A multi-agv cooperative intelligent warehouse scheduling management method and system

Abstract

This application relates to an intelligent warehouse scheduling and management method and system for multi-AGV collaboration. It includes: constructing a digital twin model based on the Unity engine, comprising shelves, multi-position AGVs, and conveyors; constructing a joint optimization model for storage location allocation and AGV path planning, with the objective of minimizing the maximum completion time for all AGVs to complete all transportation tasks; employing a multi-population adaptive simulated annealing genetic algorithm for storage location pre-allocation, and a warehouse-transportation coupled-driven adaptive large neighborhood search algorithm for path planning; adjusting storage location weights through a collaborative iterative mechanism until convergence conditions are met; loading the collaborative optimization scheme into the digital twin model for simulation verification before execution; collecting dynamic events in real time during operation, with local conflicts resolved by AGV autonomous obstacle avoidance and replanning, and global disturbances triggering a centralized coordination layer for task redistribution.

Description

Technical Field

[0001] This application relates to the field of intelligent warehousing technology, and in particular to an intelligent warehousing scheduling and management method and system for multi-AGV collaborative operation. Background Technology

[0002] With the rapid development of e-commerce and the deepening of the flexible transformation of the manufacturing industry, traditional warehousing systems, due to their low efficiency of manual operation, insufficient space utilization, and slow response speed, are no longer able to meet the high-efficiency demands of modern logistics. Against this backdrop, intelligent warehousing systems centered on automated guided vehicles (AGVs) are gradually becoming the mainstream direction for industry upgrades. Among them, multi-position AGVs, with their multi-position parallel transportation capabilities, demonstrate significant advantages in high-density warehousing scenarios.

[0003] Currently, the main method for scheduling and controlling AGVs in warehousing adopts a rule-based fixed scheduling strategy: goods are randomly allocated to available storage locations according to the "first-in, first-out" principle based on the order of goods entering the warehouse; when each transportation task is triggered, a nearest neighbor greedy algorithm is used to plan the path for the AGV, that is, selecting the available AGV closest to the current task point to execute the task, generating scheduling instructions with the goal of minimizing the response time of a single task. This method manages the AGV task queue uniformly through a central control server, and the AGVs only act as execution terminals, traveling according to the preset path, and do not have autonomous decision-making capabilities.

[0004] The aforementioned technology uses a first-in-first-out random storage location allocation method, which does not take into account the differences in cargo turnover rates. High-frequency goods may be placed in high-rise or remote storage locations, resulting in a significant increase in AGV storage and retrieval time.

[0005] Based on this, this application provides a method and system for intelligent warehouse scheduling and management of multi-AGV collaboration. Summary of the Invention

[0006] To address the issue that the first-in-first-out random storage location allocation method does not consider the differences in cargo turnover rates, and high-frequency goods may be placed in high-rise or remote storage locations, resulting in a significant increase in AGV access time, this application provides a smart warehouse scheduling and management method and system for multi-AGV collaboration.

[0007] Firstly, this application provides an intelligent warehouse scheduling and management method for multi-AGV collaboration, employing the following technical solution: including: Obtain the physical structure parameters and equipment motion parameters of the warehousing system, and build a digital twin model containing shelves, multi-position AGVs and conveyors based on the Unity engine; A joint optimization model for storage location allocation and AGV path planning is constructed with the goal of minimizing the maximum completion time of all AGVs and all transportation tasks. The model sets constraints such as storage location uniqueness, AGV carrying capacity, multi-vehicle yard coordination, and rack center of gravity stability. Obtain the list of goods to be put into storage, use simulated annealing genetic algorithm to pre-allocate storage locations and generate a storage location allocation scheme; use the storage location allocation scheme as the transportation target, use warehousing-transportation coupled large neighborhood search algorithm to perform path planning, generate an initial path planning scheme, and calculate the maximum completion time of the plan. If the maximum completion time of the plan exceeds a preset threshold, adjust the storage location allocation weight and re-optimize until the convergence condition is met to obtain a collaborative optimization scheme. The collaborative optimization scheme is loaded into the digital twin model for simulation. If the throughput indicator meets the preset target, the collaborative optimization scheme is sent to the actual warehousing system for execution; otherwise, the parameters are adjusted and the collaborative optimization scheme is regenerated. During the execution of tasks by AGVs, dynamic event information is collected in real time. When a dynamic event is detected, a rescheduling strategy is executed: for local conflict events, the distributed execution layer performs autonomous obstacle avoidance and path replanning through information interaction between AGVs; for global disturbance events, the centralized coordination layer is triggered to perform local task reallocation and path replanning.

[0008] Preferably, the acquisition of the physical structure parameters and equipment motion parameters of the warehousing system, and the construction of a digital twin model including shelves, multi-position AGVs, and conveyors based on the Unity engine, includes: Obtain the physical structure parameters and equipment motion parameters of the warehousing system. The physical structure parameters include the number of rows, columns, and layers of shelves, the depth, width, and height of the storage cell, the length of the conveyor, and the spacing between goods. The equipment motion parameters include the moving speed of the AGV on the ground, the translational and rotational speed of the AGV robot arm, and the conveyor conveying speed. Three-dimensional geometric models of shelves, multi-position AGVs and conveyors are constructed using 3D modeling software, and then imported into the Unity engine. The models are then assembled in the Unity scene according to the actual layout to form a virtual warehousing environment. Configure the physics engine component for the 3D geometry model in the Unity engine, add collision body components for the shelves and conveyors, add rigid body components and navigation proxy components for the AGV, and set the kinematic parameters, collision properties and walkable area of ​​the navigation mesh for the AGV. The script implements the cargo transportation logic of the conveyor, including the generation of cargo at the beginning of the conveyor at set intervals, the conveyor belt driving the movement of cargo, and the end trigger controlling the stopping and starting of cargo; at the same time, it implements the cargo loading and unloading logic of the AGV, including the extension and retraction of the robotic arm, the grabbing and releasing of cargo, the binding and unbinding of cargo in the AGV storage position, and the autonomous path planning and motion control of the AGV based on the navigation grid, generating an executable simulation verification digital twin model.

[0009] Preferably, the construction of the joint optimization model for storage location allocation and AGV path planning includes: Set up a set of storage locations, a set of task points, and a set of AGVs. Define a first decision variable to indicate whether the k-th AGV travels from node i to node j. Define a second decision variable to indicate whether storage location i is served by the k-th AGV. Set the carrying capacity of each AGV, the total number of goods, the change in goods at each node, and the three-dimensional coordinates of each node. Set the height of the task point to zero. The optimization objective is to minimize the maximum completion time for all AGVs to complete all transportation tasks. The maximum completion time is equal to the maximum transportation time among all AGVs. The transportation time of each AGV consists of the sum of its horizontal travel time and vertical lifting time on its travel path. The horizontal travel time is equal to the Manhattan distance divided by the AGV's ground movement speed, and the vertical lifting time is equal to the target warehouse floor height multiplied by the cell height and then divided by the robot arm's lifting speed. Set uniqueness constraints for storage locations, ensuring that each storage location is served by only one AGV; set AGV carrying capacity constraints, ensuring that the number of goods loaded by each AGV at any node does not exceed its carrying capacity; set multi-vehicle-farm coordination constraints, ensuring that each AGV must start from a certain task point and eventually return to the task point; set rack center of gravity stability constraints, ensuring that the overall center of gravity height of all goods does not exceed a preset safety threshold, and the overall center of gravity height is calculated by the weighted average of the weight of each goods and the height of its storage location. By combining the objective function with various constraints, a joint optimization model is formed, with the storage location allocation result and the AGV path as joint decision variables.

[0010] Preferably, the step of obtaining the list of goods to be received and using a simulated annealing genetic algorithm to pre-allocate storage locations and generate a storage location allocation scheme includes: The warehouse management system reads the information of goods to be put into storage. The list of goods to be put into storage includes the type identifier, turnover rate, weight and number of boxes of goods to be put into storage for each type of goods. The turnover rate is used to characterize the frequency of goods leaving the warehouse. The higher the turnover rate, the greater the demand for the goods to be accessed at a high frequency. With the optimization objectives of goods entry and exit efficiency, clustering of similar goods, and stability of shelf center of gravity, the multi-objective problem is transformed into a single-objective optimization problem using the linear weighted method. The objective function of storage location allocation is established. The entry and exit efficiency objective is achieved by minimizing the weighted sum of the turnover rate and storage and retrieval time of all goods. The clustering of similar goods objective is achieved by minimizing the spatial dispersion of similar goods in the warehouse. The shelf center of gravity stability objective is achieved by minimizing the overall center of gravity height of all goods. Using sequential integer encoding, all storage locations in the warehouse are numbered one-dimensionally according to row, column, and layer order. Each item to be stored corresponds to a storage location number, and the sequence of storage location numbers for all items constitutes a chromosome. Multiple subpopulations are set up, and each subpopulation contains a preset number of chromosomes as the initial population. Multiple subpopulations are controlled to evolve independently in parallel. Within each subpopulation, a tournament selection strategy is used to select parent individuals. The crossover probability and mutation probability are adaptively adjusted based on the individual fitness. The crossover probability and mutation probability change non-linearly in a Sigmoid manner with the difference between the individual fitness and the average fitness of the population. Every preset number of generations, the elite individuals in the current subpopulation are perturbed. Several consecutive gene loci on the chromosome are randomly selected, their library codes are randomly shuffled, and then refilled to generate a new solution. If the objective function value of the new solution is better than that of the original elite individual, the new solution is directly accepted to replace the original elite individual. If the objective function value of the new solution is worse than that of the original elite individual, the new solution is accepted with a preset probability to replace the individual with the lowest fitness in the population. The preset probability decreases as the number of iterations increases. Every preset number of generations, each subpopulation exchanges a preset proportion of high-quality individuals with its neighboring subpopulations according to a circular topology; the multi-population parallel evolution and new solution generation steps are repeated until the maximum number of iterations is reached, and the chromosome with the optimal objective function value is selected from all subpopulations as the storage allocation scheme.

[0011] Preferably, the step of using the storage location allocation scheme as the transportation target and employing a large neighborhood search algorithm driven by warehousing-transportation coupling for path planning to generate an initial path planning scheme includes: Receive the storage location allocation scheme, take the target storage location of the goods to be put into storage as the customer point set, take the loading point as the vehicle yard set, set the AGV quantity, carrying capacity and movement speed parameters, initialize the damage operator set and repair operator set and assign the same initial weight. A greedy random adaptive search algorithm is used to construct an initial path scheme. All unassigned customer points are traversed, and the transportation time cost between each customer point and each parking lot is calculated. The transportation time cost includes the Manhattan distance travel time and the robotic arm lifting service time. A restricted candidate list is constructed based on preset randomization parameters, and a customer point-parking lot pair is randomly selected from the list for insertion to form an initial path scheme. The roulette wheel strategy is used to select a destruction operator based on the operator weight, which removes a preset number of customer points from the current path; then the repair operator is selected to re-insert the removed customer points into feasible positions to form a new path scheme. Calculate the maximum completion time of the new path scheme, and assign different levels of reward scores to the corresponding operators based on the comparison results of the new path scheme with the historical best path scheme and the current path scheme; update the weight of each operator according to the cumulative reward score every preset number of iterations. The record-oriented migration criterion is used to determine whether to accept a poor path scheme where the maximum completion time of the new path scheme is greater than that of the current path scheme. The ratio of the maximum completion time of the new path scheme to the historical best path scheme is calculated. If it is less than a preset threshold, the new path scheme is accepted. The iteration is repeated until the maximum number of iterations is reached to obtain the initial path planning scheme.

[0012] Preferably, the maximum completion time of the calculation plan is exceeded. If the maximum completion time exceeds a preset threshold, the storage location allocation weights are adjusted and the plan is re-optimized until the convergence condition is met, resulting in a collaborative optimization scheme, including: Based on the initial path planning scheme, the path information of each AGV is extracted, the transportation time required for each AGV to complete the assigned task is calculated, and the maximum value of the transportation time of all AGVs is taken as the maximum completion time of the plan. The maximum completion time of the plan is compared with a preset threshold, which is determined based on the throughput index; if the maximum completion time of the plan is not greater than the preset threshold, it is determined that the current storage location allocation scheme and the current path planning scheme meet the requirements of collaborative optimization. If the maximum completion time of the plan is greater than the preset threshold, the bottleneck factors in the current path planning scheme that cause the maximum completion time to be too long are analyzed. Bottleneck factors include excessive load on a single AGV, overly scattered distribution of storage locations in a specific area, or excessive distance between storage locations for high-frequency goods. Based on the type of bottleneck factor, the target weight of storage location allocation optimization is adjusted, and the weight value of optimization targets related to bottleneck factors is increased. Based on the adjusted target weights, a new storage location allocation scheme is generated; the new storage location allocation scheme is then used as the transportation target to generate a new route planning scheme. The generation of new path planning schemes is repeatedly iterated, and the maximum completion time of the plan is recorded after each iteration. When the decrease in the maximum completion time of the plan is less than a preset percentage in a series of preset iterations, convergence is determined, and the collaborative optimization scheme is obtained.

[0013] Preferably, during the AGV's task execution, dynamic event information is collected in real time; when a dynamic event is detected, a rescheduling strategy is executed, including: During the AGV's task execution, environmental information, task status information, and AGV power and health status information are collected in real time through the sensors on the AGV. The collected dynamic events are divided into local conflict events and global disturbance events according to their scope of influence. When a local conflict event is detected, the distributed execution layer of the AGV involved initiates an autonomous coordination mechanism. The AGVs involved exchange information on their current position, target storage location, remaining tasks, and current path. Each AGV determines the passage order based on the exchanged information using priority avoidance rules. AGVs with lower priority perform local path replanning and replan a detour path between their current position and the target storage location. When a global disturbance event is detected, the affected equipment reports the event information to the centralized coordination layer. The centralized coordination layer assesses the scope of the event's impact, determines the list of affected AGVs and the list of unexecuted tasks, suspends the current tasks of the affected AGVs, performs local reallocation of unexecuted tasks, and reassigns tasks affected by the faulty equipment to idle or low-load AGVs. It also performs path planning on the reassigned tasks, generates and executes a new local path planning scheme. After the rescheduling is completed, the updated task allocation information and path planning information will be synchronized to the centralized coordination layer, and the new path plan will be issued to the corresponding AGV; the AGV will continue to execute the task according to the new path plan.

[0014] Secondly, this application discloses an intelligent warehouse scheduling and management device for multi-AGV collaboration, which adopts the following technical solution, including: The 3D model module is used to obtain the physical structure parameters and equipment motion parameters of the warehousing system, and to build a digital twin model of the shelving, multi-position AGV and conveyor based on the Unity engine; The joint optimization module is used to build a joint optimization model for storage location allocation and AGV path planning. The goal is to minimize the maximum completion time of all AGVs and complete all transportation tasks. It sets constraints such as storage location uniqueness, AGV carrying capacity, multi-vehicle yard coordination, and rack center of gravity stability. The collaborative optimization module is used to obtain a list of goods to be put into storage, use simulated annealing genetic algorithm to pre-allocate storage locations, and generate a storage location allocation scheme. The storage location allocation scheme is used as the transportation target, and a large neighborhood search algorithm driven by warehousing-transportation coupling is used for path planning to generate an initial path planning scheme. The maximum completion time of the planning is calculated. If the maximum completion time of the planning exceeds a preset threshold, the storage location allocation weight is adjusted and re-optimized until the convergence condition is met, and a collaborative optimization scheme is obtained. The simulation execution module is used to load the collaborative optimization scheme into the digital twin model for simulation execution. If the throughput indicator meets the preset target, the collaborative optimization scheme is sent to the actual warehousing system for execution; otherwise, the parameters are adjusted and the collaborative optimization scheme is regenerated. The conflict resolution module is used to collect dynamic event information in real time during the AGV's task execution. When a dynamic event is detected, a rescheduling strategy is executed: for local conflict events, the distributed execution layer performs autonomous obstacle avoidance and path replanning through information interaction between AGVs; for global disturbance events, the centralized coordination layer is triggered to perform local task reallocation and path replanning.

[0015] Thirdly, this application also provides a control device, the device comprising: It includes a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and executed as described above for intelligent warehouse scheduling and management of multi-AGV collaboration.

[0016] Fourthly, this application also provides a computer-readable storage medium storing a computer program that can be loaded by a processor and executed as described above for the intelligent warehouse scheduling and management method of multi-AGV collaboration.

[0017] In summary, this application constructs a digital twin model based on the Unity engine, including shelves, multi-position AGVs, and conveyors, providing a simulation verification environment for subsequent optimization. Based on this, a joint optimization model for storage location allocation and AGV path planning is built, aiming to minimize the maximum completion time for all AGVs to complete all transportation tasks. Constraints such as storage location uniqueness, load capacity, multi-vehicle yard collaboration, and shelf center of gravity stability are set. A hierarchical collaborative optimization architecture is adopted for the solution: the centralized coordination layer uses a multi-population adaptive simulated annealing genetic algorithm for storage location pre-allocation, prioritizing high-frequency goods to lower-level, near-exit storage locations; the distributed execution layer uses a warehouse-transportation coupled adaptive large neighborhood search algorithm for path planning, and feeds the path planning results back to the storage location allocation layer through a collaborative iteration mechanism. If the maximum completion time exceeds a threshold, the storage location weights are adjusted and re-optimized until the convergence condition is met. Finally, the collaborative optimization scheme was loaded into the digital twin model for simulation verification. After meeting the standards, it was deployed to the actual system for execution. During operation, dynamic events were collected in real time. For local conflicts, the AGVs autonomously avoided obstacles and replanned the route. For global disturbances, the centralized coordination layer performed task reassignment and path replanning. Thus, through joint optimization of storage locations and paths, hierarchical collaborative solution, and dynamic rescheduling mechanism, the overall transportation efficiency, environmental adaptability, and equipment coordination capabilities of the warehousing system were significantly improved. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating an intelligent warehouse scheduling and management method that utilizes multiple AGVs in collaboration.

[0019] Figure 2 This is a structural block diagram of an intelligent warehouse scheduling and management device that enables multi-AGV collaboration. Detailed Implementation

[0020] The following combination Figures 1-2 This application will be described in further detail.

[0021] This application provides a multi-AGV collaborative intelligent warehouse scheduling and management method, with the control system as the execution entity. Its core concept is as follows: a high-fidelity verification environment for optimization is provided by constructing a digital twin simulation model; a joint optimization model for storage location allocation and AGV path planning is established, breaking the limitation of independent optimization of the two in traditional methods; a hierarchical collaborative optimization architecture is adopted, with a centralized coordination layer for global storage location pre-allocation and a distributed execution layer for local path planning, and a collaborative iteration mechanism to achieve bidirectional feedback and deep coupling between storage location layout and transportation path; in a dynamic environment, a hierarchical rescheduling strategy is used to achieve an organic combination of autonomous response to local conflicts and centralized scheduling of global disturbances.

[0022] Reference Figure 1 The embodiments of this application include at least steps S10 to S50.

[0023] S10: Obtain the physical structure parameters and equipment motion parameters of the warehousing system, and build a digital twin model containing shelves, multi-position AGVs and conveyors based on the Unity engine.

[0024] S20. Construct a joint optimization model for storage location allocation and AGV path planning. With the goal of minimizing the maximum completion time of all AGVs and completing all transportation tasks, set constraints such as storage location uniqueness, AGV carrying capacity, multi-vehicle yard collaboration, and rack center of gravity stability.

[0025] S30: Obtain the list of goods to be put into storage, use simulated annealing genetic algorithm to pre-allocate storage locations, and generate a storage location allocation scheme; use the storage location allocation scheme as the transportation target, use the large neighborhood search algorithm driven by warehousing-transportation coupling to perform path planning, generate an initial path planning scheme, and calculate the maximum completion time of the plan. If the maximum completion time of the plan exceeds the preset threshold, adjust the storage location allocation weight and re-optimize until the convergence condition is met, and obtain a collaborative optimization scheme.

[0026] S40: Load the collaborative optimization scheme into the digital twin model for simulation. If the throughput indicator meets the preset target, the collaborative optimization scheme is sent to the actual warehousing system for execution; otherwise, adjust the parameters and regenerate the collaborative optimization scheme.

[0027] S50 collects dynamic event information in real time during AGV task execution; when a dynamic event is detected, a rescheduling strategy is executed: for local conflict events, the distributed execution layer performs autonomous obstacle avoidance and path replanning through information interaction between AGVs; for global disturbance events, the centralized coordination layer is triggered to perform local task reassignment and path replanning.

[0028] Specifically, the system builds a warehouse digital twin model based on Unity, providing a simulation verification foundation for optimization. A joint optimization model for storage location allocation and path planning is established, aiming to minimize the maximum completion time, and setting constraints on storage location uniqueness, carrying capacity, multi-vehicle yard collaboration, and rack center of gravity stability. Then, a simulated annealing genetic algorithm is used for storage location pre-allocation, with the storage location scheme as the transportation objective. A large neighborhood search algorithm is used for path planning, and a collaborative iteration mechanism adjusts the storage location weights until convergence conditions are met. The collaborative optimization scheme is then loaded into the digital twin model for simulation verification, and after achieving the target, it is deployed to the actual system for execution. Finally, dynamic events are collected in real time during operation; local conflicts are resolved by AGV autonomous obstacle avoidance and replanning, while global disturbances trigger centralized coordination layer reassignment. Thus, through digital twin model simulation, storage location and path collaborative optimization, and hierarchical dynamic rescheduling, the transportation efficiency, environmental adaptability, and equipment collaboration capabilities of the warehouse system are significantly improved. Simultaneously, through the joint optimization model of storage location allocation and path planning, based on the differences in cargo turnover rates, high-frequency goods are placed in easily accessible storage locations, reducing AGV access time.

[0029] In some embodiments, step S10 specifically includes the following steps: obtaining the physical structure parameters and equipment motion parameters of the warehousing system. The physical structure parameters include the number of rows, columns, and layers of the shelves, the depth, width, and height of the storage cell, the length of the conveyor, and the spacing between goods. The equipment motion parameters include the moving speed of the AGV on the ground, the translational and rotational speed of the AGV robot arm, and the conveyor conveying speed. Using 3D modeling software, 3D geometric models of the shelves, multi-position AGVs, and conveyors are constructed respectively, and the 3D geometric models are imported into the Unity engine and assembled in the Unity scene according to the actual layout to form a virtual warehousing environment. Configure physics engine components for the 3D geometric model in Unity, add collision body components for the shelves and conveyors, add rigid body components and navigation proxy components for the AGV, and set the kinematic parameters, collision attributes, and walkable area of ​​the navigation mesh for the AGV; implement the cargo transportation logic of the conveyor through scripts, including generating cargo at set intervals at the beginning of the conveyor, driving the cargo movement with the conveyor belt, and controlling the stopping and starting of cargo with the end trigger; at the same time, implement the cargo loading and unloading logic of the AGV, including the extension and retraction of the robotic arm, cargo grabbing and releasing, binding and unbinding of cargo in the AGV storage position, as well as the autonomous path planning and motion control of the AGV based on the navigation mesh, and generate a digital twin model that can be executed for simulation verification.

[0030] Specifically, the system collects the physical structure parameters and equipment motion parameters of the warehousing system as the basis for modeling. It uses 3D modeling software to construct 3D geometric models of shelves, multi-position AGVs, and conveyors, imports them into the Unity engine, and assembles them according to the actual layout to form a virtual warehousing environment. It configures physics engine components for the model, adds colliders, rigid bodies, and navigation proxy components, and sets kinematic parameters and walkable areas based on the navigation mesh. Through scripts, it implements the generation, movement, and end-point start / stop control of conveyors, while simultaneously realizing the AGV robot's telescopic grasping, loading and unloading, and autonomous path planning and motion control based on the navigation mesh, generating a complete digital twin model. This provides a simulation verification platform for subsequent storage location allocation and path planning algorithms, ensuring that the optimization scheme is fully verified in the virtual environment before being deployed, greatly reducing the risk of actual deployment.

[0031] In some embodiments, step S20 specifically includes the following steps: setting a storage location set, a task point set, and an AGV set, and defining a first decision variable. Indicates whether the k-th AGV has traveled from node i to node j, where A value of 1 indicates driving, and a value of 0 indicates not driving; define a second decision variable. Indicates whether storage location i is served by the k-th AGV, where A value of 1 indicates service, and a value of 0 indicates no service; the carrying capacity l of each AGV, the total number of goods n, and the change in goods at each node are set. and the three-dimensional coordinates of each node The height of the task point is set to zero, i.e. .

[0032] The optimization objective is to minimize the maximum completion time of all AGVs in completing all transportation tasks (by optimizing storage space allocation and AGV paths, the completion time of the AGV with the longest completion time is minimized). The maximum completion time is equal to the maximum value of the transportation time among all AGVs. The specific formula for this objective function is as follows: ; in, Represents a set of AGVs. This represents the transport time for the k-th AGV to complete all its tasks. Maximum completion time. It equals the maximum transport time among all AGVs.

[0033] The transportation time of each AGV consists of the sum of its horizontal travel time and vertical lifting time along its path. The transportation time of the k-th AGV is... It can be represented as: ; in, Represents the set of edges for feasible paths. This represents the speed at which the AGV moves on the ground. The vertical lifting speed of the AGV robot arm. This represents the Manhattan distance between node i and node j. The height of the target storage location is multiplied by the cell height. This refers to the time required for the robotic arm to lift the goods to the target storage location.

[0034] Setting a unique constraint for each storage location, ensuring that each location is served by exactly one AGV, can be represented as: ; in, This represents the set of storage locations, and this constraint ensures that each storage location is served by exactly one AGV.

[0035] Set AGV carrying capacity constraints: the number of goods loaded by each AGV at any node must not exceed its carrying capacity; set multi-vehicle coordination constraints: each AGV must start from a certain task point and eventually return to the task point; set shelf center of gravity stability constraints: the overall center of gravity height of all goods must not exceed a preset safety threshold. The overall center of gravity height is calculated by the weighted average of the weight of each item and the height of its storage location, as shown in the following formula: ; in, Let be the weight of the i-th item. Let h be the layer number of the storage location where the i-th item is located, and h be the height of a single storage location cell. This is a preset safety threshold.

[0036] By combining the objective function with various constraints, a joint optimization model is formed, with the storage location allocation result and the AGV path as joint decision variables.

[0037] Specifically, the algorithm first sets up a storage location set, a task point set, and an AGV set, defining decision variables to identify AGV path selection and storage location service assignment. It then sets the carrying capacity, cargo quantity, node coordinates, and cargo variation, with the task point height set to zero. The optimization objective is to minimize the maximum completion time for all AGVs to complete all transportation tasks, minimizing the completion time of the longest-running AGV. Transportation time is composed of the horizontal travel time corresponding to the Manhattan distance and the vertical lifting time corresponding to the storage location height. Constraints are set on storage location uniqueness, AGV carrying capacity, multi-vehicle-yard coordination, and rack center of gravity stability, with the center of gravity height calculated as the weighted average of cargo weight and storage location height. Combining the objective function and constraints forms a joint optimization model with storage location allocation results and AGV paths as joint decision variables, facilitating collaborative optimization of storage locations and paths.

[0038] In some embodiments, step S30 specifically includes the following steps: reading the information of goods to be received from the warehouse management system. The list of goods to be received includes the type identifier, turnover rate, weight, and number of boxes of goods to be received for each type of goods; wherein the turnover rate is used to characterize the frequency of goods leaving the warehouse, and the higher the turnover rate, the greater the demand for the goods to be accessed frequently; taking the efficiency of goods leaving and leaving the warehouse, the clustering degree of similar goods, and the stability of the shelf center of gravity as optimization objectives, the multi-objective problem is transformed into a single-objective optimization problem using a linear weighted method, and an objective function for storage location allocation is established. The goal of efficiency of goods leaving and leaving the warehouse is achieved by minimizing the weighted sum of the turnover rate and access time of all goods, the goal of clustering of similar goods is achieved by minimizing the spatial dispersion of similar goods in the warehouse, and the goal of shelf center of gravity stability is achieved by minimizing the overall center of gravity height of all goods; the objective function for storage location allocation is as follows: ; in, These are the weighting coefficients for each objective, which are summed to 1 and can be flexibly adjusted according to management strategies. To achieve the goal of efficient inbound and outbound operations, the goal is to minimize the weighted sum of the turnover rate and retrieval time of all goods. The formula is as follows: ; in, Let w be the turnover rate of the i-th item, and w be the width of the storage location cell. Let be the column number of the storage location where the i-th item is located. This represents the horizontal distance the AGV travels from the task point to the storage location, where h is the height of the storage location cell. Let be the layer number of the storage location where the i-th item is located. The objective of clustering similar goods is achieved by minimizing the spatial dispersion of similar goods in the warehouse, as shown in the following formula: ; in, Let be the three-dimensional coordinates of the i-th item. Let be the center coordinates of the j-th type of goods, calculated from the average coordinates of all goods in that type. Let be the mapping function, which maps the i-th item to its corresponding category number.

[0039] The goal of achieving rack center of gravity stability is achieved by minimizing the overall center of gravity height of all goods, as shown in the following formula: ; in, Let be the weight of the i-th item. This formula calculates the weighted center of gravity height of all items. The lower the center of gravity, the higher the stability of the shelf.

[0040] Using sequential integer encoding, all storage locations in the warehouse are numbered one-dimensionally according to row, column, and layer order. Each item to be stored corresponds to a storage location number, and the sequence of storage location numbers for all items constitutes a chromosome. Multiple subpopulations are set up, each containing a preset number of chromosomes as the initial population. The multiple subpopulations are controlled to evolve independently in parallel, and a tournament selection strategy is used to select parent individuals within each subpopulation. The crossover probability and mutation probability are adaptively adjusted based on individual fitness. The crossover probability and mutation probability change non-linearly in a Sigmoid manner with the difference between individual fitness and the average fitness of the population, so that better individuals have a lower probability of being destroyed, while average individuals retain a higher evolutionary probability. Every preset number of generations, the elite individuals in the current subpopulation are perturbed. Several consecutive gene loci on the chromosome are randomly selected, their library codes are randomly shuffled, and then refilled to generate a new solution. If the objective function value of the new solution is better than that of the original elite individual, the new solution is directly accepted to replace the original elite individual. If the objective function value of the new solution is worse than that of the original elite individual, the new solution is accepted with a preset probability to replace the individual with the lowest fitness in the population. The preset probability decreases as the number of iterations increases. This acceptance probability adopts a simulated annealing mechanism and can be expressed as: ; in, The difference between the objective function value of the new solution and the objective function value of the original solution is represented by T, which is the current temperature and gradually decreases with the number of iterations according to the annealing coefficient.

[0041] Every preset number of generations, each subpopulation exchanges a preset proportion of high-quality individuals with its neighboring subpopulations according to a circular topology; the process of multipopulation parallel evolution and new solution generation is repeated until the maximum number of iterations is reached, and the chromosome with the optimal objective function value is selected from all subpopulations as the storage allocation scheme.

[0042] It should be understood that the new solution specifically refers to the storage allocation scheme after the disturbance.

[0043] Specifically, the system retrieves a list of goods to be received from the warehouse management system. Turnover rate is a core parameter used to quantify the frequency of goods access. A higher turnover rate means the goods are accessed more frequently, resulting in more easily accessible storage locations. The optimization objectives include three dimensions: inbound / outbound efficiency, clustering of similar goods, and shelf center of gravity stability, corresponding to the timeliness of warehousing operations, ease of management, and structural security, respectively. A linear weighted method is used to integrate these three objectives into a single objective function. By adjusting the weights, different management strategies can be flexibly adapted. For example, increasing the weight of inbound / outbound efficiency can prioritize the allocation of high-frequency goods to lower-level, near-exit storage locations.

[0044] To address the combinatorial optimization characteristics of storage location allocation, the algorithm uses sequential integer encoding to map three-dimensional storage locations into a one-dimensional sequence. Each item to be stored corresponds to a storage location number, and the allocation schemes for the entire batch of goods constitute a chromosome. To enhance global search capabilities, multiple subpopulations are set up for parallel and independent evolution, and high-quality individuals are periodically exchanged through a circular topology to avoid the population getting trapped in local optima. During the evolutionary process, the crossover and mutation probabilities are adaptively adjusted based on the difference between the individual fitness and the average fitness of the population. A sigmoid-type nonlinear curve is used to ensure that high-quality individuals are destroyed with a low probability, while average individuals maintain high exploration activity, thus balancing the algorithm's development and exploration capabilities.

[0045] To prevent premature convergence of the algorithm, a simulated annealing mechanism is introduced to locally perturb elite individuals. This randomly shuffles the storage location codes of consecutive gene loci to generate new solutions. If the new solution is superior, it is directly accepted; if inferior, it is accepted with a decreasing probability as the temperature decreases. This allows the algorithm to potentially escape local optima in later iterations. Through this multi-population parallel evolution and adaptive adjustment strategy, the final storage location allocation scheme achieves an optimal balance between access efficiency, management convenience, and structural safety, providing a reasonable transportation target for subsequent AGV path planning.

[0046] Furthermore, step S30 also includes the following steps: receiving the storage location allocation scheme, taking the target storage locations of the goods to be stored as the customer point set, taking the loading points as the vehicle yard set, setting the AGV quantity, carrying capacity, and movement speed parameters, initializing the destruction operator set and the repair operator set and assigning them the same initial weight; constructing an initial path scheme using a greedy random adaptive search algorithm, traversing all unassigned customer points, calculating the transportation time cost between each customer point and each vehicle yard, the transportation time cost including the Manhattan distance travel time and the robotic arm lifting service time; the transportation time cost between customer point i and vehicle yard j can be expressed as: ; in, The ground movement speed of the AGV. The lifting speed of the robotic arm.

[0047] According to preset randomization parameters Construct a restricted candidate list and calculate the cost threshold: ; Customer point-parking lot pairs with costs below the threshold are added to the restricted candidate list, and a random pair is selected from the list for insertion to form the initial route scheme.

[0048] A roulette wheel strategy is adopted to select a disruptive operator based on operator weights, removing a preset number of customer points from the current path; then, a repair operator is selected to reinsert the removed customer points into feasible positions, forming a new path solution; the maximum completion time of the new path solution is calculated, and different levels of reward scores are given to the corresponding operators based on the comparison results of the new path solution with the historical best path solution and the current path solution; every preset number of iterations, the weights of each operator are updated based on the accumulated reward scores. The operator weight update formula can be expressed as: ; in, Let i be the current weight of the i-th operator. The cumulative reward score for this operator, This is the preset weight increment.

[0049] The record-oriented migration criterion is used to determine whether to accept a poor path scheme where the maximum completion time of the new path scheme is greater than that of the current path scheme. The ratio of the maximum completion time of the new path scheme to the historical best path scheme is calculated. If it is less than a preset threshold, the new path scheme is accepted. The iteration is repeated until the maximum number of iterations is reached to obtain the initial path planning scheme.

[0050] Specifically, the system uses the target storage location in the storage allocation scheme as the customer point and the loading point as the parking lot, establishing the basic input for path planning. An initial solution is constructed using a greedy stochastic adaptive search algorithm. The transportation cost is calculated by combining the Manhattan distance travel time with the robotic arm lifting service time, achieving a balance between greed and randomness to generate diverse initial path schemes. Subsequently, an iterative optimization phase is entered. A roulette wheel strategy dynamically selects destructive and repair operators based on their historical performance, performing customer point removal and re-insertion operations respectively to form new path schemes. Operators are given different levels of rewards based on the comparison results of the new scheme with the historical best solution and the current solution, and their weights are updated periodically, causing the algorithm to adaptively favor efficient operator combinations. A record-oriented migration criterion is adopted to allow for a moderate acceptance of inferior solutions, preventing getting trapped in local optima. Thus, through adaptive neighborhood search and feedback mechanisms, high-quality initial path planning schemes are generated, laying the foundation for subsequent collaborative optimization of storage locations and paths.

[0051] Furthermore, step S30 also includes the following steps: based on the initial path planning scheme, extract the path information of each AGV, calculate the transportation time required for each AGV to complete the assigned task, and take the maximum value of the transportation time of all AGVs as the planned maximum completion time; compare the planned maximum completion time with a preset threshold, which is determined based on the throughput index; if the planned maximum completion time is not greater than the preset threshold, it is determined that the current storage location allocation scheme and the current path planning scheme meet the requirements of collaborative optimization; if the planned maximum completion time is greater than the preset threshold, analyze the bottleneck factors in the current path planning scheme that cause the maximum completion time to be too long, including excessive load on a single AGV, overly dispersed storage location distribution in a specific area, or excessive distance between storage locations for high-frequency goods; adjust the target weight of storage location allocation optimization according to the type of bottleneck factor, and increase the weight value of the optimization target related to the bottleneck factor; Based on the adjusted target weights, a new storage location allocation scheme is regenerated; the new storage location allocation scheme is used as the transportation target to regenerate a new route planning scheme; the generation of the new route planning scheme is iterated repeatedly, and the maximum completion time of the plan is recorded after each iteration. When the decrease in the maximum completion time of the plan is less than a preset percentage in a series of preset iterations, convergence is determined, and a collaborative optimization scheme is obtained.

[0052] Specifically, the system calculates the maximum completion time of all AGVs based on the initial path planning scheme, serving as a key indicator of system efficiency. If this time exceeds a preset threshold, the system analyzes the root cause of the bottleneck to identify whether it's an AGV overloaded, a scattered distribution of storage locations, or excessively long distances for high-frequency goods. The target weights in the storage location allocation optimization model are dynamically adjusted according to the bottleneck type. For example, if high-frequency goods are too far away, the weight for inbound / outbound efficiency is increased, making subsequent storage location allocation more targeted at alleviating the current bottleneck. Then, based on the adjusted weights, storage location allocation and path planning are re-performed, forming a new iteration. This process is repeated until the maximum completion time converges to a stable range, resulting in a joint optimization scheme for storage locations and paths. This bidirectional feedback mechanism achieves deep coupling between storage location layout and transportation paths, ensuring the system reaches optimal balance at the global level.

[0053] In some embodiments, step S50 specifically includes the following steps: during the AGV's task execution, environmental information, task status information, and AGV power and health status information are collected in real time through sensors mounted on the AGV; the collected dynamic events are divided into local conflict events and global disturbance events according to their impact range; when a local conflict event is detected, the distributed execution layer of the AGV involved initiates an autonomous coordination mechanism, and the AGVs involved exchange information on their current position, target warehouse location, remaining tasks, and current path. Each AGV determines the passage order according to the exchanged information using priority avoidance rules, and the AGV with lower priority performs local path replanning and replans a detour path between the current position and the target warehouse location. When a global disturbance event is detected, the affected equipment reports the event information to the centralized coordination layer. The centralized coordination layer assesses the scope of the event's impact and determines the list of affected AGVs and the list of unexecuted tasks. The current tasks of the affected AGVs are suspended, and the unexecuted tasks are partially reallocated, with tasks affected by the faulty equipment being reassigned to idle or low-load AGVs. Path planning is performed on the reallocated tasks, generating and executing a new local path planning scheme. After the rescheduling is completed, the updated task allocation information and path planning information are synchronized to the centralized coordination layer, and the new path scheme is issued to the corresponding AGVs. The AGVs continue to execute tasks according to the new path scheme.

[0054] Local conflict events include temporary path blockage, approach conflict between AGVs, and local area congestion, while global disturbance events include AGV failure and offline, emergency order insertion, conveyor failure and shutdown, and batch changes to task priorities.

[0055] In this embodiment, the local path replanning adopts the improved A* algorithm. After dynamically adding temporary obstacle nodes in the navigation grid, a feasible path is searched again. After the replanning is completed, each AGV continues to execute the task according to the new path.

[0056] Specifically, the system continuously collects environmental, task, and equipment status information through AGV sensors and monitoring modules, and handles events in a tiered manner according to their impact. When local events such as temporary path blockages or AGV approach conflicts occur, the distributed execution layer initiates an autonomous coordination mechanism. The affected AGVs exchange position, path, and task information, autonomously avoid obstacles and locally replan detours based on priority rules, and can quickly resume operation without upper-level intervention. When global disturbances such as AGV failures, emergency order insertions, or conveyor shutdowns occur, the centralized coordination layer takes over, assesses the scope of the event's impact, suspends affected tasks, and locally reallocates unexecuted tasks, transferring tasks to idle or low-load AGVs. Subsequently, path replanning is performed, and the new plan is synchronized to the relevant equipment. This hierarchical mechanism combines rapid response to local events with centralized optimization scheduling for global events, effectively improving the system's robustness and operational efficiency in dynamic environments.

[0057] The implementation principle of the intelligent warehouse scheduling and management method based on multi-AGV collaboration in this application is as follows: A digital twin model containing shelves, multi-position AGVs, and conveyors is constructed based on the Unity engine to provide a simulation verification environment for subsequent optimization. On this basis, a joint optimization model of storage location allocation and AGV path planning is constructed, aiming to minimize the maximum completion time for all AGVs to complete all transportation tasks. Constraints such as storage location uniqueness, load capacity, multi-vehicle yard collaboration, and shelf center of gravity stability are set. A hierarchical collaborative optimization architecture is adopted for solving the problem: the centralized coordination layer uses a multi-population adaptive simulated annealing genetic algorithm for storage location pre-allocation, prioritizing the allocation of high-frequency goods to lower-level, near-exit storage locations; the distributed execution layer uses a warehouse-transportation coupled-driven adaptive large neighborhood search algorithm for path planning, and feeds back the path planning results to the storage location allocation layer through a collaborative iteration mechanism. If the maximum completion time exceeds a threshold, the storage location weights are adjusted and re-optimized until the convergence condition is met. Finally, the collaborative optimization scheme was loaded into the digital twin model for simulation verification. After meeting the standards, it was deployed to the actual system for execution. During operation, dynamic events were collected in real time. For local conflicts, the AGVs autonomously avoided obstacles and replanned the route. For global disturbances, the centralized coordination layer performed task reassignment and path replanning. Thus, through joint optimization of storage locations and paths, hierarchical collaborative solution, and dynamic rescheduling mechanism, the overall transportation efficiency, environmental adaptability, and equipment coordination capabilities of the warehousing system were significantly improved.

[0058] Figure 1 This is a flowchart illustrating an intelligent warehouse scheduling and management method involving multi-AGV collaboration in one embodiment. It should be understood that, although... Figure 1 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows; unless explicitly stated otherwise, there is no strict order requirement for the execution of these steps, and they can be executed in other orders; and Figure 1 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0059] Based on the same technical concept, referring to Figure 2 This application also provides an intelligent warehouse scheduling and management device for multi-AGV collaboration, which adopts the following technical solution: The device includes: The 3D model module is used to obtain the physical structure parameters and equipment motion parameters of the warehousing system, and to build a digital twin model of the shelving, multi-position AGV and conveyor based on the Unity engine; The joint optimization module is used to build a joint optimization model for storage location allocation and AGV path planning. The goal is to minimize the maximum completion time of all AGVs and complete all transportation tasks. It sets constraints such as storage location uniqueness, AGV carrying capacity, multi-vehicle yard coordination, and rack center of gravity stability. The collaborative optimization module is used to obtain a list of goods to be put into storage, use simulated annealing genetic algorithm to pre-allocate storage locations and generate a storage location allocation scheme; use the storage location allocation scheme as the transportation target, use a large neighborhood search algorithm driven by warehousing-transportation coupling to perform path planning, generate an initial path planning scheme, and calculate the maximum completion time of the plan. If the maximum completion time of the plan exceeds a preset threshold, the storage location allocation weights are adjusted and re-optimized until the convergence condition is met, and a collaborative optimization scheme is obtained. The simulation execution module is used to load the collaborative optimization scheme into the digital twin model for simulation execution. If the throughput indicator meets the preset target, the collaborative optimization scheme is sent to the actual warehousing system for execution; otherwise, the parameters are adjusted and the collaborative optimization scheme is regenerated. The conflict resolution module is used to collect dynamic event information in real time during the AGV's task execution. When a dynamic event is detected, a rescheduling strategy is executed: for local conflict events, the distributed execution layer performs autonomous obstacle avoidance and path replanning through information interaction between AGVs; for global disturbance events, the centralized coordination layer is triggered to perform local task reallocation and path replanning.

[0060] In some embodiments, the 3D model module is specifically used to obtain the physical structure parameters and equipment motion parameters of the warehousing system. The physical structure parameters include the number of rows, columns, and layers of shelves, the depth, width, and height of the storage cell, the length of the conveyor, and the spacing between goods. The equipment motion parameters include the moving speed of the AGV on the ground, the translational and rotational speed of the AGV robot arm, and the conveyor conveying speed. Three-dimensional geometric models of shelves, multi-position AGVs and conveyors are constructed using 3D modeling software, and then imported into the Unity engine. The models are then assembled in the Unity scene according to the actual layout to form a virtual warehousing environment. Configure the physics engine component for the 3D geometry model in the Unity engine, add collision body components for the shelves and conveyors, add rigid body components and navigation proxy components for the AGV, and set the kinematic parameters, collision properties and walkable area of ​​the navigation mesh for the AGV. The script implements the cargo transportation logic of the conveyor, including the generation of cargo at the beginning of the conveyor at set intervals, the conveyor belt driving the movement of cargo, and the end trigger controlling the stopping and starting of cargo; at the same time, it implements the cargo loading and unloading logic of the AGV, including the extension and retraction of the robotic arm, the grabbing and releasing of cargo, the binding and unbinding of cargo in the AGV storage position, and the autonomous path planning and motion control of the AGV based on the navigation grid, generating an executable simulation verification digital twin model.

[0061] In some embodiments, the joint optimization module is specifically used to set the storage location set, the task point set, and the AGV set, define a first decision variable to indicate whether the k-th AGV travels from node i to node j; define a second decision variable to indicate whether storage location i is served by the k-th AGV; set the carrying capacity of each AGV, the total number of goods, the change in goods at each node, and the three-dimensional coordinates of each node, and set the height of the task point to zero. The optimization objective is to minimize the maximum completion time for all AGVs to complete all transportation tasks. The maximum completion time is equal to the maximum transportation time among all AGVs. The transportation time of each AGV consists of the sum of its horizontal travel time and vertical lifting time on its travel path. The horizontal travel time is equal to the Manhattan distance divided by the AGV's ground movement speed, and the vertical lifting time is equal to the target warehouse floor height multiplied by the cell height and then divided by the robot arm's lifting speed. Set uniqueness constraints for storage locations, ensuring that each storage location is served by only one AGV; set AGV carrying capacity constraints, ensuring that the number of goods loaded by each AGV at any node does not exceed its carrying capacity; set multi-vehicle-farm coordination constraints, ensuring that each AGV must start from a certain task point and eventually return to the task point; set rack center of gravity stability constraints, ensuring that the overall center of gravity height of all goods does not exceed a preset safety threshold, and the overall center of gravity height is calculated by the weighted average of the weight of each goods and the height of its storage location. By combining the objective function with various constraints, a joint optimization model is formed, with the storage location allocation result and the AGV path as joint decision variables.

[0062] In some embodiments, the collaborative optimization module is specifically used to read the information of goods to be put into storage from the warehouse management system. The list of goods to be put into storage includes the type identifier, turnover rate, weight and number of boxes of goods to be put into storage for each type of goods; wherein the turnover rate is used to characterize the frequency of goods leaving the warehouse, and the higher the turnover rate, the greater the demand for the goods to be accessed at a high frequency. With the optimization objectives of goods entry and exit efficiency, clustering of similar goods, and stability of shelf center of gravity, the multi-objective problem is transformed into a single-objective optimization problem using the linear weighted method. The objective function of storage location allocation is established. The entry and exit efficiency objective is achieved by minimizing the weighted sum of the turnover rate and storage and retrieval time of all goods. The clustering of similar goods objective is achieved by minimizing the spatial dispersion of similar goods in the warehouse. The shelf center of gravity stability objective is achieved by minimizing the overall center of gravity height of all goods. Using sequential integer encoding, all storage locations in the warehouse are numbered one-dimensionally according to row, column, and layer order. Each item to be stored corresponds to a storage location number, and the sequence of storage location numbers for all items constitutes a chromosome. Multiple subpopulations are set up, and each subpopulation contains a preset number of chromosomes as the initial population. Multiple subpopulations are controlled to evolve independently in parallel. Within each subpopulation, a tournament selection strategy is used to select parent individuals. The crossover probability and mutation probability are adaptively adjusted based on the individual fitness. The crossover probability and mutation probability change non-linearly in a Sigmoid manner with the difference between the individual fitness and the average fitness of the population. Every preset number of generations, the elite individuals in the current subpopulation are perturbed. Several consecutive gene loci on the chromosome are randomly selected, their library codes are randomly shuffled, and then refilled to generate a new solution. If the objective function value of the new solution is better than that of the original elite individual, the new solution is directly accepted to replace the original elite individual. If the objective function value of the new solution is worse than that of the original elite individual, the new solution is accepted with a preset probability to replace the individual with the lowest fitness in the population. The preset probability decreases as the number of iterations increases. Every preset number of generations, each subpopulation exchanges a preset proportion of high-quality individuals with its neighboring subpopulations according to a circular topology; the process of multipopulation parallel evolution and new solution generation is repeated until the maximum number of iterations is reached, and the chromosome with the optimal objective function value is selected from all subpopulations as the storage allocation scheme.

[0063] In some embodiments, the collaborative optimization module is specifically used to receive the storage location allocation scheme, take the target storage location of the goods to be put into storage as the customer point set, take the loading point as the vehicle yard set, set the AGV quantity, carrying capacity and movement speed parameters, initialize the damage operator set and the repair operator set and assign the same initial weight. A greedy random adaptive search algorithm is used to construct an initial path scheme. All unassigned customer points are traversed, and the transportation time cost between each customer point and each parking lot is calculated. The transportation time cost includes the Manhattan distance travel time and the robotic arm lifting service time. A restricted candidate list is constructed based on preset randomization parameters, and a customer point-parking lot pair is randomly selected from the list for insertion to form an initial path scheme. The roulette wheel strategy is used to select a destruction operator based on the operator weight, which removes a preset number of customer points from the current path; then the repair operator is selected to re-insert the removed customer points into feasible positions to form a new path scheme. Calculate the maximum completion time of the new path scheme, and assign different levels of reward scores to the corresponding operators based on the comparison results of the new path scheme with the historical best path scheme and the current path scheme; update the weight of each operator according to the cumulative reward score every preset number of iterations. The record-oriented migration criterion is used to determine whether to accept a poor path scheme where the maximum completion time of the new path scheme is greater than that of the current path scheme. The ratio of the maximum completion time of the new path scheme to the historical best path scheme is calculated. If it is less than a preset threshold, the new path scheme is accepted. The iteration is repeated until the maximum number of iterations is reached to obtain the initial path planning scheme.

[0064] In some embodiments, the collaborative optimization module is specifically used to extract the path information of each AGV based on the initial path planning scheme, calculate the transportation time required for each AGV to complete the assigned task, and take the maximum value of the transportation time of all AGVs as the maximum planned completion time. The maximum planned completion time is compared with a preset threshold, which is determined based on the throughput index. If the maximum planned completion time is not greater than the preset threshold, the current storage location allocation scheme and the current path planning scheme are determined to meet the requirements of collaborative optimization. If the maximum completion time exceeds the preset threshold, analyze the bottleneck factors in the current path planning scheme that cause the maximum completion time to be too long. Bottleneck factors include excessive load on a single AGV, overly scattered distribution of storage locations in a specific area, or excessive distance between storage locations for high-frequency goods. Based on the type of bottleneck factor, adjust the target weight of storage location allocation optimization and increase the weight value of optimization targets related to bottleneck factors. Based on the adjusted target weights, a new storage location allocation scheme is generated; the new storage location allocation scheme is then used as the transportation target to generate a new route planning scheme. The process iterates through the generation of new path planning schemes, recording the maximum completion time of the plan after each iteration. When the decrease in the maximum completion time of the plan is less than a preset percentage in a series of preset iterations, convergence is determined, and a collaborative optimization scheme is obtained.

[0065] In some embodiments, the conflict resolution module is specifically used to collect environmental information, acquire task status information, and acquire AGV power and health status information in real time through sensors mounted on the AGV during the AGV's task execution; and to classify the collected dynamic events into local conflict events and global disturbance events according to their scope of influence. When a local conflict event is detected, the distributed execution layer of the AGV involved initiates an autonomous coordination mechanism. The AGVs involved exchange information on their current position, target storage location, remaining tasks, and current path. Each AGV determines the passage order based on the exchanged information using priority avoidance rules. AGVs with lower priority perform local path replanning and replan a detour path between their current position and the target storage location. When a global disturbance event is detected, the affected equipment reports the event information to the centralized coordination layer. The centralized coordination layer assesses the scope of the event's impact, determines the list of affected AGVs and the list of unexecuted tasks, suspends the current tasks of the affected AGVs, performs local reallocation of unexecuted tasks, and reassigns tasks affected by the faulty equipment to idle or low-load AGVs. It also performs path planning on the reassigned tasks, generates and executes a new local path planning scheme. After the rescheduling is completed, the updated task allocation information and path planning information will be synchronized to the centralized coordination layer, and the new path plan will be issued to the corresponding AGV; the AGV will continue to execute the task according to the new path plan.

[0066] This application also discloses a control device.

[0067] Specifically, the control device includes a memory and a processor. The memory stores a computer program that can be loaded by the processor and executed to perform the aforementioned intelligent warehouse scheduling and management method for multi-AGV collaboration.

[0068] This application also discloses a computer-readable storage medium.

[0069] Specifically, the computer-readable storage medium stores a computer program that can be loaded by a processor and executed as described above for intelligent warehouse scheduling and management of multi-AGV collaboration. The computer-readable storage medium includes, for example, various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0070] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A method for intelligent warehouse scheduling and management involving multiple AGVs, characterized in that, include: Obtain the physical structure parameters and equipment motion parameters of the warehousing system, and build a digital twin model containing shelves, multi-position AGVs and conveyors based on the Unity engine; A joint optimization model for storage location allocation and AGV path planning is constructed with the goal of minimizing the maximum completion time of all AGVs and all transportation tasks. The model sets constraints such as storage location uniqueness, AGV carrying capacity, multi-vehicle yard coordination, and rack center of gravity stability. Obtain the list of goods to be put into storage, use simulated annealing genetic algorithm to pre-allocate storage locations and generate a storage location allocation scheme; use the storage location allocation scheme as the transportation target, use warehousing-transportation coupled large neighborhood search algorithm to perform path planning, generate an initial path planning scheme, and calculate the maximum completion time of the plan. If the maximum completion time of the plan exceeds a preset threshold, adjust the storage location allocation weight and re-optimize until the convergence condition is met to obtain a collaborative optimization scheme. The collaborative optimization scheme is loaded into the digital twin model for simulation. If the throughput indicator meets the preset target, the collaborative optimization scheme is sent to the actual warehousing system for execution; otherwise, the parameters are adjusted and the collaborative optimization scheme is regenerated. During the execution of tasks by AGVs, dynamic event information is collected in real time. When a dynamic event is detected, a rescheduling strategy is executed: for local conflict events, the distributed execution layer performs autonomous obstacle avoidance and path replanning through information interaction between AGVs; for global disturbance events, the centralized coordination layer is triggered to perform local task reallocation and path replanning.

2. The intelligent warehouse scheduling and management method for multi-AGV collaboration according to claim 1, characterized in that, The process involves acquiring the physical structure parameters and equipment motion parameters of the warehousing system, and constructing a digital twin model based on the Unity engine, including shelves, multi-position AGVs, and conveyors. Obtain the physical structure parameters and equipment motion parameters of the warehousing system. The physical structure parameters include the number of rows, columns, and layers of shelves, the depth, width, and height of the storage cell, the length of the conveyor, and the spacing between goods. The equipment motion parameters include the moving speed of the AGV on the ground, the translational and rotational speed of the AGV robot arm, and the conveyor conveying speed. Three-dimensional geometric models of shelves, multi-position AGVs and conveyors are constructed using 3D modeling software, and then imported into the Unity engine. The models are then assembled in the Unity scene according to the actual layout to form a virtual warehousing environment. Configure the physics engine component for the 3D geometry model in the Unity engine, add collision body components for the shelves and conveyors, add rigid body components and navigation proxy components for the AGV, and set the kinematic parameters, collision properties and walkable area of ​​the navigation mesh for the AGV. The script implements the cargo transportation logic of the conveyor, including the generation of cargo at the beginning of the conveyor at set intervals, the conveyor belt driving the movement of cargo, and the end trigger controlling the stopping and starting of cargo; at the same time, it implements the cargo loading and unloading logic of the AGV, including the extension and retraction of the robotic arm, the grabbing and releasing of cargo, the binding and unbinding of cargo in the AGV storage position, and the autonomous path planning and motion control of the AGV based on the navigation grid, generating an executable simulation verification digital twin model.

3. The intelligent warehouse scheduling and management method for multi-AGV collaboration according to claim 1, characterized in that, The joint optimization model for storage location allocation and AGV path planning includes: Set up a set of storage locations, a set of task points, and a set of AGVs. Define a first decision variable to indicate whether the k-th AGV travels from node i to node j. Define a second decision variable to indicate whether storage location i is served by the k-th AGV. Set the carrying capacity of each AGV, the total number of goods, the change in goods at each node, and the three-dimensional coordinates of each node. Set the height of the task point to zero. The optimization objective is to minimize the maximum completion time for all AGVs to complete all transportation tasks. The maximum completion time is equal to the maximum transportation time among all AGVs. The transportation time of each AGV consists of the sum of its horizontal travel time and vertical lifting time on its travel path. The horizontal travel time is equal to the Manhattan distance divided by the AGV's ground movement speed, and the vertical lifting time is equal to the target warehouse floor height multiplied by the cell height and then divided by the robot arm's lifting speed. Set uniqueness constraints for storage locations, ensuring that each storage location is served by only one AGV; set AGV carrying capacity constraints, ensuring that the number of goods loaded by each AGV at any node does not exceed its carrying capacity; set multi-vehicle-farm coordination constraints, ensuring that each AGV must start from a certain task point and eventually return to the task point; set rack center of gravity stability constraints, ensuring that the overall center of gravity height of all goods does not exceed a preset safety threshold, and the overall center of gravity height is calculated by the weighted average of the weight of each goods and the height of its storage location. By combining the objective function with various constraints, a joint optimization model is formed, with the storage location allocation result and the AGV path as joint decision variables.

4. The intelligent warehouse scheduling and management method for multi-AGV collaboration according to claim 3, characterized in that, The process of obtaining the list of goods to be received and using a simulated annealing genetic algorithm to pre-allocate storage locations to generate a storage location allocation scheme includes: The warehouse management system reads the information of goods to be put into storage. The list of goods to be put into storage includes the type identifier, turnover rate, weight and number of boxes of goods to be put into storage for each type of goods. The turnover rate is used to characterize the frequency of goods leaving the warehouse. The higher the turnover rate, the greater the demand for the goods to be accessed at a high frequency. With the optimization objectives of goods entry and exit efficiency, clustering of similar goods, and stability of shelf center of gravity, the multi-objective problem is transformed into a single-objective optimization problem using the linear weighted method. The objective function of storage location allocation is established. The entry and exit efficiency objective is achieved by minimizing the weighted sum of the turnover rate and storage and retrieval time of all goods. The clustering of similar goods objective is achieved by minimizing the spatial dispersion of similar goods in the warehouse. The shelf center of gravity stability objective is achieved by minimizing the overall center of gravity height of all goods. Using sequential integer encoding, all storage locations in the warehouse are numbered one-dimensionally according to row, column, and layer order. Each item to be stored corresponds to a storage location number, and the sequence of storage location numbers for all items constitutes a chromosome. Multiple subpopulations are set up, and each subpopulation contains a preset number of chromosomes as the initial population. Multiple subpopulations are controlled to evolve independently in parallel. Within each subpopulation, a tournament selection strategy is used to select parent individuals. The crossover probability and mutation probability are adaptively adjusted based on the individual fitness. The crossover probability and mutation probability change non-linearly in a Sigmoid manner with the difference between the individual fitness and the average fitness of the population. Every preset number of generations, the elite individuals in the current subpopulation are perturbed. Several consecutive gene loci on the chromosome are randomly selected, their library codes are randomly shuffled, and then refilled to generate a new solution. If the objective function value of the new solution is better than that of the original elite individual, the new solution is directly accepted to replace the original elite individual. If the objective function value of the new solution is worse than that of the original elite individual, the new solution is accepted with a preset probability to replace the individual with the lowest fitness in the population. The preset probability decreases as the number of iterations increases. Every preset number of generations, each subpopulation exchanges a preset proportion of high-quality individuals with its neighboring subpopulations according to a circular topology; the multi-population parallel evolution and new solution generation steps are repeated until the maximum number of iterations is reached, and the chromosome with the optimal objective function value is selected from all subpopulations as the storage allocation scheme.

5. The intelligent warehouse scheduling and management method for multi-AGV collaboration according to claim 4, characterized in that, The step involves using the storage location allocation scheme as the transportation target, employing a large neighborhood search algorithm driven by warehousing-transportation coupling for path planning, and generating an initial path planning scheme, including: Receive the storage location allocation scheme, take the target storage location of the goods to be put into storage as the customer point set, take the loading point as the vehicle yard set, set the AGV quantity, carrying capacity and movement speed parameters, initialize the damage operator set and repair operator set and assign the same initial weight. A greedy random adaptive search algorithm is used to construct an initial path scheme. All unassigned customer points are traversed, and the transportation time cost between each customer point and each parking lot is calculated. The transportation time cost includes the Manhattan distance travel time and the robotic arm lifting service time. A restricted candidate list is constructed based on preset randomization parameters, and a customer point-parking lot pair is randomly selected from the list for insertion to form an initial path scheme. The roulette wheel strategy is used to select a destruction operator based on the operator weight, which removes a preset number of customer points from the current path; then the repair operator is selected to re-insert the removed customer points into feasible positions to form a new path scheme. Calculate the maximum completion time of the new path scheme, and assign different levels of reward scores to the corresponding operators based on the comparison results of the new path scheme with the historical best path scheme and the current path scheme; update the weight of each operator according to the cumulative reward score every preset number of iterations. The record-oriented migration criterion is used to determine whether to accept a poor path scheme where the maximum completion time of the new path scheme is greater than that of the current path scheme. The ratio of the maximum completion time of the new path scheme to the historical best path scheme is calculated. If it is less than a preset threshold, the new path scheme is accepted. The iteration is repeated until the maximum number of iterations is reached to obtain the initial path planning scheme.

6. The intelligent warehouse scheduling and management method for multi-AGV collaboration according to claim 5, characterized in that, The maximum completion time of the calculation plan is determined. If the maximum completion time exceeds a preset threshold, the storage location allocation weights are adjusted and the plan is re-optimized until the convergence condition is met, resulting in a collaborative optimization scheme, including: Based on the initial path planning scheme, the path information of each AGV is extracted, the transportation time required for each AGV to complete the assigned task is calculated, and the maximum value of the transportation time of all AGVs is taken as the maximum completion time of the plan. The maximum completion time of the plan is compared with a preset threshold, which is determined based on the throughput index; if the maximum completion time of the plan is not greater than the preset threshold, it is determined that the current storage location allocation scheme and the current path planning scheme meet the requirements of collaborative optimization. If the maximum completion time of the plan is greater than the preset threshold, the bottleneck factors in the current path planning scheme that cause the maximum completion time to be too long are analyzed. Bottleneck factors include excessive load on a single AGV, overly scattered distribution of storage locations in a specific area, or excessive distance between storage locations for high-frequency goods. Based on the type of bottleneck factor, the target weight of storage location allocation optimization is adjusted, and the weight value of optimization targets related to bottleneck factors is increased. Based on the adjusted target weights, a new storage location allocation scheme is generated; the new storage location allocation scheme is then used as the transportation target to generate a new route planning scheme. The generation of new path planning schemes is repeatedly iterated, and the maximum completion time of the plan is recorded after each iteration. When the decrease in the maximum completion time of the plan is less than a preset percentage in a series of preset iterations, convergence is determined, and the collaborative optimization scheme is obtained.

7. The intelligent warehouse scheduling and management method for multi-AGV collaboration according to claim 5, characterized in that, During the AGV's task execution, dynamic event information is collected in real time; when a dynamic event is detected, a rescheduling strategy is executed, including: During the AGV's task execution, environmental information, task status information, and AGV power and health status information are collected in real time through the sensors on the AGV. The collected dynamic events are divided into local conflict events and global disturbance events according to their scope of influence. When a local conflict event is detected, the distributed execution layer of the AGV involved initiates an autonomous coordination mechanism. The AGVs involved exchange information on their current position, target storage location, remaining tasks, and current path. Each AGV determines the passage order based on the exchanged information using priority avoidance rules. AGVs with lower priority perform local path replanning and replan a detour path between their current position and the target storage location. When a global disturbance event is detected, the affected equipment reports the event information to the centralized coordination layer. The centralized coordination layer assesses the scope of the event's impact, determines the list of affected AGVs and the list of unexecuted tasks, suspends the current tasks of the affected AGVs, performs local reallocation of unexecuted tasks, and reassigns tasks affected by the faulty equipment to idle or low-load AGVs. It also performs path planning on the reassigned tasks, generates and executes a new local path planning scheme. After the rescheduling is completed, the updated task allocation information and path planning information will be synchronized to the centralized coordination layer, and the new path plan will be issued to the corresponding AGV; the AGV will continue to execute the task according to the new path plan.

8. A multi-AGV collaborative intelligent warehouse scheduling and management device, characterized in that, The device includes: The 3D model module is used to obtain the physical structure parameters and equipment motion parameters of the warehousing system, and to build a digital twin model of the shelving, multi-position AGV and conveyor based on the Unity engine; The joint optimization module is used to build a joint optimization model for storage location allocation and AGV path planning. The goal is to minimize the maximum completion time of all AGVs and complete all transportation tasks. It sets constraints such as storage location uniqueness, AGV carrying capacity, multi-vehicle yard coordination, and rack center of gravity stability. The collaborative optimization module is used to obtain a list of goods to be put into storage, use simulated annealing genetic algorithm to pre-allocate storage locations, and generate a storage location allocation scheme. The storage location allocation scheme is used as the transportation target, and a large neighborhood search algorithm driven by warehousing-transportation coupling is used for path planning to generate an initial path planning scheme. The maximum completion time of the planning is calculated. If the maximum completion time of the planning exceeds a preset threshold, the storage location allocation weight is adjusted and re-optimized until the convergence condition is met, and a collaborative optimization scheme is obtained. The simulation execution module is used to load the collaborative optimization scheme into the digital twin model for simulation execution. If the throughput indicator meets the preset target, the collaborative optimization scheme is sent to the actual warehousing system for execution; otherwise, the parameters are adjusted and the collaborative optimization scheme is regenerated. The conflict resolution module is used to collect dynamic event information in real time during the AGV's task execution. When a dynamic event is detected, a rescheduling strategy is executed: for local conflict events, the distributed execution layer performs autonomous obstacle avoidance and path replanning through information interaction between AGVs; for global disturbance events, the centralized coordination layer is triggered to perform local task reallocation and path replanning.

9. A control device, characterized in that, The device includes: A memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and executed as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer program is stored that can be loaded by a processor and executed as described in any one of claims 1 to 7.

Images