Dispersed hybrid storage strategy optimization method and system considering SKU relevance

Abstract

The invention relates to a decentralized hybrid storage strategy optimization method and system considering SKU relevance. According to the method, a demand association pattern (DCP) is identified by analyzing historical order data, a mixed storage allocation strategy is combined, a plurality of related SKUs are clustered and stored so as to reduce the sorting time, and meanwhile, cross-roadway SKU dispersion is realized so as to balance resource utilization. A proposed mixed integer programming model (MIP) is combined with constraint conditions of storage capacity, goods channel traversal and SKU relevance, and the weighted sorting distance is minimized. By simplifying distance modeling and variable conversion and combining a two-stage heuristic solution algorithm, optimization of an SKU allocation scheme is realized. In addition, real-time adaptive adjustment is supported, and the storage allocation scheme can be dynamically updated according to new orders and replenishment. According to the invention, the efficiency of warehouse management is improved, and reliable technical support is provided for modern logistics.

Description

Technical Field

[0001] The present invention relates to the field of warehouse management and logistics, and in particular to a method and system for optimizing a decentralized hybrid storage strategy considering SKU relevance, focusing on decentralized and hybrid storage strategies considering SKU relevance. Background Art

[0002] Traditional storage strategies are typically suitable for scenarios where orders contain only a single item. However, in real-world applications, such as industrial manufacturing and e-commerce, orders often contain multiple stock-keeping units (SKUs). These SKUs are often correlated with each other, requiring efficient clustering during storage to reduce order picking time. Existing methods such as deterministic SKU allocation and isolated SKU clustering struggle to adequately address the dynamic and stochastic nature of modern warehousing operations. For example, the widely used traversal-based picking strategy has limited adaptability under high-density SKU correlations and cannot optimize the total walking distance during the picking process.

[0003] In recent years, some studies have attempted to introduce hybrid storage strategies to improve the decentralized storage of related SKUs. However, these approaches often fail to fully account for the dynamic association patterns triggered by fluctuations in customer demand. Therefore, a comprehensive approach is urgently needed to balance the clustering and dispersion of SKUs while minimizing order picking time. Summary of the Invention

[0004] The present invention proposes a method and system for optimizing a decentralized hybrid storage strategy that considers SKU relevance, for optimizing storage space allocation in a decentralized hybrid storage warehousing system that considers SKU relevance. The method includes:

[0005] Step S101: Based on historical order data, the Apriori algorithm is used to extract association rules between SKUs. All frequent item sets with a value greater than or equal to 2 are deduplicated and used as demand association patterns (DCPs). The support of the frequent item sets is used as the DCP weight. The association information between SKUs, the DCP set, and warehouse status parameters are obtained from the historical order data, including the number of SKUs stored in each aisle in the initial state and the available aisle capacity. The DCP is used as the core basis for optimizing warehouse storage and picking. A hybrid storage allocation strategy clusters SKUs in the DCP and stores them in adjacent locations, while dispersing the DCP to balance warehouse load.

[0006] Step S102: Using the parameters, a decentralized hybrid storage strategy optimization model considering SKU relevance is established. A mixed integer programming model is constructed. First, a series of observation points evenly distributed in the warehouse are introduced, represented by a set D. Minimizing the weighted sum of the picking travel distances of the DCP from all observation points is proposed as an alternative objective. Constraints are introduced into the model, including storage capacity restrictions, aisle traversal rules, and SKU relevance.

[0007] Step S103, using the decentralized hybrid storage strategy optimization model to obtain a storage solution for each SKU, including: proposing a two-stage heuristic algorithm, in the first stage, first constructing an initial feasible solution to ensure basic coverage of the DCP, and further optimizing the SKU allocation solution through an iterative fixed optimization algorithm; in the second stage, allocating the remaining products that have not been allocated in the first stage.

[0008] A decentralized hybrid storage strategy optimization system considering SKU relevance, the system comprising:

[0009] The parameter acquisition module is used to extract association rules between SKUs based on historical order data using the Apriori algorithm. All frequent item sets with a value greater than or equal to 2 are deduplicated as demand association patterns (DCPs), and the support of frequent item sets is used as the DCP weight. The module also obtains association information between SKUs, DCP sets, and warehouse status parameters from historical order data, including the number of SKU products stored in each aisle in the initial state and the available aisle capacity information.

[0010] The model building module uses the above parameters to establish a decentralized hybrid storage strategy optimization model that considers SKU relevance. A mixed integer programming model is constructed, which first introduces a series of observation points evenly distributed throughout the warehouse, represented by the set D. Minimizing the weighted sum of the picking travel distances of the DCP from all observation points is proposed as an alternative objective. The model introduces constraints, including storage capacity limits, aisle traversal rules, and SKU relevance.

[0011] The algorithm solution module uses the decentralized hybrid storage strategy optimization model to obtain a storage solution for each SKU, including: proposing a two-stage heuristic algorithm. In the first stage, an initial feasible solution is first constructed to ensure basic coverage of the DCP, and an iterative fixed optimization algorithm is used to further optimize the SKU allocation solution. In the second stage, the remaining products that have not been allocated in the first stage are allocated.

[0012] A computing device comprises: at least one processor and a memory storing program instructions; when the program instructions are read and executed by the processor, the computing device executes the decentralized hybrid storage strategy optimization method considering SKU relevance.

[0013] Beneficial effects: According to the technical solution of the present invention, the above components are integrated into a modular software platform, which supports real-time decision-making and is seamlessly integrated into the existing warehouse management system. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 This is a flow chart of a method for optimizing a distributed hybrid storage strategy taking into account SKU relevance provided by an embodiment of the present invention;

[0015] Figure 2 This is a schematic diagram of a warehouse plan provided by one embodiment of the present invention;

[0016] Figure 3 is an illustration of the actual distance and the alternative distance provided by an embodiment of the present invention;

[0017] Figure 4 The graph showing the regression analysis results of the surrogate distance and the true distance is shown;

[0018] Figure 5 Show the impact of associativity strength on access efficiency of different schemes;

[0019] Figure 6 Shows the impact of allowing mixing of cargo locations on storage and retrieval efficiency. DETAILED DESCRIPTION

[0020] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with specific embodiments and with reference to the accompanying drawings.

[0021] Figure 1 This is a flow chart of a method for optimizing a distributed hybrid storage strategy taking into account SKU relevance provided by an embodiment of the present invention. Figure 1 As shown, the method includes:

[0022] Step S101: Based on historical order data, the Apriori algorithm is used to extract association rules between SKUs. All frequent item sets greater than or equal to 2 are deduplicated as demand association patterns (DCPs), and the support of frequent item sets is used as the DCP weight. The association information between SKUs, the DCP set, and warehouse status parameters are obtained from the historical order data, including the number of SKU products stored in each lane in the initial state and the available capacity of the lane.

[0023] Use DCPs as the core basis for optimizing warehouse storage and picking. A hybrid storage allocation strategy clusters SKUs in the DCPs and stores them in adjacent locations, reducing picking time. It also disperses the DCPs to balance warehouse loads, avoid congestion, and improve picking efficiency.

[0024] In this step, DCP contains the association information between SKUs extracted from historical order data. A DCP is represented by a set of SKUs. The SKUs in the DCP often appear in orders at the same time. The DCP set is represented by C. ( ) represents a DCP.

[0025] By reading historical order data from the database, a certain analysis algorithm is used to obtain the association information between SKUs, and it is expressed in the form of a demand correlation pattern (DCP). The complete set of DCPs is denoted as C. Each DCP is a complete set of SKUs. At the same time, it is necessary to give each DCP a certain weight according to its importance. ( ). Let the lane set be A. In addition, it is necessary to read the current warehouse status information, including each SKU In different lanes The amount of storage in , the available capacity of each lane .

[0026] In step S102, the aforementioned parameters are used to establish a decentralized hybrid storage strategy optimization model (SCSAwC) that considers SKU relevance. A mixed integer programming (MIP) model is constructed. This model first introduces a set of observation points evenly distributed throughout the warehouse, represented by the set D. The observation points' horizontal coordinates, τ, are evenly spaced along the length of the warehouse, and the number of observation points is set based on the warehouse size. Minimizing the weighted sum of the picking travel distances from all observation points in the DCP is proposed as an alternative objective. Constraints introduced into the model include storage capacity limits, aisle traversal rules, and SKU relevance.

[0027] First of all, it should be noted that when building the model, the picking path strategy adopts the S-shaped path strategy, that is, entering an aisle requires completely passing through the entire aisle and leaving the aisle at the other end. Figure 2 As shown. Establish a rectangular coordinate system with the lower left corner of the warehouse as the origin. The length and width of each storage location and the width of all aisles are equal to 1. An aisle is represented by its corresponding horizontal coordinate. A given number of observation points D are evenly distributed throughout the warehouse, and each observation point is represented by its horizontal coordinate.

[0028] Define the optimization goal:

[0029] ,

[0030] in, From the observation point Departure picking DCP Picking distance. Indicates the weight of each DCP.

[0031] The decentralized hybrid storage strategy optimization model includes a first constraint condition, which includes:

[0032]

[0033] Among them, the decision variables Indicates SKU Assigned to lane The number of Indicates the SKU to be stored The first constraint requires that each SKU The sum of the quantities is equal to the given quantity to be stored .parameter Indicates lane The second constraint requires that the available storage capacity be allocated to the lanes The sum of the number of products does not exceed the available capacity of the lane. Indicates the initial state of the warehouse, that is, before adding a new SKU, the aisle SKUs stored in The number of . From the observation point Departure picking DCP Is it necessary to access the lane? , the third constraint states that for any DCP Any SKU in , from any observation point The condition for being picked is that the aisle in the initial state of the warehouse Store SKU Or SKUs to be stored Assigned to the lane . From the observation point When picking DCP c, the leftmost lane selected is the same as the observation point The horizontal distance, From the observation point When picking DCP c, the rightmost lane selected is the same as the observation point The fourth to sixth constraints describe the horizontal distance from the observation point Departure picking DCP The shortest walking distance after determining the lanes to be visited. is a set of natural numbers. The seventh constraint represents the variable is an integer, the eighth constraint represents the variable It is a 0-1 variable.

[0034] The second constraint and the fourth to sixth constraints in the above steps are nonlinear constraints, which bring great difficulties to the solution of the model. Next, the present invention will construct a nonlinear 0-1 programming model based on the above model.

[0035] ,

[0036] in, Indicates the weight of each DCP.

[0037] In this model, from the observation point Departure picking DCP Picking distance is defined as follows:

[0038] ,

[0039] in, represents the horizontal coordinate of the lane, Indicates the horizontal coordinate of the observation point. Indicates the depth of the roadway. Decision variables From the observation point Departure picking DCP Is it necessary to access the lane? .

[0040] This is actually an approximate method of calculating the path length. Figure 3 As shown, the left side is the shortest walking route after the given lanes to be visited. In the present invention, the horizontal distance from the observation point to each lane and the sum of the depths of all visited lanes approximately represent the picking distance. Figure 4 shows the correlation between the true distance and the surrogate distance, The results show that the surrogate distance is highly consistent with the true distance.

[0041] The second constraint of the nonlinear 0-1 programming model includes the following conditions:

[0042]

[0043] Decision variables Defined as whether the roadway Allocate SKU The first constraint states that the SKUs assigned to all lanes The quantity does not exceed the given SKU Number to be stored The third constraint states that for any DCP Any SKU in , from any observation point The condition for being picked is that the aisle in the initial state of the warehouse Store SKU Or SKUs to be stored Assigned to the lane Here, the lanes are divided into two categories, and the SKUs are stored in the initial state. The lane is , other lanes use Indicates. Because in this model we only care whether we need to move to the lane Allocate SKU If the lane already contains SKU in the initial state , then there is no need to assign SKU to this lane Therefore, we only need to decide whether to move to the lane Assign SKUs The fourth constraint requires that the variable It is a 0-1 variable.

[0044] Step S103 uses the decentralized hybrid storage strategy optimization model to determine a storage solution for each SKU. This includes proposing a two-stage heuristic algorithm. In the first stage, an initial feasible solution is constructed to ensure basic DCP coverage. An iterative fix-and-optimize algorithm is then used to further optimize the SKU allocation solution. In the second stage, any products not yet allocated in the first stage are allocated.

[0045] Since this problem is NP-hard, existing commercial solution software cannot solve it within a limited time. Therefore, the present invention proposes a two-stage solution algorithm to quickly solve large-scale instances.

[0046] First, we need to construct an initial feasible solution. Before constructing the initial feasible solution, we need to define some concepts. The weight of SKU ( Defined as the maximum value of the DCP weight that includes this SKU ( = SKU correlation is defined as the number of times two SKUs appear in the same DCP. SKU The similarity ( ) is defined as the initial SKU and SKU in the lane The sum of the correlations ( When SKU i appears in DCP c, Take 1, otherwise take 0). It should be noted that if the lane Initially, the SKU already exists , then the lane SKU The similarity is 0.

[0047] Constructing the initial solution includes: by converting the SKU association information represented by the DCP into pairwise SKU related information, maximizing the association between the lanes and the SKUs to construct the initial solution. When constructing the initial feasible solution, the products of each SKU are stored in the lane with the greatest similarity to the SKU in descending order of SKU weight. If the maximum similarity between the lane and SKU 1 is less than the maximum similarity between the lane and SKU 2, then start processing SKU 2. If there are still unassigned products for SKU 1, SKU 1 is added to the end of the queue, otherwise SKU 1 is removed from the queue. In order to ensure that all SKUs are included in the initial solution and the initial state, it is first necessary to store at least one product for the SKUs not included in the initial state according to the above rules. Finally, the remaining unassigned products are placed in the lane with the greatest similarity until all products are stored or all similarities are 0.

[0048] Algorithm 1 Initial solution construction

[0049] Input: Warehouse Layout , initial state , demand and cost , SKU collection ,in is the cargo space capacity,

[0050] Output: An initial SKU allocation plan

[0051] Calculate the weight of each SKU and the correlation between SKUs .

[0052] Building a Collection , including all SKUs that appear in the DCP.

[0053] According to the weight of SKU right Sort in descending order, and get .

[0054] For the sorted SKU, if the SKU does not exist in the warehouse yet:

[0055] Calculate the affinity between the SKU and each lane;

[0056] Assign an item of this SKU to the channel with the highest affinity ;

[0057] If there are no items left for that SKU, remove it from the list.

[0058] When there are at least two more SKUs in the sorted list, repeat the following process:

[0059] Take out SKU1 at the head of the list, if there are still pieces left:

[0060] Take out another SKU2;

[0061] Find the channels with the highest affinity and the corresponding affinity respectively , ;

[0062] like , then assign an item to SKU1:

[0063] If SKU2 items have been allocated, remove them from the list;

[0064] Otherwise, assign an item to SKU2;

[0065] If there are any SKU2 items left, put them back to the head of the list; otherwise, remove them.

[0066] If there are any SKU1 items left, put them back to the end of the list;

[0067] Jump out of the inner loop.

[0068] For the remaining SKUs, the remaining items are allocated to the channel with the highest affinity in turn until all the items are allocated.

[0069] In the Iterative Fix-and-Optimize algorithm, the DCP set is divided into several smaller non-overlapping subsets and the decision variables related to a DCP subset are optimized in each iteration. The selection of the DCP subset for each optimization needs to consider two factors: the weight of the DCP and the similarity of the DCP. In each iteration, the DCPs are first sorted from large to small according to their weight, and the DCP with the largest weight is selected to be added to the subset. Then, at each step, a DCP with the greatest similarity to the existing DCPs in the subset is selected to be added to the subset until the number of DCPs in the subset reaches a given upper limit. .

[0070] Algorithm 2 Iterative Fix-and-Optimize

[0071] Input: Initial solution , warehouse layout , initial warehouse state , demand pattern information , SKU and its storage requirements , algorithm parameters ,in is the upper limit of the number of given DCPs, is the algorithm running time limit (in seconds).

[0072] Output: Partial allocation plan

[0073] Create an optimization model and add all constraints.

[0074] Defining decision variables and , are binary variables.

[0075] Will is set to the value in the initial solution, that is, , for all .

[0076] Group and sort the demand-related patterns (DCPs) .

[0077] when Not empty and time has elapsed , repeat the following steps:

[0078] Select and remove The first group in .

[0079] Find out with this The associated SKU set is recorded as .

[0080] For all ,Will The value range is set to [0,1].

[0081] Update and solve the optimization model.

[0082] Update with new results ,Right now , for all .

[0083] Will Fixed to the value of the current optimal solution, for all .

[0084] In the Iterative Fix-and-Optimize algorithm, how to divide the DCP subsets is an important issue. For a given parameter, the upper limit of the number of DCPs is , a group sorting scheme that considers both DCP weight and DCP overlap is proposed. The greater the weight of the DCP, the greater the impact of the DCP on the objective function value, so it should be given priority. The greater the overlap of DCPs in the same group, the better the solution will be if these DCPs are processed simultaneously, because overlapping SKUs need to take into account several DCPs at the same time, and there is more room for optimization if these DCPs are processed simultaneously. In short, in the proposed group sorting scheme, the DCPs are first sorted according to their weights, and the DCPs with larger weights are processed first. At the same time, when selecting the DCPs in the same group When adding a DCP, the overlap between the newly added DCP and the existing DCP is maximized.

[0085] Algorithm 3 DCP sorting and grouping

[0086] Input: Demand pattern information , process-specific parameters ( )

[0087] Output: DCP sorting and grouping results

[0088] According to weight Sort DCP in descending order and get .

[0089] Initialize an empty list Used to store sorted and grouped DCPs.

[0090] when When not empty, repeat the following:

[0091] Initialize an empty group .

[0092] if The length is exactly equal to :

[0093] Will Add to .

[0094] otherwise:

[0095] take out The first DCP in .

[0096] when The length is less than and When non-empty, repeat:

[0097] exist Find the current The DCP with the largest overlap among the SKU sets of all DCPs .

[0098] Will join in and from Removed.

[0099] Will Add to .

[0100] Through the previous procedure, a part of the products have been allocated. The remaining products do not affect the objective function value of the original problem, but in actual operation, storage locations must be allocated for these products. In addition, the allocation plan for these products will also affect the picking efficiency. For the SKUs included in the DCP, the remaining products should be evenly distributed to the aisles that already contain the same SKUs, because the previous optimization plan has already told these SKUs the aisles where they should be stored. SKUs that do not exist in the DCP are allocated to aisles according to the principle of random allocation and as much dispersion as possible. Finally, a random allocation strategy is used within the aisles to allocate products to specific storage locations, and a complete SKU to storage location allocation plan is obtained. For SKUs that are not in the DCP, they are preferentially stored in aisles that do not contain such SKUs, because we hope that the same products of these SKUs are as dispersed as possible so that these SKUs can be more easily obtained from any location during picking.

[0101] Algorithm 4: Allocation of remaining items

[0102] Input: Warehouse Layout , initial warehouse state and partial allocation plan

[0103] Output: Distribution plan of all items

[0104] Get the SKU set included in the DCP .

[0105] Combine Calculate the new warehouse status and count the aisle sets that appear and do not appear for each SKU in the current status and .

[0106] Calculate the remaining available capacity of each lane in the new state .

[0107] Create a collection of remaining items to be allocated .

[0108] when When not empty, repeat the following:

[0109] from Take out a SKU from , and initialize Empty. yes Assigned lanes.

[0110] if belong :

[0111] From the included and Randomly select an aisle from the shelf , and Assign to this channel.

[0112] otherwise:

[0113] Never contains and Randomly select an aisle from the shelf , and Assign to this channel.

[0114] If the above two steps cannot find an available channel:

[0115] Random from all Select one of the channels ,Will Assign to this channel.

[0116] Below, the decentralized hybrid storage strategy proposed in the present invention is compared with the random storage strategy commonly used in practice and a decentralized storage strategy proposed by Weidinger and Boysen (2018) to further illustrate the technical solution of the present invention:

[0117] To validate the proposed alternative objective function's performance in improving storage and retrieval efficiency, simulation experiments were conducted under various warehouse settings and routing strategies. Weidinger and Boysen (2018) proposed a modified "nearest neighbor heuristic" algorithm for generating picking and replenishment routes between locations, fixing the number of items per picking and replenishment task to six (determined by the capacity of the picking cart). This simulation setting is referred to as the fixed-piece location-to-location setting (FCSS). It is assumed that pickers complete only one customer order per pick, while replenishment tasks remain fixed at six items per pick. Therefore, the number of items per picking task is not fixed. Furthermore, both pickers and replenishers employ an S-shaped routing strategy. This simulation setting is referred to as the order-picking aisle-to-aisle setting (OPAA).

[0118] The alternative objective function SSA proposed by Weidinger and Boysen (2018) is defined as: the weighted sum of the maximum distances from any measurement point of each SKU to the nearest product (taking into account the current location and the location of newly added products), that is, decentralized storage allocation (SSA). The alternative objective function proposed in this invention is called SCSAwC. In addition, the surveyed companies adopt a greedy storage strategy in actual operations: the replenishers randomly store the SKUs in any available location and record the information through handheld devices. Under this strategy, the replenishers will naturally give priority to the nearest available location and store as many products as possible (limited by capacity). This strategy is called random strategy allocation (RSA). In the simulation, it is assumed that the location capacity is a positive integer. In a non-compound setting (NC), each shelf can only store multiple units of a single SKU; in a mixed setting (C), each shelf can store multiple units of multiple different SKUs. This invention focuses on the picking distance during the outbound process and the total distance during the replenishment process.

[0119] exist Figure 5 Middle, horizontal axis Indicates the richness of SKU correlation information, and the vertical axis represents the distance. Figure 5 As shown in Figure 2, under the OPAA setting, SCSAwC shows a more significant advantage than SSA. This is because SCSAwC incorporates SKU correlation information; As for picking distance, RSA and SSA are almost unaffected by SKU relevance information because they do not utilize SKU relevance information. With the impact of changes. As the performance gap between SCSAwC and RSA and SSA gradually widens, reflecting the improvement brought by utilizing additional information. In terms of total distance, the advantage of SCSAwC is more obvious, and this advantage will increase with This result undoubtedly shows that whether the goal is to speed up order fulfillment or reduce overall operating costs, managers should pay more attention to the use of SKU correlation information, and SCSAwC is a better choice.

[0120] Next, we analyze the benefits of SKU commingling. Figure 6 As shown, when the cargo space capacity As the number of locations increases (while the number of shelves remains unchanged), the overall warehouse capacity also increases. Since the picking and replenishment paths under full capacity are always recorded in the simulation, both the picking distance and the total distance will increase with the increase of the warehouse capacity. The main focus is on the improvement ratio (IMP) of the hybrid (C) scenario relative to the non-hybrid (NC) scenario, where , which is indicated by dotted lines in the figure, The distance set under which the mixing of cargo locations is not allowed, It is the distance under which the mixing of cargo locations is allowed. As the number of SKUs increases, the SKU mixing degree in scenario C also increases, resulting in a higher IMP. This shows that allowing multiple SKUs to be stored in one location can significantly improve picking distance and overall workload, and the higher the mixing degree, the greater the improvement.

[0121] Although mixed SKU storage will bring additional search costs (i.e., the time to find the target SKU in the same location), research and actual industry experience show that as long as the number of SKU types in each location is kept within a reasonable range (for example, no more than 7), the search cost can be ignored. When the system is used, the improvement in picking distance is approximately 40%, and the improvement in total distance is approximately 12%. This represents a significant improvement in warehouse operating efficiency.

[0122] Another advantage of SKU intermixing is that it effectively reduces warehouse space requirements. Mixing smaller quantities of SKUs in the same location reduces the total number of locations, further reducing picking and replenishment path lengths. Therefore, from a practical perspective, as long as the degree of SKU intermixing is moderately increased without significantly increasing search costs, picking efficiency can be effectively improved and warehouse operating costs can be reduced.

[0123] The present invention also provides a storage strategy optimization system in a decentralized hybrid storage warehouse considering SKU relevance, comprising the following modules:

[0124] The parameter acquisition module is used to extract the association rules between SKUs based on historical order data using the Apriori algorithm. All frequent item sets greater than or equal to 2 are deduplicated as demand association patterns (DCPs), and the support of frequent item sets is used as the DCP weight. The module also obtains the association information between SKUs, DCP sets, and warehouse status parameters from historical order data, including the number of SKU products stored in each aisle in the initial state and the available capacity of the aisle.

[0125] The model building module uses the aforementioned parameters to develop a decentralized hybrid storage strategy optimization model (SCSAwC) that considers SKU relevance. This model constructs a mixed integer programming (MIP) model. This model first introduces a set of observation points evenly distributed throughout the warehouse, represented by the set D. Minimizing the weighted sum of the picking travel distances from all observation points in the DCP is proposed as an alternative objective. Constraints introduced into the model include storage capacity limits, aisle traversal rules, and SKU relevance.

[0126] The algorithm solution module uses the decentralized hybrid storage strategy optimization model to determine a storage solution for each SKU. This involves proposing a two-stage heuristic algorithm. In the first stage, an initial feasible solution is constructed to ensure basic coverage of the DCP. An iterative fix-and-optimize algorithm is then used to further optimize the SKU allocation solution. In the second stage, any products not yet allocated in the first stage are allocated.

[0127] The present invention also provides a computing device comprising: at least one processor and a memory storing program instructions; when the program instructions are read and executed by the processor, the computing device executes the decentralized hybrid storage strategy optimization method considering SKU relevance as described above.

[0128] In the description provided herein, numerous specific details are described. However, it is understood that embodiments of the present invention may be practiced without these specific details. In some instances, well-known methods, structures, and techniques are not shown in detail so as not to obscure the understanding of this description.

[0129] Although the present invention has been described with respect to a limited number of embodiments, those skilled in the art, having benefit of the foregoing description, will appreciate that other embodiments are contemplated within the scope of the invention thus described. Furthermore, it should be noted that the language used in this specification has been selected primarily for readability and instructional purposes, and not for the purpose of explaining or limiting the subject matter of the present invention.

Claims

1. A method for optimizing a distributed hybrid storage strategy considering SKU relevance, characterized in that: The method comprises: Step S101: Based on historical order data, the Apriori algorithm is used to extract association rules between SKUs. All frequent item sets with a value greater than or equal to 2 are deduplicated and used as demand association patterns (DCPs). The support of the frequent item sets is used as the DCP weight. The association information between SKUs, the DCP set, and warehouse status parameters are obtained from the historical order data, including the number of SKUs stored in each aisle in the initial state and the available aisle capacity. The DCP is used as the core basis for optimizing warehouse storage and picking. A hybrid storage allocation strategy clusters SKUs in the DCP and stores them in adjacent locations, while dispersing the DCP to balance warehouse load. Step S102: Using the parameters, a decentralized hybrid storage strategy optimization model considering SKU relevance is established. A mixed integer programming model is constructed. First, a series of observation points evenly distributed in the warehouse are introduced, represented by a set D. Minimizing the weighted sum of the picking travel distances of the DCP from all observation points is proposed as an alternative objective. Constraints are introduced into the model, including storage capacity restrictions, aisle traversal rules, and SKU relevance. Step S103, using the decentralized hybrid storage strategy optimization model to obtain a storage solution for each SKU, including: proposing a two-stage heuristic algorithm, in the first stage, first constructing an initial feasible solution to ensure basic coverage of the DCP, and further optimizing the SKU allocation solution through an iterative fixed optimization algorithm; in the second stage, allocating the remaining products that have not been allocated in the first stage.

2. The method for optimizing a distributed hybrid storage strategy considering SKU relevance according to claim 1, characterized in that: The DCP contains the association information between SKUs extracted from historical order data. A DCP is represented by a set of SKUs. The SKUs in the DCP often appear in orders at the same time. The DCP set is represented by C. Indicates a DCP, .

3. The method for optimizing a distributed hybrid storage strategy considering SKU relevance according to claim 2, characterized in that: The decentralized hybrid storage strategy optimization model uses the following formula as an alternative optimization target: ， in, Indicates DCP The weight of From the observation point Departure picking DCP The picking walking distance is 20 km, and the picking path strategy adopts the S-shaped path strategy, that is, entering an aisle requires completely passing through the entire aisle and leaving the aisle from the other end.

4. The method for optimizing a distributed hybrid storage strategy considering SKU relevance according to claim 3 is characterized in that: It is given by: ， in, represents the horizontal coordinate of the lane, represents the horizontal coordinate of the observation point, represents the depth of the roadway, the decision variable From the observation point Departure picking DCP Is it necessary to access the lane? .

5. The method for optimizing a distributed hybrid storage strategy considering SKU relevance according to claim 2, characterized in that: The decentralized storage strategy optimization model includes the following constraints: ， ， ， ， ， ， ， ， Among them, the decision variables Indicates SKU Assigned to lane the number of It is the complete set of SKUs. It is a collection of lanes. Indicates the SKU to be stored The first constraint requires that each SKU The sum of the quantities is equal to the given quantity to be stored ;parameter Indicates lane The available storage capacity, the second constraint requires that it be allocated to the lanes The sum of the number of products does not exceed the available capacity of the lane; parameter Indicates the initial state of the warehouse, that is, before adding a new SKU, the aisle SKUs stored in The number of From the observation point Departure picking DCP Is it necessary to access the lane? The third constraint indicates that for any SKU in any DCP c , from any observation point The conditions for being picked are that the aisle in the initial state of the warehouse Store SKU Or SKUs to be stored Assigned to the lane ; From the observation point When picking DCP c, the leftmost lane selected is the same as the observation point The horizontal distance, From the observation point When picking DCP c, the rightmost lane selected is the same as the observation point The fourth to sixth constraints describe the horizontal distance from the observation point Departure picking DCP When , the shortest walking distance after determining the lanes to be visited; is a set of natural numbers, and the seventh constraint represents the variable is an integer, the eighth constraint represents the variable It is a 0-1 variable.

6. The method for optimizing a distributed hybrid storage strategy considering SKU relevance according to claim 3 is characterized in that: The two-stage algorithm specifically includes: Initial solution construction: by converting the SKU association information represented by DCP into pairwise SKU related information, the initial solution is constructed by maximizing the association between lanes and SKUs; The Iterative Fix-and-Optimize algorithm optimizes the allocation of SKUs between aisles. In the Iterative Fix-and-Optimize algorithm, the DCP set is divided into several smaller, non-overlapping subsets, and the decision variables associated with one DCP subset are optimized in each iteration. The selection of each optimized DCP subset takes into account both the DCP weight and the DCP similarity. In each iteration, the DCPs are first sorted from largest to smallest by weight, and the DCP with the largest weight is selected to be added to the subset. Then, at each step, the DCP with the greatest similarity to an existing DCP in the subset is selected and added to the subset until the number of DCPs in the subset reaches a given upper limit. Phase 2: For SKUs included in the DCP, the remaining products are evenly distributed to the aisles that already contain the same SKUs. SKUs that do not exist in the DCP are randomly assigned to aisles in a dispersed manner. Finally, a random allocation strategy is used within the aisles to allocate products to specific storage locations, resulting in a complete SKU to shelf allocation plan.

7. The method for optimizing a distributed hybrid storage strategy considering SKU relevance according to claim 6, characterized in that: In the Iterative Fix-and-Optimize algorithm, the upper limit of the number of DCPs for a given parameter is , a grouping sorting scheme that considers both DCP weight and DCP overlap is proposed: in the proposed grouping sorting scheme, DCPs are first sorted according to their weights, with DCPs with larger weights being prioritized; at the same time, When adding a DCP, the overlap between the newly added DCP and the existing DCP is maximized.

8. The method for optimizing a distributed hybrid storage strategy considering SKU relevance according to claim 6, characterized in that: SKUs assigned to all lanes The quantity does not exceed the given SKU Number to be stored .

9. A decentralized hybrid storage strategy optimization system considering SKU relevance, characterized in that: The system comprises: The parameter acquisition module is used to extract association rules between SKUs based on historical order data using the Apriori algorithm. All frequent item sets with a value greater than or equal to 2 are deduplicated as demand association patterns (DCPs), and the support of the frequent item sets is used as the DCP weight. The module also obtains association information between SKUs, DCP sets, and warehouse status parameters from historical order data, including the number of SKU products stored in each aisle in the initial state and the available aisle capacity. The model building module uses the above parameters to establish a decentralized hybrid storage strategy optimization model that considers SKU relevance. A mixed integer programming model is constructed, which first introduces a series of observation points evenly distributed throughout the warehouse, represented by the set D. Minimizing the weighted sum of the picking travel distances of the DCP from all observation points is proposed as an alternative objective. The model introduces constraints, including storage capacity limits, aisle traversal rules, and SKU relevance. The algorithm solution module uses the decentralized hybrid storage strategy optimization model to obtain a storage solution for each SKU, including: proposing a two-stage heuristic algorithm. In the first stage, an initial feasible solution is first constructed to ensure basic coverage of the DCP, and an iterative fixed optimization algorithm is used to further optimize the SKU allocation solution. In the second stage, the remaining products that have not been allocated in the first stage are allocated.

10. A computing device, characterized in that include: at least one processor and memory storing program instructions; When the program instructions are read and executed by the processor, the computing device executes the distributed hybrid storage strategy optimization method considering SKU relevance according to any one of claims 1 to 8.

Images