A method, system and device for efficient storage of pallets in a logistics store

Abstract

This invention relates to the field of intelligent logistics warehousing and automated storage technology, specifically a method, system, and apparatus for efficient pallet storage in logistics storage. The method includes: collecting actual outbound instructions and pallet flow records as initial data packets; performing time-series correlation analysis to obtain the spatiotemporal coupling characteristics of orders; extracting pallet flow probabilities and calculating predictive cache hit rates; classifying heat levels based on historical outbound frequencies and assigning migration priorities; calculating the physical topology entropy reduction index of the storage area in conjunction with pallet flow probabilities; extracting idle time slices of handling equipment and calculating fragmented time rebalancing conversion rates; constructing a multi-objective optimization function and solving it to obtain a dynamic topology rearrangement strategy; generating migration trigger conditions and cache execution instructions, and executing them when the conditions are met; and calculating scheduling preemption rollback delays and formulating rearrangement interruption recovery strategies when sudden tasks have higher priority. This invention thereby improves the accessibility and scheduling stability of high-frequency pallets.

Description

Technical Field

[0001] This invention relates to the field of intelligent logistics warehousing and automated storage technology, specifically to a method, system and device for efficient pallet storage in logistics storage. Background Technology

[0002] As a crucial foundation for modern warehousing and regional distribution, logistics storage systems are experiencing continuous expansion in scale and increasingly significant order fluctuations. To ensure timely outbound response and stable in-warehouse scheduling, it is typically necessary to combine handling equipment, storage areas, and pallet location status to rationally store and dynamically schedule pallets.

[0003] However, in the pallet storage management process for multi-region automated warehouses, existing methods mostly rely on static warehouse location planning, historical experience zoning, or passive addressing after order arrival. Equipment needs to be temporarily called upon to complete retrieval and handling only after the actual outbound task is triggered. Pallet storage efficiency is not only related to the historical outbound frequency of goods, but also closely related to the outbound time distribution, warehouse location depth, regional topology, and equipment idle time utilization. In order to improve outbound efficiency, pallet adjustment is usually carried out by moving pallets forward according to popularity, placing them forward according to frequency, or manually setting priorities. Although these methods can improve local retrieval efficiency to a certain extent, they do not make sufficient use of the correlation between order sequence and spatial location. Moreover, during the rearrangement process, they are easily affected by factors such as equipment busy / idle status, aisle conflicts, and sudden urgent order insertions, resulting in low migration hit rate, interference between back-end handling and main tasks, and thus unstable pallet rearrangement optimization effect, increased equipment scheduling resource load, and difficulty in balancing the accessibility of high-frequency outbound pallets with the overall operational stability of the logistics storage system. Summary of the Invention

[0004] The purpose of this invention is to provide a method, system, and apparatus for efficient pallet storage in logistics storage, and to solve the following technical problems:

[0005] The original passive picking process that relied on static storage locations has been transformed into a proactive rescheduling process that is predictable, interruptible, and recoverable. This enables advance adaptation to order peaks and improves the accessibility and scheduling stability of high-frequency pallets without changing the existing hardware layout.

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] A method for efficient pallet storage in logistics storage, applied to a logistics storage system including handling equipment, outbound gates, and multiple storage areas, comprising:

[0008] The initial data packet is obtained by collecting actual outbound instructions and pallet transfer records in the logistics storage system. The transfer records contain the historical storage location coordinates of the pallets. The initial data packet is subjected to time series correlation analysis to obtain the spatiotemporal coupling characteristics of the orders. The pallet transfer probability is extracted from the spatiotemporal coupling characteristics of the orders and the predictive cache hit rate is calculated.

[0009] The pallet is divided into multiple heat levels using a heat assessment algorithm based on historical outbound frequency. Each heat level is assigned a decreasing migration priority in descending order of heat. The physical topology entropy reduction index of each storage area is calculated by using the migration priority and the pallet flow probability.

[0010] Idle time slices of handling equipment are extracted and the fragment time rebalancing conversion rate is calculated. A multi-objective optimization function is constructed based on the physical topology entropy reduction index and the fragment time rebalancing conversion rate. The multi-objective optimization function is solved by an optimization algorithm to obtain a dynamic topology rearrangement strategy.

[0011] By combining migration priority and dynamic topology reordering strategy, migration trigger conditions and corresponding cache execution instructions are generated;

[0012] The system collects the real-time scheduling status of the transport equipment. When the real-time scheduling status meets the migration triggering conditions, it executes the corresponding cached execution instructions. At the same time, it monitors the real-time scheduling status for sudden tasks. When a sudden task is detected, it obtains the priority of the sudden task. When the priority of the sudden task is higher than the migration priority corresponding to the current cached execution instruction, it calculates the scheduling preemption rollback delay and formulates a reordering interrupt recovery strategy.

[0013] Preferably, the steps of extracting pallet flow probability from order spatiotemporal coupling characteristics and calculating predictive cache hit rate include:

[0014] By fitting the probability distribution of the outbound time series in the spatiotemporal coupling characteristics of orders, the pallet circulation probability of the outbound time series under different preset time windows is obtained.

[0015] Within the preset prediction period, the first number of target pallets at a specific heat level that were moved in advance and subsequently hit by actual outbound instructions, and the total number of actual outbound instructions issued, are counted as the second number.

[0016] Calculate the ratio of the first count to the second count, and use the ratio as the predictive cache hit rate.

[0017] Preferably, the step of calculating the physical topology entropy reduction index of each storage region based on migration priority and tray flow probability includes:

[0018] The variance of the distance between the current storage location of the pallet and the corresponding target outbound port in each storage area is calculated as the initial arrangement disorder, and the expected consecutive outbound order of the pallets in each storage area is obtained as the time-series associated outbound sequence.

[0019] The pallet turnover probability is weighted according to the migration priority of each heat level to obtain a weighted value. The weighted value is then combined with the depth characteristics of the current storage location of the pallet to calculate the pallet picking complexity.

[0020] Calculate the difference between the initial arrangement disorder and the picking complexity, and normalize the difference using the length of the time-series outbound sequence to obtain the physical topological entropy reduction index of each storage area.

[0021] Preferably, the step of extracting idle time slices of the handling equipment and calculating the fragment time rebalancing conversion rate includes:

[0022] Monitor the operating cycle of the handling equipment and separate the idle time window without tasks from the operating cycle as idle time slices;

[0023] The statistics include the cumulative duration of idle time slices used to perform tray migration operations, and the total duration of the running cycle;

[0024] Calculate the ratio of cumulative duration to total duration, and use the ratio as the fragmented time rebalancing conversion rate.

[0025] Preferably, the steps for constructing a multi-objective optimization function based on the physical topological entropy reduction exponent and the fragmented time rebalancing conversion rate include:

[0026] After normalizing the parameters corresponding to the predictive cache hit rate, physical topology entropy reduction exponent, and optimization objectives, a multi-objective initial function is constructed.

[0027] The optimization objectives of the multi-objective initial function include maximizing the smoothness of the flow, minimizing the picking time consumption, and minimizing the fatigue wear of the handling equipment.

[0028] The weight coefficients of the multi-objective initial function are adjusted based on the fragmented time rebalancing conversion rate. The adjusted weight coefficients are then substituted into the multi-objective initial function to obtain the multi-objective optimization function.

[0029] Preferably, the steps for solving the multi-objective optimization function using an optimization algorithm to obtain the dynamic topology reshuffling strategy include:

[0030] Initialize the algorithm parameters of the optimization algorithm, including the initial temperature, cooling coefficient, and maximum number of iterations;

[0031] Generate an initial rearrangement strategy as the current rearrangement strategy, and begin iteration;

[0032] In each iteration, perform the following steps:

[0033] Candidate rearrangement strategies are generated by adjusting the pallet target storage location in the current rearrangement strategy.

[0034] The candidate rearrangement strategy and the current rearrangement strategy are respectively used as inputs to the multi-objective optimization function for calculation, and the function value of the candidate rearrangement strategy and the function value of the current rearrangement strategy are subtracted to obtain the function difference value;

[0035] Calculate the probability of accepting the candidate rearrangement strategy based on the function difference and the temperature of the current iteration;

[0036] The decision to adopt a candidate rearrangement strategy is based on probability, and the current rearrangement strategy is updated accordingly. The temperature for the next iteration is updated based on the cooling coefficient.

[0037] When the number of iterations reaches the maximum number of iterations, the iteration stops, and the current rearrangement strategy is output as the dynamic topology rearrangement strategy.

[0038] Preferably, when a sudden task is detected, the priority of the sudden task is obtained. When the priority of the sudden task is higher than the migration priority corresponding to the currently cached execution instruction, the steps of calculating the scheduling preemption rollback delay and formulating a reordering interrupt recovery strategy include:

[0039] In response to the detection of a burst task with a higher priority than the currently cached execution instruction, the interruption point of the currently cached execution instruction is recorded, and the time difference from pausing the current cached execution instruction to the complete release of the currently occupied path by the transport equipment and the start of execution of the burst task is calculated. The time difference is used as the scheduling preemption rollback delay.

[0040] Determine the relationship between the scheduling preemption rollback delay and the preset maximum tolerable delay threshold. The maximum tolerable delay threshold is determined based on the upper limit of the physical time taken for the handling equipment to exit the currently occupied path and perform avoidance actions.

[0041] When the scheduling preemption rollback delay is less than or equal to the maximum tolerable delay threshold, a first reordering interrupt recovery strategy is formulated. The first reordering interrupt recovery strategy includes suspending the currently cached execution instructions, switching to execute a burst task, and resuming the migration operation from the interrupt breakpoint after the burst task ends.

[0042] When the scheduling preemption rollback delay exceeds the maximum tolerable delay threshold, a second reordering interrupt recovery strategy is formulated. The second reordering interrupt recovery strategy includes rejecting the preemption request of the burst task and continuing to execute the current cached execution instruction until the tray corresponding to the current cached execution instruction is in place before responding to the burst task.

[0043] A system for efficient pallet storage in logistics storage, applicable to a logistics storage system including handling equipment, an outbound port, and multiple storage areas, comprising:

[0044] The data acquisition and processing module is used to collect actual outbound instructions and pallet flow records in the logistics storage system as initial data packets. The flow records contain the historical storage location coordinates of the pallets. The module performs time-series correlation analysis on the initial data packets to obtain the spatiotemporal coupling characteristics of orders. The pallet flow probability is extracted from the spatiotemporal coupling characteristics of orders, and the predictive cache hit rate is calculated.

[0045] The heat assessment calculation module is used to divide the pallet into multiple heat levels using a heat assessment algorithm based on historical outbound frequency. According to the order of heat from high to low, each heat level is assigned a decreasing migration priority. The physical topology entropy reduction index of each storage area is calculated by using the migration priority and the pallet flow probability.

[0046] The rearrangement optimization solution module is used to extract the idle time slices of the handling equipment and calculate the fragment time rebalancing conversion rate. Based on the physical topology entropy reduction index and the fragment time rebalancing conversion rate, a multi-objective optimization function is constructed, and the multi-objective optimization function is solved by the optimization algorithm to obtain the dynamic topology rearrangement strategy.

[0047] The instruction generation module is used to combine migration priority and dynamic topology reordering strategy to generate migration trigger conditions and corresponding cache execution instructions.

[0048] The scheduling execution monitoring module is used to collect the real-time scheduling status of the handling equipment. When the real-time scheduling status meets the migration trigger condition, the corresponding cached execution instruction is executed. At the same time, the real-time scheduling status is monitored for sudden tasks. When a sudden task is detected, the priority of the sudden task is obtained. When the priority of the sudden task is higher than the migration priority corresponding to the current cached execution instruction, the scheduling preemption rollback delay is calculated and a reordering interrupt recovery strategy is formulated.

[0049] An apparatus comprising: a memory and a processor;

[0050] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the method for efficiently storing pallets in logistics storage are implemented.

[0051] The beneficial effects of this invention are:

[0052] 1. This invention performs time-series correlation analysis on outbound instructions and circulation records to extract pallet circulation probability and calculate predictive cache hit rate. This mechanism overcomes the lag of passive addressing after order arrival, realizes the transformation from passive picking to active predictive migration, can accurately lock pallets with high probability of outbound within a specific time window and place them in advance, significantly improves the actual hit rate of cache operation, and effectively reduces invalid handling.

[0053] 2. This invention uses heat-based stratification to assign priority to pallet migration and combines the flow probability to calculate the physical topology entropy reduction index of each storage area. This method breaks the limitations of traditional local optimization that relies solely on placing a single pallet at the front of the historical frequency. It quantifies and reduces the disorder of physical space arrangement from the perspective of the entire region, making the rearranged outbound path more regular and greatly improving the overall smoothness of continuous outbound tasks and the regional rearrangement optimization index.

[0054] 3. This invention extracts idle time slices of handling equipment to calculate the fragmented time rebalancing conversion rate, and dynamically adjusts the multi-objective optimization function to solve the reordering strategy accordingly. This solution solves the problem of mutual interference and resource contention between background reordering and forward outbound main task. Under the premise of ensuring the scheduling of main task, fragmented equipment idle time is transformed into safe pre-migration execution force, realizing a dynamic balance between improving outbound efficiency and reducing equipment fatigue wear.

[0055] 4. This invention monitors sudden tasks in real time when executing cache instructions, compares priorities and calculates scheduling preemption rollback delays, and formulates differentiated reordering interruption recovery strategies. This mechanism effectively resolves the problem of repeated equipment rollback and lane deadlock caused by sudden urgent order insertions. While ensuring that high-priority orders are responded to in a timely manner, it makes the reordering interruption process completely controllable and smooth and recoverable, thus ensuring scheduling stability in complex logistics environments. Attached Figure Description

[0056] The invention will now be further described with reference to the accompanying drawings.

[0057] Figure 1 A flowchart illustrating an efficient pallet storage method in logistics storage, provided as an embodiment of this application;

[0058] Figure 2 This is a schematic diagram of a system for efficient pallet storage in logistics storage provided in an embodiment of this application. Detailed Implementation

[0059] 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 embodiments of the present invention, and not all embodiments. 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.

[0060] Please see Figure 1A method for efficient pallet storage in logistics storage is applied to a logistics storage system that includes handling equipment, outbound ports and multiple storage areas. The method includes: collecting actual outbound instructions and pallet flow records in the logistics storage system as initial data packets. The flow records contain the historical storage location coordinates of the pallets. The method also includes performing time-series correlation analysis on the initial data packets to obtain the spatiotemporal coupling characteristics of orders, extracting the pallet flow probability from the spatiotemporal coupling characteristics of orders and calculating the predictive cache hit rate.

[0061] The pallet is divided into multiple heat levels using a heat assessment algorithm based on historical outbound frequency. Each heat level is assigned a decreasing migration priority in descending order of heat. The physical topology entropy reduction index of each storage area is calculated by using the migration priority and the pallet flow probability.

[0062] Idle time slices of the handling equipment are extracted and the fragment time rebalancing conversion rate is calculated. A multi-objective optimization function is constructed based on the physical topology entropy reduction index and the fragment time rebalancing conversion rate. The multi-objective optimization function is solved by an optimization algorithm to obtain a dynamic topology reordering strategy. By combining migration priority and dynamic topology reordering strategy, migration triggering conditions and cache execution instructions corresponding to the migration triggering conditions are generated.

[0063] The system collects the real-time scheduling status of the transport equipment. When the real-time scheduling status meets the migration triggering conditions, it executes the corresponding cached execution instructions. At the same time, it monitors the real-time scheduling status for sudden tasks. When a sudden task is detected, it obtains the priority of the sudden task. When the priority of the sudden task is higher than the migration priority corresponding to the current cached execution instruction, it calculates the scheduling preemption rollback delay and formulates a reordering interrupt recovery strategy.

[0064] This embodiment provides an efficient pallet storage mechanism for automated warehouses in regional distribution centers. Specifically, a continuously operating main scenario is set: a regional warehouse of a pharmaceutical e-commerce company enters its evening order collection peak period from 18:00 to 22:00 every day. The warehouse contains three storage areas: A, B, and C, each corresponding to a different aisle. The handling equipment includes a stacker crane, a shuttle car, and two automated guided vehicles. The outbound ports are set as O1 and O2. The pallets in the warehouse contain both room temperature medicines and high-frequency promotional sets.

[0065] This mechanism does not passively address each order after it arrives, but rather, when there is available idle space on the equipment, it moves pallets that are likely to be shipped out to shallower or closer locations that are easier to retrieve, based on the order sequence. Specifically, it collects actual shipping instructions and pallet flow records to form an initial data packet.

[0066] Actual outbound instructions can include at least the order number, issuance time, target outbound port, category identifier, and priority; circulation records can include at least the pallet number, historical inbound time, historical outbound time, historical storage location coordinates, and handling route;

[0067] For ease of explanation, a simplified data extrapolation model can be used: In an observation window before 19:00, the system records the historical behavior of four pallets P1, P2, P3, and P4. Among them, P1 was located in the deep position A-05-07 of Zone A at 18:12, was moved to the shallow position A-02-02 of Zone A at 18:43, and was shipped out from O1 at 18:49; P2 was shipped directly from O2 from Zone B B-04-06 at 18:15.

[0068] P3 was moved from C-06-08 to C-03-01 in Zone C at 18:18 and then shipped out at 18:51; P4 remained untouched at 18:20; Meanwhile, the shipment instruction sequence was as follows: Order D1 requested P1 at 18:40, Order D2 requested P3 at 18:46, and Order D3 requested P2 at 18:47; The system correlates the temporal sequence with the spatial location changes and can extract the spatiotemporal coupling characteristics that, within a specific time window, pallets moved to shallower locations are more likely to match with subsequent orders.

[0069] Specifically, the spatiotemporal coupling characteristics of orders are represented as a multi-dimensional data matrix, whose feature dimensions include at least: the initial layer depth identifier of the pallet, the linear displacement difference before and after the warehouse location transfer, the time interval between the transfer operation and the subsequent order call, and the quantity distribution vector of other pallets with the same outbound destination within the same time window. Through the structured representation of the above dimensions, discrete historical flow records are transformed into standard inputs that can be used for probabilistic model calculations;

[0070] Based on this, the system extracts the pallet turnover probability. An hour can be divided into multiple time windows. For example, the frequency of each type of pallet changing from its current storage location to its outbound status within a 10-minute window can be counted. If 35 out of 100 historical samples of a certain type of promotional set are outbound within the 18:30-18:40 window, the turnover probability corresponding to that time window can be recorded as 0.35. If only 8 outbound samples of another type of regular medicine are outbound, the probability can be recorded as 0.08.

[0071] Furthermore, within the preset prediction period, the number of times a task is migrated in advance and then hit by the actual outbound shipment is counted. For example, if 20 tasks are migrated in advance within one hour, and 14 of them are hit in subsequent actual orders, while the total outbound shipment task is 40, then the predictive cache hit rate can be taken as 14 / 40, which is 0.35.

[0072] It is emphasized here that the hit statistics are based on whether the actual outbound task is matched with the pre-ready pallet, rather than simply counting whether the migration action is executed, thus reflecting the effectiveness of the pre-migration; pallets are stratified by popularity; popularity can be comprehensively evaluated based on historical outbound frequency, recent outbound growth rate, and the frequency of order-related occurrences;

[0073] For ease of understanding, three layers can be set: high-heat layer H1, medium-heat layer H2, and low-heat layer H3; if P1 has been requested 12 times in the past two hours and often leaves the warehouse together with P3, it is classified into H1; if P2 has been requested 5 times, it is classified into H2; if P4 has only been requested once, it is classified into H3; correspondingly, the migration priorities of the three layers are assigned to 3, 2, and 1 respectively.

[0074] Based on the pallet flow probability, the physical topology entropy reduction index of each storage area is calculated. Intuitively, this index measures whether the outbound path is more orderly and whether the disorder is reduced after the area is rearranged. For example, the distances from the existing pallets in area A to the outbound port O1 are 8 meters, 2 meters, 9 meters and 3 meters respectively. The variance exceeds the preset dispersion threshold, indicating that the disorder of the arrangement is high.

[0075] If the high-heat pallet is moved to a position near the outbound exit, the distance distribution tends to be consistent with the expected outbound sequence, and the physical topological entropy reduction index increases accordingly. The system extracts idle time slices of the handling equipment; for example, the total operating cycle of the stacker crane from 18:00 to 19:00 is 3600 seconds, of which 2200 seconds are actually used to execute the main task, 600 seconds are used for standby and interleaving scheduling, and the rest are for relocation and avoidance time; if 420 seconds of these 600 seconds are used to complete the background entropy reduction migration, then the fragmented time rebalancing conversion rate can be taken as 420 / 3600.

[0076] The physical topology entropy reduction index and the conversion rate are then fed into a multi-objective optimization function to simultaneously balance flow smoothness, picking time, and equipment fatigue. After solving, a dynamic topology rearrangement strategy is output. For example, during 18:35-18:50, the H1 pallet in area A is moved to the first two shallow columns of area A, and unnecessary handling of the low-heat pallets in area C is temporarily suspended. Furthermore, the system generates migration triggering conditions and cached execution instructions based on migration priority and dynamic topology rearrangement strategy.

[0077] Migration triggering conditions may include: current equipment utilization rate is below 80%, target roadway is unobstructed, expected turnover probability is above 0.3 within the next 15 minutes, and there are currently no higher priority outbound tasks;

[0078] The cache execution instruction can be as specific as the stacker crane moving P1 from A-05-07 to A-02-02, and then the shuttle moving P3 from C-06-08 to C-03-01; when the real-time scheduling status is collected and these conditions are met, the system automatically executes the corresponding instruction;

[0079] During execution, the system continuously monitors for sudden tasks; for example, at 19:02, an urgent hospital order E1 is suddenly received, which requires the immediate retrieval of the anti-infective drug tray PX from area B, and this task has a higher priority than the background cache migration that is currently being executed.

[0080] At this point, the system calculates the scheduling preemption rollback delay, which is the time required from pausing the current migration action, recording the interruption position, releasing the roadway, to the equipment starting to execute the urgent order. If the delay meets the control boundary, the interruption is allowed and the switch is initiated. If not, the current migration is completed to the safe placement point before the switch is initiated, in order to avoid equipment jamming or interruption during execution that would cause the roadway to be occupied.

[0081] As an anomaly handling mechanism, if the initial data packet is missing, such as a pallet lacking complete historical storage location coordinates, the system will mark the pallet as a low-confidence sample and it will only participate in basic heat statistics in this round and will not participate in high-precision spatiotemporal coupling analysis.

[0082] If the number of samples in a certain time window is too small, such as less than 5, the system can expand to adjacent windows for merging and estimation to avoid distortion of the flow probability due to accidental data. If real-time monitoring finds that the equipment power is below the safety threshold, the roadway is occupied by other equipment, or the sensor status is abnormal, the corresponding cached execution instruction will automatically be put into a suspended state and wait for recovery before re-evaluation, rather than being forcibly executed.

[0083] In the aforementioned pharmaceutical e-commerce regional warehouse, before 19:00, the system judged based on historical data that the circulation probabilities of P1 and P3 in the next 20 minutes were 0.42 and 0.38, respectively, both of which were in the high-heat layer; the topology optimization efficiency brought by the rearrangement of areas A and C was higher than that of area B. Therefore, during the 90 seconds of idle time of the stacker crane, the system first moved P1 from A-05-07 to A-02-02, and then moved P3 from C-06-08 to C-03-01.

[0084] At 19:06, after the real order arrives, P1 and P3 can be directly retrieved by shallow positioning, shortening the average pickup path; at 19:08, an urgent delivery task E1 appears. The system detects that the current background migration has not yet started the second step, and the rollback delay is only 4 seconds, which is lower than the set threshold of 8 seconds. Therefore, the background task is suspended and the urgent delivery is responded to first.

[0085] The purpose of this step is to transform the original passive picking process that relies on static storage locations into a predictable, interruptible, and recoverable proactive rescheduling process, thereby enabling advance adaptation to order peaks and improving the accessibility and scheduling stability of high-frequency pallets without changing the existing hardware layout.

[0086] In a preferred embodiment of the present invention, the step of extracting pallet flow probability from order spatiotemporal coupling features and calculating predictive cache hit rate includes: performing probability distribution fitting on outbound time series in order spatiotemporal coupling features to obtain pallet flow probability of outbound time series under preset different time windows;

[0087] Within a preset prediction period, the first number of target pallets at a specific popularity level that were migrated in advance and hit by actual outbound commands, and the total number of actual outbound commands issued are counted as the second number; the ratio of the first number to the second number is calculated, and the ratio result is used as the predictive cache hit rate.

[0088] This embodiment provides a mechanism for extracting pallet flow probability and calculating predictive cache hit rate; specifically, in the aforementioned evening pharmaceutical area warehouse scenario, relying solely on the experience that high-frequency products should be placed in the front is prone to failure when promotions switch or the shuttle bus dispatch rhythm changes.

[0089] For example, a certain antipyretic drug has a high fever during the day, but the urgent refills at night are more concentrated in surgical consumables. If the time series of the outbound shipment is not subdivided and fitted, problems such as invalid migration actions and increased frequency of equipment running idle will occur. Therefore, this embodiment further introduces probability distribution fitting based on time window modeling, so that the system can determine when to issue the shipment rather than just whether it will be issued.

[0090] Specifically, the system extracts the outbound time series corresponding to each type of pallet from the spatiotemporal coupling features of the order. The outbound time series is a set of timestamps extracted from the spatiotemporal coupling features of the order, reflecting the occurrence of each actual outbound action of the same type of pallet during the historical observation period. This set is arranged in chronological order and mapped to the corresponding natural day coordinate axis to characterize the activity rhythm of this type of pallet over time.

[0091] An observation day can be divided into multiple consecutive time windows, for example, W1 is 18:00-18:15, W2 is 18:15-18:30, W3 is 18:30-18:45, and W4 is 18:45-19:00;

[0092] For a certain high-heat pallet category T1, the number of times it was shipped out in W1, W2, W3, and W4 in the past 20 days were 8, 15, 30, and 27, respectively. After fitting the probability distribution of these frequencies, the circulation probabilities of this category in the four windows are 0.10, 0.19, 0.38, and 0.33, respectively.

[0093] For another category T2, if the frequency is 18, 12, 6, 4, the transfer probability will be more biased towards the earlier window; in this way, different pallets belonging to the same high-heat layer can also be arranged in different migration order according to the time window; the hit situation is counted within the preset prediction period; here the prediction period can be set to 30 consecutive minutes; if the system performs 12 advance migrations for the target pallet in the high-heat layer within this period, and 7 of them are hit by the actual outbound command within the prediction window after the pallet reaches the target cache position, then the first count is 7;

[0094] If the system receives a total of 20 real outbound instructions within the same period, then the second count is 20, and the ratio of the two is 0.35, which is the predictive cache hit rate for that period. If you want to observe the effects of different heat levels separately, you can also count H1, H2, and H3 separately so that you can adjust the heat threshold later.

[0095] The hit rate here should be limited to: the pallet is migrated in advance and is called by the actual outbound instruction within the corresponding forecast period; if a pallet has been migrated in advance but has not been outbound due to order cancellation, or has been outbound after the forecast period has expired, it will not be counted in the first count; by limiting it in this way, the hit rate can be artificially increased by lengthening the observation window.

[0096] Specifically, if there is very little data in a certain time window, resulting in unstable fitting, such as only one sample in the past 20 days, the system can use adjacent window smoothing to combine W2 and W3 for estimation; if no effective samples appear for several consecutive periods, the system can temporarily revert to a coarse-grained probability model based on the overall frequency.

[0097] If the second count is 0, meaning there are no actual outbound orders in a certain period, the hit rate for that period will not be directly recorded as 0, but can be marked as an invalid statistical period to avoid misleading subsequent parameter adjustments;

[0098] In this pharmaceutical warehouse, the system found that the probability of bandage combo packs being transferred during the 18:30-19:00 window was 0.41, significantly higher than the 0.17 during the 18:00-18:30 window. Therefore, the transfer of these packs was scheduled during the idle period after 18:20. Within a 30-minute prediction cycle, the system transferred high-heat trays 9 times in advance, 6 of which were matched by actual orders.

[0099] There were 16 actual outbound orders during the same period, so the predictive cache hit rate was 6 / 16. If this ratio drops significantly in the next cycle, it means that the current time window model has failed to accurately reflect the rhythm of the new round of orders and needs to be refitted. The purpose of this step is to refine the coarse historical popularity judgment into the actual outbound tendency judgment in different time periods, so as to achieve more accurate advance migration and use the hit rate to quantify the effectiveness of the migration action in a closed loop.

[0100] In a preferred embodiment of the present invention, the step of calculating the physical topology entropy reduction index of each storage area by migration priority and pallet flow probability includes: calculating the variance of the distance between the current storage location of the pallet and the corresponding target outbound port in each storage area as the initial arrangement disorder, and obtaining the expected consecutive outbound order of the pallets in each storage area as the time-series associated outbound sequence.

[0101] The pallet flow probability is weighted according to the migration priority of each heat level to obtain a weighted value. The weighted value is then combined with the layer depth feature of the current storage location of the pallet to calculate the pallet retrieval complexity. The difference between the initial arrangement disorder and the retrieval complexity is calculated, and the difference is normalized using the length of the time-series outbound sequence to obtain the physical topology entropy reduction index of each storage area.

[0102] This embodiment provides a physical topological entropy reduction exponential calculation mechanism; specifically, in the previous embodiment, the system was able to identify which pallets might be out of the warehouse at what time, but if the migration is made directly based on this, local optima may still occur, that is, after moving a single hot pallet closer, other pallets that are about to be out of the warehouse in the same aisle will interfere with each other, resulting in an increase in the probability of overall path intersection conflict.

[0103] Therefore, this embodiment further measures whether a certain region is worth rearranging and uses the physical topological entropy reduction index to rank the region-level optimization effects;

[0104] Specifically, first calculate the initial disorder of each storage area; the distance from the current pallet location to the target exit in a certain area can be used as the basic quantity; taking area A as an example, assuming that the equivalent walking distances from the four pallets to be observed to O1 are 2 meters, 8 meters, 3 meters and 11 meters respectively, with a mean of 6 meters, then the distance variance can reflect the degree of distribution dispersion; the larger the variance, the higher the dispersion of the distribution of the same batch of potential outbound pallets and the higher the disorder of the arrangement;

[0105] If the distances between the four pallets in Zone B are 4 meters, 5 meters, 6 meters, and 5 meters respectively, the variance is significantly smaller, indicating that its layout is relatively regular; obtain the time-series consecutive outbound sequence; this sequence can be derived from order association features, representing a group of pallets that are likely to be outbound consecutively within a certain period of time; for example, if the system finds from historical data that P1, P3, and P5 are frequently outbound sequentially within the same bus wave, then a consecutive sequence of length 3 can be formed;

[0106] Another pair, P2 and P8, might only appear occasionally in pairs, so the sequence length is 2. This length reflects the degree of continuous optimization in the local layout. The longer the sequence, the more orderly the pallet arrangement of this group, and the better the subsequent outbound actions can be optimized. Then, the flow probability is weighted according to the migration priority of the heat level to obtain the picking complexity. For example, if P1 belongs to H1, has a priority of 3, and a flow probability of 0.42, its weighted value can be calculated as follows: The result is 1.26;

[0107] P2 belongs to H2, has a priority of 2, and a flow probability of 0.25, so the result is 0.50; P4 belongs to H3, has a priority of 1, and a flow probability of 0.08, so the result is 0.08; combined with their current layer depth, distance from the aisle entrance, and whether they need to be moved, the pallet picking complexity can be generated.

[0108] For ease of explanation, it can be simplified to the higher the weighting value and the deeper the location, the more complex the inherent theoretical obstacles to picking up goods, that is, the higher the picking complexity. It is necessary to clarify the closed-loop logic of the principle of subtracting the two: the initial disordered arrangement measures the disordered quantification index of the current physical spatial distribution of the pallet, while the picking complexity represents the benchmark value of the inherent flow resistance that is unavoidable for the normal flow of the target high-frequency pallet.

[0109] The difference between the two precisely isolates the costs that must be paid due to objective operational rules, and extracts the optimizable redundant disorder caused only by the current messy layout, which can be completely eliminated by rearrangement; in order to ensure that the subtraction of these two data with different measurement scales will not cause data distortion or miscalculation, the system automatically performs threshold truncation and extreme value standardization mapping on both of them based on the historical regional sample distribution before calculating the difference.

[0110] The difference between the initial arrangement disorder after alignment and the picking complexity is calculated, and normalized using the time-series associated outbound sequence length to obtain the physical topological entropy reduction index for each region; specifically, let the standardized initial arrangement disorder be... The standardized pickup complexity is The time-series outbound sequence length is , If the function is the natural logarithm, then the physical topological entropy reduction exponent is... The structured computation logic is as follows:

[0111]

[0112] By introducing a natural logarithm term to smooth and normalize the length, the optimization increment brought about by the increase in sequence length is reasonably reduced. For example, after the transformation, the initial disorder of area A is relatively rated as 10, and the necessary complexity of the comprehensive evaluation and retrieval of key pallets in the area is 4. The difference between the two is 6, which indicates that a high proportion of redundant disorder can be reduced by rearranging.

[0113] If the time-series outbound sequence length of area A is 3, then after substituting into the above logic normalization, a specific index evaluation can be obtained; the disorder of area B is 4, the picking complexity is 3, the difference is 1, and the sequence length is 1, so the index is 1; therefore, it can be judged that area A is more worthy of priority rearrangement; the index here does not require the use of a unique fixed formula, as long as it can stably express the degree to which the current layout can reduce disorder and improve continuous outbound after rearrangement;

[0114] Specifically, if there is no identifiable outbound sequence in a certain area, the sequence length can be treated as 1 to avoid a denominator of 0; if all the pallets to be observed in a certain area are almost the same distance, resulting in a variance close to 0, but some of the high-temperature pallets are actually buried deep and require complex relocation operations, then the layer depth weight in the retrieval complexity can be increased to prevent areas with small variance but difficult retrieval from being mistakenly ignored; if the difference is negative, it means that the current layout is already better or the rearrangement effect is less than the disturbance resource load, then this area can be downgraded or temporarily excluded from rearrangement in this round of optimization;

[0115] In this pharmaceutical regional warehouse, there are currently three high-frequency promotional sets in Zone A, which are scattered in different locations and are often shipped out consecutively before the last shift. The system calculates that the distance variance from Zone A to O1 is relatively high, and the associated outbound sequence length is 4. Although Zone C also has high-temperature pallets, they are not often shipped out consecutively, and the sequence length is only 1.

[0116] After weighting, the physical topological entropy reduction index of area A is significantly higher than that of area C. Therefore, the rearrangement strategy prioritizes area A. This avoids the scattered movement of equipment across multiple areas and instead focuses on resolving the area most likely to cause outbound congestion.

[0117] The purpose of this step is to shift from a single-pallet perspective to a regional layout perspective, using a quantifiable metric to identify the most valuable locations for rearrangement, thereby enabling more effective resource allocation.

[0118] In a preferred embodiment of the present invention, the step of extracting idle time slices of the handling equipment and calculating the fragment time rebalancing conversion rate includes: monitoring the operating cycle of the handling equipment and separating the idle time window without tasks waiting from the operating cycle as an idle time slice.

[0119] The cumulative duration of idle time slices used for tray migration operations and the total duration of the running cycle are calculated. The ratio of the cumulative duration to the total duration is used as the fragment time rebalancing conversion rate.

[0120] This embodiment provides a mechanism for calculating the conversion rate of fragmented time rebalancing. Specifically, in the previous embodiment, the system already knows which areas have higher rescheduling efficiency. However, if the actual idle status of the equipment is not considered, background rescheduling may compete with forward outbound tasks for the same stacker crane or the same aisle, which may lead to new waiting. Therefore, this embodiment further extracts the idle time slices that can be safely utilized from the equipment operation log and performs entropy reduction migration only in these gaps. In detail, the system establishes an operation cycle record for each handling equipment.

[0121] Taking a stacker crane as an example, an operating cycle can be defined as a continuous one-hour period, which includes at least the main task execution period, the standby period, the turning or returning to zero period, and the avoidance waiting period. Only the time window that does not undertake the main task and does not conflict with the scheduled task can be separated into a scheduling idle time slice.

[0122] To illustrate this more clearly, we can assume that the stacker crane experiences the following time slices between 18:00 and 19:00: 18:05-18:07 no task, 18:19-18:20 waiting for the next instruction, 18:32-18:35 idle after returning to zero, and 18:48-18:49 short pause after the obstacle avoidance operation ends; then the lengths of these time slices are 120 seconds, 60 seconds, 180 seconds, and 60 seconds respectively, with a total idle time of 420 seconds.

[0123] The total time actually used for tray entropy reduction migration in these idle time slices is calculated; if P1 is moved forward from 18:05 to 18:07, it takes 100 seconds; if P3 is moved sideways from 18:32 to 18:35, it takes 150 seconds.

[0124] The remaining time slices were not used due to roadway congestion, so the cumulative effective rebalancing time is 250 seconds. Comparing this with the total operating cycle time of 3600 seconds, the fragmented time rebalancing conversion rate is 250 / 3600. If it is necessary to reflect the utilization quality inside the equipment, the used idle time / total idle time can also be calculated at the same time, for example, 250 / 420, to help observe whether the system has fully scheduled the available idle time slices.

[0125] The reason for using the ratio of cumulative duration to total duration, rather than just comparing it with idle time, is to uniformly reflect how much controllable resources the background optimization actually occupies throughout the entire operation cycle. The higher this indicator is, the better the system is at converting fragmented and easily wasted time into effective pre-migration actions. However, if it is too high, it may also indicate that the background task scheduling frequency is too high, which needs to be weighed against equipment fatigue and sudden order response.

[0126] As supplementary handling logic for abnormal situations, if the length of the detected idle time slice is less than the minimum executable migration time, such as less than 10 seconds, the system can only cache it as a candidate segment and not immediately issue a migration instruction, but wait for adjacent short segments to be spliced ​​together to form an executable window; if the same time slice is nominally idle, but the path planning shows that it will potentially intersect with another device, then the time slice will not be included in the executable idle set.

[0127] If the total operating cycle is shortened due to downtime maintenance, the total duration in the ratio calculation should be the actual total operating duration, rather than the natural hourly duration, to avoid distortion of the indicator; in this pharmaceutical warehouse, the shuttle bus has multiple short pauses ranging from 20 to 90 seconds within one hour at night.

[0128] After screening, the system selects only segments that will not affect the urgent delivery channel to perform shallow adjustments; the total running time of the shuttle is 3300 seconds in one hour, of which the cumulative time used for pallet entropy reduction migration is 180 seconds, so the fragment time rebalancing conversion rate of the equipment is 180 / 3300; if the stacker crane has a significantly higher indicator in the same period, it means that the background reordering task can be prioritized for the stacker crane in subsequent optimization.

[0129] The purpose of this step is to clarify that background reordering should be based on utilizing available idle time rather than crowding out the main task, thereby achieving safe conversion and quantitative measurement of fragmented device resources.

[0130] In a preferred embodiment of the present invention, the step of constructing a multi-objective optimization function based on the physical topology entropy reduction index and the fragmented time rebalancing conversion rate includes: normalizing the predictive cache hit rate, the physical topology entropy reduction index, and the parameters corresponding to the optimization objectives, and then constructing a multi-objective initial function; wherein, the optimization objectives of the multi-objective initial function include maximizing flow smoothness, minimizing retrieval time, and minimizing fatigue loss of the mechanical architecture; adjusting the weight coefficients of the multi-objective initial function based on the fragmented time rebalancing conversion rate, and substituting the adjusted weight coefficients into the multi-objective initial function to obtain the multi-objective optimization function.

[0131] This embodiment provides a multi-objective optimization function construction mechanism. Specifically, under the aforementioned mechanism, the system can obtain the hit effect, the regional rearrangement evaluation score, and the idle resource utilization level respectively. However, if there is no unified objective function, the scheduler may be biased towards a single indicator, such as frequently moving due to optimizing the hit rate, or simply saving equipment actions and giving up the due forward movement benefits.

[0132] Therefore, this embodiment organizes these factors into a multi-objective optimization function, so that the reordering strategy achieves a balance between high fluency, high address hit rate, stable operation and low device wear.

[0133] Specifically, we can first construct a multi-objective initial function, whose core objectives include at least three directions: First, maximizing the smoothness of the flow, reflecting whether the pallet outbound path is shorter and whether continuous orders are easier to respond to quickly; second, minimizing the picking time, reflecting the average addressing and handling time required when performing actual outbound tasks; and third, minimizing the fatigue wear of the mechanical structure, reflecting the additional number of actions, turning times, and high load duration of stacker cranes, shuttles, and automated guided vehicles.

[0134] In specific quantitative calculations, the smoothness score is the normalized value of the sum of the expected reduction in the travel distance of all high-heat pallets to the outbound gate after the strategy is implemented; the picking time is obtained by simulating the sum of the average addressing time and the load-bearing travel time of the equipment in each aisle after the strategy is implemented; the fatigue loss score of the mechanical structure is calculated by comprehensively weighting and summing the number of equipment start-ups and shutdowns, the number of right-angle turns, and the load lifting height caused by high-position picking and placing.

[0135] To facilitate micro-level analysis, we can set the three evaluation metrics for a candidate strategy S1 as follows: an improvement of 8 points in flow smoothness, a reduction of 12 seconds in estimated picking time, and an increase of 3 points in equipment fatigue; another strategy S2 corresponds to an improvement of 6 points, a reduction of 9 seconds, and an increase of 1 point in fatigue. If only the first two items are considered, S1 is better; if equipment wear is sensitive, then S2 may be more reasonable.

[0136] Based on this, the predictive cache hit rate and physical topology entropy reduction exponent are introduced into the initial function; a high hit rate indicates that early migration is more likely to translate into improved actual order outbound efficiency.

[0137] A high entropy reduction index indicates a more orderly regional layout rearrangement. Therefore, the base score of a candidate strategy can be understood as the probability of success multiplied by the layout benefit, minus the expected retrieval time and equipment fatigue resource load. In practice, weighted summation, hierarchical scoring, or constrained comprehensive evaluation can be used, as long as a unified comparison of each candidate strategy can be made.

[0138] It is important to note that predictive cache hit rate is limited by... The probability parameters of the interval, while the scores of core objectives such as smooth flow and data such as expected picking time consumption and mechanical operation wear and tear have numerical differences across orders of magnitude in absolute value and basic units;

[0139] If we directly apply absolute difference and other forms to the addition and subtraction formula, it will inevitably result in some small data indicators being covered by large absolute indicators such as time in seconds, which will lead to numerical deviations in the multi-objective solution.

[0140] To eliminate the computational bias caused by this difference in numerical magnitude, the system additionally sets up a boundary scaling logic. Before performing weight adjustment and merging on the final comprehensive conversion function of each parameter input, it forces a nonlinear range normalization transformation on each heterogeneous target system, mapping them to the basic reference scale. This ensures the numerical stability of the fusion of multiple target indicators;

[0141] To clearly define the calculation rules, let the flow smoothness after range normalization be denoted as . The pickup time is Fatigue loss is Predictive cache hit rate The physical topological entropy reduction exponent is The dynamic weight of the smoothness of circulation is The dynamic weight of pickup time is The dynamic weight of fatigue loss is The system constructs a multi-objective optimization function. It can be represented as:

[0142]

[0143] The system adjusts the evaluation weights of each aligned metric based on the fragmented time rebalancing conversion rate. This is because whether background rebalancing is worth actively pursuing is directly related to the current system's ability to safely handle these actions during idle periods. For example, a high fragmented time rebalancing conversion rate indicates sufficient available idle time on the devices, allowing the system to appropriately increase the dynamic weight of workflow smoothness. ;

[0144] When the conversion rate is low, it indicates a shortage of available resources, and the dynamic weight of pickup time should be increased. Dynamic weights of fatigue loss To suppress aggressive migration; for ease of explanation, we can assume that the conversion rate for a certain period is 0.12, then the weights are set as follows: smoothness of circulation 0.4, picking time 0.35, and fatigue loss 0.25; if the conversion rate drops to 0.04 in the next period, the weights are adjusted to 0.3, 0.3, and 0.4.

[0145] In this way, the same rearrangement action will receive different evaluations under different busy and idle states of the equipment. As a supplementary adjustment mechanism for system operation, if the hit rate is low for a long time but the entropy reduction index is high, it means that although the system has made the layout more orderly, it has failed to translate into an improvement in the actual order outbound task. At this time, the maximum score of this type of strategy can be limited to avoid executing redundant rearrangement strategies that do not significantly improve the actual outbound efficiency.

[0146] If the rebalancing conversion rate of fragmented time suddenly increases abnormally, it is necessary to check simultaneously whether it is due to an abnormally low main task volume or missing equipment statistics, so as to avoid amplifying the background migration intensity simply because of data abnormality; if a certain evaluation quantity in the multi-objective function cannot be obtained, such as the fatigue value is difficult to accurately estimate in a short period of time, the rolling average of the recent periods can be used as a substitute and marked as a conservative estimate in this round.

[0147] In this pharmaceutical warehouse, there are two candidate rearrangement strategies around 19:00 in the evening; the first option is to move the high-heat pallets in areas A and C forward at the same time, which is expected to shorten the average picking time of the next ten orders by 15 seconds, but will require the stacker crane to make an additional 20 turns.

[0148] The second option only moves the outbound pallets in area A forward, shortening the time by 10 seconds, but requiring 8 additional turns. Since the system detects that the fragmented time rebalancing conversion rate is only 0.05, indicating that the equipment has limited idle time, the optimization function increases the fatigue loss weight, ultimately favoring the second option.

[0149] The purpose of this mechanism is to unify multiple seemingly conflicting engineering indicators into a comparable framework, thereby achieving a dynamic balance between outbound efficiency and equipment lifespan.

[0150] In a preferred embodiment of the present invention, the step of solving a multi-objective optimization function by an optimization algorithm to obtain a dynamic topology rearrangement strategy includes: initializing the algorithm parameters of the optimization algorithm, including initial temperature, cooling coefficient and maximum number of iterations; generating an initial rearrangement strategy as the current rearrangement strategy and starting iteration; in each iteration, performing the following steps: generating candidate rearrangement strategies by adjusting the pallet target storage location in the current rearrangement strategy;

[0151] The candidate rearrangement strategy and the current rearrangement strategy are respectively used as inputs to the multi-objective optimization function for calculation. The function value of the candidate rearrangement strategy is subtracted from the function value of the current rearrangement strategy to obtain the function difference. Based on the function difference and the temperature of the current iteration, the probability of accepting the candidate rearrangement strategy is calculated.

[0152] The system determines whether to adopt a candidate rearrangement strategy based on probability and updates the current rearrangement strategy accordingly. The temperature for the next iteration is updated based on the cooling coefficient. When the number of iterations reaches the maximum number of iterations, the iteration stops, and the current rearrangement strategy is output as the dynamic topology rearrangement strategy.

[0153] This embodiment provides a dynamic topology rearrangement strategy solution mechanism. Specifically, in the previous embodiment, the multi-objective optimization function has given the evaluation rules. However, the actual number of pallets in the warehouse is large, and the number of target storage location combinations is huge. If a simple exhaustive search method is adopted, the computational load will increase rapidly with the pallet size, making it difficult to complete within the real-time scheduling cycle. Therefore, this embodiment uses an iterative optimization algorithm to search for a better rearrangement strategy within a limited time. For ease of explanation, the following description uses a solution method with initial temperature, cooling coefficient, and maximum number of iterations.

[0154] Specifically, the algorithm parameters are initialized; for example, the initial temperature can be set to 100, the cooling coefficient to 0.85, and the maximum number of iterations to 50. Here, the initial temperature is used to control the possibility of accepting poor solutions in the early search, so as to avoid the search getting trapped in local optima too early; the cooling coefficient is used to reduce the exploration intensity round by round; and the maximum number of iterations limits the computational cost; the initial rearrangement strategy is generated.

[0155] The initial strategy can be to not migrate, or to use a heuristic approach that moves high-heat trays to the nearest available shallow positions in each area according to their popularity from high to low. For example, in the initial strategy S0, P1 is moved to A-02-02, P3 is moved to C-03-01, while P2 remains unchanged.

[0156] In each iteration, candidate rearrangement strategies are generated by fine-tuning the current strategy; for example, in the first round, the target location of P1 is changed from A-02-02 to A-01-03, or a new side movement action of P2 is added to form candidate strategy S1.

[0157] The adjustment actions when the system generates candidate rearrangement strategies specifically include the following three fine-tuning operators: The first is single-point random mutation, which randomly selects one pallet in the current strategy and changes its target storage location to another available empty storage location in the same area; the second is two-point neighborhood exchange, which randomly selects two pallets of the same size in the current strategy and swaps their target storage locations.

[0158] The third type is strategy addition and deletion, which means that without violating the constraint of the maximum handling capacity of the equipment, the migration instructions for a certain edge hot tray are randomly added or removed; in each iteration, the system randomly calls one of the above fine-tuning operators according to the preset probability distribution to generate candidate rearrangement strategies.

[0159] Input S0 and S1 into the aforementioned multi-objective optimization function to obtain evaluation values; assuming S0 scores 72 and S1 scores 75, the function difference is +3, indicating that the candidate strategy is better and can be directly accepted to update the current strategy; if in a certain round the candidate strategy S2 scores 70, which is lower than the current strategy 75, the function difference is -5.

[0160] At this point, rejection is not always necessary; rather, the probability of acceptance is calculated based on the current temperature. When the temperature is high in the early stages, even if the candidate strategy is slightly inferior, a certain probability of acceptance can still be retained, thus escaping the local optimum. As the temperature decreases, the system gradually moves towards stable convergence. Furthermore, the probability acceptance process can be explained using specific calculation rules.

[0161] Let the difference of the functions be _____. That is, the candidate policy function value minus the current policy function value, and the current iteration temperature is... ;when When, the probability of accepting a worse candidate strategy. The calculation rules are as follows:

[0162]

[0163] Assuming the current temperature is 50 and the function difference is -5, the candidate strategy may still be... It has a certain probability of being accepted; if the temperature drops to 5 later, facing the same -5 difference, its probability of being accepted is... It will significantly reduce;

[0164] To prevent extreme mutations from causing floating-point overflows on the algorithm system side due to degenerate computation, the system incorporates numerical overflow protection verification for exponential boundaries: once the system detects a severely deviated inferior solution generated by a single iteration, for example... This causes the calculated penalty amount of the bad state function to exceed the preset tolerance lower limit. Even if it is in the early stage of high-temperature iteration with a high probability of accepting a poor solution, it will forcibly intervene and clear and truncate the adoption probability for this extreme deviation.

[0165] Only changes in the difference of regular functions remaining within the safe calculation space are properly input into the random number generation and verification mechanism, and the system then determines the random number based on the generated result. Uniformly distributed random numbers and the probability of secure reception Perform a numerical comparison; if the random number is less than or equal to Finally, accept the candidate strategy change after the iteration; otherwise, reject it.

[0166] Then, the temperature is updated according to the cooling coefficient for the next round, for example, from 100 to 85, then to 72.25, until the maximum number of iterations is reached and then it stops. The current optimal or stable rearrangement strategy is output as the dynamic topology rearrangement strategy. In warehousing applications, the generation of candidate strategies should not arbitrarily violate the safety boundary. Therefore, basic constraints should be maintained each time the target storage location is adjusted. For example, two pallets cannot be pointed to the same storage location.

[0167] Oversized pallets cannot be moved into mismatched locations; conflicting paths that simultaneously occupy the same aisle cannot be scheduled; in other words, the optimization algorithm searches for a set of feasible solutions that satisfy engineering constraints, rather than theoretically arbitrary exchanges; specifically, if no feasible candidate strategy is found after initialization, for example, all shallow locations are full and high-priority pallets cannot be replaced, the current strategy is to maintain the status quo and wait for the next scheduling cycle to solve the problem again.

[0168] If the function value does not improve for several consecutive rounds during the iteration process, it can be stopped early to reduce invalid calculations; if a candidate strategy has a high score, but the path planning verification shows that it requires high-conflict collaboration across devices, it should be eliminated before entering the acceptance judgment to avoid misleading the scheduling with high-scoring but low-feasibility strategies.

[0169] In this pharmaceutical regional warehouse, the system calculates the order peak around 19:10 in the evening. The initial strategy is to move only P1 and P3 forward. In the third iteration, the candidate solution attempts to move P5 out of the deep zone as well. Although this can improve the local hit rate, it will occupy the B zone turning channel, resulting in a significant increase in fatigue loss and a decrease in the overall score.

[0170] In the 7th iteration, the system moved P1 to a storage location closer to O1 without affecting the passage of P6, which improved the overall score, so the current strategy was updated; after the 30th iteration, there was no significant improvement for 5 consecutive iterations, so it converged early and output the final dynamic topology rearrangement strategy for scheduling execution.

[0171] The purpose of this step is to quickly search for an executable rearrangement scheme that balances operational efficiency and resource consumption in complex storage location combinations, thereby achieving a feasible dynamic layout optimization.

[0172] In a preferred embodiment of the present invention, when a sudden task is detected, the priority of the sudden task is obtained. When the priority of the sudden task is higher than the migration priority corresponding to the currently cached execution instruction, the steps of calculating the scheduling preemption rollback delay and formulating a reordering interrupt recovery strategy include:

[0173] In response to the detection of a burst task with a higher priority than the currently cached execution instruction, the interruption point of the currently cached execution instruction is recorded, and the time difference from pausing the current cached execution instruction to the complete release of the currently occupied path by the transport equipment and the start of execution of the burst task is calculated. The time difference is used as the scheduling preemption rollback delay. The relationship between the scheduling preemption rollback delay and the preset maximum tolerable delay threshold is determined. The maximum tolerable delay threshold is determined based on the upper limit of the physical time taken for the transport equipment to exit the currently occupied path and perform the avoidance action.

[0174] When the scheduling preemption rollback delay is less than or equal to the maximum tolerable delay threshold, a first reordering interrupt recovery strategy is formulated. The first reordering interrupt recovery strategy includes suspending the currently cached execution instructions, switching to execute a burst task, and resuming the migration operation from the interrupt breakpoint after the burst task ends.

[0175] When the scheduling preemption rollback delay exceeds the maximum tolerable delay threshold, a second reordering interrupt recovery strategy is formulated. The second reordering interrupt recovery strategy includes rejecting the preemption request of the burst task and continuing to execute the current cached execution instruction until the tray corresponding to the current cached execution instruction is in place before responding to the burst task.

[0176] This embodiment provides a mechanism for interruption recovery from sudden task preemption and rescheduling; specifically, in the previous embodiment, the system can already perform background rescheduling during idle periods, but in real logistics sites, sudden tasks such as urgent delivery, replenishment correction or abnormal recall may be inserted at any time.

[0177] If all background tasks give way, the equipment may frequently interrupt and repeatedly roll back, resulting in lane occupation and wasted actions; if all background tasks do not give way, it will affect the response of high-priority orders. Therefore, this embodiment introduces scheduling preemption rollback delay and hierarchical recovery strategies to ensure that the interruption process is controllable.

[0178] Specifically, when the system detects the arrival of a new sudden task, it obtains its priority and compares it with the priority of the currently cached background execution instructions;

[0179] When an emergency task is detected, the system parses the data message of the emergency task, extracts the business type tag and time requirement carried in it, and obtains the absolute priority of the emergency task by querying the global priority mapping table issued by the warehouse management system; for example, an emergency task carrying a medical emergency consumables allocation tag will be directly assigned the highest priority weight.

[0180] If the priority of the sudden task is not higher than that of the current task, the existing execution sequence is maintained and no preemption is required; if it is higher, the breakpoint of the currently cached execution instruction is recorded immediately.

[0181] The breakpoint should include at least: the number of the pallet currently being moved, the current location of the equipment, the completed path segment, the unfinished target storage location, and the current aisle occupancy status; in this way, when the background task is resumed in the future, it does not need to be replanned from scratch, but can continue execution from the safe breakpoint.

[0182] Calculate the scheduling preemption and rollback delay; this delay is not simply the response time after receiving a new task, but the time difference between the start of pausing the background migration and the time between the transport equipment completely releasing the physical tunnel and actually starting to process the sudden task.

[0183] For example, a stacker crane is moving P1 from A-05-07 to A-02-02. When it is halfway through the process, it receives an urgent hospital order E2. The system finds that the stacker crane needs to retract the forks, temporarily place the pallet in the nearest safe position, exit the cross passage, and then switch to the execution path of E2. The above process takes a total of 6 seconds, so the scheduling preemption rollback delay is 6 seconds.

[0184] The delay is compared with the preset maximum tolerable delay threshold. If the delay is less than or equal to the threshold, for example, if the threshold is set to 8 seconds and the current delay is only 6 seconds, the first reordering interrupt recovery strategy is adopted, that is, the current background task is suspended and the sudden task is immediately switched to be executed. After the sudden task is completed, the original migration is restored according to the interrupt breakpoint. During the recovery, it can be verified whether the original target storage location is still valid. If it is still valid, it continues.

[0185] If the tray is already occupied by other tasks, the nearest available replacement slot will be reassigned to the tray. If the scheduling preemption rollback delay is greater than the threshold, for example, it is expected to take 12 seconds to roll back safely, and the target tray of the sudden task is located in another aisle behind the current device, forcibly preempting will cause a longer total delay. In this case, the second reordering interrupt recovery strategy will be adopted, allowing the background task to send the current tray to the destination first, and then responding to the sudden task.

[0186] The key to this distinction is that true scheduling efficiency does not depend on whether the response is immediate, but on whether the switching is completed within a controllable time without creating further congestion. Especially in the case of a three-dimensional warehouse with narrow aisles and equipment that cannot avoid each other in place, incomplete interruption operations are more likely to cause deadlock or collision risks than complete migration. Therefore, setting a delay threshold can establish interruption control on physical executability.

[0187] Specifically, if the system finds that the current pallet has been lifted from its original location but has not yet reached any safe position when the interruption point is recorded, the system should first place the pallet in a temporary safe position, and then determine whether to switch. It is not allowed to hover in an unsafe posture for more than the preset safe dwell time.

[0188] If a sudden task has a high priority, but the target roadway is physically isolated from the current device and another device is actually needed to execute it, the current device does not need to preempt the task and simply forwards it to the matching device. If the sudden task is canceled during the waiting period, the suspended background task will resume from the original breakpoint. If the original breakpoint information is invalid during the resumption, such as when the position sensor is reset, the system will use the most recent safety confirmation bit as the starting point for the resumption to avoid blindly continuing from an untrusted state.

[0189] In the pharmaceutical warehouse, at 19:18, the stacker crane was executing the background forward movement operation of P3. At this time, the hospital's urgent delivery order E2 arrived, with a higher priority than the background task. The system recorded the current breakpoint as P3 having left C-06-08, the equipment being located in the middle of the C lane, and not yet placed in C-03-01.

[0190] Calculations showed that it would take 5 seconds from pausing to releasing the roadway and turning to E2, which is less than the 8-second threshold. Therefore, the system suspended the background task and completed E2 first. After the emergency matching was completed at 19:22, the system checked that C-03-01 was still empty, so it resumed the P3 migration from the breakpoint. At another time, if the equipment was close to the target position and it would take 11 seconds to back up, the system would choose to put P3 in place first and then perform the emergency matching to avoid the roadway being occupied beyond the preset safe dwell time.

[0191] The purpose of this mechanism is to provide a measurable and recoverable interrupt handling path when there is a conflict between sudden tasks and background rescheduling, thereby achieving a balance between high-priority response and system stability.

[0192] Please see Figure 2 A system for efficient pallet storage in logistics storage is applied to a logistics storage system that includes handling equipment, outbound ports and multiple storage areas. The system includes: a data acquisition and processing module, which is used to collect actual outbound instructions and pallet flow records in the logistics storage system as initial data packets. The flow records contain the historical storage location coordinates of the pallets. The module performs time-series correlation analysis on the initial data packets to obtain the spatiotemporal coupling characteristics of orders. The module extracts the pallet flow probability from the spatiotemporal coupling characteristics of orders and calculates the predictive cache hit rate.

[0193] The heat assessment calculation module is used to divide the pallet into multiple heat levels using a heat assessment algorithm based on historical outbound frequency. According to the order of heat from high to low, each heat level is assigned a decreasing migration priority. The physical topology entropy reduction index of each storage area is calculated by using the migration priority and the pallet flow probability.

[0194] The rearrangement optimization solution module is used to extract the idle time slices of the handling equipment and calculate the fragment time rebalancing conversion rate. Based on the physical topology entropy reduction index and the fragment time rebalancing conversion rate, a multi-objective optimization function is constructed, and the multi-objective optimization function is solved by the optimization algorithm to obtain the dynamic topology rearrangement strategy.

[0195] The instruction generation module is used to combine migration priority and dynamic topology reordering strategy to generate migration trigger conditions and corresponding cache execution instructions.

[0196] The scheduling execution monitoring module is used to collect the real-time scheduling status of the handling equipment. When the real-time scheduling status meets the migration trigger condition, the corresponding cached execution instruction is executed. At the same time, the real-time scheduling status is monitored for sudden tasks. When a sudden task is detected, the priority of the sudden task is obtained. When the priority of the sudden task is higher than the migration priority corresponding to the current cached execution instruction, the scheduling preemption rollback delay is calculated and a reordering interrupt recovery strategy is formulated.

[0197] This embodiment provides an efficient pallet storage system for logistics storage to implement the aforementioned method; specifically, the system can be deployed on the warehouse control server of the aforementioned pharmaceutical e-commerce regional warehouse and communicate with the warehouse control system, equipment programmable logic controller, aisle sensors and handheld terminals.

[0198] The entire system forms a closed loop around data acquisition, strategy evaluation, rearrangement solution, instruction generation, and scheduling monitoring. It does not require any new special hardware and only needs to be modularly deployed on the basis of existing warehouse automation. Specifically, the data acquisition and processing module is used to receive the actual outbound instructions issued by the order center and simultaneously read the pallet flow records.

[0199] This module can extract order time, minimum inventory unit, pallet number and outbound information from the warehouse control system, and extract historical storage location coordinates, path trajectory and action completion time from the warehouse control system or equipment logs; internally, the module aligns these data according to a unified timeline, and then performs time-series correlation analysis to generate order spatiotemporal coupling characteristics;

[0200] For example, this module can identify the correlation between P1 and P3 being frequently requested after 18:30 and the average outbound response time shortening after moving to a shallower position, and output the corresponding flow probability and hit rate statistics; the heat assessment calculation module receives the aforementioned output results, divides the pallets into heat levels and assigns migration priorities.

[0201] This module can not only be stratified according to historical frequency, but also dynamically corrected by combining recent growth trends. Then, the module calculates the physical topological entropy reduction index according to the region dimension and outputs the ranking results of which regions have a higher rearrangement effect. For example, the module may determine that region A has the highest index, region C is second, and region B has the lowest index, providing region-level input for subsequent solutions.

[0202] The reordering optimization solution module is responsible for reading equipment operation logs, identifying idle time slices, calculating fragmented time rebalancing conversion rates, and constructing a multi-objective optimization function by combining factors such as hit rate and entropy reduction index. The module obtains dynamic topology reordering strategies through iteration. The solution results can be refined into a mapping table of target pallet - target storage location - expected execution window - required equipment. For example, the output is that from 18:35 to 18:38, the stacker crane will perform P1 forward movement, and from 18:40 to 18:42, the shuttle car will perform P3 lateral movement.

[0203] The instruction generation module is used to convert the solution results into cached execution instructions that can be executed by the device. This module can generate trigger conditions, such as the device being idle, the lane conflict flag being empty, and the probability of flow in the next 15 minutes being greater than a threshold. It can also generate specific action sequences, such as picking up goods—lifting—moving—placing—returning to zero. If there are multiple feasible target positions for the same pallet, the module can include a primary position and a backup position for quick switching during the scheduling phase.

[0204] The scheduling execution monitoring module monitors equipment status, roadway occupancy status, and sudden task inputs in real time. When the triggering conditions are met, the module sends the cached execution instructions to the corresponding equipment. During execution, if a higher-priority sudden task is detected, a preemption evaluation is initiated, the scheduling preemption rollback delay is calculated, and an appropriate interruption recovery strategy is selected. The module can also write the interruption point back to the database to support execution recovery and post-event traceability.

[0205] Specifically, if communication between modules is temporarily interrupted, each module should retain the most recent stable output as a read-only snapshot; for example, if the solution module cannot obtain the latest device status, it will not generate a new rearrangement strategy, but will keep the previous strategy frozen; if the instruction generation module finds that the downstream device interface returns an abnormality, the instruction status will be marked as unsuccessfully issued and returned to the queue, and will not be directly regarded as execution failure; if the scheduling execution monitoring module finds that the priority definitions from different sources are inconsistent, it can first normalize them according to the system's built-in priority mapping table and then compare them to avoid confusion in the logic of sudden task preemption.

[0206] In this pharmaceutical regional warehouse, the data acquisition and processing module received a batch of orders and pallet trajectories before the evening peak at 18:50; the heat assessment and calculation module identified P1, P3, and P5 as high-heat layers, and the physical topological entropy reduction index of areas A and C as high; the rearrangement optimization solution module formed a set of rearrangement strategies based on the idle time slices of the stacker crane and shuttle at that time; the instruction generation module converted them into two cached execution instructions;

[0207] The scheduling execution monitoring module executes the first instruction after confirming that the lane is empty at 18:56, and decides whether to preempt an urgent order when it encounters one at 19:02 based on the rollback delay threshold. The purpose of this system is to connect prediction, evaluation, solution and execution into a complete engineering link in a modular way, so as to achieve stable deployment of the aforementioned method in the actual logistics equipment environment.

[0208] An apparatus includes: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, wherein the steps of a method for efficient pallet storage in logistics storage are implemented when the computer-executable instructions are executed by the processor.

[0209] This embodiment provides an apparatus for performing an efficient pallet storage method; specifically, the apparatus may be an industrial server deployed in a warehouse control center or an edge computing controller integrated with a warehouse control system; the apparatus includes a memory and a processor, the memory storing computer-executable instructions for implementing the aforementioned steps, historical data cache, device status table, and interrupt recovery records;

[0210] After the processor invokes these instructions, it can perform operations such as order spatiotemporal correlation analysis, flow probability calculation, heat leveling, optimization solution, instruction generation, and scheduling monitoring;

[0211] Specifically, the memory can use non-volatile storage media to store basic model parameters, such as heat level threshold, flow probability time window division rules, rearrangement effect weight, maximum tolerance latency threshold, etc.; at the same time, it can also use volatile storage areas to store real-time data in the current cycle, such as equipment location, roadway occupancy marker, pending instruction queue and interrupt breakpoint.

[0212] When executing instructions, the processor can first read order records and pallet flow records from memory to construct an initial data packet; then send it to the corresponding algorithm module to output a dynamic topology reordering strategy; and send the generated cache execution instructions to the downstream control interface.

[0213] To facilitate understanding, a micro-execution process can be constructed: The processor reads 16 outbound records and 12 pallet migration records from the last 30 minutes, calculates the flow probability of P1 as 0.42 and P3 as 0.38; it identifies that the rearrangement effect of areas A and C is relatively high, and after detecting that there is a 90-second idle window for the stacker crane, it obtains a set of forward movement strategies;

[0214] The processor then sends a forward P1 instruction to the device control bus, and simultaneously writes the priority, target bit, and recoverable breakpoint field of the instruction into memory; if an urgent matching task is inserted during execution, the processor calculates the rollback delay based on the threshold in memory and the real-time status, and chooses to suspend or delay the response.

[0215] The key to implementing this device is that the aforementioned methods and steps do not rely on manual intervention one by one, but can be executed automatically by the processor; as long as the corresponding instructions and parameters are preset in the memory, the processor can run the above process periodically or in an event-triggered manner; the device can also be connected to a database, programmable logic controller, MES or upper-level scheduling platform to adapt to different warehouse control architectures.

[0216] Specifically, if the historical records in memory are missing, the processor can prioritize executing a simplified mode based on recent data and restore the complete strategy after the data is completed. If the processor encounters a device interface timeout during execution, it will mark the corresponding instruction as pending retry or transfer it to manual confirmation, instead of directly overwriting the original interrupt breakpoint. If the device restarts, the non-volatile areas in memory can still retain the previous stable strategy and snapshots of incomplete tasks, which the processor can use to restore the scheduling context after restarting.

[0217] In this pharmaceutical regional warehouse, industrial servers are deployed as the device in the control room; after 18:45 in the evening, the processor periodically runs the tray rearrangement program, reads the daily order flow and equipment logs from the memory, and generates a forward strategy for the 19:00 peak.

[0218] When an urgent hospital order is encountered at 19:08, the processor quickly reads the interrupt breakpoint of the currently cached instruction from memory, calculates the rollback delay to be only 5 seconds, suspends the background task and prioritizes the urgent order; after the urgent order is completed, the original rearrangement action is restored according to the breakpoint information in memory; the purpose of this device is to provide executable hardware and instruction carrier entities for the aforementioned method, thereby realizing the programmatic implementation of the efficient pallet storage strategy in the actual logistics control environment.

[0219] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A method for efficient pallet storage in logistics storage, applied to a logistics storage system including handling equipment, an outbound port, and multiple storage areas, characterized in that, include: The initial data packet is obtained by collecting actual outbound instructions and pallet transfer records in the logistics storage system. The transfer records contain the historical storage location coordinates of the pallets. The initial data packet is subjected to time series correlation analysis to obtain the spatiotemporal coupling characteristics of the orders. The pallet transfer probability is extracted from the spatiotemporal coupling characteristics of the orders and the predictive cache hit rate is calculated. The pallet is divided into multiple heat levels using a heat assessment algorithm based on historical outbound frequency. Each heat level is assigned a decreasing migration priority in descending order of heat. The physical topology entropy reduction index of each storage area is calculated by using the migration priority and the pallet flow probability. Idle time slices of handling equipment are extracted and the fragment time rebalancing conversion rate is calculated. A multi-objective optimization function is constructed based on the physical topology entropy reduction index and the fragment time rebalancing conversion rate. The multi-objective optimization function is solved by an optimization algorithm to obtain a dynamic topology rearrangement strategy. By combining migration priority and dynamic topology reordering strategy, migration trigger conditions and corresponding cache execution instructions are generated; The system collects the real-time scheduling status of the transport equipment. When the real-time scheduling status meets the migration triggering conditions, it executes the corresponding cached execution instructions. At the same time, it monitors the real-time scheduling status for sudden tasks. When a sudden task is detected, it obtains the priority of the sudden task. When the priority of the sudden task is higher than the migration priority corresponding to the current cached execution instruction, it calculates the scheduling preemption rollback delay and formulates a reordering interrupt recovery strategy.

2. The method for efficient pallet storage in logistics storage according to claim 1, characterized in that, The steps for extracting pallet flow probability and calculating predictive cache hit rate from order spatiotemporal coupling characteristics include: By fitting the probability distribution of the outbound time series in the spatiotemporal coupling characteristics of orders, the pallet circulation probability of the outbound time series under different preset time windows is obtained. Within the preset prediction period, the first number of target pallets at a specific heat level that were moved in advance and subsequently hit by actual outbound instructions, and the total number of actual outbound instructions issued, are counted as the second number. Calculate the ratio of the first count to the second count, and use the ratio as the predictive cache hit rate.

3. The method for efficient pallet storage in logistics storage according to claim 2, characterized in that, The steps for calculating the physical topology entropy reduction index of each storage region based on migration priority and tray flow probability include: The variance of the distance between the current storage location of the pallet and the corresponding target outbound port in each storage area is calculated as the initial arrangement disorder, and the expected consecutive outbound order of the pallets in each storage area is obtained as the time-series associated outbound sequence. The pallet turnover probability is weighted according to the migration priority of each heat level to obtain a weighted value. The weighted value is then combined with the depth characteristics of the current storage location of the pallet to calculate the pallet picking complexity. The difference between the initial arrangement disorder and the picking complexity is calculated, and the difference is normalized using the length of the time-series outbound sequence to obtain the physical topological entropy reduction index of each storage area.

4. The method for efficient pallet storage in logistics storage according to claim 3, characterized in that, The steps for extracting idle time slices from the handling equipment and calculating the fragment time rebalancing conversion rate include: Monitor the operation cycle of the handling equipment and separate the idle time window without tasks from the operation cycle as idle time slices; The statistics include the cumulative duration of idle time slices used to perform tray migration operations, and the total duration of the running cycle; Calculate the ratio of cumulative duration to total duration, and use the ratio as the fragmented time rebalancing conversion rate.

5. The method for efficient pallet storage in logistics storage according to claim 4, characterized in that, The steps for constructing a multi-objective optimization function based on the physical topological entropy reduction exponent and the fragmented time rebalancing conversion rate include: After normalizing the parameters corresponding to the predictive cache hit rate, physical topology entropy reduction exponent, and optimization objectives, a multi-objective initial function is constructed. The optimization objectives of the multi-objective initial function include maximizing the smoothness of the flow, minimizing the picking time consumption, and minimizing the fatigue wear of the handling equipment. The weight coefficients of the multi-objective initial function are adjusted based on the fragmented time rebalancing conversion rate. The adjusted weight coefficients are then substituted into the multi-objective initial function to obtain the multi-objective optimization function.

6. The method for efficient pallet storage in logistics storage according to claim 5, characterized in that, The steps to obtain the dynamic topology reshuffling strategy by solving the multi-objective optimization function using optimization algorithms include: Initialize the algorithm parameters of the optimization algorithm, including the initial temperature, cooling coefficient, and maximum number of iterations; Generate an initial rearrangement strategy as the current rearrangement strategy, and begin iteration; In each iteration, perform the following steps: Candidate rearrangement strategies are generated by adjusting the pallet target storage location in the current rearrangement strategy. The candidate rearrangement strategy and the current rearrangement strategy are respectively used as inputs to the multi-objective optimization function for calculation, and the function value of the candidate rearrangement strategy and the function value of the current rearrangement strategy are subtracted to obtain the function difference value; Calculate the probability of accepting the candidate rearrangement strategy based on the function difference and the temperature of the current iteration; The decision to adopt a candidate rearrangement strategy is based on probability, and the current rearrangement strategy is updated accordingly. The temperature for the next iteration is updated based on the cooling coefficient. When the number of iterations reaches the maximum number of iterations, the iteration stops, and the current rearrangement strategy is output as the dynamic topology rearrangement strategy.

7. The method for efficient pallet storage in logistics storage according to claim 6, characterized in that, When a sudden task is detected, its priority is obtained. If the priority of the sudden task is higher than the migration priority of the currently cached execution instruction, the steps for calculating the scheduling preemption rollback delay and formulating a reordering interrupt recovery strategy include: In response to the detection of a burst task with a higher priority than the currently cached execution instruction, the interruption point of the currently cached execution instruction is recorded, and the time difference from pausing the current cached execution instruction to the complete release of the currently occupied path by the transport equipment and the start of execution of the burst task is calculated. The time difference is used as the scheduling preemption rollback delay. Determine the relationship between the scheduling preemption rollback delay and the preset maximum tolerable delay threshold. The maximum tolerable delay threshold is determined based on the upper limit of the physical time taken for the handling equipment to exit the currently occupied path and perform avoidance actions. When the scheduling preemption rollback delay is less than or equal to the maximum tolerable delay threshold, a first reordering interrupt recovery strategy is formulated. The first reordering interrupt recovery strategy includes suspending the currently cached execution instructions, switching to execute a burst task, and resuming the migration operation from the interrupt breakpoint after the burst task ends. When the scheduling preemption rollback delay exceeds the maximum tolerable delay threshold, a second reordering interrupt recovery strategy is formulated. The second reordering interrupt recovery strategy includes rejecting the preemption request of the burst task and continuing to execute the current cached execution instruction until the tray corresponding to the current cached execution instruction is in place before responding to the burst task.

8. A system for efficient pallet storage in logistics storage, applied to a logistics storage system including handling equipment, an outbound port, and multiple storage areas, characterized in that, include: The data acquisition and processing module is used to collect actual outbound instructions and pallet flow records in the logistics storage system as initial data packets. The flow records contain the historical storage location coordinates of the pallets. The module performs time-series correlation analysis on the initial data packets to obtain the spatiotemporal coupling characteristics of orders. The pallet flow probability is extracted from the spatiotemporal coupling characteristics of orders, and the predictive cache hit rate is calculated. The heat assessment calculation module is used to divide the pallet into multiple heat levels using a heat assessment algorithm based on historical outbound frequency. According to the order of heat from high to low, each heat level is assigned a decreasing migration priority. The physical topology entropy reduction index of each storage area is calculated by using the migration priority and the pallet flow probability. The rearrangement optimization solution module is used to extract the idle time slices of the handling equipment and calculate the fragment time rebalancing conversion rate. Based on the physical topology entropy reduction index and the fragment time rebalancing conversion rate, a multi-objective optimization function is constructed, and the multi-objective optimization function is solved by the optimization algorithm to obtain the dynamic topology rearrangement strategy. The instruction generation module is used to combine migration priority and dynamic topology reordering strategy to generate migration trigger conditions and corresponding cache execution instructions. The scheduling execution monitoring module is used to collect the real-time scheduling status of the handling equipment. When the real-time scheduling status meets the migration trigger condition, the corresponding cached execution instruction is executed. At the same time, the real-time scheduling status is monitored for sudden tasks. When a sudden task is detected, the priority of the sudden task is obtained. When the priority of the sudden task is higher than the migration priority corresponding to the current cached execution instruction, the scheduling preemption rollback delay is calculated and a reordering interrupt recovery strategy is formulated.

9. An apparatus comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the method for efficient pallet storage in logistics storage as described in any one of claims 1 to 7.

Images