Multi-layer three-dimensional warehouse automatic seeding path optimization scheduling method and system

Abstract

The present application relates to the field of three-dimensional warehouse automatic seeding, and more particularly to a multi-layer three-dimensional warehouse automatic seeding path optimization scheduling method and system, comprising the following steps: obtaining seeding associated data and three-dimensional warehouse associated data through a data interaction module to construct a multi-dimensional mapping table; calling a preset material characteristic and path constraint mapping library, filtering paths that meet the constraints through an improved A algorithm and calculating path costs to output an initial path; inputting AGV real-time load states, outputting regional task allocation plans through a global scheduling center, combining multi-source sensing data, optimizing local obstacle avoidance, and outputting an optimized task allocation table; calculating task priorities, allocating resources according to priorities and processing path conflicts to output final task scheduling instructions. The present application calculates task priorities through multi-dimensional parameters, balances AGV loads through distributed global computing power scheduling, effectively solves the problem of AGV task overload and idle coexistence, and avoids job interruption caused by hardware failure.

Description

Technical Field

[0001] This invention relates to the field of automated seeding in multi-level automated warehouses, and more particularly to a method and system for optimizing and scheduling automated seeding paths in multi-level automated warehouses. Background Technology

[0002] With the logistics industry experiencing a sustained annual growth of 15%-20% in business volume, multi-level automated warehouses have become the mainstream warehousing form due to their high-density storage and fast-paced turnover. The efficiency of their automated loading and unloading operations (the process of precisely transferring materials from multiple storage levels to the target loading point) directly determines the overall operational level of the warehouse. Current technologies have established a basic framework of path planning, task allocation, and hardware execution, but path planning largely relies on traditional methods. The algorithm, Dijkstra's algorithm, or its simple improved version, takes the shortest physical distance as the core objective, without integrating material characteristics and dynamic costs of inter-layer transfer; task scheduling is only based on the urgency of orders, without associating with material sensitive attributes; and there is a delay in the interaction between AGV perception data and the algorithm, resulting in insufficient overall adaptability to complex scenarios.

[0003] Existing technologies suffer from three core problems: First, path planning lacks multi-dimensional constraint integration and fails to adjust path parameters based on material size and fragility level, resulting in long materials scraping against shelves during travel and fragile materials experiencing a breakage rate as high as 5%-8%. Second, dynamic environment response is lagging, with sensing data being uploaded in batches at 1-second intervals, making it difficult to avoid dynamic obstacles such as inspection vehicles and scattered materials in a timely manner, easily causing AGV congestion and stagnation. Third, task scheduling does not coordinate with equipment load, allocating AGV resources only based on order urgency, causing some AGVs to be overloaded and others to be idle, extending the average transfer time for a single material to 12-15 minutes, and the overall operating efficiency to only 60%-70% of the designed capacity, which is insufficient to meet actual warehousing needs.

[0004] For example, the publication CN114648267B, 'Optimization Method and System for Scheduling Paths in Automated Warehouses,' optimizes warehouse outbound scheduling paths through order priority analysis, spatiotemporal map models, recurrent neural network material allocation, and time window algorithms. However, its technical solution still has significant limitations in adapting to multi-level automated warehouse scenarios: First, path planning is based solely on fitting the shortest path using basic coordinates, without integrating material characteristics (such as size and fragility level) and the dynamic costs of cross-level transfers in multi-level warehouses. This leads to problems such as long materials scratching against shelves and fragile materials being damaged in multi-level storage scenarios. The problems are as follows: First, frequent occurrences still occur; second, dynamic environment response relies solely on time window algorithms to handle static path conflicts, without addressing real-time load balancing of AGVs (remaining power, assigned task time) and real-time obstacle avoidance (such as inspection vehicles and scattered materials), which can easily lead to localized congestion and stagnation when multiple AGVs operate in parallel in multi-layer warehouses; third, task scheduling is based solely on the basic order of orders, without considering material sensitivity attributes (such as fragile or excessively long) and equipment load status, resulting in some AGVs being overloaded and others being idle, making it difficult to adapt to the special requirements of multi-layer warehouse automatic seeding for 'multiple materials, cross-layer, and high precision'.

[0005] Therefore, it is necessary to design an automatic seeding path optimization scheduling method and system for multi-layer three-dimensional warehouses. Summary of the Invention

[0006] To address the technical deficiencies in the background technology, this invention proposes a method and system for optimizing and scheduling the automatic seeding path of multi-layer three-dimensional warehouses, which solves the aforementioned technical problems and meets practical needs. The specific technical solution is as follows:

[0007] The method for optimizing and scheduling the automatic seeding path in multi-level automated warehouses includes the following steps:

[0008] The data interaction module is used to obtain seeding-related data and three-dimensional warehouse-related data, and a multi-dimensional mapping table is constructed.

[0009] Based on a multi-dimensional mapping table, a pre-defined material property and path constraint mapping library is called. An improved A algorithm is used to filter paths that meet the constraints and calculate path costs, and output the initial path.

[0010] Based on the initial path, the real-time load status of the AGV is input, and the regional task allocation plan is output through the global scheduling center. Combined with multi-source perception data, the optimized task allocation table is output through local obstacle avoidance optimization.

[0011] Based on the task allocation table, calculate task priorities, allocate resources according to priorities and handle path conflicts, configure backup AGVs, and output the final task scheduling instructions.

[0012] Furthermore, after outputting the final task scheduling instruction, the following steps are also included:

[0013] Based on the final task scheduling instruction, the AGV extracts material characteristics and switches the end execution head, and completes material grabbing, transfer and delivery after triple verification;

[0014] When an abnormality is detected during AGV operation, the abnormality handling process is triggered, and a new final task scheduling instruction is regenerated.

[0015] After the AGV operation is completed, the seeding accuracy is determined, the warehouse data is updated synchronously, and the entire process data is recorded in the algorithm iteration database. Based on the algorithm iteration database, the parameters of the A algorithm and the local obstacle avoidance optimization are optimized and improved.

[0016] Furthermore, the specific process for constructing the multi-dimensional mapping table is as follows:

[0017] By connecting the data interaction module with the warehouse management system and the production execution system, the system can obtain sowing task orders and material lists, extract order urgency, target sowing position 3D coordinates, material size, material weight and material fragility level, and generate sowing-related data.

[0018] The basic coordinates of the material storage location are read by RFID tags, corrected by the positioning module, and the physical existence is confirmed by taking images of the storage location with an industrial camera to generate the three-dimensional coordinates of the physical storage location.

[0019] The temperature, dust concentration and ground condition inside the automated storage and retrieval system are collected at a set frequency by environmental monitoring sensors to generate associated data for the automated storage and retrieval system.

[0020] By using the association unit of the data interaction module according to the set structure, the seeding association data, the three-dimensional coordinates of the physical storage location, and the three-dimensional warehouse association data are associated and bound together to generate a multi-dimensional mapping table.

[0021] Furthermore, the construction process of the material property and path constraint mapping library is as follows:

[0022] By connecting with enterprise ERP systems, warehouse management systems and material receiving and inspection processes, core characteristic parameters of all categories of materials are collected to form a standardized dataset.

[0023] Based on the physical layout and operational safety specifications of the automated warehouse, the inherent and dynamic constraint parameters of all paths are extracted to form a quantifiable constraint rule base.

[0024] By using a multi-dimensional matching algorithm, a unique mapping relationship is established between the standardized dataset and the constraint rule base, forming structured mapping entries, and a material characteristic and path constraint mapping base is established with the material ID as the hash index.

[0025] Based on the material property and path constraint mapping library, simulation verification is performed through a simulation platform, and the material property and path constraint mapping library is updated based on the simulation verification results.

[0026] Furthermore, the specific steps for outputting the initial path using the improved A algorithm are as follows:

[0027] Based on the multi-dimensional mapping table, the corresponding constraint parameters are retrieved from the material properties and path constraint mapping library to filter out candidate paths.

[0028] Based on a multi-dimensional mapping table, the real-time status of the inter-layer transfer unit is extracted, and the cross-layer transfer time is calculated.

[0029] Based on the improved A algorithm, combined with cross-layer transfer time, the estimated total cost of candidate paths is calculated using the path cost calculation formula, and the candidate path with the minimum estimated total cost is selected as the initial path.

[0030] The formula for calculating the path cost is as follows:

[0031] ,

[0032] in, This represents the estimated total cost from the starting node through the current node n to the target node. This represents the actual distance cost from the starting node to the current node n. This represents the actual transit cost from the starting node to the current node n. This represents the actual congestion cost from the starting node to the current node n. This represents the heuristic distance cost from the current node n to the target node. This represents the heuristic transfer cost from the current node n to the target node. Let represent the heuristic congestion cost from the current node n to the target node. This is the distance cost weighting coefficient. This is the weighting coefficient for transportation costs. This is the congestion cost weighting coefficient.

[0033] Furthermore, the specific process of the output area task allocation plan is as follows:

[0034] The load status of AGVs on each floor of the automated warehouse is collected in real time through several distributed regional scheduling nodes, including the remaining power of the AGVs and the total time of the assigned tasks, and its own CPU utilization is monitored. The load data is then fed back to the global scheduling center in real time.

[0035] Based on load data and initial path, when the CPU utilization of a certain distributed area scheduling node exceeds the preset value, it is determined to be computing power overload. The list of tasks currently being processed by the distributed area scheduling node is extracted, and non-core tasks are filtered out.

[0036] The global scheduling center filters out distributed regional scheduling nodes with CPU utilization below a preset value from all distributed regional scheduling nodes, and migrates the data packets of the filtered non-core tasks to the distributed regional scheduling nodes with CPU utilization below the preset value via Ethernet.

[0037] Once the task migration is complete, the task allocation information of all distributed regional scheduling nodes is integrated through the global scheduling center to generate a regional task allocation plan that includes the number of AGV tasks allocated to each region, the task type, and the expected start time.

[0038] Furthermore, the specific process of the local obstacle avoidance optimization is as follows:

[0039] Collect obstacle distance data, ground image data, and moving target data to generate multi-source perception data, and form a dynamic environment map through local path planning units;

[0040] Based on the regional task allocation plan and dynamic environment map, overlay analysis is performed through the collision detection module;

[0041] If a collision risk is detected, the detour distance is calculated. If the detour distance is less than the preset value, the AGV parameters are adjusted using a dynamic window method combined with reinforcement learning, and the optimized AGV parameters are output.

[0042] If the detour distance exceeds the preset value, the position and movement parameters of the obstacle will be fed back to the global scheduling center, triggering incremental replanning and forming an optimized AGV path list;

[0043] By integrating and optimizing the AGV parameters and AGV path list through local path planning units, a task allocation table is generated that includes the optimized path and estimated arrival time of each AGV.

[0044] Furthermore, the calculation process for the task priority is as follows:

[0045] Based on the task allocation table, extract the order urgency, material characteristics, and remaining time for each task, and substitute them into the task priority calculation formula to calculate the task priority, as follows:

[0046] ,

[0047] in, This represents the task priority value. The larger the value, the higher the task priority. The urgency level of the order. For material sensitivity coefficient, This represents the risk factor for task timeout.

[0048] Furthermore, the specific process for outputting the final task scheduling instruction is as follows:

[0049] Based on the task priorities and task allocation table of all tasks, resources matching the path attributes are allocated to each task in descending order of priority, and a preliminary resource allocation scheme is output.

[0050] Based on the initial resource allocation scheme, compare the optimized paths of all tasks. If the AGV path of a low-priority task intersects with the AGV path of a high-priority task and the time overlaps, schedule the low-priority AGV to stop at the nearest avoidance zone and continue driving after the high-priority task has passed. Output the updated resource allocation scheme.

[0051] Based on the task priority of all tasks, bind a spare AGV to the task with a priority greater than the preset value, and output a redundancy configuration table;

[0052] The initial resource allocation scheme, task allocation table, and redundancy configuration table are integrated into a structured final task scheduling instruction.

[0053] The multi-layer automated seeding path optimization and scheduling system for automated storage and retrieval systems includes:

[0054] The data interaction module includes multi-protocol interfaces, multi-source data acquisition units, data association units, and a real-time database. It achieves data transmission through 5G industrial Ethernet, providing basic data support for the entire system.

[0055] The path planning module includes a global path planning unit and a local path planning unit. It interacts with the data interaction module in real time and outputs the initial path, regional task allocation plan, and optimized path list.

[0056] The task scheduling module includes a priority calculation unit, a resource allocation unit, and a redundancy configuration unit. It receives the output data from the path planning module and outputs the final task scheduling instruction to the automatic seeding execution module.

[0057] The automatic seeding execution module includes an intelligent AGV, a modular end effector, and a vision recognition unit. It receives instructions from the task scheduling module and completes material grabbing, transfer, and delivery.

[0058] The environmental monitoring and fault self-diagnosis module includes an environmental data monitoring unit, a hardware fault detection unit, a backup resource scheduling unit, and a maintenance work order generation unit, which handles anomalies and faults.

[0059] Compared with existing technologies, the multi-layer automated seeding path optimization scheduling method and system for warehouses provided by this invention have the following beneficial effects:

[0060] This invention constructs a multi-dimensional mapping table that associates material characteristics, warehousing environment, and path parameters. It calls a material characteristic-path constraint mapping library and combines it with an improved A algorithm to filter candidate paths. This makes path planning no longer solely based on physical distance, but fully adapts to attributes such as material size and fragility. At the same time, it optimizes local obstacle avoidance to handle dynamic obstacles in real time, avoiding situations such as long materials scratching against shelves, damage to fragile materials, and AGV collisions and congestion. This significantly improves the adaptability of path planning and the safety of the operation process.

[0061] This invention addresses the shortcomings of existing task scheduling methods, such as lack of coordination with equipment load and low efficiency. It calculates task priorities using multi-dimensional parameters, balances AGV load by combining distributed global computing power scheduling, and configures backup AGVs for critical tasks and establishes an exception handling process. This effectively solves the problem of AGV task overload and idleness coexisting, avoiding operation interruptions caused by hardware failures. Furthermore, through full-process data recording and algorithm iteration optimization, it continuously improves the system's adaptability to complex warehousing scenarios, shortens the overall material transfer cycle, improves the space utilization and operational efficiency of multi-level automated warehouses, and reduces labor and maintenance costs. It has strong practicality and scalability. Attached Figure Description

[0062] Figure 1 This is a flowchart of the automatic seeding path optimization and scheduling method for multi-layer automated warehouses in this invention.

[0063] Figure 2 This is a schematic diagram of the modules of the multi-layer automated seeding path optimization and scheduling system for warehouses in this invention. Detailed Implementation

[0064] In the description of this invention, it should be understood that the terms "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "middle," and "inner," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, it should be noted that unless otherwise explicitly specified and limited, the terms "installed," "connected," and "joined" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention through specific circumstances.

[0065] The embodiments of the present invention will be described below with reference to the accompanying drawings and related examples. The embodiments of the present invention are not limited to the following examples, and the present invention relates to the relevant necessary components in this technical field, which should be regarded as well-known technology in this technical field and can be known and mastered by those skilled in this technical field.

[0066] See Figure 1 The method for optimizing and scheduling the automatic seeding path of multi-level automated warehouses includes the following steps:

[0067] Step S100: Obtain seeding-related data and three-dimensional warehouse-related data through the data interaction module, and construct a multi-dimensional mapping table;

[0068] The data interaction module connects with the warehouse management system, production execution system, and various sensing devices within the automated warehouse to collect and correlate two types of key data, ultimately generating a multi-dimensional mapping table. Seeding-related data is obtained from the warehouse management system and production execution system, including seeding task orders and bills of materials; automated warehouse-related data includes the three-dimensional coordinates of material storage locations and environmental data within the automated warehouse. The multi-dimensional mapping table is a structured data table generated by the correlation unit of the data interaction module, binding all key data according to a set format.

[0069] Step S200: Based on the multi-dimensional mapping table, call the preset material property and path constraint mapping library, filter the paths that meet the constraints through the improved A algorithm and calculate the path cost, and output the initial path;

[0070] Based on the material data in the multi-dimensional mapping table, a pre-defined material characteristic and path constraint mapping library is invoked. The library is used to match the path constraint rules corresponding to the current material, ensuring that the path is adapted to the material characteristics from the source. According to the constraint rules in the mapping library, paths that do not meet the conditions are eliminated from all available paths in the automated warehouse, retaining physically feasible and safe and compliant paths as candidate paths. An improved A algorithm is used to quantitatively calculate the estimated total cost of each candidate path, and finally, the candidate path with the minimum estimated total cost is selected as the initial path. The improved A algorithm is based on the traditional... This invention introduces a path search method with a multi-objective cost function based on the existing algorithm, used to find the optimal or near-optimal path under constraints. In this invention, the improved A algorithm evaluates search nodes by extending the heuristic function and incorporating multiple weighting factors such as distance, time, and risk.

[0071] Step S300: Based on the initial path, input the real-time load status of the AGV, output the regional task allocation plan through the global scheduling center, combine multi-source perception data, and output the optimized task allocation table through local obstacle avoidance optimization.

[0072] The initial path and real-time AGV load status are input into the global scheduling center. Non-core tasks from overloaded nodes are migrated to low-load nodes via Ethernet. The task allocation of all nodes is integrated, and a regional task allocation plan is output, specifying the number of tasks, task types, and estimated start times for AGVs in each region. This ensures balanced computing power during large-scale AGV scheduling and avoids equipment overload or idleness. Multi-source sensing data is used to capture various types of sensing data of the dynamic environment of the automated warehouse. Local obstacle avoidance optimization is a real-time path adjustment mechanism for dynamic obstacles. The task allocation table is the final path allocation scheme output after local obstacle avoidance optimization, containing the optimized path and estimated arrival time of each AGV, providing accurate basis for subsequent task priority calculation and resource allocation.

[0073] Step S400: Based on the task allocation table, calculate the task priority, allocate resources according to priority and handle path conflicts, configure the backup AGV, and output the final task scheduling instruction.

[0074] Task priority is a comprehensive score that measures the urgency and importance of task execution. It is used to determine the scheduling order and the direction of resource allocation. The final task scheduling instruction is an executable instruction output by the task scheduling module, which integrates task priority sorting, AGV allocation results, path parameters and redundancy configuration information. The standby AGV refers to the redundant AGV configured for critical tasks. The standby AGV is selected from the idle AGV pool according to the principle of proximity or performance matching.

[0075] In one specific embodiment, the system receives a large number of urgent orders, including fragile household appliances and extra-long building materials. The data interaction module acquires the attributes of these materials and their target placement information in real time, and constructs a multi-dimensional mapping table based on the occupancy status of each level of the automated warehouse. The system calls the deceleration and passage rules for fragile items and the width-limited path strategy for extra-long items stored in the mapping library, and uses an improved A / B algorithm to avoid narrow turning sections and generate a safe initial path. The global scheduling center rationally divides and allocates tasks according to the current AGV load distribution, and adjusts the driving trajectory through local obstacle avoidance optimization based on temporary obstacles detected by LiDAR. For high-priority orders, the system automatically assigns nearby idle AGVs as backups, which immediately take over the task in case of a main vehicle failure. After the final scheduling command is issued, the AGV group works together to complete an efficient and zero-damage automated placement operation.

[0076] This invention acquires seeding-related data and automated warehouse-related data and constructs a multi-dimensional mapping table. It calls a material characteristic and path constraint mapping library and generates an initial path using an improved A algorithm. It inputs the real-time load status of AGVs and generates regional task allocation plans through a global scheduling center. It combines multi-source perception data to perform local obstacle avoidance optimization and outputs a task allocation table. Based on the task allocation table, it calculates task priorities, handles path conflicts, configures backup AGVs, and outputs final scheduling instructions. By integrating material characteristics and warehousing environment information, it achieves path constraint response. It enhances path safety by embedding multi-dimensional cost evaluation with the improved A algorithm, supports dynamic obstacle avoidance to reduce collision risks using multi-source perception data, achieves reasonable resource allocation by combining task priorities and load status, and enhances system fault tolerance by configuring backup AGVs. This can achieve the technical effects of shortening material transfer cycles, improving space utilization, and increasing operational stability.

[0077] In one embodiment of the present invention, after outputting the final task scheduling instruction, the method further includes the following steps:

[0078] Step S500: Based on the final task scheduling instruction, the AGV extracts the material characteristics and switches the end execution head. After triple verification, the material grabbing, transfer and delivery are completed.

[0079] After receiving the final task scheduling instruction, the AGV parses the core characteristic parameters of the current material from the instruction, specifically including the material's weight, size, shape, and fragility level. It then calls the preset material characteristic-end effect head mapping rules to automatically switch to the appropriate modular end effect head. Specifically, for ultra-light materials, it switches to a vacuum anti-slip suction cup; for irregularly shaped materials, it switches to a multi-degree-of-freedom mechanical claw; and for extra-long or extra-heavy materials, it switches to a forklift-type end effect head. Once the AGV reaches the material storage location, it initiates a triple verification mechanism to avoid picking up the wrong material or deviating from the path. The triple verification includes visual verification, RFID verification, and positioning verification. Specifically, it scans the material's barcode or QR code with an industrial camera to obtain the material ID and compares it with the material ID in the scheduling instruction; it reads the material tag information using an RFID reader to confirm that the material's weight and fragility level match the instruction; and it uses a UWB positioning module to confirm that the current storage location coordinates match the storage location coordinates in the instruction. After passing the triple verification, the end effector grabs materials according to the preset parameters; the AGV strictly follows the optimized path in the scheduling instructions and receives dynamic environmental data from the local path planning unit in real time during the journey, and fine-tunes the driving speed and direction; after reaching the target seeding position, it completes the placement according to the preset placement accuracy, and then sends a work progress signal back to the task scheduling module.

[0080] Step S600: When an abnormality is detected during AGV operation, the abnormality handling process is triggered, and a new final task scheduling instruction is regenerated.

[0081] In the triple verification process, if the AGV fails to recognize the barcode, the RFID information is inconsistent with the instructions, or the positioning deviation is greater than ±15mm, it is immediately identified as a material abnormality. The current action is stopped and a local alarm is triggered. The preset emergency path for abnormal materials is invoked to transfer the abnormal materials to the abnormal material processing area. The staff scans the materials using a handheld PDA terminal or manually enters the material ID, weight, size, and other correction information. The PDA terminal uploads the corrected data to the data interaction module in real time. The data interaction module updates the multi-dimensional mapping table and re-triggers the generation of a new final task scheduling instruction, scheduling the AGV to re-execute the operation to ensure that materials do not accumulate.

[0082] When an AGV detects a malfunction, follow these steps to handle it:

[0083] The AGV battery management system monitors the remaining power in real time (low power fault is determined when SOC < 22V), the motor controller monitors the speed deviation (motor fault is determined when > ±5%), and the fault information is transmitted to the environmental monitoring and fault self-diagnosis module in real time via the vehicle Ethernet.

[0084] The fault self-diagnosis module generates fault codes and synchronizes them to the redundant configuration unit of the task scheduling module;

[0085] The redundant configuration unit calls the binding relationship between the task and the standby AGV, activates the standby AGV within 10 seconds, and issues a fault point takeover command to the standby AGV.

[0086] The fault self-diagnosis module automatically generates a maintenance work order and sends it to the maintenance personnel's terminal. After the maintenance is completed, a 100m path test must be performed, and the system can only be reconnected after passing the test.

[0087] Step S700: After the AGV operation is completed, determine the seeding accuracy, update the storage data synchronously, record the entire process data to the algorithm iteration database, and optimize and improve the parameters of the A algorithm and local obstacle avoidance optimization based on the algorithm iteration database.

[0088] After material placement is completed, the visual recognition unit captures an image of the target planting location. An image comparison algorithm confirms that the material quantity and location match the scheduling instructions, ensuring planting accuracy. The data interaction module synchronizes the planting completion signal to the warehouse management system, updating the material inventory data. Simultaneously, it synchronizes to the transportation management system, which generates an outbound transportation plan based on this signal, achieving seamless data flow from planting to outbound. The entire process data includes path planning data, obstacle avoidance optimization data, operation execution data, and anomaly handling data. The algorithm iteration module periodically initiates statistical analysis of the data in the algorithm iteration database, calculating the path cost deviation rate of the improved Algorithm A and the obstacle avoidance success rate of the local obstacle avoidance optimization. If the path cost deviation rate is greater than 10%, the weight coefficient of the improved Algorithm A is adjusted. If the frequency of dynamic obstacles is greater than 5 times / hour, the refresh rate of the local obstacle avoidance optimization is increased from 10 times / second to 15 times / second.

[0089] This invention improves the stability, task continuity, and self-evolution capabilities of multi-level automated warehouses in complex and demanding operating environments by enabling AGVs to extract material characteristics and switch end-effectors based on the final task scheduling instructions after outputting the instructions. It also incorporates a triple verification mechanism to complete material grabbing, transfer, and delivery. During operation, it detects anomalies or malfunctions in real time and triggers anomaly handling processes to regenerate scheduling instructions. After the operation is completed, it determines the seeding accuracy, updates storage data, and stores the entire process data in an algorithm iteration database. Based on this database, it adjusts the parameters of the improved A algorithm and local obstacle avoidance optimization. By dynamically adapting the grabbing tools, verifying operational accuracy through multiple layers, autonomously reconstructing scheduling tasks under abnormal conditions, and using actual operating data to drive algorithm parameter optimization, this invention achieves the technical effect of improving the stability, task continuity, and self-evolution capabilities of multi-level automated warehouses in complex and demanding operating environments.

[0090] In one embodiment of the present invention, the specific process of constructing the multi-dimensional mapping table is as follows:

[0091] Step S101: Connect the warehouse management system and the production execution system through the data interaction module to obtain the sowing task order and material list, extract the order urgency, the three-dimensional coordinates of the target sowing position, the material size, the material weight and the material fragility level, and generate sowing-related data;

[0092] A warehouse management system (WMS) is an information system responsible for managing warehouse operations, inventory status, and order execution. It can be used to output operational instructions such as seeding task orders. A manufacturing execution system (MES) is an industrial information system used to control and track material flow and operational execution during the manufacturing process. It can provide production-related bills of materials (BOMs) and task-related information. A seeding task order is an operational instruction that instructs the transfer of specific materials from storage locations to designated seeding locations. It can be used as the task source for automated seeding operations. A BOM is a data table listing all materials required for a task and their attributes. It can be used to provide key attributes such as material size, weight, and fragility level. Order urgency is a priority indicator reflecting the time limit for task completion and can be used to influence priority calculations in task scheduling. The three-dimensional coordinates of the target seeding location are the precise X / Y / Z coordinates of the target seeding location in the automated warehouse space, and can be used for endpoint positioning in path planning.

[0093] Step S102: Read the basic coordinates of the material storage location through the RFID tag, make corrections in conjunction with the positioning module, take images of the storage location with an industrial camera to help confirm its physical existence, and generate the three-dimensional coordinates of the physical storage location.

[0094] RFID tags are pre-installed at material storage locations on each shelf. RFID readers use radio frequency technology to read the basic coordinates of the storage location stored in the tag. These coordinates are theoretical coordinates preset during warehouse construction and have an installation deviation within ±50mm. A positioning module is invoked to correct the basic coordinates read by the RFID in real time: the UWB positioning module communicates with a pre-set positioning base station within the warehouse to calculate the actual three-dimensional coordinates of the storage location, correcting the deviation to within ±10mm to ensure a perfect match between the coordinates and the physical location. An industrial camera is activated to capture an image of the material storage location. Image recognition algorithms confirm whether the storage location actually contains material, preventing the misallocation of empty storage locations. If no material is detected, the storage location is immediately marked as empty and fed back to the warehouse management system to update the inventory status. The corrected three-dimensional coordinates and physical presence markers are then integrated to generate the three-dimensional coordinates of the material storage location.

[0095] Step S103: Collect temperature, dust concentration and ground condition inside the automated storage and retrieval system at a set frequency using environmental monitoring sensors to generate associated data for the automated storage and retrieval system.

[0096] The temperature inside the automated storage and retrieval system is detected using a thermocouple sensor, the dust concentration is detected using a laser dust sensor, and the ground condition is detected using a combination of ultrasonic and infrared sensors to check the flatness of the ground.

[0097] Step S104: Through the association unit of the data interaction module, the seeding association data, the three-dimensional coordinates of the physical storage location and the three-dimensional warehouse association data are associated and bound according to the set structure to generate a multi-dimensional mapping table.

[0098] The association unit of the data interaction module uniquely binds the above three types of data according to a preset structured format, generating a multi-dimensional mapping table. The association unit performs integrity verification on the bound data. After the verification is successful, the multi-dimensional mapping table is stored in the real-time database and synchronized to the path planning module as the sole data input source for the initial path planning.

[0099] This invention obtains seeding task orders and material lists by connecting with the warehouse management system and the production execution system. It then extracts order urgency, target seeding location 3D coordinates, material size, material weight, and material fragility level to generate seeding-related data. The basic coordinates of the material storage location are read using RFID tags and corrected using a positioning module. Simultaneously, images of the storage location are captured by an industrial camera to confirm its physical existence and generate its 3D coordinates. Environmental monitoring sensors collect data on temperature, dust concentration, and ground conditions within the automated storage and retrieval system at a set frequency to generate automated storage and retrieval system-related data. Finally, the data interaction module's association unit links the seeding-related data, the 3D coordinates of the physical storage location, and the automated storage and retrieval system-related data to generate a multi-dimensional mapping table. This allows the system to comprehensively model automated seeding tasks based on multi-dimensional information such as task attributes, material characteristics, actual location, and environmental conditions. This achieves comprehensive data modeling of the automated seeding task, improving the data integrity, accuracy, and environmental adaptability of path planning and scheduling decisions.

[0100] In one embodiment of the present invention, the construction process of the material property and path constraint mapping library is as follows:

[0101] Step S201: By connecting with the enterprise's ERP system, warehouse management system and material receiving inspection process, collect the core characteristic parameters of all categories of materials to form a standardized dataset;

[0102] The process involves integrating with the enterprise's ERP system to extract basic material information, ensuring unique and traceable material identification. It also integrates with the warehouse management system to supplement historical material storage data, assisting in the adaptation of subsequent constraint rules. The material receiving inspection process is a physical property testing and quality verification procedure performed when materials enter the warehouse, which can be used to generate accurate initial material characteristic data. In this process, a laser dimension measuring instrument collects the material's length, width, and height; an electronic scale collects the material's weight (in meters); a material testing report determines the material's fragility level; visual inspection equipment determines the material's shape; and special attributes are recorded through the material specification sheet.

[0103] Step S202: Based on the physical layout and operational safety specifications of the automated warehouse, extract the inherent and dynamic constraint parameters of all paths to form a quantifiable constraint rule base.

[0104] By analyzing 3D laser scanning and warehouse layout drawings, the physical parameters of all paths and transfer units are obtained, including channel attributes and transfer unit attributes. Path attributes are then transformed into rigid constraints corresponding to material characteristics, forming a constraint rule base, including dimensional constraints, load-bearing constraints, speed constraints, and special constraints. Inherent constraints are immutable path limitations determined by the physical structure of the automated warehouse, and can be used to define the basic accessibility of paths, such as height limits and turning radii. Dynamic constraints are path usage limitations that change with the operational process, and can be used to reflect the availability and risk level of paths within a specific time period.

[0105] Step S203: Through a multi-dimensional matching algorithm, establish a unique mapping relationship between the standardized dataset and the constraint rule base to form structured mapping entries and establish a material characteristic and path constraint mapping base with material ID as the hash index.

[0106] Multi-dimensional matching algorithms are computational methods that semantically align and logically associate material properties with path constraints. They can be used to automatically generate unique mapping entries from material IDs to path constraints. A structured mapping entry is a data unit representing the relationship between a material and its corresponding path constraint, and can be used as the basic storage unit for constructing a material property-path constraint mapping library.

[0107] Step S204: Based on the material property and path constraint mapping library, perform simulation verification through a simulation platform, and update the material property and path constraint mapping library based on the simulation verification results.

[0108] The simulation platform is a digital twin environment used to simulate the entire warehousing operation process, and can be used to verify the effectiveness and security of the mapping library in complex scenarios. Taking the initial path adaptation of a new type of large-sized home appliance upon warehousing as an example, the system obtains the basic information of the material through ERP and combines it with the actual measurement data of the warehousing inspection station to form standardized entries. The constraint rule library has defined an inherent constraint that "extra-long materials are prohibited from passing through narrow bends". A multi-dimensional matching algorithm identifies that the refrigerator belongs to the "extra-long" category and automatically maps it to a wide-aisle dedicated path set. The simulation platform simulates the material's transfer process during peak hours, discovers a conflict between the original path and the inspection vehicle's movement line, and triggers a dynamic constraint upgrade. The system updates the mapping entries accordingly, adds avoidance time restrictions, and synchronizes the optimized rules to the production environment to ensure the safe completion of the first order.

[0109] It should be noted that the specific steps for outputting the initial path using the improved A algorithm are as follows:

[0110] Step S205: Based on the multi-dimensional mapping table, retrieve the corresponding constraint parameters from the material properties and path constraint mapping library to filter out candidate paths;

[0111] The core characteristic parameters of the current material are extracted from the multi-dimensional mapping table. Based on the core characteristic parameters of the material, a unique corresponding constraint parameter is matched in the mapping library. From all available paths in the automated warehouse, paths that do not meet the above constraint parameters are eliminated, and only physically feasible and safe and compliant paths are retained to form a candidate path set.

[0112] Step S206: Based on the multi-dimensional mapping table, extract the real-time status of the inter-layer transfer unit and calculate the cross-layer transfer time;

[0113] When a transfer unit has no tasks, the basic transfer time is a fixed value, and the transfer waiting time is 0. If the transfer unit has a task queue, the estimated completion time of the preceding tasks in the queue is extracted from the multi-dimensional mapping table to obtain the transfer waiting time. The cross-layer transfer time equals the number of transfers. (Basic transit time + transit waiting time) + weight penalty time.

[0114] Step S207: Based on the improved A algorithm and combined with the cross-layer transfer time, calculate the estimated total cost of the candidate path using the path cost calculation formula, and select the candidate path with the smallest estimated total cost as the initial path.

[0115] The formula for calculating the path cost is as follows:

[0116] ,

[0117] in, This represents the estimated total cost from the starting node through the current node n to the target node. This represents the actual distance cost from the starting node to the current node n. Paths within the same layer are calculated using Manhattan distance, while cross-layer paths require the addition of the vertical distance cost between layers.

[0118] This represents the actual transfer cost from the starting node to the current node n, which is the cross-layer transfer time calculated in step S206.

[0119] This represents the actual congestion cost from the starting node to the current node n. The real-time AGV density of the road segment containing the current node n is extracted from the multi-dimensional mapping table. , = Number of AGVs in the current road segment / Maximum number of AGVs that the road segment can accommodate, Maximum number of AGVs that the road segment can accommodate = Road segment length / 1.5m, 1.5m is the width of the channel occupied by a single AGV;

[0120] This represents the heuristic distance cost from the current node n to the target node, and its calculation logic is similar to... Consistent, except that the starting node coordinates are replaced with the target seed position coordinates;

[0121] This represents the heuristic transfer cost from the current node n to the target node, and the calculation logic is the same as... Consistent with this, the transfer waiting time has been replaced with the historical average waiting time of the past 1 hour.

[0122] This represents the heuristic congestion cost from the current node n to the target node, and its calculation logic is similar to... Consistent, Replace with the 1-hour historical average AGV density;

[0123] This is the distance cost weighting coefficient. This is the weighting coefficient for transportation costs. This is the congestion cost weighting coefficient.

[0124] This invention uses a multi-dimensional mapping table to retrieve constraint parameters to filter candidate paths, extracts the real-time status of inter-level transfer units to calculate cross-level transfer time, combines this time with the path cost calculation formula to evaluate the estimated total cost of each candidate path, and selects the path with the minimum cost as the initial path. By integrating material characteristics, real-time equipment status, and a multi-dimensional weighted cost model, it achieves refined modeling of path decision-making. Through dynamic weight configuration and collaborative calculation of actual / heuristic costs, it improves the algorithm's adaptability to the three-dimensional warehousing environment, achieving the technical effect of generating high-quality initial paths that balance transportation safety, vertical movement efficiency, and traffic load balance under high-density operation conditions.

[0125] In one embodiment of the present invention, the specific process of the output area task allocation plan is as follows:

[0126] Step S301: Collect the load status of AGVs on each floor of the automated warehouse in real time through several distributed regional scheduling nodes, including the remaining power of the AGVs and the total time of the assigned tasks, and monitor their own CPU utilization, and generate load data to be fed back to the global scheduling center in real time.

[0127] By using distributed regional scheduling nodes divided according to the floors of the automated warehouse, the load status of AGVs and their own computing power data within the jurisdiction are collected in real time, forming a load data pool for the entire system. Each regional node feeds back the collected AGV load status and its own CPU utilization rate to the global scheduling center in real time via 5G industrial Ethernet. The global scheduling center stores the data in a real-time database according to the structured format of "region number-timestamp-AGVID-load parameter-computing power parameter", forming a dynamically updatable "global load view" to provide data support for subsequent computing power judgment and task allocation.

[0128] Step S302: Based on load data and initial path, when the CPU utilization of a certain distributed area scheduling node is greater than the preset value, it is determined to be computing power overload. Extract the list of tasks currently being processed by the distributed area scheduling node and filter out non-core tasks.

[0129] The global scheduling center sets a CPU utilization threshold and checks each node in each region individually. If a node's CPU utilization is ≥80%, it is considered overloaded, and delays are occurring in processing the current task, requiring task migration to alleviate the pressure. If the node's CPU utilization is 20% ≤ 80%, it is considered "normal load," maintaining the current task processing status without migration. If the node's CPU utilization μ < 20%, it is considered "low load" and marked as a potential migration target for subsequent non-core tasks. For nodes with overloaded computing power, the global scheduling center extracts a list of all currently processed tasks and, combined with material characteristics and order urgency from a multi-dimensional mapping table, filters out non-core tasks.

[0130] Step S303: The global scheduling center selects distributed regional scheduling nodes with CPU utilization below a preset value from all distributed regional scheduling nodes, and migrates the data packets of the selected non-core tasks to the distributed regional scheduling nodes with CPU utilization below the preset value via Ethernet.

[0131] The global scheduling center matches low-load regional nodes or cloud computing power nodes across the entire system and completes the migration of non-core tasks through a reliable data transmission mechanism, ensuring task continuation and data consistency. The global scheduling center selects migration targets according to the principle of "regional proximity → computing power redundancy", giving priority to matching low-load regional nodes on the same floor or adjacent floor as the overloaded nodes to reduce cross-regional data transmission latency. If there are not enough low-load regional nodes, cloud computing power nodes are called up and connected to the global scheduling center through public network dedicated lines to ensure the flexibility of computing power replenishment.

[0132] Step S304: After the task migration is completed, the task allocation information of all distributed regional scheduling nodes is integrated through the global scheduling center to generate a regional task allocation plan that includes the number of AGV tasks allocated to each region, the task type, and the expected start time.

[0133] The global scheduling center integrates the task processing status of all regional nodes to generate structured regional task allocation plans, providing a task allocation benchmark for subsequent local obstacle avoidance optimization. For example, during the midday peak operation period, the CPU utilization of a third-layer distributed scheduling node surged to 90% due to handling a large number of seeding tasks, triggering a computing power overload judgment. The node immediately reported its status, and after receiving the feedback, the global scheduling center identified and migrated the 12 non-core replenishment tasks it was processing to idle scheduling nodes in the first and fifth layers with CPU utilization below 20%. After the migration was completed, each region resynchronized its task list, and the global scheduling center generated a new regional task allocation plan based on this, including the number and type of tasks for each AGV and the estimated start time, avoiding the overall efficiency decline caused by local congestion.

[0134] This invention uses several distributed regional scheduling nodes to collect the load status of AGVs on each floor of an automated warehouse in real time and monitor their own CPU utilization, generating load data that is fed back to the global scheduling center. Based on the load data and the initial path, when the CPU utilization exceeds a preset value, it is determined that the computing power is overloaded, and non-core tasks are filtered out. The global scheduling center selects idle scheduling nodes with low CPU utilization and migrates non-core tasks to the target nodes using Ethernet. After the migration is completed, the global scheduling center integrates the task allocation of each node to generate a regional task allocation plan. By using dual indicators to collaboratively monitor and identify local computing power bottlenecks, non-critical tasks are automatically removed and cross-node computing power is rebalanced, ultimately forming a unified and coordinated task allocation scheme. This can improve the stability and responsiveness of the system in high-concurrency scenarios and enhance the elasticity and fault tolerance of the scheduling architecture.

[0135] It should be noted that the specific process of the local obstacle avoidance optimization is as follows:

[0136] Step 305: Collect obstacle distance data, ground image data, and moving target data to generate multi-source perception data, and form a dynamic environment map through local path planning units;

[0137] The system collects distance and contour data of obstacles around the AGV using lidar, generates obstacle distance data, collects ground image data using a panoramic camera, collects moving target data using fixed sensing devices inside the AGV, and unifies the format of the above data through a local path planning unit to generate multi-source sensing data. After fusion processing, it generates a frequently updated dynamic environment map, achieving accurate depiction of obstacle status.

[0138] Step 306: Based on the regional task allocation plan and dynamic environment map, perform overlay analysis through the collision detection module;

[0139] The collision detection module is an analysis component used to evaluate whether there is spatial overlap between the AGV's predetermined path and obstacles in the dynamic environment map. It can be used to identify potential conflict points in advance and trigger corresponding obstacle avoidance strategies.

[0140] Step 307: If a collision risk is detected, calculate the detour distance. If the detour distance is less than the preset value, use the dynamic window method combined with reinforcement learning to adjust the AGV parameters and output the optimized AGV parameters.

[0141] The detour distance is the additional distance required to bypass an obstacle from the current position when the main path is blocked. It can be used as a basis for decision-making regarding local adjustments or global replanning. The dynamic window method combined with reinforcement learning is a speed-space constraint obstacle avoidance technique that integrates classical control methods with intelligent learning mechanisms. It can be used to achieve smooth obstacle avoidance under real-time parameter adjustments and adapt to complex local scenarios. The collision detection method is as follows: The planned path node sequence of a single AGV and the estimated arrival time of each node are extracted from the regional task allocation plan. Obstacle information within a 5m radius of the AGV's planned path at the current timestamp is extracted from the dynamic environment map. If the AGV travels along the planned path and the distance to an obstacle at a certain moment is <0.5m, and the time difference is ≤2s, then a collision risk is identified. The specific method for adjusting AGV parameters is as follows: Extract the ground status from the dynamic environment map. If it is an "oil slick area", the AGV's driving speed is reduced to 0.7 times the original speed, and the braking distance is extended to 1.5 times the original braking distance. If it is a "seam protrusion area", the AGV's driving speed is reduced to 0.5 times the original speed. Introduce a reward function to filter the parameter combination with the minimum detour distance and the minimum speed loss, output the optimized AGV driving speed, direction, and braking distance parameters, and send them to the AGV controller for execution.

[0142] Step 308: If the detour distance is greater than the preset value, the position and movement parameters of the obstacle are fed back to the global scheduling center to trigger incremental replanning and form an optimized AGV path list.

[0143] Incremental replanning only replans the "obstacle-affected segments" rather than the entire path, reducing computational consumption. It adopts an improved A algorithm, combined with obstacle constraints in the dynamic environment map, to recalculate the node sequence of the path segment, generate a new road segment, and splice the new road segment with the "non-affected segments" of the original planned path to form a complete optimized AGV path list, avoiding the time waste caused by full path replanning.

[0144] Step 309: Integrate the optimized AGV parameters and AGV path list through the local path planning unit to generate a task allocation table containing the optimized path and estimated arrival time of each AGV.

[0145] In one specific embodiment, during the concentrated outbound phase at night, a toolbox temporarily stored in a certain aisle for equipment maintenance was identified as a static obstacle by the LiDAR. The local path planning unit integrates this distance data with images of water accumulation on the ground captured by a camera to construct a dynamic environment map. The collision detection module detects that the original path will pass through this area and calculates the detour distance. Since the detour requires crossing three layers of shelving aisles, the distance exceeds a preset threshold. The system feeds back the obstacle coordinates and the blocked area to the global scheduling center, triggering incremental replanning. The scheduling center reallocates adjacent available paths and issues a new path list. Simultaneously, the affected AGVs smoothly dock and await instruction updates based on deceleration and fine-tuning steering parameters output by the dynamic window method combined with reinforcement learning. After the new path is issued, the local unit integrates the parameters and path to generate a task allocation table containing the estimated arrival time, ensuring that the overall operation rhythm is not significantly affected.

[0146] It should be noted that the calculation process for the task priority is as follows:

[0147] Based on the task allocation table, extract the order urgency, material characteristics, and remaining time for each task, and substitute them into the task priority calculation formula to calculate the task priority, as follows:

[0148] ,

[0149] in, This represents the task priority value. The larger the value, the higher the task priority. This is the order urgency coefficient, which corresponds one-to-one with the order urgency level. Urgency level 5 → α=5.0, level 4 → α=4.0, level 3 → α=3.0, level 2 → α=2.0, level 1 → α=1.0. The assignment logic is that "for every increase of urgency level, the coefficient increases by 1.0", ensuring that the priority weight of urgent orders is significantly higher than that of ordinary orders. The material sensitivity coefficient distinguishes between sensitive materials and ordinary materials. Materials with "fragile grade ≥ 4" or "extra-long materials" → β = 1.5, and other ordinary materials → β = 1.0. The assignment logic is that "sensitive materials need to have an additional priority weight" to avoid damage or delays in transportation of sensitive materials due to resource contention. The task timeout risk coefficient is linked to the remaining time of the task. If the remaining time is ≤1 hour, γ=2.0; if 1 hour < remaining time ≤2 hours, γ=1.0; if the remaining time is >2 hours, γ=0.5. The assignment logic is that "the shorter the remaining time, the higher the timeout risk and the larger the coefficient", which forces the system to prioritize tasks that are about to expire, thereby reducing the risk of order default.

[0150] In one embodiment of the present invention, the specific process of outputting the final task scheduling instruction is as follows:

[0151] Step S401: Based on the task priority and task allocation table of all tasks, allocate resources matching path attributes to each task in descending order of priority, and output a preliminary resource allocation scheme.

[0152] Resources are allocated in descending order of task priority. High-priority tasks are given priority to use "premium resources", including wide channels, idle inter-layer transfer units, and AGVs with ≥80% remaining power. Low-priority tasks are allocated "regular resources", including standard channels, idle transfer units, and AGVs with ≥50% remaining power.

[0153] Step S402: Based on the preliminary resource allocation scheme, compare the optimized paths of all tasks. If the AGV path of the low-priority task intersects with the AGV path of the high-priority task and the time overlaps, schedule the low-priority AGV to stop at the nearest avoidance zone, and continue driving after the high-priority task passes. Output the updated resource allocation scheme.

[0154] The system retrieves the AGV path node coordinates and estimated arrival times from the task allocation table and compares the paths of different AGVs. If the AGV paths of low-priority tasks and high-priority tasks intersect at a node, and the time difference between their arrival times at that node is ≤2, a path conflict is identified. Following the principle of lower-priority tasks yielding to higher-priority tasks, if both conflicting tasks are high-priority, the task with the shorter remaining time takes priority; if both are low-priority, the task allocated resources first takes priority. For low-priority AGVs that need to yield, the system automatically matches the nearest yield zone from the warehouse layout database and generates a docking yield zone instruction.

[0155] Step S403: Based on the task priority of all tasks, bind a spare AGV to the task whose task priority is greater than the preset value, and output the redundancy configuration table;

[0156] Configure 10% of the total number of AGVs in the automated warehouse as standby AGVs. The standby AGVs must be in standby mode. Bind one standby AGV to each critical task to form an association table of "task ID-primary AGVID-standby AGVID" and store it in the redundancy configuration unit of the task scheduling module to ensure that the standby resources can be quickly located in case of failure.

[0157] Step S404: Integrate the preliminary resource allocation scheme, task allocation table, and redundancy configuration table into a structured final task scheduling instruction.

[0158] See Figure 2 The present invention also provides an automatic seeding path optimization and scheduling system for multi-layer automated warehouses, comprising:

[0159] The data interaction module includes multi-protocol interfaces, multi-source data acquisition units, data association units, and a real-time database. It achieves data transmission through 5G industrial Ethernet, providing basic data support for the entire system.

[0160] The multi-protocol interface supports two commonly used industrial protocols, HTTP and MQTT, which respectively connect to the warehouse management system, production execution system, and transportation management system. It extracts seeding task orders and bill of materials data from the warehouse management system and production execution system via HTTP, and synchronizes seeding completion information to the transportation management system via MQTT, thus resolving protocol barriers for data interaction between different systems. The multi-source data acquisition unit includes multi-source unit cells and environmental monitoring sensors, collecting material storage location coordinates and warehouse environmental data at a set frequency to ensure data real-time performance. The data association unit generates a multi-dimensional mapping table, binding data according to the structure of "material ID - seeding association data - storage location coordinates - environmental data," and automatically verifies data integrity.

[0161] The path planning module includes a global path planning unit and a local path planning unit. It interacts with the data interaction module in real time and outputs the initial path, regional task allocation plan, and optimized path list.

[0162] The global path planning unit is responsible for initial path planning and global computing power scheduling, while the local path planning unit deploys dynamic windowing and reinforcement learning algorithms and is responsible for local obstacle avoidance optimization.

[0163] The task scheduling module includes a priority calculation unit, a resource allocation unit, and a redundancy configuration unit. It receives the output data from the path planning module and outputs the final task scheduling instruction to the automatic seeding execution module.

[0164] The priority calculation unit is used to calculate the priority of each task, the resource allocation unit is used to allocate resources according to the priority, and the redundancy configuration unit is used to configure backup AGVs for high-priority tasks.

[0165] The automatic seeding execution module includes an intelligent AGV, a modular end effector, and a vision recognition unit. It receives instructions from the task scheduling module and completes material grabbing, transfer, and delivery.

[0166] The automated seeding execution module serves as an operational terminal for the physical handling and placement of materials, translating scheduling instructions into actual operational actions. Equipped with a modular end effector on an intelligent AGV, the module precisely grasps and places materials under the guidance of a vision recognition unit.

[0167] The environmental monitoring and fault self-diagnosis module includes an environmental data monitoring unit, a hardware fault detection unit, a backup resource scheduling unit, and a maintenance work order generation unit, which handles anomalies and faults.

[0168] The environmental data monitoring unit collects real-time data on warehouse temperature, dust concentration, and ground conditions, synchronizing the data to the data interaction module to provide environmental constraints for path planning. The hardware fault detection unit monitors the status of intelligent AGVs and inter-level transfer units in real time, generating fault codes containing "faulty equipment ID, type, and location." Upon receiving the fault code, the backup resource scheduling unit calls upon the backup AGV or backup transfer unit from the task scheduling module within 10 seconds to take over the tasks of the faulty equipment, ensuring uninterrupted operation. The maintenance work order generation unit automatically generates maintenance work orders based on the fault codes, including fault cause analysis, required repair parts models, and repair steps, and sends them to the maintenance personnel's terminal.

[0169] The above description is only a preferred embodiment of the present invention. It should be noted that those skilled in the art can make several improvements and modifications without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for optimizing and scheduling the automatic seeding path in a multi-layer automated storage and retrieval system, characterized in that: Includes the following steps: The data interaction module acquires sowing-related data and automated warehouse-related data to construct a multi-dimensional mapping table; the sowing-related data is obtained from the warehouse management system and the production execution system, including sowing task orders and bills of materials; The associated data for automated storage and retrieval systems includes the three-dimensional coordinates of the material storage locations and environmental data within the automated storage and retrieval system. Based on a multi-dimensional mapping table, a pre-defined material property and path constraint mapping library is called. An improved A algorithm is used to filter paths that meet the constraints and calculate path costs, and output the initial path. The improved A algorithm is based on the traditional Based on the algorithm, a path search method with a multi-objective cost function is introduced to quantify the estimated total cost of each candidate path and finally select the candidate path with the minimum estimated total cost as the initial path. Based on the initial path, the real-time load status of the AGV is input, and the regional task allocation plan is output through the global scheduling center. Combined with multi-source perception data, the optimized task allocation table is output through local obstacle avoidance optimization. Based on the task allocation table, calculate the task priority, allocate resources according to the priority and handle path conflicts, configure backup AGVs, and output the final task scheduling instruction. The specific process for constructing the multi-dimensional mapping table is as follows: By connecting the data interaction module with the warehouse management system and the production execution system, the system can obtain sowing task orders and material lists, extract order urgency, target sowing position 3D coordinates, material size, material weight and material fragility level, and generate sowing-related data. The basic coordinates of the material storage location are read by RFID tags, corrected by the positioning module, and the physical existence is confirmed by taking images of the storage location with an industrial camera to generate the three-dimensional coordinates of the physical storage location. The temperature, dust concentration and ground condition inside the automated storage and retrieval system are collected at a set frequency by environmental monitoring sensors to generate associated data for the automated storage and retrieval system. By using the association unit of the data interaction module according to the set structure, the seeding association data, the three-dimensional coordinates of the physical storage location, and the three-dimensional warehouse association data are associated and bound together to generate a multi-dimensional mapping table.

2. The method for optimizing and scheduling the automatic seeding path of a multi-layered automated warehouse according to claim 1, characterized in that, After outputting the final task scheduling instruction, the following steps are also included: Based on the final task scheduling instruction, the AGV extracts material characteristics and switches the end execution head, and completes material grabbing, transfer and delivery after triple verification; When an abnormality is detected during AGV operation, the abnormality handling process is triggered, and a new final task scheduling instruction is regenerated. After the AGV operation is completed, the seeding accuracy is determined, the warehouse data is updated synchronously, and the entire process data is recorded in the algorithm iteration database. Based on the algorithm iteration database, the parameters of the A algorithm and the local obstacle avoidance optimization are optimized and improved.

3. The method for optimizing and scheduling the automatic seeding path of a multi-layered automated warehouse according to claim 1, characterized in that, The construction process of the material property and path constraint mapping library is as follows: By connecting with enterprise ERP systems, warehouse management systems and material receiving and inspection processes, core characteristic parameters of all categories of materials are collected to form a standardized dataset. Based on the physical layout and operational safety specifications of the automated warehouse, the inherent and dynamic constraint parameters of all paths are extracted to form a quantifiable constraint rule base. By using a multi-dimensional matching algorithm, a unique mapping relationship is established between the standardized dataset and the constraint rule base, forming structured mapping entries, and a material characteristic and path constraint mapping base is established with the material ID as the hash index. Based on the material property and path constraint mapping library, simulation verification is performed through a simulation platform, and the material property and path constraint mapping library is updated based on the simulation verification results.

4. The method for optimizing and scheduling the automatic seeding path of a multi-layered automated warehouse according to claim 1, characterized in that, The specific steps for outputting the initial path using the improved A algorithm are as follows: Based on the multi-dimensional mapping table, the corresponding constraint parameters are retrieved from the material properties and path constraint mapping library to filter out candidate paths. Based on a multi-dimensional mapping table, the real-time status of the inter-layer transfer unit is extracted, and the cross-layer transfer time is calculated. Based on the improved A algorithm, combined with cross-layer transfer time, the estimated total cost of candidate paths is calculated using the path cost calculation formula, and the candidate path with the minimum estimated total cost is selected as the initial path. The formula for calculating the path cost is as follows: ， in, This represents the estimated total cost from the starting node through the current node n to the target node. This represents the actual distance cost from the starting node to the current node n. This represents the actual transit cost from the starting node to the current node n. This represents the actual congestion cost from the starting node to the current node n. This represents the heuristic distance cost from the current node n to the target node. This represents the heuristic transfer cost from the current node n to the target node. Let represent the heuristic congestion cost from the current node n to the target node. This is the distance cost weighting coefficient. This is the weighting coefficient for transportation costs. This is the congestion cost weighting coefficient.

5. The method for optimizing and scheduling the automatic seeding path of a multi-layered automated warehouse according to claim 1, characterized in that, The specific process of the output area task allocation plan is as follows: The load status of AGVs on each floor of the automated warehouse is collected in real time through several distributed regional scheduling nodes, including the remaining power of the AGVs and the total time of the assigned tasks, and its own CPU utilization is monitored. The load data is then fed back to the global scheduling center in real time. Based on load data and initial path, when the CPU utilization of a certain distributed area scheduling node exceeds the preset value, it is determined to be computing power overload. The list of tasks currently being processed by the distributed area scheduling node is extracted, and non-core tasks are filtered out. The global scheduling center filters out distributed regional scheduling nodes with CPU utilization below a preset value from all distributed regional scheduling nodes, and migrates the data packets of the filtered non-core tasks to the distributed regional scheduling nodes with CPU utilization below the preset value via Ethernet. Once the task migration is complete, the task allocation information of all distributed regional scheduling nodes is integrated through the global scheduling center to generate a regional task allocation plan that includes the number of AGV tasks allocated to each region, the task type, and the expected start time.

6. The method for optimizing and scheduling the automatic seeding path of a multi-layered automated warehouse according to claim 1, characterized in that, The specific process of local obstacle avoidance optimization is as follows: Collect obstacle distance data, ground image data, and moving target data to generate multi-source perception data, and form a dynamic environment map through local path planning units; Based on the regional task allocation plan and dynamic environment map, overlay analysis is performed through the collision detection module; If a collision risk is detected, the detour distance is calculated. If the detour distance is less than the preset value, the AGV parameters are adjusted using a dynamic window method combined with reinforcement learning, and the optimized AGV parameters are output. If the detour distance exceeds the preset value, the position and movement parameters of the obstacle will be fed back to the global scheduling center, triggering incremental replanning and forming an optimized AGV path list; By integrating and optimizing the AGV parameters and AGV path list through local path planning units, a task allocation table is generated that includes the optimized path and estimated arrival time of each AGV.

7. The method for optimizing and scheduling the automatic seeding path of a multi-layered automated warehouse according to claim 1, characterized in that, The calculation process for the task priority is as follows: Based on the task allocation table, extract the order urgency, material characteristics, and remaining time for each task, and substitute them into the task priority calculation formula to calculate the task priority, as follows: ， in, This represents the task priority value. The larger the value, the higher the task priority. The urgency level of the order. For material sensitivity coefficient, This represents the risk factor for task timeout.

8. The method for optimizing and scheduling the automatic seeding path of a multi-layered automated warehouse according to claim 7, characterized in that, The specific process for outputting the final task scheduling instruction is as follows: Based on the task priorities and task allocation table of all tasks, resources matching the path attributes are allocated to each task in descending order of priority, and a preliminary resource allocation scheme is output. Based on the initial resource allocation scheme, compare the optimized paths of all tasks. If the AGV path of a low-priority task intersects with the AGV path of a high-priority task and the time overlaps, schedule the low-priority AGV to stop at the nearest avoidance zone and continue driving after the high-priority task has passed. Output the updated resource allocation scheme. Based on the task priority of all tasks, bind a spare AGV to the task with a priority greater than the preset value, and output a redundancy configuration table; The initial resource allocation scheme, task allocation table, and redundancy configuration table are integrated into a structured final task scheduling instruction.

9. A multi-layer automated seeding path optimization and scheduling system for warehouses, used to implement the multi-layer automated seeding path optimization and scheduling method for warehouses as described in any one of claims 1-8, characterized in that, include: The data interaction module includes multi-protocol interfaces, multi-source data acquisition units, data association units, and a real-time database. It achieves data transmission through 5G industrial Ethernet, providing basic data support for the entire system. The path planning module includes a global path planning unit and a local path planning unit. It interacts with the data interaction module in real time and outputs the initial path, regional task allocation plan, and optimized path list. The task scheduling module includes a priority calculation unit, a resource allocation unit, and a redundancy configuration unit. It receives the output data from the path planning module and outputs the final task scheduling instruction to the automatic seeding execution module. The automatic seeding execution module includes an intelligent AGV, a modular end effector, and a vision recognition unit. It receives instructions from the task scheduling module and completes material grabbing, transfer, and delivery. The environmental monitoring and fault self-diagnosis module includes an environmental data monitoring unit, a hardware fault detection unit, a backup resource scheduling unit, and a maintenance work order generation unit, which handles anomalies and faults. The multi-protocol interface supports two commonly used industrial protocols, HTTP and MQTT, which respectively connect to the warehouse management system, production execution system, and transportation management system. It extracts seeding task orders and bill of materials data from the warehouse management system and production execution system via HTTP, and synchronizes seeding completion information to the transportation management system via MQTT, thus overcoming protocol barriers in data interaction between different systems. The multi-source data acquisition unit includes multi-source unit cells and environmental monitoring sensors, collecting material storage location coordinates and warehouse environmental data at a set frequency to ensure real-time data accuracy. The data association unit generates a multi-dimensional mapping table, binding data according to the structure of material ID, seeding association data, storage location coordinates, and environmental data, and automatically verifying data integrity. The environmental data monitoring unit collects real-time data on warehouse temperature, dust concentration, and ground conditions, synchronizing the data to the data interaction module to provide environmental constraints for path planning. The hardware fault detection unit monitors the status of intelligent AGVs and inter-level transfer units in real time, generating fault codes containing the faulty equipment ID, type, and location. Upon receiving the fault code, the backup resource scheduling unit calls upon the backup AGV or backup transfer unit from the task scheduling module within 10 seconds to take over the tasks of the faulty equipment, ensuring uninterrupted operation. The maintenance work order generation unit automatically generates maintenance work orders based on the fault codes, including fault cause analysis, required repair parts models, and repair steps, and sends them to the maintenance personnel's terminal.

Images