Dynamic planning method for goods allocation of stereoscopic warehouse

Abstract

The invention relates to a dynamic planning method for goods allocation of a stereoscopic warehouse, and the method comprises the steps: determining an in-warehouse commodity cluster on a storage target shelf in current goods allocation planning information; utilizing a sales volume prediction model to predict first sales volume data of the in-warehouse commodities in a preset period, and updating historical ex-warehouse frequency parameters corresponding to the in-warehouse commodities and historical cluster ex-warehouse frequency parameters corresponding to the in-warehouse commodity clusters according to the first sales volume data; after the commodity cluster correlation degree corresponding to any two in-library commodity clusters is determined, coding all in-library commodity clusters and target shelves based on the commodity cluster correlation degree and the updated new cluster ex-warehouse frequency parameters, and generating a population including initial particles; and performing particle swarm search iteration processing on the initial particle swarm by using a preset particle swarm optimization algorithm and a preset first fitness function until a target goods allocation code body is generated, and taking allocation schemes corresponding to all target goods allocation codes of the target goods allocation code body as dynamic planning results.

Description

Technical Field

[0001] This application relates to the field of smart logistics and warehouse management technology, and in particular to a dynamic planning method for storage locations in automated warehouses. Background Technology

[0002] In e-commerce, fluctuations in sales volume are common, influenced by factors such as changes in market demand, seasonality, promotional activities, and competitor strategies. When sales of certain products fluctuate, changes in warehouse storage planning become necessary, requiring additional manpower, resources, and operational costs. However, in e-commerce warehouses, appropriately shifting the location of some goods—that is, adjusting the inventory planning—can shorten the picking path for staff, reduce walking time, improve picking efficiency, and decrease errors when staff venture deep into warehouse storage areas. This ultimately enhances customer satisfaction and overall supply chain efficiency.

[0003] In related technologies, when the frequency of outbound shipments of some goods changes due to fluctuations in sales, warehouse managers cannot effectively adjust the layout of storage locations. The adjusted storage locations are not conducive to subsequent product picking, or the corresponding warehouse storage location planning adopts static storage location planning. The planned storage location scheme will cause picking personnel to walk long distances and take long picking time, resulting in low warehouse goods picking efficiency.

[0004] Currently, no effective solution has been proposed for the low picking and outbound efficiency of the storage location adjustment schemes used in related technologies after fluctuations in commodity sales. Summary of the Invention

[0005] This application provides a dynamic planning method for storage locations in an automated warehouse, which at least solves the problems of low picking and outbound efficiency in the storage location adjustment schemes adopted after fluctuations in commodity sales in related technologies.

[0006] This application provides a dynamic planning method for storage locations in an automated warehouse, comprising: determining, in the current storage location planning information, a cluster of in-stock goods stored on a target shelf in the automated warehouse, wherein the cluster of in-stock goods is associated with multiple in-stock goods stored in a target picking order, and the historical outbound frequency parameter of the cluster of in-stock goods is determined based on the historical outbound frequency parameters of all corresponding in-stock goods; using a trained sales forecasting model, predicting the first sales data of the in-stock goods within a preset period, and updating the historical outbound frequency parameter corresponding to the in-stock goods and the historical outbound frequency parameter corresponding to the cluster of in-stock goods based on the first sales data, wherein the sales forecasting model is a neural network model trained based on PSO and LSTM; after determining the cluster correlation degree between any two clusters of in-stock goods, updating the new cluster based on the cluster correlation degree and the updated cluster... The outbound frequency parameter is used to encode all the in-stock product clusters and the target shelves, generating an initial particle population including multiple location allocation codes. Each location allocation code represents a allocation scheme for assigning the in-stock product cluster to the corresponding target shelf, and two adjacent location allocation codes are associated with a product cluster association degree. Using a preset particle swarm optimization algorithm and a preset first fitness function, the initial particle population is subjected to particle swarm search iterative processing until a target location allocation code is generated. The allocation schemes corresponding to all target location allocation codes of the target location allocation code are used as dynamic programming results. The first fitness function is used to determine the fitness of the corresponding location code, and the fitness is used to characterize the changes in transfer and picking costs resulting from adjusting the target shelf corresponding to the in-stock product cluster.

[0007] Compared to related technologies, the dynamic planning method for automated warehouse locations provided in this application involves: determining the in-stock product clusters stored on the target shelf in the automated warehouse based on the current location planning information; using a trained sales forecasting model to predict the first sales data of all in-stock products within a preset period for each in-stock product cluster; updating the historical outbound frequency parameters corresponding to the in-stock products and the historical cluster outbound frequency parameters corresponding to the in-stock product clusters based on the first sales data; after determining the cluster correlation degree between any two in-stock product clusters, encoding all in-stock product clusters and the target shelf based on the cluster correlation degree and the updated new cluster outbound frequency parameters to generate an initial particle swarm including multiple location allocation codes; and utilizing a preset particle swarm optimization algorithm and a preset first... A fitness function is used to perform particle swarm search iterative processing on the initial particle population until a target storage location allocation code body is generated. The allocation schemes corresponding to all target storage location allocation codes in the target storage location allocation code body are used as dynamic programming results. By predicting the sales volume of in-stock goods within a preset period, the change in the outbound frequency of in-stock goods is inferred. Based on the change in outbound frequency, the product clusters that need storage location adjustment are determined. At the same time, based on the existing storage location planning, the dynamic adjustment of the shelf storage position corresponding to the product clusters is adopted as the goal, realizing warehouse storage location dynamic planning based on sales prediction. This solves the problem of low picking efficiency and outbound efficiency of the storage location adjustment scheme adopted after the fluctuation of product sales in related technologies. It achieves the beneficial effects of improving warehouse picking efficiency and overall operational efficiency, reducing picking time and improving outbound efficiency. Attached Figure Description

[0008] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a hardware structure block diagram of the terminal of the dynamic planning method for storage locations in an automated warehouse according to an embodiment of this application. Figure 2 This is a flowchart of a dynamic planning method for storage locations in an automated warehouse according to an embodiment of this application; Figure 3 This is a schematic diagram of a cargo location allocation code body according to an embodiment of this application; Figure 4 This is a schematic diagram of the commodity cluster coding in an embodiment of this application; Figure 5 This is a schematic diagram of a static decision-encoded particle according to a preferred embodiment of this application; Figure 6 This is a schematic diagram of the order product association matrix according to a preferred embodiment of this application; Figure 7This is a schematic diagram of the co-occurrence matrix of goods according to an embodiment of this application; Figure 8 This is a schematic diagram of a product association network according to an embodiment of this application; Figure 9 This is a schematic diagram of a Boolean matrix as an embodiment of this application; Figure 10 This is a schematic diagram of shell decomposition in an embodiment of this application; Figure 11 This is a schematic diagram of the 2-shell decomposition involved in the embodiments of this application; Figure 12 This is a schematic diagram of the 4-shell decomposition involved in the embodiments of this application; Figure 13 This is a structural block diagram of a dynamic planning device for automated warehouse storage locations according to an embodiment of this application. Detailed Implementation

[0009] To make the objectives, technical solutions, and advantages of this application clearer, the application is described and illustrated below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the application. All other embodiments obtained by those skilled in the art based on the embodiments provided in this application without inventive effort are within the scope of protection of this application. Furthermore, it is understood that although the efforts made in such a development process may be complex and lengthy, for those skilled in the art related to the content disclosed in this application, modifications to design, manufacturing, or production based on the technical content disclosed in this application are merely conventional technical means and should not be construed as insufficient disclosure of the content of this application.

[0010] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment that is mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this application may be combined with other embodiments without conflict.

[0011] Unless otherwise defined, the technical or scientific terms used in this application shall have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms "a," "an," "an," "the," and similar words used in this application do not indicate quantity limitation and may indicate singular or plural. The terms "comprising," "including," "having," and any variations thereof used in this application are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units not listed, or may include other steps or units inherent to these processes, methods, products, or devices. "Multiple stages" used in this application refers to two or more stages. "And / or" describes the relationship between related objects, indicating that three relationships may exist; for example, "A and / or B" can represent: A alone, A and B simultaneously, and B alone. The terms "first," "second," "third," etc., used in this application are merely to distinguish similar objects and do not represent a specific ordering of objects.

[0012] The method embodiments provided in this example can be executed on a terminal, computer, or similar computing device. Taking running on a terminal as an example, Figure 1 This is a hardware structure block diagram of the terminal for the dynamic planning method of automated warehouse storage locations according to an embodiment of this application. For example... Figure 1 As shown, a terminal may include one or more ( Figure 1 Only one is shown in the diagram. A processor 102 (which may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.) and a memory 104 for storing data are also shown. Optionally, the terminal may further include a transmission device 106 for communication functions and an input / output device 108. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the terminal described above. For example, the terminal may also include components that are more... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0013] The memory 104 can be used to store computer programs, such as application software programs and modules, like the computer program corresponding to the dynamic planning method for automated warehouse storage locations in this embodiment of the invention. The processor 102 executes various functional applications and data processing by running the computer programs stored in the memory 104, thereby implementing the aforementioned method. The memory 104 may include high-speed random access memory and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the terminal 10 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0014] The transmission device 106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of terminal 10. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 may be a Radio Frequency (RF) module, used for wireless communication with the Internet.

[0015] This embodiment provides a dynamic planning method for storage locations in an automated warehouse running on the aforementioned terminal. Figure 2 This is a flowchart of a dynamic planning method for automated warehouse storage locations according to an embodiment of this application, such as... Figure 2 As shown, the process includes the following steps: Step S201: In the current storage location planning information, determine the in-stock product clusters on the target shelves of the automated warehouse. The in-stock product clusters are associated with multiple in-stock products stored according to the target picking order. The historical outbound frequency parameter of the in-stock product clusters is determined based on the historical outbound frequency parameters of all corresponding in-stock products.

[0016] In this embodiment, the executing entity of the dynamic programming method of this application includes, but is not limited to, a warehouse management system applied to warehouse management. Simultaneously, the warehouse manager executes the dynamic programming method of this application by running the corresponding warehouse management system. In this embodiment, during the management of the automated warehouse, the warehouse manager will periodically or regularly adjust the storage locations of the goods in the warehouse to cope with changes in the sales volume of the corresponding goods in the future. In this embodiment, because the sales volume of some goods changes, the current storage location plan is no longer suitable for picking and shipping. A comprehensive adjustment of all storage locations would inevitably lead to a waste of manpower and resources and increased operating costs. Instead, dynamic storage location planning is performed by appropriately adjusting and moving the storage locations of some goods based on the existing storage location plan, with the goal of adjusting product clusters with higher shipping frequency. Therefore, when performing dynamic storage location planning, the current storage location planning information is first obtained, and the storage location of the currently stored product clusters in the automated warehouse is determined based on this information, that is, on which target shelf the product clusters are stored.

[0017] In this embodiment, the current location planning information is the location planning scheme executed by the warehouse manager before the current time. The current location planning information is determined based on the number of location planning adjustments completed before the current time. If the dynamic planning operation to be performed is the second one, then the current location planning information is the commodity storage scheme generated by the initial static location planning. At this time, the corresponding plan considers the outbound frequency of commodity clusters, the correlation between commodity clusters, and the storage location (storage location) of all in-stock commodities in the commodity cluster. If the dynamic planning operation to be performed is the third or more, then the current location planning information is the commodity storage scheme generated after the completion of the previous dynamic planning. Dynamic planning operations after the static location planning operation are all based on the commodity storage scheme generated by the static location planning, and the commodity clusters with large fluctuations in outbound frequency are moved and adjusted to generate the corresponding commodity storage scheme. In this embodiment, an in-stock commodity cluster is generated by clustering relevant commodity information before static location planning, and the number of commodities corresponding to an in-stock commodity cluster is also determined according to the location on the target shelf of the automated warehouse. It should be understood that when commodity sales fluctuate, causing When the outbound frequency of corresponding in-stock goods changes, the clustering of all goods in a single product cluster remains unchanged; that is, the outbound frequency of a product does not affect the product cluster to which it belongs. Simultaneously, when moving or adjusting an in-stock product cluster, the picking order (target picking order) of all in-stock goods in that cluster remains unchanged before and after the move. For example, before the move, in-stock goods 1 to 9 of an in-stock product cluster are stored at locations 1 to 9 of target shelf A; when moved to target shelf B, in-stock goods 1 to 9... Product 9 is still stored in location 1 to location 9 of target shelf B. In this embodiment, due to fluctuations in product sales, the outbound frequency of the corresponding product will change, which in turn will cause changes in the cluster outbound frequency of the corresponding in-stock product cluster. Moreover, the change in cluster outbound frequency is a parameter considered in the location adjustment planning. Before and after dynamic planning, each in-stock product cluster will have a corresponding matching cluster outbound frequency parameter. For example, before dynamic planning, the historical cluster outbound frequency parameter corresponding to the in-stock product cluster is calculated based on the outbound frequency parameters of all in-stock product clusters corresponding to the in-stock product cluster.

[0018] Step S202: Using the trained sales forecasting model, predict the first sales data of the goods in stock within a preset period, and update the historical outbound frequency parameters corresponding to the goods in stock and the historical cluster outbound frequency parameters corresponding to the product clusters in stock based on the first sales data. The sales forecasting model is a neural network model trained based on PSO and LSTM.

[0019] In this embodiment, after determining the goods stored in the automated warehouse, the sales volume of the goods in the warehouse over a certain period (e.g., one month) is predicted using a pre-trained sales prediction model (corresponding to the first sales data). Then, based on the corresponding sales volume, the outbound frequency parameter for each corresponding goods in the warehouse is determined, thereby determining the fluctuation of the outbound frequency for each goods in the warehouse. Subsequently, based on the outbound frequency parameters of all goods in the same goods cluster, the cluster outbound frequency parameter and the fluctuation of the cluster outbound frequency parameter are determined for each goods cluster, providing data parameters for subsequently determining which goods clusters in the warehouse need to be moved and how to adjust and move them.

[0020] In this embodiment, the sales forecasting model is a prediction network model trained using Particle Swarm Optimization (PSO) and Long Short-Term Memory (LSTM) networks. It is used to train a model that outputs the sales data corresponding to the input product. In this embodiment, PSO is used to obtain the hyperparameters with the best performance within the given range of hyperparameters. Then, the optimized LSTM model is constructed using the obtained optimal hyperparameters to predict sales.

[0021] Step S203: After determining the product cluster correlation degree corresponding to any two in-stock product clusters, based on the product cluster correlation degree and the updated new cluster outbound frequency parameter, all in-stock product clusters and target shelves are encoded to generate an initial particle population including multiple location allocation codes. Among them, a location allocation code represents an allocation scheme for allocating in-stock product clusters to the corresponding target shelves, and two adjacent location allocation codes are associated with a product cluster correlation degree.

[0022] In this embodiment, during dynamic location planning, a location allocation rule is adopted that simultaneously considers the outbound frequency of in-stock product clusters and the cluster correlation (relationship between clusters) to meet the requirement of lower product picking costs. It is understood that product picking costs mainly depend on the picking time spent in the outbound picking process and the distance traveled during picking. Product picking costs include two parts: First, the shelf storage location of in-stock product clusters is determined according to their outbound frequency. In-stock product clusters with higher outbound frequencies should be placed on target shelves closer to the warehouse entrance / exit, while in-stock product clusters with lower outbound frequencies should be placed on target shelves farther from the warehouse entrance / exit. Second, in-stock product clusters with certain correlations... The goods should be stored close together to ensure that the placement of all goods is related to their function when processing warehouse customer orders. In this embodiment, within a target shelf corresponding to a cluster of in-stock goods, changing the storage order of in-stock goods within the cluster based on changes in the frequency of goods leaving the warehouse can only slightly reduce the movement distance of picking personnel within an aisle. Therefore, this embodiment adopts a location adjustment method that only changes the storage location of the in-stock goods cluster. This means that it is necessary to consider changes in the frequency of goods leaving the cluster. A particle swarm optimization algorithm can be used to search for the solution, and a segmented encoding is used to represent the storage solution of an in-stock goods cluster corresponding to a particle. That is, one particle corresponds to one location allocation code body, and one location allocation code body (refer to...) Figure 3 (As shown) includes multiple location allocation codes. Each location allocation code corresponds to an allocation scheme for a cluster of in-stock items to a target shelf. For example, using J... i J1 represents the i-th in-stock product cluster. Encoding J1 as 11 means that the i-th in-stock product cluster will be stored on the target shelf 11. (J1, 11) corresponds to a location allocation code.

[0023] Step S204: Using a preset particle swarm optimization algorithm and a preset first fitness function, the initial particle population is subjected to particle swarm search iterative processing until the target storage location allocation code body is generated. The allocation schemes corresponding to all target storage location allocation codes of the target storage location allocation code body are used as dynamic programming results. The first fitness function is used to determine the fitness of the corresponding storage location code body. The fitness is used to characterize the changes in the transfer cost and picking cost caused by adjusting the target shelf corresponding to the in-stock commodity cluster.

[0024] In this embodiment, the particle swarm optimization algorithm and the first fitness function are used to perform particle swarm search iterative processing on the initial particle population. This is clear and executable to those skilled in the art and does not constitute an unclear limitation on the embodiments of this application. In this embodiment, the first fitness function is constructed by considering two parts: the cost of goods movement (handling distance and handling quantity) and the cost of goods picking.

[0025] Through steps S201 to S204 above, the following steps are taken: First, based on the current location planning information, the in-stock product clusters stored on the target shelves of the automated warehouse are determined. Then, using a trained sales forecasting model, the first sales data for all in-stock products within a preset period is predicted. Based on the first sales data, the historical outbound frequency parameters corresponding to the in-stock products and the historical cluster outbound frequency parameters corresponding to the in-stock product clusters are updated. After determining the cluster correlation between any two in-stock product clusters, based on the cluster correlation and the updated new cluster outbound frequency parameters, all in-stock product clusters and the target shelves are encoded to generate an initial particle population including multiple location allocation codes. Finally, using a preset particle swarm optimization algorithm and a preset first fitness function, the initial particle population is optimized. Particle swarm search iterative processing is performed until the target storage location allocation code body is generated. The allocation schemes corresponding to all target storage location allocation codes in the target storage location allocation code body are used as dynamic programming results. By predicting the sales volume of in-stock goods within a preset period, the change in the outbound frequency of in-stock goods is inferred. Based on the change in outbound frequency, the product clusters that need storage location adjustment are determined. At the same time, based on the existing storage location planning, the dynamic adjustment of the shelf storage position corresponding to the product clusters is adopted as the goal. This realizes warehouse storage location dynamic planning based on sales forecast, which solves the problems of low picking efficiency and outbound efficiency of the storage location adjustment scheme adopted after the fluctuation of product sales in related technologies. It achieves the beneficial effects of improving warehouse picking efficiency and overall operational efficiency, reducing picking time and improving outbound efficiency.

[0026] It should be noted that in this embodiment, changes in the sales volume of some products have rendered the current warehouse location plan unsuitable for picking and outbound goods. A comprehensive adjustment of all warehouse locations would inevitably lead to a waste of manpower and resources and an increase in operating costs. However, by appropriately adjusting and moving some product locations within the existing warehouse location plan, such as moving products with higher outbound frequency to positions closer to the storage area entrance and exit, the picking path for operators can be shortened. This also reduces the walking time spent by operators during the picking process, improves picking efficiency, and reduces the error rate of picking personnel going deep into the warehouse storage area. This, in turn, improves customer satisfaction and overall supply chain efficiency, achieving the goal of improving product picking and outbound efficiency while minimizing the increased costs, thus maintaining a high level of outbound picking in the automated warehouse.

[0027] In some embodiments, a preset particle swarm optimization algorithm and a preset first fitness function are used to perform particle swarm search iterative processing on the initial particle population until the target cargo location allocation code is generated, which is achieved through the following steps: Step 21: After using the location allocation code as a candidate particle, determine the first initial particle position and the preset first initial particle velocity for each candidate particle, wherein the first initial particle position is used to characterize the shelf allocation scheme of all in-stock commodity clusters corresponding to the corresponding location allocation code.

[0028] In this embodiment, after determining the corresponding encoding scheme, it is necessary to initialize the initial velocity (corresponding to the first initial particle velocity) and initial position (corresponding to the first initial particle position) of the particle swarm. In this embodiment, the first initial particle position is used to characterize an initial encoding scheme for each location allocation code, that is, an allocation scheme for all in-stock product clusters to be allocated to the corresponding target shelves, that is, to randomly allocate a target shelf to each in-stock product cluster. However, when initializing the initial particle position of each location allocation code, the correlation degree of the product clusters of the two adjacent location allocation codes of each location allocation code needs to be considered.

[0029] Step 22: According to the preset dynamic adaptive change method, determine the first inertia factor, the first learning factor and the second learning factor corresponding to the current particle search, wherein the first inertia factor, the first learning factor and the second learning factor are generated by adaptive linear change based on the number of search iterations.

[0030] In this embodiment, taking the first inertia factor as an example, the process of generating the first inertia factor corresponding to the current particle search using adaptive linear transformation is explained. In this embodiment, the following formula is used: ,in, Let {L, K} represent the first inertia factor, {L, K} represent the range of values ​​for the first inertia factor, i represent the current iteration number of the particle, and max represent the maximum number of iterations. In this embodiment, the first learning factor and the second learning factor can also be fixed learning factors.

[0031] Step 23: Based on the first initial particle position, the first initial particle velocity, the first inertia factor, the first learning factor, and the second learning factor, the particle swarm optimization algorithm is used to perform at least one population search iteration on all candidate particles until the fitness of each cargo location allocation code corresponding to the generated particle population is not greater than a preset fitness threshold, thereby obtaining a candidate particle population including at least one candidate particle, wherein the fitness of the cargo location allocation code is calculated based on the first fitness function.

[0032] In this embodiment, the velocity and position updates of candidate particles are performed in the same way as the first inertia factor update. In this embodiment, the number of population search iterations can be determined by the fitness not being greater than the corresponding fitness threshold, or by setting a maximum number of iterations and using the fitness at the maximum number of iterations as the fitness threshold, thereby limiting the number of iterations and obtaining at least one candidate particle that meets the requirements, that is, obtaining a candidate particle population.

[0033] Step 24: Select a target particle from the candidate particle population, wherein the cargo location allocation code corresponding to the target particle includes the target cargo location allocation code.

[0034] In some alternative implementations, step 24, selecting the target particle from the candidate particle population, includes the following steps: Step 241: Determine the fitness of the candidate cargo location allocation code corresponding to each candidate particle in the candidate particle population according to the first fitness function.

[0035] Step 242: Based on the fitness of the candidate storage location allocation code, select the candidate particle with the lowest fitness from multiple candidate particles in the candidate particle population to obtain the target particle, and use the corresponding candidate storage location allocation code as the storage location allocation code corresponding to the target particle.

[0036] Through steps 21 to 24 above, the corresponding storage location planning information is solved using the particle swarm optimization algorithm, so as to improve the picking efficiency and overall operational efficiency of the warehouse, reduce picking time and improve outbound efficiency.

[0037] In some embodiments, the cluster of in-stock goods stored on the target shelf of the automated warehouse is determined from the current storage location planning information through the following steps: Step 31: Determine the current storage location planning information obtained so far. The current storage location planning information includes one of the following: static storage location planning information generated by the initial static planning, or historical storage location planning information generated by completing at least one dynamic planning based on the static storage location planning information.

[0038] In this embodiment, the current storage location planning information is not the storage location planning information that has been dynamically adjusted in the current instance. The current storage location planning information is determined based on the number of storage location planning adjustments completed before the current instance. If the current dynamic adjustment corresponds to the second dynamic adjustment, then the current storage location planning information is the static storage location planning information generated by the initial static storage location planning. Furthermore, when performing the corresponding storage location planning, it is planned based on the outbound frequency of the product cluster, the association relationship of the product cluster, and the storage location (storage location) of all in-stock products of the product cluster. If the current dynamic adjustment corresponds to the third or subsequent dynamic adjustment, then the current storage location planning information is the historical storage location planning information generated by completing at least one dynamic planning based on the static storage location planning information. At the same time, the dynamic planning operations after the static storage location planning operation are all based on the static storage location planning information generated by the static storage location planning, and the product clusters with large fluctuations in outbound frequency are moved and adjusted to generate the corresponding storage location planning information.

[0039] Step 32: Obtain all storage allocation information corresponding to the current storage location planning information, where one storage allocation information corresponds to one target shelf.

[0040] In this embodiment, the corresponding location planning information represents the arrangement of the in-stock goods of the product cluster on the corresponding target shelf. That is, it includes multiple storage allocation information. One storage allocation information is associated with one target shelf. After obtaining the corresponding storage allocation information, the in-stock product cluster stored on the corresponding target shelf is determined. At the same time, the storage allocation information can also determine the product information of all in-stock goods of the in-stock product cluster on the corresponding target shelf and the storage location on the target shelf, which is the corresponding picking order.

[0041] Step 33: For each warehouse allocation information, detect the cluster information associated with all target shelves and the sorting information associated with each cluster information, and determine the in-stock product clusters stored on the corresponding target shelves based on the cluster information, and determine the target picking sorting of all in-stock products corresponding to each in-stock product cluster based on the sorting information.

[0042] Through steps 31 to 33 above, the current inventory clusters and all inventory items on each target shelf are determined, providing a data foundation for determining the cluster outbound frequency and the correlation between inventory clusters through sales forecasting.

[0043] In some embodiments, based on the first sales data, the historical outbound frequency parameters corresponding to the in-stock goods and the historical cluster outbound frequency parameters corresponding to the in-stock goods clusters are updated through the following steps: Step 41: Based on the first sales data, determine the outbound frequency parameters of the in-stock products within the preset period after the current time, and obtain the corresponding new outbound frequency parameters.

[0044] In this embodiment, the sales volume of a product determines the outbound frequency of the corresponding in-stock product. That is, by determining the first sales volume data of each in-stock product within a preset period, the outbound frequency parameter of the corresponding in-stock product can be determined.

[0045] Step 42: Determine the average value of all new outbound frequency parameters to obtain the current cluster outbound frequency parameter, and determine whether the change value between the current cluster outbound frequency parameter and the historical cluster outbound frequency parameter is greater than the preset change threshold. The historical cluster outbound frequency parameter is the average value of the corresponding historical outbound frequency parameter in the preset period before the current time. The historical outbound frequency parameter is determined based on the second sales data of the corresponding in-stock goods in the preset period before the current time.

[0046] In this embodiment, the current outbound frequency of an in-stock commodity cluster is determined by the outbound frequency of all in-stock commodities within that cluster, which is the average outbound frequency of all in-stock commodities within that cluster. In this embodiment, the current outbound frequency parameter of the corresponding in-stock commodity cluster is obtained based on the average of the new outbound frequency parameters of all in-stock commodities, and the magnitude of the current outbound frequency parameter is compared with the previous historical outbound frequency parameters to determine whether the outbound frequency parameter of the in-stock commodity cluster has fluctuated.

[0047] Step 43: If the change value is found to be greater than the preset change threshold, update the current cluster outbound frequency parameter to the new cluster outbound frequency parameter.

[0048] Step 44: If it is determined that the change value is not greater than the preset change threshold, the historical cluster outbound frequency parameter is used as the new cluster outbound frequency parameter.

[0049] In this embodiment, when it is determined that the cluster outbound frequency parameter of the corresponding in-stock product cluster fluctuates greatly, that is, when the cluster outbound frequency parameter increases, it indicates that the corresponding in-stock product cluster needs to be adjusted. At this time, the new cluster outbound frequency parameter is used as the outbound frequency parameter of the corresponding in-stock product cluster, so as to move the in-stock product cluster with the estimated increased outbound frequency to a position closer to the warehouse entrance and exit through dynamic planning adjustment. When it is determined that the cluster outbound frequency parameter of the corresponding in-stock product cluster fluctuates low, it indicates that the corresponding in-stock product cluster does not need to be moved or needs to be moved to a position away from the warehouse entrance and exit. At this time, the original cluster outbound frequency parameter of the in-stock product cluster is retained.

[0050] Through steps 41 to 44 above, the sales volume of in-stock goods in the future period is predicted, thereby inferring the changes in the frequency of goods leaving the warehouse. This determines which in-stock goods and in-stock product clusters need to be repositioned, thereby improving the picking efficiency and overall operational efficiency of the warehouse. In-stock product clusters with higher outbound frequency are repositioned to more convenient locations to reduce picking time and improve outbound efficiency.

[0051] In some embodiments, static storage location planning information is generated through the following steps: Step 51: Determine the cluster information corresponding to multiple in-stock product clusters to be stored and the shelf information corresponding to multiple target shelves. The cluster information is generated by performing correlation statistics on all SKUs of in-stock products according to a preset statistical method. The shelf information includes multiple target storage locations set on the target shelves.

[0052] In this embodiment, before generating initial static storage location planning information, corresponding data is acquired, including basic parameters, average outbound frequency parameters for product clusters, product relationships, and shelf data. In this embodiment, basic parameters include the types of goods in stock, the number of target shelves, the number of clustered product clusters, and the number of goods in each cluster (corresponding to the maximum capacity of the target shelf). The outbound frequency of goods is determined based on the sales volume of goods in the corresponding order, while the average outbound frequency of a product cluster needs to consider which goods are included in the cluster, summing and averaging the outbound frequencies of all goods within the cluster. In this embodiment, the strength of product relationships is determined based on the frequency of combined purchases of the corresponding goods. In this embodiment, shelf data includes the aisle width between shelves, the width of each target shelf, the width of each storage location, and the width of the main warehouse aisle. The acquired corresponding data is used to mathematically model the dynamic adjustment scheme for storage locations during model building, objective function determination, and model solving.

[0053] Step 52: Based on the historical outbound frequency parameters of the corresponding in-stock goods, determine the historical cluster outbound frequency parameters of the in-stock goods cluster corresponding to each cluster information and the historical cluster correlation degree between any two in-stock goods clusters.

[0054] In this embodiment, after determining the historical outbound frequency parameters for each in-stock item, the historical outbound frequency parameters for all in-stock items are summed and then averaged to obtain the historical cluster outbound frequency parameters corresponding to each cluster information. In this embodiment, the cluster correlation degree between any two in-stock item clusters is the sum of the correlations (e.g., product weighting) between all in-stock items in each in-stock item cluster and all in-stock items in another compared in-stock item cluster. The correlation C between in-stock item cluster k and in-stock item cluster l is calculated as follows: klIt is necessary to calculate the sum of the association relationships between product i in product cluster k and all other products in product cluster l, and use... This represents the association between product i in product cluster k and product j in product cluster l.

[0055] Step 53: Based on the historical outbound frequency parameters, historical cluster outbound frequency parameters, and historical cluster correlation, perform two-stage coding on the in-stock product clusters, target shelves, in-stock products, and target storage locations to generate multiple static decision coding particles. The static decision coding particles include product cluster shelf allocation sub-particles with multiple shelf allocation particles and product storage location allocation sub-particles with multiple storage location allocation particles. The product cluster shelf allocation sub-particles are used to represent a shelf allocation scheme for allocating target shelves to all in-stock product clusters, and the product storage location allocation sub-particles are used to represent a sorting scheme for allocating storage locations to in-stock products on a target shelf.

[0056] Step 54: Using the particle swarm optimization algorithm, perform particle swarm search iterative processing on multiple static decision-encoded particles until the target decision-encoded particle is obtained. The static storage location planning information includes all shelf allocation schemes and all sorting schemes corresponding to the target decision-encoded particle.

[0057] In this embodiment, the operation of the particle swarm optimization algorithm is clear to those skilled in the art.

[0058] Through steps 51 to 54 above, the initial static planning information was generated.

[0059] In some embodiments, a particle swarm optimization algorithm is used to perform particle swarm search iterative processing on multiple static decision-encoded particles until the target decision-encoded particle is obtained, including the following steps: Step 61: Determine the second initial particle position and the preset second initial particle velocity for each static decision-encoded particle, wherein the second initial particle position is used to characterize the shelf allocation scheme and picking sorting allocation scheme corresponding to the corresponding static decision-encoded particle.

[0060] Step 62: Determine the second inertia factor, the third learning factor, and the fourth learning factor corresponding to the current particle search. The second inertia factor, the third learning factor, and the fourth learning factor are generated by adaptive linear changes based on the number of search iterations.

[0061] Step 63: Based on the second initial particle position, second initial particle velocity, second inertia factor, third learning factor, and fourth learning factor, the particle swarm optimization algorithm is used to perform at least one population search iteration on all static decision-coded particles until the fitness of each static decision-coded particle corresponding to the generated particle population is not greater than a preset fitness threshold, thereby obtaining at least one candidate decision-coded particle. The fitness of the static decision-coded particle is calculated based on a preset second fitness function, which is constructed based on the picking movement distance function and the picking time function. The picking movement distance function is constructed based on the corresponding cluster outbound frequency parameter and cluster correlation degree, and the picking time function is constructed based on the sorting information of the target storage location corresponding to the in-stock goods.

[0062] Step 64: Select the candidate decision encoding particle with the smallest fitness from at least one candidate decision encoding particle to obtain the target decision encoding particle.

[0063] Through steps 61 to 64 above, the particle swarm optimization algorithm is used to solve for the target decision-encoded particles corresponding to the static cargo location planning information.

[0064] To determine the relationships between in-stock items and the cluster correlation of in-stock item clusters, in some embodiments, the following steps are performed before determining the cluster information corresponding to multiple in-stock item clusters to be stored: Step 71: Use one-hot encoding to construct a matrix of multiple historical order data and the SKUs of the corresponding products, and generate an order product association matrix with historical order data as rows and SKUs as columns.

[0065] In this embodiment, in order to obtain the relationship between orders and products from a large amount of order data, a "one-hot encoding" representation is used to form a relationship matrix between orders and products, which is defined as follows: Where N represents the total number of product types in the order, D represents the total number of historical orders, and r is the element in the i-th row and j-th column of the order-product relationship matrix. ij Defined as: The constructed order product association matrix (see reference) Figure 6 ).

[0066] Step 72: After generating the product co-occurrence matrix based on the order product association matrix, generate the corresponding product association network according to the product co-occurrence matrix. The product co-occurrence matrix includes multiple product association degrees used to represent the co-occurrence frequency of two corresponding products in all historical order data. A network graph node of the product association network is used to represent a corresponding SKU, and a node path of the product association network is used to represent the association relationship between two corresponding SKUs.

[0067] In this embodiment, the co-occurrence matrix of products is calculated based on the order product association matrix (see reference). Figure 7 Then, the corresponding product co-occurrence matrix is ​​converted into the corresponding product association network (see reference). Figure 8 ); It can be understood that in a product association network, a network graph node (refer to...) Figure 8 The degree (as indicated by the letters in the graph) refers to the edge directly connected to the node in the network graph (see reference). Figure 8 The number of other letters (connected to a given letter) in a node path is... Figure 8 The connecting lines in the diagram indicate the corresponding relationships between the numbers.

[0068] Step 73: Determine the undirected association network corresponding to the product association network. Weight the degree of each network node and the undirected path between two network nodes in the undirected association network to obtain the product weighting degree corresponding to a SKU. Based on the product weighting degree, use the preset k-shell decomposition analysis method to classify all products corresponding to the historical order data to obtain product groups at multiple shell levels. Among them, network nodes are used to represent a corresponding SKU, and undirected paths are used to represent whether two corresponding SKUs are purchased together. Each shell level corresponds to a weighting degree interval.

[0069] In this embodiment, a Boolean matrix is ​​abstracted from the product association matrix. The Boolean matrix (reference) Figure 9 The purpose of this is to determine whether there is a relationship between two products, that is, whether there is an edge between two nodes in the product association network. The formula for determining whether product i and product j have been purchased as a combination is as follows: The Boolean matrix can be used to transform the network into an undirected network. Then, based on the undirected network, the degree d of each node can be determined. i This refers to the degree of the corresponding product, and its calculation formula is: In a product association network, degree can be understood as a measure of the number of edges directly connecting nodes, reflecting the association between nodes. At the product level, degree reflects how many other products a particular product is associated with. However, product association networks only consider whether a node or product has relationships with other products, without directly considering the strength of these relationships, i.e., the weights of edges between nodes. Edge weights are crucial in product association networks. When considering a product's degree, it's necessary to consider not only the number of associated products but also the weights of these relationships. Therefore, weighted degree, as an improved measure, considers both the product's degree and the weights of edges between products, providing a more comprehensive description of a product's association within the network. We define weighted degree as... The calculation formula is as follows: ,in, This represents the sum of edge weights connecting product i to other products, reflecting the strength of the association between product i and other products.

[0070] In this embodiment, to cluster the products, it is first necessary to perform weighted network k-shell decomposition on the weighted degree of the products. Weighted network k-shell decomposition is a network analysis method used to reveal the structure and connection patterns between nodes in the network. The following references... Figures 10 to 12 The k-shell decomposition involved in the embodiments of this application will be described as follows:

[0071] In this embodiment, firstly, isolated nodes in the product association network, i.e., products with a weighted degree of 0, are classified as 0-shell products. A weighted degree of 0 indicates that these products are not purchased simultaneously with other products, their association is weak, and they are not prioritized when planning storage locations based on product association. Then, 1-shell products are separated: the weighted degree of the node (product) is calculated according to the formula for calculating the weighted degree of the product, and nodes with a weighted degree less than or equal to 1 and their corresponding edges are deleted. Then, the weighted degree of the products reflected by the remaining nodes is recalculated. This process is repeated until no nodes with a weighted degree less than or equal to 1 appear in the product association network. Finally, the products corresponding to all nodes removed in this layer are classified as 1-shell products. The 1-shell decomposition is then complete. (See reference...) Figure 10 Repeat the above operation until there are no remaining nodes and edges in the product association network, and complete the k-shell decomposition task of all product weighted networks to obtain the results belonging to products of different shells.

[0072] In this embodiment, Figure 10The diagram shows the 1-shell decomposition. Since the weighted degree of products D and E is less than or equal to 1, the corresponding nodes and edges are deleted. After recalculating the weighted degree of the remaining nodes, no nodes with a weighted degree less than or equal to 1 appear. Therefore, the 1-shell decomposition is complete. The 1-shell products are: products D and products E. Figure 11 This is a schematic diagram of 2-shell decomposition. Since only product A has a weighted degree less than or equal to 2, and there are no nodes with a weighted degree less than or equal to 2 after deleting node A and its corresponding edge, the 2-shell decomposition is complete, and the 2-shell product is: product A. After completing the 2-shell decomposition, since the weighted degree of the products corresponding to the remaining nodes is all greater than or equal to 3, the 3-shell decomposition cannot be performed according to the rules, and the 4-shell decomposition process begins directly. Figure 12 This is a schematic diagram of the 4-shell decomposition, for reference. Figure 12 The 4-shell decomposition process is as follows: the weighted degrees of the remaining nodes B, C, and F are all less than 4, so they should be deleted simultaneously. The resulting 4-shell products are: product B, product C, and product F. After deletion, the product association network has no remaining nodes or edges. Therefore, the weighted network k-shell decomposition process ends, yielding products belonging to shell 1 (see reference). Figure 10 Shell 2 (reference) Figure 11 ) and shell 4 (reference) Figure 12 (Three levels of goods)

[0073] Step 74: In each shell-level product group, select the product with the highest product weight as the core product. In all products corresponding to all product groups, select at least one alternative product that has an undirected path to each core product. Then, from the corresponding at least one alternative product, select a preset number of alternative products in descending order of product weight. Finally, cluster all selected alternative products and the corresponding core products into in-stock product clusters corresponding to the core products. The preset number is determined based on the number of target storage locations on the target shelf.

[0074] In this embodiment, the k-shell decomposition method of weighted networks can be used to obtain the decomposition results of goods in different shell layers. The higher the shell layer, the stronger the correlation between goods within that shell layer, and the greater the probability that high-level goods will be purchased in combination with other goods. In this embodiment, core goods are divided based on the k-shell decomposition results, and goods are sorted according to the shell layer level. A certain number of goods at the top of the ranking are classified as core goods. The correlation ranking of goods will take priority into account the shell layer level. If multiple goods belong to the same shell layer level, they will be further sorted according to their weighted degree values. Considering that goods need to be stored according to their correlation relationship when optimizing the storage location later, in this embodiment, core goods are clustered to form... The resulting product clusters can directly correspond to the stored products on a shelf. When product locations need to be adjusted, they will be moved in units of product clusters. In this embodiment, the top 64 high-shell products in the k-shell decomposition result are divided into core products. Each shelf can store a maximum of 9 products, so the maximum number of products in stock for a product cluster is 9. That is, based on the association matrix of the core products, the top 8 strongly related products are clustered into a product cluster. If there are fewer than 8 related products, the total number of products in stock for a product cluster is allowed to be less than 8. If the top 8 strongly related products of a core product contain other core products, the core products are removed and the cluster is extended sequentially until the total number of products in stock for the product cluster is met.

[0075] Through steps 71 to 74 above, clustering based on core products of the association network is realized. The product clusters formed by clustering core products can directly correspond to the stored products of a shelf. When it is necessary to adjust the product location, the products will be moved in units of product clusters, thereby improving warehouse operation efficiency.

[0076] In some embodiments, the product association degree corresponding to any two in-stock product clusters is determined by the following steps: Step 81: Select the first target product cluster and the second target product cluster from the two current in-stock product clusters.

[0077] Step 82: Iterate through all the in-stock items of the first target product cluster in sequence, and in the undirected association network, determine the product weighting degree between the current product and all the in-stock items of the second target product cluster in each iteration, and sum the product weighting degrees between the current product and all the in-stock items of the second target product cluster in each iteration to obtain the sub-association degree between the current product and the second target product cluster.

[0078] Step 83: Sum the sub-associations of all traversed current products and the second target product cluster to obtain the total association between the first target product cluster and the second target product cluster, and use the corresponding total association as the association between any two in-stock product clusters.

[0079] Through steps 81 to 83 above, the cluster correlation degree between two adjacent in-stock product clusters is calculated, providing data support for decisions to store two adjacent in-stock product clusters close together, thus ensuring the effectiveness of storage location planning.

[0080] In some embodiments, the training and preparation sales forecasting model is implemented through the following steps: Step 91: Preprocess and extract features from the pre-collected commodity transaction data to generate a sample data sequence. Divide the sample data sequence into a training dataset and a test dataset according to a set ratio. The preprocessing includes one of the following: data completion and data cleaning. The sample data sequence includes the actual sales data of the corresponding commodity and multiple target features used to predict the sales volume of the commodity.

[0081] In this embodiment, the commodity transaction data mainly includes the basic information of the commodity, the sales data of the commodity, and the user behavior of the commodity. The basic information of the commodity mainly includes the commodity ID, the commodity category, the brand ID of the commodity, and the supplier ID. The sales data of the commodity mainly includes the sales of all commodities every day from the data start date, including the date, the number of orders placed, the order amount, the transaction amount, and the number of items sold. The user behavior of the commodity mainly includes the user behavior generated by online consumers when they are involved with a certain commodity every day, including the number of times they browse, the number of times they are added to the cart, the number of times they are favorited, and the number of people guided by the advertisement.

[0082] In this embodiment, features are filtered and processed based on existing fields in the dataset. Data related to user behavior can be directly used for prediction without further processing. These data fields include pageview count, number of pageviewers, number of times added to cart, number of people adding to cart, number of favorites, number of comments, and comment score. Although these data fields can be directly input into the model as feature variables, the original data table still has issues such as missing data and data anomalies. The data in the above fields are all integer or floating-point data. At the same time, data such as pageview count, number of favorites, and number of comments are generated in a certain amount every day, and there is a small probability of sudden large fluctuations. In addition, this type of data shows a certain correlation with data in the preceding and following time series. Therefore, the average data method is used to fill in the abnormal data. In this embodiment, through data preprocessing, the feature variables input to the pre-trained model include: number of pageviews, number of pageviewers, number of times added to cart, number of people adding to cart, number of favorites, number of comments, comment score, price, discount, advertising intensity, and seasonal attributes.

[0083] Step 92: After using the historical combined hyperparameters used for training the current model as the corresponding historical particles, the Particle Swarm Optimization (PSO) algorithm is used to perform at least one population search iteration on the historical particles to generate the current particle and determine the current combined hyperparameters corresponding to the current particle. The historical combined hyperparameters include one of the following: randomly generated initial combined hyperparameters, or combined hyperparameters that have undergone at least one PSO optimization. Both the historical combined hyperparameters and the current combined hyperparameters include the learning rate, the number of hidden layer units, and the dimension of the hidden layer state.

[0084] Step 93: After updating the historical combined hyperparameters of the historical LSTM model constructed based on the historical combined hyperparameters to the current combined hyperparameters, the current operation of training the model using the backpropagation algorithm is performed based on the historical LSTM model, the current combined hyperparameters, and the training dataset to generate the current LSTM model. The root mean square error between the current predicted sales data and the actual sales data corresponding to the test dataset is used as the hyperparameter fitness of the current combined hyperparameters.

[0085] Step 94: Repeat the operation of searching and training the corresponding LSTM model using the Particle Swarm Optimization (PSO) algorithm for combined hyperparameters until the hyperparameter fitness of the generated combined hyperparameters is not greater than the preset fitness threshold, to obtain the target combined hyperparameters, and reconstruct the corresponding target LSTM model based on the target combined hyperparameters. The sales prediction model includes the target LSTM model.

[0086] In this embodiment, the particle swarm optimization algorithm is used to optimize two important hyperparameters in LSTM: the number of hidden layers and the number of hidden layer neurons. Specifically, the position of each particle is first randomly initialized according to the range of hyperparameter values. Then, an initial LSTM model is built using the hyperparameter values ​​corresponding to the initialized particle positions. The optimization effect of the particle swarm optimization algorithm is reflected in treating the number of hidden layers and the number of hidden layer neurons as particles. The optimal values ​​of the two hyperparameters are found through competition and cooperation among the particles. Finally, the optimized LSTM model is constructed using the obtained optimal hyperparameters for sales prediction.

[0087] In some preferred embodiments, the training process of the PSO-LSTM prediction model includes: Step 1: Import the data and preprocess it. Normalize the sequence data and convert it into supervised learning (using sales data as labels). Then, set the first 80% of the data as the training set and the last 20% as the test set.

[0088] Step 2: Initialize the LSTM neural network structure with any value within the given range of hyperparameters, and establish the mapping relationship between the particle swarm particles and the two hyperparameters that the LSTM needs to optimize.

[0089] Step 3: Initialize the parameters of PSO, including particle inertial weights, acceleration constants, particle swarm size, particle velocity and position, maximum number of iterations of the particle swarm, and number of particles.

[0090] Step 4: Train and test the initial LSTM network structure using the dataset, and use the root mean square error (RMSE) of the prediction results as the fitness value.

[0091] Step 5: Update the best position and global best position of each particle in the iteration by comparing the corresponding fitness values, i.e., the individual optimal value and the global optimal value.

[0092] Step 6: Update the velocity and position of individual particles according to the relevant formulas.

[0093] Step 7: If the maximum number of iterations is satisfied, terminate the algorithm and output the optimal solution, i.e., output the optimal hyperparameters.

[0094] Step 8: Use the hyperparameters corresponding to the optimal solution optimized by the particle swarm optimization algorithm to construct an LSTM model, and then train and test it.

[0095] Through steps 91 to 94 above, the sales forecasting model was trained, resulting in a sales forecasting model that outputs the sales data of the in-stock products within a certain future period based on the input in-stock products.

[0096] The following also describes the process of generating static planning information and the process of dynamic planning of storage locations involved in the embodiments of this application.

[0097] The process of generating static planning information

[0098] First, regarding the storage location planning problem in a static environment, the following assumptions are made: (1) The number of storage locations meets the demand for the goods to be allocated, that is, the number of storage locations is definitely greater than the number of types of goods to be allocated; (2) There is only one entrance / exit (I / O) in the storage area, from which the picking personnel start and eventually return to the entrance / exit; (3) Only the outbound situation of goods is considered, and the inbound situation is not considered; (4) Each storage location can completely store a certain type of goods; (5) The default storage SKU is medium and small items, and the difference in volume and weight between SKUs is not considered.

[0099] II. Parameter Setting and Variable Definition

[0100] The following parameters are set: N represents the total number of in-stock product clusters; M represents the total number of columns on the target shelf; S represents the total number of in-stock products in a product cluster; R represents the total number of storage locations on a target shelf; a and b represent shelf half-zone numbers (a, b = 1 or 2), 1 represents the upper half-zone, and 2 represents the lower half-zone; g and h represent shelf column numbers, g, h = 1, 2, ..., M; k and l represent the k-th or l-th in-stock product cluster; i and j represent the i-th or j-th in-stock product cluster, i, j = 1, 2, ..., S; e and f represent the storage location numbers of the e-th or f-th storage location on a target shelf, e, f = 1, 2, ..., R; P k P represents the average outbound frequency of the k-th in-stock product cluster; i D represents the frequency of outbound shipments of the i-th item in stock; ag D represents the distance from the shelving unit a_g to the warehouse entrance / exit; ag,bh This represents the distance between shelf a_g and shelf b_h; d e,f C represents the distance between storage location e and storage location f; kl This represents the association between the k-th in-stock product cluster and the l-th in-stock product cluster; H1 represents the relationship between the i-th in-stock item in the k-th in-stock item cluster and the j-th in-stock item in the l-th in-stock item cluster; H2 represents the width of the aisle between shelves; H3 represents the width of each storage location; H4 represents the width of the main warehouse aisle. Let be the decision variable, representing whether the k-th in-stock product cluster is stored on the a_g-th target shelf. If stored, ,otherwise, ; Let be the decision variable, representing whether the i-th in-stock item is stored in the e-th storage location. If it is stored, ,otherwise, .

[0101] III. Objective Function

[0102] 1) Objective function for the cluster outbound frequency parameter of in-stock commodity clusters

[0103] A high outbound frequency for a particular product indicates that it requires frequent picking within a certain timeframe. Placing these products in locations closer to the warehouse entrance / exit, while placing products with low outbound frequency in locations farther away, effectively reduces the movement distance of picking staff, increases order picking speed, and thus reduces picking time. In this embodiment of location planning, location planning is based on in-stock product clusters. Therefore, it is necessary to consider the average outbound frequency of all in-stock products within each cluster. Clusters with high outbound frequency are placed on front-position shelves, while clusters with low outbound frequency are placed on back-position shelves. Using P... k This represents the average outbound frequency of a cluster of goods in stock, while the value of 𝑃𝑘 is determined by the outbound frequency of all goods within that cluster. Use D ag and D bg Let represent the distance from the g-th target shelf to the entrance / exit of the storage area, calculated using the following formula: Where (g-1) / 2 represents the number of target shelves in front of the g-th target shelf, and (g-1) / 4 represents the number of aisles in front of the g-th target shelf; Decision variables are introduced. The objective function for the cluster outbound frequency parameter of the in-stock commodity cluster is: .

[0104] 2) Objective function regarding the association relationship of in-stock product clusters In warehouse location planning, the relationships between all goods should be fully considered. If there is a strong, non-negligible relationship between product i in inventory cluster k and product j in inventory cluster l, then inventory cluster k and inventory cluster l should be stored as close as possible during location planning. At the same time, goods with external connections should also be placed in locations close to the main warehouse aisle. (Using C...) kl This represents the association between two different in-stock product clusters, i and l. Use D ag,bh D represents the distance between shelf a_g and shelf b_h. ag,bh The calculation rules are as follows: When hg is odd: If 𝑔 is odd and 𝑎=𝑏, If g is odd and a ≠ b, If g is even and a = b, If g is even and a ≠ b, When hg is even: if 𝑎=𝑏, If a≠b, Introducing decision variables and The objective function for the association relationship of product clusters in the inventory is: .

[0105] 3) Objective function regarding the storage of goods on the shelves The location planning for all in-stock items is performed using a one-to-one correspondence between in-stock item clusters and target shelves. After determining the in-stock item cluster to which each in-stock item belongs, it is also necessary to determine the storage location of each in-stock item within that cluster on a single target shelf. The objective function for storing items on the shelf is: .

[0106] 4) Constraints 1. Constraints on the association between different in-stock product clusters 𝑘 and 𝑘: .

[0107] 2. Distance constraints between adjacent target shelves: .

[0108] 3. Assign values ​​to the horizontal distances of the 9 target storage locations, and adjust the constraints as follows: .

[0109] 4. Constraints on the storage conditions of goods in stock: .

[0110] IV. Mathematical Model Establish the following static storage location planning mathematical model: ;

[0111] Constraints: .

[0112] V. Model Solving 1. Data preparation, including basic parameters and the average outbound frequency parameter P for in-stock product clusters. k Related relationships of products The data includes shelf data (H1, H2, H3, H4); in this embodiment, the basic parameters include the types of goods in stock, the number of target shelves, the number of set clusters of goods, and the number of goods in each cluster (corresponding to the maximum capacity of the target shelves); the outbound frequency of goods is determined by the sales volume of goods in the corresponding order, and the average outbound frequency of a product cluster needs to consider which goods are included in the product cluster, and then sum and average the outbound frequencies of all goods in the product cluster; in this embodiment, the strength of the correlation between goods is determined by the frequency of combined purchases of corresponding goods; in this embodiment, the shelf data includes the aisle width H1 between shelves, the width H2 of each target shelf, the width H3 of each storage location, and the width H4 of the main warehouse aisle.

[0113] 2. Determining the weights of the objective function

[0114] In this embodiment, the multi-objective optimization model of the static storage location planning model is converted into a single-objective optimization model. Due to the different levels of importance attached to the frequency of product cluster outbound, the correlation of product cluster, and the frequency of product outbound within the shelf, different weights are assigned to the three objective functions, and they are combined into a single objective function after weighting: minF=w1minf1+w2minf2+w3minf3, with constraint adjustment: w1+w2+w3=1, w1, w2, w3∈(0,1). In this embodiment, the weights of the three objective functions are determined by the analytic hierarchy process.

[0115] 3. Data Encoding

[0116] In this embodiment, the static storage location planning model first needs to determine the shelf placement position corresponding to the in-stock product cluster, and then determine the corresponding storage location position of the in-stock product within the in-stock product cluster. In this embodiment, when using the particle swarm optimization algorithm to solve the static storage location planning model, a two-stage encoding is used to represent a solution corresponding to a particle. In this embodiment, J1, J2, ..., J N This represents the IDs of all in-stock product clusters, W1, W2, ..., W... S This indicates an in-stock item belonging to a corresponding in-stock item cluster. Each in-stock item cluster is coded, generating a unique item cluster code (see reference). Figure 4 Then, multiple product clusters are encoded to generate a static decision encoding particle. The first part represents the product cluster index (𝑔∈𝑁), with a value range of [0,65] and cannot be repeated; the second part represents the product's location index (𝑒∈𝑆), with a value range of [1,9] and the product values ​​within the same cluster cannot be repeated (see reference). Figure 5 ).

[0117] 4. Generate the initial population After determining the encoding scheme for the static storage location planning problem, it is necessary to initialize the initial velocity and position of the particle swarm. Considering the computation time, in this embodiment, the particle swarm size is set to 100, and the initial velocity and position of the particles are generated by generating random values.

[0118] 5. Determine parameters In this embodiment, before the particle swarm algorithm is executed, relevant parameters are set, including the inertia factor, learning factors c1 and c2 of the particle swarm search algorithm. The inertia factor is generated using a linear dynamic change processing method.

[0119] 6. Fitness Function The fitness of each particle is calculated using a weighted single-objective function.

[0120] 7. Algorithm terminates The particle swarm optimization (PSO) algorithm terminates its computation when it reaches the pre-set maximum number of iterations. For the static storage location planning model, the maximum number of iterations for the PSO algorithm is set to 300.

[0121] Explanation of the model construction and objective function involved in dynamic programming for storage locations: I. Model Assumptions 1. Since the dynamic location adjustment model is based on the original static location planning model, it adjusts and changes the location of some e-commerce goods in the warehouse. Therefore, the problem assumptions of the dynamic location adjustment model will be basically consistent with the problem assumptions of the static location planning model. The relevant assumptions for the handling and movement of goods should also be added. The specific contents are as follows: (1) The same number of shelves are used before and after the goods are handled, and the empty shelves without stored goods are not considered; (2) There is only one entrance and exit (I / O) in the storage area. Pickers start from here and eventually return to this entrance and exit; (3) Only the outbound situation of goods is considered, and the inbound situation is not considered; (4) Each location can completely store a certain type of goods; (5) The default storage SKU is medium and small items, and the differences in volume and weight between SKUs are not considered; (6) If the goods cluster cannot be directly moved from the original shelf to the target shelf, the goods in the cluster are temporarily placed in the empty space of the warehouse until the target shelf is empty.

[0122] 2. Supplementary parameter settings and variable definitions for the dynamic storage location adjustment model include: r ag This indicates the number of remaining in-stock items on shelf a_g; F k This represents the average frequency of outbound shipments for the k-th in-stock commodity cluster within a future period. ;F i The frequency of the i-th item in a cluster of in-stock items being removed from the warehouse within a future period is represented by α; α represents the penalty coefficient for pickers who move items to the wrong location. ag,bh Let x be the decision variable, representing whether the item on shelf a_g needs to be moved to shelf b_h. If so, x... ag,bh =1, otherwise, x ag,bh =0.

[0123] II. Objective Function

[0124] 1) Product movement module

[0125] Calculating the movement costs of goods from one storage location to another requires consideration of multiple aspects to ensure a comprehensive and reasonable assessment of the overall movement costs. These aspects include: the quantity and weight of the goods, the distance and path the goods need to be moved, and the labor costs incurred during the handling process. The following section will analyze in detail the impact of these three aspects on movement costs. The quantity of goods is a fundamental indicator for calculating movement costs. If a large number of goods need to be moved, it means that staff cannot move all goods to the target storage location at once, but must complete the movement in multiple trips using appropriate equipment. Similarly, the weight of the goods directly affects the logistics costs during movement. If the goods are heavy, staff will inevitably be unable to maintain a normal speed during handling, thus increasing the time spent on goods transfer. The time spent in the goods handling process can be considered as a cost incurred during movement. The distance the goods are moved is also a factor that needs to be considered in the movement costs. Longer movement distances mean more time and resources are needed in the goods handling process. Therefore, the distance between the storage locations before and after a goods movement directly affects the cost of moving goods. Besides the quantity of goods and the movement distance, a change in storage location disrupts the picking habits developed by pickers in the preceding period. Before the location change, pickers are already familiar with the location of certain goods due to a large number of orders. If some orders request the same type of goods after the movement, staff will subconsciously go to the previous location, increasing the movement distance and time spent picking, thus reducing efficiency. Based on the learning curve concept in human factors engineering and the actual survey of this warehouse, pickers need approximately five picking attempts to adapt to the new location. Therefore, the model needs to consider the difficulty pickers experience in locating goods after a location change and should include corresponding penalty costs. In summary, the objective function for the goods movement module is:

[0126] Among them, (𝑟 𝑎𝑔 +1 𝑏ℎ ) represents the total number of items in the cluster that needs to be moved, 𝐷 𝑎𝑔𝑏ℎ The distance to be moved for the cluster of goods that needs to be moved is represented by , where represents the penalty coefficient for the picking staff if they move the goods to the wrong storage location. 𝑎𝑔𝑏ℎ The variable is 0-1 and is used to determine whether a cluster of goods has moved from shelf 𝑎_𝑔 to shelf 𝑏_ℎ.

[0127] 2) Goods picking module

[0128] To maintain both outbound frequency and relatedness in the location allocation rules for product clusters within the dynamic location adjustment model, the model must also satisfy the requirement of low product picking costs. Product picking costs primarily depend on the picking time and distance traveled during the outbound picking process. Therefore, product picking costs are influenced by two factors: First, product clusters are stored on shelves based on their outbound frequency; clusters with higher outbound frequency should be placed on shelves closer to the warehouse entrance, while clusters with lower outbound frequency should be placed on shelves farther away. Second, product clusters with certain relatedness should still be stored close together to ensure that the placement of all products is related to their relationships when processing customer orders. The objective function for product picking costs is: .

[0129] Based on comprehensive analysis, the following mathematical model for dynamic programming of storage locations is established: ;

[0130] Constraints: .

[0131] For solving the dynamic programming problem of cargo locations, please refer to the description of generating the target cargo location allocation code body in the above embodiments, which will not be repeated here.

[0132] This embodiment also provides a dynamic planning device for automated warehouse storage locations. This device is used to implement the above embodiments and preferred embodiments, and will not be repeated as already described. As used below, the terms "module," "unit," "subunit," etc., can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0133] Figure 13 This is a structural block diagram of a dynamic planning device for automated warehouse storage locations according to an embodiment of this application, such as... Figure 13 As shown, the device includes a determining module 131, a predicting module 132, a generating module 133, and a processing module 134, wherein,

[0134] The determination module 131 is used to determine the in-stock product clusters on the target shelves of the automated warehouse in the current storage location planning information. The in-stock product clusters are associated with multiple in-stock products stored in the target picking order. The historical outbound frequency parameter of the in-stock product clusters is determined based on the historical outbound frequency parameters of all corresponding in-stock products.

[0135] The prediction module 132, coupled to the determination module 131, is used to predict the first sales data of the goods in stock within a preset period using a trained sales prediction model, and update the historical outbound frequency parameters corresponding to the goods in stock and the historical cluster outbound frequency parameters corresponding to the clusters of goods in stock based on the first sales data. The sales prediction model is a neural network model trained based on PSO and LSTM.

[0136] The generation module 133, coupled to the prediction module 132, is used to encode all in-stock product clusters and target shelves based on the product cluster correlation degree and the updated new cluster outbound frequency parameter after determining the product cluster correlation degree corresponding to any two in-stock product clusters. This generates an initial particle population including multiple location allocation codes. In this initial population, a location allocation code represents a distribution scheme that allocates in-stock product clusters to the corresponding target shelves, and two adjacent location allocation codes are associated with a product cluster correlation degree.

[0137] The processing module 134, coupled to the generation module 133, is used to perform particle swarm search iterative processing on the initial particle population using a preset particle swarm optimization algorithm and a preset first fitness function until the target storage location allocation code body is generated. The allocation schemes corresponding to all target storage location allocation codes of the target storage location allocation code body are used as dynamic programming results. The first fitness function is used to determine the fitness of the corresponding storage location code body. The fitness is used to characterize the changes in the transfer cost and picking cost caused by adjusting the target shelf corresponding to the in-stock commodity cluster.

[0138] This embodiment also provides a service platform, including a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to perform the steps in any of the above method embodiments.

[0139] Optionally, the service platform may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor and the input / output device is connected to the processor.

[0140] Optionally, in this embodiment, the processor can be configured to perform the following steps via a computer program:

[0141] S1. In the current location planning information, determine the in-stock product clusters on the target shelves of the automated warehouse. The in-stock product clusters are associated with multiple in-stock products stored in order of target picking. The historical outbound frequency parameter of the in-stock product clusters is determined based on the historical outbound frequency parameters of all corresponding in-stock products.

[0142] S2 uses a trained sales forecasting model to predict the first sales data of in-stock products within a preset period, and updates the historical outbound frequency parameters corresponding to the in-stock products and the historical cluster outbound frequency parameters corresponding to the in-stock product clusters based on the first sales data. The sales forecasting model is a neural network model trained based on PSO and LSTM.

[0143] S3. After determining the correlation degree of any two in-stock product clusters, based on the correlation degree of product clusters and the updated new cluster outbound frequency parameter, encode all in-stock product clusters and target shelves to generate an initial particle population including multiple location allocation codes. Among them, a location allocation code represents a distribution scheme that allocates in-stock product clusters to the corresponding target shelves, and two adjacent location allocation codes are associated with a product cluster correlation degree.

[0144] S4. Using a preset particle swarm optimization algorithm and a preset first fitness function, the initial particle population is subjected to particle swarm search iterative processing until the target storage location allocation code body is generated. The allocation schemes corresponding to all target storage location allocation codes of the target storage location allocation code body are used as dynamic programming results. The first fitness function is used to determine the fitness of the corresponding storage location code body. The fitness is used to characterize the changes in the transfer cost and picking cost caused by adjusting the target shelf corresponding to the in-stock product cluster.

[0145] It should be noted that the specific examples in this embodiment can refer to the examples described in the above embodiments and optional implementations, and will not be repeated here.

[0146] Furthermore, in conjunction with the dynamic planning method for automated warehouse storage locations in the above embodiments, this application embodiment can provide a storage medium for implementation. This storage medium stores a computer program; when executed by a processor, the computer program implements any of the dynamic planning methods for automated warehouse storage locations in the above embodiments.

[0147] The above embodiments merely illustrate several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A dynamic planning method for storage locations in an automated warehouse, characterized in that, include: In the current location planning information, the in-stock product clusters on the target shelves of the automated warehouse are determined. The in-stock product clusters are associated with multiple in-stock products stored in a target picking order. The historical outbound frequency parameter of the in-stock product clusters is determined based on the historical outbound frequency parameters of all the corresponding in-stock products. Using a trained sales forecasting model, the first sales data of the in-stock goods within a preset period is predicted, and based on the first sales data, the historical outbound frequency parameter corresponding to the in-stock goods and the historical cluster outbound frequency parameter corresponding to the in-stock goods cluster are updated. The sales forecasting model is a neural network model trained based on PSO and LSTM. After determining the product cluster correlation degree corresponding to any two of the in-stock product clusters, based on the product cluster correlation degree and the updated new cluster outbound frequency parameter, all the in-stock product clusters and the target shelf are encoded to generate an initial particle population including multiple location allocation codes. Here, a location allocation code of the location allocation code represents an allocation scheme for allocating the in-stock product cluster to the corresponding target shelf, and two adjacent location allocation codes are associated with a product cluster correlation degree. Using a preset particle swarm optimization algorithm and a preset first fitness function, the initial particle population is subjected to particle swarm search iterative processing until a target storage location allocation code body is generated. The allocation schemes corresponding to all target storage location allocation codes of the target storage location allocation code body are used as dynamic programming results. The first fitness function is used to determine the fitness of the corresponding storage location code body. The fitness is used to characterize the changes in the transfer cost and picking cost caused by adjusting the target shelf corresponding to the in-stock commodity cluster.

2. The method according to claim 1, characterized in that, Using a preset particle swarm optimization algorithm and a preset first fitness function, the initial particle population is subjected to particle swarm search iterative processing until the target cargo location allocation code is generated, including: After using the location allocation code as a candidate particle, a first initial particle position and a preset first initial particle velocity are determined for each candidate particle, wherein the first initial particle position is used to characterize the shelf allocation scheme of all the in-stock commodity clusters corresponding to the corresponding location allocation code. According to the preset dynamic adaptive change method, the first inertia factor, the first learning factor and the second learning factor corresponding to the current particle search are determined respectively. The first inertia factor, the first learning factor and the second learning factor are generated by adaptive linear change based on the number of search iterations. Based on the first initial particle position, the first initial particle velocity, the first inertia factor, the first learning factor, and the second learning factor, the particle swarm optimization algorithm is used to perform at least one population search iteration on all the candidate particles until the fitness of each cargo location allocation code corresponding to the generated particle population is not greater than a preset fitness threshold, thereby obtaining a candidate particle population including at least one candidate particle, wherein the fitness of the cargo location allocation code is calculated based on the first fitness function. From the candidate particle population, a target particle is selected, wherein the cargo location allocation code corresponding to the target particle includes the target cargo location allocation code.

3. The method according to claim 2, characterized in that, From the candidate particle population, target particles are selected, including: According to the first fitness function, determine the fitness corresponding to the candidate cargo location allocation code body corresponding to each candidate particle in the candidate particle population; Based on the fitness of the candidate storage location allocation code, the candidate particle with the lowest fitness is selected from the multiple candidate particles in the candidate particle population to obtain the target particle, and the corresponding candidate storage location allocation code is used as the storage location allocation code corresponding to the target particle.

4. The method according to claim 2, characterized in that, Based on the current location planning information, identify the clusters of goods in stock on the target shelves of the automated warehouse, including: The currently acquired storage location planning information is determined, wherein the current storage location planning information includes one of the following: static storage location planning information generated by initial static planning, or historical storage location planning information generated by completing at least one dynamic planning based on the static storage location planning information; Obtain all storage allocation information corresponding to the current storage location planning information, wherein one storage allocation information corresponds to one target shelf; In each of the storage allocation information, the cluster information associated with all the target shelves and the sorting information associated with each cluster information are detected, and the in-stock product clusters stored on the corresponding target shelves are determined according to the cluster information, and the target picking sorting of all the in-stock products corresponding to each in-stock product cluster is determined according to the sorting information.

5. The method according to claim 4, characterized in that, Based on the first sales data, update the historical outbound frequency parameter corresponding to the in-stock goods and the historical cluster outbound frequency parameter corresponding to the in-stock goods cluster, including: Based on the first sales data, determine the outbound frequency parameters of the in-stock goods within a preset period after the current date, and obtain the corresponding new outbound frequency parameters; The mean of all the new outbound frequency parameters is determined to obtain the current cluster outbound frequency parameter, and it is determined whether the change value between the current cluster outbound frequency parameter and the historical cluster outbound frequency parameter is greater than a preset change threshold. The historical cluster outbound frequency parameter is the mean of the corresponding historical outbound frequency parameter in the preset period before the current time. The historical outbound frequency parameter is determined based on the second sales data of the corresponding in-stock goods in the preset period before the current time. If it is determined that the change value is greater than the preset change threshold, the current cluster outbound frequency parameter is updated to the new cluster outbound frequency parameter; If it is determined that the change value is not greater than the preset change threshold, the historical cluster outbound frequency parameter is used as the new cluster outbound frequency parameter.

6. The method according to claim 4, characterized in that, Generating the static storage location planning information includes: The cluster information corresponding to the multiple in-stock product clusters to be stored and the shelf information corresponding to the multiple target shelves are determined. The cluster information is generated by performing correlation statistics on all SKUs of the in-stock products according to a preset statistical method. The shelf information includes multiple target storage locations set on the target shelves. Based on the historical outbound frequency parameters of the corresponding in-stock goods, the historical cluster outbound frequency parameters of the in-stock goods cluster corresponding to each cluster information and the historical cluster correlation degree of any two in-stock goods clusters are determined respectively. Based on the historical outbound frequency parameters, the historical cluster outbound frequency parameters, and the historical cluster correlation, the in-stock product clusters, the target shelves, the in-stock products, and the target storage locations are encoded in a two-stage manner to generate multiple static decision coding particles. The static decision coding particles include product cluster shelf allocation sub-particles with multiple shelf allocation particles and product storage location allocation sub-particles with multiple storage location allocation particles. The product cluster shelf allocation sub-particles are used to represent a shelf allocation scheme for allocating the target shelves to all in-stock product clusters, and the product storage location allocation sub-particles are used to represent a sorting scheme for allocating storage locations to the in-stock products on one of the target shelves. Using the particle swarm optimization algorithm, multiple static decision-encoded particles are subjected to particle swarm search iterative processing until the target decision-encoded particle is obtained. The static storage location planning information includes all the shelf allocation schemes and all the sorting schemes corresponding to the target decision-encoded particle.

7. The method according to claim 6, characterized in that, Using the particle swarm optimization algorithm, a particle swarm search iterative process is performed on multiple static decision-encoded particles until the target decision-encoded particle is obtained, including: Determine the second initial particle position and the preset second initial particle velocity for each static decision-encoded particle, wherein the second initial particle position is used to characterize the shelf allocation scheme and picking sorting allocation scheme corresponding to the corresponding static decision-encoded particle; The second inertial factor, the third learning factor, and the fourth learning factor corresponding to the current particle search are determined respectively, wherein the second inertial factor, the third learning factor, and the fourth learning factor are generated by adaptive linear change based on the number of search iterations; Based on the second initial particle position, the second initial particle velocity, the second inertia factor, the third learning factor, and the fourth learning factor, the particle swarm optimization algorithm is used to perform at least one population search iteration on all the static decision-coded particles until the fitness of each static decision-coded particle corresponding to the generated particle population is not greater than a preset fitness threshold, thereby obtaining at least one candidate decision-coded particle. The fitness of the static decision-coded particle is calculated based on a preset second fitness function, which is constructed based on a picking movement distance function and a picking time function. The picking movement distance function is constructed based on the corresponding cluster outbound frequency parameter and cluster correlation degree, and the picking time function is constructed based on the sorting information of the target storage location corresponding to the in-stock goods. Among at least one of the candidate decision-encoding particles, the candidate decision-encoding particle with the lowest corresponding fitness is selected to obtain the target decision-encoding particle.

8. The method according to claim 6, characterized in that, Before determining the cluster information corresponding to the multiple in-stock product clusters to be stored, the method further includes: One-hot encoding is used to construct a matrix of multiple historical order data and the SKUs of the corresponding products, generating an order-product association matrix with historical order data as rows and SKUs as columns; After generating a co-occurrence matrix based on the order product association matrix, a corresponding product association network is generated according to the product co-occurrence matrix. The product co-occurrence matrix includes multiple product association degrees used to characterize the co-occurrence frequency of two corresponding products in all historical order data. A network graph node of the product association network is used to characterize a corresponding SKU, and a node path of the product association network is used to characterize the association relationship between two corresponding SKUs. A weighted undirected association network corresponding to the product association network is determined. The degree of each network node in the undirected association network and the undirected path between two network nodes are weighted to obtain the product weighted degree corresponding to a corresponding SKU. Based on the product weighted degree, a preset k-shell decomposition analysis method is used to classify all products corresponding to the historical order data to obtain product groups at multiple shell levels. The network node is used to represent a corresponding SKU, and the undirected path is used to represent whether two corresponding SKUs are purchased together. Each shell level corresponds to a weighted degree interval. In each shell-level product group, the product with the highest weighting is selected as the core product. In all products corresponding to all product groups, at least one alternative product with an undirected path to each core product is selected. After selecting a preset number of alternative products from the corresponding at least one alternative product in descending order of product weighting, all selected alternative products and the corresponding core products are clustered into the in-stock product cluster corresponding to the core product. The preset number is determined based on the number of target storage locations on the target shelf.

9. The method according to claim 8, characterized in that, Determining the product association degree between any two of the aforementioned in-stock product clusters includes: Select the first target product cluster and the second target product cluster from the two currently available product clusters; The system sequentially traverses all the in-stock items of the first target product cluster, and in the undirected association network, it determines the product weighting degree between the current product and all the in-stock items of the second target product cluster in each traversal, and sums the product weighting degrees between the current product and all the in-stock items of the second target product cluster in each traversal to obtain the sub-association degree between the current product and the second target product cluster. Sum the sub-associations of all traversed current products and the second target product cluster to obtain the total association between the first target product cluster and the second target product cluster, and use the corresponding total association as the association between any two in-stock product clusters.

10. The method according to claim 1, characterized in that, The sales forecasting model described above includes: The pre-collected commodity transaction data is preprocessed and features are extracted to generate a sample data sequence. The sample data sequence is then divided into a training dataset and a test dataset according to a set ratio. The preprocessing includes one of the following: data completion and data cleaning. The sample data sequence includes the actual sales data of the corresponding commodity and multiple target features for predicting commodity sales volume. After taking the historical combined hyperparameters used for previous model training as the corresponding historical particles, the Particle Swarm Optimization (PSO) algorithm is used to perform at least one population search iteration on the historical particles to generate the current particle and determine the current combined hyperparameters corresponding to the current particle. The historical combined hyperparameters include one of the following: randomly generated initial combined hyperparameters, or combined hyperparameters that have undergone at least one PSO optimization. Both the historical combined hyperparameters and the current combined hyperparameters include the learning rate, the number of hidden layer units, and the dimension of the hidden layer state. After updating the historical combined hyperparameters of the historical LSTM model constructed based on the historical combined hyperparameters to the current combined hyperparameters, the current operation of training the model using the backpropagation algorithm is performed based on the historical LSTM model, the current combined hyperparameters, and the training dataset to generate the current LSTM model. The root mean square error between the current predicted sales data output by testing the current LSTM model using the test dataset and the actual sales data corresponding to the test dataset is used as the hyperparameter fitness of the current combined hyperparameters. Repeat the operation of searching and training the corresponding LSTM model using the Particle Swarm Optimization (PSO) algorithm until the fitness of the generated combined hyperparameters is not greater than a preset fitness threshold to obtain the target combined hyperparameters. Based on the target combined hyperparameters, reconstruct the corresponding target LSTM model, wherein the sales prediction model includes the target LSTM model.

Images