Dynamic slot reconfiguration and task allocation method for wave-oriented sorting

Abstract

This invention relates to the field of automatic control technology, and more specifically, to a dynamic location reconfiguration and task allocation method for wave picking. The method includes: acquiring storage facility layout and aisle network data of the warehouse environment to construct a picking area topology map; reading target order data for wave tasks to be processed to form a dynamic location reconfiguration scheme; and decomposing and recombining the target order data into regional task sets according to regional nodes and wave operation time windows based on time period color labels, and allocating the regional task sets to picking operation resources according to a load balancing strategy to drive the picking operation resources to complete the picking operation within the corresponding wave operation time window. This invention can dynamically adapt to structural changes in different wave tasks without changing the overall warehouse layout, reducing cross-area movement and lowering local aisle pressure.

Description

Technical Field

[0001] This invention belongs to the field of automatic control technology, specifically relating to a dynamic location reconstruction and task allocation method for wave picking. Background Technology

[0002] In modern warehousing and logistics operations, wave picking, as a relatively mature operational organization method, has been widely used in large and medium-sized warehouses. Typically, warehousing systems group multiple orders into a wave task based on factors such as order arrival time, delivery timeliness, and product type, and then complete the picking operation centrally within a specified time period. To reduce the walking distance of picking personnel and improve overall picking efficiency, existing technologies generally employ location optimization methods based on historical order statistics, placing frequently outbound items in fixed locations near aisle entrances or main aisles. However, this fixed location optimization method generally assumes that the demand pattern for goods is relatively stable and that the differences in the item structure between different wave tasks are not significant, making it difficult to adapt to business scenarios with frequently changing order structures.

[0003] On the other hand, some existing warehouse management systems have also introduced picking route generation technology based on area division. For example, the warehouse is divided into several manually defined static picking areas, with each picking operator or picking device performing local picking tasks within one area, and then the entire order is completed through aggregation. However, these static area division methods mostly rely on manual experience to set area boundaries, without considering the differences in connectivity between areas or whether the items in the order are related across multiple areas. For example, when two areas have a large demand for cross-area items in the same wave of tasks, picking operators need to frequently travel between the two areas, resulting in high personnel density in the connecting aisles, causing local congestion and reducing overall operational efficiency. Existing technologies usually alleviate this congestion by increasing the number of picking operators or widening the aisles, but this approach is costly and cannot eliminate the root cause of congestion at the algorithmic level. Summary of the Invention

[0004] The main objective of this invention is to provide a dynamic location reconfiguration and task allocation method for wave picking. By statistically analyzing the passage weight and busy weight values ​​generated from target order data, graph coloring operations are performed on regional nodes to generate time-period color labels corresponding one-to-one with wave operation time windows. This effectively separates highly correlated areas in time, thereby reducing the risk of aisle congestion. Furthermore, the target order data is decomposed into regional task sets according to regional nodes and wave operation time windows. A closed-loop picking path is generated and optimized using a dynamic location reconfiguration scheme to significantly shorten the picking walking distance within each region. Subsequently, based on the total passage cost, a load balancing strategy is employed to allocate the regional task sets to picking operation resources, achieving task balance and improved operational efficiency among multiple resources. Through this collaborative process, this invention can dynamically adapt to structural changes in different wave tasks without altering the overall warehouse layout, reducing cross-regional movement, lowering local aisle pressure, improving picking efficiency, and optimizing resource utilization, resulting in a significant improvement in the overall performance of warehouse wave picking operations.

[0005] To solve the above problems, the technical solution of the present invention is implemented as follows:

[0006] A dynamic location reconfiguration and task allocation method for wave picking includes: acquiring storage facility layout and channel network data of the warehousing environment; discretizing the physical space into multiple picking area units and mapping them as area nodes; establishing area adjacency edges based on physical connectivity to construct a picking area topology graph; reading target order data of the wave tasks to be processed; statistically analyzing the demand correlation of each area node and its adjacency edges based on the mapping relationship between items and storage locations to obtain the passage weight value of the area adjacency edges and the busy weight value of the area nodes; and, under a preset congestion threshold constraint, performing graph coloring operations on the area nodes based on the passage weight value and the busy weight value to allocate each area node to a wave. Each operation time window is labeled with a corresponding time period color tag. Based on the time period color tag, high-frequency demand items in the target order data are dynamically reconstructed and allocated to priority storage locations in the active area nodes within the wave operation time window, forming a dynamic storage location reconstruction scheme. The target order data is then decomposed and reorganized into area task sets according to area nodes and wave operation time windows based on the time period color tag. Closed-loop picking paths for each area task set are generated and optimized based on the dynamic storage location reconstruction scheme. The corresponding passage cost is calculated, and the area task sets are allocated to picking operation resources according to a load balancing strategy to drive the picking operation resources to complete the picking operation within the corresponding wave operation time window.

[0007] Furthermore, the calculation of the passage weight value specifically includes: initializing the passage weight value of each region's adjacent edge to zero; parsing the independent orders in the target order data one by one, determining the involved region node subsets based on the item storage location, and for the region node pairs directly connected by region adjacent edges in the subset, incrementally accumulating the passage weight value of the corresponding region adjacent edge according to the preset single passage cost, until all independent orders are processed, and obtaining the final passage weight value of each region adjacent edge.

[0008] Furthermore, the calculation of the busy weight value specifically includes: initializing the busy weight value of each regional node to zero; counting the total frequency of item demand falling within the range of each regional node in the target order data and accumulating it to the corresponding node; then weighting and accumulating the final passage weight value of the adjacent edges of the region directly connected to the node according to the preset influence factor, and the resulting value is used as the final busy weight value of the regional node, which is used to determine the node processing priority in the graph coloring operation.

[0009] Furthermore, the graph coloring operation specifically includes: establishing a color tag library containing multiple distinct time period color tags, the number of which is greater than the maximum adjacency degree of any region node; sorting all region nodes in descending order according to the final busy weight value to form a node processing priority sequence; when selecting each region node as the current processing node, searching its neighbor nodes for nodes that have completed color assignment, if the final passage weight value of the region adjacency edge between the current processing node and a colored neighbor is greater than or equal to the first congestion judgment threshold, then marking the neighbor node as a strong conflict neighbor and adding its time period color tag to the forbidden label list; removing the labels from the forbidden label list from the color tag library to obtain a candidate label subset, if the candidate label subset is not empty, selecting the time period color tag that has been used the least in the topology graph of the selected region and assigning it to the current processing node, if the candidate label subset is empty, adding an unused time period color tag to the current processing node and adding it to the color tag library.

[0010] Furthermore, after completing the initial time-period color label allocation, a local color optimization step is also included, specifically: traversing all regional adjacent edges, selecting regional adjacent edges that simultaneously satisfy the condition that the final passage weight value is greater than or equal to the second congestion judgment threshold and that the two end regional nodes have the same time-period color label as the edge to be optimized; for each edge to be optimized, comparing the final busy weight values ​​of the two end regional nodes, selecting the smaller one as the node to be adjusted, and obtaining the time-period color labels of all strongly conflicting neighbors of the node to be adjusted to form a local forbidden set; removing the local forbidden set from the color label library to obtain an adjustment feasible set; based on each potential time-period color label in the adjustment feasible set, evaluating the number of regional adjacent edges with the node to be adjusted as the endpoint and the same label at both ends after changing the label of the node to be adjusted to the potential label, selecting the potential time-period color label that minimizes this number as the new label to update the time-period color label of the node to be adjusted, and completing the above adjustment in descending order of the final passage weight value of the edge to be optimized to obtain the final regional node time-period color label allocation table.

[0011] Furthermore, the dynamic storage location reconfiguration scheme includes: for each regional node, calculating the walking distance of each storage location within its physical range relative to the aisle entrance, and generating a high-quality storage location index sequence table by sorting the walking distances from smallest to largest; based on the regional node time period color label allocation table and the mapping relationship between time period color labels and wave operation time windows, dividing the item demand in the target order data into multiple regional wave demand groups corresponding to specific regional nodes and specific wave operation time windows; for each regional wave demand group, counting the demand quantity of each item within the group and sorting it from largest to smallest to obtain a list of high-frequency items to be allocated, mapping the high-frequency items to be allocated sequentially to the storage locations in the high-quality storage location index sequence table, allocating all items to the top-ranked storage locations when the total number of items is not greater than the total number of storage locations, and retaining the excess items in their original storage locations when the total number of items is greater than the total number of storage locations, and recording the mapping change relationship between items and storage locations, as the dynamic storage location reconfiguration scheme for each regional node under the corresponding wave operation time window.

[0012] Furthermore, the decomposition and reorganization of the regional task set specifically includes: dividing each parent order in the target order data according to the regional node to which the item belongs and the wave operation time window, generating child order fragments that only contain the same regional node and need to pick items within the same wave operation time window; merging each child order fragment with the regional node and wave operation time window as the aggregation key to obtain an independent regional task set for each regional node under each wave operation time window.

[0013] Furthermore, generating and optimizing the closed-loop picking path specifically includes: reading the physical coordinates of all items to be picked in the independent area task set, determined according to the dynamic location reconstruction scheme, and setting the channel entrance coordinates of the area nodes as the path start and end points; using the nearest neighbor construction algorithm, starting from the path start point, selecting the physical coordinates closest to the current position in a straight line or Manhattan distance from the unvisited physical coordinates and adding them sequentially until all physical coordinates have been visited and connected back to the path end point, thus obtaining the initial path; then using the path topology exchange optimization algorithm, selecting two non-adjacent path connection segments in the initial path and exchanging their connection order. If the total path length decreases after the exchange, the exchange is retained and the iteration continues until the total path length no longer decreases within a preset number of iterations, and the resulting path is determined as the optimized closed-loop picking path.

[0014] Furthermore, the calculation of the total passage value and task allocation specifically includes: calculating the total passage value of each independent area task set based on the sum of the physical lengths of each path segment in the optimized closed-loop picking path, and storing each independent area task set in the task queue to be allocated in descending order of total passage value; establishing a resource status list, recording the cumulative passage value of each picking operation resource's allocated tasks in real time, and sorting them in ascending order of the cumulative value; adopting a greedy water-filling allocation strategy, sequentially taking the independent area task set from the head of the task queue to be allocated and allocating it to the picking operation resource ranked first in the resource status list, while updating the cumulative passage value and sorting position of the picking operation resource, until the task queue to be allocated is cleared, generating the final task allocation instruction set containing specific execution instructions.

[0015] The dynamic location reconstruction and task allocation method for wave picking of this invention has the following beneficial effects: By introducing graph coloring operations based on passage weight values ​​and busy weight values ​​on the picking area topology map, this invention enables the time-dimensional separation and scheduling of different area nodes in multiple wave operation time windows. This effectively reduces the high-frequency cross-area walking behavior generated by adjacent area nodes within the same time window while maintaining the warehouse physical layout, avoiding instantaneous high-density personnel flow at the passageway connecting two area nodes. In the process of allocating time period color labels, this invention actively disperses potentially congested area nodes into different wave operation time windows by identifying high passage weight values ​​related to the adjacent edges of the areas. This results in a more dispersed spatial distribution of area nodes within the same wave operation time window, fundamentally reducing the passage pressure on local passageways. After obtaining the color labels for regional nodes, this invention divides the target order data into multiple regional task sets according to regional nodes and wave operation time windows. Based on a dynamic location reconfiguration scheme, it generates optimized closed-loop picking paths for each regional task set. This allows path planning for each regional task set to be completed independently within the regional node, avoiding complex cross-regional jumps and reducing unnecessary movement of picking resources throughout the warehouse. Simultaneously, by prioritizing the allocation of high-frequency demand items to priority storage locations near aisle entrances, the average path length of regional task sets is significantly shortened, further improving picking efficiency. During the task allocation phase, this invention establishes a load balancing strategy using total passage cost as a metric, ensuring that the workload of multiple picking resources is balanced within the same wave operation time window. This avoids the problem of some resources being overloaded while others are idle, improving overall resource utilization. This invention organically combines regional node time period color label allocation, dynamic storage location reconstruction, regional task set decomposition, optimized closed-loop picking path generation, and load balancing scheduling to form a full-link collaborative optimization system from regional planning to storage location configuration and from task organization to resource scheduling. This enables the warehousing system to maintain stable and efficient operation under conditions of large order fluctuations or frequent changes in regional density, significantly improving wave picking efficiency and reducing the risk of aisle congestion. Attached Figure Description

[0016] Figure 1 This is a schematic diagram illustrating the distribution of regional adjacent edge passage weights and regional node busy weights in an embodiment of the present invention.

[0017] Figure 2 This is a schematic diagram showing the comparison before and after optimization of the closed-loop picking path for the regional task set provided in an embodiment of the present invention;

[0018] Figure 3 This is a schematic diagram of the graph coloring operation results provided in an embodiment of the present invention. Detailed Implementation

[0019] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0020] A dynamic location reconfiguration and task allocation method for wave picking includes: acquiring storage facility layout and channel network data of the warehousing environment; discretizing the physical space into multiple picking area units and mapping them as area nodes; establishing area adjacency edges based on physical connectivity to construct a picking area topology graph; reading target order data of the wave tasks to be processed; statistically analyzing the demand correlation of each area node and its adjacency edges based on the mapping relationship between items and storage locations to obtain the passage weight value of the area adjacency edges and the busy weight value of the area nodes; and, under a preset congestion threshold constraint, performing graph coloring operations on the area nodes based on the passage weight value and the busy weight value to allocate each area node to a wave. Each operation time window is labeled with a corresponding time period color tag. Based on the time period color tag, high-frequency demand items in the target order data are dynamically reconstructed and allocated to priority storage locations in the active area nodes within the wave operation time window, forming a dynamic storage location reconstruction scheme. The target order data is then decomposed and reorganized into area task sets according to area nodes and wave operation time windows based on the time period color tag. Closed-loop picking paths for each area task set are generated and optimized based on the dynamic storage location reconstruction scheme. The corresponding passage cost is calculated, and the area task sets are allocated to picking operation resources according to a load balancing strategy to drive the picking operation resources to complete the picking operation within the corresponding wave operation time window.

[0021] First, the system acquires storage facility layout and aisle network data from the electronic floor plan generated during the warehouse design or on-site surveying phase. The electronic floor plan records the coordinates of the four corner points of each set of shelves, shelf height information, and the polygonal coordinates of the aisle centerline in a two-dimensional planar coordinate system. For ease of subsequent processing, the system uses a fixed corner point on the warehouse floor as the origin, the direction extending along the main aisle as the first axis, and the direction perpendicular to the main aisle as the second axis, uniformly transforming the coordinates of all shelves and aisles. For example, if the long side of a warehouse is 60 meters along the first axis and the short side is 40 meters along the second axis, this is recorded in the electronic floor plan as the first axis ranging from 0 to 60 and the second axis ranging from 0 to 40. In this way, the storage facility layout and aisle network data are converted into a unified set of planar coordinate data, where the location of each set of shelves is described as a rectangular area, and each aisle is described as one or more polygonal lines formed by connecting several coordinate points.

[0022] After completing the coordinateization of the storage facility layout and aisle network data, the system discretizes the physical space, dividing the continuous storage space into multiple picking area units. To ensure that the picking area units reflect the actual walking stride of picking personnel without being too fine and affecting the efficiency of subsequent algorithms, the system sets the discretized grid size based on the average aisle width and shelf length of the warehouse. For example, when the main aisle width is 3 meters, the branch aisle width is 2 meters, and the typical spacing between shelves is 2 meters, the planar space can be divided into several rectangular grids with a length of 2 meters and a width of 2 meters. Starting from the origin of the coordinate system, the system generates grid dividing lines along the first axis with a step size of 2 meters, and then generates grid dividing lines along the second axis with the same step size of 2 meters, thus obtaining a regular grid covering the entire storage area. For each grid, the system calculates the overlap between the grid and the rectangular area of ​​the shelving and the aisle lines. If a grid is adjacent to the picking side of at least one set of shelving units and the nearest distance between the grid and the centerline of the aisle is less than 1 meter, the grid is marked as a picking-related area and defined as a picking area unit. If a grid is completely occupied by a wall or equipment and picking personnel are not allowed to enter, the grid is skipped and not considered a picking area unit. Using these rules, the system can generate approximately 600 candidate grids in a 60m x 40m warehouse, of which approximately 300 form valid picking relationships with the shelving and aisles; these 300 grids are identified as picking area units.

[0023] To facilitate the subsequent mapping of picking area units to area nodes, the system assigns a unique identifier to each picking area unit and records its geometric and business attributes. Geometric attributes may include the coordinates of the four corner points of the rectangular grid containing the picking area unit, the coordinates of the center point of the picking area unit, and the distance from the center line of the nearest aisle. Business attributes may include the shelf identifier associated with the picking area unit, the number of storage locations on one side of the shelf, and the maximum number of picking personnel allowed to enter the picking area unit simultaneously. In terms of data structure, the system creates a record for each picking area unit, writing its unique identifier, geometric attributes, and business attributes into the record. At this point, each picking area unit logically forms a candidate for an area node.

[0024] After the candidate regional nodes are established, the system determines which picking area units can be considered adjacent and reachable based on the physical connectivity between them, thus establishing regional adjacency edges. The determination of physical connectivity is not simply a comparison of grid numbers, but rather based on the actual walking paths of picking personnel within the warehouse. When determining the adjacency of any two picking area units, the system first checks whether they share a common boundary or corner point on the plane. If the two picking area units are only connected by an edge or corner on the planar grid, they are marked as geometrically adjacent regions. For each pair of geometrically adjacent regions, the system further checks whether there are obstacles between them, such as walls, enclosed fire doors, or fixed equipment. If the common boundary of the two is covered by an obstacle for more than half its length, it is considered that picking personnel cannot pass directly between these two picking area units, and these two picking area units are not considered physically connected. If the common boundary of the two is entirely located on the aisle area, or the distance between the common boundary and the centerline of the aisle is less than 0.5 meters, it is considered that picking personnel can freely pass through this location, and these two picking area units are considered physically connected. In this case, the system establishes a regional adjacency edge between the corresponding regional nodes of the two.

[0025] To improve picking efficiency and maintain a concise topology, the system can prioritize establishing adjacency edges between picking area units with longer shared boundary lengths. For example, if two picking area units share a 2-meter common boundary, while two other units share a 0.2-meter common boundary, the former is more likely to form a natural passageway, making it more likely that picking personnel will move between these two areas during actual operations. This approach avoids introducing numerous adjacency relationships rarely used in actual operations due to overly fine grid division, reducing the computational burden in subsequent graph coloring and path planning.

[0026] After establishing the regional adjacency edges based on shared boundaries, the system can further perform connectivity checks based on channel network data. In some warehouse layouts, the grids of two picking area units are not geometrically adjacent, but can be directly connected by a short channel. For example, one area unit is located on the left side of the main channel, and another area unit is located on the right side of the main channel, separated by the centerline of the main channel, with no common boundary between the grids. In this case, the system can calculate the nearest intersection point from the center point of each picking area unit to the centerline of the channel based on the coordinates of the channel centerline, and consider this intersection point as the entrance for picking personnel to enter the channel. When the nearest entrance points of two picking area units are located on the same channel polygon and the distance between them along the channel direction is less than 3 meters, the system considers that picking personnel can move quickly between these two areas without passing through additional picking area units, thus establishing a regional adjacency edge between the two corresponding area nodes. This supplementary connectivity check makes the picking area topology map closer to the actual walking path, rather than being limited to the geometric adjacency between regular grids.

[0027] After establishing adjacency edges between all eligible picking area units, the system completes the picking area topology graph. The picking area topology graph can be stored using an adjacency list, where each area node has a record entry listing the unique identifiers of all other area nodes that are adjacent to it. For example, in the aforementioned 300 picking area units, a picking area unit near the main warehouse aisle entrance is numbered 1, and its adjacent picking area units are numbered 2, 3, and 5. In the record for area node number 1, 2, 3, and 5 are written as adjacent node identifiers into the adjacency list. Alternatively, the system can establish a two-dimensional adjacency matrix, where the rows and columns correspond to the unique identifiers of the area nodes. When there is an adjacency edge between any two corresponding rows and columns, the value at the corresponding position in the matrix is ​​set to 1; otherwise, it is set to 0. Adjacency lists are suitable for warehouse layouts with a large number of nodes but sparse adjacency relationships, while adjacency matrices are suitable for application scenarios with a relatively small number of nodes but requiring quick determination of whether any pair of nodes is adjacent. The system can choose one of these implementation methods based on the actual warehouse size.

[0028] Through the discretization and adjacency relationship construction process described above, the originally continuous physical warehouse space is transformed into a picking area topology graph composed of region nodes and region adjacency edges. The picking area topology graph abstracts the locations accessible to picking personnel within the warehouse and the direct reachable relationships between these locations. Region nodes in the graph structure correspond to the stopping areas of picking operations, and region adjacency edges correspond to the area range traversed in a single continuous walking motion. In subsequent graph coloring operations, using region nodes as coloring objects and region adjacency edges as conflict constraints, time-period color labels can be directly assigned on the graph structure, thus avoiding complex geometric judgments based on the original coordinate data for each calculation and reducing computational complexity. Simultaneously, in the subsequent picking path planning stage, the system can perform path searches based on region nodes on the picking area topology graph, using the adjacency relationships of region nodes to determine the reasonable movement order of picking personnel or picking equipment. This ensures that the entire method maintains consistency with the physical layout while possessing high computational efficiency.

[0029] In one alternative implementation, the process of discretizing the physical space into multiple picking area units does not employ regular grid partitioning, but rather a shelf segmentation method. Specifically, the system divides each shelf along its length using a fixed number of storage locations, for example, every 5 consecutive storage locations constitute a picking area unit. The aisle segments adjacent to each picking area unit are considered the passable area of ​​that unit. Thus, each 20-meter-long shelf containing 40 storage locations can be divided into 8 picking area units. Subsequently, the system establishes area nodes based on the shelf segments, creating adjacent edges between adjacent shelf segments and intersecting aisle segments. Compared to regular grid partitioning, this method makes each picking area unit more closely resemble the range of storage locations that a picker can cover in one go, making it suitable for warehouses with relatively uniform storage location sizes and regular shelf arrangements.

[0030] In another alternative implementation, the storage facility layout and aisle network data are not entirely derived from design drawings, but are automatically constructed on-site by operating mobile devices through laser scanning or visual recognition. Specifically, autonomous mobile robots deployed in the warehouse continuously collect laser distance information or image information during their movement, and generate a grid map containing obstacle outlines and passable area outlines using environmental mapping algorithms. The system identifies areas considered obstacles in the grid map as shelves, walls, and equipment, and identifies continuous passable areas as aisles, determining the aisle centerline based on the width of the passable area and the relative position of the shelves. Subsequently, the system divides the grid map into picking area units and establishes adjacent edges according to the aforementioned rules, thereby constructing a picking area topology map. The advantage of this implementation is that when the warehouse is expanded or the shelves are rearranged, there is no need to manually update the design drawings; simply having the autonomous mobile robots retrace the warehouse generates a new picking area topology map, adapting to dynamically changing storage environments.

[0031] With the picking area topology map already constructed, the system first reads the target order data for the current wave of tasks from the warehouse management system. The target order data can be stored in structured record format, with each record containing at least an order identifier, the wave identifier to which the order belongs, multiple item identifiers, and the required quantity for each item identifier. For example, a wave of tasks may contain 500 independent orders, each order involving an average of 10 items, with each item requiring 1 to 5 units in the order. Based on the mapping relationship between items and storage locations, the system converts each item identifier into a corresponding storage location identifier, and then queries the area node identifier of the storage location based on the storage location identifier, thereby obtaining the set of area nodes associated with each order.

[0032] Based on this, the system begins to statistically analyze the demand relationships of adjacent edges in different areas to calculate the passage weight value. Specifically, the system creates a record for each adjacent edge in the picking area topology map, initializing the passage weight value field to 0. The passage weight value characterizes the frequency with which picking personnel need to cross areas via that adjacent edge under the current target order data. The system processes each independent order in the target order data one by one. For the current order, it first collects all the area nodes involved in the order, forming a subset of area nodes. For example, an order may involve area node 3, area node 7, and area node 9. Within this subset of area nodes, the system enumerates all pairs of area nodes directly connected by adjacent edges in the picking area topology map. For example, there is an adjacent edge between area node 3 and area node 7, and an adjacent edge between area node 7 and area node 9, but no direct adjacent edge between area node 3 and area node 9. For each pair of region nodes with adjacent edges, the system adds a preset single-passage cost to the passage weight value of the corresponding adjacent edge. This single-passage cost can be 1 or set to another fixed integer based on the warehouse size. For example, if the single-passage cost is set to 1, then in the above order, the passage weight value of the adjacent edge between region node 3 and region node 7 is increased by 1, and the passage weight value of the adjacent edge between region node 7 and region node 9 is also increased by 1. As the system completes traversal of all independent orders, some adjacent edges may have their passage weight values ​​accumulated multiple times. For example, if the passage weight value of a certain adjacent edge eventually reaches 120, it indicates that a large number of orders in the current wave of tasks simultaneously involve the region nodes at both ends of that adjacent edge. This means that picking personnel are highly likely to generate dense passage behavior along that adjacent edge.

[0033] The calculation process of the passage weight value directly reflects the strength of inter-regional linkage caused by the current target order data. If two regional nodes appear simultaneously in a large number of orders, the passage weight value of the corresponding adjacent edges will be higher. This indicates that if two regional nodes are performing picking operations simultaneously within the same wave of operation time window, the probability of picking personnel frequently traveling between these two regions will significantly increase, easily causing congestion at the corresponding passage locations. Therefore, this simple accumulation method can use an integer value to express the potential congestion risk level, facilitating subsequent threshold-based control.

[0034] After calculating the access weights of adjacent edges in each region, the system continues to analyze the demand relationships of each region node to calculate its busy weight. The busy weight characterizes the concentration of picking tasks at a given region node within the current target order data. Specifically, the system creates a record for each region node, initializing its busy weight to 0. The system then iterates through the target order data again. For each item in each order, it determines its corresponding region node based on the mapping relationship between the item and its storage location, and accumulates the busy weight of that region node according to the quantity of the item required. For example, if an order requires 3 of a certain item, and the region node for that item is region node 5, then the busy weight of region node 5 increases by 3. By iterating through all orders, the system can calculate the total quantity of items required by each region node in the current wave of tasks. For example, the busy weight of region node 5 might be 800, while the busy weight of region node 12 might only be 60. This indicates that region node 5 will handle a large number of picking operations, while region node 12 will be relatively less busy.

[0035] To simultaneously reflect both the demand density of an area and its interconnectivity with neighboring areas, the system can also add the final passage weight value of the adjacent edges directly connected to a given area node to the busy weight value of that area node. Specifically, for a given area node, the system iterates through all adjacent edges connected to that area node, multiplies the final passage weight value of each adjacent edge by a preset influence factor, and then adds it to the busy weight value of that area node. For example, when the preset influence factor is 0.5, the passage weight value of each adjacent edge can be divided by 2 before being added to the busy weight value. The reason for this approach is that even if an area node itself does not have the highest demand for items, if it is connected to multiple adjacent edges with high passage weight values, it means that the area is frequently used as a transit point, resulting in more stops and passes by picking personnel in the vicinity of the area, and thus a higher risk of congestion. By incorporating the passage weight value of adjacent edges into the busy weight value, the busy weight value of a region node can comprehensively reflect "the number of times it stops picking in this region" and "the number of times it passes through this region as a path", thereby more accurately identifying the region nodes that truly need to be prioritized in the entire picking region topology.

[0036] After calculating the passage weight and busy weight, the system performs graph coloring operations on the regional nodes under a preset congestion threshold constraint. To this end, the system first establishes a color tag library containing multiple distinct time-segment color labels. Each time-segment color label in the library corresponds one-to-one with a wave of operation time windows in the warehouse. For example, six wave of operation time windows can be pre-set: 8:00-9:00, 9:00-10:00, 10:00-11:00, 13:00-14:00, 14:00-15:00, and 15:00-16:00, corresponding to six time-segment color labels, denoted as color label 1 to color label 6. The number of time-segment color labels in the color tag library is set to be greater than the maximum adjacency degree of any regional node, that is, greater than the number of adjacent regional nodes of any regional node in the picking area's topology graph. This provides sufficient selection space during the subsequent coloring process, avoiding the forced placement of adjacent high-load areas within the same wave of operation time window due to insufficient color options.

[0037] The system sorts all regional nodes in descending order based on their final busy weight values, forming a node processing priority sequence. Regional nodes with higher busy weight values ​​are ranked higher, meaning they have a greater impact on overall picking efficiency and congestion risk, and should be given priority in obtaining the best time-period color label selection opportunity in the graph coloring operation. Subsequently, the system selects regional nodes as the current processing nodes according to the node processing priority sequence. For each current processing node, it retrieves its neighbor node set from the picking region topology graph and filters out neighbor nodes that have already been assigned time-period color labels, forming a colored neighbor set. For each neighbor node in the colored neighbor set, the system finds the final passage weight value of the regional adjacency edge between the current processing node and that neighbor node, and compares this passage weight value with a preset first congestion judgment threshold. For example, the first congestion threshold can be set to 20. When the final passage weight value of an adjacent edge in a certain area is greater than or equal to 20, it indicates that this adjacent edge is very likely to become a high-frequency passage channel in the current wave of tasks. If the current processing node and this neighboring node are assigned the same time period color label, then within the corresponding wave of operation time window, the picking personnel will move back and forth between these two areas many times, easily causing local channel congestion. Therefore, when the passage weight value of the adjacent edge corresponding to a colored neighboring node is greater than or equal to the first congestion threshold, the system marks this neighboring node as a strong conflicting neighbor of the current processing node and adds the time period color label already obtained by this neighboring node to the current processing node's prohibited label list.

[0038] refer to Figure 1This diagram visually illustrates the load status calculated and mapped by the system on the picking area topology after acquiring the target order data for the wave of tasks to be processed. The diagram contains multiple circular area nodes and line-like adjacent edges connecting these nodes. Specifically, each circular node in the diagram represents a discretized picking area unit. The size (i.e., diameter) of the circular node corresponds to the busy weight value of that area node. As shown in the diagram, some nodes are significantly larger than others (e.g., some large circles in the lower left or central area of ​​the diagram), indicating that in the current wave of tasks, the items stored in these area nodes are in high demand by a large number of orders, or that the area node is carrying a high item picking density. According to the method of the present invention, the system calculates the busy weight value by statistically analyzing the demand quantity of each item in the target order data and combining it with the access weight influence factor of adjacent edges. Larger nodes mean that the area will become a "hotspot area" where picking personnel or equipment frequently linger. The lines connecting the nodes in the diagram represent adjacent edges, i.e., physically traversable path relationships. The thickness of the lines (i.e., line width) corresponds to the passage weight value of adjacent edges in a region. As shown in the figure, the lines connecting certain large nodes are very thick (e.g., the thick black lines in the figure), while the lines in the edge areas are relatively thin. This difference in thickness intuitively reflects the statistical results of the system on the target order data: a thick line indicates that in a large number of orders, the nodes at both ends of this edge appear simultaneously, meaning that picking personnel need to frequently travel back and forth between the two areas via this path. Figure 3 The weight distribution mapping shown in this invention allows for the intuitive identification of potential congestion risk points. Local areas that are both large-sized nodes (high busy weight) and connected by thick lines (high throughput weight) are the areas most prone to channel congestion. These quantified weight values ​​are then input into the graph coloring module as constraints (e.g., a first congestion threshold) to ensure that two high-load area nodes connected by thick lines are assigned different time-period color labels, thereby staggering operations in the time dimension and achieving congestion avoidance.

[0039] After constructing the forbidden label list, the system removes all time-period color labels from the forbidden label list in the color label library. The remaining time-period color labels constitute the candidate label subset for the current processing node. If the candidate label subset is not empty, the system counts how many regional nodes each time-period color label has been assigned to in the candidate label subset, and prioritizes assigning the time-period color label with the least usage to the current processing node. This selection method is based on the consideration of avoiding using the same time-period color label with strongly conflicting neighbors as much as possible, and also ensuring that the usage of each time-period color label is as balanced as possible in the entire picking area topology, thereby avoiding a situation where too many regional nodes are gathered in one wave of operation time window, while there are almost no tasks in another wave of operation time window. If the candidate label subset is empty, it means that the time-period color labels of all the colored neighbor nodes of the current processing node are related to the regional adjacency edges with high passage weight values, and it is impossible to find an existing time-period color label that avoids all strongly conflicting neighbors. In this scenario, the system can select a time period color label from the color label library that has not yet been used by any regional node, assign it to the current processing node, and add the time period color label to the available set of the color label library. This allows the entire system to extend a new wave of operation time window in the color dimension, thereby providing a more distributed scheduling time space for congested areas.

[0040] After coloring all regional nodes according to their processing priority sequence, the system obtains an initial time-segment color label allocation result for the regional nodes. In some complex layouts and order distribution scenarios, even following the strong conflict neighbor avoidance strategy described above, it's still possible that some adjacent regional nodes on the same boundary may be assigned the same time-segment color label, and the final passage weight of this boundary may be very high. This means that on a certain channel line, two highly correlated regional nodes in the order may still simultaneously execute picking tasks within the same wave operation time window, causing significant congestion in actual operation. To further reduce this risk of local congestion, the system can perform local color optimization based on the initial time-segment color label allocation result.

[0041] During local color optimization, the system traverses all adjacent edges in the picking area topology and selects those edges that simultaneously satisfy a final passage weight value greater than or equal to a preset second congestion threshold and have the same time-period color label at both ends as the set of edges to be optimized. For example, the second congestion threshold can be set to 40, higher than the first congestion threshold, specifically used to identify adjacent edges in areas most likely to experience severe congestion. For each adjacent edge in the set of edges to be optimized, the system compares the final busy weight values ​​of the adjacent nodes at both ends and selects the node with the smaller busy weight value as the node to be adjusted. The reason for selecting the node with the smaller busy weight value for time-period color label adjustment is that the picking workload of this area is relatively low, and scheduling it to different wave operation time windows has a smaller impact on overall efficiency. At the same time, it is also easier to find suitable resource arrangements for it in other wave operation time windows.

[0042] The system obtains the time-period color labels of all strongly conflicting neighbors of the node to be adjusted, forming a local forbidden set. The definition of a strongly conflicting neighbor remains consistent with the initial graph coloring, i.e., a colored neighbor node whose regional adjacency edge passage weight with the node to be adjusted is greater than or equal to the first congestion judgment threshold. Subsequently, the system removes all time-period color labels from the local forbidden set in the color label library, obtaining a feasible adjustment set. For each potential time-period color label in the feasible adjustment set, the system assumes that the time-period color label of the node to be adjusted will be changed to that potential time-period color label, and then counts the number of regional adjacency edges with the node to be adjusted as the endpoint and the same time-period color labels at both ends in the selection area topology graph. Specifically, the system traverses all regional adjacency edges directly connected to the node to be adjusted, and checks whether the time-period color label of the regional node at the other end of each regional adjacency edge is the same as the current potential time-period color label. If they are the same, the count value is incremented by 1. By performing this simulated statistical analysis on all potential time-period color labels, the system obtains a correspondence between a set of candidate labels and the number of adjacent edges with the same label. The system selects the potential time-period color label that minimizes the number of adjacent edges with the same label, and uses this as the new time-period color label for the node to be adjusted, updating the time-period color label of the node to be adjusted to this label. Through this local adjustment, the system can minimize the use of the same time-period color label at both ends of adjacent edges in high-passage-weight areas without changing the overall coloring structure, thereby further reducing the probability of local channel congestion.

[0043] refer to Figure 3 , Figure 3The graph contains 12 region nodes, labeled Node 1 to Node 12, which are regularly distributed on a two-dimensional plane. Specifically, Nodes 1, 2, 3, and 4 are located in the first row, arranged from left to right; Nodes 5, 6, 7, and 8 are located in the second row, arranged from left to right; and Nodes 9, 10, 11, and 12 are located in the third row, arranged from left to right. Adjacent region nodes are connected by region adjacency edges, forming the basic structure of the picking region topology graph. It can be observed in the graph that adjacent region nodes in the same row have horizontal region adjacency edges, such as between Node 1 and Node 2, Node 2 and Node 3, and Node 3 and Node 4; adjacent region nodes in the same column have vertical region adjacency edges, such as between Node 1 and Node 5, and Node 5 and Node 9. After the graph coloring operation, each region node is assigned a time period color label. In this embodiment, three different time period color labels are set, each corresponding to one of the three wave operation time windows: the first time period color label corresponds to the wave operation time window from 8:00 to 9:00, the second time period color label corresponds to the wave operation time window from 9:00 to 10:00, and the third time period color label corresponds to the wave operation time window from 10:00 to 11:00. To facilitate differentiation of the different time period color labels in the black and white diagram, Figure 3 Different fill patterns are used to represent the time periods: the first time period color label uses horizontal lines, the second uses vertical lines, and the third uses diagonal lines. The specific allocation results are as follows: Nodes 1, 3, and 6 are assigned the first time period color label, with their circular nodes filled with horizontal lines; Nodes 2, 5, 8, and 10 are assigned the second time period color label, with their circular nodes filled with vertical lines; Nodes 4, 7, 9, and 12 are assigned the third time period color label, with their circular nodes filled with diagonal lines; Node 11 is assigned the first time period color label, with its circular node filled with horizontal lines. Through observation... Figure 3It can be observed that adjacent area nodes are assigned different time-period color labels as much as possible. For example, node 1 is assigned the first time-period color label, its right-adjacent node 2 is assigned the second time-period color label, node 3, the right-adjacent node of node 2, is also assigned the first time-period color label, and node 4, the right-adjacent node of node 3, is assigned the third time-period color label. Vertically, node 1 is assigned the first time-period color label, its lower-adjacent node 5 is assigned the second time-period color label, and its lower-adjacent node 9 is assigned the third time-period color label. This assignment method ensures that area nodes directly connected by adjacent edges are in different wave operation time windows in most cases, thereby avoiding the need for picking personnel to frequently move back and forth between adjacent areas within the same time period, effectively reducing the risk of aisle congestion. Figure 3 In the upper left corner, there is a legend area for time period color labels. This area sequentially displays the representation methods of the three time period color labels and their corresponding wave operation time windows. The first legend item shows a circular marker filled with a horizontal line pattern, labeled "Time Period 1 (8:00-9:00)"; the second legend item shows a circular marker filled with a vertical line pattern, labeled "Time Period 2 (9:00-10:00)"; and the third legend item shows a circular marker filled with a slanted line pattern, labeled "Time Period 3 (10:00-11:00)". This legend clearly establishes the correspondence between the filled patterns of each area node in the diagram and its corresponding wave operation time window.

[0044] In the aforementioned local color optimization, the system can process the edges to be optimized sequentially in descending order of their final passage weight values. The purpose of this is to prioritize addressing the adjacent edges of areas most prone to severe congestion, as these areas have the largest concentration of picking workers and the most significant path conflicts. By prioritizing the adjustment of time-period color labels for nodes near these high-risk edges, a relatively significant improvement in congestion can be achieved within a limited number of adjustments. After all edges to be optimized have been processed, the system obtains the final time-period color label allocation table for the regional nodes.

[0045] After the regional node time-slot color label allocation table is generated, each regional node is associated with a specific wave operation time window. Based on the mapping relationship between time-slot color labels and wave operation time windows, the system can determine which regional nodes are active within any given time period, and thus schedule picking operations during that time period. Next, the system further breaks down and analyzes the target order data based on the time-slot color labels, dynamically reconstructing and allocating high-frequency demand items to priority storage locations within the active regional nodes corresponding to the wave operation time window, forming a dynamic storage location reconstruction scheme.

[0046] Specifically, for each area node, the system first reads the spatial coordinates of all storage locations within the physical range of that area node, as well as the location coordinates of the corresponding passage entrance. The system can then calculate the walking distance from each storage location to the passage entrance based on the direction of the passage network. For example, when the passage is basically linear, the walking distance can be approximated as the distance from the center point of the storage location to the passage entrance along the passage direction. When the passage has a broken line, discrete path points on the passage centerline can be pre-calculated, the center point of the storage location can be projected onto the nearest path point, and the path length can be accumulated along the path points to obtain the walking distance. In this way, the system obtains a walking distance value for each storage location; for example, the walking distance for a storage location near the passage entrance is 3 meters, the walking distance for a storage location in a normal location is 12 meters, and the walking distance for the farthest storage location is 25 meters.

[0047] The system sorts all storage locations within the area node according to their walking distance from shortest to longest, generating a priority storage location index sequence list. The first few storage locations in the priority storage location index sequence list, i.e., those with relatively short walking distances, are considered priority storage locations. These priority storage locations can significantly reduce the walking distance between shelves for picking personnel after entering the area node in actual operations, and are therefore suitable for storing items with high demand frequency within the current wave of operations. For example, if there are 100 storage locations in a certain area node, the system defines the top 20 storage locations with the shortest walking distance as priority storage locations, which are used to prioritize the allocation of high-frequency demand items in dynamic storage location reconfiguration schemes.

[0048] Next, the system divides the item requirements in the target order data according to the regional node time period color label allocation table and the mapping relationship between time period color labels and wave operation time windows. Specifically, for each combination of regional node and each wave operation time window, the system filters all items that need to be picked within that wave operation time window and stored within that regional node from the target order data, forming a regional wave demand group. Each record in the regional wave demand group can contain the item identifier and the total quantity that needs to be picked within that regional node and that wave operation time window. For example, under the combination of regional node 5 and the wave operation time window from 8:00 to 9:00, the system calculates that the demand quantity for item A is 30 pieces, the demand quantity for item B is 25 pieces, the demand quantity for item C is 5 pieces, and the demand quantity for item D is 2 pieces.

[0049] For each regional wave demand group, the system sorts items from highest to lowest demand quantity within that group, resulting in a list of high-frequency items to be allocated. Items ranked higher in the high-frequency item list are picked more frequently within the corresponding wave operation time window for that regional node. Placing these items in priority storage locations with shorter walking distances minimizes the average walking distance for picking personnel. For example, in the aforementioned regional wave demand group, if item A has a demand of 30 units, item B 25 units, item C 5 units, and item D 2 units, the high-frequency item list to be allocated would be in the order of A, B, C, and D. The system then maps these items sequentially to storage locations in the priority storage location index sequence table. For instance, item A is allocated to the first few priority storage locations with the shortest walking distances, item B to the next batch of priority storage locations, and items C and D to the remaining priority storage locations or storage locations with relatively moderate walking distances.

[0050] When the total number of items in a regional wave demand group is no greater than the total number of storage locations within that regional node, the system can allocate all items to the highest-ranked storage locations, reserving the remaining unoccupied locations as backups for dynamic adjustments in subsequent wave tasks. When the total number of items in a regional wave demand group exceeds the total number of storage locations, meaning the prime storage location index sequence cannot accommodate all items, the system prioritizes allocating the items with the highest demand to priority storage locations, while retaining the items with lower demand in their original storage locations. In this case, recording the mapping changes between items and storage locations is crucial. The system maintains a mapping change table, recording the item identifier, the original storage location identifier, and the new storage location identifier, for picking personnel to query during actual operations.

[0051] Through the aforementioned dynamic location reconfiguration process, frequently requested items within each wave of operations can be concentrated in the priority storage locations of active area nodes under the corresponding time-period color labels. This allows for adaptive adjustment of the micro-location layout based on changes in wave tasks, without altering the overall warehouse layout. On one hand, graph coloring operations driven by passage and busyness weights distribute highly correlated area nodes across different wave of operations, reducing large-scale personnel movement between adjacent areas within the same timeframe. On the other hand, the dynamic location reconfiguration scheme places the most frequently picked items as close as possible to the aisle entrance in each time period, shortening the single picking path. The combination of these two aspects significantly alleviates aisle congestion and noticeably improves picking efficiency in actual operation.

[0052] In one optional implementation, the determination of high-frequency demand items can be based not only on the quantity of demand within a regional wave of demand groups, but also on a weighted ranking according to the importance level of the items. For example, for critical materials with high importance, they can be prioritized and placed higher in the list of high-frequency items to be allocated, provided that the demand quantity is the same. This ensures that critical materials are allocated to priority storage locations closer to the aisle entrance within the corresponding wave of operation time window, thereby shortening the replenishment and picking time for critical materials. In another optional implementation, the number of priority storage locations can be dynamically adjusted based on the busy weight value of regional nodes. Regional nodes with higher busy weight values ​​can reserve more priority storage locations to accommodate the reconfiguration needs of more high-frequency demand items, while regional nodes with lower busy weight values ​​can retain only a few priority storage locations to reduce unnecessary location adjustments. Through these optional implementation methods, details can be flexibly adjusted within a unified basic framework to adapt to warehousing environments of different sizes and with different business characteristics.

[0053] With the aforementioned regional node time-slot color label allocation table and dynamic storage location reconstruction scheme already generated, the system first reads the target order data for the wave of tasks to be processed. In the target order data, each parent order contains a unique order identifier, the customer identifier to which the order belongs, and several item records. Each item record includes an item identifier and the required quantity. At this point, the dynamic storage location reconstruction scheme has already provided the storage location identifier corresponding to each item identifier in the current wave of tasks. Each storage location identifier is then associated with a unique regional node identifier, and a one-to-one correspondence is established between the regional node identifier and the time-slot color label through the aforementioned graph coloring operation. Thus, for any item in the target order data, the system can determine three basic pieces of information: the regional node to which it belongs, the corresponding wave of operation time window, and the current storage location.

[0054] Based on this information, the system begins to break down and reorganize the target order data into regional task sets according to regional nodes and wave operation time windows. Specifically, the system reads the parent order one by one, treating each item in the parent order as a divisible unit. For a given item, the system finds its storage location identifier according to the dynamic storage location reconstruction scheme, and then finds its corresponding regional node identifier from the storage location identifier, for example, an item belongs to regional node 5. Subsequently, the system looks up the time period color label of regional node 5 from the regional node time period color label allocation table, for example, it is the second color label, and the wave operation time window corresponding to the second color label is 9:00 to 10:00. Thus, the system assigns the item to the child order fragment under the key combination "regional node 5 + 9:00 to 10:00". If other items in the same parent order belong to area node 5 and should also be picked during the 9:00 to 10:00 wave operation time window, these items will be added to the same child order fragment. If some items belong to area node 7 and the corresponding time period color label is the third color label, i.e., the 10:00 to 11:00 wave operation time window, the system will create a separate child order fragment for the combination key "area node 7 + 10:00 to 11:00" and add these items to it.

[0055] Through the above breakdown, the parent order, originally organized by customer dimension, is broken down into multiple child order fragments organized by "regional nodes + wave operation time windows". The reason for this breakdown is that parent orders often span multiple regional nodes and multiple channels. Directly generating picking paths based on parent orders would result in the path repeatedly jumping around the entire warehouse, resulting in long travel distances. By breaking down the parent order into multiple child order fragments, each child order fragment contains only items from the same regional node that need to be picked within the same wave operation time window. This confines the picking path within a single regional node, forming a relatively compact local travel route, which is more conducive to controlling the execution time of a single task and also facilitates subsequent parallel task allocation based on regional nodes.

[0056] After all parent orders have been broken down, the system uses the region node and wave operation time window as the aggregation key to merge all child order fragments into multiple independent region task sets. For each combination of "region node + wave operation time window," the system summarizes the items involved and their required quantities in all its child order fragments, forming an independent region task set. For example, under region node 5 and the wave operation time window from 9:00 to 10:00, the system may summarize a region task set containing 80 different items, totaling 400 pieces. This organization of independent region task sets means that when generating closed-loop picking routes for each region task set, only the location layout and aisle conditions within a single region node need to be considered, greatly simplifying the complexity of the route planning problem.

[0057] After the independent regional task sets are generated, the system generates closed-loop picking paths for each regional task set based on a dynamic location reconfiguration scheme. For a specific regional task set, the system first obtains the physical coordinates of the storage locations of all items to be picked within that regional node according to the dynamic location reconfiguration scheme, and also obtains the physical coordinates of the corresponding aisle entrance for that regional node. The aisle entrance is set as the start and end point of the picking path for that regional task set because picking personnel or resources typically enter the regional node from the aisle entrance, complete the picking, and then leave from that location or move to other regional nodes. Starting from and returning to the aisle entrance allows the task to form a complete closed loop in both time and space, facilitating task management and scheduling.

[0058] Based on this, the system uses a nearest neighbor algorithm to generate an initial closed-loop picking path. Specifically, taking the aisle entrance as the current path position, the estimated travel distance between each unvisited physical coordinate point and the current path position is calculated sequentially. The estimated travel distance can be obtained in two ways: in scenarios where warehouse aisles are relatively regular, a simple Manhattan distance can be used, which is the sum of the distances along the first and second axes; in scenarios where warehouse aisles have multiple turns or bends, the aisle centerline can be pre-extracted as a broken line, and the path length from the current path position along the aisle centerline to the projected position of that physical coordinate point can be calculated as the estimated travel distance. The system selects the unvisited physical coordinate point with the smallest estimated travel distance, adds it to the picking path, updates the current path position to that physical coordinate point, and then repeats the above process for the remaining unvisited physical coordinate points until all unvisited physical coordinate points have been visited once. Finally, the system connects the path endpoint to the aisle entrance with a path segment, forming a closed-loop picking path that starts from the aisle entrance, passes through all storage locations sequentially, and returns to the aisle entrance.

[0059] Paths generated using only the nearest neighbor algorithm are typically much shorter than random paths, but they can still exhibit a noticeable "backtracking" phenomenon. This means that the order in which certain physical coordinates are visited is unreasonable, causing picking personnel to walk a long distance and then turn back to a position near the starting point. To further shorten the picking path length, the system performs path topology swapping optimization based on the initial closed-loop picking path. Specifically, the system selects two non-adjacent path segments in the closed-loop picking path. For example, if four consecutive physical coordinates in the path are A, B, C, and D, the original connection relationship is from A to B, then from B to C, and then from C to D. The system attempts to adjust the connection order to from A to C, then from C to B, and then from B to D, that is, swapping the topology relationship of the two middle connection segments, and recalculates the total path length before and after the adjustment. If the adjusted total path length is less than the original total path length, it means that this topology swap reduced unnecessary detours. The system retains this swap result and uses the updated path as the basis for the next topology swap attempt. If the total path length is not improved, the topology swap is canceled, and the original connection relationship is restored. The system performs the above attempts sequentially according to all possible combinations of non-adjacent path segments until no more topology swap combinations that can shorten the total path length are found in a complete traversal, or until the preset iteration limit is reached. The closed-loop picking path obtained at this point can be regarded as the optimized closed-loop picking path for the task set in that region.

[0060] The purpose of path topology swapping optimization is to correct the local greedy selection problem caused by the nearest neighbor construction algorithm. The nearest neighbor construction algorithm only considers the next closest point at each step, failing to predict the overall path direction and easily leading to local detours. By selecting two non-adjacent connecting segments in the path and attempting to swap their access order, it's equivalent to straightening or rearranging a small "broken line" on the path. This breaks the unreasonable structure caused by local greedy selection, making the picking path smoother, reducing the number of turns, and significantly shortening the overall walking distance of picking personnel within the area nodes.

[0061] After generating the optimized closed-loop picking path for each independent regional task set, the system calculates the corresponding travel cost based on the physical length of each path segment. Specifically, the system breaks down the optimized closed-loop picking path into several continuous path segments, each connecting two adjacent physical coordinate points. For example, one path segment runs from the aisle entrance to the first storage location, and another runs from the first storage location to the second. The system calculates the physical length of each path segment based on aisle network data and coordinate data, which can be measured using the actual aisle centerline length. Subsequently, the system sums the physical lengths of all path segments, and the total length is taken as the total travel cost for the independent regional task set. The total travel cost describes the total distance that picking resources need to travel within the regional nodes when executing the independent regional task set. For example, if the optimized closed-loop picking path for a certain regional task set consists of 50 path segments with a total physical length of 600 meters, then the total travel cost for that regional task set can be recorded as 600. The higher the pass value, the longer the execution time of the task set in that area, and the higher the degree of occupation of picking operation resources.

[0062] After calculating the total throughput value of all regional task sets, the system begins allocating these task sets to picking operation resources according to a load balancing strategy. The goal of this load balancing strategy is to ensure that the total throughput value borne by each picking operation resource is as similar as possible throughout the entire wave of tasks. This avoids situations where some picking operation resources are over-scheduled and unable to complete within the wave's time window, while other picking operation resources remain idle. To achieve this, the system first establishes a queue of tasks to be allocated and writes all independent regional task sets into this queue in descending order of total throughput value. This is done because prioritizing tasks with higher total throughput values ​​allows for the even distribution of "heavyweight" tasks across multiple picking operation resources from the initial allocation stage, reducing the difficulty of compensatory scheduling later.

[0063] Simultaneously, the system establishes a resource status list to record the current load status of each picking operation resource. Each record in the resource status list corresponds to a picking operation resource, including its unique identifier, the cumulative passage cost of currently assigned tasks, and the available wave operation time window information. Initially, the cumulative passage cost of all picking operation resources is 0. The resource status list is sorted in ascending order of cumulative passage cost, ensuring that the picking operation resources at the beginning of the list represent the resources with the lightest current load.

[0064] After the aforementioned data structure is prepared, the system employs a greedy allocation strategy for task distribution. Specifically, the system retrieves the set of independent area tasks with the largest total access cost from the head of the task queue, and reads the corresponding wave operation time window and the area node to which the task set belongs. Subsequently, the system selects the picking operation resource with the smallest current cumulative access cost and which is available within the current wave operation time window from the resource status list, assigns the independent area task set to that picking operation resource, and adds the cumulative access cost of that picking operation resource to the total access cost of the task set. For example, if the total access cost of an independent area task set is 600, and the system assigns it to a picking operation resource with a current cumulative access cost of 1200, then the cumulative access cost of that picking operation resource is updated to 1800. After the update, the system re-sorts the resource status list according to the cumulative access cost in ascending order, ensuring that the new sorting reflects the latest load distribution.

[0065] The greedy water-filling allocation strategy behaves similarly to pouring water blocks of decreasing volume into multiple containers sequentially, always adding the largest block to the currently emptyest container. This results in all containers eventually having similar water levels, achieving load balancing. In this method, the total throughput value can be understood as the "volume" of the task, and the picking operation resources as "containers." By prioritizing the allocation of independent area tasks with higher throughput values ​​to picking operation resources with lower current cumulative throughput values, a near-ideal load balancing effect can be achieved through simple greedy computation, without needing to solve complex global optimization planning problems. This approach is suitable for wave picking scenarios with high real-time requirements.

[0066] As the queue of tasks to be assigned is gradually cleared, all independent area task sets are allocated to specific picking operation resources. Based on this, the system generates the final task allocation instruction set. For each picking operation resource, the system organizes the multiple independent area task sets allocated to that resource according to wave operation time windows and execution order. The system writes the area node identifier, wave operation time window, critical path point sequence on the optimized closed-loop picking path, and estimated execution time for each independent area task set into a task instruction list. Subsequently, the system sends the final task allocation instruction set to the handheld terminal or vehicle controller of each picking operation resource via a wireless communication network. The handheld terminal can display the start and end times, target area nodes, and critical path points of each independent area task set in the form of a task list, guiding human pickers to sequentially enter the corresponding area nodes within the corresponding wave operation time window and perform picking operations along the optimized closed-loop picking path. The vehicle controller can generate control instructions based on the channel entrance and physical coordinate point sequence given in the optimized closed-loop picking path to drive the vehicle automatically along the predetermined path and stop at the designated storage location to complete the retrieval.

[0067] refer to Figure 2 , Figure 2 This diagram illustrates the process of generating picking paths within the same regional node for a single set of tasks in a given region. The diagram includes a triangle marker representing the entrance to the regional node (i.e., the start and end point of the path); the scattered squares represent the physical coordinates of the storage locations where the items to be picked are located within that regional task set. (Refer to...) Figure 2 The section corresponding to (a) shows the initial closed-loop picking path generated based on the nearest neighbor construction algorithm. In this path, the system starts from the aisle entrance and greedily selects the nearest unvisited storage location as the next target. Although this method is computationally simple, as shown in the figure, the generated path trajectory appears rather chaotic, with obvious path intersections and backtracking. For example, the picking path may first traverse the entire area to reach a distant point, and then backtrack to visit the missed point in the middle, resulting in the overall walking route exhibiting a zigzag or complex winding shape, increasing unnecessary passage costs. (Refer to...) Figure 2 The part corresponding to (b) in the text shows that... Figure 2 Based on the part corresponding to (a) in the present invention, the optimized closed-loop picking path is optimized through path topology exchange. The system eliminates the problem by iteratively selecting non-adjacent connecting segments in the path for topology order exchange (e.g., the 2-opt algorithm). Figure 2 The path intersections that exist in the part corresponding to (a) in the example. Figure 2As shown in section (b), the optimized path forms a clear, smooth convex hull or near-circular closed-loop structure. Picking resources start from the aisle entrance, sequentially visit the storage locations along their distribution outline, and finally return directly to the aisle entrance, without any obvious detours or repetitive routes. Through comparison... Figure 2 The part corresponding to (a) in the middle and Figure 2 As can be seen from part (b) in the diagram, the total physical length of the optimized closed-loop picking path is significantly shorter than that of the initial path. This process corresponds to the step of calculating and optimizing the "passage cost" in this invention. Figure 2 The path scheme shown in part (b) will be transformed into a sequence of critical path points in the final task allocation instruction set, guiding picking resources to complete the picking of all items within the area node with the shortest travel distance, thereby effectively improving the execution efficiency of a single operation.

[0068] In one optional implementation, the system can also consider the type and capability differences of picking resources when executing the greedy water-filling allocation strategy. For example, in a scenario combining human pickers and automated guided vehicles (AGVs), the system can set different maximum acceptable total passage value thresholds for different types of picking resources, such as setting the maximum total passage value for human pickers to 5000 and the maximum total passage value for AGVs to 8000. When allocating tasks, if the cumulative passage value of a picking resource exceeds its maximum acceptable total passage value after a set of tasks for a specific area has been assigned to it, no new tasks will be assigned to that picking resource temporarily. In this way, the mismatch between actual execution capabilities due to different types of picking resources can be avoided.

[0069] In another optional implementation, when generating closed-loop picking paths for independent regional task sets, the system can select different path generation strategies based on the number of items in the regional task sets. When the number of items to be picked in a certain independent regional task set is small, for example, no more than 10, the system can directly calculate all possible access sequences through exhaustive search and select the access sequence with the lowest total passage cost as the closed-loop picking path. When the number of items to be picked in a certain independent regional task set is large, for example, more than 50, the system can add an insertion-based improvement step to the nearest neighbor construction algorithm. That is, after generating the initial path, it attempts to insert some distant physical coordinate points into more suitable positions in the path, and then performs path topology exchange optimization to balance computational speed and path quality. Through these optional implementations, this method can achieve better adaptability in warehousing environments of different sizes and resource configurations.

[0070] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for dynamic slot reconfiguration and task assignment for wave-oriented picking, characterized in that, include: Obtain the storage facility layout and channel network data of the warehouse environment, discretize the physical space into multiple picking area units and map them as area nodes, establish area adjacency edges based on physical connectivity, and construct a picking area topology map; Read the target order data of the pending wave of tasks, and statistically analyze the demand association of each regional node and regional adjacent edge according to the mapping relationship between items and storage locations, so as to obtain the passage weight value of the regional adjacent edge and the busy weight value of the regional node. Under the constraint of a preset congestion threshold, graph coloring operation is performed on the regional nodes based on the passage weight value and busy weight value, and each regional node is assigned a different time period color label corresponding to the wave operation time window. Among them, the local area that has both a regional node with a high busy weight and is composed of regional adjacent edges connected by a high passage weight is the high-load area most prone to congestion. The preset congestion threshold constraint ensures that two regional nodes connected by regional adjacent edges in the high-load area are assigned different time period color labels. Based on the time period color labels, the high-frequency demand items in the target order data are dynamically reconstructed and allocated to the priority storage locations in the active area nodes within the wave operation time window, forming a dynamic storage location reconstruction scheme. Based on the time period color label, the target order data is broken down and reorganized into regional task sets according to regional nodes and wave operation time windows. Based on the dynamic location reconstruction scheme, the closed-loop picking path of each regional task set is generated and optimized, the corresponding passage value is calculated, and the regional task sets are allocated to picking operation resources according to the load balancing strategy to drive the picking operation resources to complete the picking operation within the corresponding wave operation time window. The calculation of the passage weight value of the adjacent edge of the region specifically includes: initializing the passage weight value of each adjacent edge of the region to zero; parsing the independent orders in the target order data one by one, determining the subset of regional nodes involved based on the storage location of the goods, and for the pairs of regional nodes directly connected by the adjacent edges of the region in the subset, incrementally accumulating the passage weight value of the corresponding adjacent edge of the region according to the preset single passage cost, until all independent orders are processed, and obtaining the final passage weight value of each adjacent edge of the region; The calculation of the busy weight value of the regional node specifically includes: initializing the busy weight value of each regional node to zero; counting the total frequency of item demand falling within the range of each regional node in the target order data and accumulating it to the corresponding node; then weighting and accumulating the final passage weight value of the adjacent edges of the region directly connected to the node according to a preset influence factor, and using the resulting value as the final busy weight value of the regional node, which is used to determine the node processing priority in the graph coloring operation.

2. The wave-oriented sorting dynamic bin reconfiguration and task assignment method of claim 1, wherein, Graph coloring operations specifically include: Establish a color tag library containing multiple distinct time period color tags, the number of which is greater than the maximum adjacency degree of any region node; All regional nodes are sorted in descending order according to their final busy weight values ​​to form a node processing priority sequence. When selecting each region node as the current processing node in sequence, the nodes that have completed color assignment among its neighbor nodes are retrieved to form a set of colored neighbors. Each colored neighbor node in the set of colored neighbors is judged. If the final passage weight value of the regional adjacency edge between the current processing node and the colored neighbor node is greater than or equal to the first congestion judgment threshold, the colored neighbor node is marked as a strong conflict neighbor and the time period color label of the colored neighbor node is added to the list of prohibited labels. Remove the labels from the forbidden label list from the color label library to obtain a subset of candidate labels. If the subset of candidate labels is not empty, select the time period color label that has been used the least in the picking area topology map and assign it to the current processing node. If the subset of candidate labels is empty, add an unused time period color label, assign it to the current processing node, and add it to the color label library.

3. The dynamic location reconstruction and task allocation method for wave picking as described in claim 2, characterized in that, After assigning time-period color labels to each region node according to the method described in claim 2, a local color optimization step is further included, specifically including: Traverse all adjacent edges of regions and select the adjacent edges of regions that simultaneously satisfy the condition that the final passage weight value is greater than or equal to the second congestion judgment threshold and that the nodes of the two regions at both ends have the same time period color label as the edge to be optimized. For each edge to be optimized, compare the final busy weight values ​​of the nodes at both ends, select the smaller one as the node to be adjusted, and obtain the time period color labels of all the strongly conflicting neighbors of the node to be adjusted to form a local forbidden set. The local forbidden set is removed from the color label library to obtain the feasible set for adjustment. Each time period color label in the feasible set is taken as a potential time period color label. The number of regional adjacent edges with the node to be adjusted as the endpoint and the same time period color label at both ends is evaluated after changing the time period color label of the node to be adjusted to the potential time period color label. The potential time period color label that minimizes this number is selected as the new time period color label to update the time period color label of the node to be adjusted. The above adjustment is completed in descending order of the final passage weight value of the edge to be optimized to obtain the final regional node time period color label allocation table.

4. The dynamic location reconstruction and task allocation method for wave picking as described in claim 3, characterized in that, The dynamic storage location reconfiguration solution includes: For each region node, calculate the walking distance of each storage location within its physical range relative to the entrance of the passage, and generate a high-quality storage location index sequence table by sorting the walking distances from smallest to largest. Based on the color label allocation table for regional nodes and the mapping relationship between time period color labels and wave operation time windows, the item demand in the target order data is divided according to the combination of regional nodes and wave operation time windows. Each combination of regional node and each wave operation time window corresponds to a regional wave demand group. For each regional demand group, the demand quantity of each item within the group is counted and sorted from largest to smallest to obtain a list of high-frequency items to be allocated. These high-frequency items are then mapped sequentially to storage locations in the prime storage location index sequence table. When the total number of items is not greater than the total number of storage locations, all items are assigned to the storage locations with the highest order. When the total number of items is greater than the total number of storage locations, the excess items are kept in their original storage locations. The mapping change relationship between items and storage locations is recorded as a dynamic storage location reconstruction scheme for each regional node under the corresponding wave operation time window.

5. The dynamic location reconstruction and task allocation method for wave picking as described in claim 4, characterized in that, The decomposition and reorganization of the regional task set specifically includes: Each parent order in the target order data is divided according to the region node to which the item belongs and the wave operation time window, generating a child order fragment that only contains items with the same region node and needs to be picked within the same wave operation time window; Using the regional node and wave operation time window as the aggregation key, the order fragments of each sub-level are merged to obtain the independent regional task set of each regional node under each wave operation time window.

6. The dynamic location reconstruction and task allocation method for wave picking as described in claim 5, characterized in that, Generating and optimizing closed-loop picking paths specifically includes: Read the physical coordinates of all items to be picked in the independent area task set, determined by the dynamic location reconstruction scheme, and set the channel entrance coordinates of the area node as the path start and end points; The nearest neighbor construction algorithm is used to select the physical coordinate point with the closest straight-line distance or Manhattan distance to the current position from the unvisited physical coordinate points starting from the path origin and add them in turn until all physical coordinate points have been visited and connected back to the path end point to obtain the initial path. Then, a path topology swapping optimization algorithm is used. Two non-adjacent path segments in the initial path are selected and their connection order is swapped. If the total path length decreases after the swap, the swap is retained and the iteration continues until the total path length no longer decreases within a preset number of iterations. The path obtained after the iteration is determined as the optimized closed-loop picking path.

7. The dynamic location reconstruction and task allocation method for wave picking as described in claim 6, characterized in that, The calculation of the corresponding passage cost and the allocation of the regional task set to picking operation resources according to the load balancing strategy specifically include: The total passage value of each independent area task set is calculated based on the sum of the physical lengths of each path segment in the optimized closed-loop picking path, and each independent area task set is stored in the task queue to be assigned in descending order of total passage value. Establish a resource status list, record the cumulative passage value of each picking operation resource assigned to a task in real time, and sort them in ascending order of the cumulative passage value; Sequentially retrieve the set of independent area tasks from the head of the task queue to be assigned and assign them to the picking operation resource that is first in the resource status list. At the same time, update the cumulative passage value and sorting position of the picking operation resource until the task queue to be assigned is cleared, and generate the final task assignment instruction set containing specific execution instructions.

Images