A multi-bin collaborative-oriented intelligent inventory allocation management method and system

Abstract

The application relates to the technical field of warehouse logistics, and discloses an intelligent inventory allocation management method and system for multi-warehouse cooperation, aiming at the problem that a high cross-warehouse order splitting rate is caused by not considering associated purchase and packaging restrictions in multi-warehouse cooperation. The method performs commodity association analysis on the basis of historical transaction records to construct an initial commodity cluster. Physical attribute parameters of independent inventory units in the initial commodity cluster are extracted, physical constraint verification is performed in combination with the upper limit of a logistics package specification, and an effective commodity cluster is determined. Real-time available inventory data is acquired, a physical warehouse inventory barycenter drift index is generated on the basis of the effective commodity cluster complete set inventory, and when the index is greater than a tolerance threshold, a mixed integer linear programming model is constructed to solve a cross-warehouse allocation decision matrix. The application integrates physical packaging constraints and business association into the allocation decision, and improves the overall complete set delivery rate of the multi-warehouse system on the premise of meeting the trunk line transport capacity restriction.

Description

Technical Field

[0001] This invention relates to the field of warehousing and logistics technology, specifically to an intelligent inventory allocation management method and system for multi-warehouse collaboration. Background Technology

[0002] With the development of warehousing networks, multi-warehouse distributed collaborative operations have become a common way to improve fulfillment efficiency. In actual operation scenarios, consumer orders typically contain multiple different products. Existing warehouse management systems mainly use independent forecasting models based on the historical sales of a single inventory unit when planning inventory allocation and cross-warehouse transfers. This method of setting inventory levels on a per-item basis severs the inherent joint purchasing relationships between products in actual transactions. When a consumer purchases multiple strongly related items in the same order, it is highly likely that some items will be out of stock in one physical warehouse, forcing shipments from other warehouses, resulting in the initial order being split into multiple logistics packages.

[0003] Furthermore, some inventory scheduling methods, when attempting to introduce the concept of product bundling to reduce order splitting, rely solely on probability statistics at the data level for logical binding, failing to incorporate the physical packaging constraints of the final logistics fulfillment stage into the allocation decision boundary. When the total volume or weight of the allocated combination output by the system exceeds the carrying capacity limit of existing standard logistics parcels or automated sorting equipment, the hardware equipment at the warehouse site cannot execute the bundling task, and physical order splitting is still required during the outbound operation stage. This inventory management approach, where algorithmic planning and physical execution are not effectively integrated, increases the risk of overload overflow and transportation costs in the trunk transportation stage, and reduces the overall completeness of shipments in a multi-warehouse collaborative network. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an intelligent inventory allocation management method and system for multi-warehouse collaboration. It solves the problems that if inventory allocation schemes are based solely on historical sales of individual products without considering related consumer purchasing behavior and physical logistics packaging limitations, they are prone to high cross-warehouse order splitting rates, oversized logistics parcels, and increased trunk transportation costs.

[0005] To address the above problems, the present invention provides the following technical solution:

[0006] The first aspect of this invention provides an intelligent inventory allocation management method for multi-warehouse collaboration, the method comprising:

[0007] Obtain historical transaction records and perform product association analysis to identify a set of independent inventory units with purchase associations and construct an initial product cluster;

[0008] Extract the physical attribute parameters of each independent inventory unit in the initial product cluster, and perform physical constraint verification in combination with the preset upper limit of logistics package specifications. The initial product cluster that meets the verification conditions is determined as a valid product cluster.

[0009] Obtain real-time available inventory data for each independent inventory unit in each physical warehouse, and generate an inventory center of gravity drift index for each physical warehouse based on the complete inventory of the effective commodity cluster in the corresponding physical warehouse and historical regional demand data.

[0010] By comparing the inventory center of gravity drift index with a preset tolerance threshold, when the inventory center of gravity drift index is greater than the tolerance threshold, a mixed integer linear programming model is constructed with the objective function of minimizing the weighted sum of trunk transportation costs and business deviation penalties, and the cross-warehouse transfer decision matrix is ​​obtained by solving the model.

[0011] Based on the cross-warehouse transfer decision matrix, a structured transfer task message is generated, and the automated warehousing equipment is driven to perform physical material transfer actions.

[0012] Preferably, the step of identifying a set of independent inventory units with purchase associations and constructing an initial product cluster includes:

[0013] Calculate the joint purchase support and association confidence of different independent inventory unit combinations in the transaction dataset;

[0014] When both the joint purchase support and the association confidence are greater than or equal to the preset corresponding thresholds, the corresponding set of independent inventory units is defined as the initial product cluster.

[0015] Based on the demand of each independent inventory unit within the initial product cluster in historical orders, determine the standard demand quantity of each independent inventory unit within the corresponding product cluster.

[0016] Preferably, after determining the valid product clusters, the method further includes the step of establishing a product cluster relationship mapping table, specifically:

[0017] Identify multiple valid product clusters belonging to the same independent inventory unit;

[0018] Based on the association confidence level corresponding to each effective product cluster, the weight allocation function of the independent inventory unit for different effective product clusters is calculated, which is used to perform logical isolation and pre-allocation of available inventory during the allocation planning stage.

[0019] Preferably, the step of combining a preset upper limit for logistics parcel specifications to perform physical constraint verification, and determining the initial product clusters that meet the verification conditions as valid product clusters, includes:

[0020] Extract the three-dimensional dimension data and mass data of each independent inventory unit within the initial product cluster;

[0021] By introducing a preset spatial stacking conversion factor, the theoretical combined volume and total mass of the initial commodity cluster under the theoretical combined packaging state are calculated;

[0022] If the theoretical combined volume is less than or equal to the preset upper limit of logistics package volume, and the total mass is less than or equal to the preset upper limit of load capacity, then the initial product cluster is determined to be a valid product cluster.

[0023] Preferably, the step of generating the inventory center of gravity shift index for each physical warehouse based on the completeness inventory of the effective commodity cluster in the corresponding physical warehouse and historical regional demand data includes:

[0024] Based on the ratio of the real-time available inventory of each independent inventory unit within the effective commodity cluster in a single physical warehouse to the standard demand quantity, the minimum value of the ratio is taken as the complete inventory of the effective commodity cluster in the physical warehouse.

[0025] Based on the completeness distribution of inventory for each effective commodity cluster in each physical warehouse, the absolute value of the difference between the actual inventory distribution ratio and the historical demand distribution ratio is calculated to generate the inventory center of gravity drift index for the corresponding physical warehouse.

[0026] Preferably, the constraints of the mixed-integer linear programming model include:

[0027] Trunk route logical constraints: By summing the product of the physical volume of a single item and the quantity allocated, the total volume is limited to not exceeding the maximum cargo capacity of the transport vehicle, and this constraint is associated with a Boolean variable for the route.

[0028] Structural kitting constraint: Introduce non-negative integer variables to represent the expected kitting inventory after allocation, and establish a linear inequality relationship between the expected kitting inventory and the inventory level after allocation of each independent inventory unit in the cluster.

[0029] Matching Gap Penalty Boundary Constraint: Establish a difference constraint between the expected matching gap quantity and the expected matching stock and historical regional demand benchmark to activate the business deviation penalty term in the objective function.

[0030] Preferably, the step of driving the automated warehousing equipment to perform physical material transfer operations includes:

[0031] The cross-warehouse transfer decision matrix is ​​converted into a structured transfer task message supported by the underlying logistics control system.

[0032] The structured allocation task message is written into a preset instruction interaction interface path, which is then polled and read by the warehouse management system to generate hardware control instructions.

[0033] Preferably, before the warehouse management system polls, reads, and generates hardware control instructions, it further includes:

[0034] Determine whether the current items to be picked exceed the limits based on the maximum volume and load capacity of the standard turnover container;

[0035] If the limit is exceeded, the current materials to be picked will be forcibly divided into multiple sub-batches for picking, and the packages will be combined according to the effective commodity cluster relationship during the outbound verification process.

[0036] A second aspect of the present invention provides an intelligent inventory allocation management system for multi-warehouse collaboration, the system comprising:

[0037] The data mining module is used to acquire historical transaction records and perform product association analysis to build initial product clusters;

[0038] The physical decoupling module is used to extract physical attribute parameters and verify them in conjunction with the upper limit of logistics package specifications to determine the valid product clusters;

[0039] The status monitoring module is used to obtain real-time available inventory data and generate the inventory center of gravity drift index for each physical warehouse.

[0040] The operations planning module is used to compare the inventory center of gravity drift index with the tolerance threshold, and to construct and solve the mixed integer linear programming model to output the cross-warehouse transfer decision matrix.

[0041] The instruction issuance module is used to generate structured allocation task messages and drive automated warehousing equipment to perform physical material allocation.

[0042] Preferably, the data mining module is further used for:

[0043] Calculate the joint purchase support and association confidence of different independent inventory unit combinations in the transaction dataset;

[0044] When both the joint purchase support and the association confidence are greater than or equal to the preset corresponding thresholds, the corresponding set of independent inventory units is defined as the initial product cluster.

[0045] Based on the demand of each independent inventory unit within the initial product cluster in historical orders, determine the standard demand quantity of each independent inventory unit within the corresponding product cluster.

[0046] This invention provides an intelligent inventory allocation management method and system for multi-warehouse collaboration. It offers the following advantages:

[0047] 1. This invention extracts the physical attribute parameters of each independent inventory unit in the initial product cluster and performs physical constraint verification by combining the volume and weight limit of logistics packages. This combines data-level correlation analysis with physical-level packaging restrictions. This feature eliminates inventory unit combinations that have purchase relevance but cannot be physically combined into packages, avoiding passive order splitting operations caused by package over-limit in subsequent fulfillment processes, and improving the physical executability of inventory transfer instructions at automated sorting terminals.

[0048] 2. This invention calculates completeness inventory based on the real-time available inventory and standard demand quantity of each independent inventory unit within a single physical warehouse within an effective product cluster, and generates an inventory center of gravity drift index by combining historical regional demand data. This feature transforms traditional single-item independent inventory monitoring into a completeness fulfillment capability assessment for related orders, directly quantifying structural deviations in multi-warehouse inventory distribution, thereby identifying in advance the business risk of overall order non-delivery due to shortages of some related items. Attached Figure Description

[0049] Figure 1 This is a structural block diagram of an intelligent inventory allocation management system for multi-warehouse collaboration according to an embodiment of the present invention;

[0050] Figure 2 This is a flowchart of an intelligent inventory allocation management method for multi-warehouse collaboration according to an embodiment of the present invention;

[0051] Figure 3 This is a flowchart illustrating the dynamic decoupling of product cluster validity based on physical packaging constraints according to an embodiment of the present invention.

[0052] Figure 4 This is a flowchart of the commodity cluster kitting status assessment and inventory center of gravity shift index calculation in an embodiment of the present invention;

[0053] Figure 5 This is a flowchart of the collaborative allocation process of mixed integer linear programming based on a hierarchical penalty mechanism according to an embodiment of the present invention.

[0054] Figure 6 This is a flowchart illustrating the instruction parsing and physical warehousing operation issuance process according to an embodiment of the present invention.

[0055] Figure 7 This is a directed graph of the flow direction and load of the multi-warehouse collaborative allocation network in an embodiment of the present invention;

[0056] Figure 8 This is a graph showing the convergence curve of objective function optimization iteration and matching gap descent in an embodiment of the present invention. Detailed Implementation

[0057] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0058] See attached document Figure 1 , Figure 1 This is a structural block diagram of an intelligent inventory allocation management system for multi-warehouse collaboration according to an embodiment of the present invention. The present invention provides an intelligent inventory allocation management system for multi-warehouse collaboration, including a data mining module, a physical decoupling module, a status monitoring module, an operations planning module, and an instruction issuance module. The intelligent inventory allocation management system for multi-warehouse collaboration is deployed in a central control server. The central control server establishes data communication with the order management system and the warehouse management systems of multiple physical warehouses distributed in different geographical locations through the enterprise's internal network. Data interaction between various modules within the system and communication with external systems are all performed through file read / write via preset data interaction paths.

[0059] See attached document Figure 2 , Figure 2 This is a flowchart of an intelligent inventory allocation management method for multi-warehouse collaboration according to an embodiment of the present invention. The present invention provides an intelligent inventory allocation management method for multi-warehouse collaboration, comprising the following steps:

[0060] S10: Access the preset historical data storage path to read historical transaction details records according to the set time period, execute the association rule mining algorithm on the extracted records, calculate the joint purchase support and confidence of each independent inventory unit, and when the joint purchase support and confidence of the independent inventory unit combination reaches the set threshold, define the corresponding independent inventory unit set as the initial product cluster.

[0061] S20: Access the preset warehouse master data storage path to read the three-dimensional dimension data and mass data of each independent inventory unit. Calculate the theoretical combined packaging volume and total mass of the initial product cluster according to the space stacking conversion factor. Compare the theoretical combined packaging volume and total mass with the preset upper limit of logistics parcel volume and upper limit of load capacity. If the theoretical combined packaging volume or total mass of the initial product cluster exceeds the corresponding upper limit, remove it from the planning queue. Output the initial product clusters that do not exceed the upper limit as valid product clusters.

[0062] S30: Obtain real-time available inventory data for each physical warehouse. For effective product clusters, calculate the complete inventory of the effective product cluster in a single physical warehouse based on the minimum inventory ratio of each independent inventory unit within the cluster. Based on the complete inventory of the effective product cluster and historical regional demand data, calculate the absolute value of the difference between the actual inventory distribution ratio and the demand distribution ratio, and generate the inventory center of gravity drift index of the corresponding physical warehouse.

[0063] S40: Compare the inventory center of gravity drift index of each physical warehouse with the preset tolerance threshold. When the inventory center of gravity drift index is greater than the preset tolerance threshold, access the preset configuration parameter storage path to read the business penalty weight parameter and the remaining available warehouse capacity parameter of each physical warehouse, and construct a mixed integer linear programming model. The model takes the minimization of the weighted sum of trunk transportation cost and business deviation penalty as the objective function, and combines inventory conservation constraints, structural matching constraints, index correction constraints and physical three-dimensional warehouse capacity constraints to perform mathematical solutions and output the cross-warehouse transfer decision matrix.

[0064] S50 receives the cross-warehouse transfer decision matrix, converts the cross-warehouse transfer decision matrix into a structured transfer task message supported by the underlying logistics control system, writes the structured transfer task message into a preset instruction interaction interface path, and the warehouse management system polls and reads the structured transfer task message in the preset instruction interaction interface path to generate hardware control instructions that drive automated warehousing equipment to perform physical material outbound, cross-warehouse trunk line transportation and inbound actions.

[0065] In this embodiment, the present invention provides an intelligent inventory allocation management method for multi-warehouse collaboration, and the steps of historical transaction data mining and commodity cluster topology construction involved in the method are described in detail with reference to the accompanying drawings.

[0066] The specific steps involved in mining historical transaction data and constructing commodity cluster topologies include:

[0067] S101, the central control server retrieves transaction records from a preset historical data storage path via a data access interface. As a preferred approach, considering the spatiotemporal differences in a multi-warehouse collaborative environment, the system first performs multi-source data alignment, converting transaction times from shipping points distributed across different geographical locations and time zones into Coordinated Universal Time (UTC), and normalizing operational data at the daily level. Transaction records include a unique order identifier, an independent inventory unit code, transaction time, and corresponding physical warehouse shipping point information. The central control server preprocesses the transaction records, removing abnormal entries with missing key fields, and aggregates the independent inventory unit codes based on the unique order identifier, generating a transaction dataset categorized by order. This order-based aggregation process aims to transform discrete sales records into data clusters reflecting the correlation of consumer purchasing behavior. For data cleaning and deduplication, those skilled in the art can employ existing data preprocessing algorithms; the specific processing steps are well-known in the field and will not be elaborated upon here.

[0068] S102, Execute an association rule mining algorithm on the transaction dataset to calculate the joint distribution characteristics between different combinations of independent inventory units. This step is based on the law of large numbers and uses statistical methods to mine the co-occurrence probability between products. During this process, the system identifies sets of independent inventory units that frequently co-occur in the same order and counts the frequency of these sets in the global transaction dataset. The specific calculation logic is as follows: ;

[0069] in, Indicates independent inventory units The initial commodity cluster formed Support level; This indicates the absolute frequency of the occurrence of this set of independent inventory units in the transaction dataset; This represents the total number of order samples in the transaction dataset. The formula incorporates a non-zero denominator check to ensure that the algorithm logic will not be interrupted by a division-by-zero exception during system initialization or in extreme cases where there is no order data.

[0070] S103 calculates the association confidence level between each independent inventory unit within the initial product cluster to quantify the purchase certainty of the product combination. Technically, confidence level represents the conditional probability that one product will appear given the existence of another. The formula for calculating confidence level is as follows:

[0071] ;

[0072] in, Indicates that the purchase of independent inventory units is known. At the same time, purchase the initial product cluster. The correlation strength index of other members; This represents the total number of times all members of the initial product cluster appear together in the same order. As an independent inventory unit The total frequency of occurrences in the transaction dataset. The physical meaning of this metric is to identify the core dominant product within a product cluster, providing a sovereign reference for subsequent inventory completeness assessments.

[0073] S104, the central control server accesses the preset configuration parameter storage path and extracts the support threshold. With confidence threshold The system compares the calculated support and confidence scores with their corresponding thresholds. In this embodiment, the support threshold is... The value range is set to [0.01, 0.05], which is based on balancing the system's computing power load and correlation significance, and filtering low-frequency random combinations; confidence threshold. The value range is set to [0.5, 0.8] to ensure that the output product clusters have strong causal logic in consumption. The initial product cluster is retained only if the support of the initial product cluster is greater than or equal to the preset support threshold, and the confidence of the association between its internal members is greater than or equal to the preset confidence threshold.

[0074] S105, construct a product cluster topology based on the set of independent inventory units filtered by a threshold. This topology is represented in system memory as a many-to-many mapping model. In this embodiment, the topology further includes the standard demand quantity of each independent inventory unit within the cluster. The parameter is calculated by extracting the average demand for this item within the corresponding product cluster orders from historical orders and then rounding it up. The database stores the matching relationships of each independent inventory unit within the product cluster. The product cluster topology serves as the basic logical unit for subsequent inventory completeness assessment and is persistently stored in the database.

[0075] S106, Generate a product cluster relationship mapping table and output it to the cache layer. The mapping table records the product cluster set to which each independent inventory unit belongs. Considering that an independent inventory unit may exist in multiple valid product clusters simultaneously, this embodiment introduces a weight allocation function. This is to resolve inventory competition conflicts. The weighting is based on a weighted proportion of association confidence levels, and the calculation logic is as follows: ;

[0076] in, For including independent inventory units The system comprises all valid product clusters. Through weight allocation, the system can prioritize the completeness requirements of high-weight product clusters during the allocation planning stage, avoiding decision-making conflicts caused by multiple logical clusters competing for a single product, thereby achieving logical isolation and pre-allocation. Logical isolation and pre-allocation refer to the following: during the allocation planning stage, the system logically divides and locks the available inventory of the same independent inventory unit in each physical warehouse according to its different valid product clusters based on the weight allocation function; the inventory allocated to a specific product cluster is only used to meet the completeness requirements of that product cluster in this round of allocation decisions and cannot be preempted by allocation decisions of other product clusters. This pre-allocation state persists throughout the entire mathematical programming solution process until the current allocation decision matrix is ​​completed or timed out. Regarding the underlying iterative optimization of the association rule algorithm, such as using a hash tree structure to accelerate frequent itemset searches, those skilled in the art can choose appropriate mature solutions based on actual computing power requirements. The specific algorithm structure is a well-known technology in this field and will not be elaborated here.

[0077] For each independent inventory unit s, there is a valid cluster of goods. The weighting coefficients. This indicates that when it is known that an independent inventory unit s is purchased, a product cluster is also purchased. The association confidence of all other members. For the k-th valid product cluster, For the j-th valid product cluster, This indicates that when it is known that an independent inventory unit s is purchased, a product cluster is also purchased. The association confidence of all other members. The normalization factor represents the number of individual inventory units (s) purchased simultaneously with a product cluster. The sum of the association confidence scores of all other members.

[0078] In this embodiment, for the initial product clusters generated through association rule mining, the present invention provides a dynamic decoupling method for product cluster validity based on physical packaging constraints. This method aims to eliminate invalid product clusters that are logically strongly correlated but cannot be delivered as single packages at the physical fulfillment level through hard filtering at the physical dimension, thereby ensuring the executability of subsequent allocation instructions in warehousing operations.

[0079] See attached document Figure 3 , Figure 3 This is a flowchart illustrating the dynamic decoupling of product cluster validity based on physical packaging constraints according to an embodiment of the present invention. Specifically, it may include the following steps:

[0080] S201, the central control server accesses the preset warehouse master data storage path through the data retrieval interface and extracts the physical attribute parameters that match each independent inventory unit within the initial product cluster. In this embodiment, these parameters constitute the physical envelope characteristics of the inventory unit, and the physical attribute parameters include at least the length of each independent inventory unit. ,Width ,high and the quality of individual items Specifically, for irregularly shaped independent inventory units, their length, width, and height parameters are taken as the dimensions of their smallest three-dimensional bounding rectangle in their natural flat state. Considering the data differences between multi-source heterogeneous systems, the system performs unit standardization after reading the parameters, converting length units to millimeters (mm) and mass units to grams (g). This unification of dimensions is a logical prerequisite for establishing the physical calculation model, aiming to eliminate calculation deviations caused by different measurement habits in cross-regional warehouse management systems. For inventory units lacking physical parameters in the master data, the system automatically triggers an abnormal data completion mechanism. The specific data completion process is a well-known technique in this field and will not be elaborated here.

[0081] S202, based on the physical parameters of each member within the product cluster, calculate the total volume and total mass of the product cluster under the theoretical combined packaging state. Considering the irregular rectangular gaps caused by irregular shapes and the space occupied by shock-absorbing filling materials during the actual boxing process of the storage unit, this embodiment introduces a space stacking conversion factor. The total volume is corrected. This correction logic is based on the principle of spatial geometric envelope, aiming to convert the sum of the volumes of discrete individual items into the pre-occupied space value of a physical package. The calculation formula is as follows:

[0082] ;

[0083] ;

[0084] In the above formula, Represents the initial product cluster Theoretical combined volume; , , These represent independent inventory units within the cluster. Length, width, and height; Represents the initial product cluster Total mass; As an independent inventory unit The quality; This is the spatial stacking conversion factor. In this embodiment, The value range is strictly set within the interval (0,1). Based on statistical analysis of past packaging material filling rates, Typically, a value between [0.6, 0.85] is selected. As a preferred implementation, when configuration parameters are missing... When the value approaches or equals 0, the system automatically corrects it to the default value of 1.0 to avoid division by zero. The preferred value range [0.6, 0.85] is an empirical range derived from statistical experiments on the filling rate of various common packaging consumables (such as corrugated boxes, bubble bags, and foam granules). Those skilled in the art can adjust this range according to the specifications of the standard turnover box used and the characteristics of the filling material. The reason is that the total physical volume of individual items is often less than the actual boxed volume. By adjusting this coefficient, it is possible to effectively prevent secondary order splitting caused by package overflow in subsequent processes.

[0085] S203, the system accesses the configuration parameter storage path to obtain the maximum parcel size parameters that can be handled in the current logistics process. These parameters include the preset maximum parcel volume. With upper limit of load capacity These upper limits are not set randomly, but are physical boundaries determined by the mechanical strength of the levers of the automated sorting line in the downstream logistics chain, the rated power of the conveyor belt motor, and the standard cargo volume of the terminal delivery vehicle.

[0086] S204, execute the product cluster validity determination logic, identify and process initial product clusters that do not meet physical constraints. The system will calculate... and Each item is compared to its corresponding upper limit. The judgment process uses a logical OR operation; if either the volume or weight exceeds the limit, the item cluster is deemed inoperable under the current logistics hardware environment. The judgment criteria are as follows:

[0087] ;

[0088] in, This is a Boolean decision variable, with a value of 1 indicating invalidity. Based on the above decision, the system performs a dynamic decoupling operation, marking the initial product cluster as an invalid combination and removing it from the subsequent collaborative planning queue. The technical purpose of this step is to proactively intercept hardware execution layer anomalies caused by excessive aggregation of logical associations. For the decoupled inventory units, the system reverts them to independent single items and automatically constructs them into single-item logical clusters for inclusion in the subsequent allocation planning model. In the subsequent S403 step, the association confidence of this special product cluster is... Assigned a preset baseline value (such as 1.0), the completeness gap penalty is simplified to a business penalty for single item shortages, thereby ensuring that the system can still take into account the allocation needs of bulk single items during the collaborative planning phase and avoid decision omissions.

[0089] S205, For the initial product cluster with a judgment result of 0, the system confirms it as a valid product cluster and stores it in the set of valid product clusters. In this embodiment, a valid product cluster is considered the smallest business decision unit that satisfies both consumer-related purchasing characteristics and the physical operational limits of the warehouse. As a preferred approach, the system generates a feature vector for each valid product cluster, containing both volumetric load rate and weight load rate. These feature vectors are passed as core state variables to subsequent stages for accurate calculation of physical warehouse capacity occupancy.

[0090] Based on the aforementioned filtering mechanism constrained by physical packaging, this embodiment introduces three-dimensional spatial and weight limitations on entities, achieving dimensionality reduction at the logical level in the early stages of algorithm decision-making. This dynamic decoupling process solves the problem of increased order splitting rates that traditional algorithms easily cause when handling bulk commodities or overweight combinations, ensuring that each generated allocation suggestion corresponds to a combination of entities that can be physically contained in a single standard package. In this embodiment, this judgment logic based on physical causality has higher engineering reliability than judgment methods that simply rely on data probability, providing physically authentic input data for subsequent allocation models centered on completeness, thus fundamentally supporting the technical solution for improving inventory allocation accuracy as stated in the claims.

[0091] See attached document Figure 4 , Figure 4 This is a flowchart illustrating the assessment of commodity cluster completeness and the calculation of inventory center of gravity drift index according to an embodiment of the present invention. In this embodiment, based on effective commodity clusters selected through physical decoupling, the present invention provides a method for assessing commodity cluster completeness and calculating the inventory center of gravity drift index. This method aims to quantitatively assess the degree of deviation between the inventory distribution in each physical warehouse and the actual demand distribution, providing quantitative numerical indicators to support subsequent operational allocation. Specifically, it may include the following steps:

[0092] S301, the central control server obtains real-time snapshot data of available inventory from multiple physical warehouses through the system's internal data bus interface. Considering the differences in data reporting frequencies among the underlying warehouse management systems of each physical warehouse, a timestamp alignment and distributed locking mechanism are introduced as a preferred approach. The system freezes the inventory change status of each warehouse under a unified Coordinated Universal Time (UTC), generating a globally consistent available inventory matrix. In this embodiment, to ensure that the extracted inventory has genuine physical transferability attributes, the system performs a deduction logic on the underlying total physical inventory when generating the snapshot. Specifically, the available inventory... The calculation excluded frozen inventory under quality inspection or damaged conditions, as well as pre-held inventory locked by front-end orders but not yet packaged and shipped. This matrix records the real-time available inventory of each independent unit within each valid product cluster in a specific physical warehouse. The technical purpose of extracting this globally consistent snapshot is to eliminate inventory read deviations caused by data transmission delays or state asynchronization, ensuring a unified data benchmark for cross-warehouse state assessment. For the concurrent read and locking mechanisms of the underlying inventory database, those skilled in the art can employ conventional distributed database transaction processing schemes; the concurrency control methods are well-known in the field and will not be elaborated upon here.

[0093] S302, for each product cluster in the effective product cluster set, the system calculates its completeness inventory within a single physical warehouse. This step is based on the shortest-board constraint principle in operations research. In this embodiment, the completeness inventory of a product cluster depends on the minimum proportion of each independent inventory unit within the cluster. This calculation logic essentially transforms the discrete single-item inventory dimension into a set indicator based on completeness fulfillment capability. The specific calculation formula is as follows: ;

[0094] in, Indicates a valid product cluster In the physical warehouse Complete set of inventory within; Indicates independent inventory unit In the physical warehouse The absolute amount of real-time available inventory; This indicates that the single item is assigned to the product cluster. The weighted proportion; Indicates independent inventory unit In the effective commodity cluster The standard required ratio quantity; To find the minimum value of the function; This indicates a rounding down operation. In data processing logic, because... Derived from historical real orders and with a minimum value of 1, this naturally avoids calculation anomalies caused by a denominator of zero. The physical meaning of introducing the floor function is that any non-integer half-set of goods cannot constitute a complete logistics delivery unit; the system only recognizes the integer quantity of sets sufficient to physically complete the package. This strict limitation of the floor feature ensures that the complete set inventory parameter strictly corresponds to the actual physical shippable status.

[0095] S303, based on the complete inventory in a single warehouse, the system calculates the actual inventory distribution ratio and historical regional demand distribution ratio of this effective product cluster in a multi-warehouse environment across the entire network. Actual Inventory Distribution Ratio This reflects the current absolute distribution of resources in physical space. Historical regional demand distribution ratio. This represents the expected shipping pressure in the regions where each warehouse is located, mapped from past consumer behavior. The corresponding calculation logic is as follows:

[0096] ;

[0097] ;

[0098] In the above formula, This represents the total number of physical warehouses in the network topology; This refers to the index tags for physical warehouses. As a preferred method, historical total demand... Instead of using unlimited historical data, the system extracts the cumulative order value from a specific time window (e.g., the past 30 calendar days) preceding the current point in time. The rationale for introducing the time window parameter is that the demand distribution in the consumer market has significant time sensitivity. Extracting recent operational data effectively filters out long-tail interference caused by outdated data, thus making the calculated demand distribution ratio closer to the current actual supply and demand structure. Both formulas include pre-defined logic for determining if the denominator is non-zero. When the total inventory or total demand across the entire network is zero, the system safely assigns the distribution ratio to zero, avoiding division by zero errors that could cause system crashes. This joint calculation of the distribution ratio elevates the system's decision-making basis from simply focusing on the absolute inventory of a single warehouse to a relative topological analysis of the entire network's supply and demand structure.

[0099] S304, Based on the aforementioned actual inventory distribution ratio and demand distribution ratio, the system further calculates the inventory center-of-gravity drift index for each physical warehouse. In this embodiment, the inventory center-of-gravity drift index is defined as the absolute value of the difference between the actual inventory distribution and the expected demand distribution. This index, as a core quantitative feedback indicator, is used to intuitively characterize the degree to which the current physical inventory distribution deviates from the ideal network structure. Its calculation formula is as follows: ;

[0100] in, Indicates a valid product cluster In the physical warehouse The system uses an inventory center of gravity drift index. The absolute value sign is used to eliminate interference from positive and negative signs of the deviation direction (overstock or shortage). The system uses scalar characteristics to uniformly measure the objective magnitude of the mismatch. The theoretical range of this index is between [0,1]. When the index approaches 0, it indicates a high degree of matching between the physical reserves of the warehouse and its regional delivery tasks. Conversely, when the index increases, it means that there is a significant supply-demand mismatch in the warehouse, manifested as regional inventory backlog or the business risk of potential stockouts or supply disruptions. Based on this, the system does not rely solely on extreme alarms for a single warehouse for scheduling. Instead, it compiles the inventory center of gravity drift indices of all physical warehouses and their corresponding commodity clusters into a status monitoring matrix. This matrix is ​​then passed to the subsequent operations planning module as a pre-verification and weight input parameter for initiating the global allocation model, effectively avoiding inefficient and high-frequency allocations caused by local threshold alarms.

[0101] See attached document Figure 5 , Figure 5 This is a flowchart of a mixed-integer linear programming (MILP) collaborative allocation process based on a hierarchical penalty mechanism, according to an embodiment of the present invention. In this embodiment, based on the commodity cluster fitting status monitoring matrix produced in the preceding steps, the present invention transforms discrete logistics allocation decisions into a mathematical optimization problem constrained by physical boundaries by constructing a mixed-integer linear programming (MILP) model. Specifically, it may include the following steps:

[0102] S401, the system constructs a model-triggered judgment mechanism based on the inventory center-of-gravity drift index of each physical warehouse recorded in the status monitoring matrix. The system calculates the inventory center-of-gravity drift index for each physical warehouse for a specific valid product cluster. With the preset transfer trigger threshold A comparison was conducted. Considering the economies of scale in trunk logistics transportation, The value is typically set within the range of [0.15, 0.30]. This threshold is determined to balance the fixed logistics costs of a single allocation with the potential order losses caused by inventory mismatch. Based on the above comparison, when any drift index in the monitoring matrix exceeds the trigger threshold, the system activates the mixed-integer linear programming solver. To achieve optimal allocation of network resources while ensuring local demand, the system extracts all valid product clusters and corresponding physical warehouse node states that are logically related to the trigger warehouse across the entire network, and incorporates them into a unified decision matrix for global collaborative optimization, rather than performing local allocations only for alarm nodes. This eliminates local deviations through dynamic reorganization of the global inventory center of gravity.

[0103] In step S402, the central control server initializes the decision variables of the mathematical programming model in the memory pool. This process aims to map the flow relationships of physical entities into an algebraic expression. Specifically, the system defines non-negative integer decision variables. This is used to indicate the shipment originating from the warehouse within the current allocation cycle. Transfer to destination warehouse Independent inventory units The quantity allocated; Boolean decision variables defined simultaneously. Used to characterize physical warehouses and Is there a substantial allocation of trunk line capacity between them? The technical purpose of introducing Boolean variables is to accurately model the fixed costs of transportation routes and avoid generating overly fragmented and uneconomical allocation paths.

[0104] S403 constructs a model objective function with an embedded hierarchical penalty mechanism to achieve the optimal balance between overall network cost and kitting level. The penalty factor transforms qualitative business priorities into quantitative cost items. The objective function is expressed as follows:

[0105] ;

[0106] In the above formula, The objective cost of solving the system; For physical warehouse to Fixed start-up costs for the route; For unit inventory unit Variable transportation costs; A collection of inventory units awaiting allocation; This represents the set of valid product clusters. The third term on the right side of the formula is the innovation penalty term of this invention, wherein... For the destination warehouse after allocation Targeting effective product clusters The expected number of matching gaps. As a preferred approach, the penalty coefficient... The value is determined using weighted logic and is equal to the historical average gross profit of the members within the product cluster and the correlation confidence level obtained in step S103. The product of these parameters. The reason for choosing this composite parameter is that it establishes a causal correlation between physical allocation and commercial value, as well as the probability of consumption synergy, ensuring that the system prioritizes the complete supply of high-contribution, strongly correlated commodity clusters when resource allocation is limited.

[0107] S404 configures a constraint matrix to ensure that the allocation plan conforms to objective physical laws. Based on modeling the actual operational limits of the warehousing environment, the system sets at least the following five constraints:

[0108] Firstly, the logical constraints of the trunk line: ;

[0109] in, This represents the maximum cargo capacity of current trunk line transport vehicles. For a single item The physical volume. By introducing physical parameters, this formula replaces the purely mathematical maximum constant with a capacity constraint that has real physical meaning, strictly guaranteeing that the capacity is only determined when the channel Boolean variable... When the value is 1 (i.e., actual vehicle dispatch), the total volume of physical displacement of the corresponding route is allowed to be greater than 0 and not exceed the vehicle volume limit, thus locking in the relationship between fixed cost and flow rate under the premise of complete unit uniformity.

[0110] Secondly, the destination warehouse receiving capacity constraints: ;

[0111] in, For a single item The physical volume; For the purpose of storage Total available storage capacity; This represents the current occupied space in the warehouse. The physical meaning of this constraint is that the total amount of all transferred materials must not exceed the remaining physical carrying capacity of the destination warehouse, thus avoiding the risk of congestion at the receiving site due to overloaded inventory.

[0112] Third, the flow conservation constraint: ;

[0113] in, and Physical warehouses before and after the allocation execution The available inventory level. This equation ensures the conservation of physical quality of goods throughout the entire network during their circulation.

[0114] Fourth, the linearization transformation of nonlinear logic into a single constraint:

[0115] Considering that the objective function includes the expected number of matching gaps This gap is derived from a nonlinear homogenized inventory that includes minimization and floor operations. To meet the algebraic input specifications of the Mixed Integer Linear Programming (MILP) solver engine, auxiliary non-negative integer variables are introduced internally. The complete set of stock can be equivalently expressed by the following system of linear inequalities:

[0116] ;

[0117] in: Represents commodity clusters After the allocation was executed, in the physical warehouse The expected matching stock within the range is defined as a non-negative integer variable; Indicates independent inventory unit In the physical warehouse The initial available inventory; This indicates the available inventory competition weight assigned to the product cluster via step S106; This indicates that from other physical warehouses across the entire network Transferred to destination warehouse The total quantity of this single item; Indicates from this physical warehouse The total quantity of this item transferred to other warehouses; Indicates independent inventory unit In product clusters The standard required ratio quantity.

[0118] In the above constraints, due to Defined as a non-negative integer variable, driven by the objective function's minimization of the penalty term (or maximization of combined returns), it spontaneously approaches the upper bound (i.e., the minimum available proportion in the "barrel effect") formed by the inventory proportions of each item within the cluster. Simultaneously, due to the integer constraint, this variable mathematically achieves the physical semantics of rounding down. This transformation avoids directly using non-linear logical operators in the model, eliminates logical gaps in the objective function, and transforms the problem into a standard mixed-integer programming form, thus ensuring the solvability of the operations research mathematical model in the engineering solution process.

[0119] Fifth, the boundary constraint for the fitting gap penalty: considering that the objective function requires a penalty for the expected number of fitting gaps. To minimize the penalty, the system must set a compliant mathematical lower bound to prevent the solver from directly setting it to zero. The system incorporates the historical regional demand distribution benchmark extracted in step S303. Construct the following linear inequality:

[0120] ;

[0121] ;

[0122] Driven by both the aforementioned constraints and the minimization of the objective function, It can be correctly activated and accurately represent the actual number of matching gaps in the destination warehouse after allocation.

[0123] S405, the central control server loads the above objective function and constraint matrix, and calls the preset mixed integer programming solver engine to perform the calculation. For the planning problem, the system uses a branch and bound algorithm for iterative optimization. The solver ultimately outputs the optimal decision matrix. The system automatically converts these into corresponding logistics operation instructions, including picking task sheets, loading lists, and inbound pre-notification forms. Through this direct mapping from abstract algebraic models to concrete physical instructions, this embodiment achieves dynamic correction of the inventory center of gravity in a multi-warehouse collaborative environment, significantly improving the matching level of effective commodity clusters across all nodes in the network while reducing total transportation costs.

[0124] See attached document Figure 6 , Figure 6 This is a flowchart illustrating the instruction parsing and physical warehousing operation issuance process according to an embodiment of the present invention. In this embodiment, based on the optimal decision variable matrix output by an operations research model, the present invention provides a method for instruction parsing and physical warehousing operation issuance. This method converts the optimization solution in pure mathematical space into control instructions that can be directly executed by the underlying warehousing equipment and transport vehicles, thus realizing the implementation of model decisions into physical operations. Specifically, it may include the following steps:

[0125] S501, the central control server receives the optimal decision matrix output by the mixed-integer linear programming solver and initiates the data inverse analysis program. Because the multi-warehouse collaborative model exhibits highly sparse matrix characteristics in actual computation, the system extracts all decision elements with values ​​greater than zero through a traversal operation, constructing a set of executable tasks. The general technical principle of this operation lies in filtering nodes in a static state in the network topology, accurately locating the storage units that need to undergo physical displacement and their corresponding routes. Its extraction expression is as follows: ;

[0126] In the above formula, This represents the set of entity transfer tasks that need to be executed within the current allocation cycle; The solver output regarding the originating warehouse destination warehouse and independent inventory units The optimal number of entities to be transferred is determined. Based on the above extraction results, the system groups the task set into first-level groups according to the physical code of the originating warehouse, so as to issue dedicated work orders to physical warehousing systems in different geographical locations.

[0127] S502, for the grouped tasks of each originating warehouse, the system generates physical outbound batches and picking work orders. In this embodiment, to ensure the physical load safety of automated guided vehicles (AGVs) or manual picking carts within the warehouse, the system needs to calculate the total volume and total weight of a single task batch before generating the work order to establish a reasonable work segmentation boundary. Originating Warehouse To the destination warehouse The calculation logic for the total physical load of the batch of materials sent is as follows:

[0128] ;

[0129] ;

[0130] in, and These represent the estimated combined total volume and total mass of the batch of materials, respectively. and Each represents an independent inventory unit The standard physical volume and standard mass of a single item; This is a set of subtasks targeting specific origin and destination nodes. After calculating the total load, the system needs to further compare it with the upper limit of the physical properties of standard turnover boxes in the warehouse and perform a segmentation determination.

[0131] ;

[0132] In the above determination formula, This is the work order segmentation flag. and These correspond to the maximum volume and maximum load capacity of the standard turnover containers used in the warehouse, respectively. When this flag is 1, the system, based on the above multi-dimensional hard over-limit results, forcibly divides the batch into multiple sub-batches according to the descending first-fit packing algorithm. This work order segmentation control based on the underlying equipment physical load-bearing constraints effectively prevents equipment downtime or physical damage caused by overloading during on-site operations.

[0133] S503, the central control server issues trunk line loading plans to the transportation management system based on grouped tasks. The economic efficiency of trunk line transportation is highly dependent on the utilization rate of vehicle space; therefore, the system needs to calculate the expected loading rate indicator before issuing departure instructions. This indicator integrates both quality and spatial dimensions, and its calculation formula is as follows: ;

[0134] In this formula, and These are the maximum rated volume and maximum rated load of the vehicles assigned to the trunk transportation dispatching system, respectively. This is a function for maximizing the value. As a preferred approach, to address potential gaps in vehicle master data maintenance in multi-source heterogeneous systems, the system pre-configures a denominator fault-tolerant logic: if a fault is detected... or If the value approaches 0, the system will automatically apply the default parameters for a 9.6-meter standard van to ensure that the underlying division operation does not throw an error. The system will further adjust the loading rate... Compared with the preset economic departure threshold A comparison is performed. In this embodiment, The value is set between [0.4, 0.6], and the specific value is dynamically calculated based on the current route's fuel rate. When the load factor is lower than this threshold, the system triggers a capacity protection mechanism, partially suspending the allocation task to wait for the order merging in the next cycle; otherwise, a formal transportation dispatch order is generated.

[0135] S504, the system sends a message to the destination physical warehouse. Inbound advance notification forms are issued. These forms include details of the expected arriving goods, quantities, and estimated arrival timestamps dynamically calculated based on vehicle GPS trajectories. This step aims to drive the destination warehouse's storage allocation module to run in advance, locking and organizing available storage locations to absorb peak throughput at the receiving platform.

[0136] S505 executes a phased state transition update of the entire network's inventory state machine. In a distributed warehousing network, data consistency and physical operation progress must maintain strict transaction alignment. To prevent subsequent orders from repeatedly occupying materials that have already been allocated, the system only converts the available inventory in the originating warehouse to a pre-held state at the moment the picking instruction is issued, rather than directly determining it as outbound. The corresponding primary transition formula is as follows:

[0137] ;

[0138] ;

[0139] in, Available inventory; The term "pre-held / frozen inventory" represents the state mapping of the aforementioned 'logical isolation and pre-held' mechanism at the entity execution level. From the time the system issues the transfer instruction until the physical outbound scan is completed, this portion of inventory is frozen and cannot be consumed by front-end orders or other transfer tasks. Only when the originating warehouse platform completes physical loading and triggers the departure scan signal will the system execute a secondary state transition: deducting the pre-held inventory at the originating warehouse and simultaneously increasing the in-transit inventory at the destination warehouse. This phased update mechanism, which introduces intermediate states, bridges the time gap between the issuance of instructions from the information system and the physical operation of goods, ensuring the quality conservation and consistency between inventory records and actual inventory in a multi-warehouse collaborative environment.

[0140] For ease of understanding, this embodiment uses the multi-warehouse collaborative network of a large 3C electronics e-commerce platform as an example to demonstrate the actual operation process of the present invention.

[0141] Scenario preset.

[0142] Physical warehouse network: including Beijing warehouse ( ), Shanghai warehouse ( ), Guangzhou warehouse ( ).

[0143] Target product: Smartphone ), dedicated fast charging head ( Custom phone cases ).

[0144] Core execution steps

[0145] Step S10 (Data Mining): The system analyzes order data from the past 30 days and discovers... , , These items are frequently purchased together. The calculated support for joint purchases is 0.04 (threshold 0.02), and the confidence level is 0.75 (threshold 0.6). The system defines this as a valid product cluster. The standard demand ratio is .

[0146] Step S20 (Physical Filtering): Retrieve the physical parameters of the three components and calculate the theoretical combined volume to be 1200 cubic centimeters and the total mass to be 450 grams. This value is far below the upper limit of express delivery parcel volume (e.g., 3000 cubic centimeters) and the upper limit of weight (e.g., 2000 grams), therefore... The product cluster was confirmed as valid and was not decoupled or removed.

[0147] Step S30 (Status Monitoring): At the current moment, Beijing warehouse ( smartphones ( There is a large inventory, but fast charging adapters ( There is a severe shortage of goods; while the Shanghai warehouse ( fast charging adapter ( Inventory backlog. Calculate the complete set inventory in the Beijing warehouse according to the formula. The extremely low inventory level caused a significant deviation between the actual inventory distribution ratio (10%) and the historical regional demand distribution ratio (35%), resulting in the calculation of the inventory center of gravity drift index. .

[0148] Step S40 (Operations Planning): Since the drift index of 0.25 is greater than the set tolerance threshold (0.15), the system triggers a mixed-integer linear programming (MILP) solution. The model comprehensively considers the trunk transportation cost from Beijing to Shanghai and the penalty weight for failure to ship complete sets. The solver's final output decision is: to generate a transfer instruction to move 2000 fast charging heads from the Shanghai warehouse to the Beijing warehouse.

[0149] Step S50 (Instruction Issuance): The Shanghai warehouse receives the structured message, the WMS system automatically segments the batches, assigns AGV carts to pick the goods, merges and loads them onto trucks to be sent to Beijing, and issues an inbound forecast to the Beijing warehouse in advance.

[0150] To verify the effectiveness of the present invention, a simulation test environment was used to conduct a retrospective experiment based on the anonymized historical transaction and warehousing data of an e-commerce company.

[0151] 1. Experimental environment and parameter settings:

[0152] Test dataset: Contains 5 physical warehouses, extracts 100,000 real orders, involving 2,000 independent inventory units. The time window is set to 30 calendar days.

[0153] 2. Comparison scheme setup

[0154] To highlight the technical advantages of this invention, the following three comparative groups are set up:

[0155] Model A (Traditional Safety Stock Level Model): Each warehouse operates independently. When the inventory of a single item falls below the safety stock level, it triggers direct replenishment from the supplier. There is no cross-warehouse collaboration.

[0156] Model B (Association Transfer Model without Physical Constraints): It considers the correlation between goods for transfer, but lacks the "dynamic decoupling of the validity of the commodity cluster based on physical packaging constraints" logic in step S20 of this invention.

[0157] Model C (the model of this invention): includes complete physical packaging decoupling constraints and a kitting gap penalty. Guided hierarchical penalty MILP allocation model

[0158] 3. Comparison table of experimental results data:

[0159] Comparison Dimensions Model A (Traditional Single Warehouse) Model B (Unconstrained Collaboration) Model C (This Invention) Comprehensive allocation cost (yuan / day) 0 (No allocation, but extremely high order loss rate) 45200 31500 Product cluster fulfillment rate 62.4% 89.1% 96.8% Secondary order splitting and shipment rate 15.3% 22.5% 1.2% Cross-warehouse logistics loading rate - 65.4% 88.7%

[0160] The following technical effects can be derived from the above experimental data:

[0161] This invention significantly improves the fulfillment rate of complete product sets: Compared to Model A, Model C, through cross-warehouse dynamic reorganization, increases the fulfillment rate of complete product sets from 62.4% to 96.8%, significantly reducing the predicament of an entire order being unable to be shipped due to the lack of a single component. While Model B improves the fulfillment rate, the indiscriminate bundling and allocation of large or overweight items results in the inability to fit them into a single package during final packaging, causing the secondary order splitting rate to soar to 22.5%. This invention introduces physical volume and weight constraints (…). This reduced the secondary order splitting rate to below 1.2%, proving the physical feasibility of the solution. Through the logical constraints of the trunk line (carriage volume)... (Constraints), the average loading rate of the logistics vehicles in this invention reaches 88.7%, effectively avoiding the phenomenon of scattered vehicles appearing in Model B, and reducing the overall allocation cost by about 30%.

[0162] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A smart inventory allocation management method for multi-warehouse collaboration, characterized in that, The method includes: Obtain historical transaction records and perform product association analysis to identify a set of independent inventory units with purchase associations and construct an initial product cluster; Extract the physical attribute parameters of each independent inventory unit in the initial product cluster, and perform physical constraint verification in combination with the preset upper limit of logistics package specifications. The initial product cluster that meets the verification conditions is determined as a valid product cluster. Obtain real-time available inventory data for each independent inventory unit in each physical warehouse, and generate an inventory center of gravity drift index for each physical warehouse based on the complete inventory of the effective commodity cluster in the corresponding physical warehouse and historical regional demand data. By comparing the inventory center of gravity drift index with a preset tolerance threshold, when the inventory center of gravity drift index is greater than the tolerance threshold, a mixed integer linear programming model is constructed with the objective function of minimizing the weighted sum of trunk transportation costs and business deviation penalties, and the cross-warehouse transfer decision matrix is ​​obtained by solving the model. Based on the cross-warehouse transfer decision matrix, a structured transfer task message is generated, and the automated warehousing equipment is driven to perform physical material transfer actions.

2. The intelligent inventory allocation management method for multi-warehouse collaboration according to claim 1, characterized in that, The step of identifying a set of independent inventory units with purchase associations and constructing an initial product cluster includes: Calculate the joint purchase support and association confidence of different independent inventory unit combinations in the transaction dataset; When both the joint purchase support and the association confidence are greater than or equal to the preset corresponding thresholds, the corresponding set of independent inventory units is defined as the initial product cluster. Based on the demand of each independent inventory unit within the initial product cluster in historical orders, determine the standard demand quantity of each independent inventory unit within the corresponding product cluster.

3. The intelligent inventory allocation management method for multi-warehouse collaboration according to claim 2, characterized in that, After determining the valid product clusters, the process also includes the step of establishing a product cluster relationship mapping table, specifically: Identify multiple valid product clusters belonging to the same independent inventory unit; Based on the association confidence level corresponding to each effective product cluster, the weight allocation function of the independent inventory unit for different effective product clusters is calculated, which is used to perform logical isolation and pre-allocation of available inventory during the allocation planning stage.

4. The intelligent inventory allocation management method for multi-warehouse collaboration according to claim 1, characterized in that, The step of combining the preset upper limit of logistics parcel specifications for physical constraint verification, and determining the initial product clusters that meet the verification conditions as valid product clusters, includes: Extract the three-dimensional dimension data and mass data of each independent inventory unit within the initial product cluster; By introducing a preset spatial stacking conversion factor, the theoretical combined volume and total mass of the initial commodity cluster under the theoretical combined packaging state are calculated; If the theoretical combined volume is less than or equal to the preset upper limit of logistics package volume, and the total mass is less than or equal to the preset upper limit of load capacity, then the initial product cluster is determined to be a valid product cluster.

5. The intelligent inventory allocation management method for multi-warehouse collaboration according to claim 1, characterized in that, The step of generating the inventory center of gravity drift index for each physical warehouse based on the completeness of the effective commodity clusters in the corresponding physical warehouses and historical regional demand data includes: Based on the ratio of the real-time available inventory of each independent inventory unit within the effective commodity cluster in a single physical warehouse to the standard demand quantity, the minimum value of the ratio is taken as the complete inventory of the effective commodity cluster in the physical warehouse. Based on the completeness distribution of inventory for each effective commodity cluster in each physical warehouse, the absolute value of the difference between the actual inventory distribution ratio and the historical demand distribution ratio is calculated to generate the inventory center of gravity drift index for the corresponding physical warehouse.

6. The intelligent inventory allocation management method for multi-warehouse collaboration according to claim 1, characterized in that, The constraints of the mixed-integer linear programming model include: Trunk route logical constraints: By summing the product of the physical volume of a single item and the quantity allocated, the total volume is limited to not exceeding the maximum cargo capacity of the transport vehicle, and this constraint is associated with a Boolean variable for the route. Structural kitting constraint: Introduce non-negative integer variables to represent the expected kitting inventory after allocation, and establish a linear inequality relationship between the expected kitting inventory and the inventory level after allocation of each independent inventory unit in the cluster. Matching Gap Penalty Boundary Constraint: Establish a difference constraint between the expected matching gap quantity and the expected matching stock and historical regional demand benchmark to activate the business deviation penalty term in the objective function.

7. The intelligent inventory allocation management method for multi-warehouse collaboration according to claim 1, characterized in that, The steps for driving automated warehousing equipment to perform physical material transfer operations include: The cross-warehouse transfer decision matrix is ​​converted into a structured transfer task message supported by the underlying logistics control system. The structured allocation task message is written into a preset instruction interaction interface path, which is then polled and read by the warehouse management system to generate hardware control instructions.

8. The intelligent inventory allocation management method for multi-warehouse collaboration according to claim 7, characterized in that, Before the warehouse management system polls, reads, and generates hardware control commands, it also includes: Determine whether the current items to be picked exceed the limits based on the maximum volume and load capacity of the standard turnover container; If the limit is exceeded, the current materials to be picked will be forcibly divided into multiple sub-batches for picking, and the packages will be combined according to the effective commodity cluster relationship during the outbound verification process.

9. An intelligent inventory allocation management system for multi-warehouse collaboration, executed by the intelligent inventory allocation management method for multi-warehouse collaboration as described in any one of claims 1-8, characterized in that, The system includes: The data mining module is used to acquire historical transaction records and perform product association analysis to build initial product clusters; The physical decoupling module is used to extract physical attribute parameters and verify them in conjunction with the upper limit of logistics package specifications to determine the valid product clusters; The status monitoring module is used to obtain real-time available inventory data and generate the inventory center of gravity drift index for each physical warehouse. The operations planning module is used to compare the inventory center of gravity drift index with the tolerance threshold, and to construct and solve the mixed integer linear programming model to output the cross-warehouse transfer decision matrix. The instruction issuance module is used to generate structured allocation task messages and drive automated warehousing equipment to perform physical material allocation.

10. The intelligent inventory allocation management system for multi-warehouse collaboration according to claim 9, characterized in that, The data mining module is also used for: Calculate the joint purchase support and association confidence of different independent inventory unit combinations in the transaction dataset; When both the joint purchase support and the association confidence are greater than or equal to the preset corresponding thresholds, the corresponding set of independent inventory units is defined as the initial product cluster. Based on the demand of each independent inventory unit within the initial product cluster in historical orders, determine the standard demand quantity of each independent inventory unit within the corresponding product cluster.

Images