Intelligent three-dimensional warehousing system based on dynamic partitioning and real-time mapping

Abstract

The invention relates to the technical field of intelligent warehousing, and discloses an intelligent three-dimensional warehousing system based on dynamic partitioning and real-time mapping. The inventory partition module of the three-dimensional warehouse obtains physical feature information of warehoused articles, classifies and judges storage partitions according to storage attributes; the storage location mapping module extracts space coordinates of the storage partitions and constructs a mapping relation with storage location coordinates; a path combination module extracts a high-frequency access area from the mapping relation, calculates the shortest path matching degree between storage location coordinates, and generates a fork deviation path; the dynamic updating module verifies the coordinates of the storage location in real time on the basis of the deviation path, identifies the displacement change of the article and updates the mapping relation; and the scheduling management module determines a storage target and a carrying sequence of the warehouse-in articles according to the updated mapping relationship, and generates a warehouse scheduling database. The storage can optimize storage partition and path planning, update the storage location information in real time, and improve the storage operation efficiency and flexibility.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent warehousing, in particular to an intelligent stereoscopic warehousing system based on dynamic partitioning and real-time mapping. BACKGROUND

[0002] Under the background of the rapid development of modern logistics and manufacturing industry, the warehousing system as the core link of the flow of materials, its operation efficiency directly affects the turnover speed of the whole supply chain. The traditional stereoscopic warehouse adopts the operation mode of fixed partitioning and preset path, when facing the diversified storage demand of goods, the problem of mismatch between storage partitioning and goods attributes often occurs. For example, some warehousing systems only roughly partition goods according to the size or weight of goods, without fully considering the key storage attributes of goods such as access frequency and shelf life, which leads to the high-frequency access goods being allocated to the storage location far away from the operation channel, increasing the forklift carrying path length and operation time, and reducing the overall turnover efficiency of the warehouse.

[0003] The traditional warehouse management adopts a static mapping method, that is, once the correspondence between the coordinates of the storage location and the stored goods is determined, it will be fixed for a long time, and it cannot respond to the displacement changes of the goods in the storage location in real time. In actual operation, due to slight deviation in the forklift carrying process, position adjustment in the goods stacking process and other factors, the actual position of the goods in the storage location often differs from the initial recorded coordinates. If this difference cannot be updated in time, it will be difficult for the forklift to accurately locate the goods in the subsequent access operation, not only increasing the operation failure rate, but also possibly prolonging the operation cycle due to repeated position adjustment.

[0004] The traditional warehouse lacks targeted optimization of high-frequency access areas in path planning, and usually uses a unified shortest path algorithm to calculate the forklift path without considering the access frequency difference of different areas. When multiple forklifts are operating in the warehouse space at the same time, the high-frequency access area is prone to path congestion, which increases the waiting time of the forklift, further reducing the operation efficiency. The scheduling management module also generates a carrying sequence based on the fixed mapping relationship of the storage location, which cannot dynamically adjust the operation plan according to the real-time updated storage location information, and is difficult to adapt to the dynamic changes of the goods access demand in the warehouse system, ultimately affecting the overall operation flexibility and efficiency of the warehouse system. SUMMARY

[0005] The present application aims to provide an intelligent stereoscopic warehousing system based on dynamic partitioning and real-time mapping to solve the problems raised in the background.

[0006] To achieve the above-mentioned purpose, the present application provides an intelligent stereoscopic warehousing system based on dynamic partitioning and real-time mapping, which comprises:

[0007] The inventory partition module is configured to acquire physical characteristic information of the warehousing goods, classify the goods according to storage attributes of the goods, and determine storage partitions of the goods relative to the warehouse space;

[0008] The storage location mapping module is configured to extract space coordinates of the storage partitions and construct a mapping relationship between the storage partitions and the storage location coordinates;

[0009] The path combination module is configured to extract high-frequency access areas from the storage partitions and the storage location coordinates, calculate a shortest path matching degree between the storage location coordinates according to the high-frequency access areas, and generate a fork bias path;

[0010] The dynamic updating module is configured to perform real-time verification on the storage location coordinates based on the fork bias path, identify displacement changes of the goods in the storage location coordinates, and update the mapping relationship between the storage partitions and the storage location coordinates according to the displacement changes;

[0011] The scheduling management module is configured to determine storage targets and handling sequences of the warehousing goods according to the updated mapping relationship between the storage partitions and the storage location coordinates, and generate a warehouse scheduling database.

[0012] Preferably, the implementation of the inventory partition module comprises:

[0013] For any physical characteristic information of the warehousing goods, a pre-trained attribute classification model is called;

[0014] The physical characteristic information is hierarchically divided by the attribute classification model, and at least one attribute category is acquired;

[0015] The size-related items, weight-related items and material-related items in the physical characteristic information corresponding to the attribute category are identified to form a feature sample set;

[0016] The size-related items, weight-related items and material-related items in the feature sample set are respectively clustered to sequentially generate a size clustering area, a weight analysis area and a material association area as the storage partitions of the goods relative to the warehouse space.

[0017] Preferably, the implementation of generating the size clustering area, the weight analysis area and the material association area further comprises:

[0018] The size-related items, weight-related items and material-related items in the feature sample set are merged according to the attribute categories to obtain a plurality of attribute merging results;

[0019] The size similarity pairs in the attribute merging results are extracted, and the size clustering area is generated after comparison with a standard size library;

[0020] The distribution dispersion of the weight-related items and the combination strength of the material-related items in the attribute merging results are calculated, and the weight analysis area and the material association area are divided according to the distribution dispersion and the combination strength.

[0021] Preferably, the implementation of the warehouse position mapping module comprises:

[0022] Call the historical warehouse position coordinates and the current warehouse position state associated with the storage partition to generate an unlabeled warehouse position recognition result;

[0023] Determine whether the unlabeled warehouse position recognition result meets the idle condition, and if so, take the warehouse position recognition result meeting the condition as the warehouse position coordinates of the storage partition.

[0024] Preferably, the implementation of the warehouse position mapping module comprises:

[0025] Take the feature representation of the size clustering area, the weight analysis area and the material association area as input, and take the spatial attribute and the load-bearing level of the warehouse position coordinates as output to establish a multi-dimensional mapping relationship from the storage partition to the warehouse position coordinates.

[0026] Preferably, the implementation of the path combination module comprises:

[0027] Hierarchical cluster the storage partition and the warehouse position coordinates according to the access frequency, the path length and the number of turns, and set the maximum cluster center as a high-frequency access area;

[0028] Extract the spatial nodes of the high-frequency access area, calculate the Manhattan distance between each node, and generate a public path sequence related to the distance;

[0029] Extract adjacent warehouse position coordinates based on the public path sequence, set the shortest common sub-path between each warehouse position, and take the sub-path length value as the path matching degree;

[0030] Generate a fork bias path according to the path matching degree and the time access weight.

[0031] Preferably, the implementation of the dynamic updating module comprises:

[0032] Extract the time access weight of each warehouse position coordinate from the fork bias path;

[0033] Divide the target trajectory of the fork bias path according to the time period corresponding to the time access weight;

[0034] Compare the access density of the target trajectory in each time period with the actual item distribution of the warehouse position coordinates, mark the difference value, and update the warehouse position coordinate state according to the difference value.

[0035] Preferably, the implementation of the scheduling management module comprises:

[0036] According to the updated mapping relationship between the storage partition and the warehouse position coordinates, extract the storage priority of the warehousing items;

[0037] Generate a carrying sequence according to the time urgency and the path optimization coefficient of the storage priority;

[0038] The storage scheduling database is generated by encoding the storage priority and the carrying sequence in a tree structure.

[0039] Preferably, the intelligent stereoscopic warehouse further comprises:

[0040] The real-time position data of the incoming goods is time-aligned with the predicted path of the warehouse scheduling database.

[0041] When the actual path deviates from the predicted trajectory, the re-mapping of the bin coordinate and the iterative optimization of the fork deviation path are triggered.

[0042] Preferably, according to the intelligent stereoscopic warehouse system based on dynamic partitioning and real-time mapping described above, the implementation method for updating the bin coordinate state further comprises:

[0043] The bin coordinates are dynamically classified based on the difference value, and the classification result is fed back to the inventory partitioning module to trigger the re-division of the storage partition.

[0044] Compared with the prior art, the present application has the following advantages:

[0045] The bin mapping module extracts the spatial coordinates of the storage partition and constructs a mapping relationship with the bin coordinates, providing a basis for the accurate management of the warehouse system. Compared with the static bin record method in traditional warehouses, this mapping relationship clearly presents the correspondence between each storage partition and the specific bin coordinates, allowing management personnel to quickly query the bin location of any goods. It also provides accurate coordinate basis for the precise positioning of the fork, reduces positioning errors caused by ambiguous bin information, improves the precision of fork access operations, reduces the risk of collision during the operation process, and ensures the safety of warehouse equipment and goods.

[0046] The path combination module calculates the shortest path matching degree between bin coordinates for high-frequency access areas and generates a fork deviation path, which can significantly optimize the fork operation path. In traditional warehouses, path planning does not fully consider the frequency of area access, which can easily cause congestion in high-frequency areas. The deviation path generated by the module can guide the fork to preferentially select the route that passes through the high-frequency access area and has the shortest path, reducing the waiting time of the fork in the high-frequency area, while avoiding the extension of the operation time caused by the long path. When multiple forks are operating simultaneously, the deviation path can also disperse the operation routes of the forks to a certain extent, reducing the probability of congestion of different forks in the same area, and improving the smoothness of the overall operation process.

[0047] The dynamic updating module is based on the forklift deflection path to perform real-time checking on the storage location coordinates and identify the displacement change of the goods, and then update the mapping relationship between the storage partition and the storage location coordinates, effectively solving the drawbacks of the static library information in traditional warehousing. In actual operation, the displacement change of the goods is difficult to avoid, and the dynamic updating module can capture this change in real time and update the mapping relationship synchronously, ensuring that the storage location information is always consistent with the actual location of the goods. This enables the forklift to accurately position the goods based on the latest storage location information during each access operation, eliminating the need for repeated position adjustments, reducing the likelihood of operational errors, and shortening the time required for a single operation. At the same time, it provides reliable real-time data for the subsequent scheduling management module to develop operational plans, avoiding scheduling deviations caused by outdated storage location information.

[0048] The scheduling management module determines the storage target and handling sequence of the incoming goods based on the updated mapping relationship and generates a warehouse scheduling database, significantly improving the operational flexibility and dynamic adaptability of the warehouse system. Traditional warehouse scheduling plans are typically based on fixed storage location information and are difficult to adapt to changes in goods access demand. However, this module can dynamically adjust the storage target and handling sequence based on real-time updated storage location information. For example, if the access frequency of goods in a certain storage partition suddenly increases, the scheduling management module can promptly allocate subsequent incoming goods of the same type to adjacent idle storage locations to prevent overcrowding in that partition. At the same time, the generated warehouse scheduling database can record and analyze historical operational data, providing a reference for subsequent optimization of storage partitions and adjustment of path planning, enabling the warehouse system to continuously adapt to changes in business demand and maintain a high-efficiency and stable operational state in the long term. BRIEF DESCRIPTION OF DRAWINGS

[0049] Figure 1 The working principle diagram of the intelligent vertical warehouse with upper storage and lower picking described in the present application;

[0050] Figure 2 The flowchart for the inventory partition module;

[0051] Figure 3 The flowchart generated for the size clustering area, weight analysis area, and material association area. DETAILED DESCRIPTION

[0052] The technical solutions in the embodiments of the present application will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0053] Please refer to Figure 1The application provides an intelligent stereoscopic warehouse system based on dynamic partitioning and real-time mapping, which comprises the cooperative operation of an inventory partition module, a storage location mapping module, a path combination module, a dynamic updating module and a scheduling management module.

[0054] The inventory partition module obtains physical characteristic information of the warehousing goods, classifies the goods according to the storage attributes of the goods, and determines the storage partition of the goods relative to the warehouse space. The storage location mapping module extracts the spatial coordinates of the storage partition, and establishes the mapping relationship between the storage partition and the storage location coordinates. The path combination module identifies the high-frequency access area from the storage partition and the storage location coordinates, calculates the shortest path matching degree between the storage location coordinates based on the high-frequency access area, and then generates a fork bias path. The dynamic updating module verifies the storage location coordinates in real time based on the fork bias path, identifies the displacement change of the goods, and updates the mapping relationship between the storage partition and the storage location coordinates according to the displacement change. The scheduling management module determines the storage target and the carrying sequence of the warehousing goods according to the updated mapping relationship, and finally generates a warehouse scheduling database.

[0055] Embodiment 1: see Figure 2 The primary task of the inventory partition module is to process the continuously input warehousing goods information. The module obtains the original physical characteristic data of the goods through the built-in data interface. These data may come from a multi-dimensional sensor array installed on the conveyor belt or at the entrance. The information collected by the sensor includes the point cloud profile of the object, the weight reading and the material spectral characteristics. After the physical characteristic information is transmitted to the system in the form of a data stream, a pre-trained attribute classification model will be triggered immediately. This model is not a simple rule engine, but a deep neural network structure trained with a large amount of labeled data. It can perform nonlinear transformation and hierarchical abstraction on the input high-dimensional features. The training data of the model comes from the feature library of historical warehouse goods. The output layer uses the softmax function to generate the probability distribution of each attribute category. The system selects the category with a probability exceeding the set threshold as the effective output, and at least ensures that one attribute category is activated for subsequent processing.

[0056] The attribute category can be regarded as a macro-level definition of the basic properties of the item, such as "electronic product", "fragile item" or "bulk cargo" classification. Each category is closely related to a set of processing rules and parameters. Once the attribute category of an item is clearly determined, the system will carry out more detailed and in-depth feature analysis based on the semantic space defined by the category. Specifically, the system will identify a number of highly relevant associated items, including size, weight and material. In terms of size, the system will not only consider the basic geometric dimensions of the item, such as length, width, height, etc., but also further derive more complex indicators, such as volume, surface area and the complexity of the shape profile. These indicators can more comprehensively describe the geometric characteristics of the item, providing important reference for subsequent storage, transportation and packaging.

[0057] The analysis of weight-related items starts with the static mass of the item and expands to more dynamic physical properties such as density and center of gravity. Accurate identification of these properties is crucial to ensuring the stability and safety of the item during handling and transportation. As for the analysis of material-related items, it is more in-depth and complex. It not only involves the physical properties of the material itself, such as hardness, brittleness, temperature resistance, etc., but also considers the chemical compatibility of the material when it comes into contact with other materials. A comprehensive understanding of these information helps to avoid damage or chemical reactions caused by improper material. All these refined feature items will eventually be organized by the system into a structured feature sample set. This sample set not only provides detailed data support for the classification and management of items, but also provides a solid foundation for subsequent intelligent decision-making and optimization.

[0058] The clustering processing of the feature sample set is the core step of the partition generation. The system starts three independent clustering processes in parallel to process the size, weight and material data respectively. The clustering of size-related items uses an optimization algorithm based on Euclidean distance. This algorithm can automatically determine the optimal number of clusters and merge items with similar sizes into the same group, thereby forming several size clustering areas, each representing a set of items with similar spatial occupation. The generation of weight analysis areas relies on different mathematical tools. The system calculates the statistical distribution characteristics of weight data and uses parameters such as mode, median and standard deviation to divide different weight intervals, each interval corresponding to a weight level analysis area. The processing of material-related areas is more challenging. It requires the use of association rule learning techniques to discover the coexistence and mutual exclusion relationships between different material characteristics, thereby aggregating items with compatible or similar material properties into the same material-related area. The generation of size clustering areas, weight analysis areas and material-related areas is not a completely independent process. The system establishes a cross-validation mechanism to ensure the reasonableness of the partitioning. For example, the size of an item may be classified into the A clustering area, but its weight characteristics may belong to the boundary of two weight analysis areas. In this case, the system will refer to its material properties to make the final partitioning decision. The final generated storage partition is a combination of multiple dimensions, which reflects the characteristics of the item in size, weight and material. This composite partition will serve as the logical attribution of the item in the warehouse space. After determining the storage partition of the item, the inventory partition module synchronizes the partition results to the WMS system. The WMS system combines the storage partition information and the material information to allocate different floor target storage locations for the incoming items based on the pre-set storage strategy. The storage partitioning will adapt to the structural characteristics of the different floors of the main access rack. For example, items that need to be accessed frequently will be divided into storage partitions close to the elevator or the first temporary storage area, so that subsequent four-way shuttle vehicles, AMR and other equipment can efficiently complete the storage and access tasks.

[0059] In this embodiment, the intelligent stereoscopic warehouse hardware system served by the inventory partition module specifically includes the main access rack, the high-altitude tray handling equipment four-way shuttle vehicle, the ground tray handling equipment AMR, the tray layer changing equipment elevator, and the sorting station system. The main access rack adopts a layered design, and the second layer and above are tray four-way vehicle high-density access racks for storing incoming trays and materials. The number of rack floors can be set to N layers according to the height of the factory building space, the height of the stored materials, and the design specifications of the vertical warehouse to meet the high-density storage requirements. The first layer of the rack serves as a temporary storage area for picking out the tray materials to be picked. The area near the rack can also be equipped with N sorting workstations according to the sorting efficiency requirements, each of which is equipped with a sorting workstation, a sorting display unit, a code reader (PDA), and a sorting container for the operator to pick the goods from the picking tray according to the order. The size clustering area, the weight analysis area, and the material association area divided by the inventory partition module need to be compatible with the above-mentioned rack layering structure and sorting station layout, for example, the larger and heavier items are divided into the second layer and above storage areas with stronger load-bearing capacity, and the high-frequency picking items are divided into the storage areas near the first layer temporary storage area or sorting station, laying the foundation for subsequent equipment access operations.

[0060] Embodiment 2: Refer to Figure 3 , the system does not consider the size association items, weight association items, and material association items identified in the previous stage in isolation, but merges them according to the same attribute category they belong to. The merging process uses a feature weighting fusion algorithm that assigns a dynamic weight coefficient to each feature item. The weight value depends on the importance of the feature item in its attribute category. Multiple attribute merging results are generated through weighted summation and vector splicing technology, each representing a condensed expression of all features in an attribute category. After generating the attribute merging results, the system needs to identify size combinations with high similarity, i.e., size similarity pairs. The identification process is completed by calculating the multidimensional similarity between the size features of each two items. The similarity measure uses an improved distance function to consider both absolute size and shape ratio matching. The identified size similarity pairs need to be compared one by one with the pre-stored standard size library in the system. The standard size library is a constantly updated knowledge base that contains the standard size specifications of all rack positions in the warehouse space and the common size patterns of historical stored items. The comparison operation is not a simple numerical comparison, but a fuzzy matching strategy based on tolerance, which allows classification within a certain error range. Finally, the successfully matched item groups are divided into a size clustering area.

[0061] For the weight and material features, different strategies are adopted. The system calculates the statistical distribution characteristics of all weight-related items in the attribute merging result, focusing on the distribution dispersion. The dispersion is quantified by calculating the ratio of the standard deviation to the average value of the weight values. The system divides the weight data into several consecutive intervals according to the preset dispersion threshold. Each interval corresponds to a weight analysis zone, which identifies the weight concentration range of the group of items and the required shelf load capacity. The processing of material-related items is more complex, as it requires evaluating the combination strength between different material attributes. The combination strength is measured by analyzing the co-occurrence frequency and conditional probability of materials. High-strength combined material attributes are aggregated into the same material-related zone, which defines the material compatibility requirements and storage environment preferences.

[0062] After successfully generating the size clustering area, weight analysis area and material correlation area, the system needs to map these logical partitions to the specific location coordinates of the physical world. The construction of this mapping relationship is the core link to realize intelligent storage. The mapping process takes a structured feature representation vector as input, which integrates the spatial demand features of the size clustering area, the load bearing demand features of the weight analysis area, and the environmental demand features of the material correlation area. The output is the spatial attributes of the location coordinates (including its three-dimensional position, spatial volume, and physical orientation) and the load bearing level (including the static load limit and dynamic operation load capacity). The system uses a supervised learning model to establish the mapping from the input features to the output attributes. This model is trained through a large number of historical storage records, learning the complex nonlinear relationship between storage partition features and optimal location selection, thereby forming a robust multidimensional mapping relationship that can dynamically recommend the most suitable storage location based on real-time location status and item features. In establishing the multidimensional mapping relationship between the size clustering area, weight analysis area, material correlation area and location coordinates, the location mapping module synchronously associates the spatial coordinates of the hardware devices of the warehouse system, including the three-dimensional coordinates of the main access shelves at each floor, the entrance coordinates of the pallet layer changing equipment elevators, the operation channel coordinates of the high-altitude pallet handling equipment four-way shuttle vehicles, and the movement path node coordinates of the ground pallet handling equipment AMR between the first floor temporary storage area and the sorting workstation. For example, the partitions with high load bearing demand in the weight analysis area are preferentially mapped to the location coordinates of the high-density access shelves on the second floor and above, which match the load bearing level, and ensure that the location coordinates are close to the entrance coordinates of the elevators, facilitating the rapid transfer of pallets by four-way shuttle vehicles. When mapping the partitions of items that need to be frequently sorted in the material correlation area to the location coordinates of the first floor temporary storage area, the key node coordinates on the AMR operation path are associated to ensure that the location coordinates are compatible with the pickup and delivery paths of the AMR, so that the coordinate data in the mapping relationship can be directly called during subsequent equipment scheduling, reducing the secondary calculation of the path. After the mapping relationship is established, it is not fixed and the system will continuously monitor the effectiveness of actual storage operations, such as the storage and retrieval efficiency of items, the safety load state of shelves, and the physical preservation status of items. These monitoring data are fed back to the mapping model for fine-tuning and optimization to ensure that the mapping relationship can adapt to the dynamic changes in inventory and fluctuations in equipment status, thereby continuously optimizing the utilization of warehouse space and operation efficiency.

[0063] Take a large home appliance warehouse center receiving a batch of goods into storage as an example, including refrigerators, washing machines, microwave ovens and other different categories. The warehouse management system first obtains the physical characteristic information of each commodity, including the size, weight and material composition, which are collected by 3D scanners and weight sensors installed on the conveyor belt. The system classifies this batch of home appliances by product type, refrigerators into the "major appliances" category, washing machines into the "washing equipment" category, and microwave ovens into the "kitchen appliances" category. Under each category, the system further extracts size-related items (length, width, height, volume), weight-related items (net weight, gross weight), and material-related items (metal shell, plastic components, glass parts) to form a structured feature sample set. Each item of data in the feature sample set is merged by attribute category, and the size data of the refrigerator is aggregated into a group, the weight data into another group, and the material data into a separate group. Washing machines and microwave ovens also go through the same processing procedure. The system extracts similar combinations of refrigerator size data and finds that side-by-side refrigerators and single-door refrigerators form two distinct groups in terms of size. These size combinations are compared with standard shelf sizes to determine the need for different specifications of storage compartments.

[0064] The distribution of washing machine weight data is calculated, and it is found that the weight of drum washing machines is concentrated in the range of 70-90 kg, and the weight of pulsator washing machines is in the range of 45-65 kg. The system divides washing machines into two weight analysis zones according to the weight distribution dispersion. At the same time, the material combination strength is analyzed, and it is found that metal shell washing machines and plastic shell washing machines require different storage environment requirements, and accordingly material-related zones are generated. After establishing size clustering zones, weight analysis zones and material-related zones, the system begins to construct the mapping relationship with specific storage locations. Take the side-by-side refrigerator as an example, its feature representation vector contains large size requirement, heavy load requirement and metal material characteristics, the system matches these features with the spatial attributes of the storage location coordinates (large cargo grid, reinforced support) and the bearing level (more than 500 kg per square meter). Through the pre-trained mapping model, the system maps this type of refrigerator to the reinforced shelves in a specific area.

[0065] For small household appliances such as microwave ovens, the system adopts different mapping strategies, and the feature vectors contain small and medium size, light weight, and mixed material information, and the mapping model points to the standard storage grid of the medium-sized shelf, which has appropriate space size and load capacity. Each mapping decision considers the multidimensional matching of item characteristics and storage location attributes to ensure the optimization of storage safety and space utilization. The system finds the most suitable storage location for each type of household appliance. New refrigerators are arranged in the heavy-duty shelf area, washing machines are distributed to different load-bearing areas according to type, and microwaves are placed in the medium-sized goods area. After the mapping relationship is established, the system continuously monitors the actual storage effect, adjusts the feature weights and mapping parameters according to the actual load of the shelf and the access operation frequency, so that the correspondence between the storage partition and the location coordinates remains optimal.

[0066] In Example 3, the system calls historical storage location coordinate data associated with the storage partition, which comes from the operation log of the warehouse management system, and obtains real-time state information of the current location through the sensor network deployed on the shelf, including occupancy, device health status, and accessibility. After data cleaning and fusion processing, a set of unlabeled location recognition results are generated, which contain basic information of the location but have not yet established a binding relationship with the specific storage partition.

[0067] The system needs to judge the idle condition compliance of these unlabeled recognition results, and the judgment logic is based on a set of pre-defined multi-dimensional rules, which cover the physical integrity (such as no structural damage), operation availability (such as not being locked for maintenance), and logical availability (such as not being reserved by other tasks) of the location. The location recognition results that meet the idle condition will be marked as available and used as candidate location coordinates for the corresponding storage partition, which ensures that only the locations in an operable state will enter the subsequent allocation process. After successfully establishing the mapping between the storage partition and the location coordinates, the system starts the path optimization process, which analyzes the storage partition and its mapped location coordinates as a whole data set. This data set is hierarchically clustered according to three key indicators: access frequency, path length, and number of turns. Access frequency is obtained by counting the number of historical access times, path length is calculated by geometric distance, and the number of turns depends on the path topology. The clustering algorithm automatically determines the number of categories and aggregates locations with similar access patterns, where the category with the largest cluster center is defined as the high-frequency access area, which represents the most active operation hotspot in the warehouse.

[0068] Extract the spatial nodes contained in the high-frequency access area, including location coordinate points, path intersection points, and turning points, and calculate the Manhattan distance between each node to evaluate the degree of spatial separation. The Manhattan distance is calculated as follows:

[0069] DM (n i ,n j )=|x i -x j |+|y i -y j |+|z i -z j |

[0070] wherein: n i and n j represent two spatial nodes, x, y, z represent the coordinate values of the nodes in a three-dimensional coordinate system respectively. Based on these distance values, the system generates a set of common path sequences sorted by distance, which defines the basic connectivity relationship between nodes.

[0071] On the basis of the common path sequence, the system further extracts adjacent pair of coordinates of the storage locations, and calculates the shortest common sub-path for each adjacent pair of storage locations. The calculation of the sub-path takes into account the actual shelf layout and equipment movement constraints, and the length value of the sub-path is quantified as a path matching degree, which reflects the accessibility efficiency between two storage locations. Finally, the system generates a fork bias path by combining the path matching degree and the time access weight, and the time access weight is determined by analyzing the historical access patterns in different time periods. The bias path provides an optimized guide for the movement of the warehouse equipment. The system collects actual path data and equipment operation indicators while performing storage tasks, compares these data with the prediction model, and triggers the adjustment of the clustering algorithm and path calculation parameters when the deviation exceeds the threshold, ensuring that the path planning always matches the actual operating state. The fork bias path generated by the path combination module will adapt to the job requirements of different equipment. For four-way shuttles, the path will fit the channel layout of two-layer and above high-density access shelves, reducing the number of turns between shelves. For AMRs, the path will preferentially plan the shortest route between the first-layer temporary storage area and each sorting workstation, while avoiding the elevator entrance and conveyor line area to avoid equipment operation conflicts. For elevators, the path will clearly indicate the connection coordinate points with four-way shuttles and conveyor lines to ensure smooth pallet transfer.

[0072] Taking a large e-commerce warehouse center during the 618 promotion as an example, the system needs to handle tens of thousands of goods in and out of the warehouse operations. The warehouse management system first calls the historical storage location coordinate data, which comes from the operation records in the past six months, and at the same time obtains the real-time state information of all shelves through the Internet of Things sensor network, including the occupancy of the storage location, the load-bearing state of the shelf, and the working state of the conveying equipment. The system detects that there are multiple storage locations on the A area shelf showing an idle state, but these storage locations often store high-turnover electronic products in historical records, and the system marks these storage locations as unmarked storage location recognition results. Through idle condition judgment on these storage locations, the system confirms that their physical structure is intact, wireless communication is normal, and they are not reserved for special goods, so these storage locations are allocated to high-value electronic products such as smartphones and tablets that will be warehoused.

[0073] After completing the storage location allocation, the system begins to analyze the access mode of the storage partition and storage location coordinates, and the system performs hierarchical clustering analysis on the historical operation data of the past three months according to access frequency, path length, and turn number. The access frequency data shows that the daily access frequency of the digital product area reaches more than 500 times, the path length analysis shows that the average distance between this area and the packaging area is the shortest, and the turn number statistics find that the main path to this area only needs one right-angle turn. These data are processed by the clustering algorithm to identify the digital product area as the largest clustering center, i.e. the high-frequency access area. The system extracts all spatial nodes in the high-frequency access area, including shelf coordinate points, path intersection points, and turning points, and calculates the spatial relationship between nodes. By measuring the position difference of nodes in the three-dimensional coordinate system, the system obtains a set of measurement values reflecting the actual moving distance, and generates the optimal public path sequence based on these measurement values, which defines the shortest connected path between nodes.

[0074] Based on the public path sequence, the system further analyzes the relationship between adjacent storage locations and calculates the optimal sub-path for each pair of adjacent digital product storage locations. These sub-paths take into account the actual motion characteristics of the stacker, including acceleration, deceleration, and turning operation restrictions, and quantify the calculated path length into path matching degree values. Finally, the system generates a fork bias path by combining the path matching degree and the time access weight, and the time access weight is determined by analyzing the operation frequency in different time periods, giving higher weight coefficients to high-frequency access areas during the morning order peak period. The entire implementation process is completed during the night inventory period, and the optimal path scheme generated by the system is put into practical application in the next day's promotion activities. Warehouse management personnel observe that the operation efficiency of the fork device in the digital product area has significantly improved, the device's empty mileage has decreased, and the parallel processing capacity of multiple orders has increased. The system continuously monitors actual running data, and when it detects that the actual passing time of some paths differs significantly from the predicted value, it automatically triggers path recalculation and optimization adjustment to ensure that warehouse operations remain efficient at all times.

[0075] In a practical application scenario of a large e-commerce warehouse center storing tens of thousands of goods, the electronic product area and high-value item area usually have a higher access operation frequency, and the system needs to continuously optimize the storage management strategy of these areas. After generating the fork bias path, the system parses the time access weight of each storage coordinate. The calculation of the time access weight relies on the time series analysis of the historical operation log. The system divides a day into multiple time windows (e.g., 08:00-10:00, 10:00-12:00, etc.), and counts the number of accesses, operation duration, and operation type (storage / taking) of each storage in each time window. Then, through normalization, a weight coefficient is assigned to each storage in different time periods, which reflects the relative activity level of the storage in that time period. For example, during the morning sorting peak period, the access weight of the A area shelf may reach 0.85, while during the night period, it may drop to 0.20. When identifying the displacement changes of the goods and updating the mapping relationship, the dynamic update module synchronously collects the real-time positions of the four-way shuttles, the operation states of the AMRs, and the running conditions of the elevators. When detecting that the four-way shuttles have position deviations when moving pallets on the second floor and above, or that the AMRs have path deviations when moving between the first floor temporary storage area and the sorting workstation, the system integrates these device state information into the difference value calculation, so that the storage coordinate state update is not only based on the displacement of the goods, but also combined with the actual operation conditions of the devices, to improve the accuracy of the mapping relationship.

[0076] The target trajectory of the fork bias path is divided according to the time period corresponding to the time access weight. The target trajectory is composed of a series of continuous storage access sequences, each sequence is marked with a timestamp, and the system aggregates the continuous access points with similar time weights into the same trajectory segment. For example, all high-frequency access storages from 08:00 to 10:00 are connected to form the morning peak trajectory, and the medium-frequency access points from 14:00 to 16:00 are connected to form the afternoon operation trajectory. Each trajectory segment carries the predicted access density expectation value. The access density is obtained by counting the planned number of accesses per unit time, and the actual goods distribution is collected in real time by the weight sensors, vision recognition cameras, and RFID readers installed on the shelves. The system calculates the absolute deviation of the predicted number of accesses and the actual number of operations for each storage in each time window, and marks it as the difference value. The storages with a difference value exceeding the set threshold are marked as state abnormal, as shown in Table 1.

[0077] Table 1: Storage state difference analysis

[0078]

[0079]

[0080] The inventory coordinates are dynamically classified based on the difference values, and the classification standard adopts a multi-level threshold strategy, for example, a difference value below 5 is marked as "normal", 5-15 is marked as "attention", and above 15 is marked as "abnormal". The classification results and specific values are fed back to the inventory partition module in real time, and the inventory partition module triggers the re-division of the storage partition according to the received state information. For example, when the B-02-03 area continuously shows an abnormal state, the system may reclassify it from the "high-frequency access electronic area" to the "medium-frequency access area" and adjust its corresponding path planning priority. The system performs a full warehouse area scan every 15 minutes, updates the difference value data table and recalculates the inventory status classification. This continuous adaptive adjustment mechanism enables the warehouse system to respond to seasonal fluctuations in sales, changes in traffic caused by promotional activities, and equipment state abnormalities, and maintain an optimal balance between operation efficiency and space utilization.

[0081] Embodiment 5: Processing the updated storage partition and inventory coordinate mapping relationship to coordinate all warehouse operations. Consider a large distribution center in actual operation scenario, which needs to handle multi-source warehousing demand from online orders, offline replenishment and return processing at the same time, and each kind of goods has different processing timeliness and space requirements. According to the dynamically updated storage partition and inventory coordinate mapping relationship, the system extracts the storage priority of the warehousing goods, and the priority calculation is based on multi-dimensional factors including order deadline, goods value coefficient, customer level identification and special processing requirements. For example, fresh goods have a higher time urgency coefficient, high-value electronic products have a higher safe storage weight, and VIP orders have a priority processing identifier. The system generates a quantitative priority score for each warehousing goods through a weighted algorithm that combines these factors. The carrying sequence is generated according to the time urgency of the storage priority and the path optimization coefficient, the time urgency is determined by calculating the difference between the current time and the promised delivery time, and the path optimization coefficient considers the path length, device energy consumption estimation and parallel operation conflict probability.

[0082] The system generates a carrying sequence using a multi-objective optimization algorithm to ensure that high-priority items are processed quickly while maintaining efficient device movement paths. For example, high-priority items in the same area are processed in batches to reduce device back-and-forth movement. The storage priority and carrying sequence are encoded using a tree structure, which uses a multi-way tree form. The root node represents the scheduling task entry, and the child nodes represent specific carrying operation steps. Each node contains the storage location coordinates, operation type, time window, and device allocation information. The tree structure is traversed and optimized using breadth-first search to generate a structured warehouse scheduling database. This database not only contains the current task sequence but also saves the dependency relationships between operations and resource allocation states. The real-time location data of the incoming items is time-aligned with the predicted path from the warehouse scheduling database. The real-time location data is collected using UWB positioning tags and visual recognition systems, and the predicted path is obtained from the scheduling database. The warehouse scheduling database generated by the scheduling management module is linked with the WMS system, WCS control system, and AMR scheduling system. When performing warehouse operations, the database updates the mapping relationship based on the updated mapping relationship to support the WMS system issuing warehouse tasks to the WCS control system. The WCS schedules the conveyor line and elevator to lift the pallet to the target floor and then schedules the four-way shuttle vehicle to carry the pallet to the designated storage location. When performing out-of-stock replenishment operations, the database provides the storage location information of the replenishment pallet to assist the WMS system in issuing replenishment tasks. The WCS schedules the four-way vehicle and elevator to deliver the pallet to the temporary storage area on the first floor, and the AMR scheduling system arranges the AMR to carry the pallet to the buffer storage location. When performing picking operations, the database provides the location data of the buffer picking pallet to support the WMS system issuing picking tasks to the AMR scheduling system, which schedules the AMR to carry the pallet to the sorting workstation for the operator to pick.

[0083] The alignment operation uses the dynamic time warping algorithm to match the actual timestamp sequence with the planned timeline, identifying differences in advance, delay, or synchronization. When detecting a deviation between the actual path and the predicted trajectory, the system triggers the remapping of the storage location coordinates and the iterative optimization of the fork deviation path. Deviation detection is achieved by comparing the Euclidean distance between the actual device location and the planned path, and the correction program is started when the deviation exceeds the threshold. The remapping process calls the algorithm of the storage location mapping module to recalculate the optimal storage location based on the current actual storage location state. Iterative optimization recalculates the path matching degree and time weight to generate an updated fork deviation path. The entire scheduling management process forms a real-time response control loop, with the system refreshing the scheduling state every 30 seconds and dynamically adjusting the node order and parameters in the tree structure based on the latest device location and task execution. This dynamic adjustment mechanism enables the warehouse system to adapt to abnormal situations such as device sudden failure, order priority changes, and traffic peaks, maintaining efficient operation of warehouse operations.

[0084] It is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting; it is not intended to exclude myriad other embodiments of the present application that other inventors can develop based on the same general inventive concepts embodied by the described embodiments. That is, although the present application is described in terms of particular embodiments and illustrative figures, it should be apparent that the scope of the present application is not limited to these specific embodiments.

[0085] While the embodiments of the application have been shown and described herein, it is to be understood that the scope of the application, jointly pointed out in the appended claims, is not to be limited to the above-described embodiments but can be otherwise variously changed, modified, replaced, and altered within the principles and spirit of the present application.

Claims

1. An intelligent automated storage and retrieval system based on dynamic zoning and real-time mapping, characterized in that, include: The inventory partitioning module is used to obtain the physical characteristics of incoming items, classify them according to their storage attributes, and determine the storage partition of the items relative to the warehouse space. The storage location mapping module is used to extract the spatial coordinates of storage partitions and construct the mapping relationship between storage partitions and storage location coordinates; The path combination module is used to extract high-frequency access areas from storage partitions and storage location coordinates, calculate the shortest path matching degree between storage location coordinates according to the high-frequency access areas, and generate fork bias paths. The dynamic update module is used to perform real-time verification of storage location coordinates based on the fork offset path, identify the displacement changes of items in the storage location coordinates, and update the mapping relationship between storage partitions and storage location coordinates according to the displacement changes. The scheduling management module is used to determine the storage target and handling sequence of incoming items based on the updated mapping relationship between storage partitions and storage location coordinates, and to generate a warehouse scheduling database.

2. The intelligent automated storage and retrieval system based on dynamic partitioning and real-time mapping according to claim 1, characterized in that, The implementation methods of the inventory partitioning module include: For any physical feature information of an item entering the warehouse, call the pre-trained attribute classification model; The physical feature information is hierarchically divided using an attribute classification model to obtain at least one attribute category. Identify the size-related items, weight-related items, and material-related items in the physical feature information under the corresponding attribute category to form a feature sample set; Clustering is performed on the size-related items, weight-related items, and material-related items in the feature sample set respectively, generating size clustering area, weight analysis area, and material-related area in sequence, which serve as storage partitions for items relative to the storage space.

3. The intelligent automated storage and retrieval system based on dynamic partitioning and real-time mapping according to claim 2, characterized in that, Other methods for generating size clustering regions, weight resolution regions, and material association regions include: The size-related items, weight-related items, and material-related items in the feature sample set are merged according to attribute categories to obtain a multi-attribute merged result; Extract size similarity pairs from the attribute merging results, compare them with the standard size library, and generate size cluster regions; The weight-related items and material-related items in the merged attribute results are calculated and their distribution dispersion and combination strength are used to divide the weight analysis region and the material-related region.

4. The intelligent automated storage and retrieval system based on dynamic partitioning and real-time mapping according to claim 1, characterized in that, The implementation methods of the storage location mapping module include: Retrieve the historical storage location coordinates and current storage location status associated with the storage partition to generate unlabeled storage location identification results; Determine whether the unmarked storage location identification results meet the free condition. If they do, use the storage location identification results that meet the condition as the storage location coordinates of the storage partition.

5. The intelligent automated storage and retrieval system based on dynamic partitioning and real-time mapping according to claim 3, characterized in that, The methods for establishing the mapping relationship between storage partitions and library location coordinates include: Using the feature representations of size clustering region, weight analysis region and material association region as input, and the spatial attributes and load-bearing level of storage location coordinates as output, a multidimensional mapping relationship from storage partition to storage location coordinate is established.

6. The intelligent automated storage and retrieval system based on dynamic partitioning and real-time mapping according to claim 1, characterized in that, The path combination module can be implemented in the following ways: The storage partitions and location coordinates are hierarchically clustered according to access frequency, path length and number of turns, and the largest cluster center is set as the high-frequency access area. Extract spatial nodes from frequently accessed areas, calculate the Manhattan distance between each node, and generate a sequence of common paths related to the distances. The coordinates of adjacent storage locations are extracted based on the common path sequence, the shortest common sub-path between each storage location is set, and the sub-path length value is used as the path matching degree. Forklift bias paths are generated based on path matching degree and time access weight.

7. The intelligent automated storage and retrieval system based on dynamic partitioning and real-time mapping according to claim 6, characterized in that, The implementation methods for the dynamic update module include: Extract the time access weights of each storage location coordinate from the forklift bias path; The target trajectory of the fork bias path is divided according to the time period corresponding to the time access weight; The access density of the target trajectory in each time period is compared with the actual item distribution at the warehouse location coordinates, the difference value is marked, and the warehouse location coordinate status is updated based on the difference value.

8. The intelligent automated storage and retrieval system based on dynamic partitioning and real-time mapping according to claim 1, characterized in that, The implementation methods of the scheduling management module include: Based on the updated mapping relationship between storage partitions and storage location coordinates, extract the storage priority of incoming items; A transport sequence is generated based on the time urgency of storage priority and the path optimization coefficient; A tree structure is used to encode storage priorities and handling sequences to generate a warehouse scheduling database.

9. The intelligent automated storage and retrieval system based on dynamic partitioning and real-time mapping according to claim 8, characterized in that, Also includes: Align the real-time location data of incoming goods with the predicted paths in the warehouse scheduling database in a time sequence. When the actual path deviates from the predicted trajectory, the remapping of the storage location coordinates and the iterative optimization of the forklift bias path are triggered.

10. The intelligent automated storage and retrieval system based on dynamic partitioning and real-time mapping according to claim 7, characterized in that, Other methods for updating the storage location coordinates include: The storage location coordinates are dynamically classified based on the difference value, and the classification results are fed back to the inventory partitioning module to trigger the re-division of storage partitions.

Images