Intelligent boxing method and system

Abstract

The invention provides an intelligent boxing method and system. The intelligent boxing method comprises the steps of obtaining boxing task information and generating a boxing scheme. The boxing task information comprises the size, the weight, the stacking limitation, the number and the priority of each cargo, and the size, the bearing capacity and the occupation state of each loading unit. And under the condition that the size, bearing and stacking constraint conditions are met, a plurality of candidate boxing combinations are generated, the loading utilization rate, the residual space distribution and the shape matching degree are evaluated based on a preset optimization target, and an optimal scheme is selected. The system comprises an input terminal, a task information receiving module, a data storage module, a task and scheme storage module, an operation processing module and a final boxing scheme generation module. According to the method, multi-dimensional constraint matching and weight balance are considered, and space waste and transportation risks are avoided; the maximum greedy algorithm and the adaptive large neighborhood search optimization are combined to realize the combination of automation and experience. The method can be expanded to batch task scheduling and grouping sequencing, so that the calculation complexity is reduced, and the overall boxing and transportation efficiency is improved.

Description

Technical Field

[0001] This application relates to the field of intelligent logistics, and in particular to an intelligent packing method and system. Background Technology

[0002] With the rapid development of logistics, e-commerce, and warehousing and transportation within industrial parks, cargo loading efficiency directly impacts transportation costs and overall operational efficiency. Traditional packing methods often rely on manual experience or simple rule matching, typically considering only cargo size and spatial constraints of loading units while neglecting multi-dimensional factors such as weight distribution, stacking limitations, and priorities. This can easily lead to low utilization rates of loading units, severe space fragmentation, and even overloading or cargo damage. Furthermore, most existing automated packing algorithms remain at the single-objective optimization stage, such as pursuing maximum volume utilization without comprehensive balance calculations for multiple objectives. Some methods lack support for dynamic adjustment and manual modification after generating packing plans, resulting in insufficient flexibility during actual execution. Especially in scenarios with a large number of concurrent packing tasks within industrial parks, existing technologies struggle to balance overall transportation efficiency and individual vehicle loading quality, thus hindering the level of intelligence in warehousing and logistics scheduling. Summary of the Invention

[0003] The purpose of this application is to solve the problems of low loading utilization, severe fragmentation of remaining space, uneven weight distribution, and lack of flexible adjustment in the existing packing methods.

[0004] According to one aspect of this application, a smart packing method is provided, suitable for allocating multiple goods to multiple available loading units, including:

[0005] S10. Obtain packing task information, which includes the size, weight, stacking limit, quantity and priority of each item, as well as the size, load-bearing capacity and occupied status of each loading unit.

[0006] S20. Under the conditions of satisfying the size constraints, load-bearing constraints and stacking limitations of the loading unit, generate multiple candidate packing combination schemes.

[0007] S30. Evaluate each of the candidate combination schemes based on the preset optimization target, and calculate the loading utilization rate and the remaining space distribution;

[0008] S40. Based on the evaluation results, select the optimal packing combination from the candidate packing combination schemes as the final packing scheme.

[0009] Preferably, the process further includes the following step before step S20:

[0010] S11. Based on the size, weight, stacking restrictions, and priority of the goods, divide the goods into multiple groups with similar attributes;

[0011] S12. Arrange the goods in each group according to the preset sorting rules to reduce the search space and improve the matching degree.

[0012] Preferably, the preset optimization objective is:

[0013] The loading scheme with the highest volume utilization, lowest fragmentation of remaining space, and least differentiation of loading units is selected. The optimization target value is obtained by weighted summation of the three factors.

[0014] Preferably, S20 includes:

[0015] S21. According to the cargo sorting, the cargo is sequentially allocated to loading units that meet the size, load-bearing and stacking restrictions;

[0016] S22. When direct placement is not possible, generate new candidate combinations by rotating the goods or adjusting the position of existing goods.

[0017] Preferably, the initial binning combination is generated in S20 using a maximum greedy algorithm, specifically including:

[0018] Calculate the objective function gain ΔF for each triplet in all feasible sets of "goods, placement location, placement posture";

[0019] In each iteration, the triplet with the largest gain ΔF of the objective function is selected for placement, and triplets that have geometric or load-bearing conflicts with this placement are removed from the feasible set.

[0020] Repeat the iteration until there are no feasible triples with ΔF>0 or all goods have been placed;

[0021] The objective function F is defined by the following formula:

[0022] F=α×U-β×Frag-γ×Imb

[0023] In the formula, U is the volume utilization rate, Frag is the fragmentation measure of the remaining space, Imb is the measure of the uneven weight distribution among loading units, α, β, and γ are non-negative weighting coefficients, and ΔF is the difference of F before and after the placement operation.

[0024] Preferably, the calculation of the remaining spatial distribution in S30 includes:

[0025] Record the occupied and unoccupied areas in the loading unit according to three-dimensional coordinates;

[0026] Perform connectivity analysis on unoccupied regions and calculate the volume and shape of each unoccupied region;

[0027] The availability of remaining space is evaluated based on shape matching.

[0028] Preferably, the process of selecting the optimal combination in S40 employs an adaptive large neighborhood search algorithm, including:

[0029] Destruction operator: Removes some cargo loading positions from the current combination;

[0030] Repair operator: Reselects a loading location for removed cargo;

[0031] Repeat the destruction and repair process until the number of iterations or the convergence condition is met;

[0032] The destruction operators include: random removal, removal of the smallest item by volume, and removal of the item with the greatest fragmentation of remaining space;

[0033] The repair operator includes: randomly selecting a location for loading and selecting a location with the highest remaining space matching degree for loading.

[0034] Preferably, the contents of the packing solution output include:

[0035] The three-dimensional placement and orientation of the goods within each loading unit;

[0036] Volume utilization rate, weight utilization rate, and remaining space distribution of each loading unit;

[0037] List of goods that could not be loaded and the reasons.

[0038] Preferably, in step S40, after selecting the optimal combination, the packing scheme is visualized and rendered to generate a three-dimensional packing diagram for manual review or direct control of automated loading equipment. The three-dimensional packing diagram allows users to make manual adjustments via an input terminal, and the results of the manual adjustments are rewritten into the packing scheme and stored.

[0039] The present invention also provides an intelligent packing system suitable for distributing multiple goods to multiple loading units, comprising:

[0040] An input terminal is used to receive packing task information, which includes the size, weight, stacking limits and priority of each item, as well as the size, load-bearing capacity and occupied status of each loading unit.

[0041] A data storage module, connected to the input terminal, is used to store the packing task information and packing plan;

[0042] The processing module, connected to the data storage module, is used to execute the above method and generate a packing scheme based on the packing task information.

[0043] This application offers the following advantages: By considering cargo size, weight, stacking limitations, and priorities, it achieves multi-dimensional constraint matching between loading unit space and load-bearing capacity, avoiding space waste caused by simply relying on size matching. The inclusion of remaining space distribution and shape matching degree analysis in candidate scheme evaluation effectively reduces the generation of irregular gaps, thereby improving space availability. Weight balance metrics are incorporated into the optimization objective to avoid loading unit tilting or transportation safety risks caused by uneven weight distribution. It supports generating initial combinations using a maximum greedy algorithm and iteratively optimizing using an adaptive large neighborhood search, while allowing users to manually adjust the scheme through a visual interface, combining automation with human experience. It can be extended to scheduling scenarios with multiple vehicles and batch tasks in a park, reducing computational complexity through grouping and sorting, and improving the overall efficiency of container loading and transportation scheduling. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0045] Figure 1 This is a logic block diagram of an intelligent packing method according to an embodiment of this application. Detailed Implementation

[0046] To facilitate understanding of this application, a more complete description will be provided below with reference to the accompanying drawings. Preferred embodiments of this application are shown in the drawings. However, this application can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the disclosure of this application.

[0047] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein in the specification of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0048] Please refer to Figure 1 One embodiment of this application provides an intelligent packing method suitable for allocating multiple goods to multiple available loading units, including:

[0049] S10. Obtain packing task information, including the dimensions, weight, stacking limits, quantity, and priority of each cargo, as well as the dimensions, load-bearing capacity, and occupied status of each loading unit. In this step, it is important to note the following: Standardize and validate the task information from the upstream system / manual input, including unit unification (length / weight / volume), dimensional tolerance handling, and field integrity checks; pre-calculate the set of possible placement orientations for each cargo (a subset of all permutations of length × width × height that satisfy stacking and orientation restrictions); abstract stacking restrictions into a rule matrix (e.g., "stackable / non-stackable," "maximum number of layers," "threshold for full bottom support ratio," etc.); establish a three-dimensional coordinate system for each loading unit and introduce safety boundary gaps (to allow for operational margins for loading machinery); calculate the "candidate loading unit set" for each cargo based on load-bearing capacity and occupied status; establish stable comparison rules for task priorities (e.g., lexicographical order of priority → volume → weight) to ensure the repeatability of subsequent solutions.

[0050] S20. Under the conditions of satisfying the size constraints, load-bearing constraints, and stacking limitations of the loading unit, generate multiple candidate packing combination schemes. In this step, it should be noted that: a "set of available space" is maintained for each loading unit, initially the space being the inner cavity of the loading unit (excluding the occupied volume and safety gap); when a cargo is to be placed, a placement position (cargo, space, posture triplet) that is geometrically compatible with a certain available space and meets the load-bearing / stacking / support ratio thresholds is selected from its set of possible placement positions; after successful placement, a cutting strategy (such as "split-type / guillotine-type") is used to deduct the occupied space from the parent space and generate several non-overlapping subspaces, and adjacent and mergeable subspaces are merged to reduce the number of fragments; under the premise of satisfying the constraints, several different packing paths are generated through multiple starting points and perturbation mechanisms to form multiple candidate combinations (e.g., priority-driven sequences, volume-decreasing sequences, sequences interspersed with random perturbations, etc.), and each path fully records all placement decisions and intermediate space states for subsequent evaluation.

[0051] S30. Evaluate each candidate combination scheme based on the preset optimization target, and calculate the loading utilization rate and the distribution of remaining space. In this step, it should be noted that: the loading utilization rate is calculated as "placed volume / effective volume of loading unit"; the distribution of remaining space is extracted from the set of available space. First, a three-dimensional connectivity analysis (six-neighbor or twenty-six-neighbor) is performed on the unoccupied area to obtain the volume, surface area, and circumscribed dimensions of several connected cavities. Then, a "fragmentation metric" is constructed using shape factors (such as length, width, and height dispersion, surface area / volume ratio) and the matching degree with common cargo dimensions. In the scenario of multiple loading units, each loading unit can be evaluated one by one and summarized into a total scheme score. To ensure the stability of the comparison, in addition to the main index, disambiguation rules for parallel candidates are set (such as prioritizing those with lower fragmentation, then those with more concentrated remaining space, and if they are still the same, choosing those with lower average height, etc.). The evaluation results and the key intermediate quantities that generated the results are saved together for easy traceability.

[0052] S40. Based on the evaluation results, select the optimal packing combination from the candidate packing combinations as the final packing scheme. In this step, it should be noted that: candidate combinations are sorted and the optimal one is selected according to a weighted or sequential priority strategy based on the preset optimization objective; a comprehensive consistency check is performed before output, including: all geometries do not overlap, safety clearances are met, load-bearing and stacking rules are not violated, and the placement order is executable (i.e., each step has a feasible loading path and gripping posture); the final output includes the three-dimensional placement position (x, y, z reference points), posture (discrete rotation around the coordinate axes), loading sequence number, and a summary of the utilization rate and remaining space distribution of each loading unit for each item within the corresponding loading unit; simultaneously, metadata (such as timestamps, algorithm parameters, candidate scheme numbers, and scores) is generated for execution and auditing, for subsequent reproduction and comparison.

[0053] The technical solution implemented in this embodiment can systematically generate multiple candidate packing paths while strictly meeting multiple constraints such as size, load-bearing capacity, and stacking. It also allows for comparable optimization through quantified utilization and fragmentation indicators, significantly improving the volume utilization of loading units. Spatial cutting and connectivity analysis suppress the generation of invalid, fragmented gaps, making the remaining space more regular and reusable, reducing the difficulty of subsequent additional loading. The use of safety gaps and consistency verification mechanisms ensures the feasibility and stability of the solution on actual loading equipment, reducing the risk of rework due to collisions or insufficient support. Reproducible sorting and disambiguation rules make the solution process robust and the results traceable, facilitating rapid comparison and deployment in multi-batch task environments within the park. Without relying on manual experience for fine-tuning, it provides structured output for subsequent integration with visual auditing and automated loading equipment, reducing overall manual planning costs and improving park operational efficiency.

[0054] Furthermore, prior to S20, it also includes:

[0055] S11. Based on the size, weight, stacking restrictions, and priority of the goods, divide them into multiple groups with similar attributes. In this step, it should be noted that to ensure the stability of subsequent solutions and controllable computational load, the original task data is first standardized and cleaned, including unit unification (length / width / height, weight), field integrity verification, outlier removal, and dimensional tolerance handling. Subsequently, each item is mapped to an attribute vector v = (L, W, H, m, stack_class, prio), where L / W / H are normalized dimensions (sorted according to the rule of length ≥ width ≥ height to eliminate the influence of orientation differences), m is the weight, stack_class is the stacking rule category (such as "stackable / non-stackable," maximum number of layers, bottom support required ratio, etc., abstracted as rule codes), and prio is the priority. Based on this attribute vector, the following grouping process is adopted:

[0056] Size and weight classification: Discrete intervals are set according to the nominal internal dimensions and load-bearing capacity of the loading unit (e.g., the bins are divided according to the volume interval V and the weight interval m). Goods first fall into the corresponding "size-weight" combined interval.

[0057] Stacking rule constraints: Within each "size-weight" bucket, further subdivide by stack_class to ensure that goods in the same group have the same or compatible stacking restrictions (such as no stacking, maximum number of layers = K, bottom full support ≥ θ, etc.) to avoid frequent rule conflicts during subsequent placement stages;

[0058] Priority aggregation: Combined with PIO, goods with the same or adjacent priorities are clustered into the same subgroup to facilitate coordinated packing according to task urgency and loading wave.

[0059] Distance metrics and merging / splitting: For small groups with a small sample size and a feature distance d (e.g., cosine distance of size vector or Mahalanobis distance) to neighboring groups below a threshold τ, perform neighbor merging; for large groups with excessively large intra-group variance (e.g., coefficients of variation of length, width or height exceeding a set threshold), use hierarchical clustering or K-means (K is determined by the elbow method) for further splitting until the intra-group dispersion meets the threshold.

[0060] Feasible loading unit filtering: Combine the loading unit list (internal cavity size, load-bearing capacity, and occupied status) to generate a "set of reachable loading units" for each group. If the entire group is not feasible for a certain loading unit, the group is removed from the candidates for that unit to reduce subsequent invalid searches.

[0061] S12. Within each group, arrange the goods according to a preset sorting rule to reduce the search space and improve the matching accuracy. In this step, it should be noted that:

[0062] To improve the success rate and space utilization during the placement phase, deterministic serialization rules are applied to the goods within each group to form a placement sequence. Hierarchical lexicographical order is used by default, supplemented by "placement difficulty" correction.

[0063] Primary order rule: Sort by priority (prio) from highest to lowest; if priorities are the same, sort by volume (V = L × W × H) from largest to smallest; if volumes are the same, sort by weight (m) from largest to smallest. This primary order ensures that urgent tasks are prioritized and large / heavy "placeholder" items are placed in advance.

[0064] Difficulty in releasing correction item: Calculate a "difficulty in releasing" score h for each item, used to determine the order of release when the order of release is equal. h can be weighted by the following indicators:

[0065] Shape mismatch: The deviation between the length, width, and height of the cargo and the aspect ratio of the typical cavity (such as the inner cavity or common remaining space) of the loading unit;

[0066] Rotational degree of freedom penalty: The fewer the number of feasible placement orientations (e.g., only 1-2 effective orientations), the larger h is;

[0067] Stacking vulnerability: h is increased for goods that cannot be stacked or require a high proportion of bottom support;

[0068] Center of gravity and load-bearing sensitivity: For those with heavy weight and sensitive to the bottom support ratio, h increases.

[0069] Placing items with higher h values ​​first can reduce the probability of subsequent space fragmentation and rule conflicts.

[0070] Business rule revisions: Optional introduction of on-site rules, such as "priority for bottom-load components" (goods with strong load-bearing capacity and good pressure resistance are prioritized at the beginning of the sequence), "fragile components at the end" (fragile and non-stackable goods are arranged at the end of the sequence), and "grouping goods of the same order / customer together" (to facilitate outbound shipments in the same wave).

[0071] Stability and reproducibility: When all metrics are identical, a fixed, stable hash key (such as cargo ID) is used as the final disambiguation criterion to ensure that the same sequence can be reproduced across different batches.

[0072] Output: Each group outputs a determined sequence of items to be placed (including tie-break criteria and h score), which is used for item-by-item placement and space updates during the S20 candidate generation stage.

[0073] The technical solution implemented in this embodiment can significantly compress the candidate search space through two-level preprocessing of "grouping and sorting," enabling subsequent placement and evaluation to be carried out within a smaller feasible domain, thereby improving computational efficiency and convergence stability. By imposing stacking constraints, load-bearing capacity, and priority at the grouping stage, rule conflicts and backtracking frequency during placement are reduced, increasing the success rate of one-time placement. The "difficulty of placement" is used to correct the main order, prioritizing goods with low rotational freedom, high shape mismatch, fragile stacking, or load-bearing sensitivity, reducing space fragmentation and improving overall volume utilization and solution feasibility. A stable sorting and disambiguation mechanism ensures the reproducibility of results across different batches, facilitating the review, backtracking, and parameter tuning of park operations. Combined with feasible loading unit filtering, each group generates candidates only on reachable units, reducing invalid attempts and further shortening the overall solution time and improving solution quality.

[0074] In one optional embodiment, the preset optimization objective is to select the loading scheme with the highest volume utilization rate, the lowest degree of fragmentation of remaining space, and the least differentiation among loading units. The optimization objective value is obtained by weighted summation of the three factors. It should be noted that the volume utilization rate is calculated as follows:

[0075]

[0076] Where, ∑V cargo V represents the total volume of goods already placed. unit U represents the effective volume of the loading unit. By calculating U in real-time during candidate combination generation and scheme selection, the degree of space utilization of each scheme can be intuitively reflected. During evaluation, a threshold setting can be used to ensure that schemes with utilization rates below a certain level are eliminated. Furthermore, volume utilization is not a single indicator and must be comprehensively evaluated in conjunction with subsequent spatial distribution to avoid the problem of "high utilization rate but no further loading."

[0077] Fragmentation of unused space refers to the distribution of unused space across multiple small, irregular areas, making it difficult to effectively utilize for subsequent cargo. Calculation methods may include:

[0078] Perform three-dimensional connectivity analysis on the unoccupied space to count the number of connected components and volume distribution of the remaining space;

[0079] The matching degree between the shape of each connected component and the size of typical cargo is calculated. If the size of most connected components is much smaller than the average size of cargo, the degree of fragmentation is considered to be high.

[0080] Define a fragmentation metric, Frag, for example:

[0081] Among them, V i Let α be the volume of the i-th remaining space. iThis is its shape irregularity factor. By controlling the size of the Frag, situations where "there seems to be space but it is difficult to load" can be avoided, thereby improving the sustainability of subsequent packing.

[0082] Different loading units have varying load-bearing capacities. If the weight distribution is severely uneven, some units may be overloaded while others are underutilized. Therefore, a weight distribution balance metric, Imb, is introduced into the optimization objective, which can be reflected using, for example, the standard deviation or range. Among them, W k The loaded weight of the k-th loading unit. Where Imb represents the average load weight, and m represents the total number of loading units. By controlling the size of Imb, a more reasonable distribution can be achieved within the overall load-bearing range, thereby reducing safety risks caused by off-center loading and center of gravity shift during transportation.

[0083] The technical solution implemented in this embodiment can ensure full utilization of loading space through volume utilization index, reduce empty consumption during transportation, and improve overall transportation efficiency; by suppressing fragmentation, the remaining space maintains good connectivity and regularity, thereby improving the placement of subsequent goods and avoiding the dead end of "high utilization but not scalable" packing; by balancing weight distribution constraints, the loading scheme is made more in line with transportation safety and actual engineering needs, avoiding hidden dangers caused by overloading or uneven loading of some units; the three optimization objectives can be used individually or combined into a comprehensive objective function through weighted combination, so that the packing scheme achieves a balance between efficiency, sustainability and safety, thereby improving the overall practicality and feasibility of the scheme.

[0084] Furthermore, S20 includes:

[0085] S21. Based on the cargo sorting order, sequentially allocate the cargo to loading units that meet the size, load-bearing capacity, and stacking limitations. In this step, it should be noted that the preprocessed and sorted sequence of cargo to be placed in S11 and S12 serves as input. Each loading unit maintains a "set of available space" (which can be represented as several axis-aligned cuboid subspaces) and state variables such as the currently loaded weight and volume; each cargo has its set of possible placement orientations, mass, stacking rules, and priority markers. For the current cargo to be placed, enumerate possible axis-aligned placement orientations (commonly six axial orientations) according to its orientation set, and within each available subspace, probe placement positions using several anchor points (e.g., the lower left front corner, lower right front corner, layer height reference point, etc.). Each probe first performs a size compatibility check (whether the cargo bounding box can be completely accommodated within the subspace in this orientation, considering the safety margin δ), and then performs geometric conflict detection (based on bounding box fast determination or voxelization detection). If the goods need to be stacked or placed on top of other goods, calculate the ratio of the actual load-bearing area at the bottom to the bottom surface (bottom support ratio) and compare it with the minimum support threshold of the goods; at the same time, check the remaining load-bearing capacity of the load-bearing goods below to ensure that local static constraints (such as single-point load exceeding limits, risk of overturning due to center of gravity deviation) are not violated. If necessary, calculate the overall center of gravity change of the unit after placement for subsequent weight balance assessment. Update the loaded weight of the loading unit before and after placement; if it exceeds the rated load of the loading unit or causes overloading of a certain layer, the placement position is judged as infeasible. When placement is successful, cut the volume of the goods from the occupied subspace and generate a new set of subspaces according to the agreed spatial division strategy (such as splitting or rectangular division), and then perform a merging operation on the adjacent mergeable subspaces to reduce fragmentation; at the same time, record the placement decision (goods ID, loading unit ID, attitude, coordinates, timestamp) to the intermediate solution sequence. If the current cargo cannot be placed feasiblely after trying all available loading units and all attitudes / anchor points, a backtracking or backup strategy is triggered (e.g., recorded as unloaded, added to a delay queue, or the position adjustment mechanism in S22 is triggered). To control computational load, a maximum number of attempts and a time threshold are set for the placement of a single cargo (e.g., no more than N anchor point attempts or no more than T seconds per cargo). If the timeout occurs, the subsequent processing logic is invoked. All attempts (successful and unsuccessful) are recorded with their corresponding intermediate spatial states and scores, facilitating backtracking decisions during subsequent local optimization or manual review.

[0086] S22. When direct placement is not possible, new candidate combinations are generated by rotating the cargo or adjusting the position of existing cargo. It should be noted that S22 is triggered when a cargo in S21 is deemed infeasible to place at all anchor points and allowed orientations in all loading units (or placement would significantly lower the overall score). S22 includes two basic methods: rotation attempts (changing only the current cargo orientation) and local rearrangement (adjusting placed cargo to make room for available space). These can be used individually or in combination, and both are subject to controllable search depth and time limits to prevent explosive growth of combinations.

[0087] Prioritize the use of discrete rotation sets (up to six orthogonal orientations) aligned with the axis, and the rotation priority can be defined based on the shape of the cargo and the length, width and height relationship of the remaining space (e.g., prioritize rotations that align the long side of the cargo with the long side of the space); perform the same size compatibility, geometric conflict and stacking / load-bearing checks as in S21 after each rotation; for non-orthogonal placement, limited angles (such as discrete angles other than 90°) may be tried where permitted, but it must be noted in the specification that such practice is limited by mechanical gripping capabilities and loading constraints.

[0088] If simple rotation is not feasible, then fine-tuning or rearranging of existing goods is performed within a defined neighborhood. The neighborhood can be selected based on spatial adjacency (the k nearest placed goods to the current goods), a set of goods on the same level, or all goods within the same subspace. Local rearrangements include, but are not limited to:

[0089] For goods within a neighborhood, attempt to translate them along the X / Y / Z directions within their subspace by a certain number of steps (the step size is determined by the safety margin) to explore gaps; swap the position of the goods to be placed with a goods already placed in the neighborhood (the feasibility of placing them in each other's positions must be satisfied simultaneously); if there are multiple small goods in the same layer of space in the neighborhood, create more continuous cavities by merging and rearranging the order of the goods in that layer (e.g., grouping the small items to one side to free up sufficient width); under neighborhood constraints, revoke several most recently placed decisions (e.g., revoke the most recently placed items 1 to M), and replay a sequence in a maximum greedy or heuristic order to obtain different spatial partitions.

[0090] Each local rearrangement operation generates one or more new local spatial states and corresponding candidate combinations. The system scores these candidates using the same evaluation function as S30 (volume utilization, fragmentation, weight balancing, etc.). If the score after rearrangement is significantly better than the non-rearrangement scheme or enables the current goods to be successfully placed, the local rearrangement is accepted and written into the current solution; otherwise, the operation is rolled back and the next rearrangement scheme is tried. To avoid loops and resource waste, the system should maintain a set of attempted rearrangement codes and limit the number of rearrangements in the same neighborhood.

[0091] The search order for local rearrangements can be determined by a heuristic cost function, such as prioritizing adjustments that maximize the release of the largest contiguous volume or the maximum increase in the gain of F (objective function). Simultaneously, global and local time limits and rearrangement depth limits are set (e.g., a maximum of M = 3 undoes of the most recently placed elements) to ensure algorithm responsiveness. Stacking support, weight-bearing capacity, and grasping feasibility checks are performed on all rearrangement operations to prevent the generation of bin packing sequences that are mechanically unexecutable or pose safety risks.

[0092] Each partial rearrangement operation saves a change log (including undone placement decisions, repositioned goods, and new space status) before and after application, so that errors can be quickly rolled back or used for manual review.

[0093] The technical solution implemented in this embodiment can improve the success rate of single placement by finely enumerating the posture and anchor points before placement and combining support / load-bearing verification to reduce direct placement failures caused by rule conflicts; reduce space fragmentation and backtracking times by immediately executing subspace merging and fragmentation reduction strategies after placement to make the remaining space more valuable, and reduce subsequent backtracking by combining a "difficulty-to-place" priority placement strategy; provide controlled extended search when direct placement is not possible by using limited-depth rotation attempts and local rearrangement within the neighborhood, adopting heuristic priority and cost constraints to release continuous space in a targeted manner, thereby significantly improving the packing success rate under the premise of controllable computational overhead; ensure the feasibility and safety of the solution by including grabbing and loading feasibility, stacking support, load-bearing and safety gap verification in all placement and rearrangement steps, and ensuring that the output packing order and position are feasible on automated loading equipment; and support traceable and auditable operations by recording intermediate space status, placement decisions and rearrangement logs to achieve solution reproduction, backtracking and manual intervention, facilitating mixed automated and manual collaborative operations in actual park operations.

[0094] Furthermore, the initial binning combination is generated in S20 using a maximum greedy algorithm, specifically including:

[0095] Calculate the objective function gain ΔF for each triplet in the set of all feasible "cargo, placement location, placement posture" triplets.

[0096] In each iteration, the triplet with the largest objective function gain ΔF is selected for placement, and triplets that have geometric or load-bearing conflicts with this placement are removed from the feasible set.

[0097] Repeat the iteration until there are no feasible triples with ΔF>0 or all goods have been placed.

[0098] The objective function F is defined by the following formula:

[0099] F=α×U-β×Frag-γ×Imb

[0100] In the formula, U represents volume utilization, Frag is a measure of remaining space fragmentation, Imb is a measure of uneven weight distribution among loading units, and α, β, and γ are non-negative weighting coefficients. ΔF is the difference between F before and after the placement operation.

[0101] In this embodiment, it should be noted that the maximum greedy algorithm is used to generate the initial packing combination (the initial constructor of S20). The algorithm uses the "set of feasible triples" as the search space. The feasible triples refer to candidate placement decisions in the form of (cargo, placement position, placement orientation), where "placement position" represents the coordinates of an anchor point in a certain available subspace of a loading unit, and "placement orientation" represents a discrete orientation or allowed rotation of the cargo.

[0102] The system first enumerates or heuristically generates a set of initial feasible triples. Each triple must pass a size compatibility check (considering the safety margin δ), a preliminary support / stack check, and a load-bearing boundary check. Triples that are clearly infeasible (e.g., exceeding the unit's load-bearing capacity, conflicting with existing occupied spaces, or having insufficient support ratios) are directly discarded to reduce subsequent computational load. To control scale, a grouping / bucketing strategy (e.g., binning by cargo volume / attitude) can be adopted, and the number of anchor points for each cargo in each subspace can be limited.

[0103] For each feasible triplet, calculate its gain ΔF with respect to the overall objective function F, and sort all feasible triplets in descending order of ΔF value (a priority queue / heap can be used to improve efficiency). In each iteration, take the triplet with the largest current gain and place it; after successful placement, update the loaded volume, loaded weight, and available subspace set of the corresponding loading cell (using a strategy of space cutting and merging adjacent subspaces), and record the placement decision; remove all triplets that have geometric or load-bearing conflicts with the placement from the feasible triplet set (these conflict determinations can be quickly detected by bounding boxes or based on subspace overlap determination); only recalculate ΔF for affected triplets (i.e., those whose dependent subspaces have been changed or whose weight state of the cell has changed significantly), and re-push the updated value into the heap; other unaffected triplets retain their original values ​​to save computation.

[0104] Iteration termination conditions and backup strategies: The iteration terminates when any of the following conditions are met: all goods have been placed; or the maximum ΔF in the pile is ≤0 (i.e. no placement can bring positive target gain); or the preset maximum number of iterations or time threshold is reached.

[0105] Once an initial solution is constructed using a greedy strategy, unplaced goods or goods that are stopped from being placed due to ΔF≤0 can be handled by subsequent optimizers (e.g., Adaptive Large Neighborhood Search ALNS), or a safety net strategy can be adopted (e.g., continuing to place goods even if the minimum negative gain is allowed, or adding the goods to a delay queue for a second attempt).

[0106] Parallelism and disambiguation rules: If multiple triples have the same or nearly the same ΔF, a preset secondary disambiguation rule is adopted: the triple that can reduce Frag the most is selected first; if they are still the same, the triple that can increase U is selected first; if they are still the same, the triple that can reduce Imb is selected; if there is still no difference, the decision is made based on a stable hash (such as the item ID) to ensure the reproducibility of the execution.

[0107] Efficiency optimization and implementation details: After each placement, ΔF is only updated incrementally for the affected triples to avoid full recalculation; during placement, triples with low space utilization efficiency or infeasible support / grabbing are penalized by downgrading or directly removed; ΔF can be calculated in parallel for different loading units or different groups, and the stack can be constructed in parallel; the intermediate space state and scoring details of each placement are saved for easy backtracking and human / machine review; weights α, β, γ, maximum number of attempts per item, stack depth M, etc. are all configurable items to adapt to different business requirements and computing resources.

[0108] The purpose of using the maximum greedy algorithm as an initial constructor is to quickly generate a high-quality, executable initial solution, which can then be further improved using iterative optimization methods such as ALNS. In implementation, the solution obtained through the greedy algorithm (including the list of unplaced goods and candidate residual space) should be used as the initial input for ALNS.

[0109] The technical solution implemented in this embodiment can greedily select each candidate placement decision using the objective function gain ΔF as a metric, thereby obtaining an executable initial solution that balances volume utilization and fragmentation in a short time. By utilizing feasible triplet pre-screening, incremental ΔF updates, and conflict triplet removal, unnecessary redundant calculations are avoided, improving construction efficiency. Through support, load-bearing, and safety clearance verification, as well as consistency disambiguation rules, the generated solution is guaranteed to have high feasibility for implementation on automated loading equipment. The greedy initial solution typically has a good local structure (low fragmentation, high utilization), enabling subsequent metaheuristic algorithms such as ALNS to converge faster and achieve better improvements within a finite number of iterations.

[0110] Furthermore, the calculation of the remaining space distribution in S30 includes: recording the occupied and unoccupied areas in the loading unit according to three-dimensional coordinates; performing connectivity analysis on the unoccupied areas to calculate the volume and shape of each unoccupied area; and evaluating the availability of the remaining space based on shape matching.

[0111] In this embodiment, it should be noted that the spatial state within the loading unit is saved in a three-dimensional representation. Based on this, connectivity and shape analysis are performed on unoccupied areas to quantify the availability of remaining space. First, a three-dimensional coordinate system is established for each loading unit, and the placed goods are geometrically projected into this coordinate system to form a three-dimensional occupancy representation of the loading unit. The occupancy representation can be implemented using one or both of the following methods: one is voxelgrid representation, which discretizes the cavity into voxel units and marks them as occupied or vacant; the other is analytic subspace representation, which records the currently available subspaces using axis-aligned cuboid sets. The voxelgrid method is easy to perform connectivity analysis and shape measurement, while the analytic representation facilitates accurate volume calculation and fast space cutting. To balance accuracy and computational overhead, the voxel resolution or the minimum size of the analytic subspace (configurable parameter) should be set according to the scenario, and the implementation should support compression from the voxel structure to the analytic subspace (e.g., octree / octree merging or cuboid merging algorithms).

[0112] After obtaining the 3D representation of occupied / vacant areas, connectivity analysis is performed on the unoccupied regions to extract several connected cavities. Connectivity analysis can be based on voxel adjacency rules (e.g., using 6-neighborhood or 26-neighborhood connectivity) to flood / fill the voxelized representation, or it can be performed by constructing a connected graph based on spatial intersection relationships in the analytic subspace representation and extracting connected components. Each connected cavity is further calculated for its geometric characteristics: total volume (number of voxels × voxel volume or sum of analytic subspace volumes), circumscribed length × width × height, surface area, shortest / longest axis length, aspect ratio, and the fill ratio of the circumscribed minimum cuboid to the actual cavity volume. To characterize shape regularity and elongation, shape factors, such as the surface area to volume ratio, can be calculated. Or the maximum / minimum circumscribed axis ratio, etc., these quantitative indicators are used for subsequent fragmentation measurement and shape matching.

[0113] Based on the aforementioned connected cavities and shape characteristics, a fragmentation metric for remaining space (Frag) is constructed. One feasible implementation is as follows: First, all connected cavities are sorted by volume from largest to smallest, and a usable volume threshold V is set. th (As related to the minimum volume of typical goods), the proportion of total cavity volume less than a threshold is calculated as a direct indicator of fragmentation; at the same time, a shape factor weighting is introduced, such as Frag, which reflects both the proportion of small-volume cavities and the impact of cavity shape on usability. The implementation can provide multiple Frag calculators to adapt to different business focuses (such as scenarios that emphasize pluggability or scenarios that emphasize volume recycling).

[0114] A shape matching evaluation is performed on each connected cavity to assess its suitability for candidate cargo. The shape matching is obtained through the following steps: First, calculate the minimum bounding box (or the minimum bounding box viaPCA of the optimal orientation) for the cavity to obtain the oriented dimensions (Lc, Wc, Hc); then, calculate the bounding box dimensions (lg, wg, hg) of the candidate cargo within its allowed placement postures (multiple postures are generated depending on whether the cargo is allowed to rotate); for each posture, calculate the size matching ratio r = min{Lc / lg, Wc / wg, Hc / hg} (if any dimension < 1, the posture cannot be placed); combine the size matching ratio with the volume ratio (Vc / Vg) and the accessibility factor (as described below) to form a shape matching score Smatch, for example:

[0115] Among them, A access For reachability or crawling feasibility scoring (between 0 and 1), ω1, ω2, and ω3 are weight parameters. If S match If the value exceeds a preset threshold, the cavity is marked as usable for the cargo, and a specific placement posture suggestion is returned. In practice, the feasibility of the specific placement can be verified through greedy or local search (e.g., simulating placement within the cavity and detecting whether it interferes with existing cargo and meets the support ratio).

[0116] Furthermore, the availability assessment of remaining space should also incorporate loading executability constraints, including but not limited to: gripping path and operational accessibility, gripping / placement posture limitations of the loading equipment's end effector, support ratio requirements for vertical stacking, and safety clearance requirements. Therefore, in the shape matching assessment process, "executability factors" should be calculated in parallel: gripping / placement path accessibility (based on the loading machine's workspace or a path model from the entrance to the cavity); bottom support ratio satisfaction (bottom force-bearing area of ​​the placement location / bottom surface area of ​​the cargo); and whether the minimum safety clearance with adjacent cargo is met. These executability factors should be merged into S... match This helps avoid decisions that are theoretically feasible but practically inoperable.

[0117] To improve computational efficiency and scalability, this embodiment proposes several implementation optimizations: using octrees or voxel compression representations to reduce memory overhead; employing parallel flood-fill or partitioned parallel processing for connectivity analysis; adopting a candidate-first strategy for shape matching (e.g., first attempting to fit the largest cavity to a larger item) to reduce unnecessary combination attempts; and incrementally maintaining cavity indexes (e.g., R-trees) to quickly locate affected cavities and re-evaluate only them after placing / removing items. For scenarios with high real-time requirements, a hierarchical strategy of coarse-resolution fast evaluation + fine-resolution fine evaluation can be used to balance speed and accuracy.

[0118] The output of this embodiment includes, but is not limited to: a list of connected cavities for each loading unit (including coordinate boundaries, volume, and shape factor), a candidate cargo matching list for each cavity, and a corresponding S... match Scores, overall Frag values, and total available volume (cumulative volume that can be utilized by a specific cargo group). These data serve as the evaluation results of S30 and are input into the final decision-making module of S40. They can also be used in a human-visualized interface to support review and manual adjustments. All intermediate data (such as voxelization resolution, connectivity threshold, and shape matching weight) are recorded as configurable parameters for reproducibility and optimization.

[0119] The technical solution implemented in this embodiment can accurately quantify the unoccupied space within the loading unit in units of connected cavities, obtaining the actual usable volume and shape characteristics that can be used for loading, thereby providing a refined basis for subsequent placement decisions; by combining the theoretically configurable cavities with actual operability through shape matching degree and executability scoring, the probability of generating undesirable solutions that are "loadable but inoperable" is significantly reduced; by using fragmentation measurement (Frag) and cavity priority ranking, the initial constructor and iterative optimization algorithm are guided to prioritize the use of well-structured large cavities, reducing the generation of fragmented spaces and improving the overall volume utilization rate; it supports incremental and parallel computing, enabling acceptable computational response time in batch or real-time packing scenarios; and it provides clear cavity-cargo matching suggestions and visualization basis for manual review and automated execution, facilitating seamless integration of automation and manual collaboration in actual park operations.

[0120] Furthermore, the process of selecting the optimal combination in S40 employs an adaptive large neighborhood search algorithm, including: a destruction operator: removing some cargo loading positions from the current combination; and a repair operator: reselecting loading positions for the removed cargo. The destruction and repair processes are repeated until the required number of iterations or convergence conditions are met. The destruction operators include: random removal, removal of cargo with the smallest volume, and removal of cargo with the highest degree of fragmentation in the remaining space. The repair operators include: random selection of loading positions and selection of loading positions with the highest matching degree in the remaining space.

[0121] In this embodiment, it should be noted that the Adaptive Large Neighborhood Search (ALNS) is used as the main iterative optimizer in S40 to select the optimal bin packing combination from the candidate combinations. The core idea of ​​ALNS is to alternately use several ruin / destroy operators and several repair operators to escape local optima and explore a better solution domain through local "destruction-reconstruction". The following is a detailed explanation of the key implementation points.

[0122] Overall Iterative Framework and Termination Condition: Before the algorithm starts, the initial solution (initial packing combination) from S20 / S30 and related intermediate quantities are input (the remaining cavity list of each loading unit, the S value of each cavity). match (List, current Frag, U, Imb, etc.). ALNS uses this initial solution as the current solution S. cur And record the current optimal solution S. best The main loop runs according to the maximum number of iterations or the time limit. Common termination conditions include: reaching the maximum number of iterations N. max The time consumed exceeds the time threshold T. max Or, if there is no improvement within a certain number of consecutive iterations (stagnation determination). Each iteration includes: selecting a destruction operator → performing destruction (removing the positions of k goods) → selecting a repair operator → performing repair (reassigning positions to the removed goods) → evaluating the generated new solution → deciding whether to replace the current solution with the new solution according to the acceptance criteria.

[0123] Types and parameterization of destruction operators: Destruction operators are used to remove a set of cargo placement decisions from the current solution, with the aim of freeing up space and creating new combinations for repair. The destruction operators used in the embodiments include (but are not limited to):

[0124] Random Removal: Randomly select k items and remove them from the container. k can be a fixed value or randomly sampled proportionally (e.g., p% of the total number of items) to maintain diversity.

[0125] Smallest-Volume Removal: Prioritizes the removal of items with the smallest volume (or volume × placement difficulty) to reduce space fragmentation and facilitate relocation.

[0126] Fragmentation-driven Removal: Identify the placements that contribute the most to the current fragmentation (such as boundary objects that cause fragmentation in the cavity or goods that create a large number of small cavities when cut), and remove these goods to reduce overall fragmentation.

[0127] Each destruction operator can be further subdivided (e.g., hierarchical random, nearest neighbor removal, same group removal, etc.); the removal scale k can be dynamically adjusted (small / medium / large levels) to balance exploration and exploitation. After removal, the available subspace and connected cavity index of the affected loaded cell need to be updated.

[0128] Repair operator type and implementation details: The repair operator is responsible for restoring removed goods back to certain loading units, restoring or improving the quality of the current solution. The repair operator in the embodiment includes: RandomRepair, which randomly selects feasible placement locations from its candidate cavity / attitude set for each removed goods and attempts to place it, suitable for exploratory repair to introduce diversity.

[0129] Loading is performed at the location with the highest remaining space matching degree (GreedyShape-MatchRepair): Utilizing the cavity-cargo S calculated in S30 match Scores, ranked from highest to lowest S for each removed item. match Try placing items sequentially until successful placement or the maximum number of attempts is reached. This strategy is more insightful and can quickly improve local scores.

[0130] The repair process is typically performed using either sequential insertion or parallel insertion. Sequential insertion places items one by one in a sequence (e.g., in descending order of difficulty). Parallel insertion uses a greedy or matching algorithm to allocate cavities to multiple items simultaneously, preventing the placement of one item from blocking subsequent placements. To improve the success rate, local neighborhood searches can be embedded in the repair process (e.g., attempting a small number of local rearrangements or rotations), but this is limited by the repair time budget.

[0131] New Solution Evaluation and Acceptance Criteria: The new solution S obtained after repair new The objective function value F(S) needs to be calculated. new (same as αU-βFrag-γImb as defined in S20), and compared with the current solution S cur F(S) cur Comparison. Accepting a new solution can be done using one or a combination of the following strategies:

[0132] Greedy acceptance: If F(S) new )>F(S cur Then accept;

[0133] Simulated annealing (Metropolis) accepts: if F(S) new )≤F(S cur ), in probability Accept (T is the temperature, which decreases with iteration) to escape local optima;

[0134] Threshold acceptance / gradual shrinkage threshold: If F(S) new )≥F(S cur If -∈, then accept; ∈ gradually decreases with iteration.

[0135] If F(S) is accepted, new ) is better than F(S) bestThen update the global optimal solution and record the relevant metadata.

[0136] Adaptive Operator Selection and Weight Update Mechanism: To enhance the effectiveness of operator combinations, ALNS introduces an adaptive operator selection mechanism. It maintains a set of weights or scores for each destruction / repair operator, selecting the operator in each iteration based on a probability distribution (normalized by weight). The weights are dynamically updated based on the operator's contribution feedback. For example, if a destruction-repair pair used in an iteration produces a significant improvement to the global optimum (a substantial increase in F), the corresponding operator weight receives a large reward; if it only improves the current solution but does not update the global optimum, a moderate reward is given; if no improvement is produced, the operator weight is gradually reduced.

[0137] A commonly used weight update rule is as follows: assign the weight w to w←(1-ρ)w+ρr, where ρ is the memory parameter and r is the reward value obtained by this operator (which can be a discrete level). This adaptive mechanism allows the algorithm to automatically favor combinations of operators that perform well during operation.

[0138] Embedded constraints and executability checks: During the repair phase, an executability check (geometric non-overlap, support ratio, load-bearing limits, safety clearance, reachability of the grab path, etc.) is performed on each placement. Infeasible placements are rejected and counted as failed attempts. For solutions that are feasible but mechanically risky, a penalty term can be introduced to reduce their F-value or they can be directly prohibited. This embedding ensures that the solutions generated by the algorithm are not only theoretically excellent but also executable on actual loading equipment.

[0139] Incremental Assessment and Efficiency Optimization: To meet engineering response requirements, the implementation of ALNS should employ incremental assessment as much as possible: after damage, only the Frag / U / Imb contributions of the affected loading unit or cavity need to be recalculated, without performing a global recalculation from scratch; spatial indexes such as R-trees / octrees are used to quickly locate the affected cavity; the recalculation of ΔF uses the local influence range (neighborhood of removeset) to save computation. Parallelization allows different repair strategies or different damage intensities to be tried in parallel on multiple CPU cores or nodes to shorten the wall-clock time.

[0140] Preventing Cycles and Diversity Control: To avoid getting stuck in repeating cycles, the system should maintain a recent solution history (tabu-like list or hash set) and temporarily disable destructive-repair combinations that have been tried and yielded poor results; at the same time, it should maintain solution diversity by setting random removal / repair probabilities and varying removal scales.

[0141] Logs, backtracking, and human interaction: Key operations in each iteration (operators used, set removal, repair actions, new solution scoring) are logged for easy auditing and human replay. If human intervention is required (see S40's visualization and manual adjustment functions), users are allowed to manually mark prohibited / forced placement areas, and ALNS will respect these human constraints in subsequent iterations.

[0142] Parameter configuration suggestions: Commonly used empirical values ​​(can be optimized according to the scenario): Maximum number of iterations N max =10 4 Time threshold T max =30s (real-time scenario) or higher (offline batch optimization); damage scale k is divided into levels within the range of 5%–30% of the number of goods; adaptive weight memory parameter ρ∈[0.05,0.2]; the initial temperature T can be set to several times the scale of the objective function, and the temperature decreases linearly or geometrically with iteration.

[0143] The technical solution implemented in this embodiment can achieve exploration of a large solution domain and high-quality local reconstruction through diverse destruction operators (random, volume-based, fragment contribution-based, etc.) combined with heuristic repair. Using Frag and shape matching to guide the repair operation prioritizes filling well-structured cavities during reconstruction, reducing the generation of unusable small gaps and thus improving the overall U. Explicitly considering Imb in the objective function and prioritizing Imb reduction during repair yields a more balanced multi-unit loading scheme, reducing transportation safety risks. By embedding executability verification into the repair process and employing incremental evaluation and adaptive operator weight updates, the algorithm maintains computational overhead suitable for engineering while ensuring solution quality. Iteration logs, backtracking mechanisms, and manual constraint interfaces facilitate integration with human decision-making in actual park operations, improving the reliability and acceptability of deployment and maintenance.

[0144] Furthermore, the packing plan output includes: the three-dimensional placement and orientation of the goods within each loading unit; the volume utilization rate, weight utilization rate, and remaining space distribution of each loading unit; and a list of goods that could not be loaded and the reasons therefor.

[0145] In this embodiment, it should be noted that the output of the packing plan includes not only the final placement result, but also a series of detailed data related to the quality of the plan. First, for each loading unit, the system outputs the three-dimensional placement position and orientation information of the goods within it. Specifically, the placement position of each item is described by three-dimensional coordinates (X, Y, Z) and rotational attitude parameters, thus accurately reflecting its positional relationship and placement status within the loading unit. This information can be used to generate three-dimensional visualization images for operator review, and can also serve as input for execution commands from automated loading equipment, achieving seamless integration from the plan to actual operation.

[0146] Secondly, the system quantitatively evaluates the loading effect of each loading unit, outputting volume utilization rate, weight utilization rate, and remaining space distribution. Volume utilization rate measures the tightness of cargo filling in space, weight utilization rate reflects the utilization of the loading unit's load-bearing capacity, and the remaining space distribution is calculated based on three-dimensional connectivity analysis and spatial shape matching, displaying the shape and location of unused space. These indicators not only serve as feedback data for optimization goals but also provide a basis for decision-making in subsequent scheduling or manual adjustments.

[0147] In addition, the system generates a list of goods that failed to load, noting the specific reason why each item was not loaded, such as exceeding size limits, weight exceeding the unit's load capacity, not meeting stacking restrictions, or insufficient loading space. In this way, managers can quickly locate problematic goods and decide whether to replace the loading unit, split the goods, or manually intervene to load them, thereby improving the transparency and controllability of the overall packing operation.

[0148] The technical solution implemented in this embodiment provides comprehensive packing information at the output stage, ensuring that the final solution includes not only the placement results but also detailed quality assessments and explanations of the reasons for any unloaded goods. On one hand, this guarantees the feasibility and traceability of the solution, enabling operators to quickly understand and verify the packing results. On the other hand, by outputting a list of unloaded goods and the reasons for their absence, it helps park managers take timely remedial measures, improving the flexibility and adaptability of the packing process. Overall, this embodiment significantly enhances the operability, controllability, and application value of the intelligent packing method in actual park logistics operations.

[0149] Furthermore, in step S40, after selecting the optimal combination, the packing scheme is visualized and rendered to generate a three-dimensional packing diagram for manual review or direct control of automated loading equipment. The three-dimensional packing diagram allows users to make manual adjustments via an input terminal, and the results of the manual adjustments are rewritten into the packing scheme and stored.

[0150] In this embodiment, it should be noted that in step S40, after the system selects the optimal packing combination, it further visualizes and renders the packing plan. Specifically, the system uses the three-dimensional dimensions, position coordinates, and spatial data of the loading unit to generate a corresponding three-dimensional packing image. This three-dimensional packing image can intuitively display the placement of goods in the loading unit, including the relative position, orientation, and spatial relationship of each item with surrounding goods and the boundaries of the loading unit. In this way, users can understand the packing plan more intuitively without relying on complex data tables or abstract descriptions.

[0151] Based on this, the 3D packing diagram can not only serve as a visual display but also be used for manual review. Users can interact with the 3D diagram through input terminals (such as tablets, touch screens, or computer interfaces). When users find that the automatically generated packing plan has unreasonable aspects in certain local areas (such as goods facing inconvenient orientation, uneven weight distribution, or potential stacking risks), they can manually adjust the position and orientation of some goods by dragging, rotating, or other operations. This manual adjustment operation is recorded by the system in real time, and the modified packing result is rewritten into the packing plan, thereby updating the stored data. This not only retains the efficiency of automated calculation but also introduces the corrective capability of human experience.

[0152] Furthermore, the visualized 3D packing diagram can be directly input as instructions into automated loading equipment. For example, robotic arms or automated stacking equipment can directly perform loading operations based on the coordinates and orientation information in the 3D diagram, thereby reducing manual intervention and improving operational efficiency. This approach allows packing solutions to go beyond the calculation and display stage and be directly implemented in practice.

[0153] The technical solution implemented in this embodiment organically combines computer automation with human experience in the packing result output stage. On the one hand, 3D visualization rendering improves the intuitiveness and understandability of the packing plan, enabling users to quickly identify the rationality of the plan; on the other hand, the manual adjustment function ensures flexible handling of complex or special packing scenarios, improving the practicality and reliability of the plan. Simultaneously, by directly transmitting the final result to the automated loading equipment, this solution achieves seamless integration between the plan and execution, significantly enhancing the practical value and operational efficiency of the intelligent packing method in actual park logistics operations.

[0154] This invention also provides an intelligent packing system suitable for allocating multiple goods to multiple loading units, comprising: an input terminal, a data storage module, and a processing module. The input terminal receives packing task information, including the size, weight, stacking limits, and priority of each goods, as well as the size, load-bearing capacity, and occupied status of each loading unit. The data storage module is connected to the input terminal and stores the packing task information and packing plan. The processing module is connected to the data storage module and executes any one of the methods from 1 to 9 based on the packing task information to generate a packing plan.

[0155] In this embodiment, it should be noted that the intelligent packing system provided by the present invention is mainly used to efficiently allocate multiple goods to multiple loading units, thereby improving packing efficiency, optimizing space utilization, and reducing manual intervention. The system includes three core components: an input terminal, a data storage module, and a processing module.

[0156] The input terminal is used to receive packing task information, which includes the dimensions, weight, stacking restrictions, and priority of each item, as well as the dimensions, load-bearing capacity, and occupied status of each loading unit. Through the input terminal, operators can quickly input or import the various parameters required for the packing task, ensuring the accuracy of subsequent calculations.

[0157] The data storage module connects to the input terminal and stores the received packing task information and the packing scheme generated by the system. The data storage module not only provides reliable data preservation but also supports the subsequent retrieval, modification, and reuse of packing schemes, giving the system continuity and flexibility.

[0158] The processing module is connected to the data storage module and is used to execute any of the above methods based on the stored packing task information, and generate the final packing plan. The processing module can optimize the layout of the goods in each loading unit through algorithms according to the size, weight, stacking restrictions of the goods, and the load-bearing capacity and available space of the loading unit, so as to ensure that the packing plan meets both physical constraints and priority requirements.

[0159] The technical solution implemented in this embodiment can achieve automatic allocation and layout optimization of multiple goods, significantly improving packing efficiency; reducing manual operation and human error; improving the space utilization of loading units; and the system has the ability to store and reuse solutions, providing convenient support for subsequent packing tasks, thus having high practical value in logistics, warehousing and automated loading scenarios.

[0160] The embodiments described above are merely illustrative of several implementations of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the scope of protection of this application. Therefore, the scope of protection of this patent application should be determined by the appended claims.

Claims

1. A smart packing method, suitable for allocating multiple goods to multiple available loading units, characterized in that, include: S10. Obtain packing task information, which includes the size, weight, stacking limit, quantity and priority of each item, as well as the size, load-bearing capacity and occupied status of each loading unit. S20. Under the conditions of satisfying the size constraints, load-bearing constraints and stacking limitations of the loading unit, generate multiple candidate packing combination schemes. S30. Evaluate each of the candidate combination schemes based on the preset optimization target, and calculate the loading utilization rate and the remaining space distribution; S40. Based on the evaluation results, select the optimal packing combination from the candidate packing combination schemes as the final packing scheme.

2. The intelligent packing method according to claim 1, characterized in that, The following is included before S20: S11. Based on the size, weight, stacking restrictions, and priority of the goods, divide the goods into multiple groups with similar attributes; S12. Arrange the goods in each group according to the preset sorting rules to reduce the search space and improve the matching degree.

3. The intelligent packing method according to claim 1, characterized in that, The preset optimization objective is: The loading scheme with the highest volume utilization, the lowest degree of fragmentation of remaining space, and the least differentiation of loading units is selected, and the optimization target value is obtained by weighted summation of the three factors.

4. The intelligent packing method according to claim 3, characterized in that, S20 includes: S21. According to the cargo sorting, the cargo is sequentially allocated to loading units that meet the size, load-bearing and stacking restrictions; S22. When direct placement is not possible, generate new candidate combinations by rotating the goods or adjusting the position of existing goods.

5. The intelligent packing method according to claim 4, characterized in that, The initial bin packing combination generated in S20 uses a maximum greedy algorithm, specifically including: Calculate the objective function gain ΔF for each triplet in all feasible sets of "goods, placement location, placement posture"; In each iteration, the triplet with the largest gain ΔF of the objective function is selected for placement, and triplets that have geometric or load-bearing conflicts with this placement are removed from the feasible set. Repeat the iteration until there are no feasible triples with ΔF>0 or all goods have been placed; The objective function F is defined by the following formula: F=α×U-β×Frag-γ×Imb In the formula, U is the volume utilization rate, Frag is the fragmentation measure of the remaining space, Imb is the measure of the uneven weight distribution among loading units, α, β, and γ are non-negative weighting coefficients, and ΔF is the difference of F before and after the placement operation.

6. The intelligent packing method according to claim 5, characterized in that, The calculation of the remaining spatial distribution in S30 includes: Record the occupied and unoccupied areas in the loading unit according to three-dimensional coordinates; Perform connectivity analysis on unoccupied regions and calculate the volume and shape of each unoccupied region; The availability of remaining space is evaluated based on shape matching.

7. The intelligent packing method according to claim 1, characterized in that, The process of selecting the optimal combination in step S40 employs an adaptive large neighborhood search algorithm, including: Destruction operator: Removes some cargo loading positions from the current combination; Repair operator: Reselects a loading location for removed cargo; Repeat the destruction and repair process until the number of iterations or the convergence condition is met; The destruction operators include: random removal, removal of the smallest item by volume, and removal of the item with the greatest fragmentation of remaining space; The repair operator includes: randomly selecting a location for loading and selecting a location with the highest remaining space matching degree for loading.

8. The intelligent packing method according to claim 1, characterized in that, The contents of the packing solution output include: The three-dimensional placement and orientation of the goods within each loading unit; Volume utilization rate, weight utilization rate, and remaining space distribution of each loading unit; List of goods that could not be loaded and the reasons.

9. The intelligent packing method according to claim 1, characterized in that, In step S40, after selecting the optimal combination, the packing scheme is visualized and rendered to generate a three-dimensional packing diagram for manual review or direct control of automated loading equipment. The three-dimensional packing diagram allows users to make manual adjustments via an input terminal, and the results of the manual adjustments are rewritten into the packing scheme and stored.

10. An intelligent packing system suitable for distributing multiple goods to multiple loading units, characterized in that, include: An input terminal is used to receive packing task information, which includes the size, weight, stacking limits and priority of each item, as well as the size, load-bearing capacity and occupied status of each loading unit. A data storage module, connected to the input terminal, is used to store the packing task information and packing plan; The processing module is connected to the data storage module and is used to execute the method described in any one of the packing task information (1 to 9) and generate a packing scheme.

Images