined from the bid combination that does not have resource time period conflicts.

[0024] Preferably, the local rearrangement includes: keeping the order of tasks with immovable delivery boundaries in the committed tasks of the candidate resources unchanged, inserting the tasks to be assigned into the candidate gaps formed by the remaining movable tasks, and simultaneously determining the pickup connection time, delivery transfer time and resource empty running time of the tasks before and after the insertion position.

[0025] When the task to be assigned is completed across two execution entities, the arrival time window corresponding to the handover point is written into the commitment status record of the preceding and following execution entities respectively. Only the local rearrangement sequence with a common handover interval between the arrival time windows and without intruding into the protection time period is retained, and the local rearrangement sequence is encapsulated into the execution sequence corresponding to the candidate bid.

[0026] Preferably, when the remaining adjustable time of any sequential execution node is less than zero, a set of tasks that share resources with the sequential execution node and whose delivery boundaries are associated with its time occupation are extracted according to the task resource association graph, and the resource allocation outside the task set is frozen.

[0027] Within the task set, releaseable time intervals are identified according to the propagation direction of the time commitment chain based on the negative remaining adjustable time amount. Bid requests are then reissued only to candidate sorting or delivery resources with the releaseable time intervals. The reissued bids are then compared with the original commitment status records for version verification before updating the resource allocation of the corresponding sequential execution nodes and their subsequent sequential execution nodes.

[0028] Preferably, for the rolling adjusted resource allocation maintenance commitment version sequence, each commitment version records the task resource association graph version, time commitment chain version, quotation vector source, resource executable time period and the identifier of the previous commitment version that was replaced;

[0029] When receiving new inventory, sorting, delivery robot or channel status, the corresponding status timestamp is compared with the generation time of the current committed version, and an alternative offer is formed only based on status data whose timestamp is not earlier than the generation time of the current committed version;

[0030] When multiple local adjustments involve the same resource time period, the adjustment scope is merged according to the dependencies between committed versions, and candidate allocations with circular resource occupancy dependencies are rolled back to the most recent committed version without resource conflicts before the resource allocation is re-determined.

[0031] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0032] 1. By establishing task relationships between surgical tasks, kits, inventory, sorting resources, delivery resources, and operating rooms, the execution stages of the surgical kit delivery process are transformed into a continuous task chain with time dependencies. The time commitment boundaries for kit preparation, sorting, retrieval, delivery, and receiving stages are determined in reverse order based on the latest delivery time of the surgical task, enabling all executing entities to participate in resource competition based on unified time constraints. When submitting task quotations, resource entities no longer rely solely on their own execution time but consider the impact of task insertion on existing task time redundancy, delivery boundaries, and resource transfer times to form a comprehensive quotation. When the actual execution time of a stage changes, the corresponding time redundancy change is propagated along the task chain, ensuring that related resources are updated with their executable time ranges synchronously. Through this time commitment chain and resource competition coordination mechanism, individual resource adjustments are incorporated into the overall task chain constraints, reducing subsequent time compression caused by local scheduling and maintaining continuous time coordination in the kit preparation, sorting, and delivery processes.

[0033] 2. By combining task association relationships and time commitment chains, the system can adjust the time allocation relationship between tasks according to changes in resource status. This allows the time released by early-completed tasks to participate in subsequent task scheduling, and ensures that delayed tasks are promptly reflected in the resource competition process of related tasks. Introducing task time redundancy in the distributed bidding process allows the task selection results submitted by each resource entity to simultaneously reflect current execution capabilities and their impact on the commitment boundaries of other tasks, reducing overall task conflicts caused by optimal selection of a single resource. When multiple surgical tasks are executed in parallel, maintaining time commitment relationships between different tasks can limit the occupation of critical time windows by low-urgency tasks and adjust the resource allocation range according to the task chain status. Simultaneously, based on commitment version and associated status update mechanisms, resource changes do not require reordering all tasks; instead, the status of relevant tasks is updated according to the impact range of the changed resources, ensuring that the task adjustment range during scheduling corresponds to actual resource changes. This process improves the task status transmission method in a multi-resource collaborative environment, enabling continuous data association between resource competition, time constraints, and task execution status in the package delivery scheduling process. Attached Figure Description

[0034] Figure 1 A flowchart illustrating the overall scheduling process for the delivery of surgical kits based on a multi-agent system;

[0035] Figure 2 A time-limited commitment chain boundary recursive calculation tracing graph;

[0036] Figure 3 A flowchart of distributed pricing and partial resource rearrangement under time-limited commitment chain constraints;

[0037] Figure 4 Calculate a tracking graph to rank candidate resource bids;

[0038] Figure 5 A flowchart illustrating the process of redundancy propagation, local renegotiation, and committed version update under perturbation. Detailed Implementation

[0039] In one embodiment, a surgical kit delivery scheduling method based on a multi-agent system includes: acquiring surgical schedules, surgical procedures, kit inventory, sorting load, delivery robot task status, and channel occupancy information; establishing task associations between surgical tasks, kits, inventory, sorting resources, delivery resources, and operating rooms; generating a time-limit commitment chain covering kit assembly, sorting, pickup, delivery, and receiving stages based on the latest delivery time of the surgical task; having participating resources generate distributed bids based on their local execution status, and determining resource allocation results by combining the task time redundancy occupancy relationships corresponding to the bids; updating the remaining redundancy of each stage when the execution status changes, and propagating the updated redundancy constraints along the time-limit commitment chain; and rolling adjustment of subsequent resource allocation based on the updated redundancy constraints.

[0040] Specifically, please refer to Figure 1 The multi-agent system establishes state objects at the logical level corresponding to surgical tasks, kits, inventory, sorting resources, delivery resources, and the operating room. Each state object has an independent data recording area and event processing procedure, but they are linked through a unified task identifier, kit identifier, resource identifier, and time base. The surgical task state object stores the planned surgical time, the kit requirement corresponding to the surgical procedure, the latest delivery time, and the risk level. The kit state object stores the components, the kit status, and the pending sorting status. The inventory state object stores the inventory batches available for the target kit and their available status. The sorting resource state object stores the order of accepted tasks and the expected completion time. The delivery resource state object stores the task path corresponding to the current location, unfinished tasks, and the time interval that can be inserted. The operating room state object stores the time window during which the target kit can be received. The system sorts data from different data sources according to the task occurrence time and binds it to a unified task identifier, so that each subsequent state change can locate the associated task without re-parse all historical data.

[0041] In this embodiment, after a surgical task is established, it does not directly form a single priority queue according to the planned surgical time. Instead, it establishes a continuous task chain along the processing direction of kit assembly, sorting, retrieval, delivery, and receiving in the operating room. Each link records the expected start time, expected end time, allowed latest end time, predecessor link identifier, successor link identifier, and currently occupied resource identifier. When a surgical task requires multiple kits, the assembly and sorting nodes of each kit are established separately and then converged into the same delivery constraint. When multiple kits are allowed to be delivered in batches, the corresponding delivery nodes and receiving boundaries are maintained respectively. The operating room state object maintains multiple receiving boundaries according to the latest receiving requirements of each kit, so that the time constraint can propagate along the actual task dependency relationship instead of simply reserving a fixed lead time forward based on the surgical start time. At the same time, the occupation intervals of the same resource between different tasks are established to establish an overlapping relationship, so that the subsequent quotation process can determine the impact range of a new task on existing tasks.

[0042] In one embodiment, establishing task associations includes: taking each surgical task as a task node, and taking package assembly, sorting, robot retrieval, delivery and transfer, and operating room reception as sequential execution nodes, and taking available inventory batches, available sorting time periods, available delivery robot execution time periods, and available passageway time periods as resource nodes; connecting task nodes and sequential execution nodes according to their execution order, and connecting sequential execution nodes and resource nodes according to their resource availability; establishing competitive associations for tasks that share the same resource and have overlapping execution times, forming a task resource association graph that is updated synchronously with scheduling and resource status; in actual storage, an adjacency list can be used to record the execution nodes and resource nodes directly connected to each task node, and record the start time, end time, occupancy type, and current version for each association, so that the system can directly obtain the set of affected tasks based on the adjacency relationship when it receives a resource change event.

[0043] Furthermore, the resource occupancy relationship in the task resource association graph does not only indicate whether a task uses a certain resource, but also stores the time position of the task's resource occupancy. For example, inventory resources correspond to the locked interval of available batches, sorting resources correspond to the expected processing interval of a task in the current sorting queue, delivery resources correspond to the continuous time interval of picking up, moving, handing over, and returning to accept tasks again, and channel resources correspond to the occupancy interval of the relevant channel that the delivery task is expected to pass through. When two tasks are connected to the same resource node and the corresponding intervals do not overlap in time, no real-time competition flag is established. When the occupancy interval of any task moves and overlaps due to scheduling adjustment, a competition flag is added and the time slice of the competition is recorded. In this way, tasks that share resources but do not actually compete at the same time and tasks that need to compete within the same time window can be distinguished. Subsequent local reordering only needs to process the set of tasks with real-time competition flags.

[0044] In a preferred embodiment, the system performs incremental maintenance on the task association graph. When a surgical schedule is added, canceled, or changed, the changed task and its directly connected execution nodes are added to the set to be updated. When an inventory batch becomes unavailable, the matching nodes referencing the batch are added to the set to be updated. When the sorting order changes, the task to be moved and its adjacent tasks are added to the set to be updated. When the execution status of the delivery robot or the channel occupancy status changes, the delivery nodes occupying the corresponding time interval are added to the set to be updated. Then, update requests are propagated to adjacent nodes according to the predecessor and successor relationships of the execution nodes. The expansion stops when the resource occupancy interval of the adjacent nodes is consistent with that before the update. This ensures that the recalculation range of the association graph is determined by the actual change propagation boundary. The task relationships that have been determined and are not affected continue to use the original version data, thus preserving a stable reference state for subsequent local quotations and commitment updates.

[0045] In one embodiment, generating a time-limit commitment chain includes: using the latest delivery time of the surgical task as the final commitment boundary, reading the current queue occupancy, committed tasks, and executable time windows of each stage in reverse order of receiving, delivering, picking up, sorting, and kitting in the operating room, and determining the latest delivery boundary between adjacent stages step by step; setting a front-to-back dependency relationship for adjacent delivery boundaries of the same task, limiting the time interval allowed for the upstream stage to be before the downstream delivery boundary, and writing the delivery boundaries of each stage and the adjustable time between the boundaries into the commitment status record of the corresponding task. Each latest delivery boundary is obtained by subtracting the expected processing time of the subsequent stage, the known queuing waiting time, and the transfer time occupied to maintain the continuity between the preceding and following stages from the boundary of the subsequent stage, so that the time boundary passed forward from the final delivery constraint is consistent with the current actual resource occupancy status.

[0046] The latest delivery boundary between adjacent stages can be calculated according to the following relationship:

[0047] ;

[0048] in, This represents the latest delivery boundary of the i-th surgical task at the k-th execution stage. This indicates the latest delivery boundary of the preceding execution stage. This represents the estimated actual processing time for the k-th execution stage, determined based on the current task status. This represents the waiting time caused by committed tasks before entering the k-th execution stage. This indicates the connection time required to transition from the previous execution stage to the kth execution stage. The minus sign indicates that the determined necessary time is deducted from the latest allowed time downstream. For example, if a task requires the operating room to be received before 10:00, the current delivery processing takes 12 minutes, the estimated waiting time before delivery is 3 minutes, and the connection time between pickup and delivery is 2 minutes, then the latest delivery boundary corresponding to the previous stage is 9:43. If the waiting time changes from 3 minutes to 7 minutes due to changes in the robot queue, then the corresponding boundary moves forward to 9:39. The aforementioned movement continues to be passed to the sorting and kitting stages, so that the new waiting time will not be stuck at a single delivery node.

[0049] Please refer to Figure 2 , Figure 2 The horizontal axis represents the sequence number of the time commitment chain boundary calculation tracking record, which records the reverse boundary calculation process originating from sequential execution nodes such as kitting, sorting, pickup, delivery, and operating room reception; the vertical axis represents the amount of minutes deducted when the boundary is pushed upstream, in minutes. In the legend, the solid line represents the total deduction formed when the boundary is passed from the downstream boundary to the upstream boundary, and the long dashed line represents the expected processing time. The dotted line indicates the waiting time caused by the committed task. The dotted lines represent the connection time occupied when adjacent links transition into each other. .

[0050] from Figure 2 It can be seen that the total deduction curve exhibits periodic peak-trough variations, with the peak values ​​mainly occurring near records 2, 7, 12, 17, 22, 27, 32, and 37. Record 2 shows a total deduction of 37.00 minutes, indicating that this type of process... Towards When iterating forward, a large time boundary needs to be given up all at once. In contrast, the total deduction for some records is reduced to about 10 minutes, indicating that the compression of the upstream boundary in the corresponding link is smaller, and the task chain still retains sufficient local adjustment space.

[0051] Further comparison of the individual curves reveals the expected processing time. The significant contribution at most peak values ​​indicates that sorting, delivery, or receiving processes within the surgical kit distribution chain are the primary sources of time for boundary pushback; waiting time Overall, the processing time is lower than the processing time, but significant fluctuations are still observed at several points, reflecting the continuous pressure exerted by the committed task queue on the upstream delivery boundary of the current task; connection time The curve is near the bottom but not zero, indicating that pickup, transfer, handover or receiving window connections still need to be written into the commitment status as independent time occupancy.

[0052] Preferably, each commitment state record, in addition to storing the latest delivery boundary, also stores the current estimated completion time and the remaining adjustable time amount formed by the difference between the two. When the estimated completion time is earlier than the latest delivery boundary, the difference is recorded as a time slack that can be used for subsequent local reordering. When the estimated completion time is later than the latest delivery boundary, the difference is recorded as a negative value and the corresponding node is marked to enter the pending adjustment state. To avoid state overwriting caused by upstream and downstream modifying the same task at the same time, each commitment state record is accompanied by a generation version and a parent version identifier. Only the state calculation results of the same parent version can be merged in the same round of updates. If a resource generates a new state during the calculation, the corresponding link boundary is recalculated based on the new state instead of directly overwriting the generated new version.

[0053] In a preferred embodiment, for a task that includes multiple packages that need to be completed together before being received in the operating room, each package establishes a completion node and a sorting node, and establishes a convergence constraint before delivery. The system uses the latest expected completion time of all predecessor nodes that enter the convergence constraint as the current delivery start time, and retains the latest delivery boundary for each predecessor node. When a package is completed ahead of schedule, the entire advance is not immediately transferred to other packages, but is first recorded as a partial margin for the corresponding predecessor node. Only when other predecessor nodes do not exceed their own delivery boundaries are they allowed to use the partial margin in subsequent resource quotations. This avoids the situation where an early-completed package masks the continuous delay of another package, and keeps the convergence time of multiple package tasks traceable to their respective processing chains.

[0054] In one embodiment, forming a distributed quote includes: a candidate sorting resource or a candidate delivery resource reading the commitment status record of the task to be assigned and its own set of committed tasks; partially rearranging the execution sequence after inserting the task to be assigned; determining the additional time occupied by the task to be assigned, the adjustable time of the compressed existing task, the connection time occupied when a task transfer occurs, and the change in the delivery boundary of each task; encoding the additional time occupied, the adjustable time, the connection time occupied, and the change in the delivery boundary into the same quote vector; and binding the quote vector with the corresponding executable time period and submitting it to the task node corresponding to the surgical task, so that the content reported by each candidate resource includes both its own execution cost and the time occupied change caused to the existing commitment chain after receiving the task.

[0055] Specifically, please refer to Figure 3 When a sorting resource generates a quote, it retrieves all movable and immovable tasks from its committed task queue. Immovable tasks include those whose start time has entered the locked interval or whose movement would directly cause a mismatch in subsequent delivery boundaries. Tasks to be assigned are simulated and added within the insertable gaps formed between immovable tasks. For each feasible insertion position, the estimated start and end times of adjacent tasks are recalculated. When a delivery resource generates a quote, in addition to performing the same time insertion process, it also writes the current remaining delivery path, the movement time to the pickup location after completing the existing task, and the connection time to return to the next existing task path after executing the task to be assigned into the quote status. If the task to be assigned needs to be transferred to another delivery resource through a handover point, the current resource only generates a quote for its responsible interval and records the handover point arrival window for subsequent matching.

[0056] In this embodiment, the adjustable time of existing tasks in the bidding vector is calculated separately according to the affected range of the tasks. For a candidate insertion position, the estimated completion time is recalculated starting from the first existing task after the insertion point. If the recalculated estimated completion time is the same as the original estimated completion time, the propagation stops. If there are still changes, the next task is calculated until an immovable task is encountered or the change is completely absorbed by the remaining capacity of the current task. This obtains the time compression range actually caused by the task to be assigned. The system will not include all committed tasks of the candidate resource in the bidding, so as to avoid historical tasks that are not related to the task to be assigned from interfering with the current resource competition. At the same time, the task identifier, compression amount and corresponding delivery boundary are recorded for the compressed task, so that the task node can compare the impact of different resources on the same existing task after receiving multiple candidate bids.

[0057] In one embodiment, candidate bids can be sorted according to the time occupancy component value after unification of dimensions, and the calculation relationship is as follows:

[0058] ;

[0059] in, This represents the bid ranking value when selecting the r-th candidate resource for the i-th surgical task. This indicates the additional time occupied by the task to be assigned due to the selection of candidate resources. This represents the available time for a task from the moment of resource allocation to the corresponding latest delivery boundary. This indicates the total adjustable time available for compressing existing tasks on the candidate resources. This represents the total remaining adjustable time for all currently movable tasks of the candidate resource. This indicates the connection time required for the task to be assigned due to resource transfer. This indicates the amount of time the delivery boundary of the task to be assigned is compressed after selecting candidate resources. This indicates the amount of unprotected time currently allowed for resource competition in the task to be assigned. The plus sign indicates that the standardized time occupancy of various types is accumulated in the same sorting value. The division sign is used to normalize the time occupancy corresponding to different tasks and resources to their respective available ranges. For example, if two candidate delivery resources generate 6 minutes and 8 minutes of additional occupancy respectively, but the former will compress the existing task by an additional 5 minutes while the latter will not compress the existing task, then the former will not automatically obtain priority simply because the additional occupancy is smaller. The system determines its sorting position based on the complete standardized occupancy calculation result. If the available time corresponding to a certain denominator is 0, the corresponding candidate bid is directly marked as infeasible and will not enter the sorting set.

[0060] Please refer to Figure 4 , Figure 4The horizontal axis represents the candidate resource bid calculation tracking record number, with each record corresponding to one bid calculation for the surgical task on the candidate sorting or delivery resources; the vertical axis represents the contribution of the bid ranking input, without an independent unit of measurement. The solid line in the legend represents the final ranking value. The long dashed line represents the additional time occupied. Available time for the task The contribution after normalization; the dotted line represents the amount of existing task compression. Remaining movable resources The contribution after unification, with dots and lines representing the transition and connection time. The corresponding contribution, indicated by the dashed line, represents the delivery boundary compression. Non-protected competition time The contributions after unification.

[0061] from Figure 4 As can be seen, the final ranking value curve exhibits a pattern of rising and falling. The final ranking value at the 7th marked point is 2.554, the highest value among the currently tracked samples, indicating that the candidate resource at this insertion position has the heaviest impact on task time margin and existing commitment chains. Observing the ranking logic of resource selection, a higher final ranking value usually indicates that the candidate solution exerts greater time pressure on the current or existing task; the section where the curve falls to near zero indicates that the contribution of that group of bids is low, or that the corresponding record does not form effective ranking pressure in the current data.

[0062] The sub-curves show that both the newly added occupancy contribution and the existing compression contribution rise simultaneously near most peaks, indicating that candidate resources do not only incur costs due to the time required to execute their assigned tasks, but also compress the remaining adjustable time of committed tasks through local rearrangement. The transfer occupancy contribution is generally low but persistent, reflecting that the movement and connection between the current location of delivery resources, pickup points, handover points, or operating rooms will be included in the bidding ranking; the boundary compression contribution is relatively small, but occurs during tight protection periods or non-protection competition periods. At lower levels, its effect on The impact will be amplified.

[0063] In one embodiment, determining the resource allocation result based on the bid vector includes: verifying the delivery boundary of the task to be allocated, the delivery boundary of the existing task, and the protection time period for the execution sequence corresponding to the candidate bids, and eliminating candidate bids that cause mismatch of any delivery boundary; forming a bid ranking sequence for the remaining bids according to the new time occupied by the task to be allocated, the adjustable time consumption of the existing task, the time occupied by the task transfer and connection, and the delivery boundary compression amount, and establishing a mutual exclusion selection relationship when multiple bids share resources, and determining the task allocation set corresponding to the executable time periods of each resource from the bid combination without resource time period conflicts, wherein the mutual exclusion selection relationship uses resource identifier and time interval as the judgment basis, the same resource can correspond to different task bids in non-overlapping time periods, and bids with overlapping time intervals cannot enter the final task allocation set at the same time.

[0064] Furthermore, when multiple tasks submit competitive requests to the same candidate resource in the same round of scheduling, the resource state object generates an independent copy of the offer for each request and records the local execution sequence assuming that other requests have not yet been accepted. After the task nodes summarize the offers, they establish a conflict set according to the shared resources and overlapping intervals. The feasibility of the offer combinations in the conflict set is verified. After selecting one offer, the new execution sequence of the corresponding resource is rebroadcast as a temporary commitment state to the tasks in the conflict set that have not yet made a selection. The remaining tasks re-check whether their offers are still valid based on the temporary commitment state. If the insertion gap of the original offer has been occupied, only the offer corresponding to that resource is recalculated without recalculating the offers of other resources. This allows each task to gradually converge to a non-overlapping allocation result based on the latest commitment state of the same resource during the parallel bidding process.

[0065] In one embodiment, partial reordering includes: maintaining the order of tasks with immovable delivery boundaries in the committed tasks of candidate resources, inserting the task to be assigned into the candidate gap formed by the remaining movable tasks, and simultaneously determining the pickup connection time, delivery transfer time, and resource idle time of the tasks before and after the insertion position; when the task to be assigned is completed across two execution entities, the arrival time window corresponding to the handover point is written into the commitment status record of the previous execution entity and the next execution entity respectively, retaining only the partial reordering sequence with a common handover interval between the arrival time windows before and after and without intruding into the protection time period, and encapsulating the partial reordering sequence into the execution sequence corresponding to the candidate price, so that the partial adjustment of resources still maintains the time consistency with the upstream and downstream task chains.

[0066] Specifically, in the delivery resource execution sequence, each task records the estimated time of entering the pickup location, the departure time after pickup, the estimated time of arriving at the handover location or operating room, and the next available time after execution. When a new task is inserted between two existing tasks, the system extrapolates from the end position of the previous task to the pickup location of the new task, and from the delivery location of the new task to the start position of the next task, and adds these two connecting intervals to the resource occupation time axis. If the movement of the next task due to the insertion of the new task exceeds its adjustable time, the current insertion position is deleted. If the movement can be absorbed by the remaining capacity of the next task, the system continues to check whether it will affect the tasks further down the line. Sorting resources use the same time axis insertion logic, but their connecting intervals are composed of the processing switching and queuing of the preceding and following tasks, expressing the local execution sequence of different resources under a unified time constraint framework.

[0067] In a preferred embodiment, when a delivery task needs to be executed consecutively by two delivery resources, the preceding delivery resource generates a time interval that can reach the handover point based on its own local execution sequence, and the following delivery resource generates a time interval that can take over based on its existing tasks and the time required to reach the handover point. The system calculates the intersection of the two time intervals as the available handover interval. If the intersection is empty, the current resource combination does not form a candidate offer. If the intersection exists, the handover time that will not break the subsequent commitment boundary of both parties is selected and written into the commitment record of the two resources. The waiting time from the arrival of the preceding resource to the takeover by the following resource is included in the connection time of the corresponding task. When the state of any resource changes and the original handover time no longer falls within the intersection interval, only the two resources and corresponding tasks associated with the handover record are recalculated, without modifying other delivery tasks.

[0068] In one embodiment, the propagation redundancy constraint includes: comparing the actual completion time, the current estimated completion time, and the delivery boundary of each sequential execution node to obtain the remaining adjustable time of the corresponding sequential execution node; when an upstream sequential execution node releases time in advance, the release amount is limited to the commitment interval not yet occupied by the adjacent downstream sequential execution nodes and transferred step by step; when an upstream sequential execution node occupies time late, the existing adjustable time of the adjacent sequential execution nodes is deducted sequentially; based on the surgical risk level, a protection period is divided in the adjustable time of the corresponding task that does not participate in the bidding of other tasks, and the effective time interval for each sequential execution node to participate in the next round of resource competition is synchronously corrected after each propagation, wherein the protection period is saved with the task chain version as part of the commitment state, and only the non-protected part is allowed to be read as the time reserve that can be occupied by other tasks during the resource bidding stage.

[0069] In this embodiment, when the kitting process is completed earlier than the current estimated completion time, the system calculates the lead time and updates the remaining adjustable time of the kitting node. Then, it reads the estimated start time of the subsequent sorting node. If the sorting node cannot start earlier due to existing task constraints, the release time is retained at the kitting node and does not continue to propagate. If there is an unused time interval at the sorting node, the available portion is transferred to the sorting node and the allowable time range for pickup and delivery is recalculated. When a certain process is delayed, the reverse processing is used, deducting the delay amount from the current node's remaining capacity. The insufficient portion continues to occupy the unprotected capacity of the subsequent nodes until the delay amount is completely absorbed or the remaining adjustable time of a certain node becomes negative. This forms a time change propagation process consistent with the execution chain, preventing the fixed lead time from remaining unchanged after the task is executed.

[0070] Preferably, the protection time period corresponding to the risk level is not written to the task by a simple fixed value, but is established by creating a hierarchical record based on the total adjustable time between the latest delivery boundary of the task and the current expected completion time. The system marks a portion of this time period as protected and removes it from the remaining time available for bidding on other tasks. Tasks with higher risk levels retain the time period closer to the delivery boundary and do not participate in the transfer. Tasks with lower risk levels can release more non-protected time while meeting their own delivery boundaries. When different risk tasks compete for the same resource, the bidding calculation can only consume the non-protected interval. If a candidate insertion scheme requires the use of the protected interval to be valid, the scheme is directly excluded. The protected interval itself can still be used by the corresponding task when it experiences a delay, so that the available range for resource competition transfer and task time recovery is different.

[0071] In one embodiment, updating the task resource association graph includes: when the inventory batch status, sorting queuing order, robot task sequence, or channel occupancy range changes, only the sequential execution nodes directly connected to the changed resource nodes are recalculated, and the process is extended upstream and downstream along the task execution sequence to the first sequential execution node whose resource occupancy relationship has not changed; the original task competition associations outside the extended range are retained, resource competition relationships that are added, removed, or have changed time overlap relationships within the extended range are remarked, and the association graph update version number is written into the commitment status record of the corresponding task. When a subsequent bid is submitted, the version number carried by the bid is verified to be consistent with the association graph update version number in the commitment status record, so as to avoid candidate resources from continuing to participate in the allocation based on invalid association relationships.

[0072] Specifically, when the end time of an existing task for a delivery resource is delayed, the system first updates the time axis corresponding to the delivery resource, then searches for the delivery execution nodes connected within the time axis change interval, and re-determines the time overlap relationship between these nodes and other tasks on the same resource. For tasks that originally had no overlap but now overlap after the update, competitive associations are added; for tasks that originally had overlap but have now separated after the update, competitive associations are removed. Then, it checks whether the predecessor pickup node and successor receiving node of the updated delivery node have changed their resource occupancy relationship due to time movement. If the associated resources and occupancy intervals have not changed, the expansion stops; if there are still changes, the expansion continues along the task chain. After the expansion is completed, a new partial version of the association graph is formed, and the commitment status of the affected tasks is pointed to this version. Tasks that have not entered the expansion scope continue to reference the original partial relationship without regenerating all their association data.

[0073] In a preferred embodiment, the quotation request carries the version number of the association graph currently used by the task and the committed version number. After receiving the request, the candidate resource reads its own occupied time axis according to the corresponding version and generates a quotation. When the task node receives the quotation, it checks again whether the versions are consistent. If the association graph is updated during the period but the update scope does not involve the candidate resource and the task, the quotation is allowed to be mapped to the new version and continue to be used. If the update scope has already included the time interval of the resource bound to the current task or quotation, the old quotation is discarded and a new quotation request is only initiated for the candidate resource. This avoids the failure of all parallel quotations as a whole due to a local state change, and at the same time prevents the old version quotation from overwriting the newly formed resource occupancy relationship.

[0074] In one embodiment, please refer to Figure 5 When the remaining adjustable time of any sequential execution node is less than zero, the task set that shares resources with the sequential execution node and whose delivery boundary is associated with its time occupation is extracted according to the task resource association graph, and the resource allocation outside the task set is frozen; within the task set, the releaseable time interval is identified according to the propagation direction of the negative remaining adjustable time along the time commitment chain, and the quotation request is re-initiated only to the candidate sorting resources or candidate delivery resources with releaseable time intervals. After the re-obtained quotation is verified with the original commitment status record, the resource allocation of the corresponding sequential execution node and its subsequent sequential execution nodes is updated, and the scope of renegotiation is limited to resources and tasks that can directly absorb the current time gap.

[0075] Specifically, when a negative remaining adjustable time occurs, the system reads the current resource node and adjacent competing tasks from the execution node where the negative value occurs, forming an initial impact set. Then, it determines whether the adjacent tasks have shifted their expected completion time due to the current node's delay. Only tasks that have shifted and whose shift continues to encroach on their unprotected time intervals are added to the impact set. If an adjacent task has sufficient unprotected margin to absorb the shift and its subsequent resource occupancy interval remains unchanged, then that task is used as the propagation boundary to stop further expansion. After the impact set is formed, the current commitment status of resources outside the set is marked as frozen. Tasks within the set read the time interval of alternative resources and generate new bids. After the bid selection is completed, the delivery boundary of the negative value node is restored first, and then the remaining time change is re-propagated along its successor nodes. Only when the new allocation result causes the impact boundary to expand are new related tasks added, avoiding the rescheduling of all surgical package tasks when a single task is delayed.

[0076] Furthermore, when the negative remaining adjustable time comes from inventory kitting delay, the system does not directly re-bid on all delivery resources. Instead, it first determines the time gap after the kitting delay is expected to propagate to the sorting node along the current task chain. If the existing unprotected margin of the sorting node can absorb the gap, the delivery commitment remains unchanged. If the sorting node cannot absorb the gap, the boundary changes of the pickup node and the delivery node are calculated. Only when the executable range of the delivery node moves and forms a new competitive relationship with the existing task is a re-bid initiated on the corresponding delivery resource. By confirming layer by layer along the task chain whether the time gap has really changed the downstream resource occupancy relationship, the same local renegotiation logic based on the propagation of time effects is adopted for material shortage, sorting congestion and delivery delay.

[0077] In one embodiment, a sequence of commitment versions is maintained for the resource allocation after rolling adjustments. Each commitment version records the task resource association graph version, time commitment chain version, quotation vector source, resource executable time period, and the identifier of the previous commitment version being replaced. When receiving new inventory, sorting, delivery robot, or channel status, the corresponding status timestamp is compared with the generation time of the current commitment version, and alternative quotations are formed only based on status data whose timestamps are not earlier than the generation time of the current commitment version. When multiple local adjustments involve the same resource time period, the adjustment range is merged according to the dependency relationship between commitment versions, and candidate allocations with cyclic resource occupancy dependencies are rolled back to the most recent commitment version without resource conflicts before the resource allocation is re-determined, so that rolling scheduling can maintain a consistent order of resource occupancy data under continuous state changes.

[0078] In this embodiment, commitment versions are stored using a parent-child relationship. Whenever a task accepts a new resource offer, a child version is generated. The child version does not copy the entire historical state, but records the task nodes, resource nodes, time intervals, and offer sources that have changed relative to the parent version. When the complete commitment state needs to be read later, the uncovered data is merged according to the reference relationship from the child version to the parent version. If two parallel local adjustments modify different resources and their time intervals do not overlap, they can be merged into the same successor version after verifying that the parent version is consistent. If two local adjustments modify the same resource in overlapping time intervals, conflicts are identified according to the generation time and dependency relationship, and direct merging is prevented. The task within the conflict range is waited for to re-offer. This avoids the situation where different event processing processes generate valid local results but the same resource is repeatedly occupied after merging.

[0079] In a preferred embodiment, circular resource occupancy dependency manifests as multiple tasks waiting for each other to release resources through partial reordering. For example, a new delivery arrangement for one task depends on a second task releasing a certain delivery time slot, while the new arrangement for the second task depends on the first task releasing another time slot. When generating candidate commitment versions, the system establishes a dependency edge from the reassigned task to its dependency on the release action of the old commitment, and checks whether the dependency edge forms a closed path. When a closed path appears, it is determined that the current multiple candidate adjustments cannot be submitted simultaneously in the current order. Then, it rolls back along the parent version pointer of each candidate version to the most recent commitment version that does not have resource time conflicts and has not introduced closed dependencies. The completed or locked resource occupancy in the rollback version is retained, and then a bid is reissued for the resources involved in the closed path based on the available time slots after the rollback. This allows the circular dependency to be resolved through version state rollback and partial reallocation without affecting tasks outside the closed path.

[0080] In one embodiment, the multi-agent system performs the above processing in an event-driven manner during continuous operation. When a new surgical task enters, a task resource association graph is established and an initial time limit commitment chain is generated. Subsequently, the participating sorting and delivery resources form a distributed quotation and determine the initial allocation. When the task enters actual execution, the completion of kitting, sorting, pickup, change of delivery location, handover, and reception in the operating room all form time-stamped state events. Each event only modifies the current state of the corresponding execution node and triggers redundant recalculation. When the new state is still within the original commitment range, only the remaining adjustable time is updated. When the state crosses the original commitment range and changes the resource occupancy relationship, the incremental update of the association graph and the local requote process are entered. This ensures that small time changes during normal execution and time changes that require modification of resource allocation have different processing paths in the same commitment data structure.

[0081] Preferably, for urgent surgical tasks that are temporarily inserted, the system uses the same task nodes and time commitment chain structure as the planned surgery, and establishes the delivery boundaries of each link in reverse according to its latest delivery time. When the inserted task enters the resource bidding, it reads the unprotected time interval and resource executable gap of the existing tasks, and does not directly cancel the commitment of the existing tasks. If the available gap can accommodate the inserted task, the inserted task is added to the local rearrangement sequence and the remaining adjustable time of the affected tasks is updated. If there is no complete executable gap, the system searches for existing tasks that can transfer unprotected time according to the task resource association graph and forms a new candidate bid combination. Combinations involving protected time periods or causing mismatch of the delivery boundaries of existing tasks are excluded, so that the insertion process of temporary tasks is still jointly restricted by the time commitment chain, resource competition relationship and protection time constraints.

[0082] In this embodiment, if congestion occurs in sorting resources and causes the expected completion times of multiple sorting tasks to shift simultaneously, the system performs an incremental update of the association graph for the changed sorting timeline. It establishes competitive associations for tasks with newly overlapping times and recalculates the allowable time ranges for pickup and delivery along the time commitment chains of each task. For tasks that can absorb sorting delays with their own non-protection margins, the original delivery allocation is maintained. Only tasks whose allowable delivery time ranges have shifted are included in the delivery resource re-quote set. If multiple tasks affect the same delivery resource simultaneously, a mutually exclusive quotation relationship is formed according to the shared resource time interval, and temporary commitment versions are submitted one by one. The queue changes on the sorting side are converted into delivery side adjustments corresponding to their actual propagation results, rather than a unified rearrangement of all delivery tasks.

[0083] In a preferred embodiment, when a delivery resource is unable to continue undertaking the current task during execution, the system marks the resource as temporarily unavailable from the executable time period after the current moment. The task resource association graph is then decoupled from the original resource associations of the dependent time period, and the remaining delivery boundary is calculated starting from the execution node where the current task has not yet been completed. If the task has been completed and the pickup has been completed, a new candidate delivery resource is generated with a quote based on the takeover status corresponding to the current location. If the task has not been completed and the pickup has not been completed, a new candidate quote is generated from the original pickup node. The system retains the completed upstream kitting and sorting commitments unchanged, performs partial renegotiation only on the delivery and subsequent receiving nodes, and uses the original delivery commitment version as the parent version to generate a replacement version, so that the completed task status is not rewritten due to changes in delivery resources.

[0084] Furthermore, when changes in channel occupancy information render the original delivery time interval unusable, the system maps the occupancy interval corresponding to the changed channel to all delivery execution nodes passing through that channel. For the executed portions that have not yet entered the changed interval, the original record is maintained. For the portions that have not yet been executed, the passable time interval is redefined. The system also checks whether the adjusted estimated delivery completion time still falls before the current delivery boundary. If the changes can be absorbed by the delivery node's own non-protection time, only the estimated completion time of the current delivery execution node is updated. If it is necessary to change the resource execution order or cause the occupancy intervals of other tasks to overlap again, a new competitive association is generated and a local bid is triggered, so that the channel state change enters the existing multi-agent scheduling process through time interval constraints without the need to establish an independent scheduling process.

[0085] In one embodiment, the operating room receiving status is synchronized with the end commitment of the delivery task. The operating room status object maintains the receiving window of each task based on the current planned surgery time and the received package status. After the delivery completion event arrives, the actual arrival time is written to the corresponding receiving node and the subsequent occupied interval of delivery resources is released. If the operating room receiving window changes, the time commitment chain of the current task is recalculated with the new latest receiving boundary. When the boundary moves backward, the added time is not directly and evenly distributed to all upstream links, but is formed according to the execution chain starting from the current downstream node that has not yet been completed, and the decision on whether to continue to open up to the upstream is made in combination with the resource occupancy status of each node. When the boundary moves forward, the non-protection margin is deducted from the current node and the task association relationship is used to determine whether a new quotation is needed, thereby keeping the end time change consistent with the actual uncompleted task range.

[0086] Specifically, the system saves the final commitment version, the actual completion time of each execution node, the boundary changes and resource rearrangement records that have occurred at each node for each task completion. These records do not directly change the existing scheduling rules, but serve as the data source for the initial expected processing time and waiting time of subsequent similar tasks. When generating a new time-limit commitment chain, the system can select completion status data from historical records under the same formula and similar resource occupancy conditions, and filter the record intervals corresponding to the current conditions according to the current sorting queue length, the number of delivery tasks, and the channel occupancy relationship. The actual processing time in the records is used as the reference value for the current expected processing time. When there are no records with the same conditions, the queuing time and execution time that can be directly derived from the current resource status are used, and prediction results that cannot be traced are not used, so that the expected processing time and waiting time in the formula have a corresponding data basis.

[0087] In a preferred embodiment, to prevent consecutive state events from causing frequent version generation for the same task, the system adopts a commitment transaction interval during event processing. Multiple state events with similar arrival times on the same task chain that have not yet triggered changes in resource contention are merged into the same transaction to update the current estimated completion time and the remaining adjustable time. When any event causes a delivery boundary mismatch, a change in the overlapping relationship of resource occupation intervals, or a need to re-formulate a bid, the current transaction is immediately terminated and a new commitment version is generated. Subsequent events continue to be processed based on the new version. By mapping state updates and resource allocation modifications to different version generation conditions, the execution data can be continuously updated while resource commitments only change when the scheduling relationship actually changes.

[0088] In this embodiment, a complete closed loop is formed from task creation to task completion. The surgical task status object provides the end-time constraint, the package and inventory status objects determine the conditions for complete set execution, the sorting resources and delivery resources generate candidate bids through the local time axis, the task resource association graph limits the scope of resource competition, and the time limit commitment chain transmits the final delivery boundary to each execution stage step by step. The bid selection process filters executable combinations based on new task occupancy, compression of existing task surplus, transfer occupancy, and changes in delivery boundary. The early or late states generated during the execution process update the remaining adjustable time along the commitment chain. When the time change is insufficient to change the resource relationship, the original commitment is maintained. When the time change causes the resource relationship to change, bidding is only conducted within the corresponding impact set. Each substitution relationship and resource occupancy change is recorded through the commitment version, so that package complete set, sorting, picking, delivery, and receiving under the condition of multi-task parallelism are always continuously updated according to the same time constraint system.

Claims

1. A method for scheduling the delivery of surgical kits based on a multi-agent system, characterized in that, include: Obtain information on surgical schedules, surgical procedures, kit inventory, sorting load, delivery robot task status, and channel occupancy; establish task relationships between surgical tasks, kits, inventory, sorting resources, delivery resources, and operating rooms. Based on the latest delivery time of the surgical task, a time limit commitment chain covering the complete set, sorting, picking up, delivery and receiving stages is generated in reverse. The participating resources form a distributed bid based on their local execution status, and the resource allocation result is determined by combining the task time redundancy occupancy relationship corresponding to the bid. When the execution status changes, the remaining redundancy of each stage is updated, and the updated redundancy constraints are propagated along the time commitment chain. The subsequent resource allocation is then adjusted on a rolling basis according to the updated redundancy constraints.

2. The surgical kit delivery scheduling method based on a multi-agent system according to claim 1, characterized in that, Establishing the task association includes: taking each surgical task as a task node, and taking package assembly, sorting, robot retrieval, delivery and transfer and operating room reception as sequential execution nodes, and taking available inventory batches, available sorting time periods, available delivery robot execution time periods and available passage time periods as resource nodes. The task nodes are connected to the sequential execution nodes according to their execution order, and the sequential execution nodes are connected to the resource nodes according to their resource availability. Competitive relationships are established for tasks that share the same resource and have overlapping execution times, forming a task-resource association graph that is updated synchronously with the scheduling and resource status.

3. The surgical kit delivery scheduling method based on a multi-agent system according to claim 1, characterized in that, Generating the time commitment chain includes: taking the latest delivery time of the surgical task as the end commitment boundary, reading the current queue occupancy, committed tasks and executable time windows of each link in the reverse order of receiving, delivering, picking up, sorting and assembling in the operating room, and determining the latest delivery boundary between adjacent links level by level. Set up pre- and post-delivery dependencies for adjacent delivery boundaries of the same task, limit the time interval allowed for upstream links to be before the downstream delivery boundary, and write the delivery boundaries of each link and the adjustable time between the boundaries into the commitment status record of the corresponding task.

4. The surgical kit delivery scheduling method based on a multi-agent system according to claim 1, characterized in that, The formation of the distributed quotation includes: reading the commitment status record of the task to be assigned and the set of its own committed tasks from the candidate sorting resources or candidate delivery resources; performing local rearrangement on the execution sequence after inserting the task to be assigned; and determining the additional time occupied by the task to be assigned, the adjustable time of the compressed existing task, the connection time occupied when the task transfer occurs, and the change in the delivery boundary of each task. The newly added occupied time, adjustable time, connection occupied time and delivery boundary change are encoded into the same quotation vector, and the quotation vector is bound to the corresponding executable time period and submitted to the task node corresponding to the surgical task.

5. The surgical kit delivery scheduling method based on a multi-agent system according to claim 2, characterized in that, Updating the task resource association graph includes: when the inventory batch status, sorting queuing order, robot task sequence, or channel occupancy range changes, only the sequential execution nodes directly connected to the changed resource nodes are recalculated, and the process is extended upstream and downstream along the task execution sequence to the first sequential execution node whose resource occupancy relationship has not changed; The original task competition relationships outside the extended scope are retained. Resource competition relationships that are added, removed, or have changed time overlap relationships within the extended scope are re-marked, and the updated version number of the association graph is written into the commitment status record of the corresponding task. When submitting a bid, the version number carried by the bid is verified to be consistent with the updated version number of the association graph in the commitment status record.

6. The surgical kit delivery scheduling method based on a multi-agent system according to claim 1, characterized in that, Propagating the redundancy constraints includes: comparing the actual completion time, the current estimated completion time, and the delivery boundary of each sequential execution node to obtain the remaining adjustable time for the corresponding sequential execution node; When an upstream sequential execution node releases its time ahead of schedule, the release amount is limited to the committed interval that has not yet been occupied by the adjacent downstream sequential execution nodes and transferred level by level. When the upstream sequential execution node occupies its time late, the amount is deducted sequentially according to the existing adjustable time of the adjacent sequential execution nodes. Based on the surgical risk level, a protection period is allocated within the adjustable time of the corresponding task that does not participate in bidding for other tasks. After each propagation, the effective time interval for each sequential execution node to participate in the next round of resource competition is synchronously adjusted.

7. The surgical kit delivery scheduling method based on a multi-agent system according to claim 4, characterized in that, Determining the resource allocation result based on the quoted price vector includes: verifying the delivery boundary of the task to be allocated, the delivery boundary of the existing task, and the protection time period for the execution sequence corresponding to the candidate quoted price, and eliminating the candidate quoted price that causes any delivery boundary mismatch; The remaining bids are sorted into a bid sequence based on the additional time occupied by the tasks to be assigned, the adjustable time consumption of existing tasks, the time occupied by task transfer and connection, and the delivery boundary compression. When multiple bids share resources, a mutual exclusion selection relationship is established, and a task allocation set corresponding to the executable time periods of each resource is determined from the bid combination that does not have resource time period conflicts.

8. A method for scheduling the delivery of surgical kits based on a multi-agent system according to claim 7, characterized in that, The local rearrangement includes: keeping the order of tasks with immovable delivery boundaries in the committed tasks of candidate resources unchanged, inserting the tasks to be assigned into the candidate gaps formed by the remaining movable tasks, and simultaneously determining the pickup connection time, delivery transfer time and resource empty running time of the tasks before and after the insertion position. When the task to be assigned is completed across two execution entities, the arrival time window corresponding to the handover point is written into the commitment status record of the preceding and following execution entities respectively. Only the local rearrangement sequence with a common handover interval between the arrival time windows and without intruding into the protection time period is retained, and the local rearrangement sequence is encapsulated into the execution sequence corresponding to the candidate bid.

9. A method for scheduling the delivery of surgical kits based on a multi-agent system according to claim 1, characterized in that, When the remaining adjustable time of any sequential execution node is less than zero, extract the set of tasks that share resources with the sequential execution node and whose delivery boundaries are associated with its time occupation according to the task resource association graph, and freeze the resource allocation outside the task set. Within the task set, releaseable time intervals are identified according to the propagation direction of the time commitment chain based on the negative remaining adjustable time amount. Bid requests are then reissued only to candidate sorting or delivery resources with the releaseable time intervals. The reissued bids are then compared with the original commitment status records for version verification before updating the resource allocation of the corresponding sequential execution nodes and their subsequent sequential execution nodes.

10. A method for scheduling the delivery of surgical kits based on a multi-agent system according to claim 1, characterized in that, For the rolling adjusted resource allocation maintenance commitment version sequence, each commitment version records the task resource association graph version, time commitment chain version, quotation vector source, resource executable time period, and the identifier of the previous commitment version that was replaced. When receiving new inventory, sorting, delivery robot or channel status, the corresponding status timestamp is compared with the generation time of the current committed version, and an alternative offer is formed only based on status data whose timestamp is not earlier than the generation time of the current committed version; When multiple local adjustments involve the same resource time period, the adjustment scope is merged according to the dependencies between committed versions, and candidate allocations with circular resource occupancy dependencies are rolled back to the most recent committed version without resource conflicts before the resource allocation is re-determined.

Images