Three-dimensional warehouse goods scheduling method and device based on large model agent and medium

Abstract

The application discloses a three-dimensional warehouse goods scheduling method and device based on a large model agent, and a medium, the method integrates order data, inventory data and real-time state data of equipment of a three-dimensional warehouse to obtain scheduling input data; a scheduling optimization algorithm is called as a tool by a large model agent through arrangement cycles to obtain operation indexes; target operation parameters are obtained by iteratively optimizing operation parameters within a preset adjustment range according to the operation indexes; time efficiency is estimated based on the target operation parameters and the real-time state data of the equipment to jointly determine a starting time of goods preparation and a time of calling a vehicle for warehouse delivery; a goods arrangement and transfer, goods preparation and vehicle calling scheme for warehouse delivery is generated and integrated into a target scheduling scheme; when a preset degradation condition is not met, the target scheduling scheme is converted into a control instruction and is executed, otherwise, hierarchical degradation is performed. The application realizes scheduling self-adaptation, goods preparation and vehicle arrival occlusion by means of core operation of a scheduling optimization algorithm and limited parameter adjustment of a large model agent, and avoids unreliable decisions from directly acting on equipment.

Description

Technical Field

[0001] This invention relates to the field of logistics and distribution technology, and in particular to a method, apparatus and medium for scheduling goods in an automated warehouse based on a large-scale intelligent model. Background Technology

[0002] Automated storage and retrieval systems (AS / RS) are automated warehousing systems that utilize stacker cranes to automatically store and retrieve goods on high-rise racks, use shuttle cars to transport goods between platforms within the warehouse, and complete loading and unloading at platforms. They are widely used in the logistics and distribution field. Outbound operations in AS / RS typically consist of two stages: sorting and transferring goods within the warehouse according to orders; and loading and unloading, where goods are removed from the racks, transported to the platforms via shuttle cars, and then loaded onto trucks. The overall efficiency of these outbound operations highly depends on the coordinated scheduling of AS / RS equipment such as stacker cranes, shuttle cars, and platforms.

[0003] Currently, cargo scheduling in automated warehouses (AS / RS) largely relies on scheduling optimization algorithms with fixed operating parameters. These parameters are typically determined manually through repeated adjustments based on experience. However, the daily order structure and real-time equipment load of AS / RS fluctuate dynamically. Algorithms with fixed operating parameters struggle to adapt to changes in order structure and equipment load. Furthermore, the timing of inventory preparation and vehicle dispatching is difficult to dynamically match with the real-time equipment status. This leads to operational congestion during peak preparation periods, waiting times between goods and transport vehicles at the platform, and redundant in-warehouse transfers. Consequently, AS / RS equipment utilization is low, and outbound operation cycles are long, failing to meet the logistics and distribution requirements for timely outbound delivery and operational efficiency. Summary of the Invention

[0004] This invention provides a method, apparatus, and medium for scheduling goods in automated warehouses based on a large model intelligent agent. It aims to solve the technical problems in the prior art where the scheduling optimization algorithm used for goods scheduling in automated warehouses has fixed operating parameters, which are difficult to adaptively adjust with order structure and real-time equipment load. Furthermore, the timing of goods preparation and outbound vehicle dispatch is difficult to dynamically match with the real-time status of the equipment, resulting in low equipment utilization and long outbound operation cycles.

[0005] In a first aspect, embodiments of the present invention provide a method for scheduling goods in an automated warehouse based on a large-scale intelligent agent, including: Acquire order data, inventory data, and real-time equipment status data from the automated warehouse, integrate the acquired data, and obtain the scheduling input data. The scheduling input data is sent to the large model agent, which then uses a scheduling optimization algorithm as a tool through a cyclical arrangement to obtain job metrics. Adjust the running parameters of the scheduling optimization algorithm within a preset adjustment range according to the work indicators, and rerun the iterative optimization to obtain the target running parameters; Based on the target operating parameters and the real-time status data of the equipment, the time consumption of the operation is estimated in order to jointly determine the start time of preparation and the time of dispatching the truck. Based on the stock preparation start time and the outbound vehicle call time, a sorting and transfer plan, a stock preparation plan and an outbound vehicle call plan are generated respectively, and then integrated to obtain a target scheduling plan. If the preset degradation conditions are not met, the target scheduling scheme is converted into control instructions for the automated warehouse equipment and issued for execution. If the preset degradation conditions are met, the scheduling source is switched to perform hierarchical degradation.

[0006] Secondly, embodiments of the present invention provide a three-dimensional warehouse cargo scheduling device based on a large-scale intelligent agent, comprising: The data acquisition unit is used to acquire order data, inventory data, and real-time equipment status data of the automated warehouse, and integrate the acquired data to obtain scheduling input data. The agent orchestration unit is used to send the scheduling input data to the large model agent, which then uses the scheduling optimization algorithm as a tool through an orchestration loop to obtain job metrics. The parameter optimization unit is used to adjust the running parameters of the scheduling optimization algorithm within a preset adjustment range according to the job indicators, and rerun the iterative optimization to obtain the target running parameters. The time-efficiency estimation unit is used to estimate the time-efficiency of the operation based on the target operating parameters and the real-time status data of the equipment, so as to jointly determine the start time of preparation and the time of dispatching the truck. The scheme generation unit is used to generate a sorting and transfer scheme, a stock preparation scheme, and an outbound vehicle dispatching scheme based on the stock preparation start time and the outbound vehicle dispatching time, and integrate them to obtain a target scheduling scheme. The scheduling execution unit is used to convert the target scheduling scheme into control instructions for the automated warehouse equipment and issue them for execution when the preset degradation conditions are not met, and to switch the scheduling source to perform hierarchical degradation when the preset degradation conditions are met.

[0007] Thirdly, embodiments of the present invention provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the automated warehouse cargo scheduling method based on a large model intelligent agent of the first aspect.

[0008] Fourthly, embodiments of the present invention provide a computer-readable storage medium, wherein a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, it implements the automated warehouse cargo scheduling method based on a large model intelligent agent of the first aspect.

[0009] This invention provides a method for scheduling goods in an automated warehouse based on a large-scale intelligent agent. The method includes acquiring order data, inventory data, and real-time equipment status data of the automated warehouse; integrating the acquired data to obtain scheduling input data; sending the scheduling input data to the large-scale intelligent agent, which then uses a scheduling optimization algorithm as a tool through a cyclical orchestration process to obtain operational indicators; adjusting the running parameters of the scheduling optimization algorithm within a preset adjustment range based on the operational indicators, and re-running iterative optimization to obtain target running parameters; estimating the time consumption of operations based on the target running parameters and the real-time equipment status data to jointly determine the start time for goods preparation and the time for outbound vehicle dispatch; generating a goods handling and transfer plan, a goods preparation plan, and an outbound vehicle dispatch plan based on the start time for goods preparation and the time for outbound vehicle dispatch, respectively, and integrating them to obtain a target scheduling plan; converting the target scheduling plan into control instructions for the automated warehouse equipment and issuing them for execution when preset degradation conditions are not met; and switching the scheduling source to perform tiered degradation when preset degradation conditions are met. This invention utilizes a scheduling optimization algorithm as a tool, which is then orchestrated and iteratively invoked by a large model agent. The algorithm adjusts and iteratively optimizes the operating parameters within a preset range based on operational indicators, enabling the algorithm's parameters to adaptively adjust according to order structure and real-time equipment load, overcoming the insufficient adaptability caused by fixed operating parameters. By estimating operation time based on target operating parameters and real-time equipment status data, the invention jointly determines the start time for goods preparation and the time for dispatching vehicles, ensuring that the time when goods are prepared coincides with the time when transport vehicles arrive at the platform, reducing mutual waiting between goods and vehicles at the platform. Furthermore, by switching the scheduling source to implement tiered degradation when preset degradation conditions are met, unreliable scheduling decisions are prevented from directly impacting the automated warehouse equipment, ensuring the continuity and stability of scheduling execution. This improves the equipment utilization rate of the automated warehouse, shortens the outbound operation cycle, and reduces redundancy in warehouse transfers.

[0010] This invention also provides a three-dimensional warehouse cargo scheduling device, computer equipment, and storage medium based on a large model intelligent agent, which have the same beneficial effects as described above. Attached Figure Description

[0011] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 A flowchart illustrating a three-dimensional warehouse cargo scheduling method based on a large-scale intelligent agent, provided in an embodiment of the present invention; Figure 2 This is a schematic block diagram of a three-dimensional warehouse cargo scheduling device based on a large model intelligent agent, provided as an embodiment of the present invention. Detailed Implementation

[0013] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0014] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0015] 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 invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0016] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0017] Please see below. Figure 1 , Figure 1 The flowchart of a three-dimensional warehouse cargo scheduling method based on a large model intelligent agent provided in an embodiment of the present invention specifically includes steps S101 to S106.

[0018] S101. Obtain order data, inventory data, and real-time equipment status data from the automated warehouse, and integrate the obtained data to obtain scheduling input data; S102. The scheduling input data is sent to the large model agent, which then uses the scheduling optimization algorithm as a tool through a scheduling loop to obtain job indicators. S103. Adjust the running parameters of the scheduling optimization algorithm within a preset adjustment range according to the work indicators, and rerun the iterative optimization to obtain the target running parameters; S104. Based on the target operating parameters and the real-time status data of the equipment, the time consumption of the operation is estimated in order to jointly determine the start time of preparation and the time of dispatching the truck. S105. Generate a sorting and transfer plan, a stock preparation plan, and an outbound vehicle dispatching plan based on the stock preparation start time and the outbound vehicle dispatching time, and integrate them to obtain a target scheduling plan. S106. If the preset degradation conditions are not met, the target scheduling scheme is converted into a control command for the automated warehouse equipment and issued for execution. If the preset degradation conditions are met, the scheduling source is switched to perform hierarchical degradation.

[0019] In step S101, an automated warehouse (AS / RS) refers to an automated warehousing system that automatically completes the storage and retrieval of goods on high-rise racks using stacker cranes, transports goods between stations within the warehouse using circular shuttles, and completes loading and unloading via platforms. A stacker crane is an automated device that runs along aisles and stores and retrieves goods between rack locations and work stations; a circular shuttle is an automated device that transports goods between picking and unloading stations on a circular track; and a platform is a location for transport vehicles to park and load. Order data describes the composition and quantity requirements of goods to be shipped; inventory data describes the storage location, batch affiliation, and availability of each item within the AS / RS; and real-time equipment status data describes the current operating load and availability of AS / RS equipment such as stacker cranes and circular shuttles. This step structurally integrates these three types of data to form unified scheduling input data for subsequent scheduling, enabling subsequent scheduling decisions to simultaneously consider order demand, inventory distribution, and real-time equipment load.

[0020] In one embodiment, step S101 includes: Obtain the order data for the next day, and analyze the outbound items and required quantities for each order to obtain the order data for the automated warehouse; Based on the order data of the automated warehouse, the location, batch, entry time and availability of each item are read to obtain the inventory data; For each stacker crane corresponding to the inventory data, the work queue length and availability status data are collected separately and integrated to obtain stacker crane status data; The stacker crane status data is integrated with the load level data of the circular shuttle to obtain real-time equipment status data. Unavailable stacker cranes are excluded based on the real-time status data of the equipment, and the various data are integrated to obtain the scheduling input data.

[0021] In this embodiment, the next-day order data is first obtained. This next-day order data refers to the detailed data of each order with a shipment date of the following day. The next-day order data is then parsed to obtain the items to be shipped for each order and the corresponding required quantities for each item, thus obtaining the automated warehouse's order data. For example, after parsing, an order reveals that its shipment items are item A and item B. The required quantity for item A is nine pallets (a pallet is a unit of measurement for goods, the same below), and the required quantity for item B is six pallets. These shipment items and required quantities are then used as the order data for that order. Using the next-day order data as scheduling input allows sufficient time for scheduling decisions and inventory handling operations before the shipment date.

[0022] Subsequently, based on the order data of the automated warehouse, the location, batch number, entry time, and availability status of each item are retrieved to obtain inventory data. The location refers to the specific storage position of the goods on the high-level shelves of the automated warehouse; the batch number identifies the production or entry batch of the goods; the entry time reflects the order in which the goods entered the warehouse; and the availability status indicates whether the goods can be retrieved normally, for example, goods in a locked or frozen state are unavailable.

[0023] For each stacker crane corresponding to the inventory data, the work queue length and availability status data are collected separately and integrated to obtain stacker crane status data. The work queue length refers to the number of tasks that the stacker crane has not yet completed, reflecting its workload. The availability status of the stacker crane indicates whether it is currently in normal, faulty, or under maintenance. The stacker crane status data is then integrated with the collected load level of the circular shuttle to obtain real-time equipment status data. The load level of the circular shuttle characterizes the current workload of its transport tasks; for example, it can be the ratio of the number of tasks en route to its rated concurrent tasks. All of the above status data can be periodically collected and updated via a data interface connected to the automated storage and retrieval system (AS / RS) equipment, ensuring that the real-time equipment status data reflects the latest operational status of the AS / RS in a timely manner.

[0024] Finally, unavailable stacker cranes are excluded based on real-time equipment status data. For example, stacker cranes with a faulty or under-maintenance status are excluded so that they are no longer optional equipment for subsequent scheduling, thereby avoiding the generation of job arrangements that are bound to be unexecuted. Then, order data, inventory data and real-time equipment status data are integrated to obtain scheduling input data.

[0025] In step S102, the large model agent refers to an autonomous executor that uses a large language model as its core, capable of reasoning and selecting actions based on observation; the scheduling optimization algorithm refers to a deterministic algorithm that produces a definite output given input, contains no randomness, and is used to perform core scheduling operations such as cargo handling and transfer, stock preparation, and outbound vehicle dispatch. In this step, the large model agent is positioned as an orchestration layer rather than a computation layer: the large model agent does not directly execute core scheduling operations, nor does it access the internal computation process of the scheduling optimization algorithm. Instead, it uses the scheduling optimization algorithm as a callable tool through an orchestration loop. The scheduling optimization algorithm completes the core scheduling operations based on the scheduling input data and returns job indicators. Job indicators refer to quantitative indicators used to characterize the quality of the scheduling optimization algorithm's performance, such as one or more of the following: planned completion time, load balancing among stacker cranes, peak concurrent stock preparation quantity, average mutual waiting time between goods and transport vehicles on the platform side, and total number of transfers. Thus, the large model agent obtains adaptive orchestration capabilities for the scheduling process, while the scheduling optimization algorithm ensures the determinism and reproducibility of the core scheduling operation results.

[0026] In one embodiment, step S102 includes: The scheduling input data is sent to the large model agent, and analysis tools, parameter tuning tools, query tools and interface tools are configured for the large model agent respectively, and integrated to obtain a tool set; By setting a step limit and action budget for calling the aforementioned toolset, a bounded reasoning-action orchestration loop is obtained; The scheduling optimization algorithm is run through the analysis tool in the bounded reasoning-action orchestration loop to obtain the running results; The running results are fed back into the large model agent as observation results, and the process proceeds alternately with observation, reasoning and action. The loop ends when the final conclusion is output or the upper limit of the number of steps is reached, and the operation index is output.

[0027] In this embodiment, scheduling input data is sent to the large model agent, and analysis tools, parameter tuning tools, query tools, and closing tools are configured for the large model agent, integrating them into a toolset. The analysis tool runs scheduling optimization algorithms related to inventory management, transfer, or stock preparation and returns operational metrics; the parameter tuning tool adjusts the running parameters of the scheduling optimization algorithm; the query tool queries detailed information such as abnormal operation documents; and the closing tool provides operational conclusions at the end of the scheduling process. Each tool in the toolset is provided to the large model agent through a standardized calling interface. The large model agent can only obtain the operational results such as operational metrics and adjustable parameters of the scheduling optimization algorithm through tool calls, but cannot access the internal calculation process of the scheduling optimization algorithm.

[0028] By setting a step limit and action budget for toolkit invocation, a bounded inference-action orchestration loop is obtained. The step limit restricts the maximum number of steps a large model agent can invoke tools in a single orchestration process; the action budget limits the total number of actions allowed to be executed. The bounded inference-action orchestration loop refers to an orchestration process constrained by the step limit and action budget, progressing alternately through observation, inference, and action until it concludes. By setting clear boundaries for the orchestration loop, large model agents can avoid endless repetitive invocations, ensuring the orchestration process ends within predictable overhead.

[0029] In the bounded reasoning-action orchestration loop, the scheduling optimization algorithm is run through the analysis tool to obtain the results. These results are then fed back to the large model agent as observations, providing the large model agent with new observations for further reasoning and action selection. This process alternates between observation, reasoning, and action: at each step, the large model agent reasons based on current task indicators and historical observations, selects an action from the tool set, executes it, and feeds back the execution result as a new observation. The loop ends when the large model agent outputs a conclusion or reaches the step limit, and the task indicators are output. For example, if the large model agent observes a high total number of database moves in the returned task indicators after running the scheduling optimization algorithm through the analysis tool at a certain step, it will select the corresponding query or analysis action in the next step of reasoning to locate the cause and continue to observe the changes in the task indicators until a conclusion is reached.

[0030] In one embodiment, several operational constraints are set for the bounded reasoning-action orchestration loop, including: not allowing the loop to close before any operational results are generated, performing a confirmation before closing the loop, intercepting repeated actions that are exactly the same as the previous step, and terminating the loop when the large model agent continuously outputs illegal actions for a preset number of times, thereby ensuring that the operation of the large model agent is bounded and controllable.

[0031] In one embodiment, the orchestration of the large model agent is set in the nighttime or periodic planning phase, while the stock preparation and transfer control, executed according to adjusted operating parameters, is set in the real-time control phase that operates on a fixed cycle. The large model agent is not invoked in the real-time control phase: In the real-time control phase, current inventory data and real-time equipment status data are repeatedly read at fixed cycles, and a scheduling optimization algorithm is run using operating parameters determined in the planning phase to obtain the current cycle's work arrangement and execute it accordingly. Thus, the large model agent does not enter the hot path of real-time control, achieving adaptive orchestration capabilities without affecting the real-time performance and stability of real-time control.

[0032] In step S103, the operating parameters refer to the configurable parameters upon which the scheduling optimization algorithm operates. The values ​​of the operating parameters directly affect the algorithm's execution results and corresponding job performance indicators. In this step, the large model agent, based on the job performance indicators obtained in step S102, determines whether there is still room for improvement in the algorithm's execution results under the current operating parameters, and adjusts the operating parameters within a preset adjustment range. The preset adjustment range refers to a pre-defined restricted range for parameter adjustment, used to limit the allowed parameter items and the range of allowed values ​​for each parameter. After each adjustment, the scheduling optimization algorithm is rerun and the changes in job performance indicators are observed. This iterative optimization process is repeated to gradually approach the parameter values ​​that improve job performance indicators, ultimately obtaining the target operating parameters. By strictly limiting the influence of the large model agent on the scheduling optimization algorithm to parameter adjustments within the preset adjustment range, the operating parameters can be adaptively adjusted according to the order structure and real-time equipment load, while preventing out-of-bounds parameter values ​​from affecting the safety and stability of the scheduling operation.

[0033] In one embodiment, step S103 includes: Based on the work indicators, the operating parameters that are allowed to be adjusted within the preset adjustment range will be set as pre-set whitelist parameters, while parameters outside the whitelist will be refused adjustment. The adjusted whitelist parameters are clamped to a preset safe range and recorded to obtain the clamped parameter values. The scheduling optimization algorithm is rerun using the clamped parameter values ​​to obtain the adjusted job performance indicators. The whitelist parameters are iteratively optimized based on the changes in the work indicators and the adjusted work indicators to obtain the optimized parameters. The optimized parameters are provided in the form of suggestions and written into the production configuration through preset switch control to obtain the target operating parameters.

[0034] In this embodiment, based on the operational indicators, the adjustable operating parameters within a preset adjustment range are designated as a pre-defined whitelist of parameters, while adjustments to parameters outside the whitelist are rejected. The whitelist of parameters refers to a pre-defined set of operating parameters that are allowed to be adjusted by large model agents. For example, the whitelist parameters include at least the night advance preparation quantity, the upper limit of the prepared but not loaded temporary storage for each work loop, the maximum number of batches for a single order, the assumed travel time from the factory to the platform for reverse calculation, and the queue compensation coefficient. Among them, the work loop refers to the work area obtained by dividing the automated warehouse according to the outbound path; the night advance preparation quantity is used to limit the upper limit of the work quantity that can be prepared in advance during the night handling stage; the upper limit of the prepared but not loaded temporary storage is used to limit the upper limit of the temporary storage quantity of goods that have been prepared but not yet loaded in each work loop; the maximum number of batches for a single order is used to limit the upper limit of the number of batches that can be selected when selecting goods for the same order, so as to avoid increasing the picking and consolidation work by a single order spanning too many batches; the assumed travel time from the factory to the platform is used to estimate the travel time required for the transport vehicle to arrive at the platform before it enters the factory; the queue compensation coefficient is used to compensate for the queuing wait caused by the stacker crane operation queue or the load of the circular shuttle in the timeliness estimation in the subsequent step S104. The larger the value, the more conservative the estimate of the operation time. When a large model agent requests to adjust parameters that are not on the whitelist, the request is rejected directly, thus strictly limiting parameter adjustments to a preset range.

[0035] The adjusted parameter values ​​of the whitelist parameters are clamped to a preset safe range and recorded to obtain the clamped parameter values. The safe range refers to a pre-defined range of values ​​for each whitelist parameter; clamping means correcting the adjusted parameter value to the boundary value of that safe range when it exceeds the corresponding safe range. For example, if the safe range for a whitelist parameter is [a, b], when the adjustment value given by the large model agent is greater than b, the parameter value is clamped to b; when the adjustment value is less than a, the parameter value is clamped to a. The occurrence of clamping is recorded and communicated to the large model agent. Rejecting adjustments to non-whitelist parameters and clamping out-of-bounds parameter values ​​are both routine parameter protection processes. Through the dual constraints of whitelist parameters and safe ranges, the actual operating parameters used by the scheduling optimization algorithm are always within a safe and controllable range.

[0036] The scheduling optimization algorithm was rerun using the clamped parameter values ​​to obtain the adjusted operational metrics. During the execution of the scheduling optimization algorithm, each whitelist parameter was substituted into its corresponding calculation stage: the amount of goods prepared in advance at night and the upper limit of the temporary storage of goods prepared but not loaded were substituted into the calculation related to off-peak advance preparation; the maximum number of batches for a single order was substituted into the calculation related to goods selection; the assumed travel time from the factory to the platform was substituted into the calculation of the outbound dispatch timing; and the queuing compensation coefficient was substituted into the calculation related to timeliness prediction. Thus, the large model agent can influence the running result of the scheduling optimization algorithm by adjusting the whitelist parameters without ever touching the internal calculations of the scheduling optimization algorithm.

[0037] The whitelist parameters are iteratively optimized based on the operational indicators and the changes in the adjusted operational indicators to obtain the optimized parameters. Specifically, the changes in operational indicators before and after each adjustment are compared: if an adjustment improves the operational indicators, such as reducing the total number of warehouse transfers or shortening the average mutual waiting time between goods and transport vehicles on the platform side, then optimization continues based on that adjustment; if an adjustment worsens the operational indicators, then that adjustment is rolled back and other adjustment directions are tried; this process is repeated repeatedly within a bounded reasoning-action orchestration loop until the operational indicators tend to stabilize or reach the step limit, thus obtaining the optimized parameters.

[0038] The optimized parameters are provided as suggestions and written into the production configuration via a preset switch to obtain the target operating parameters. Specifically, the optimized parameters do not take effect immediately but are output as parameter adjustment suggestions. When the preset switch is on, the optimized parameters are automatically written into the production configuration of the scheduling optimization algorithm and thus take effect. When the preset switch is off, the optimized parameters are not automatically written into the production configuration and can be reviewed and confirmed by operations personnel before adoption. Therefore, the impact of parameter adjustments on production operations remains under control.

[0039] In step S104, timeliness estimation refers to the prediction of when the operation will be completed based on the real-time status of the equipment. The operation time in the automated warehouse outbound process includes at least the stacker crane picking time and the circular shuttle transportation time, and its actual value changes dynamically with the real-time status of the equipment, such as the stacker crane operation queue length and the circular shuttle load level. This step estimates the operation time based on the target operating parameters and real-time equipment status data obtained in step S103, obtains the estimated time when the goods are ready, and then jointly determines the start time of goods preparation and the time of outbound vehicle dispatch. Among them, the start time of goods preparation refers to the time when the goods preparation operation of each order begins to be executed, and the time of outbound vehicle dispatch refers to the time when the transport vehicle is summoned to the platform for loading. By jointly determining the start time of goods preparation and the time of outbound vehicle dispatch based on the same timeliness estimation, the time when the goods are ready and the time when the transport vehicle arrives at the platform can be synchronized, avoiding long waiting times for goods or long waiting times for vehicles.

[0040] In one embodiment, step S104 includes: Based on the target operating parameters and the real-time status data of the equipment, the automated warehouse is divided into several operating loops according to the outbound path, and the historical transportation records of the circular shuttle car are collected. Based on the historical transportation records, a multi-level time consumption comparison relationship is constructed between platform layer, operation loop layer and default layer; Search layer by layer according to the multi-level time consumption comparison relationship and drop the layer when the sample is insufficient, and take the high quantile value to obtain the estimated time consumption of the task. The estimated time consumption is amplified based on the load level of the circular shuttle, and the latest start time for preparing goods for each order is calculated in reverse. The start time for stock preparation is determined based on the latest start time, and the timing for dispatching a vehicle is calculated by reverse calculation using the ideal vehicle arrival time at the platform.

[0041] In this embodiment, based on target operating parameters and real-time equipment status data, the automated warehouse is divided into several operating loops according to the outbound path, and historical transportation records of the circular shuttle are collected. The operating loops can be divided, for example, according to the aisle where the stacker crane is located and the access platforms it can reach, so that each operating loop includes a group of stacker cranes and one or more platforms corresponding to that group of stacker cranes. Retrieval and outbound operations within the same operating loop are completed through the platform corresponding to that operating loop. The granularity of the operating loop division can be determined based on the aisle layout and the number of platforms in the automated warehouse; this embodiment does not limit the specific method of dividing the operating loops. Historical transportation records refer to the time consumption records of each segment of transportation previously completed by the circular shuttle. Each historical transportation record includes at least the retrieval platform, the release platform, and the actual time consumed for that segment of transportation.

[0042] Based on historical transportation records, a multi-level time consumption comparison relationship is constructed, comprising platform-to-platform, operation loop-to-operation loop, and default layers. The platform-to-platform layer uses platform pairs (picking and releasing platforms) as the granular statistical time consumption sample; the operation loop-to-operation loop layer uses operation loop pairs (picking and releasing operation loops) as the granular statistical time consumption sample; and the default layer is a coarser-grained bottom layer, which can further include default layers within the same operation loop, default layers across operation loops, and a global default layer. This forms a multi-level time consumption comparison relationship from fine to coarse: the finer the granularity, the closer it is to the actual time consumption of a specific transportation route; the coarser the granularity, the more abundant the time consumption sample.

[0043] The search proceeds layer by layer according to the multi-level time consumption comparison relationship, and when the sample size is insufficient, it moves to the next lower level with the highest quantile value to obtain the estimated operation time. Specifically, when estimating the time consumption of a certain transportation segment, the search is performed sequentially according to the platform level, the operation loop level, and the default level. When the number of samples in the searched level is lower than a preset threshold (e.g., less than twenty records), the search moves to the next lower level with a coarser granularity, i.e., it moves to the next lower level. The selected time consumption estimate is the highest quantile value of the historical transportation record at the corresponding level, i.e.: in, This is an estimated transport time from entrance platform u to exit platform v. For the corresponding historical time consumption sample set, This indicates that the 90th percentile value is used. Using a high percentile value instead of the average value as the estimated time consumption allows for a margin when subsequently using it to reverse-calculate the timing of inventory preparation and dispatch, reducing the risk of delays in dispatch due to underestimation of time consumption.

[0044] The estimated time consumption is amplified based on the load level of the circular shuttle, i.e.: In the formula, This is the amplified time-consuming estimate. To adapt to real-time load levels An increasing amplification factor. As an example, the amplification factor can be taken as a linear increase with the load level. ,in To normalize the real-time load level, for example, the ratio of the number of tasks en route to the rated concurrent tasks of the circular shuttle, c is a non-negative coefficient, which can be determined, for example, by the queuing compensation coefficient in the target operating parameters. The larger c is, the greater the amplification under the same load level; the amplification coefficient can also be chosen as... Other forms, such as monotonically increasing piecewise functions or exponential functions, are not limited to this embodiment. Furthermore, as historical transportation records continue to accumulate, the multi-level time consumption comparison relationship is refitted, so that the time consumption estimate gradually becomes more accurate with the accumulation of operational data.

[0045] Based on the amplified time-consuming estimates, the latest start time for preparing inventory for each order is calculated in reverse. Specifically, for each order, the latest start time for inventory preparation is obtained by subtracting the sum of the stacker crane picking time and the circular shuttle transportation time, which are estimated based on the required completion time for inventory preparation. The completion time for inventory preparation can be determined based on the order's outbound schedule or the estimated arrival time of its transport vehicles. As the time-consuming estimates are updated with real-time equipment status data, the latest start time for inventory preparation for each order is also dynamically adjusted accordingly.

[0046] The start time for order preparation is determined based on the latest start time. In one embodiment, access control is performed separately for each work loop, and the concurrent order preparation limit for each work loop is set to the number of platforms corresponding to that work loop. Within a preset look-ahead time window, the latest start time for order preparation for each order is mapped to several time buckets. The look-ahead time window refers to a preset duration of time window observed from the current time, and the time bucket refers to a time segment obtained by dividing the look-ahead time window with a fixed width. The width of the look-ahead time window and the time bucket can be preset, for example, the look-ahead time window is several hours and the width of the time bucket is fifteen to thirty minutes. When the number of orders in any time bucket exceeds the concurrent order preparation limit, it is determined that the work loop is congested within the look-ahead time window. Only when congestion is determined to occur is the order preparation operation of subsequent orders allowed to start in advance to the current time, and the number of orders advanced does not exceed the temporary storage limit of the prepared but not loaded order for that work loop. Orders that have arrived at the factory but have not yet been prepared are placed in the highest order preparation priority. Therefore, when a peak in inventory preparation is anticipated, inventory can be prepared in advance to smooth out the peak operating time, and temporary storage space will not be occupied in advance when no congestion occurs.

[0047] The ideal vehicle arrival time is used to calculate the dispatch timing from the warehouse. The ideal vehicle arrival time refers to the optimal time for a transport vehicle to arrive at the platform, assuming the goods are ready and the vehicle arrives at the same time. In one embodiment, the ideal vehicle arrival time is the later of the following two times: the time from the completion of goods preparation to the platform (minus the total loading time) and the initial delivery arrival time. in, For the ideal train's arrival time at the platform; The estimated time for the last shipment of goods in the order to arrive at the platform is the time when the goods are prepared for delivery. The vehicle loading time can be determined based on historical loading records or a preset value. The estimated arrival time of the first shipment at the platform is the time when the first shipment of goods in the order is expected to arrive at the platform. and All of these can be calculated by adding the time required for stacker crane retrieval and the time required for circular shuttle transportation, which are estimated based on the start time of goods preparation.

[0048] After determining the ideal vehicle's arrival time at the platform, the time to call for a vehicle is calculated in reverse. The time to call for a vehicle is the specific time corresponding to the opportunity to call for a vehicle upon leaving the depot. When the transport vehicle has entered the factory but has not yet been called to the platform, the time to call for a vehicle is calculated in reverse from the ideal vehicle's arrival time at the platform, based on the travel time from the parking lot to the platform. When the transport vehicle has not yet entered the factory but has an estimated arrival time, the earliest time the vehicle can reach the platform is determined by adding the estimated arrival time to the assumed travel time from entering the factory to the platform. If the ideal vehicle's arrival time at the platform is not later than this earliest time, the vehicle is called immediately; otherwise, the time to call for a vehicle is calculated in reverse from the ideal vehicle's arrival time at the platform after the vehicle has entered the factory. The time to call for a vehicle satisfies the following relationship: in, For the scheduled vehicle call time, A represents the vehicle's arrival time at the factory or its estimated arrival time, R represents the estimated cargo readiness time, L represents the safe lead time for the cargo to arrive at the platform earlier than the vehicle, and T represents the travel time from the scheduled call to the vehicle's arrival at the platform. The sum of the estimated cargo readiness time R and the safe lead time L represents the ideal vehicle arrival time at the platform. That is, the scheduled vehicle call time is calculated by subtracting the travel time T from the ideal vehicle arrival time at the platform, and then extrapolating backwards using the vehicle's arrival time at the factory or its estimated arrival time A as a lower bound. This relationship, compared to the previously mentioned method of taking the later of the two, is simply a different representation of the same ideal vehicle arrival time at the platform. The thus determined departure call timing ensures that the arrival time of the transport vehicle at the platform coincides with the cargo readiness time, reducing mutual waiting between cargo and vehicles at the platform.

[0049] In step S105, based on the stock preparation start time and outbound vehicle dispatch time determined in step S104, the scheduling optimization algorithm organizes specific operation plans according to a deterministic process. Specifically, the stock handling and transfer plan describes the stock handling operation arrangement of gathering and redistributing goods among stacker cranes within the warehouse before stock preparation; the stock preparation plan describes the stock preparation operation arrangement of retrieving goods from each stacker crane according to the stock preparation start time and transporting them to the platform via a circular shuttle; and the outbound vehicle dispatch plan describes the dispatching arrangement of calling transport vehicles to the corresponding platform for loading according to the outbound vehicle dispatch time. These three plans respectively cover the stock handling, stock preparation, and loading connection links of outbound operations. After numerical verification, they are integrated to form a complete target scheduling plan. Furthermore, given the same input data and operating parameters, the scheduling optimization algorithm produces deterministic and reproducible outputs, ensuring that the scheduling plan is traceable and verifiable.

[0050] In one embodiment, step S105 includes: Based on the stock preparation start time and the outbound vehicle call time, the combination of order and inventory unit is used as the processing unit, and the demand of multiple details in the same processing unit is merged. The automated warehouse is divided into several operational loops according to the outbound path, and a dominant operational loop is selected for the processing unit. The balanced share is calculated based on the number of stacker cranes available within the dominant operating loop, and batch grouping of goods is carried out according to the first-in-first-out rule; Goods exceeding the equilibrium share are moved to stacker cranes with insufficient holdings, and integrated to obtain a sorting and transfer scheme, a stocking scheme, and an outbound dispatching scheme. The numerical compliance of the number of tasks for each device in each generated scheme is verified to obtain the target scheduling scheme.

[0051] In this embodiment, based on the stock preparation start time and the outbound dispatch time, the combination of order and inventory unit is used as the processing unit, and the demand quantities of multiple details within the same processing unit are merged. Here, an inventory unit refers to an aggregated inventory unit in an automated warehouse that is uniformly managed according to attributes such as item type; for example, one inventory unit corresponds to one item. A processing unit refers to a scheduling object composed of one order and one inventory unit. A detail refers to an entry in an order that records the specific item and quantity. When multiple details exist within the same inventory unit of the same order, the demand quantities of each detail are merged into the total demand quantity of that processing unit, avoiding fragmented and repetitive operations within the same processing unit.

[0052] The automated warehouse is divided into several operational loops according to the outbound route, and a dominant operational loop is selected for each processing unit. The dominant operational loop refers to the operational loop selected from the available operational loops for a given processing unit, used for the outbound selection of goods by that processing unit. For example, it can be determined based on the outbound operational loop specified in the order or the main distribution of the goods in that processing unit.

[0053] The equilibrium share is calculated based on the number of available stacker cranes within the dominant work loop, and batch grouping is performed according to a first-in, first-out (FIFO) rule. Unavailable stacker cranes are excluded from the source and destination of goods handling and transfer operations, as well as from the candidate list for goods preparation and selection, based on real-time equipment status data. They are not included in the available stacker crane count, thus avoiding inevitably failed transfer or retrieval operations. The equilibrium share is calculated by dividing the demand of the processing unit by the number of available stacker cranes within the dominant work loop and rounding up. in, To balance the share, To determine the demand for processing units, K represents the number of stacker cranes available within the dominant work loop. First-in, first-out (FIFO) refers to a selection method where goods are selected based on their earliest arrival time in the batch, and goods from the same batch are selected in groups. Selecting goods in groups according to the FIFO rule ensures that goods that arrive first are removed first, while also reducing the number of batches spanned by the same order. The number of batches selected within the same order is limited by the maximum number of batches per single order mentioned above.

[0054] Goods exceeding the equilibrium share are moved to stacker cranes with insufficient holdings, and integrated into three schemes: a goods handling and transfer scheme, a goods preparation scheme, and a goods dispatch scheme. Specifically, for each stacker crane in the dominant operating loop: when the quantity of target goods held by it does not exceed the equilibrium share, no transfer occurs; when the quantity of target goods held by it exceeds the equilibrium share, only the portion exceeding the equilibrium share is moved to a stacker crane with a holding below the equilibrium share; for goods that need to be replenished from other operating loops, they are aggregated to stacker cranes in the dominant operating loop with a holding below the equilibrium share. For example, if a processing unit's demand is nine pallets, and the number of available stacker cranes K in the dominant operating loop is three, then the equilibrium share is three pallets; if the three stacker cranes currently hold five pallets, three pallets, and one pallet of target goods respectively, then only the two pallets exceeding the equilibrium share from the first stacker crane are moved to the third stacker crane with insufficient holdings, and the second stacker crane does not undergo a transfer. Therefore, goods in the same processing unit are distributed to multiple stacker cranes within the main operating loop according to a balanced share before preparation. This allows multiple stacker cranes to pick up goods in parallel during preparation, avoiding sequential waiting caused by concentrated picking on a few stacker cranes. Simultaneously, only the portion exceeding the balanced share is transferred, reducing the amount of goods transferred. After distribution, the picking operations of each stacker crane and the transportation operations of the circular shuttle are arranged into a preparation plan based on the start time of preparation. The dispatching arrangements for each order are arranged into an outbound dispatching plan based on the outbound dispatching time. Together with the sorting and transfer plan, these constitute the plan to be validated.

[0055] The numerical compliance of the number of operations for each device in each generated plan is verified to obtain the target scheduling plan. The numerical compliance verification refers to checking whether the number of operations for each device in the generated plan meets preset numerical constraints. For example, it verifies that the number of warehouse transfers and pickups for each stacker crane does not exceed its available inventory and available operational capacity, and does not exceed the balanced share. Plans that pass the numerical compliance verification are integrated into the target scheduling plan. The above plan generation process is carried out under the premise of meeting numerical constraints such as the available operational capacity and available inventory of each device. Its optimization objectives are reflected in: balancing the load of each stacker crane in the dominant operational loop through the distribution of the balanced share; reducing the peak concurrent inventory preparation quantity by organizing inventory preparation according to the start time of preparation; and reducing the mutual waiting time between goods and transport vehicles on the platform side by organizing vehicle dispatching according to the dispatching time.

[0056] In one embodiment, a large model agent generates operational attribution statements for frontline operators, which are explanatory texts in natural language describing the composition and rationale of scheduling arrangements. The operational attribution statements undergo numerical consistency verification, specifically verifying whether the number of jobs stated in the statements equals the actual number of equipment jobs in the generated plan. If they are inconsistent, the inconsistency is fed back to the large model agent as an observation to generate a revised operational attribution statement. If the inconsistency persists after a preset step limit, a deterministic template is used to generate the operational attribution statement. The result of the numerical consistency verification is only used to determine whether to regenerate the operational attribution statement and does not change the generated scheduling plan. Furthermore, the operational attribution statements can be verified to ensure their word count is within a preset range, that they contain preset prohibited terms, and that they cover one or more of the preset essential elements. Prohibited terms include, for example, exaggerated or speculative wording, and essential elements include, for example, the stacker crane identification, the number of warehouse transfer or retrieval operations, and the target platform identification.

[0057] In another embodiment, the automated warehouse is a hazardous chemical warehouse, which is divided into several isolated storage areas. The division of storage areas is determined according to existing hazardous chemical zoning management rules. In this case, when generating a cargo handling and transfer plan, transfer constraints are also imposed based on the isolation relationship between storage areas, prohibiting the movement of goods between isolated storage areas; the distribution of equal shares and the collection of goods across work loops are both limited to the same storage area. When the goods required to replenish the leading work loop are located in other storage areas that are isolated from it, cross-storage transfer is not carried out; and when selecting goods in batches according to the first-in-first-out rule, the shelf life and quality inspection qualification status of the goods are considered for screening, and goods that have exceeded their shelf life or have failed quality inspection are excluded from the selection.

[0058] In step S106, the target scheduling scheme is implemented and execution security is guaranteed. Preset degradation conditions are used to determine the reliability of the scheduling result obtained from the large model agent as the scheduling source: if the preset degradation conditions are not met, it indicates that the target scheduling scheme is reliable and executable, and it is converted into control commands for the automated storage and retrieval system (AS / RS) equipment and issued for execution; if the preset degradation conditions are met, it indicates that the current scheduling result has a risk of unreliability, and it is not applied to the AS / RS equipment. Instead, the scheduling source is switched to implement a graded degradation, and a scheduling source with verified reliability takes over the scheduling decision. Therefore, regardless of the reliability of the large model agent's output, the scheduling execution of the AS / RS can proceed safely and continuously.

[0059] In one embodiment, step S106 includes: If the preset degradation conditions are not met, the target scheduling scheme is converted into control instructions for the stacker crane, the circular shuttle car and the platform; the control instructions are sent to the corresponding equipment controllers through the communication interface and executed after a feasibility self-check; wherein, the preset degradation conditions include numerical verification failure, budget exhaustion or confidence level lower than a preset threshold; When the preset degradation conditions are met, the scheduling optimization algorithm replaces the large model agent in outputting scheduling decisions; when the output scheduling decisions do not meet the preset pass conditions, the scheduling and allocation are handled by the warehouse management system or warehouse control system.

[0060] In this embodiment, when the preset degradation conditions are not met, the target scheduling scheme is converted into control instructions for stacker cranes, circular shuttles, and platforms: the sorting and transfer scheme is converted into transfer instructions indicating the source stacker crane, target stacker crane, and quantity of goods; the stocking scheme is converted into stocking instructions instructing each stacker crane to retrieve the corresponding goods and transport them to the platform via the circular shuttle; the outbound vehicle calling scheme is converted into a vehicle calling instruction instructing a transport vehicle to be summoned to the corresponding platform at the designated time. These control instructions are sent to the corresponding equipment controllers via the communication interface and executed after a feasibility self-assessment. Here, the communication interface refers to the standardized data transmission interface that interfaces with each equipment controller; the equipment controller refers to the control device that directly controls the actions of equipment such as stacker cranes, circular shuttles, and platforms; the feasibility self-assessment refers to the review of whether the control instructions can be actually executed under the current equipment and inventory conditions before execution. For example, the warehouse management system or warehouse control system reviews the control instructions and reserves the right to refuse execution. The warehouse management system refers to the information system that manages warehouse orders, inventory, and other business operations, while the warehouse control system refers to the information system that controls and coordinates the operations of various equipment in the automated warehouse. In one embodiment, the specific allocation of storage locations and containers is determined automatically by the warehouse management system or warehouse control system within the range of the stacker crane indicated by the control command.

[0061] The preset degradation conditions include numerical verification failure, budget exhaustion, or confidence level below a preset threshold. Numerical verification failure means the generated solution fails the numerical compliance check in step S105, indicating a numerical error in the solution itself and its inability to be deployed. Budget exhaustion means the upper limit of the bounded inference-action orchestration loop's steps or the action budget has been exhausted without a reliable conclusion. Confidence level below the preset threshold means the confidence level of the large model agent's output is below the preset threshold. The confidence level characterizes the credibility of the scheduling decision output by the large model agent. It can be determined by one or more methods, such as the large model agent's self-assessed confidence score for the generated solution at the conclusion, the consistency ratio between solutions obtained from multiple runs of the bounded inference-action orchestration loop on the same input, and the proportion of solutions generated during iterative optimization that pass the numerical compliance check. The preset threshold can be calibrated based on historical running data; for example, when normalizing the confidence level to the range of 0 to 1, the preset threshold can be 0.7.

[0062] When preset degradation conditions are met, the scheduling source is switched to perform tiered degradation. Tiered degradation refers to fault-tolerant processing that switches scheduling sources sequentially according to a pre-defined order. This embodiment sets up three levels of scheduling sources for sequential degradation: the first level is a large model agent, the second level is a scheduling optimization algorithm, and the third level is a warehouse management system or warehouse control system. When the first level meets the preset degradation conditions, the scheduling optimization algorithm replaces the large model agent in outputting scheduling decisions. That is, the results tuned by the large model agent are no longer used, and the scheduling optimization algorithm independently completes the scheduling calculation with verified operating parameters. When the scheduling decision output by the scheduling optimization algorithm does not meet the preset pass conditions, the warehouse management system or warehouse control system performs scheduling and allocation according to its own scheduling and allocation logic. The preset pass conditions mean that the scheduling decision output by the scheduling optimization algorithm passes the scheme numerical compliance verification and is feasible and can be issued. For example, when the scheduling optimization algorithm cannot generate a scheduling decision that passes the scheme numerical compliance verification and can be issued due to insufficient available operational capacity or available inventory, it is determined that the preset pass conditions are not met.

[0063] Through the above-mentioned hierarchical degradation, when the output of the large model agent is unreliable, the scheduling decisions that may have biases are not directly applied to physical equipment such as stacker cranes, circular shuttle cars and platforms. Instead, they are gradually degraded to deterministic scheduling optimization algorithms or even existing warehouse management systems or warehouse control systems, thus ensuring the safety and continuity of automated warehouse scheduling execution.

[0064] In summary, the automated warehouse cargo scheduling method based on a large-scale intelligent agent provided in this invention integrates order data, inventory data, and real-time equipment status data to obtain scheduling input data. The large-scale intelligent agent then uses a scheduling optimization algorithm as a tool through orchestration and looping, adjusting and iteratively optimizing operating parameters within a preset range. This allows the scheduling strategy to adaptively adjust to changes in order structure and real-time equipment load. By jointly determining the start time for goods preparation and the time for dispatching vehicles based on the time-sensitivity prediction of target operating parameters and real-time equipment status data, the timing of goods preparation and the arrival time of transport vehicles at the platform are synchronized, reducing mutual waiting between goods and vehicles at the platform. The target scheduling scheme is obtained by generating a sorting and transfer plan, a goods preparation plan, and a dispatching plan, and integrating them after verifying the numerical compliance of the schemes. This enables multiple stacker cranes to operate in parallel and reduces transfer redundancy. By switching the scheduling source and implementing hierarchical degradation when preset degradation conditions are met, unreliable scheduling decisions are prevented from directly affecting the automated warehouse equipment. Therefore, this invention improves the equipment utilization rate of the automated warehouse, shortens the outbound operation cycle, and balances the adaptability, interpretability, and execution security and stability of cargo scheduling.

[0065] Combination Figure 2 As shown, Figure 2 This invention provides a schematic block diagram of an automated warehouse cargo scheduling device based on a large-scale intelligent agent. The automated warehouse cargo scheduling device 200 based on a large-scale intelligent agent includes: The data acquisition unit 201 is used to acquire order data, inventory data and real-time equipment status data of the automated warehouse, and integrate the acquired data to obtain scheduling input data. The agent orchestration unit 202 is used to send the scheduling input data to the large model agent, and the large model agent calls the scheduling optimization algorithm as a tool through the orchestration loop to obtain the job indicators. The parameter optimization unit 203 is used to adjust the running parameters of the scheduling optimization algorithm within a preset adjustment range according to the job indicators, and rerun the iterative optimization to obtain the target running parameters. The timeliness estimation unit 204 is used to estimate the timeliness of the operation based on the target operating parameters and the real-time status data of the equipment, so as to jointly determine the start time of preparation and the time of dispatching the truck. The scheme generation unit 205 is used to generate a sorting and transfer scheme, a stocking scheme and an outbound vehicle call scheme according to the stocking start time and the outbound vehicle call time, and integrate them to obtain a target scheduling scheme. The scheduling execution unit 206 is used to convert the target scheduling scheme into control instructions for the automated warehouse equipment and issue them for execution when the preset degradation conditions are not met, and to switch the scheduling source to perform hierarchical degradation when the preset degradation conditions are met.

[0066] In this embodiment, the data acquisition unit 201 acquires order data, inventory data, and real-time equipment status data from the automated warehouse, integrates the acquired data to obtain scheduling input data; the intelligent agent orchestration unit 202 sends the scheduling input data to the large model intelligent agent, which uses the scheduling optimization algorithm as a tool through an orchestration loop to obtain work indicators; the parameter optimization unit 203 adjusts the running parameters of the scheduling optimization algorithm within a preset adjustment range according to the work indicators, and re-runs the iterative optimization to obtain target running parameters; the timeliness prediction... The estimation unit 204 estimates the operation time based on the target operating parameters and the real-time status data of the equipment to jointly determine the start time of preparation and the time of dispatching the vehicle; the scheme generation unit 205 generates a sorting and transfer scheme, a preparation scheme and a dispatching scheme respectively according to the start time of preparation and the time of dispatching the vehicle, and integrates them to obtain the target scheduling scheme; the scheduling execution unit 206 converts the target scheduling scheme into a control command for the automated warehouse equipment and issues it for execution when the preset degradation conditions are not met, and switches the scheduling source to perform hierarchical degradation when the preset degradation conditions are met.

[0067] In one embodiment, the data acquisition unit 201 is specifically used for: Obtain the order data for the next day, and analyze the outbound items and required quantities for each order to obtain the order data for the automated warehouse; Based on the order data of the automated warehouse, the location, batch, entry time and availability of each item are read to obtain the inventory data; For each stacker crane corresponding to the inventory data, the work queue length and availability status data are collected separately and integrated to obtain stacker crane status data; The stacker crane status data is integrated with the load level data of the circular shuttle to obtain real-time equipment status data. Unavailable stacker cranes are excluded based on the real-time status data of the equipment, and the various data are integrated to obtain the scheduling input data.

[0068] In one embodiment, the intelligent agent orchestration unit 202 is specifically used for: The scheduling input data is sent to the large model agent, and analysis tools, parameter tuning tools, query tools and interface tools are configured for the large model agent respectively, and integrated to obtain a tool set; By setting a step limit and action budget for calling the aforementioned toolset, a bounded reasoning-action orchestration loop is obtained; The scheduling optimization algorithm is run through the analysis tool in the bounded reasoning-action orchestration loop to obtain the running results; The running results are fed back into the large model agent as observation results, and the process proceeds alternately with observation, reasoning and action. The loop ends when the final conclusion is output or the upper limit of the number of steps is reached, and the operation index is output.

[0069] In one embodiment, the parameter optimization unit 203 is specifically used for: Based on the work indicators, the operating parameters that are allowed to be adjusted within the preset adjustment range will be set as pre-set whitelist parameters, while parameters outside the whitelist will be refused adjustment. The adjusted whitelist parameters are clamped to a preset safe range and recorded to obtain the clamped parameter values. The scheduling optimization algorithm is rerun using the clamped parameter values ​​to obtain the adjusted job performance indicators. The whitelist parameters are iteratively optimized based on the changes in the work indicators and the adjusted work indicators to obtain the optimized parameters. The optimized parameters are provided in the form of suggestions and written into the production configuration through preset switch control to obtain the target operating parameters.

[0070] In one embodiment, the timeliness estimation unit 204 is specifically used for: Based on the target operating parameters and the real-time status data of the equipment, the automated warehouse is divided into several operating loops according to the outbound path, and the historical transportation records of the circular shuttle car are collected. Based on the historical transportation records, a multi-level time consumption comparison relationship is constructed between platform layer, operation loop layer and default layer; Search layer by layer according to the multi-level time consumption comparison relationship and drop the layer when the sample is insufficient, and take the high quantile value to obtain the estimated time consumption of the task. The estimated time consumption is amplified based on the load level of the circular shuttle, and the latest start time for preparing goods for each order is calculated in reverse. The start time for stock preparation is determined based on the latest start time, and the timing for dispatching a vehicle is calculated by reverse calculation using the ideal vehicle arrival time at the platform.

[0071] In one embodiment, the scheme generation unit 205 is specifically used for: Based on the stock preparation start time and the outbound vehicle call time, the combination of order and inventory unit is used as the processing unit, and the demand of multiple details in the same processing unit is merged. The automated warehouse is divided into several operational loops according to the outbound path, and a dominant operational loop is selected for the processing unit. The balanced share is calculated based on the number of stacker cranes available within the dominant operating loop, and batch grouping of goods is carried out according to the first-in-first-out rule; Goods exceeding the equilibrium share are moved to stacker cranes with insufficient holdings, and integrated to obtain a sorting and transfer scheme, a stocking scheme, and an outbound dispatching scheme. The numerical compliance of the number of tasks for each device in each generated scheme is verified to obtain the target scheduling scheme.

[0072] In one embodiment, the scheduling execution unit 206 is specifically used for: If the preset degradation conditions are not met, the target scheduling scheme is converted into control instructions for the stacker crane, the circular shuttle car and the platform; the control instructions are sent to the corresponding equipment controllers through the communication interface and executed after a feasibility self-check; wherein, the preset degradation conditions include numerical verification failure, budget exhaustion or confidence level lower than a preset threshold; When the preset degradation conditions are met, the scheduling optimization algorithm replaces the large model agent in outputting scheduling decisions; when the output scheduling decisions do not meet the preset pass conditions, the scheduling and allocation are handled by the warehouse management system or warehouse control system.

[0073] Since the embodiments of the apparatus and the embodiments of the method correspond to each other, please refer to the description of the embodiments of the method for the embodiments of the apparatus, which will not be repeated here.

[0074] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed, can perform the steps provided in the above embodiments. The storage medium may include various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0075] This invention also provides a computer device, which may include a memory and a processor. The memory stores a computer program, and when the processor calls the computer program in the memory, it can implement the steps provided in the above embodiments. Of course, the computer device may also include various network interfaces, a power supply, a graphics card, etc., to utilize the graphics card's performance to operate the model, such as for inference and training.

[0076] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to in the method section. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from the principles of this application, and these improvements and modifications also fall within the protection scope of the claims of this application.

[0077] It should also be noted that, in this specification, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

Claims

1. A method for scheduling goods in an automated warehouse based on a large-scale intelligent agent model, characterized in that, include: Acquire order data, inventory data, and real-time equipment status data from the automated warehouse, integrate the acquired data, and obtain the scheduling input data. The scheduling input data is sent to the large model agent, which then uses a scheduling optimization algorithm as a tool through a cyclical arrangement to obtain job metrics. Adjust the running parameters of the scheduling optimization algorithm within a preset adjustment range according to the work indicators, and rerun the iterative optimization to obtain the target running parameters; Based on the target operating parameters and the real-time status data of the equipment, the time consumption of the operation is estimated in order to jointly determine the start time of preparation and the time of dispatching the truck. Based on the stock preparation start time and the outbound vehicle call time, a sorting and transfer plan, a stock preparation plan and an outbound vehicle call plan are generated respectively, and then integrated to obtain a target scheduling plan. If the preset degradation conditions are not met, the target scheduling scheme is converted into control instructions for the automated warehouse equipment and issued for execution; if the preset degradation conditions are met, the scheduling source is switched to perform hierarchical degradation. The method of estimating the time consumption of operations based on the target operating parameters and the real-time status data of the equipment to jointly determine the start time of inventory preparation and the time of dispatching outbound vehicles includes: dividing the automated warehouse into several operation loops according to the outbound path based on the target operating parameters and the real-time status data of the equipment, and collecting historical transportation records of the circular shuttle; constructing multi-level time consumption comparison relationships between platforms, operation loops, and default layers according to the historical transportation records; searching layer by layer according to the multi-level time consumption comparison relationships and dropping layers when the samples are insufficient, taking the high quantile value to obtain the estimated time consumption of operations; amplifying the estimated time consumption according to the load level of the circular shuttle, and back-calculating the latest start time of inventory preparation for each order; determining the start time of inventory preparation based on the latest start time of inventory preparation, and back-calculating the time of dispatching outbound vehicles based on the ideal vehicle arrival time at the platform.

2. The automated warehouse cargo scheduling method based on a large-scale intelligent agent according to claim 1, characterized in that, The process involves acquiring order data, inventory data, and real-time equipment status data from the automated warehouse, integrating the acquired data to obtain scheduling input data, including: Obtain the order data for the next day, and analyze the outbound items and required quantities for each order to obtain the order data for the automated warehouse; Based on the order data of the automated warehouse, the location, batch, entry time and availability of each item are read to obtain the inventory data; For each stacker crane corresponding to the inventory data, the work queue length and availability status data are collected separately and integrated to obtain stacker crane status data; The stacker crane status data is integrated with the load level data of the circular shuttle to obtain real-time equipment status data. Unavailable stacker cranes are excluded based on the real-time status data of the equipment, and the various data are integrated to obtain the scheduling input data.

3. The automated warehouse cargo scheduling method based on a large-scale intelligent agent according to claim 1, characterized in that, The process involves sending the scheduling input data to a large model agent, which then uses a scheduling optimization algorithm as a tool through a orchestration loop to obtain job metrics, including: The scheduling input data is sent to the large model agent, and analysis tools, parameter tuning tools, query tools and interface tools are configured for the large model agent respectively, and integrated to obtain a tool set; By setting a step limit and action budget for calling the aforementioned toolset, a bounded reasoning-action orchestration loop is obtained; The scheduling optimization algorithm is run through the analysis tool in the bounded reasoning-action orchestration loop to obtain the running results; The running results are fed back into the large model agent as observation results, and the process proceeds alternately with observation, reasoning and action. The loop ends when the final conclusion is output or the upper limit of the number of steps is reached, and the operation index is output.

4. The automated warehouse cargo scheduling method based on a large-scale intelligent agent according to claim 1, characterized in that, The step of adjusting the running parameters of the scheduling optimization algorithm within a preset adjustment range according to the job indicators, and re-running the iterative optimization to obtain the target running parameters includes: Based on the work indicators, the operating parameters that are allowed to be adjusted within the preset adjustment range will be set as pre-set whitelist parameters, while parameters outside the whitelist will be refused adjustment. The adjusted whitelist parameters are clamped to a preset safe range and recorded to obtain the clamped parameter values. The scheduling optimization algorithm is rerun using the clamped parameter values ​​to obtain the adjusted job performance indicators. The whitelist parameters are iteratively optimized based on the changes in the work indicators and the adjusted work indicators to obtain the optimized parameters. The optimized parameters are provided in the form of suggestions and written into the production configuration through preset switch control to obtain the target operating parameters.

5. The automated warehouse cargo scheduling method based on a large-scale intelligent agent according to claim 1, characterized in that, The process generates a sorting and transfer plan, a stock preparation plan, and an outbound vehicle dispatch plan based on the stock preparation start time and the outbound vehicle dispatch time, respectively, and integrates them to obtain a target scheduling plan, including: Based on the stock preparation start time and the outbound vehicle call time, the combination of order and inventory unit is used as the processing unit, and the demand of multiple details in the same processing unit is merged. The automated warehouse is divided into several operational loops according to the outbound path, and a dominant operational loop is selected for the processing unit. The balanced share is calculated based on the number of stacker cranes available within the dominant operating loop, and batch grouping of goods is carried out according to the first-in-first-out rule; Goods exceeding the equilibrium share are moved to stacker cranes with insufficient holdings, and integrated to obtain a sorting and transfer scheme, a stocking scheme, and an outbound dispatching scheme. The numerical compliance of the number of tasks for each device in each generated scheme is verified to obtain the target scheduling scheme.

6. The automated warehouse cargo scheduling method based on a large-scale intelligent agent according to claim 1, characterized in that, The step of converting the target scheduling scheme into control commands for the automated warehouse equipment and issuing them for execution when the preset degradation conditions are not met, and switching the scheduling source to perform tiered degradation when the preset degradation conditions are met, includes: If the preset degradation conditions are not met, the target scheduling scheme is converted into control instructions for the stacker crane, the circular shuttle car and the platform; the control instructions are sent to the corresponding equipment controllers through the communication interface and executed after a feasibility self-check; wherein, the preset degradation conditions include numerical verification failure, budget exhaustion or confidence level lower than a preset threshold; When the preset degradation conditions are met, the scheduling optimization algorithm replaces the large model agent in outputting scheduling decisions; when the output scheduling decisions do not meet the preset pass conditions, the scheduling and allocation are handled by the warehouse management system or warehouse control system.

7. A three-dimensional warehouse cargo scheduling device based on a large-scale intelligent agent model, characterized in that, include: The data acquisition unit is used to acquire order data, inventory data, and real-time equipment status data of the automated warehouse, and integrate the acquired data to obtain scheduling input data. The agent orchestration unit is used to send the scheduling input data to the large model agent, which then uses the scheduling optimization algorithm as a tool through an orchestration loop to obtain job metrics. The parameter optimization unit is used to adjust the running parameters of the scheduling optimization algorithm within a preset adjustment range according to the job indicators, and rerun the iterative optimization to obtain the target running parameters. The time-efficiency estimation unit is used to estimate the time-efficiency of the operation based on the target operating parameters and the real-time status data of the equipment, so as to jointly determine the start time of preparation and the time of dispatching the truck. The scheme generation unit is used to generate a sorting and transfer scheme, a stock preparation scheme, and an outbound vehicle dispatching scheme based on the stock preparation start time and the outbound vehicle dispatching time, and integrate them to obtain a target scheduling scheme. The scheduling execution unit is used to convert the target scheduling scheme into control instructions for the automated warehouse equipment and issue them for execution when the preset degradation conditions are not met; and to switch the scheduling source to perform hierarchical degradation when the preset degradation conditions are met. The timeliness estimation unit is specifically used for: dividing the automated warehouse into several operation loops according to the outbound path based on the target operating parameters and the real-time status data of the equipment, and collecting historical transportation records of the circular shuttle; constructing multi-level time consumption comparison relationships between platform layers, operation loop layers, and default layers according to the historical transportation records; searching layer by layer according to the multi-level time consumption comparison relationships and dropping layers when the samples are insufficient, taking the high quantile value to obtain the estimated time consumption of the operation; amplifying the estimated time consumption according to the load level of the circular shuttle, and reverse-calculating the latest preparation start time for each order; determining the preparation start time according to the latest preparation start time, and reverse-calculating the outbound call time through the ideal car arrival time at the platform.

8. A computer device, characterized in that, The system includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the automated warehouse cargo scheduling method based on a large model intelligent agent as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the automated warehouse cargo scheduling method based on a large model intelligent agent as described in any one of claims 1 to 6.

Images