A warehouse multi-point addressing task allocation method based on an improved hybrid frog-leaping algorithm

By improving the hybrid frog-jumping algorithm and combining fuzzy membership degree and spatial load interference degree matrix to optimize the allocation of multi-point addressing tasks in the warehouse, the problem of heterogeneous robot capacity and channel interference is solved, and efficient task allocation and path planning are achieved, thereby improving the collaborative efficiency of the warehouse system.

CN123117594APending Publication Date: 2026-11-17WUHAN NARI LIABILITY OF STATE GRID ELECTRIC POWER RES INST
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611152952.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-31
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing warehouse task allocation and path planning methods are unable to simultaneously address complex constraints such as heterogeneous robot rated capacity, large differences in task point load requirements, limited shelf aisle space, and cross-interference of multiple robot paths. This leads to problems such as the splitting of highly correlated tasks, excessive load on low-capacity robots, local congestion in aisles, and ineffective detours.

Method used

An improved hybrid frog-jumping algorithm is adopted. By acquiring warehouse multi-point addressing task data, channel topology data and robot state data, a fuzzy membership matrix and a spatial load interference matrix are constructed to generate a hybrid frog-jumping algorithm initial population. Iterative optimization is then performed to generate the optimal task allocation result and path.

Benefits of technology

It achieves capacity constraints, task uniqueness, and low-interference collaboration in task allocation under complex scenarios, reducing channel congestion, lowering the total travel distance, and improving overall operational efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123117594A_ABST
    Figure CN123117594A_ABST
Patent Text Reader

Abstract

This application discloses a warehouse multi-point addressing task allocation method based on an improved hybrid frog-jumping algorithm, comprising: acquiring warehouse task data and robot state data; performing fuzzy pre-allocation through capacity-weighted adaptation distance to obtain a fuzzy membership matrix; constructing a spatial load interference matrix by fusing spatial proximity weights, membership consistency weights, and co-robot load weights; and constructing a task point interference map for group mining to obtain task density groups; finally, embedding the fuzzy membership matrix, spatial load interference matrix, and task density groups into a hybrid frog-jumping algorithm for population initialization; and iteratively optimizing the algorithm using interference-sensing discrete jump updates and global dynamic enhancement processes to generate an optimal task allocation and path access scheme that satisfies capacity constraints, task uniqueness constraints, and low-interference collaboration requirements. This invention can effectively reduce channel congestion, decrease total travel distance, and improve the overall efficiency of multi-robot collaborative operations.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of intelligent warehousing and logistics technology, specifically to a method for allocating multi-point addressing tasks in warehousing based on an improved hybrid frog-jumping algorithm. Background Technology

[0002] With the continuous improvement of warehouse automation and intelligence, the collaborative execution of multi-point picking, replenishment or handling tasks by multiple mobile robots has become a key mode for improving logistics efficiency. However, existing task allocation and path planning methods mostly rely on clustering or heuristic allocation based on a single distance or time index, which makes it difficult to simultaneously cope with complex constraints such as heterogeneous robot rated load capacity, large differences in load demand at task points, limited shelf aisle space and cross-robot path interference. This can easily lead to problems such as highly correlated tasks being split into different robots, low-capacity robots being overloaded, local aisle congestion and ineffective detours.

[0003] However, existing warehouse task allocation and path planning methods mostly rely on static allocation based on single distance, time, or clustering indicators, making it difficult to simultaneously address the complex physical constraints of actual warehouse operations, such as heterogeneous robot rated capacity, large differences in task point loads, limited shelf aisle space, and interference between multiple robot paths. This easily leads to highly correlated tasks that should be completed continuously by the same robot (such as adjacent task points within the same aisle) being mechanically split and assigned to different robots, resulting in low-capacity robots being overloaded while high-capacity robots are idle. At the same time, due to the lack of dynamic prediction of aisle topology interference, multiple robots are prone to frequent encounters, deadlocks, or even ineffective detours in narrow aisles or intersections, causing a sharp decline in the overall efficiency of the multi-robot collaborative system.

[0004] Therefore, it is necessary to propose a warehouse multi-point addressing task allocation method based on an improved hybrid frog-jumping algorithm to solve the problems of unreasonable task allocation, easy local congestion and deadlock, and low collaborative efficiency in existing technologies under complex scenarios such as robot heterogeneous capacity limitations, highly correlated task splitting, channel constraints, and multi-machine interference conflicts. Summary of the Invention

[0005] This invention provides a warehouse multi-point addressing task allocation method based on an improved hybrid frog-jumping algorithm, which is used to solve the complex scenarios of discrete spatial distribution of task points, heterogeneous robot capacity and obvious channel interference in the collaborative operation of multiple robots in warehouses. It solves the problems of load imbalance, channel congestion and low efficiency caused by existing methods that ignore capacity matching, split highly correlated tasks and lack adaptability of search strategies.

[0006] This invention provides a warehouse multi-point addressing task allocation method based on an improved hybrid frog-leaping algorithm, comprising:

[0007] Acquire warehouse multi-point addressing task data, warehouse aisle topology data, and warehouse mobile robot status data to establish a warehouse multi-point addressing task dataset for optimization calculations; the warehouse multi-point addressing task data includes task point coordinates and task load requirements, and the warehouse mobile robot status data includes the robot's initial position and rated maximum load capacity;

[0008] Based on the warehouse multi-point addressing task dataset, a fuzzy pre-assignment based on capacity-weighted adaptation distance is performed to obtain a fuzzy membership matrix that reflects the adaptation relationship between task points and warehouse mobile robots.

[0009] Based on the fuzzy membership matrix, task point coordinates, task load requirements, and rated maximum load capacity, a spatial load interference matrix is ​​constructed. The spatial load interference matrix is ​​used to characterize the necessity and feasibility of any two task points being continuously served by the same warehouse mobile robot.

[0010] Based on the spatial load interferometry matrix, a task point interferogram is constructed, and the task point interferogram is subjected to cluster mining to obtain task density clusters used to maintain the local task association structure.

[0011] Based on the fuzzy membership matrix, spatial load interference matrix, and task density grouping, a hybrid frog-jumping algorithm is generated to initialize the population. Through interference-sensing discrete jump updates and global dynamic enhancement, the multi-robot task allocation scheme is iteratively optimized to generate the optimal task allocation result and multi-point access path that can simultaneously meet the capacity constraints, task uniqueness constraints, and low-interference collaboration requirements.

[0012] This invention also provides a warehouse multi-point addressing task allocation system based on an improved hybrid frog-leaping algorithm, used to implement the above method, including:

[0013] The data acquisition module is used to acquire warehouse multi-point addressing task data, warehouse aisle topology data, and warehouse mobile robot status data, and to establish a warehouse multi-point addressing task dataset for optimization calculations; the warehouse multi-point addressing task data includes task point coordinates and task load requirements, and the warehouse mobile robot status data includes the robot's initial position and rated maximum load capacity.

[0014] The fuzzy matrix calculation module is used to perform fuzzy pre-allocation based on capacity-weighted adaptation distance according to the warehouse multi-point addressing task dataset, and obtain a fuzzy membership matrix that reflects the adaptation relationship between task points and warehouse mobile robots.

[0015] The spatial interference matrix construction module is used to construct a spatial load interference matrix based on the fuzzy membership matrix, task point coordinates, task load requirements, and rated maximum load capacity. The spatial load interference matrix is ​​used to characterize the necessity and feasibility of any two task points being continuously served by the same warehouse mobile robot.

[0016] The task density clustering analysis module is used to construct a task point interferogram based on the spatial load interferometry matrix, and to perform cluster mining on the task point interferogram to obtain task density clusters used to maintain the local task association structure.

[0017] The optimized output module is used to generate a hybrid frog-jumping algorithm initial population based on the fuzzy membership matrix, spatial load interference matrix, and task density grouping. Through interference-sensing discrete jump update and global dynamic enhancement, the multi-robot task allocation scheme is iteratively optimized to generate the optimal task allocation result and multi-point access path that can simultaneously meet the capacity constraint, task uniqueness constraint, and low interference cooperation requirements.

[0018] Compared with the prior art, the beneficial effects of the present invention include:

[0019] 1. Encode the heterogeneous capacity of the robot and the task load requirements into the weighted adaptation distance in fuzzy clustering, so that the pre-allocation process takes into account both spatial proximity and capacity matching, and avoids pure distance clustering ignoring load constraints.

[0020] 2. Construct a three-dimensional fusion interference matrix that integrates spatial proximity, consistent affiliation, and co-operation, quantify the channel contention and load imbalance risks that may be caused when task points are split into services, and provide a refined structural prior for combinatorial optimization.

[0021] 3. A discrete jump update mechanism based on interference divergence degree is proposed to dynamically adjust the length of task segments inherited from excellent individuals, effectively breaking low-quality combinations while preserving local high interference structures.

[0022] 4. A global interference balance index is proposed to dynamically adjust the global learning probability and elite perturbation strategy, so that the algorithm strengthens local exploration when the cross-robot interference is strong and accelerates global convergence when the interference is low, thus achieving an adaptive balance of search behavior. Attached Figure Description

[0023] Figure 1 This is a flowchart illustrating the warehouse multi-point addressing task allocation method based on the improved hybrid frog-leaping algorithm provided by the present invention.

[0024] Figure 2 A schematic diagram of a typical scenario for warehouse multi-point addressing provided by the present invention;

[0025] Figure 3The spatial load interference degree three-dimensional surface analysis diagram provided for this invention;

[0026] Figure 4 A schematic diagram illustrating the process by which nine high-interference task points within the same channel are grouped into the same task density cluster, as provided by this invention.

[0027] Figure 5 A scatter plot showing the relationship between the merging path increment and spatial proximity weights provided by this invention;

[0028] Figure 6 A schematic diagram of the stacked area of ​​the cost components of the solution provided by the present invention. Detailed Implementation

[0029] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.

[0030] This invention provides a warehouse multi-point addressing task allocation method based on an improved hybrid frog-leaping algorithm, such as... Figure 1 As shown, Figure 1 This is a flowchart illustrating the warehouse multi-point addressing task allocation method based on the improved hybrid frog-leaping algorithm, the method comprising:

[0031] Step S1: Obtain warehouse multi-point addressing task data, warehouse aisle topology data, and warehouse mobile robot status data to establish a warehouse multi-point addressing task dataset for optimization calculation; the warehouse multi-point addressing task data includes task point coordinates and task load requirements, and the warehouse mobile robot status data includes the robot's initial position and rated maximum load capacity.

[0032] Step S2: Perform fuzzy pre-assignment based on capacity-weighted adaptation distance according to the warehouse multi-point addressing task dataset to obtain a fuzzy membership matrix that reflects the adaptation relationship between task points and warehouse mobile robots.

[0033] Step S3: Combine the fuzzy membership matrix with the task point coordinates, task load requirements and rated maximum load capacity to construct a spatial load interference matrix. The spatial load interference matrix is ​​used to characterize the necessity and feasibility of any two task points being continuously served by the same warehouse mobile robot.

[0034] Step S4: Construct a task point interferogram based on the spatial load interferometry matrix, and perform cluster mining on the task point interferogram to obtain task density clusters used to maintain the local task association structure;

[0035] Step S5: Based on the fuzzy membership matrix, spatial load interference matrix, and task density grouping, generate a hybrid frog-jumping algorithm to initialize the population. Iteratively optimize the multi-robot task allocation scheme through interference-sensing discrete jump update and global dynamic enhancement to generate the optimal task allocation result and multi-point access path that can simultaneously meet the capacity constraint, task uniqueness constraint, and low interference cooperation requirements.

[0036] This embodiment provides a warehouse multi-point addressing task allocation method based on an improved hybrid frog-jumping algorithm. Addressing the challenges of discrete task point spatial distribution, heterogeneous robot capacities, and significant channel interference in multi-robot collaborative warehouse scenarios, this method pre-encodes capacity adaptation relationships, spatial load interference, and task density structures, integrating them into the population initialization, discrete jump update, and elite dynamic enhancement processes of the hybrid frog-jumping algorithm. This generates collaborative picking and handling path schemes that satisfy unique task allocation, robot rated load constraints, executable access order, and low cross-robot interference, thereby reducing channel congestion, lowering total travel distance, and improving overall operational efficiency.

[0037] As a specific embodiment, in step S1, warehouse multi-point addressing task data, warehouse aisle topology data, and warehouse mobile robot status data are read from the warehouse management system, warehouse map system, and robot scheduling system;

[0038] The set of warehouse multi-point addressing tasks is denoted as follows: , ,in, This indicates the total number of task points in the current executable task batch, with values ​​ranging from 80 to 200; The two-dimensional coordinate vector of each task point is denoted as , It refers to the first The position coordinates of each task point in the warehouse plane coordinate system can be obtained from the coordinates of the storage location, electronic map coordinates, or positioning system coordinates; The task load requirement of each task point is denoted as . , Used to characterize the Each task point corresponds to the weight of the material or its equivalent load.

[0039] The set of status data of the warehouse mobile robot is denoted as , ,in, This indicates the total number of warehouse mobile robots participating in the current batch of executable tasks, with values ​​ranging from 4 to 12; The initial position of each warehouse mobile robot is denoted as: , It refers to the first The position coordinates of each warehouse mobile robot on the warehouse map before executing the current batch of executable tasks; The rated maximum load capacity of each warehouse mobile robot is denoted as... , Used to characterize the The maximum weight of materials that a warehouse mobile robot is allowed to carry is, for example, 30 kg to 80 kg.

[0040] In practical implementation, the set of warehouse multi-point addressing tasks is read from the warehouse management system. Two-dimensional coordinate vector of each task point and task load requirements Read the warehouse mobile robot set from the robot scheduling system. Initial position of each warehouse mobile robot and rated maximum load capacity Read the channel topology, channel length, restricted areas, and one-way traffic rules from the warehouse map system.

[0041] Specifically, warehouse aisles, intersections, charging points, and cargo location entrances are converted into a warehouse aisle topology map. , This refers to a map graph structure composed of walkable nodes and walkable edges. If the task point coordinates or robot position are not on a walkable node, they are first projected to the nearest walkable node or storage location entrance node, and then the shortest path algorithm is used to calculate the walkable distance between any two storage plane positions. The walkable distance function is denoted as... ,in, and These represent the coordinates of two locations within the warehouse plane. Indicates from position Arrive at the location The passable path length is used to replace the simple Euclidean distance, so that the subsequent pre-allocation results conform to the warehouse channel constraints.

[0042] coordinates and Projected onto warehouse aisle topology map Find the nearest walkable node between the two nodes, and then run a shortest path algorithm (such as Dijkstra's algorithm or A* algorithm) between these two walkable nodes. The total path length obtained is the result. , The units used are consistent with the units in the warehouse map coordinate system (e.g., meters). For example, there are task points... and Its two-dimensional coordinates are not on the channel node. First, in Find the distance The nearest node (e.g., the entrance node of a certain storage location) and distance The nearest node Then, Dijkstra's algorithm is used in... Calculate the passable edges from arrive The shortest path, if the path passes through nodes in sequence. If the lengths of each side segment are 5 meters, 8 meters, and 6 meters respectively, then rice.

[0043] By uniformly modeling the task point coordinates, task load requirements, robot rated maximum load capacity, and travel distance, a warehouse multi-point addressing task dataset that can be used for optimization calculations is obtained.

[0044] It should be noted that before implementing this method, if the total task load requirement of the current order exceeds the total rated maximum load capacity of all warehouse mobile robots, the warehouse management system will split the current order into multiple executable task batches according to order priority, storage location aisle area, or material batch, ensuring that each executable task batch meets the requirements. If the task load demand of a single task point exceeds the maximum rated maximum load capacity of all warehouse mobile robots, then that task point will be marked as a manual processing task or a split handling task, and will not be included in the current executable task batch.

[0045] like Figure 2 As shown, Figure 2 The diagram illustrates a typical scenario of multi-point addressing in warehousing.

[0046] Warehouse multi-point addressing tasks are typically characterized by discrete spatial distribution of task points, large differences in task load requirements, heterogeneous robot capacity, and significant channel interference. Conventional random initialization multi-robot path optimization methods are difficult to utilize the spatial proximity relationship of task points and the task load adaptation relationship, which can easily lead to problems such as highly correlated task points in the same channel being dispersed to different robot paths, low-capacity robots being assigned excessive tasks, and frequent intersections across robot paths.

[0047] Since the rated maximum load capacity and initial position of warehouse mobile robots often differ, pre-allocating loads solely based on the distance from the task point to the robot's initial position can easily lead to low-capacity robots being overloaded or high-capacity robots failing to fully utilize their load capacity. To address this issue, in step S2, a capacity-weighted adaptation distance is established based on the task point coordinates, task load requirements, robot initial position, and robot rated maximum load capacity. A fuzzy membership matrix is ​​then obtained through fuzzy clustering, ensuring that the adaptation degree of the task point to different warehouse mobile robots simultaneously reflects both spatial proximity and capacity adaptation relationships. The specific steps are as follows:

[0048] Step S21: Calculate the first... The task point and the first The normalized travel distance between the cluster centers and the first cluster center The normalized capacity of each warehouse mobile robot. Specifically:

[0049] The first Initial position of a warehouse mobile robot As the first The initial coordinates of each cluster center are obtained. ,in, , Indicates the first The center of the task adaptation region corresponding to each warehouse mobile robot in the 0th iteration; the iteration number index is denoted as... , ,in, This represents the maximum number of iterations for fuzzy pre-assignment, with an example value of 100.

[0050] Furthermore, in the In the nth iteration, calculate the th The task point and the first Normalized travel distance between cluster centers ,Right now .in, Indicates the first During the nth iteration The task point and the first Normalized travel distance between cluster centers is used to scale travel distances in different storage areas to a uniform numerical range. Indicates the first During the nth iteration The coordinates of the cluster center for each warehouse mobile robot; Indicates the first The maximum travel distance between all task points and all cluster centers in the next iteration; This represents the distance normalization stability constant, used to prevent the denominator from being zero. Examples of its values ​​are shown below. .

[0051] At the same time, calculate the first Normalized capacity of a warehouse mobile robot ,Right now .in, Indicates the first The normalized capacity of a warehouse mobile robot is used to characterize the capacity level of that warehouse mobile robot relative to all warehouse mobile robots. This represents the robot comparison index, used to traverse the set of warehouse mobile robots. ; Indicates the first The rated maximum load capacity of each warehouse mobile robot; This represents the capacity normalization stability constant, used to prevent the denominator from being zero. Examples of its values ​​are shown below. .

[0052] Step S22: Calculate the capacity-weighted adaptation distance based on the normalized travel distance, normalized capacity, and task load requirements. Capacity-weighted adaptation distance Used to characterize the During the nth iteration The task point is determined by the first The degree of spatial proximity and load adaptability of the warehouse mobile robot service is considered; the smaller the value, the more suitable the warehouse mobile robot is for handling the first task. The number of task points is calculated as follows: ;

[0053] in, This represents the load adaptation adjustment coefficient, used to control the influence of task load demand on the capacity-weighted adaptation distance. The value range is 0.3 to 1.0. Used to enhance the differentiation of task attractiveness based on spatial distance differences; This is used to enable warehouse mobile robots with larger rated maximum load capacity to have a smaller equivalent adaptation distance at the same distance; Characterizing the first The task load requirement of each task point accounts for the first... The ratio of the rated maximum load capacity of each warehouse mobile robot is used to prevent tasks with high loads from being assigned to warehouse mobile robots with lower capacity.

[0054] It should be noted that, if Explanation of the first A single warehouse mobile robot cannot handle the task alone. If there are 1 task point, then... Set to the maximum value or set the first The warehouse mobile robot from the first The robot is removed from the candidate robot set for each task point so that subsequent fuzzy membership calculations will not assign uncapable tasks to the warehouse mobile robot.

[0055] Step S23: Calculate fuzzy membership degree based on capacity-weighted adaptation distance. Fuzzy membership Used to characterize the During the nth iteration The task point is determined by the first The degree of compatibility of a warehouse mobile robot service is indicated by a higher value.

[0056] In the specific implementation, first determine the first Capacity-feasible candidate robot set for each task point , This refers to a rated maximum load capacity of not less than A collection of warehouse mobile robots; then, using a capacity-weighted adaptation distance calculation method, only... The fuzzy membership degree is calculated internally and expressed as: (for ,make );

[0057] in, Indicates the first During the nth iteration The task point is related to the first Fuzzy membership degree of a warehouse mobile robot; This represents the stability constant for membership calculation, used to prevent numerical overflow caused by excessively small capacity-weighted fit distance. An example value is shown below. ; This represents the fuzzification coefficient, used to control the smoothness of the fuzzy membership degree distribution. The value of is greater than 1, and the preferred value is 2; Indicates the first During the nth iteration The task point and the first Capacity-weighted adaptation distance between warehouse mobile robots.

[0058] Based on this, the sum of the fuzzy membership degrees of all feasible warehouse mobile robots at the same task point is 1; warehouse mobile robots that are closer, have higher capacity, and lower load ratio correspond to higher fuzzy membership degrees.

[0059] Step S24: Iteratively update the cluster center coordinates based on the fuzzy membership degree, and repeatedly perform capacity-weighted adaptation distance calculation and fuzzy membership degree calculation until the change in fuzzy membership degree between two adjacent iterations is less than the preset convergence threshold, or the number of iterations reaches the maximum number of iterations.

[0060] In specific implementation, the first The cluster center coordinates of the warehouse mobile robots are updated using a fuzzy membership degree weighted average method, as follows: ;

[0061] in, Indicates the first During the nth iteration The coordinates of the cluster center for each warehouse mobile robot; Indicates the first The task point is related to the first The fuzzy membership weights of each warehouse mobile robot are used to make highly fit task points have a greater influence on the location of cluster centers.

[0062] The change in fuzzy membership degree between two consecutive iterations is denoted as... , is represented as: ;

[0063] in, Indicates the first The iteration relative to the first The maximum change in fuzzy membership degree in each iteration is used to determine whether the fuzzy pre-assignment has converged; the preset convergence threshold is denoted as... , Used to control the precision of iteration stopping; examples of values ​​are provided. .

[0064] Step S25: After the fuzzy pre-assignment converges, the fuzzy membership matrix is ​​obtained. , This refers to a matrix composed of the fuzzy membership degrees of all task points to all warehouse mobile robots, with dimension [missing information]. Matrix elements Indicates the first The task point is related to the first The final fuzzy membership degree of each warehouse mobile robot.

[0065] Furthermore, based on the fuzzy membership matrix... Generate task pre-assignment relationships. For the first... For each task point, the warehouse mobile robot with the highest fuzzy membership degree is prioritized as the pre-assigned robot. If the remaining capacity of this warehouse mobile robot is insufficient in the current cumulative allocation state, other warehouse mobile robots that meet the remaining capacity constraint are selected in descending order of fuzzy membership degree. If none of the warehouse mobile robots can receive the first task, the remaining task is completed. The first task point will then be the first... Each task point is then assigned to the warehouse management system for adjustment to the next batch of executable tasks.

[0066] It should be noted that the pre-assignment relationship is not directly used as the final task assignment result, but is used to provide heuristic information for subsequent spatial load interferometry matrix construction, task density clustering mining, and population initialization, i.e., fuzzy membership matrix. Preserving the coupling relationship between the spatial location of the task point, the task load requirements, and the differences in robot capacity can reduce the inefficiency caused by subsequent optimization algorithms starting the search from a completely random structure.

[0067] In one embodiment, for example, if a task point is 10 meters away from both the first and second cluster centers of warehouse mobile robots, the rated maximum load capacity of the first warehouse mobile robot is 60 kg, and the rated maximum load capacity of the second warehouse mobile robot is 30 kg, and the task load requirement of this task point is 12 kg, then under the same distance conditions, the normalized capacity of the first warehouse mobile robot is higher, and the task load requirement accounts for a lower proportion of its capacity. Therefore, the capacity-weighted adaptation distance corresponding to the first warehouse mobile robot is smaller, and the fuzzy membership degree of this task point to the first warehouse mobile robot is higher. This allows the high-capacity warehouse mobile robot to prioritize task points with higher load requirements and closer spatial proximity.

[0068] In warehousing scenarios, if task points located in the same shelf aisle, near the same intersection, or adjacent storage areas are served by different warehousing mobile robots, it is easy to cause aisle contention, path intersection, and local congestion. However, if task points are forcibly merged based solely on spatial distance, it may cause low-capacity robots to be overloaded or to detour across areas.

[0069] To address the aforementioned issues, in step S3, this method constructs a spatial load interference matrix by integrating the merging path increment between task points, fuzzy attribution consistency, and co-robot carrying feasibility. This ensures that task points with higher interference are more likely to be continuously served by the same warehouse mobile robot in subsequent optimizations. The specific steps are as follows:

[0070] Step S31: Calculate the travel distance between any two task points, and calculate the nearest neighbor distance for each task point. The task point and the first The travel distance between each task point is denoted as . ;

[0071] in, Indicates the index of the first task point. Indicates the index of the second task point, and ; , is used to characterize the reachability distance between two task points under the constraints of the storage channel.

[0072] The first The nearest neighbor distance of each task point is denoted as . ,Right now ;

[0073] in, Indicates the first The minimum travel distance from one task point to other task points is used to reflect the minimum travel distance from the first task point to other task points. The density of task points in the local space. If Smaller, indicating the first If other task points are nearby, a locally contiguous access structure is more likely to be formed; if Larger, indicating that the first If a task point is relatively isolated, directly merging the service with other task points may result in additional detours.

[0074] Step S32: Calculate the merged path increment based on the travel distance and nearest neighbor distance between the two task points. Merge path increment Used to estimate the The task point and the first The additional local path cost that may be introduced when multiple task points are visited consecutively by the same warehouse mobile robot, i.e. ;in, Indicates the first The task point and the first The incremental merged path between task points; Indicates the first The nearest neighbor distance of each task point.

[0075] It should be noted that using the merged path increment instead of just the two-point travel distance can avoid two isolated task points being mistakenly identified as high-density task points due to accidental proximity, and can also enhance the tendency of task points in the same dense storage area to merge services.

[0076] Step S33: Calculate the spatial proximity weight based on the merging path increment. Spatial proximity weight Used to characterize the The task point and the first Whether a task point has spatial merging service value is calculated as follows: ;

[0077] in, This represents the spatial proximity threshold, used to determine whether two task points have spatial merging value. The value range is 8 to 15 meters, which can be determined based on shelf spacing, aisle width, and robot turning radius. This represents the spatial proximity sharpness coefficient, which controls the intensity of the influence of the merged path increment on the spatial proximity weight. An example value is 1.5. This represents the natural exponential function, used to convert spatial proximity relationships into smooth weights.

[0078] In practical implementation, when Less than hour, A higher value indicates that the two task points have strong spatial merging service value; when Significantly greater than hour, The lower value indicates that the two task points are not suitable for being forcibly merged solely due to spatial relationships.

[0079] Step S34: Based on the fuzzy membership matrix Calculate the attribution consistency weight , It refers to the first The task point and the first The consistency weight between task points is used to suppress erroneous merging of spatially similar but capacity-adaptable objects.

[0080] In specific implementation, the first The task point and the first The average attribution preference difference of each task point across all warehouse mobile robots is denoted as . ,Right now .in, Indicates the first The task point and the first The average attribution tendency difference between two task points indicates that the two task points are more inconsistent in their suitability for the warehouse mobile robot. Indicates the first The task point is related to the first Fuzzy membership degree of a warehouse mobile robot; Indicates the first The task point is related to the first Fuzzy membership degree of a warehouse mobile robot; This indicates that two task points are related to the first task. Differences in the affiliation preferences of individual warehouse mobile robots.

[0081] In practical implementation, The calculation method is as follows: for the first... The first task point and the first Each task point is determined from the fuzzy membership matrix. Read them from the first Fuzzy membership values ​​of individual warehouse mobile robots and Then subtract the two values ​​and take the absolute value; the smaller the absolute value, the closer the two task points are to the same robot.

[0082] Furthermore, the average attribution propensity difference is mapped to an attribution consistency weight. ,Right now .in, This represents the sensitivity coefficient for attribution differences, used to control the attenuation strength of the attribution consistency weight due to the average attribution propensity difference. An example value is 2.0.

[0083] Step S35: Calculate the co-loading weight based on the feasibility of having the same warehouse mobile robot carry two task points. Same machine carrying weight Used to characterize the The task point and the first The load feasibility and adaptability of multiple task points being served by the same warehouse mobile robot, i.e. .in, Indicates that it can simultaneously carry the first The first task point and the first A collection of warehouse mobile robots at a task point is defined as... ; Indicates the first The first task point and the first The task points are jointly adapted to the first The extent of mobile warehouse robots; Used to select the highest common fit strength among all feasible robots with common capacity. If If it is empty, then let This indicates that two task points are not suitable to be forcibly merged into the same robot path.

[0084] Step S36: Merge the spatial proximity weight, the affiliation consistency weight, and the co-operation weight to obtain the first... The task point and the first Spatial load interference between task points ,Right now .in, Indicates the first The task point and the first Spatial load interference between task points, with a value ranging from 0 to 1, is used to characterize the necessity and feasibility of two task points being continuously served by the same warehouse mobile robot. The larger the value, the closer the two task points are in space, the more consistent the robot's affiliation, and the more feasible it is to carry them on the same machine. In subsequent optimization, we should avoid dispersing them to different warehouse mobile robot paths.

[0085] Step S37: Traverse all task point pairs and construct the spatial load interferometry matrix. Spatial load interferometry matrix It refers to a matrix composed of the spatial load interference degrees between all task points, with dimensions of . Matrix elements Indicates the first The task point and the first Spatial load interference between task points.

[0086] In practical implementation, the spatial load interferometry matrix Set as a symmetric matrix, i.e. diagonal elements Setting it to 0 is used to avoid the task point forming meaningless interference edges with itself. Spatial load interferometry matrix. It represents both the proximity of warehouse spaces and the feasibility of same-machine services resulting from differences in task load requirements and robot capacity.

[0087] In one embodiment, for example, two storage locations are located on the same shelf aisle and are relatively close to each other. Both locations have a high fuzzy membership degree to a third high-capacity warehouse mobile robot, and the sum of the task load requirements of the two locations is less than the rated maximum load capacity of the third high-capacity warehouse mobile robot. In this case, the spatial proximity weight, membership consistency weight, and same-robot load weight are all high, corresponding to a high degree of spatial load interference. In subsequent optimization, the two task points are tended to be assigned to the same warehouse mobile robot path. If two storage locations are spatially adjacent, but one task point has a high load and is only suitable for a high-capacity warehouse mobile robot, while the other task point is closer to the area handled by a low-capacity warehouse mobile robot, then the membership consistency weight or same-robot load weight is reduced, thereby reducing the degree of spatial load interference and avoiding forced merging that could cause capacity imbalance or cross-regional detours.

[0088] In one embodiment, such as Figure 3 As shown, Figure 3 This is a surface analysis diagram of the spatial load interferometry of this application. It displays the fluctuations of the spatial load interferometry between task point pairs on a two-dimensional task index plane in the form of a three-dimensional surface plot. Elements on the diagonal are set to zero. The two horizontal axes represent the task point index and the task point index, respectively, and are dimensionless. The vertical axis represents the spatial load interferometry, which is also dimensionless. The peak regions on the surface correspond to high-interference task point pairs, demonstrating that the interferometry exhibits a clear local clustering structure rather than a uniform distribution in the task point space.

[0089] Furthermore, although the spatial load interferometry matrix While capable of representing the local relationship between any two task points, subsequent population initialization and path updates require more stable task structure units. In step S4, this method constructs a task point interference graph using task points as graph nodes and spatial load interferometry as edge weights. It then obtains task density clusters through spectral clustering, ensuring that highly interfering task points are clustered within the same task density cluster. The steps are as follows:

[0090] Step S41: Using the task point as a graph node, and the spatial load interference degree As the first The task point and the first The edge weights between task points are used to construct a task point interference graph. The task point interference graph is used to transform the warehouse multi-point addressing task into a graph partitioning problem, so that task points with spatial proximity, capacity matching and consistent affiliation form stronger connections in the graph structure.

[0091] In practical implementation, the spatial load interferometry matrix will be used. This serves as the initial graph similarity matrix. To reduce the computational complexity of dense graphs, only the one with the highest spatial load interference can be retained for each task point. By considering the edges, we obtain the graph edge weight matrix. .in, The edge weight matrix of the task point interferogram has dimensions of . ; represents the sparsity of the matrix, indicating the number of high-interference neighbors retained for each task point, with values ​​ranging from 8 to 15. If sparsity processing is not required, it can be directly set to . This is the original spatial load interferometry matrix.

[0092] As a specific embodiment, the implementation process of the above method is as follows:

[0093] First, initialize one All-zero matrix Then, for each task point Find the degree of interference with its spatial load. ( The highest Neighbor task points ; this Interference value between task points Fill in the corresponding positions in the matrix and The weights of the remaining unselected edges remain at 0.

[0094] Step S42: Based on the graph edge weight matrix Calculate the graph degree matrix Graph degree matrix This refers to the node connection strength matrix of the task point interferogram, which is a diagonal matrix. The diagonal element represents the first... The sum of the edge weights between each task point and other task points, i.e. .in, Representing the graph degree matrix The Middle The diagonal element is used to represent the first diagonal element. The total connectivity strength of each task point in the task point interferogram; Represents the graph edge weight matrix The Middle Line 1 The edge weights of the column.

[0095] In other words, the graph degree matrix The total connectivity strength of each task point in the entire network was quantified, and this was used as a benchmark to eliminate the bias of "super nodes" on the clustering results in subsequent calculations. When a task point is surrounded by highly interfering neighbors (i.e., located in a densely packed shelving area), its connectivity strength is high; conversely, when a task point is located in a corner of the warehouse, its connectivity strength is low. (Graph degree matrix) It directly reflects the pivotal position or cluster density of the task point in the local structure.

[0096] Step S43: Based on the graph edge weight matrix And graph degree matrix Constructing the normalized graph Laplacian matrix and extract the previous eigenvectors. Among them, the normalized graphical Laplacian matrix. The overall connection structure used to describe the task point interferogram is represented as follows: ;

[0097] In the formula, Representing the graph degree matrix The inverse square root matrix.

[0098] Setting parameters The total number of task-intensive clusters can be determined based on the number of warehouse mobile robots, the number of warehouse zones, or the interval of graph feature values. An example value is shown below. to When the aisle zones in the warehouse area are clearly defined, It can be preferentially set to the larger value between the number of warehouse zones and the number of warehouse mobile robots; when the task points are relatively evenly distributed. Desirable This ensures that the number of task-intensive groups is consistent with the number of warehouse mobile robots.

[0099] Step S44: Place the front The feature vectors are arranged row-wise to form a task point embedding matrix, and clustering is performed on the task point embedding matrix to obtain task density clusters. .in, Indicates the first A task density cluster refers to a set of task points that have high spatial load interference and are suitable for co-service. This represents the task density group index, used to distinguish different task density groups.

[0100] In practical implementation, clustering methods can be adopted. Mean value clustering can be used, with Euclidean distance as the clustering distance. The maximum number of clustering iterations can be set to 100. After clustering is completed, the th iteration is calculated. Total load of groups based on task density ,Right now .in, Indicates the first The total load of a task-intensive group is used to determine whether the task-intensive group can be carried by a single warehouse mobile robot.

[0101] Furthermore, the method also includes: performing capacity feasibility verification on each task density group obtained by clustering; if the total load of a task density group exceeds the maximum value among the rated maximum load capacities of all warehouse mobile robots, then the task density group is further split until each split subgroup has at least one warehouse mobile robot capable of carrying the entire load.

[0102] Specifically, if If the load capacity exceeds the maximum value among all warehouse mobile robots' rated maximum load capacities, then the robot will be assigned a specific load capacity based on the task point's channel location, the strength of spatial load interference, or the travel distance from the anchored task. The task-intensive groups are further subdivided until each subdivided group has at least one warehouse mobile robot capable of handling the task, thus avoiding the formation of large groups with unmanageable capacity.

[0103] Step S45: Select the anchor task within each task-closeness group. Specifically, the... The anchor task index of a task-closeness group is denoted as... ,Right now .in, Indicates the first The anchor task index of each task density group is used to characterize the core task point with the strongest interference connection within the task density group. Indicates the first Index of candidate task points within a task density group; Indicates the first The index of task points associated with candidate task points within each task density group; Indicates the first The task point and the first Spatial load interference between task points; This indicates the index selection operator corresponding to the maximum value, used to select the task point with the largest sum of interference within the group.

[0104] In practical implementation, if the sum of the spatial load interference degrees within a cluster of multiple candidate task points is the same, the task point closest to the geometric center of the cluster with high task density is selected as the anchor task. If multiple candidate task points still exist, the task point with higher task load demand is selected as the anchor task. The anchor task is used for subsequent initial access path generation and local path repair, enabling high-interference task points to form a more stable access order around the anchor task.

[0105] It should be noted that the task density clustering is not the final task allocation result, but rather an intermediate structural unit used to protect the structure of highly interferometric tasks. This is achieved by using the spatial load interferometry matrix... Converting to a task point interference map and then mining task density clusters from the task point interference map can help the subsequent meme evolution algorithm retain task point sets with the same channel, same region, and same capacity adaptation relationship as much as possible during initialization and jump update, thereby reducing the probability of cross-robot interference and path intersection.

[0106] In one embodiment, assuming there are 12 task points in a certain shelf aisle, 9 of which are spatially adjacent, have moderate task load requirements, and all have a high fuzzy membership degree to the second warehouse mobile robot, the edge weights between these 9 task points in the task point interference graph are relatively high. Spectral clustering will preferentially classify them into the same task density group. The task point with the largest sum of spatial load interference degrees with other task points in this task density group is identified as the anchor task. During subsequent population initialization, local access paths can be constructed around this anchor task to reduce the probability that task points in the same aisle are scattered across multiple warehouse mobile robot paths.

[0107] like Figure 4 As shown, nine high-interference task points within the same channel are grouped into the same task density cluster and form a local access structure around the anchored task.

[0108] In one embodiment, such as Figure 5 As shown, Figure 5 A scatter plot showing the relationship between the merge path increment and spatial proximity weights is presented. The analysis examines the relationship between the merge path increment and the corresponding spatial proximity weights for task point pairs. The horizontal axis represents the merge path increment in meters, and the vertical axis represents the spatial proximity weights (dimensionless). The scatter plots show a clear negative correlation: the spatial proximity weights are higher when the merge path increment is small, and decrease rapidly when the increment is large.

[0109] Steps S2-S4 above, through capacity-aware fuzzy pre-allocation, spatial load interference matrix construction, and task density clustering mining, pre-encode the spatial distribution of task points, task load requirements, and robot capacity differences into structural information that can be used for memetic evolution optimization. This is achieved by obtaining the fuzzy membership matrix. Spatial load interferometry matrix After grouping with task density, in step S5, the present invention embeds the above-mentioned structural information into the population initialization, meme group discrete jump update and elite dynamic enhancement process of the improved hybrid frog jumping algorithm, so that the search process simultaneously satisfies capacity constraints, task uniqueness constraints and low interference cooperation requirements.

[0110] Before explaining the implementation and improvement of the hybrid frog jumping algorithm in this method, we will first introduce the meaning of each parameter in the algorithm and the calculation rules of this algorithm.

[0111] As a specific embodiment, this method will... Each frog is denoted as , It refers to the first A multi-robot task allocation and multi-point access path scheme. The frog individual is assigned to the first The set of tasks for a warehouse mobile robot is denoted as . , It refers to the first In the set of schemes, the first A set of task points served by a warehouse mobile robot; the first The first frog individual The access order of the warehouse mobile robots is denoted as follows: , It refers to the first A warehouse mobile robot The order of visits to the task points.

[0112] The task uniqueness constraint requires that each task point can only appear in the task set of one warehouse mobile robot; the capacity constraint requires... Access order feasibility constraint requirements Only contains The task points are not repeated.

[0113] To evaluate the quality of an individual frog, the first... Cost per frog Cost of the solution This is used to comprehensively measure path travel length, capacity violation degree, cross-robot interference degree, and access sequence disruption degree. The smaller the value, the higher the solution quality, expressed as: ;

[0114] in, Indicates the first The path travel cost of an individual frog is used to represent the total travel distance of all warehouse mobile robots. Indicates the first The capacity default cost for each individual frog is used to penalize capacity over-limit schemes. Indicates the first The cross-robot interference cost for each individual frog is used to penalize high-interference task points that are dispersed to different warehouse mobile robot paths. Indicates the first The cost of breaking the access sequence of an individual frog is used to penalize low spatial load interference between adjacent task points in the same robot access sequence. , , and These represent the weights for path travel cost, capacity violation cost, cross-robot interference cost, and access order disruption cost, respectively. Their values ​​can be set according to the warehouse scheduling objectives, for example... , , , .

[0115] In practical implementation, path travel cost This can be calculated by summing the travel distances of each warehouse mobile robot as it starts from its initial position, sequentially visits the task points, and returns to the specified ending position. The final position of each warehouse mobile robot is denoted as , It refers to the first After completing their tasks, the warehouse mobile robots can return to their unloading points, waiting points, or charging points. If they do not need to return after completing their tasks, the last task point can be set to... The distance parameter is set to 0.

[0116] No. The fitness score of each individual frog is denoted as: ,Right now .in, Indicates the first The fitness score of an individual frog is indicated by its fitness score. A higher value indicates a better individual frog. "Highest fitness" refers to the highest fitness score, and "lowest fitness" refers to the lowest fitness score.

[0117] In step S5, a hybrid frog-jumping algorithm is generated to initialize the population based on the fuzzy membership matrix, spatial load interference matrix, and task density clustering. Through interference-sensing discrete jump updates and global dynamic enhancement, the multi-robot task allocation scheme is iteratively optimized to generate the optimal task allocation result and multi-point access path that simultaneously satisfies capacity constraints, task uniqueness constraints, and low-interference cooperation requirements. Specific steps include:

[0118] Step S51: Guided discrete population initialization of task-dense clusters;

[0119] Step S52: Dynamically truncate the span of the meme group for discrete jump update based on the degree of interference;

[0120] Step S53: Washing and Elite Dynamic Enhancement Based on Global Interference Balance Index.

[0121] Conventional hybrid frog-leaping algorithms often use random allocation for the initial population, which easily leads to problems such as capacity overload, task points being scattered across regions, and high-interference task points being broken up. This invention utilizes task density clustering and fuzzy membership matrices. An initial frog population is generated, ensuring that the initial population satisfies the capacity constraint while preserving the local structure of high-interference task points. Therefore, the specific steps for population initialization in step S51 include:

[0122] Step S511: Set the initial frog population ,Right now .in, The initial frog population refers to the population formed by... A set of candidate solutions consisting of an initial set of individual frogs; This represents the total number of individual frogs and is used to control the population size; values ​​range from 50 to 150. Indicates the first Individual frogs, .

[0123] Step S512: For each individual frog The traversal order of task density groups is randomly shuffled, and the average fuzzy membership degree of each task density group to each warehouse mobile robot is calculated. The task's density is grouped towards the first The average fuzzy membership degree of each warehouse mobile robot is denoted as . ,Right now .in, Indicates the first The task's density is grouped towards the first Average compatibility of individual warehouse mobile robots; Indicates the first The number of task points in a task-intensive group; Indicates the first The task point is related to the first Fuzzy membership degree of a warehouse mobile robot.

[0124] Step S513: Allocate probabilities according to the overall group. Decide whether to retain the first The overall structure of task-intensive groups. Among them, This represents the overall probability of group allocation, used to control the strength of task density groups being retained as a whole in the initial frog population, with values ​​ranging from 0.6 to 0.85.

[0125] In the specific implementation, a random number between 0 and 1 is generated; if the random number is less than 1... Then try to make the first The task density group is assigned to the group with the highest average fuzzy membership degree and the remaining capacity meets the total load of the group. The required warehouse mobile robot; if the remaining capacity of the warehouse mobile robot is insufficient, other warehouse mobile robots that meet the capacity constraints are selected in descending order of average fuzzy membership degree; if no such robot exists, the entire robot can be used to carry the required capacity. Warehouse mobile robots, grouped together by task density, are anchored to specific tasks. Centered on, the first Each task density group is divided into several subgroups according to the strength of spatial load interference and travel distance, and then capacity constraint allocation is performed on each subgroup.

[0126] Based on this, high-interference task points have a high probability of being retained as a whole in the initial frog individuals, while the robot's rated maximum load capacity constraint will not be violated due to forced overall allocation.

[0127] Step S514: Perform capacity constraint random allocation for task points not allocated through the group-based allocation method. Specifically, for the first... For each unassigned task point, first filter those with remaining capacity not less than [amount missing]. The warehouse mobile robots are selected to form a candidate robot set; then, the random selection weights of the candidate robots are calculated based on fuzzy membership degree, remaining capacity ratio, and spatial proximity relationship; finally, a roulette wheel random selection method is used to determine the first robot. The robot assigns tasks to specific locations.

[0128] In one implementation, the first The task point is assigned to the first Random selection weights for individual warehouse mobile robots Calculated based on fuzzy membership degree, remaining capacity ratio, and spatial proximity, it is expressed as: ;

[0129] in, Indicates the first The task point is assigned to the first Random selection weights for individual warehouse mobile robots; Indicates the first Current remaining capacity of the warehouse mobile robots; Indicates the first The remaining capacity percentage of each warehouse mobile robot is used to avoid continuously concentrating task points on warehouse mobile robots whose capacity is about to be exhausted. The representation space is close to the weight, which is used to reduce the probability of cross-regional allocation.

[0130] Step S515: Generate an initial access sequence for each warehouse mobile robot. For the... A warehouse mobile robot, read its task set. ,like If it is empty, then the first The warehouse mobile robot in the first If no individual frog performs the task; If not empty, prioritize the position closest to the initial position. The nearest task or its anchor task within a task cluster with high density is used as the starting task for access. Then, the nearest neighbor rule is used to progressively select the next task for access, until... All task points in the process are included in the access order. .

[0131] Furthermore, to improve the diversity of the initial frog population, after the nearest neighbor access order is generated, a small amount of random exchange perturbation is performed on the access order of each warehouse mobile robot. The random exchange perturbation can select two task points in the same access order to swap positions, with the number of exchanges ranging from 5% to 10% of the length of the access order. The random exchange perturbation only changes the access order and does not change the task allocation relationship, so it does not violate task uniqueness and capacity constraints.

[0132] Step S516: Perform a feasibility check on the generated frog individuals. If a task point appears repeatedly in multiple warehouse mobile robot task sets, retain the task point in the warehouse mobile robot task set with the highest fuzzy membership degree and feasible capacity, and delete it from other warehouse mobile robot task sets; if a task point is not assigned, replenish it according to the capacity constraint random allocation method; if a warehouse mobile robot exceeds its capacity limit, prioritize removing task points with lower fuzzy membership degree and higher task load requirements for that warehouse mobile robot, and reassign them to other warehouse mobile robots with feasible capacity.

[0133] In one embodiment, for example, a task-density cluster contains 10 high-interference task points. This task-density cluster has the highest average fuzzy membership degree to the third warehouse mobile robot, and the remaining capacity of the third warehouse mobile robot can handle the total load of the task-density cluster. Therefore, the entire task-density cluster is allocated to the third warehouse mobile robot with a high probability, and a local access order is generated around the anchored task according to the nearest neighbor rule. If the remaining capacity of the third warehouse mobile robot is insufficient, other robots with feasible capacity are selected according to the average fuzzy membership degree order, or the task-density cluster is split into several sub-clusters before allocation, thereby avoiding the initial individual robot directly violating the capacity constraint.

[0134] Traditional hybrid frog-jumping algorithms typically guide weaker individuals within a meme group to update during the local search phase by using better individuals within the meme group. However, multi-robot multi-point addressing tasks in warehouses are discrete combinatorial optimization problems. Fixing the task segment length can easily disrupt high-interference task points that have already formed a reasonable continuous structure, or result in insufficient updates when low-interference task points are incorrectly bundled. To address this issue, this invention dynamically determines the task segment truncation length based on the average interference divergence degree within the candidate task subsequence, allowing the discrete jump span to adaptively change with the current local structure quality. As a preferred embodiment, step S52, the step of dynamically trunculating the span within the meme group for discrete jump updates based on interference degree, specifically includes:

[0135] Step S521: Sort the current generation of frogs from high to low fitness scores, and divide them into groups using a round-robin assignment method. One meme group. This represents the number of meme groups, used to control the size of local search groups in the hybrid frog-jumping algorithm. Examples of values ​​are 5 to 10.

[0136] The rotational allocation method refers to assigning the first sorted frog individual to the first meme group, the second frog individual to the second meme group, and so on, until the next group is reached. After each meme group is assigned a different number of meme groups, the assignment returns to the first meme group. Through this rotational allocation, each meme group contains frog individuals of different quality levels, which avoids the excessive concentration of high-quality individuals in a few meme groups.

[0137] Step S522: For each meme group, identify the frog individual with the lowest fitness within the meme group. The frog individual with the highest fitness within the meme group The frog with the highest global fitness at present .in, This represents the frog individual within the current meme group that needs improvement; This refers to a reference frog individual within the current meme group that can provide an excellent structure within the group; This represents the global reference frog individual with the highest fitness in the current population.

[0138] Step S523: From the frog individual to be updated Randomly select a non-empty access sequence of warehouse mobile robots, and extract consecutive task segments from this access sequence to form candidate task subsequences. Candidate task subsequence Used to assess the degree of interference divergence in the current local access structure.

[0139] Furthermore, based on the spatial load interferometry matrix Calculate candidate task subsequences Internal average interference divergence , is represented as: ;

[0140] in, This represents the average interference divergence degree within the candidate task subsequence, with a value ranging from 0 to 1. The larger the value, the more combinations of low-interference task points there are in the candidate task subsequence, and the more likely there is unreasonable binding in the current local structure. Indicates the number of task points contained in the candidate task subsequence; Indicates the first The task point and the first Spatial load interference between task points; Indicates the first The task point and the first Interference divergence between individual task points; This represents the stability constant used in calculating the divergence degree. It is used to avoid the denominator being 0 when the candidate task subsequence length is too short. An example value is shown below. .like Then directly order .

[0141] Step S524: Calculate the task segment truncation length based on the average interferometry. Task fragment truncation length Used to control the scale of inheriting consecutive task segments from a reference frog individual, i.e. .in, This indicates the minimum task segment truncation length, used to avoid local search stagnation caused by excessively small jumps; an example value is 2. This indicates the maximum task segment truncation length, used to avoid excessive jumps that could destroy a large number of feasible structures. An example value is 15. This represents the floor operator, used to convert consecutive calculation results into executable integer segments of length.

[0142] When there are many low-interference task points within the candidate task subsequence, the average interference divergence degree is... Larger, task segment truncation length Approaching This is to inherit longer, higher-quality segments from reference frog individuals and break the original irrational structure; when there are many high-interference task points within the candidate task subsequence, the average interference divergence degree is... Smaller, task segment truncation length Approaching This is to preserve the existing tight structure and make only minor modifications.

[0143] Step S525: Based on the global learning probability threshold Select a reference frog. Generate a random number between 0 and 1. If the random number is less than 1... Then select the frog individual with the highest global fitness. As a reference frog individual; if the random number is not less than Then select the frog individual with the highest fitness within the meme group. This is used as a reference frog individual. It should be noted that the global learning probability threshold... The global interference balance index is dynamically updated in step S53, and the initial value can be 0.5.

[0144] Step S526: Select consecutive reference task segments from the reference frog individuals and merge them into the frog individual to be updated. Here, a consecutive reference task segment refers to a segment of length [not specified] within the access sequence of a certain warehouse mobile robot in the reference frog individual. A sequence of consecutive tasks.

[0145] In practical implementation, firstly, select continuous task segments with high internal average spatial load interference from the reference frog individuals to give the inherited segments strong local structural value; then, from the frog individuals to be updated... To avoid task duplication, the task points contained in consecutive reference task fragments are removed from the entire access sequence. Then, the consecutive reference task fragments are inserted into the access sequence of the target warehouse mobile robot. The target warehouse mobile robot can preferentially select the warehouse mobile robot to which the consecutive reference task fragment originally belonged in the reference frog individual; if the capacity of the warehouse mobile robot is insufficient, the warehouse mobile robot with a higher fuzzy membership degree and feasible capacity is selected.

[0146] If the target warehouse mobile robot's capacity exceeds the limit after insertion, task points with low fuzzy membership or those causing significant path detours are removed from the target access sequence, and these removed task points are reassigned to warehouse mobile robots with sufficient capacity. If missing task points occur after insertion, they are replenished using a random allocation method based on capacity constraints. Finally, nearest neighbor repair or... -opt performs local repair, restoring the access order to meet task uniqueness, capacity limits, and path consistency.

[0147] Step S527: Calculate the fitness score of the candidate frog for update. If the fitness score of the candidate frog for update is higher than that of the frog to be updated... Then replace with candidate updated frog individuals. If the candidate updated frog does not improve its fitness score, switch to the reference frog and re-execute discrete jump updates; if consecutive If the fitness score is not improved in any of the updates, then a new feasible random frog individual will be generated to replace it. This is to prevent the current meme group from getting stuck in a local stagnation. This represents the threshold for the number of consecutive failures, with values ​​ranging from 3 to 5.

[0148] In one embodiment, for example, if most task points in the candidate task subsequence are spatially dispersed and have low spatial load interference, then the average interference divergence is high and the task segment truncation length is close to... The algorithm tends to replace the local structure with a longer, higher-quality segment from a reference frog. If task points within a candidate task subsequence are located in the same shelf aisle and have high spatial load interference, the task segment truncation length will be close to... The algorithm only makes minor corrections to avoid disrupting reasonable continuous access relationships.

[0149] Furthermore, after each meme group completes its local jump update, this invention re-mixes and sorts all frog individuals, calculates the average cross-robot interference degree between different warehouse mobile robot paths for the frog individual with the highest global fitness, and then adjusts the global learning probability and elite perturbation intensity based on this interference degree, so that the algorithm achieves a dynamic balance between convergence acceleration and diversity preservation. Specifically, in step S53, the shuffling and elite dynamic enhancement based on the global interference balance index includes:

[0150] Step S531: Merge all frog individuals after each meme group has completed the local jump update, re-sort them according to fitness score from high to low, and update the frog individual with the highest global fitness. Next, read the frog with the highest global fitness. The middle is assigned to the first A set of tasks for a warehouse mobile robot ,in, This indicates that the frog with the highest global fitness is the first one... A set of task points for warehouse mobile robot services.

[0151] Step S532: Calculate the frog individual with the highest global fitness. Global Interference Balance Index Global Interference Balance Index This is used to quantify the average degree of interference between different warehouse mobile robot paths in the frog individual with the highest global fitness. The lower the value, the more concentrated the high-interference task points are on the same warehouse mobile robot path, and the fewer cross-robot execution conflicts there are.

[0152] In the specific implementation, first construct a non-empty task robot set. , This refers to the set of warehouse mobile robots with a non-empty task set among the frog individuals currently possessing the highest global fitness. If Then the global interference balance index Represented as: ;

[0153] in, Indicates the second robot index, used in conjunction with the first... A cross-robot path comparison is formed among individual warehouse mobile robots; This indicates that the frog with the highest global fitness is assigned to the first frog. A set of tasks for a warehouse mobile robot; Indicates the first The number of task points in a set of tasks for a warehouse mobile robot; Indicates the first The number of task points in a set of tasks for a warehouse mobile robot. If Then let This avoids computational anomalies when there are no cross-robot comparison objects.

[0154] Step S533: Update the global learning probability threshold based on the global interference balance index. Global learning probability threshold Used to select the frog individual with the highest current global fitness when controlling discrete jump updates within the meme group. The probability of a reference frog individual is expressed as: ;

[0155] in, This represents the minimum global learning probability threshold, used to ensure that the algorithm always retains the ability to learn from globally excellent structures; an example value is 0.2. This represents the maximum global learning probability threshold, used to prevent premature homogenization of all meme groups. An example value is 0.9.

[0156] When the global interference balance index is low, it indicates that the frog with the highest global fitness has less cross-robot interference, so the probability of learning from the frog with the highest global fitness can be increased to speed up convergence. When the global interference balance index is high, it indicates that the frog with the highest global fitness still has a lot of cross-robot interference, so the global learning intensity needs to be reduced and the local exploration within the group needs to be enhanced.

[0157] Step S534: Perform elite dynamic enhancement on the frogs with the highest fitness ranking. The elite retention rate is denoted as... This value is used to determine the number of high-fitness frog individuals participating in elite dynamic enhancement, with examples ranging from 0.05 to 0.10; the number of frog individuals participating in elite dynamic enhancement can be set to... .

[0158] During the elite dynamic enhancement process, the disturbance intensity is first determined based on the global interference balance index. If the global interference balance index is high, it indicates that many high-interference task points are distributed across different warehouse mobile robot paths; in this case, cross-robot segment exchange, task migration, and intra-channel task reorganization are prioritized. If the global interference balance index is low, it indicates that the task interference structure is relatively stable; in this case, local reversal, adjacent exchange, and other operations within the same robot access sequence are prioritized. -opt shortens the path to reduce path detours.

[0159] In one implementation, as an example, cross-robot segment swapping includes: selecting task segments with higher spatial load interference from the access sequences of two different warehouse mobile robots; determining whether the task loads of both warehouse mobile robots after the swap do not exceed their respective rated maximum load capacities; if the capacity is feasible, swapping the two task segments and performing nearest neighbor repair; if the capacity is not feasible, attempting to migrate only the task point with the lower load and a fuzzy membership degree more suitable for the target robot. After each perturbation, a task uniqueness check, a capacity constraint check, and an access sequence repair are performed to ensure that the perturbation candidate frog individuals are still executable solutions.

[0160] Step S535: Use dynamic acceptance temperature and the Metropolis criterion to determine whether the disturbed candidate frog individuals are accepted. Specifically, the... The dynamic receiving temperature is denoted as ,Right now .in, This represents the generation index, used to record the current shuffle evolution round; This represents the initial acceptance temperature, used to control the exploration intensity in the early stages of evolution; an example value is 1.0. This represents the temperature decay coefficient, used to control the rate at which the dynamic acceptance temperature decreases with each generation. An example value is 0.95.

[0161] The incremental cost of perturbing candidate frog individuals relative to the original elite frog individuals is denoted as . ,Right now .in, This indicates the disturbed candidate frog individuals; This refers to the original elite frog individual; This represents the incremental cost of perturbing the candidate frog individual relative to the original elite frog individual.

[0162] like This indicates that the disturbed candidate frog is not inferior to the original elite frog, and the disturbed candidate frog is directly accepted; if This indicates that the disturbed candidate frogs have deteriorated to some extent, so they are accepted according to the probability. The decision to accept or reject is expressed as follows: ;

[0163] in, This represents the acceptance probability of the perturbation candidate frog individual; This represents the temperature stability constant, used to prevent the denominator from being zero when the dynamic acceptance temperature is too low. An example value is shown below. By dynamically accepting the probability of accepting deteriorating candidate frog individuals based on temperature control, the ability to escape local optima can be preserved in the early stages of evolution, and convergence stability can be gradually enhanced in the later stages of evolution.

[0164] Step S536: Determine whether the maximum number of shuffles or the convergence condition has been met. The maximum number of shuffles is denoted as... , is used to limit the termination condition of the hybrid frog jumping algorithm, with values ​​ranging from 50 to 200.

[0165] In practical implementation, if the current generation... Not achieved If the cost reduction of the optimal solution for several consecutive generations does not fall below the preset convergence threshold, then return to step S202 to continue executing the discrete jump update within the meme group; if the current evolutionary generation... achieve , or continuous The cost reduction of the optimal solution is less than the convergence threshold. Then output the frog with the highest global fitness. This serves as the final solution for multi-robot task allocation and multi-point access paths in warehousing.

[0166] in, This represents a consecutive algebraic number that stops the evaluation early, with values ​​ranging from 10 to 20. This represents the cost convergence threshold of the solution, with examples of possible values. .

[0167] In one embodiment, for example, when multiple high-interference task points of the frog individual with the highest global fitness are dispersed to different warehouse mobile robot paths, the global interference balance index increases. The algorithm lowers the global learning probability threshold and simultaneously performs strong cross-robot fragment exchange and task reorganization on elite frog individuals, giving the dispersed task points a chance to recombine into the same warehouse mobile robot path. When the global interference balance index decreases and remains stable, the algorithm increases the global learning probability threshold, allowing more meme groups to inherit the current high-quality structure, thereby accelerating convergence.

[0168] In one embodiment, such as Figure 6 As shown, Figure 6The diagram shows the stacked area of ​​the cost components of the solution, which decomposes the changes of each cost component of the globally optimal solution during the evolution process in the form of a stacked area diagram. The horizontal axis represents the evolution generation, and the vertical axis represents the cost components of the solution, all of which are dimensionless cost indices. Figure 6 It includes four components: path travel cost, capacity violation cost, cross-robot interference cost, and access sequence disruption cost. The multi-objective weighted cost design guides the algorithm to prioritize fixing hard constraints and highly interfering dispersion problems, reflecting the different contributions of various cost terms at different optimization stages.

[0169] Furthermore, as a preferred embodiment, after completing the interference-sensing meme evolution optimization, the present invention selects the frog individual with the highest current global fitness. The process involves decoding the scheduling instructions that the warehouse mobile robot can execute, and performing rolling updates during online operation based on new tasks, robot status changes, or temporary channel blockages. The specific steps are as follows:

[0170] Step S61: Read the frog individual with the highest global fitness. The data is then decoded into a warehouse multi-robot task allocation table and a multi-point access path table. The warehouse multi-robot task allocation table records the set of task points, total task load, and remaining capacity for each warehouse mobile robot; the multi-point access path table records the starting position, task point access order, travel path between task points, and ending position for each warehouse mobile robot.

[0171] In specific implementation, for the first Each warehouse mobile robot outputs its final task set. and final access order .in, Indicates the first in the final solution A set of task points served by a warehouse mobile robot; Indicates the first in the final solution The task access order of each warehouse mobile robot. If If it is empty, then the first Each warehouse mobile robot remains on standby or performs other system tasks within the currently available task batch.

[0172] Step S62: Perform constraint verification on the final solution. Constraint verification includes task uniqueness verification, capacity verification, path traversability verification, and cross-robot interference verification. Specifically, task uniqueness verification confirms that each task point is assigned to only one warehouse mobile robot; capacity verification confirms that the total task load corresponding to each warehouse mobile robot does not exceed the rated maximum load capacity; and path traversability verification confirms that each path segment exists in the warehouse aisle topology map. The cross-robot interference verification is used to confirm that high spatial load interference task points are not widely dispersed across different warehouse mobile robot paths.

[0173] If the review finds that a constraint is not met, a partial repair is performed on the abnormal task point. Partial repair can be accomplished through task migration, rearranging the access order within the same machine, exchanging tasks across robots, or postponing the abnormal task point to the next batch of executable tasks. After partial repair, the cost of the solution is recalculated. It will then output the repaired executable solution.

[0174] Step S63: Send the final approved solution to the robot scheduling system. The robot scheduling system will then process the solutions according to the final access order. For the first Each warehouse mobile robot generates an underlying navigation path and, combined with the occupancy status of warehouse aisles, generates a passage reservation table. The passage reservation table is a scheduling table that records the estimated time the robot will occupy at key aisles, intersections, and storage location entrances, used to reduce the risks of robot encounters, congestion, and waiting at intersections.

[0175] In practical implementation, the arrival time of each task point can be estimated based on the rated travel speed, turning speed, and loading / unloading dwell time of the warehouse mobile robot. If the estimated occupancy time of two warehouse mobile robots in the same channel segment or intersection conflicts, the waiting time of the path with lower cost or the local detour path will be adjusted first, without changing the already determined task allocation relationship; if the waiting time exceeds the preset threshold, the rolling update in step S564 will be triggered.

[0176] Step S64: During online operation, when a new task is added, a task is canceled, the robot's battery is low, the robot malfunctions, the channel is temporarily blocked, or the loading / unloading at a task point fails, the completed tasks and currently executing short-term tasks are frozen, and only the remaining tasks are subjected to rolling optimization. The set of frozen tasks is denoted as... , This refers to the set of tasks that have been completed or whose execution objects should not be changed in the short term; the remaining set of tasks is denoted as... , This refers to a set of tasks that still need to be re-optimized in terms of allocation and access order.

[0177] During rolling optimization, the current position of each warehouse mobile robot is used as the new initial position. The weight of the loaded materials is deducted from the corresponding rated maximum load capacity to obtain the new remaining capacity. Then, the remaining task set is optimized. Repeat steps S2 through S5. By optimizing only the remaining tasks without frequently disturbing already executed tasks, it is possible to adapt to dynamic changes in the warehouse while maintaining the stability of on-site execution.

[0178] Step S65: After all task points have been visited and materials have been handed over, the warehouse management system records the actual travel distance, actual completion time, robot waiting time, number of channel conflicts, and number of task reassignments for this batch of executable tasks, and uses the above results as the basis for subsequent parameter adjustments.

[0179] In one implementation, if the actual number of channel conflicts remains high over a long period, the weight of the cross-robot interference cost is increased. or spatial proximity threshold If the actual driving distance is high but there are few lane conflicts, then increase the weight of the route driving cost. Or reduce the overall probability of group allocation If the capacity utilization rate is low, the load adaptation adjustment coefficient should be appropriately increased. This allows for more efficient use of high-capacity warehouse mobile robots in task allocation.

[0180] This embodiment also provides a warehouse multi-point addressing task allocation system based on an improved hybrid frog-leaping algorithm, used to implement the warehouse multi-point addressing task allocation method based on the improved hybrid frog-leaping algorithm described above, including:

[0181] The data acquisition module is used to acquire warehouse multi-point addressing task data, warehouse aisle topology data, and warehouse mobile robot status data, and to establish a warehouse multi-point addressing task dataset for optimization calculations; the warehouse multi-point addressing task data includes task point coordinates and task load requirements, and the warehouse mobile robot status data includes the robot's initial position and rated maximum load capacity.

[0182] The fuzzy matrix calculation module is used to perform fuzzy pre-allocation based on capacity-weighted adaptation distance according to the warehouse multi-point addressing task dataset, and obtain a fuzzy membership matrix that reflects the adaptation relationship between task points and warehouse mobile robots.

[0183] The spatial interference matrix construction module is used to construct a spatial load interference matrix based on the fuzzy membership matrix, task point coordinates, task load requirements, and rated maximum load capacity. The spatial load interference matrix is ​​used to characterize the necessity and feasibility of any two task points being continuously served by the same warehouse mobile robot.

[0184] The task density clustering analysis module is used to construct a task point interferogram based on the spatial load interferometry matrix, and to perform cluster mining on the task point interferogram to obtain task density clusters used to maintain the local task association structure.

[0185] The optimized output module is used to generate a hybrid frog-jumping algorithm initial population based on the fuzzy membership matrix, spatial load interference matrix, and task density grouping. Through interference-sensing discrete jump update and global dynamic enhancement, the multi-robot task allocation scheme is iteratively optimized to generate the optimal task allocation result and multi-point access path that can simultaneously meet the capacity constraint, task uniqueness constraint, and low interference cooperation requirements.

[0186] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A warehouse multi-point addressing task allocation method based on an improved hybrid frog-jumping algorithm, characterized in that, include: Acquire warehouse multi-point addressing task data, warehouse aisle topology data, and warehouse mobile robot status data to establish a warehouse multi-point addressing task dataset for optimization calculations; the warehouse multi-point addressing task data includes task point coordinates and task load requirements, and the warehouse mobile robot status data includes the robot's initial position and rated maximum load capacity; Based on the warehouse multi-point addressing task dataset, a fuzzy pre-assignment based on capacity-weighted adaptation distance is performed to obtain a fuzzy membership matrix that reflects the adaptation relationship between task points and warehouse mobile robots. Based on the fuzzy membership matrix, task point coordinates, task load requirements, and rated maximum load capacity, a spatial load interference matrix is ​​constructed. The spatial load interference matrix is ​​used to characterize the necessity and feasibility of any two task points being continuously served by the same warehouse mobile robot. Based on the spatial load interferometry matrix, a task point interferogram is constructed, and the task point interferogram is subjected to cluster mining to obtain task density clusters used to maintain the local task association structure. Based on the fuzzy membership matrix, spatial load interference matrix, and task density grouping, a hybrid frog-jumping algorithm is generated to initialize the population. Through interference-sensing discrete jump updates and global dynamic enhancement, the multi-robot task allocation scheme is iteratively optimized to generate the optimal task allocation result and multi-point access path that can simultaneously meet the capacity constraints, task uniqueness constraints, and low-interference collaboration requirements.

2. The warehouse multi-point addressing task allocation method based on the improved hybrid frog-leaping algorithm according to claim 1, characterized in that, Based on the aforementioned warehouse multi-point addressing task dataset, fuzzy pre-assignment based on capacity-weighted adaptation distance is performed to obtain a fuzzy membership matrix reflecting the adaptation relationship between task points and warehouse mobile robots, including: Using the initial position of each warehouse mobile robot as the initial coordinate of the corresponding cluster center, the normalized travel distance is calculated based on the task point coordinates and warehouse aisle topology data. The normalized capacity is calculated based on the rated maximum load capacity of the warehouse mobile robot; The capacity-weighted adaptation distance is determined based on the normalized travel distance, normalized capacity, and task load requirements of each task point; The fuzzy membership degree of each warehouse mobile robot corresponding to each task point is calculated based on the capacity-weighted adaptation distance. The cluster centers are iteratively updated based on the fuzzy membership degrees until the convergence condition is met, thus obtaining the fuzzy membership degree matrix.

3. The warehouse multi-point addressing task allocation method based on the improved hybrid frog-leaping algorithm according to claim 2, characterized in that, The step of iteratively updating the cluster centers based on the fuzzy membership degrees until the convergence condition is met, to obtain the fuzzy membership degree matrix, includes: In each iteration, the coordinates of each task point are weighted and averaged based on the fuzzy membership degree of the current iteration to update the position of each cluster center. The weight of each task point is a power function of the fuzzy membership degree of the task point to the robot corresponding to the current cluster center. The higher the membership degree of the task point, the greater its contribution to the update of the cluster center position. After each cluster center update, the capacity-weighted adaptation distance of each task point to each robot is recalculated, and the fuzzy membership matrix is ​​updated accordingly. Calculate the maximum change in the fuzzy membership matrix between two consecutive iterations. When the maximum change is less than a preset convergence threshold, stop the iteration and output the current fuzzy membership matrix as the final fuzzy membership matrix; or when the number of iterations reaches a preset maximum number of iterations, stop the iteration and output the current fuzzy membership matrix.

4. The warehouse multi-point addressing task allocation method based on the improved hybrid frog-leaping algorithm according to claim 1, characterized in that, The construction of the spatial load interference matrix based on the fuzzy membership matrix, task point coordinates, task load requirements, and rated maximum load capacity includes: Calculate the merge path increment between any two task points based on the task point coordinates, and calculate the spatial proximity weight based on the merge path increment. Calculate the membership consistency weight based on the fuzzy membership matrix; Calculate the same-machine load weight of two task points being carried by the same warehouse mobile robot based on the task load requirements and the robot's rated maximum load capacity. Based on the spatial proximity weight, the affiliation consistency weight, and the co-load weight, the spatial load interference matrix is ​​obtained.

5. The warehouse multi-point addressing task allocation method based on the improved hybrid frog-leaping algorithm according to claim 4, characterized in that, The merged path increment is determined based on the minimum value of the travel distance between the coordinates of the two task points and their respective nearest neighbor distances; The membership consistency weight is determined based on the average difference between the two task points and the fuzzy membership vectors of all robots. The same-machine carrying weight is determined based on the maximum common membership degree among robots that are compatible with both task points and have feasible capacity.

6. The warehouse multi-point addressing task allocation method based on the improved hybrid frog-leaping algorithm according to claim 4, characterized in that, Based on the spatial load interferometry matrix, a task point interferogram is constructed, and cluster mining is performed on the task point interferogram to obtain task density clusters used to maintain the local task correlation structure, including: Construct a task point interference graph using task points as graph nodes and spatial load interference degree as edge weights; Clustering is performed on the task point interferogram to ensure that task points within the same task density cluster are spatially adjacent, have consistent affiliation tendencies, and are feasible to be carried on the same machine, thus obtaining the density clusters of each task. Capacity feasibility verification is performed on the clustering of each task. After verification, the anchoring task is determined based on the sum of internal interference within each task cluster.

7. The warehouse multi-point addressing task allocation method based on the improved hybrid frog-leaping algorithm according to claim 6, characterized in that, Clustering is performed on the task point interferogram to ensure that task points within the same task density group are spatially adjacent, have consistent affiliation tendencies, and are feasible for co-operation on the same aircraft, including: Calculate the normalized Laplacian matrix of the task point interferogram, extract the first K feature vectors from the matrix and perform clustering to obtain K task density clusters.

8. The warehouse multi-point addressing task allocation method based on the improved hybrid frog-leaping algorithm according to claim 1, characterized in that, The interferometric sensing discrete jump update includes: In the process of the hybrid frog jumping algorithm, the current population at each iteration is divided into several meme groups. Within each meme group, continuous task segments are extracted from the access order of the frog individuals to be updated as candidate task subsequences. The average interference divergence degree within the candidate task subsequence is calculated based on the spatial load interference degree matrix. The average interference divergence degree is used to measure the proportion of low interference degree and unreasonable binding between task points in the candidate task subsequence. The task segment truncation length is dynamically determined based on the average interference divergence degree. The task segment truncation length is positively correlated with the average interference divergence degree. It is used to increase the truncation length when there are many low interference task points in the candidate task subsequence in order to break the unreasonable structure, and to decrease the truncation length when there are many high interference task points in the candidate task subsequence in order to retain the compact structure. Based on the task segment truncation length, the corresponding continuous task segments are inherited from the best frog individual within the meme group or the global best frog individual, and these continuous task segments are merged into the access order of the frog individual to be updated.

9. The warehouse multi-point addressing task allocation method based on the improved hybrid frog-leaping algorithm according to claim 8, characterized in that, The global dynamic enhancement includes: After each iteration completes the meme group update and re-washes all frog individuals, the global interference balance index of the current global best frog individual is calculated. The global interference balance index is used to quantify the average degree of cross-robot interference between different warehouse mobile robot paths in this scheme. The global learning probability threshold is dynamically adjusted according to the global interference balance index. The global learning probability threshold is used to control the probability of selecting the globally optimal individual as the reference individual when the meme group is updated discretely. Based on the global interference balance index, adaptive perturbation is performed on elite frog individuals. When the global interference balance index is high, cross-robot task segment exchange is prioritized to reorganize dispersed high-interference task points. When the global interference balance index is low, local sequence adjustment within the same robot path is prioritized to shorten the travel distance.

10. A warehouse multi-point addressing task allocation system based on an improved hybrid frog-jumping algorithm, characterized in that, For implementing the method as described in any one of claims 1-9, comprising: The data acquisition module is used to acquire warehouse multi-point addressing task data, warehouse aisle topology data, and warehouse mobile robot status data, and to establish a warehouse multi-point addressing task dataset for optimization calculations; the warehouse multi-point addressing task data includes task point coordinates and task load requirements, and the warehouse mobile robot status data includes the robot's initial position and rated maximum load capacity. The fuzzy matrix calculation module is used to perform fuzzy pre-allocation based on capacity-weighted adaptation distance according to the warehouse multi-point addressing task dataset, and obtain a fuzzy membership matrix that reflects the adaptation relationship between task points and warehouse mobile robots. The spatial interference matrix construction module is used to construct a spatial load interference matrix based on the fuzzy membership matrix, task point coordinates, task load requirements, and rated maximum load capacity. The spatial load interference matrix is ​​used to characterize the necessity and feasibility of any two task points being continuously served by the same warehouse mobile robot. The task density clustering analysis module is used to construct a task point interferogram based on the spatial load interferometry matrix, and to perform cluster mining on the task point interferogram to obtain task density clusters used to maintain the local task association structure. The optimized output module is used to generate a hybrid frog-jumping algorithm initial population based on the fuzzy membership matrix, spatial load interference matrix, and task density grouping. Through interference-sensing discrete jump update and global dynamic enhancement, the multi-robot task allocation scheme is iteratively optimized to generate the optimal task allocation result and multi-point access path that can simultaneously meet the capacity constraint, task uniqueness constraint, and low interference cooperation requirements.