A collaborative consolidation global optimization method and system

Abstract

The application provides a collaborative case-packing global optimization method and system, and belongs to the technical field of data optimization. Specifically, a modeling module is responsible for generating all in-box allocation schemes that meet the requirement of "at least one piece of SKU is allocated", a scoring module is responsible for calculating a candidate scheme set by using a small root heap to dynamically maintain the top K candidate schemes with the highest scores, controlling the size of the candidate schemes, and a solving module is responsible for selecting an optimization scheme with the minimum complexity level that can minimize the overall remaining SKU demand as the final output scheme. On the basis of reducing transportation costs, the efficiency of identifying and processing the optimal collaborative case-packing scheme is improved.

Description

Technical Field

[0001] This invention belongs to the field of data optimization technology, and in particular relates to a collaborative bin-packing global optimization method and system. Background Technology

[0002] In existing automated consolidation order (AOSS) operations in logistics warehousing, traditional methods typically employ a greedy batching strategy, which involves selecting the locally optimal consolidation solution round by round until the remaining SKUs cannot meet the consolidation requirements. This approach easily leads to problems such as an excessively large overall remaining manual handling volume (MSS), low AOSS processing efficiency, inconsistent packing structures, and uneven production line load. Furthermore, when dealing with large-scale SKU combinations, enumerating all feasible solutions can cause combinatorial explosion, resulting in excessively high solution complexity, making it difficult to obtain a globally optimized solution within a limited time, and hindering engineering implementation.

[0003] Therefore, there is an urgent need for a collaborative global optimization method and system for container consolidation. Summary of the Invention

[0004] To achieve the objectives of this invention, the following technical solution is adopted:

[0005] Specifically, this application provides a collaborative consolidation global optimization system, which includes:

[0006] The modeling module is responsible for enumerating all SKU combinations with sizes ranging from 1 to the upper limit k of the preset complexity level according to the preset business constraints. For each SKU combination, the single box capacity is determined according to the "shortest board principle", and all box allocation schemes that satisfy "at least one piece of each SKU" are generated.

[0007] The scoring module is responsible for calculating the maximum number of boxes that can be produced by each allocation scheme within a box, and applying batch consistency constraints and maximum number of boxes constraints to filter and obtain feasible schemes. Feasible schemes are scored by "number of boxes that can be produced × capacity of a single box". The "Top-K streaming pruning" strategy is adopted, using a min-heap to dynamically maintain the top K candidate schemes with the highest scores to obtain a set of candidate schemes and control the size of the candidate schemes.

[0008] The solution module is responsible for inputting the selected candidate solution set into the optimization solver, constructing the optimization solver using a greedy strategy and setting a maximum solution time limit for each level, executing an iterative strategy that prioritizes high complexity, recording the remaining SKU requirements corresponding to each complexity level, and selecting the optimization solution with the lowest overall remaining SKU requirements as the final output solution.

[0009] Furthermore, the SKUs to be processed need to be strategically grouped according to preset constraint rules before the SKU combination enumeration process is performed.

[0010] Furthermore, the upper limit k of the preset complexity level is the number of SKUs in a single box.

[0011] Furthermore, generate all in-box allocation schemes that satisfy the requirement of "at least one item per SKU", specifically including:

[0012] Based on the enumeration range of all SKU combinations from a single SKU (size=1) to the preset complexity upper limit k, we enumerate to obtain SKU combinations;

[0013] For each enumerated SKU combination, its theoretical maximum capacity per box is determined according to the "shortest board principle";

[0014] The specific operation involves extracting the packing quantity parameter for all SKUs within the combination and taking the minimum value as the uniform packing capacity for the combination. This principle ensures that the generated allocation scheme can meet the physical packing constraints of all SKUs in actual packing operations, guaranteeing the physical feasibility of the scheme.

[0015] Given a fixed single-box capacity, a hard constraint of "at least one item per SKU" is added. An integer partitioning algorithm is used to generate a specific allocation scheme for the SKUs within the box, resulting in the box allocation scheme.

[0016] Given a fixed single-box capacity, an integer partitioning algorithm is used to generate a specific allocation scheme for the SKUs within the box. This process includes a hard constraint that "each SKU must be allocated at least once" to ensure that all SKUs within the combination are represented and to avoid invalid zero allocations. The algorithm systematically enumerates all positive integer solutions that allocate the total capacity (i.e., the single-box capacity) to each SKU, forming a complete set of allocation patterns that conform to basic business rules.

[0017] Furthermore, the system caches the calculated splitting results for SKU combinations, so that when the same parameters are encountered later, they can be read directly from the cache.

[0018] Furthermore, the flowchart of the method for obtaining the candidate solution set is as follows:

[0019] For each SKU involved in each generated in-box allocation scheme, calculate the ratio of its initial demand to the number of boxes allocated to that SKU in the scheme, round down, and take the minimum value of the calculation results for all related SKUs as the maximum number of boxes that the allocation scheme can produce without other constraints.

[0020] The calculated maximum number of boxes that can be produced is compared with the system's preset batch consistency threshold. Only when the number of boxes that can be produced is greater than or equal to the batch consistency threshold is the solution considered to meet the batch stability requirements of the production line operation, thus obtaining a solution that meets the batch consistency constraint. For solutions that meet the batch consistency constraint, they are further compared with the system's optional parameter—the upper limit of the maximum number of boxes. The number of boxes that can be produced in the solution must not exceed this upper limit, thus obtaining a feasible solution.

[0021] Each feasible solution is assigned a score, defined as the product of "number of boxes that can be produced" and "capacity per box". A priority queue-based streaming pruning mechanism is used, initializing a min-heap of capacity K to dynamically maintain the K solutions with the highest scores among the currently processed solutions. When processing each feasible solution, its score is compared with the top of the min-heap (i.e., the score of the current Kth solution). Based on the comparison result, it is determined whether to insert it into the heap and may discard the top solution.

[0022] During the streaming process, when the min-heap is full (i.e., K solutions have been included), for the SKU combination to which the newly arrived solution belongs, the theoretically highest possible score upper bound of the combination can be calculated in advance. If this theoretical upper bound is lower than the current top score of the heap, it can be determined that all unprocessed allocation solutions in the combination have no chance of entering the Top-K set. After the process is completed, the K feasible solutions stored in the min-heap constitute the final candidate solution set submitted to the optimization solver within the strategic group.

[0023] Secondly, this invention provides a collaborative consolidation global optimization method, applied to the aforementioned collaborative consolidation global optimization system, specifically including:

[0024] Obtain SKU data and, based on the deviation in the quantity of goods corresponding to each SKU within a single box, determine the recommended value for the complexity level.

[0025] Based on the recommended values, a global optimization process for collaborative consolidation is performed. According to the single-box capacity and overall MSS index of different SKU combinations under the existing complexity level, the optimization scheme of the recommended values ​​at the complexity level is determined.

[0026] The recommended value is optimized using the optimization processing scheme to obtain the optimization result. Based on the single-box capacity of the SKU combination under the optimization result, the optimization result of the collaborative consolidation SKU is determined.

[0027] Furthermore, the deviation in the quantity of goods corresponding to the SKU within a single box is determined based on the deviation between the quantity of goods corresponding to the SKU within a single box and the quantity of goods corresponding to other SKUs.

[0028] Furthermore, the method for determining the recommended value of the complexity level is as follows:

[0029] Determine the number of SKUs based on the SKU data;

[0030] Based on the deviation in the quantity of goods corresponding to the SKU in a single box, determine the quantity of SKUs in different quantity ranges;

[0031] The recommended value for the complexity level is determined based on the number of SKUs and the number of SKUs in different capacity ranges.

[0032] The beneficial effects of this invention are as follows:

[0033] Compared with existing technologies, the collaborative consolidation global optimization method and system provided by this invention have the following significant advantages:

[0034] Improving AOSS processing efficiency and significantly reducing logistics costs: Through one-time global optimization within the group and complexity-prioritized iteration, SKU demand can be utilized more fully to find LCL (Less than Container Load) solutions with smaller MSS (Maximum Size) per unit, thereby sending more orders to automated production lines. Practical applications show that AOSS processing efficiency can be improved by approximately 6.8 percentage points, and the unit logistics cost is significantly reduced.

[0035] The optimization solution is highly efficient and has strong engineering feasibility: It combines high-quality candidate generation based on the "shortest board principle + integer partitioning", intelligent pruning based on "Top-K + upper bound pruning", and robust solution based on "initial solution hints + hierarchical time limits". While ensuring optimization quality, it effectively controls computational complexity, making it possible to solve large-scale problems quickly.

[0036] Enhanced operational stability and production line balance: Batch consistency constraints ensure that multiple boxes with completely identical structures can be produced under the same scheme, reducing production line changeovers and manual intervention. The selectable maximum number of boxes constraint prevents overproduction of a single scheme, contributing to production line load balancing and cycle time stability.

[0037] Automated decision-making and efficiency improvement: The LCL planning process, which originally required hours of manual meetings for decision-making, has been compressed into automated processing at the minute level, greatly improving the decision-making efficiency and response speed of warehousing operations.

[0038] The system has good scalability: the system architecture and algorithm design support flexible expansion and are easy to integrate with new business constraints (such as brand, wave), multi-objective optimization (such as balanced box count) and advanced requirements such as capacity planning.

[0039] Other features and advantages will be set forth in the following description, and the objects and other advantages of the invention are realized and obtained through the structures particularly pointed out in the description and the drawings.

[0040] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0041] The above and other features and advantages of the present invention will become more apparent from a detailed description of exemplary embodiments thereof with reference to the accompanying drawings.

[0042] Figure 1 This is a framework diagram of a collaborative consolidation global optimization system;

[0043] Figure 2 It is a flowchart for generating all in-box allocation schemes that satisfy "at least one item per SKU";

[0044] Figure 3 This is a flowchart illustrating the method for obtaining a set of candidate solutions;

[0045] Figure 4 This is a flowchart of a collaborative consolidation global optimization method. Detailed Implementation

[0046] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0047] Example 1

[0048] like Figure 1 As shown, this application provides a collaborative consolidation global optimization system, specifically including:

[0049] The modeling module is responsible for enumerating all SKU combinations with sizes ranging from 1 to the upper limit k of the preset complexity level according to the preset business constraints. For each SKU combination, the single box capacity is determined according to the "shortest board principle", and all box allocation schemes that satisfy "at least one piece of each SKU" are generated.

[0050] The scoring module is responsible for calculating the maximum number of boxes that can be produced by each allocation scheme within a box, and applying batch consistency constraints and maximum number of boxes constraints to filter and obtain feasible schemes. Feasible schemes are scored by "number of boxes that can be produced × capacity of a single box". The "Top-K streaming pruning" strategy is adopted, using a min-heap to dynamically maintain the top K candidate schemes with the highest scores and control the size of the candidate schemes.

[0051] The solution module is responsible for inputting the selected candidate solutions into the optimization solver, constructing the optimization solver using a greedy strategy and setting a maximum solution time limit for each level, executing an iterative strategy that prioritizes high complexity, recording the remaining SKU requirements corresponding to each complexity level, and selecting the optimization solution with the lowest overall remaining SKU requirements as the final output.

[0052] Furthermore, the SKUs to be processed need to be strategically grouped according to preset constraint rules before the SKU combination enumeration process is performed.

[0053] Specifically, the preset constraint rules are store, product number, logistics method, and box size.

[0054] Furthermore, the upper limit k of the preset complexity level is the number of SKUs in a single box.

[0055] Specifically, the shortest-board principle refers to the minimum number of SKUs that a single box can hold.

[0056] Furthermore, such as Figure 2 As shown, generate all in-box allocation schemes that satisfy "at least one item per SKU", specifically including:

[0057] Based on the enumeration range of all SKU combinations from a single SKU (size=1) to the preset complexity upper limit k, we enumerate to obtain SKU combinations;

[0058] For each enumerated SKU combination, its theoretical maximum capacity per box is determined according to the "shortest board principle";

[0059] The specific operation involves extracting the packing quantity parameter for all SKUs within the combination and taking the minimum value as the uniform packing capacity for the combination. This principle ensures that the generated allocation scheme can meet the physical packing constraints of all SKUs in actual packing operations, guaranteeing the physical feasibility of the scheme.

[0060] Given a fixed single-box capacity, a hard constraint of "at least one item per SKU" is added. An integer partitioning algorithm is used to generate a specific allocation scheme for the SKUs within the box, resulting in the box allocation scheme.

[0061] Given a fixed single-box capacity, an integer partitioning algorithm is used to generate a specific allocation scheme for the SKUs within the box. This process includes a hard constraint that "each SKU must be allocated at least once" to ensure that all SKUs within the combination are represented and to avoid invalid zero allocations. The algorithm systematically enumerates all positive integer solutions that allocate the total capacity (i.e., the single-box capacity) to each SKU, forming a complete set of allocation patterns that conform to basic business rules.

[0062] Furthermore, the system caches the calculated splitting results for SKU combinations, so that when the same parameters are encountered later, they can be read directly from the cache.

[0063] Furthermore, such as Figure 3 As shown, the method for obtaining the candidate solution set is as follows:

[0064] For each SKU involved in each generated in-box allocation scheme, calculate the ratio of its initial demand to the number of boxes allocated to that SKU in the scheme, round down, and take the minimum value of the calculation results for all related SKUs as the maximum number of boxes that the allocation scheme can produce without other constraints.

[0065] The calculated maximum number of boxes that can be produced is compared with the system's preset batch consistency threshold. Only when the number of boxes that can be produced is greater than or equal to the batch consistency threshold is the solution considered to meet the batch stability requirements of the production line operation, thus obtaining a solution that meets the batch consistency constraint. For solutions that meet the batch consistency constraint, they are further compared with the system's optional parameter—the upper limit of the maximum number of boxes. The number of boxes that can be produced in the solution must not exceed this upper limit, thus obtaining a feasible solution.

[0066] Each feasible solution is assigned a score, defined as the product of "number of boxes that can be produced" and "capacity per box". A priority queue-based streaming pruning mechanism is used, initializing a min-heap of capacity K to dynamically maintain the K solutions with the highest scores among the currently processed solutions. When processing each feasible solution, its score is compared with the top of the min-heap (i.e., the score of the current Kth solution). Based on the comparison result, it is determined whether to insert it into the heap and may discard the top solution.

[0067] During the streaming process, when the min-heap is full (i.e., K solutions have been included), for the SKU combination to which the newly arrived solution belongs, the theoretically highest possible score upper bound of the combination can be calculated in advance. If this theoretical upper bound is lower than the current top score of the heap, it can be determined that all unprocessed allocation solutions in the combination have no chance of entering the Top-K set. After the process is completed, the K feasible solutions stored in the min-heap constitute the final candidate solution set submitted to the optimization solver within the strategic group.

[0068] Specifically, step one: calculate the maximum number of boxes that can be produced under the allocation scheme:

[0069] For each generated in-box allocation plan, its theoretical maximum production batch size is calculated based on the initial demand of each SKU within the group. Specifically, for each SKU involved in the allocation plan, the initial demand is calculated as the quotient of the initial demand and the number of boxes allocated to that SKU in the plan, rounded down. The minimum value among all relevant SKUs is taken as the maximum number of boxes that the allocation plan can produce without other constraints. This calculation ensures the feasibility of the plan at the material level.

[0070] Step 2: Apply batch consistency constraints for filtering:

[0071] The calculated maximum number of boxes that can be produced is compared with the system's preset batch consistency threshold (min_boxes_for_aoss). A solution is considered to meet the batch stability requirements of the production line only if its number of producible boxes is greater than or equal to this threshold. This constraint ensures that the adopted solution has value for large-scale production, avoiding frequent production line changeovers and efficiency losses due to excessively small batch sizes.

[0072] Step 3: Apply the optional maximum bin number constraint for secondary filtering:

[0073] For solutions that meet batch consistency constraints, they are further compared with the system's optional parameter—the maximum number of boxes (max_boxes_for_aoss). If this parameter is enabled, the number of boxes that the solution can produce must not exceed this limit. This constraint is mainly used to balance production line load, prevent a single solution from excessively occupying capacity, and ensure the balance of the overall work plan. Solutions that pass this step are marked as "feasible solutions".

[0074] Step 4: Calculate the score for feasible solutions:

[0075] Each feasible solution is assigned a quantitative score, defined as the product of "number of boxes that can be produced" and "capacity per box". This score intuitively represents the total number of SKUs that the solution can handle if fully implemented, i.e., its direct contribution to the optimization objective—maximizing total throughput. The score serves as a single numerical measure of solution quality, providing a basis for subsequent comparison and selection.

[0076] Step 5: Implement the Top-K streaming pruning strategy:

[0077] To control the size of candidate solutions input to the final optimization model, a priority queue-based streaming pruning mechanism is adopted. A min-heap of capacity K is initialized to dynamically maintain the K highest-scoring solutions among those processed so far. When processing each feasible solution, its score is compared with the top of the min-heap (i.e., the score of the current Kth solution), and the comparison result determines whether to insert it into the heap and may evict the top-ranked solution.

[0078] Step Six: Enhanced pruning by integrating upper bound pruning:

[0079] During streaming processing, when the min-heap is full (i.e., K solutions have been included), for the SKU combination to which a new solution belongs, the theoretically highest possible score bound for that combination can be pre-calculated. If this theoretical bound is lower than the current top score of the heap, it can be determined that all unprocessed allocation solutions within that combination have no chance of entering the Top-K set, thus prematurely terminating further enumeration and evaluation of that combination. This strategy effectively avoids unnecessary computation of a large number of low-potential solutions.

[0080] Step 7: Output the cropped, high-quality candidate set:

[0081] At the end of the process, the K feasible solutions stored in the min-heap constitute the final set of candidate solutions submitted to the optimization solver within that strategic group. This set is controlled in size and theoretically consists of the highest-scoring solutions within the current group (potentially near-optimal under specific conditions of upper-bound pruning). This compresses the potentially exponentially growing candidate space into a fixed-size, high-quality subset, fundamentally ensuring computational feasibility in subsequent optimization stages.

[0082] Furthermore, the batch consistency threshold ranges from 3 to 10, and is specifically determined based on the batch stability requirements of the production line operation.

[0083] Furthermore, the value of k ranges from 1000 to 2000, and the specific value is determined according to the user's settings.

[0084] Furthermore, the high-complexity-first iteration strategy starts with the highest allowed complexity k and iterates downwards sequentially.

[0085] Specifically, the method for determining the final output scheme is as follows:

[0086] A greedy heuristic algorithm is used to process the filtered candidate solution set to quickly construct a feasible initial solution. Based on the current complexity level, a differentiated maximum running time limit is set for the optimization solver.

[0087] The system executes the complete optimization process independently for different complexity limits in descending order. At each complexity level, the candidate set is regenerated based on the combinatorial enumeration constraints of that level and optimized independently to obtain the optimal binning scheme under that constraint.

[0088] After the optimization solution for each complexity level is completed, the system calculates the initial demand that has not yet been met for each SKU in the group after adopting its output solution, i.e. the remaining manual processing amount. The remaining demand of all SKUs is summarized to form the overall MSS index under that complexity level.

[0089] After completing independent optimization and MSS calculation for all preset complexity levels, the system compares the overall MSS index corresponding to each complexity level and selects the optimization scheme corresponding to the complexity level that minimizes the index value as the final output scheme for the current strategic group.

[0090] Furthermore, a greedy heuristic algorithm is used to process the filtered candidate solution set, specifically including:

[0091] Based on the scoring density or simple priority rules of the solutions, the solution is iteratively selected until the resources are exhausted. The generated initial solution is provided as "hot start" information to the subsequent optimization solver.

[0092] Specifically, including:

[0093] Step 1: Construct an initial feasible solution based on a greedy strategy:

[0094] Before formally invoking the optimization solver, a greedy heuristic algorithm is first used to process the filtered set of candidate solutions to quickly construct a feasible initial solution. This algorithm typically iterates through the selection of solutions based on the score density or simple priority rules until resources are exhausted. The generated initial solution serves as "warm-start" information for subsequent optimization solvers, aiming to guide the search direction, accelerate solution convergence, and increase the probability of finding a high-quality feasible solution within the time limit.

[0095] Step 2: Configure the hierarchical maximum solution time limit mechanism:

[0096] Based on the current level of complexity, differentiated maximum runtime caps are set for the solver. Typically, longer computation time is allocated to solutions of higher complexity because they have a larger candidate space and are more complex; shorter time is allocated to solutions of lower complexity. This tiered time cap strategy is a key computational resource allocation method designed to ensure that the solver has sufficient time to explore the search space of complex solutions, while avoiding the impact on overall system throughput and response performance due to excessively long solve times for individual instances.

[0097] Step 3: Perform iterative optimization using a "high complexity first" approach:

[0098] The system executes the complete optimization process (including candidate generation, screening, and solution) independently for different complexity limits (from k_max to 1) in descending order. At each complexity level, a candidate set is regenerated based on the combinatorial enumeration limit of that level (i.e., a single bin can mix at most the number of SKUs of that complexity) and independently optimized to obtain the optimal (or near-optimal) binning scheme under that constraint.

[0099] Step 4: Calculate and record the remaining requirements for each complexity level:

[0100] After the optimization solution for each complexity level is completed, the system accurately calculates the initial unmet demand for each SKU within the group after adopting its output solution, i.e., the remaining manual processing volume (MSS). The remaining demands of all SKUs are then aggregated to form the overall MSS metric for that complexity level. This metric is the core measure of the scale of "unmet needs" in the optimization solutions at that level.

[0101] Step 5: Final decision based on the minimum residual demand criterion:

[0102] After completing independent optimization and MSS calculation for all preset complexity levels, the system compares the overall MSS metric for each complexity level. The optimization scheme corresponding to the complexity level that minimizes this metric value is selected as the final output scheme for the current strategic group. This decision-making criterion directly serves the core business objective—maximizing automated processing, i.e., minimizing the remaining demands that must rely on manual processing.

[0103] Step Six: Technology Integration and Output

[0104] Finally, the system integrates and outputs the complete optimization results for the selected complexity level, including: a list of specific leasing schemes adopted, the number of boxes produced by each scheme, the total AOSS processing rate, a detailed MSS list, and the optimal complexity value to achieve this result. This closed-loop process ensures that, under multiple constraints and computational resource limitations, the system can automatically explore different complexity spaces and deliver a job scheme that achieves the best overall performance while minimizing the remaining demand objective.

[0105] In one possible specific implementation, in-warehouse consolidation optimization for footwear products:

[0106] Suppose a footwear warehouse needs to process a batch of orders from the same store using the same logistics method, involving 10 SKUs. The initial demand and the number of pieces per carton are shown in the table below:

[0107] Table 1 SKU Data

[0108]

[0109] System parameter settings:

[0110] Business constraints: Same store, same logistics method, same box size.

[0111] The upper limit of complexity is k = 3.

[0112] Batch consistency constraint min_boxes_for_aoss = 5.

[0113] Maximum number of boxes constraint max_boxes_for_aoss = 20 (optional).

[0114] The upper limit for candidate retention is K = 1000.

[0115] Optimization process:

[0116] Strategic Grouping: All SKUs that meet the same business constraints are grouped into the same strategic group.

[0117] Candidate generation and pruning (taking a complexity of 3 as an example):

[0118] The system enumerates all SKU combinations of size 1, 2, and 3. For example, the combination {A, B, C}.

[0119] Determine the capacity of a single box: Under the principle of the shortest board, take min(20, 16, 30) = 16 pairs / box.

[0120] Integer partitioning: Generate all partitioning schemes that sum to 16 and where A, B, and C are each allocated at least one pair, such as (A:7, B:5, C:4), (A:6, B:6, C:4), etc.

[0121] Calculate the number of boxes that can be produced: For scheme (A:7, B:5, C:4), the number of boxes that can be produced = min(100 / / 7, 80 / / 5, 150 / / 4) = min(14, 16, 37) = 14 boxes.

[0122] Filtering and Scoring: If 14 >= 5 and 14 <= ​​20, pass. Score = 14 boxes * 16 pairs / box = 224.

[0123] While generating candidate combinations, the system continuously maintains a min-heap of up to 1000 candidates. For candidates with low scores or combinations with low theoretical bounds, real-time pruning is performed.

[0124] Global optimization solution: Construct a 0-1 integer programming model from the final retained candidate set (e.g., 800 high-quality candidates). First, run a greedy algorithm (e.g., prioritize candidates with high scores until the requirement cannot be met) to obtain a feasible solution as the initial solution. Then, call the solver, setting a maximum time limit of 30 seconds, to perform precise optimization solution.

[0125] Complexity Iteration: The system first completes the above optimization at k=3, recording the MSS (e.g., remaining SKUs D: 10 pairs, E: 15 pairs). Then, it repeats the independent optimization at k=2 and k=1. Assume that the total MSS is minimized when k=2.

[0126] Output: The system ultimately outputs the optimal consolidation solution set with k=2 complexity, the number of boxes produced by each solution, and a detailed MSS list. The warehouse operations system uses this solution for picking, packing, and AOSS deployment.

[0127] Example 2

[0128] Secondly, such as Figure 4 As shown, this invention provides a collaborative consolidation global optimization method, applied to the aforementioned collaborative consolidation global optimization system, specifically including:

[0129] S1 acquires SKU data and, based on the deviation in the quantity of goods corresponding to the SKU within a single box, determines the recommended value for the complexity level.

[0130] S2 performs global optimization processing for collaborative consolidation based on the recommended values. According to the single-box capacity and overall MSS index of different SKU combinations under the existing complexity level, the optimization processing scheme of the recommended values ​​at the complexity level is determined.

[0131] S3 uses the optimization processing scheme to optimize the recommended value to obtain the optimization result, and determines the optimization result of the SKU of the collaborative consolidation based on the single box capacity of the SKU combination under the optimization result.

[0132] Specifically, the SKU data includes the number of SKUs.

[0133] Furthermore, the deviation in the quantity of goods corresponding to the SKU within a single box is determined based on the deviation between the quantity of goods corresponding to the SKU within a single box and the quantity of goods corresponding to other SKUs.

[0134] Specifically, the method for determining the recommended value of the complexity level is as follows:

[0135] This embodiment aims to solve the engineering problem of "how to dynamically recommend a reasonable complexity level (k value) for a collaborative consolidation optimization system". Its core logic is to adaptively predict a k value that ensures optimization effectiveness while controlling computational complexity by analyzing the size (number) and internal structure (distribution of the number of SKUs carried per container) of the set of SKUs to be processed. This logic avoids the "one-size-fits-all" problem caused by a fixed k value, achieving intelligent system configuration.

[0136] S11 uses the SKU data to determine the number of SKUs;

[0137] Basic data input and analysis, key terms explained: SKU quantity is the total number of product types to be processed; single-carton capacity refers to the maximum number of pieces required to fill a standard carton for each SKU, which is the physical basis of LCL (Less than Container Load) shipping.

[0138] The number of SKUs directly determines the size of the search space for the combinatorial optimization problem and is the primary indicator for evaluating computational complexity. Obtaining the capacity of a single container is a prerequisite for subsequent analysis of distribution characteristics; the difference in capacity among different SKUs directly affects the feasibility and efficiency of mixed packing. The significance of this step lies in forming a quantitative understanding of the input problem from both scale and physical characteristics dimensions.

[0139] Example: Suppose a batch of footwear from the same store needs to be consolidated into one box, involving 12 SKUs (SKU quantity = 12). The per-box capacity for each SKU is as follows: SKU_A (20 pairs / box), SKU_B (20 pairs / box), SKU_C (30 pairs / box), SKU_D (12 pairs / box), SKU_E (24 pairs / box), SKU_F (12 pairs / box), SKU_G (30 pairs / box), SKU_H (16 pairs / box), SKU_I (20 pairs / box), SKU_J (24 pairs / box), SKU_K (30 pairs / box), SKU_L (16 pairs / box).

[0140] S12 determines the quantity of SKUs in different quantity ranges based on the deviation of the quantity of goods corresponding to the SKU in a single box;

[0141] Statistics on the distribution range of carrying capacity:

[0142] Keyword Explanation: The carrying capacity range divides the carrying capacity value of all SKUs into several continuous ranges (e.g., 0-10, 11-20, 21-30, etc.); the number of SKUs in different carrying capacity ranges is counted, indicating the number of SKU types in each range.

[0143] Discretizing the continuous carrying capacity into different intervals is to quantify the distribution structure of the SKU group in terms of physical capacity. If the carrying capacity of all SKUs is concentrated in one or two adjacent intervals, it indicates that their capacities are similar, the "weakest link effect" is not significant when mixing containers, and it is easy to find a balanced allocation scheme. Conversely, if the distribution is very scattered, it means that the capacity differences are large, and more refined (higher complexity) combination exploration is needed to fully utilize the container space when mixing containers. The significance of this step is to transform physical characteristics into analyzable structured data, providing a basis for judging the "mixing difficulty" of the problem.

[0144] For example, the carrying capacity of the above 12 SKUs is divided into three ranges: small capacity (10-19 pairs / box), medium capacity (20-29 pairs / box), and large capacity (30 pairs / box and above). Statistics show that: the small capacity range has SKUs_D, F, H, and L, a total of 4; the medium capacity range has SKUs_A, B, E, I, and J, a total of 5; and the large capacity range has SKUs_C, G, and K, a total of 3. Therefore, the number of "effective quantity ranges" (ranges containing SKUs) is 3.

[0145] S13 determines the recommended value for the complexity level based on the number of SKUs and the number of SKUs in different capacity ranges.

[0146] It should be noted that if the number of SKUs is less than the preset quantity threshold, the complexity level is determined by directly multiplying the first preset ratio by the number of SKUs.

[0147] Rapid decision-making in large-scale scenarios. Key terms explained: The preset quantity threshold is an empirical value used to define "small-scale problems".

[0148] When the number of SKUs is small, the combinatorial enumeration size of the problem is inherently limited, and even with a high complexity value k, the computational cost is acceptable. In this case, using a large initial pre-defined ratio (such as 0.4 or 0.6) multiplied by the number of SKUs to determine k ensures sufficient exploration capability to find high-quality solutions, avoiding premature pruning of potentially good solutions due to an excessively small k value. Its significance lies in adopting a lenient strategy for small problems, prioritizing optimization quality.

[0149] For example (assuming the preset quantity threshold is 5): In our example, the number of SKUs is 12 > 5, so this branch is not entered.

[0150] Additionally, it can be understood that if the number of SKUs is not less than a preset number threshold, the number of SKUs in different carrying number intervals is obtained, and the carrying number interval containing SKUs is taken as the effective number interval. It is then determined whether the number in the effective number interval is greater than the preset interval number threshold. If so, the complexity level is determined by directly multiplying the second preset ratio by the number of SKUs. If not, the process proceeds to the next step.

[0151] Rapid decision-making for extremely dispersed or clustered distributions. Key terms explained: Preset interval quantity thresholds are used to determine whether the distribution is "extremely dispersed"; the distribution dispersion factor is an indicator that measures the proportion of SKUs in a certain interval to the total number of SKUs. The lower the proportion, the smaller the dispersion factor, indicating that the SKUs in that interval are very "sparse".

[0152] Scenario A (Too Many Effective Intervals): If the SKU capacity is scattered across many different intervals, it indicates significant differences in physical characteristics. A highly mixed (high k-value) LCL (Less than 1) solution is likely to result in a low overall loading rate due to the weakest link principle, which would be counterproductive. In this case, the system tends to use lower complexity (still using the second preset ratio) to guide the algorithm to prioritize finding SKU combinations within the same or adjacent capacity intervals. This aligns better with engineering intuition and efficiency principles.

[0153] Scenario B (High Degree of Distribution Clustering): If most SKUs are concentrated in a few adjacent capacity ranges (i.e., there are no ranges with very small dispersion factors), it indicates that their capacities are similar, and the "natural synergy" of mixed packing is good. In this case, even with lower complexity (using a smaller second preset ratio), it is possible to find a solution with a high loading rate with a higher probability, thereby significantly reducing computational costs and improving system response speed.

[0154] Example: Assuming the preset threshold for the number of intervals is 5, in our example, the number of valid intervals = 3 < 5, so proceed to the next step. Calculate the distribution dispersion factor (i.e., the proportion of SKUs within each interval): small-capacity intervals have a proportion of 4 / 12 ≈ 0.33, medium-capacity intervals have a proportion of 5 / 12 ≈ 0.42, and large-capacity intervals have a proportion of 3 / 12 = 0.25. Assuming the preset dispersion factor threshold is 0.2, then all interval proportions are greater than 0.2, and there are no "discrete intervals". At this point, the system determines that the distribution clustering degree is high, and directly uses the second preset ratio (assumed to be 0.4) to calculate the k value: 12 * 0.4 = 4.8, rounding down to recommend a complexity level of k=4.

[0155] The distribution dispersion factor of the effective quantity interval is determined by the proportion of the number of SKUs in different effective quantity intervals in the total number of SKUs. It is then determined whether there are effective quantity intervals with a distribution dispersion factor less than a preset dispersion factor threshold. If so, proceed to the next step. If not, the distribution clustering degree is high at this time, so a smaller complexity level can be used for collaborative optimization calculation, that is, the complexity level is determined by directly using the product of the second preset ratio and the number of SKUs.

[0156] The effective number interval of the distribution discrete factor less than the preset discrete factor threshold is taken as the discrete interval, and the recommended value of the complexity level is determined based on the constituent data of the discrete interval within the effective number interval.

[0157] It should be noted that the second preset ratio is smaller than the first preset ratio.

[0158] Specifically, if the proportion of the discrete interval within the effective quantity range is less than a preset proportion threshold, the complexity level is determined by directly multiplying the second preset proportion by the number of SKUs. If the proportion of the discrete interval within the effective quantity range is not less than the preset proportion threshold, the complexity level is determined by directly multiplying the first preset proportion by the number of SKUs.

[0159] It handles complex distributions with sparse discrete intervals, where the discrete interval refers to the interval with a distribution discrete factor less than a threshold and a sparse number of SKUs; a preset percentage threshold is used to determine the influence range of the sparse interval.

[0160] When there are individual SKUs (sparse intervals) with "outlier" capacity, their impact needs to be assessed more precisely. If there are very few outlier SKUs (the proportion is less than the threshold), their impact on overall optimization is limited and can be ignored, so a lower second preset proportion is adopted. If there are many outlier SKUs (the proportion is not less than the threshold), it indicates the existence of a subgroup with different capacity characteristics that cannot be ignored. In order to fully explore the possibilities of mixing within this subgroup and with the mainstream group, a higher degree of exploration freedom is needed, so a moderate preset proportion (first preset proportion) is adopted. The significance is to finely distinguish the impact of "outliers" and achieve a reasonable allocation of computing resources between the mainstream and special groups.

[0161] For example (assuming discrete intervals exist): If the proportion of large-capacity intervals (0.25) is less than the discrete factor threshold of 0.2, then it is marked as a discrete interval. If the preset proportion threshold is 0.3, then the proportion of discrete intervals within the effective intervals is 1 / 3 ≈ 0.33 > 0.3. In this case, the system determines that the discrete intervals have a significant impact and calculates using a preset proportion (assumed to be 0.6): 12 * 0.6 = 7.2, rounding down to recommend k=7.

[0162] The complexity-level recommendation algorithm demonstrated in this embodiment is an intelligent bridge that transforms business insights (physical carrying capacity distribution) into algorithm parameters (complexity k). Through multi-level analysis of SKU data (scale judgment → distribution statistics → discreteness assessment), its core value lies in:

[0163] Improve optimization efficiency: Avoid unnecessary calculations caused by configuring too high a k value for simple problems (small scale or high clustering), or a decrease in solution quality caused by configuring too low a k value for complex problems.

[0164] Enhanced system adaptability: Enables the consolidation optimization system to intelligently respond to business scenarios of different scales and characteristics without the need for repeated manual adjustments to the k-value parameter.

[0165] Balancing effectiveness and performance: A data-driven, dynamic trade-off was made between solution quality (requiring sufficient exploration complexity) and computation time (complexity is directly related to computational load), ensuring the overall availability and responsiveness of the system in complex industrial scenarios.

[0166] Specifically, the existing complexity level refers to the complexity level that has already been solved.

[0167] Furthermore, the single-box capacity of the SKU combination is determined based on the capacity utilized by the single box of the SKU combination.

[0168] Specifically, the method for determining the optimization scheme for the recommended value of the complexity level is as follows:

[0169] This embodiment aims to demonstrate how the system utilizes the optimization results at different complexity levels (K values) to perform multi-dimensional and multi-level analysis and diagnosis, thereby dynamically and intelligently adjusting the recommended K values ​​for similar future tasks. Its core logic lies in using historical optimization results as a "clinical diagnostic report." By examining the two key "symptoms" of "low space utilization" and "degree of unmet needs," and their performance patterns at different K values, the system infers the root cause of the problem and prescribes a targeted "prescription" (adjusting the K value), thus achieving continuous self-optimization of the algorithm parameters.

[0170] S21 determines the proportion of remaining space in a single box for different SKU combinations based on the single box capacity of the existing complexity level.

[0171] S21: Calculate the remaining space ratio of a single container:

[0172] Keyword Explanation: Existing complexity levels refer to the set of K values ​​that have undergone optimization calculations (e.g., K=5,4,3). The single-container remaining space ratio is calculated for a specific adopted LCL (Less than Container Load) scheme, representing the ratio of the actual number of loaded items to the scheme's theoretical maximum single-container capacity (determined by the shortest board principle). A lower ratio indicates more efficient space utilization.

[0173] This metric directly measures the space utilization of a less-than-container-load (LCL) solution. A high percentage of remaining space indicates underfilling, resulting in wasted capacity. Analyzing the distribution of this metric under different K values ​​is the primary basis for determining whether the current K value leads to ineffective mixing or failed combination searches.

[0174] For example: The system analyzes the optimization results under K=4 and calculates for each adopted scheme. It is found that the theoretical capacity of scheme P01 (containing SKU_A and SKU_B) is 20 pieces, but 15 pieces are actually loaded according to the allocation scheme, and the remaining space ratio is (20-15) / 20=25%.

[0175] S22 determines the initial demand quantity that has not yet been met for each SKU based on the overall MSS index, and determines the impact value of the SKU based on the ratio of the initial demand quantity that has not yet been met to the initial demand quantity.

[0176] Calculate the impact value for each SKU:

[0177] Key terms explained: The overall MSS metric is the sum of all remaining demand after optimization. The SKU impact value is defined as the percentage of remaining demand for that SKU relative to its initial demand.

[0178] This metric quantifies the severity of each SKU being "left behind." SKUs with high impact values ​​are the "hard nuts to crack" in the optimization process, potentially due to their inherent characteristics (such as specific load capacity) making them difficult to pair with other SKUs. Identifying high-impact SKUs helps pinpoint optimization challenges and determine whether adjusting the K value is necessary to change the handling strategy for these SKUs.

[0179] For example, in the case of K=4, SKU_X has an initial demand of 100 units and 40 units remaining, with an impact value of 40%; SKU_Y has an initial demand of 50 units and 40 units remaining, with an impact value as high as 80%. The impact value of SKU_Y is significantly higher.

[0180] S23 determines the optimization scheme for the recommended value of the complexity level based on the impact value of each SKU and the proportion of remaining space in a single box of SKU combinations.

[0181] It should be noted that the optimization scheme for determining the recommended value of the complexity level based on the impact value of each SKU and the proportion of remaining space in a single bin for SKU combinations specifically includes:

[0182] S231 Obtain the remaining space ratio of a single box of SKU combinations under the existing complexity level, and determine whether there is an SKU combination whose remaining space ratio of a single box is greater than the preset space ratio threshold. If yes, proceed to the next step; otherwise, proceed to step S233.

[0183] S232 determines whether the number of SKU combinations with a remaining space ratio greater than a preset space ratio threshold under all existing complexity levels is greater than a preset combination number threshold. If yes, then the optimization processing scheme for determining the recommended value of the complexity level using the preset optimization scheme is determined. If no, then proceed to the next step.

[0184] S231 & S232: Diagnosis of the prevalence of wasted space:

[0185] Keyword explanation: The preset space ratio threshold (e.g., 20%) is used to define "significant waste"; the preset combination quantity threshold is used to determine whether waste is "widespread".

[0186] S231 identifies the "symptom," while S232 performs a cross-valued comparison across different K values ​​to determine if the "symptom" is an inherent characteristic of the system. If the number of wasteful solutions is high (exceeding the threshold) across all tried K values, it strongly suggests that the root cause of the space waste lies in the characteristics of the order data itself (such as significant differences in SKU capacity), rather than an improperly set specific K value. In this case, increasing the K value (pre-defined optimization scheme, factor > 1) might allow for a mix of more diverse SKUs, potentially exacerbating the bottleneck effect and leading to even greater waste. Therefore, this branch directly leads to an adjustment (which could be increasing K to explore a wider range, or it could be other processing methods).

[0187] For example: Suppose the preset space ratio threshold is 20%, and the preset combination number threshold is 15. Inspection revealed that in the results with K=5, K=4, and K=3, the number of schemes with a remaining space ratio > 20% were 18, 22, and 17 respectively, all exceeding 15. The system determined that there was widespread space waste.

[0188] S233 checks whether the impact value of each SKU at the existing complexity level is greater than the preset impact threshold. If so, it determines the optimization processing scheme to determine the recommended value of the complexity level using the preset optimization scheme. If not, it proceeds to the next step.

[0189] Diagnosis of unmet overall needs:

[0190] Keyword Explanation: The preset impact threshold (e.g., 70%) is used to define "severe impact".

[0191] This step checks for the extreme case where all SKUs are severely unmet. If so, it indicates that the optimization algorithm as a whole has failed under the current problem structure and all tried K values, and cannot effectively digest any SKUs. This usually means that the problem constraints are extremely strong or the data is extremely mismatched. In this case, an aggressive strategy is needed (pre-set optimization scheme, significantly adjust the K value, such as significantly increasing it to maximize exploration, or significantly decreasing it to simplify the problem) to try to break the deadlock.

[0192] For example: the check found that when K=4, not all SKUs have an impact value >70% (as in the example above, only SKU_Y>70%), so this branch is not entered.

[0193] S234 determines the analysis deviation level in the complexity level based on the number of SKUs whose impact value is greater than a preset impact threshold under different complexity levels, and determines the optimization processing scheme of the recommended value of the complexity level based on the analysis deviation level data.

[0194] It should be noted that the analysis deviation level is the complexity level where the proportion of SKUs with an impact value greater than a preset impact threshold is greater than a preset proportion threshold.

[0195] It is understood that if there is no analysis deviation level, then it is determined that there is no need to optimize the recommended value of the complexity level. If there is an analysis deviation level, then if the analysis deviation level is greater than the preset deviation level number threshold, then it is determined that a preset optimization scheme is used to determine the optimization scheme for the recommended value of the complexity level. Otherwise, it is determined that a second preset optimization scheme is used to determine the optimization scheme for the recommended value of the complexity level.

[0196] Based on fine-tuning of the "analysis deviation level," the analysis deviation level refers to those complexity levels where the proportion of high-impact SKUs exceeds a preset threshold (e.g., 30%) at a given K value. The preset deviation level threshold is used to determine whether such "poorly performing" K values ​​constitute the majority.

[0197] This is the most sophisticated diagnosis. It no longer looks at absolute numbers, but rather at proportions, to identify which K values ​​are relatively less adept at handling the current set of SKUs.

[0198] If most K values ​​are at the analysis bias level, it indicates that the optimizer has left a large number of "difficult" SKUs in most settings. This may suggest a need to change the optimization paradigm, adopting a pre-defined optimization scheme (e.g., increasing the K value in 1.4) to give the algorithm greater combinatorial freedom to try to accommodate these difficult SKUs.

[0199] If only a few K values ​​are at the analysis bias level, it indicates that the problem is selective, and some K values ​​(such as the middle K=4) may fall into a local optimization trap. In this case, adopting a second preset optimization scheme (such as using a scaling factor slightly less than 1 to slightly reduce the K value) may avoid the inefficient area. If there are no analysis bias levels, it means that all K values ​​perform relatively evenly in meeting the requirements, and no adjustment is needed.

[0200] For example: Assume the preset percentage threshold is 30%. In the three levels K=5, 4, and 3, the percentages of high-impact (>70%) SKUs are 10%, 40%, and 15%, respectively. Therefore, only K=4 is marked as an "analysis deviation level." Since the number of analysis deviation levels (1) is less than the preset deviation level threshold (e.g., 2), the system determines the problem is not widespread and adopts the second preset optimization scheme. Assuming the original recommended K value is 4 and the second preset percentage factor is 0.8, the new recommended K value is 4 * 1.2 = 7, rounded up to K=5.

[0201] Specifically, the preset optimization scheme determines a new complexity level by multiplying a preset scaling factor by the original complexity level, and the second preset optimization scheme determines a new complexity level by multiplying a second preset scaling factor by the original complexity level.

[0202] It should be noted that the preset scaling factor is greater than the second preset scaling factor.

[0203] Preset optimization scheme and second preset optimization scheme:

[0204] Keyword Explanation: Both adjust the original complexity level by multiplying by a scaling factor, but the default scaling factor is greater than the second default scaling factor. Typically, the default scaling factor may be greater than 1 (aggressive adjustment), and the second default scaling factor may be greater than 1 (moderate adjustment).

[0205] This provides differentiated adjustment levels. The "preset optimization plan" addresses severe or widespread problems diagnosed, requiring significant adjustments. The "second preset optimization plan" addresses localized or minor problems, requiring only minor tweaks. This design ensures that the system's response matches the severity of the problem, avoiding instability caused by over-adjustment.

[0206] Summary and Value of Implementation Examples

[0207] This embodiment demonstrates a multi-level, coarse-to-fine intelligent decision-making loop:

[0208] From “symptoms” to “diagnosis”: The system does not react mechanically to a single indicator (such as remaining space), but integrates space utilization and demand satisfaction, and distinguishes between systemic problems, local problems and extreme problems through cross-level comparisons.

[0209] Differentiated “treatment” strategies: For different diagnostic results (general waste, comprehensive unmet needs, localized performance bias), the system matches adjustment strategies with different intensities (from radical adjustment to mild fine-tuning, or even no adjustment).

[0210] Achieving continuous self-optimization: This mechanism enables the system to learn from historical experience and automatically adapt to order batches with different characteristics (such as SKU carrying capacity distribution and demand scale), dynamically finding the most suitable complexity level K for the current task, thereby continuously improving the overall optimization efficiency and stability in long-term operation, which is something that fixed parameter systems cannot achieve.

[0211] Specifically, the method for determining the optimization results of the collaborative consolidation SKUs is as follows:

[0212] This embodiment aims to address a deep-seated problem in collaborative consolidation optimization: identifying and handling "problem SKUs" that consistently lead to significant space waste across multiple complexity levels. The core logic is to analyze the "empty combinations" (solutions with low space utilization) generated from optimization results with different K values ​​to identify recurring specific SKUs. If these SKUs are the "common root cause" of space waste, they are temporarily isolated from collaborative optimization and processed separately, thereby purifying the collaborative optimization environment and improving overall consolidation efficiency.

[0213] S41 determines the single-box capacity of SKU combinations at different complexity levels based on the single-box capacity of the SKU combinations under the optimization results.

[0214] Obtain single-container capacity data under multiple complexity levels. The single-container capacity of SKU combinations under different complexity levels (K values) refers to the theoretical maximum loading capacity of all LCL schemes generated and adopted by the algorithm under the corresponding K value constraints (determined by the minimum single-container capacity of SKUs in the combination).

[0215] This forms the basis of the analysis. To assess space utilization, we must first know the theoretical upper limit (single container capacity) for each scheme. Collecting this data at different K values ​​is for cross-sectional comparison, observing the capacity performance of the same batch of SKUs under different mixing constraints.

[0216] For example: The system has already optimized a certain order with K=5, K=4, and K=3. For the adopted solution under K=4, record its combination content and the corresponding single box capacity (e.g., the combination {A,B,C} has a capacity of 15 pieces).

[0217] S42 determines the SKU combinations whose remaining space ratio in the single box is greater than a preset space ratio threshold based on the single box capacity of the SKU combinations at different complexity levels, and treats them as vacant combinations.

[0218] Define and filter "empty combinations":

[0219] A preset space ratio threshold (e.g., 25%) is used to define "severely underutilized space". SKU combinations with a remaining space ratio greater than this threshold are marked as "empty combinations".

[0220] This is a crucial step in identifying problems. Idle combinations are "low-quality byproducts" in the optimization results, directly lowering the overall load factor and transportation efficiency. Identifying them separately makes them the target objects for subsequent root cause analysis. Their significance lies in accurately locating "inefficient units" from a massive number of solutions.

[0221] For example, in the result of K=4, the scheme {A,B,C} actually only contains 10 items, and the remaining space ratio = (15-10) / 15≈33.3%>25%, so it is marked as an empty combination.

[0222] S43 determines the optimization result of the collaborative binning SKU based on the overlap of SKUs in the empty combinations at different complexity levels.

[0223] Furthermore, based on the overlap of SKUs in empty combinations at different complexity levels, the optimization result of the collaborative binning SKUs is determined, specifically including:

[0224] S431 determines the vacancy ratio based on the proportion of vacancy combinations under different complexity levels in all SKU combinations, and judges whether the vacancy ratio under different complexity levels is less than the preset vacancy ratio threshold. If so, it is determined that there is no need to optimize the SKUs for collaborative consolidation; otherwise, it proceeds to the next step.

[0225] Preliminary screening (whether the vacancy problem is generally minor):

[0226] Keyword Explanation: Vacancy Rate = (Number of vacant combinations at a certain K value / Total number of combinations at that K value). A preset vacancy rate threshold (e.g., 5%) is used to determine whether the vacancy problem is "acceptable".

[0227] If the vacancy rate is low (less than the threshold) for all tried K values, it indicates that the space waste is sporadic and accidental, not a systemic problem. In this case, there's no need to initiate complex optimization processes, avoiding excessive intervention. This reflects the efficiency principle: only take action on truly serious problems.

[0228] For example, the calculated vacancy rates for K=5, 4, and 3 are 8%, 12%, and 7% respectively, all of which are greater than the preset threshold of 5%. Therefore, the problem cannot be ignored, and we will proceed to the next step.

[0229] S432 Based on the vacancy ratio at different complexity levels, determine whether the average vacancy ratio at different complexity levels is greater than the preset vacancy ratio value (greater than the preset vacancy ratio threshold). If yes, proceed to step S434; otherwise, proceed to the next step.

[0230] Assess the overall severity of the vacancy problem:

[0231] Calculate the average vacancy rate for all K values ​​and compare it with a higher "vacancy rate preset" (e.g., 15%).

[0232] This step aims to differentiate the "overall severity" of the problem. If the average vacancy rate is very high (greater than the preset value), it indicates that space waste is very serious regardless of how the K value is adjusted. This may point to a more fundamental problem (such as an overall mismatch in SKU capacity or abnormal order structure). In this case, it may be necessary to move beyond adjusting the K value, trigger a higher-level alarm, or adopt other global strategies (such as adjusting box size). This serves as a risk warning and strategy diversion mechanism.

[0233] For example: Average vacancy rate = (8% + 12% + 7%) / 3 = 9% < 15%, which does not reach an extremely serious level, so proceed to the detailed analysis step.

[0234] S433 determines whether there is a complexity range where the vacancy ratio is less than the preset vacancy ratio threshold. If so, it determines to optimize the SKU of collaborative packing by optimizing the k value, that is, to further perform the optimization process. If not, it proceeds to the next step.

[0235] The process involves finding a "safe zone," where the complexity interval refers to a continuous range of K values ​​(e.g., K from 3 to 4). It determines whether there exists an interval where the vacancy rate of all K values ​​is lower than a preset vacancy rate threshold.

[0236] If such a "safe range" can be found, it means there exists a range for setting the K value, within which co-optimization can consistently produce acceptable results. The optimal strategy then is to further refine the optimization within this range (e.g., trying different K values ​​with smaller step sizes), rather than focusing on specific SKUs. This is a lower-cost optimization path.

[0237] For example: If we check the interval from K=3 to K=5, the percentage of empty K values ​​(8%, 12%, 7%) is greater than the 5% threshold, so there is no "safe interval" and we need to proceed to the final diagnostic step.

[0238] S434 determines the complexity level of an SKU belonging to an vacant combination by the overlap of SKUs in vacant combinations under different complexity levels, and uses this as an interference level. Based on the interference level data, the optimization result of the collaborative binning SKU is determined.

[0239] It should be noted that if the proportion of interference levels in the complexity levels is greater than the preset interference level proportion threshold, then each time the SKU with the most interference levels is selected as the optimization target, that is, it is no longer used for collaborative optimization, but is stored and processed separately, and solved until the vacancy proportions under different complexity levels are all less than the preset vacancy proportion threshold. Otherwise, each time the SKU with the most interference levels is selected as the optimization target, that is, it is no longer used for collaborative optimization, but is stored and processed separately, and solved until the vacancy proportions under different complexity levels are all less than the preset vacancy proportion threshold.

[0240] It is understandable that the first quantity is greater than 1.

[0241] Root cause localization and isolation treatment (core):

[0242] Interference level refers to the set of K values ​​corresponding to the occurrence of a specific SKU in the empty combinations. For example, if SKU_X appears in both the empty combinations of K=4 and K=5, then its interference level set is {4,5}. A preset interference level proportion threshold (e.g., 50%) is used to determine whether the interference of an SKU is widespread.

[0243] This is the core innovation of this embodiment. By analyzing the overlap of SKUs in the empty combinations, "habitual offenders" can be identified—SKUs that frequently result in vacancies under multiple K values. These SKUs are likely to have special physical attributes (such as highly irregular shapes or abnormal load capacities) or business attributes, making them difficult to efficiently co-pack with most other SKUs. Continuing to include them in the co-optimization will continuously pollute the results.

[0244] Isolation strategy: Based on the prevalence of "habitual offenders" (the proportion of interference levels), different levels of isolation measures will be adopted.

[0245] Widespread interference (proportion > threshold): This indicates that multiple SKUs are causing widespread problems. In this case, isolating the most severely affected SKUs in batches (first batch, e.g., 2) can quickly clean up the environment.

[0246] Local interference (proportion ≤ threshold): There may only be a few "bad apples". In this case, isolating the most destructive SKUs one by one (at a time) is a more refined approach.

[0247] Solution: Isolated SKUs are no longer involved in complex collaborative consolidation optimization. Instead, they are handled by separate storage or simple stacking (e.g., allocating them to individual cases or using pre-packaged fixed containers). This essentially breaks down the complex problem: letting collaborative optimization handle the majority of "collaborative" SKUs, while using simple and reliable methods to handle the minority of "difficult-to-collaborate" SKUs, thereby achieving optimal overall efficiency.

[0248] For example, analysis revealed that SKU_P and SKU_Q repeatedly co-occurred in the undefined combinations of K=3, 4, and 5. SKU_P appeared in all three K values ​​(interference level ratio 100%), while SKU_Q appeared twice (67%). Assuming a preset interference level ratio threshold of 60%, both SKU_P and SKU_Q were identified as "high-interference SKUs." Due to the existence of multiple high-interference SKUs, the system decided to select the first number (e.g., two) of the most severely interfering SKUs for isolation each time. Therefore, SKU_P and SKU_Q were removed from the collaborative optimization pool and marked as "storage and processing separately."

[0249] Step 3: Iterative optimization:

[0250] Process closed loop: The new SKU set that has isolated the "problem SKU" is then re-executed with the collaborative consolidation optimization process that started from S41. This process is repeated until the vacancy rate for all K values ​​is reduced to below the acceptable threshold.

[0251] This is a gradual purification and iterative approximation process. Each isolation may improve the synergy among the remaining SKUs, thereby reducing the vacancy rate. This ensures that the system can robustly find a feasible and efficient solution under the current constraints, although it may involve special handling for specific SKUs.

[0252] Summary and Value of Implementation Examples:

[0253] This embodiment discloses a data-driven, refined strategy for optimizing resource allocation:

[0254] From “processing solution” to “processing SKU”: This involves shifting the optimization focus from adjusting algorithm parameters (K value) to identifying and processing the data object (problem SKU) itself, which is a deeper level of optimization.

[0255] The engineering wisdom of "divide and conquer": Recognizing that not all SKUs are suitable for mixed packaging, we proactively separate "incompatible" members and process them separately in the most suitable way, thereby ensuring the purity and high efficiency of the main body's collaborative optimization.

[0256] Achieving system robustness: This mechanism enables the optimization system to have "detoxification" and "adaptive" capabilities, automatically identifying and avoiding the root causes of performance bottlenecks, ensuring that a stable and usable operating plan can be output when faced with various abnormal data, greatly improving the industrial robustness of the system.

[0257] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments of apparatus, devices, and non-volatile computer storage media are basically similar to the method embodiments, so the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0258] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0259] The above description is merely one or more embodiments of this specification and is not intended to limit this specification. Various modifications and variations can be made to the one or more embodiments of this specification by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of one or more embodiments of this specification should be included within the scope of the claims of this specification.

Claims

1. A collaborative consolidation global optimization system, characterized in that, Specifically, it includes: The modeling module is responsible for enumerating all SKU combinations with sizes ranging from 1 to the upper limit k of the preset complexity level according to the preset business constraints. For each SKU combination, the single box capacity is determined according to the shortest board principle, and all box allocation schemes that satisfy the requirement that each SKU is allocated at least one piece are generated. The scoring module is responsible for calculating the maximum number of boxes that can be produced by each allocation scheme within a box, and applying batch consistency constraints and maximum number of boxes constraints to filter and obtain feasible schemes. The feasible scheme is scored by the number of boxes that can be produced × the capacity of a single box. A Top-K streaming pruning strategy is adopted, using a min-heap to dynamically maintain the top K candidate schemes with the highest scores to obtain a set of candidate schemes and control the size of the candidate schemes. The batch consistency constraint is whether the calculated maximum number of boxes that can be produced meets the system's preset batch consistency threshold. The maximum number of boxes constraint is whether the number of boxes that the scheme can produce exceeds the maximum number of boxes limit; The solution module is responsible for inputting the selected candidate solution set into the optimization solver, constructing the optimization solver using a greedy strategy and setting a maximum solution time limit for each level, executing an iterative strategy that prioritizes high complexity, recording the remaining SKU requirements corresponding to each complexity level, and selecting the optimization solution with the lowest overall remaining SKU requirements as the final output solution. The shortest-board principle is to extract the number of pieces per carton for all SKUs in the combination and take the minimum value as the uniform single-carton capacity of the combination. The Top-K streaming pruning strategy controls the size of the candidate solutions input to the optimization model. It adopts a priority queue-based streaming pruning mechanism, initializes a min-heap with a capacity of K, and dynamically maintains the K solutions with the highest scores among the currently processed solutions. When processing each feasible solution, its score is compared with the top of the min-heap. Based on the comparison result, it is decided whether to insert it into the heap and may eliminate the top solution. The upper limit k of the preset complexity level is the number of SKUs in a single box.

2. The collaborative consolidation global optimization system as described in claim 1, characterized in that, The SKUs to be processed need to be grouped according to preset constraint rules before the SKU combination enumeration process is performed.

3. The collaborative consolidation global optimization system as described in claim 1, characterized in that, Generate all in-box allocation schemes that satisfy the requirement of allocating at least one item per SKU, specifically including: Based on the enumeration range of all SKU combinations from a single SKU to a preset complexity upper limit k, we enumerate to obtain SKU combinations; For each enumerated SKU combination, its theoretical maximum capacity per box is determined according to the shortest board principle; Given a fixed single-box capacity, a hard constraint is added that each SKU must be allocated at least once. An integer partitioning algorithm is used to generate a specific allocation scheme for the SKUs within the box, resulting in the box allocation scheme.

4. The collaborative consolidation global optimization system as described in claim 2, characterized in that, The process of obtaining the candidate solution set is as follows: For each SKU involved in each generated in-box allocation scheme, calculate the ratio of its initial demand to the number of boxes allocated to that SKU in the scheme, round down, and take the minimum value of the calculation results for all related SKUs as the maximum number of boxes that the allocation scheme can produce without other constraints. The calculated maximum number of boxes that can be produced is compared with the system's preset batch consistency threshold. Only when the number of boxes that can be produced is greater than or equal to the batch consistency threshold is the solution considered to meet the batch stability requirements of the production line operation, thus obtaining a solution that meets the batch consistency constraint. For solutions that meet the batch consistency constraint, they are further compared with the system's optional parameter—the upper limit of the maximum number of boxes. The number of boxes that can be produced in the solution must not exceed this upper limit, thus obtaining a feasible solution. Each feasible solution is assigned a score, which is defined as the product of the number of boxes that can be produced and the capacity of a single box. A priority queue-based streaming pruning mechanism is adopted. A min-heap with a capacity of K is initialized to dynamically maintain the K solutions with the highest scores among the currently processed solutions. When processing each feasible solution, its score is compared with the top of the min-heap. Based on the comparison result, it is decided whether to insert it into the heap and may eliminate the top solution. During the streaming process, when the min-heap is full, for the SKU combination to which the newly arrived solution belongs, the theoretically highest possible score upper bound of the combination can be calculated in advance. If this theoretical upper bound is lower than the current top score of the heap, it can be determined that all unprocessed allocation solutions in the combination have no chance of entering the Top-K set. After the process is completed, the K feasible solutions stored in the min-heap constitute the final candidate solution set submitted to the optimization solver within the group.

5. The collaborative consolidation global optimization system as described in claim 4, characterized in that, The batch consistency threshold ranges from 3 to 10, and is determined based on the batch stability requirements of the production line operation.

6. The collaborative consolidation global optimization system as described in claim 1, characterized in that, The method for determining the final output scheme is as follows: A greedy heuristic algorithm is used to process the filtered candidate solution set to quickly construct a feasible initial solution. Based on the current complexity level, a differentiated maximum running time limit is set for the optimization solver. The system executes the complete optimization process independently for different complexity limits in descending order. At each complexity level, the candidate set is regenerated based on the combinatorial enumeration constraints of that level and optimized independently to obtain the optimal binning scheme under the constraints of that complexity level. After the optimization solution for each complexity level is completed, the system calculates the initial demand that has not yet been met for each SKU in the group after adopting its output solution, i.e. the remaining manual processing amount. The remaining demand of all SKUs is summarized to form the overall remaining manual processing amount index under that complexity level. After completing independent optimization and calculating the remaining manual processing volume for all preset complexity levels, the system compares the overall remaining manual processing volume index corresponding to each complexity level, selects the optimization scheme corresponding to the complexity level that minimizes the index value, and uses it as the final output scheme.

7. A collaborative consolidation global optimization method, applied to a collaborative consolidation global optimization system according to any one of claims 1-6, characterized in that, Specifically, it includes: Obtain SKU data and, based on the deviation in the quantity of goods corresponding to each SKU within a single box, determine the recommended value for the complexity level. Based on the recommended values, a global optimization process for collaborative consolidation is performed. According to the single-box capacity of different SKU combinations and the overall remaining manual processing volume under the existing complexity level, the optimization scheme of the recommended values ​​for the complexity level is determined. The recommended value is optimized using the optimization processing scheme to obtain the optimization result. Based on the single-box capacity of the SKU combination under the optimization result, the optimization result of the collaborative consolidation SKU is determined. The method for determining the optimization scheme of the recommended value of the complexity level is as follows: Given the existing complexity levels and the single-box capacity of different SKU combinations, determine the proportion of remaining space in a single box for each SKU combination. Based on the overall remaining manual processing volume index, determine the initial demand that has not yet been met for each SKU, and determine the impact value of the SKU based on the ratio of the initial demand that has not yet been met to the initial demand. Based on the impact value of each SKU and the proportion of remaining space in a single box of SKU combinations, an optimized processing scheme for the recommended value of the complexity level is determined. The method for determining the optimization results of the collaborative consolidation SKUs is as follows: Based on the single-box capacity of the SKU combination under the optimization results, the single-box capacity of the SKU combination under different complexity levels is determined. Based on the single-box capacity of SKU combinations at different complexity levels, determine the SKU combinations whose remaining space ratio in a single box is greater than a preset space ratio threshold at different complexity levels, and treat them as vacant combinations. The optimization result of the collaborative binning SKU is determined by the overlap of SKUs in the empty combinations at different complexity levels.

8. The collaborative consolidation global optimization method as described in claim 7, characterized in that, The method for determining the recommended value for the complexity level is as follows: Determine the number of SKUs based on the SKU data; Based on the deviation in the quantity of goods corresponding to the SKU in a single box, determine the quantity of SKUs in different quantity ranges; The recommended value for the complexity level is determined based on the number of SKUs and the number of SKUs in different capacity ranges.

Images