A three-dimensional real-time modeling digital twin warehouse management method and system

Abstract

The application belongs to the technical field of digital warehouse management, and particularly relates to a three-dimensional real-time modeling digital twin warehouse management method and system. The method comprises the following steps: initializing warehouse environment parameters, collecting and accessing multiple types of real-time data, constructing a three-dimensional state model based on point cloud and vector data, evaluating modeling error and performing fitting calibration, and finally forming an iteratively updated digital twin model. The system comprises environment initialization, data collection, time synchronization, three-dimensional modeling, error measurement and correction modules, and can realize dynamic mapping of spatial structure, logistics behavior and equipment state. By introducing a unified time source synchronization mechanism and a residual error optimization correction strategy, the application improves the modeling accuracy, response efficiency and historical traceability of the digital twin system, and is suitable for intelligent warehouse management in high dynamic scenarios.

Description

Technical Field

[0001] This invention belongs to the technical field of digital warehouse management, specifically relating to a three-dimensional real-time modeling digital twin warehouse management method and system. Background Technology

[0002] In the modern logistics and warehousing industry, with the widespread application of technologies such as the Industrial Internet of Things (IIoT), edge computing, and artificial intelligence (AI), digital twin technology is gradually permeating all aspects of warehouse management. By constructing a high-fidelity virtual mapping of real-world space, digital twins can reflect the state and behavioral characteristics of physical entities in real time, supporting dynamic prediction, intelligent scheduling, and full lifecycle management. It has become one of the key supporting technologies for intelligent warehousing. Against this backdrop, 3D modeling capabilities have become one of the key foundations for the visual interaction of digital twin platforms, determining the accuracy of the management system's response to changes in physical space and events.

[0003] Existing digital twin warehousing systems primarily construct digital mappings of the warehousing environment through static modeling or periodic updates. Typical technical approaches include generating static warehouse layout maps based on BIM (Building Information Modeling) and combining this with sensor data from RFID, cameras, or AGVs for localized state updates. However, these methods generally suffer from problems such as lag in response, coarse-grained updates, and an inability to accurately reflect dynamic changes. For example, changes in temporary material storage locations, temporary adjustments to forklift paths, and rack modifications often fail to be synchronized to the system in a timely manner, leading to system scheduling failures or operator misjudgments. Furthermore, some current 3D modeling methods rely on manual modeling or laser scanning, resulting in long modeling cycles and high update costs, making them unsuitable for high-frequency changing scenarios. The lack of integrated modeling capabilities that fuse task dynamics, physical states, and operational behaviors also limits the existing systems' ability to manage complex tasks with precision and optimize them in real time.

[0004] Therefore, there is an urgent need to build a digital twin warehouse management solution with real-time 3D modeling capabilities to meet the needs of intelligent decision-making and collaborative execution in a high-frequency dynamic environment. Summary of the Invention

[0005] To address the above problems, the present invention aims to propose a three-dimensional real-time modeling digital twin warehouse management method, comprising the following steps:

[0006] S1. Initialize warehouse environment parameters: including the three-dimensional position of the warehouse entity, shelf number, equipment coordinates, entrance and exit coordinates, and loading and unloading operation area, and establish an association mapping between the initial state information and the actual warehouse environment;

[0007] S2. Collect and access multiple types of real-time data, including data on goods entering and leaving the warehouse, handling, storage, unloading, and equipment operating status, and align and correct the collected timestamps through a synchronization mechanism based on a unified time source.

[0008] S3. Construct a three-dimensional real-time state model: The model is based on point cloud and vector data, and combines a dynamic data set that has been processed by historical state restoration to fit the current modeling state and form spatial structure constraints.

[0009] S4. Modeling Error Assessment and Fit Determination: After modeling is completed, the modeling error of the 3D state model is calculated using an error measurement method, and the modeling fit is determined based on a preset threshold.

[0010] S5. Perform inverse fitting and historical residual calibration: Extract historical input data and modeling status records of key areas within a set time window, use the residual between the current modeling result and the historical record to perform parameter back-calculation and modeling optimization, adjust the position and attribute information of model nodes through the minimum residual, generate the corrected modeling result and output it.

[0011] As a preferred technical solution, step S4 specifically includes: S4.1, if the modeling error is within the set threshold, then update the mapping data cache and synchronously write it into the modeling database as a snapshot of the digital model at the current moment; S4.2, if the modeling error exceeds the set threshold, then statistically analyze the distribution function of the modeling error under different data sources, and identify the two source regions with the largest errors as key areas for reverse correction.

[0012] As a preferred technical solution, the method further includes: S6, outputting the correction model and performing error verification iteration: taking the modeling result optimized by minimizing the residual as the latest three-dimensional model at the current moment, and comparing it with the current state of the real warehousing system again to verify the three-dimensional indicators of spatial location, timestamp and item identification; if all indicators meet the preset error threshold requirements, then updating the state snapshot and entering the next round of data access and modeling process to realize the dynamic real-time iteration of the digital twin system.

[0013] As a preferred technical solution, in step S1, the initialization operation is triggered by the start signal of the warehousing system, including automatic loading and identification of equipment coordinates, static mapping and marking of shelf numbers, and system preset correction of entrance and exit coordinates and loading and unloading operation areas.

[0014] As a preferred technical solution, in step S2, the multi-type real-time data includes at least one set of data from sensor data, RFID read / write data, and automated handling equipment control command data, and is written to the middleware cache interface in a unified format;

[0015] The synchronization mechanism performs millisecond-level time alignment processing on various data sources based on a unified time source. Specifically, it includes: providing a unified time reference through the Network Time Protocol (NTP); accurately sorting data from different sources according to the timestamps of the acquisition terminals; and using a linear time interpolation algorithm to correct the offset for data with time drift, thereby generating a unified time series on a standardized time axis.

[0016] As a preferred technical solution, in step S3, the three-dimensional real-time state model includes at least a point cloud dataset and a corresponding time series identifier. Each point cloud dataset includes spatial location, volume information and a timestamp, so as to realize the dynamic visualization construction of the three-dimensional modeling scene.

[0017] As a preferred technical solution, step S4.1, updating the mapping data cache also includes establishing a structured data set on the modeling timeline and writing it into a modeling database with a timestamp index for tracing the historical state of the fitting correction process.

[0018] As a preferred technical solution, in step S4.2, the parameters of the distribution function include the comprehensive calculation results of spatial coordinate difference, time delay and data fitting residual, and are sorted according to regional weights to lock the key regions with the greatest impact on error.

[0019] As a preferred technical solution, in step S5, the inverse fitting and historical residual calibration are performed based on a set fixed time interval, the time interval being 3 minutes in length, processed in a preset step order, and the modeling residual of the modeling node is recalculated in each interval to form a comparison table of residuals before and after optimization.

[0020] The present invention also provides a three-dimensional real-time modeling digital twin warehouse management system for implementing the method, comprising:

[0021] The environment initialization module is used to collect initial warehouse environment parameters and establish a correspondence with the actual scenario;

[0022] The data acquisition module is used to collect data, including unified collection, screening, and deduplication of data on cargo flow, equipment scheduling, environmental parameters, sensor sampling, and system events;

[0023] The time synchronization module is used to establish time series, align the time of all connected data sources, and perform synchronization processing and data association through a unified timestamp dimension;

[0024] The 3D modeling module is used to receive time-series data and generate real-time 3D modeling scenes through calculation rules.

[0025] The error measurement module is used to calculate the time deviation, spatial deviation, and object deviation between the modeling model and the actual scene, and output the residual error to provide feedback on the modeling fit.

[0026] The correction module is used to compare the accuracy of key areas in the modeling model based on the error measurement results, and to dynamically correct areas where the modeling error exceeds a preset threshold.

[0027] The cache database is used to store the access data set, the corrected modeling state set, and bind historical state information associated with timestamps.

[0028] As a preferred technical solution, the error measurement module includes multiple deviation parameters, each of which includes the time delay, spatial position offset, or item detail difference between the actual scene visual object and the modeling model.

[0029] As a preferred technical solution, the calculation method of the modeling error is based at least on the comprehensive weighted result of the ratio of spatial three-dimensional coordinate difference and time dimension deviation, so as to output the fitting degree between each model node and the actual node.

[0030] As a preferred technical solution, the correction module includes an error localization unit, a data backtracking unit, and a residual optimization unit. The error localization unit identifies key deviation nodes based on the modeling error distribution results and generates corresponding backtracking data requests. The data backtracking unit extracts the original data and historical modeling status of the target node within a set time period from the cache database and establishes a corresponding time series mapping. The residual optimization unit performs fine-tuning iterations based on the original modeling parameters and the target residual, calculates the correction vector, and updates the optimization results to the three-dimensional model structure maintained by the modeling module, forming a closed-loop correction process.

[0031] Beneficial effects

[0032] This invention significantly improves the time consistency of multi-source heterogeneous data during real-time modeling by introducing a high-precision data synchronization mechanism under a unified time source. Compared to traditional warehousing systems that rely on data acquisition sequence or local clocks, this invention uses NTP unified time synchronization combined with linear time interpolation and time drift correction strategies to accurately map device commands, sensor data, and RFID read / write information onto a unified time axis. This effectively avoids problems such as model frame breaks and object misalignment caused by time deviations, ensuring the continuity and real-time performance of 3D modeling.

[0033] This invention is the first to introduce dynamic residual evaluation and spatial partitioning correction mechanisms into the warehouse digital twin modeling process. By constructing an error distribution function, it identifies the key areas with the largest modeling residuals and performs targeted backtracking optimization on historical data within these areas. This strategy overcomes the inefficiency of traditional global reconstruction-based correction methods, achieving local high-frequency fitting and fine-tuning under error closure, significantly improving model iteration efficiency and regional modeling accuracy, especially demonstrating superior robustness and responsiveness in high-dynamic, high-precision scenarios.

[0034] The modeling snapshot system and time-indexed cache database constructed in this invention support accurate model history tracing and error source analysis. The system not only automatically generates timestamp-bound state snapshots in each round of modeling, but also writes correction residuals, error peaks, and optimization parameters into logs, forming a reviewable, comparable, and traceable model evolution chain. This structured evolution recording mechanism provides a data foundation for anomaly tracing, model verification, and system rollback, significantly enhancing the stability and controllability of the digital twin system in complex warehousing environments. Attached Figure Description

[0035] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0036] Figure 2 This is a schematic diagram of the system structure of the present invention. Detailed Implementation

[0037] To enhance understanding of the present invention, the present invention will be further described in detail below with reference to embodiments. These embodiments are only used to explain the present invention and do not constitute a limitation on the scope of protection of the present invention.

[0038] Example 1: A Digital Twin Warehouse Management Method Based on Real-Time 3D Modeling

[0039] like Figure 1 As shown, this embodiment provides a three-dimensional real-time modeling warehouse management method for digital twin systems, which is applicable to large and medium-sized automated warehouse scenarios. It can integrate multi-source dynamic data, construct a high-precision three-dimensional warehouse status model, and realize a closed-loop feedback correction mechanism for modeling errors.

[0040] This method achieves dynamic tracking of the digital twin of the warehouse status by establishing initial warehouse environment parameters, synchronously collecting dynamic data, building a modeling model in real time, and performing error measurement and correction.

[0041] S1. Initialize warehouse environment parameters:

[0042] This step is mainly used for the basic environment construction of the digital twin system. That is, when the warehousing system is started, the system automatically collects and organizes the static structure and regional distribution information of the entire warehouse, and establishes an initial mapping with the real physical environment.

[0043] First, upon receiving the start signal, the spatial positioning submodule is invoked to acquire 3D data of the warehouse's internal structure, including spatial boundary information of walls, supporting columns, ceilings, and floor areas. Structural point cloud data is collected using laser scanning equipment, and a basic structural diagram is generated through a 3D reconstruction algorithm. This diagram is then combined with BIM (Building Information Modeling) to complete structural comparison and obtain a standardized spatial coordinate system.

[0044] Based on this, the system loads a pre-set list of shelf numbers and combines it with QR codes or RFID tags to statically map and mark each shelf, generating a shelf-location mapping table.

[0045] Equipment coordinates are identified by synchronously locating the initial status signal bound to the configured AGV or conveyor device, and generating a location label record by combining the equipment number.

[0046] Simultaneously, the coordinates of entrances / exits and loading / unloading areas are calibrated through manual configuration or image recognition to ensure that the spatial constraints of the logistics path are fully expressed. After completing the above steps, the system generates an initial state information table, including a shelf distribution map, a set of static equipment coordinates, and area boundary labels, and writes it into the environment cache as the initial modeling benchmark for the digital twin model.

[0047] S2. Collect and access multiple types of real-time data:

[0048] The core of this step lies in achieving effective access and time alignment of multi-source dynamic data.

[0049] It collects real-time data on various dynamic behaviors involved in warehouse operations, including but not limited to status changes during the processes of goods entering, leaving, handling, transferring, temporarily storing, and unloading. It also collects scheduling signals and feedback status of equipment operation.

[0050] In addition, to ensure consistency between the physical scene and the modeling system, sensor data including temperature, humidity, vibration, and lighting are also integrated. The system sets up a unified data acquisition interface to classify, organize, and standardize the aforementioned heterogeneous data sources, and uses a middleware buffer architecture to integrate them into the modeling system.

[0051] The key point is that the system constructs a synchronization mechanism based on a unified time source. It provides a unified time reference for each acquisition terminal through an NTP (Network Time Protocol) server, and all data must be appended with a millisecond-level timestamp before being entered into the database. If time drift or inconsistent sampling frequency is found between device command data and sensor acquisition values, a linear time interpolation algorithm is introduced for alignment correction to ensure that all data sources form a continuous distribution on a unified time axis.

[0052] The cache refresh cycle is set to 300ms. The latest collected data is written to the intermediate data queue in chronological order each time, so that the modeling module can call it in real time and ensure that the scene reflected by the model always matches the physical system's operating state.

[0053] S3. Construct a three-dimensional real-time state model:

[0054] Based on the synchronized data access, the real-time construction process of the three-dimensional state model is initiated.

[0055] The core of the modeling includes two types of data:

[0056] First, spatial point cloud data, used to reconstruct the actual position and structural outline of goods, pallets, and equipment in three-dimensional space;

[0057] Second, there is vector behavioral data, which represents the scheduling behavior of equipment, the flow trajectory of goods, and the operation path.

[0058] Based on the two types of data mentioned above, spatial reconstruction and object association operations are performed to construct a 3D modeling model within the scene.

[0059] This model has continuous time labels, object identification information, and position-attitude coupling description capabilities, and can dynamically reflect spatial changes and object transfers during warehousing operations.

[0060] Each node in the model contains four attributes: location coordinates, volume information, status code, and timestamp. The system adopts an incremental construction mechanism, that is, each round of modeling only updates the nodes or regions that have changed, in order to improve modeling efficiency.

[0061] During the modeling process, the three most recent snapshots are automatically retrieved as reference states, and a prediction mechanism is introduced to compensate for the current data state, thereby reducing modeling jitter caused by missing samples or signal delays.

[0062] The final result is a three-dimensional real-time model covering the entire warehouse area, in which the number of visible objects, their distribution density, dynamic trajectories, and boundary conditions are consistent with the actual scene.

[0063] S4. Modeling error assessment and goodness-of-fit judgment:

[0064] After modeling is completed, the system immediately performs error evaluation and goodness-of-fit judgment operations.

[0065] The error here is not a single spatial deviation, but a joint metric that takes into account three dimensions: three-dimensional coordinate error, time alignment error, and object recognition error rate.

[0066] Introducing a weighted error scoring formula .

[0067] Where Δx represents spatial location deviation, Δt represents timestamp alignment error, Δid represents object identifier consistency score, and w1, w2, and w3 are pre-set weight ratios.

[0068] The score is compared with a set threshold. If the score is less than the threshold (e.g., 95 points), the model is considered to have high credibility and can be directly used for twin system state updates; otherwise, the error backtracking stage is entered.

[0069] This error scoring mechanism is highly versatile, adaptable to different accuracy levels and dynamically adjustable in terms of evaluation sensitivity to accommodate changing conditions in complex warehousing scenarios.

[0070] S4.1 Update the mapping data cache and synchronously write it to the modeling database:

[0071] Once the model fit evaluation is satisfactory, the system will write the current modeling model into the modeling database and update the mapping in the cache table.

[0072] The data written includes the current model snapshot number, 3D structure diagram, object attribute set and binding timestamp, forming a complete digital twin state snapshot.

[0073] At the same time, a differential summary is generated, which records the new objects, movement trajectories and status changes in this update compared to the last snapshot, for subsequent analysis modules to call.

[0074] This snapshot serves not only as a basis for tracing historical states but also as a reference standard for correction mechanisms, thereby ensuring that the system modeling state has structured support that is rollbackable, comparable, and verifiable.

[0075] S4.2 The region with the largest source of identification error is used for calibration preparation:

[0076] If the model error score fails to meet the standard, the system will trigger the error localization algorithm to perform regional attribution analysis on the modeling error.

[0077] The model bias caused by each data source is calculated, and the data block or spatial region with the largest error is determined by residual sorting. For example, the system may find that RFID data on a certain path is continuously offset, or that blurry camera images of a certain shelf area lead to point cloud recognition errors.

[0078] Extract error peak points, label their spatial ranges, generate a list of key correction areas, and record the corresponding time window and data source type for each area to provide a location basis and data range boundary for subsequent reverse correction operations.

[0079] S5. Perform inverse fitting and historical residual calibration:

[0080] After the error region is determined, the reverse fitting and historical residual calibration process is initiated.

[0081] First, retrieve all modeling records and data inputs for the error region within the first 3 minutes from the modeling database to construct a historical input-output residual matrix.

[0082] Subsequently, based on the principle of minimum residual, the locations of outlier nodes in the model are optimized in reverse, and the spatial coordinates, object pose, and recognition attributes are iteratively updated using a gradient descent strategy. During this process, the system introduces perturbation functions into key modeling parameters and performs fitting updates in conjunction with the target residual value until the overall residual falls below a dynamically adjusted threshold.

[0083] After optimization, output a corrected modeling result, and record the parameter set, convergence process and residual curve of this round of correction operation. Write this as a log to the system to support later error backtracking and iterative analysis.

[0084] S6. Output the calibration model and perform error verification iteration:

[0085] Finally, the corrected model is used as a candidate state for the current moment and re-compares and verifies it with real-time data.

[0086] The system re-performs 3D spatial location matching, timestamp consistency verification, and identification accuracy checks, and calculates the updated error score. If the score meets the update threshold criteria, it is officially written into the state database and published as a new modeling snapshot to the upper-level visualization platform and decision-making system; otherwise, the system allows up to two rounds of iterative correction. If convergence is still not achieved, a manual intervention mechanism is triggered to manually review the area.

[0087] Through the aforementioned dynamic feedback mechanism, it is ensured that the modeling process always possesses high precision, high responsiveness, and high adaptability, and that it has the ability to quickly correct and track errors when they occur.

[0088] Example 2: Digital Twin Warehouse Management System with Real-Time 3D Modeling

[0089] This embodiment provides a 3D real-time modeling digital twin warehouse management system. Built on an automated warehouse management platform, the system combines multi-source dynamic data acquisition with high-precision 3D modeling algorithms to achieve digital twin modeling, error correction, and real-time iterative updates of warehouse conditions. This system can be widely applied to highly dynamic scenarios such as large logistics warehouses, automated storage and retrieval systems, and cold chain distribution bases, offering advantages in accuracy, efficiency, and traceability.

[0090] like Figure 2 As shown, this system mainly comprises seven functional modules: environment initialization, data acquisition, time synchronization, 3D modeling, error measurement, correction, and cache database. These modules work collaboratively to form a closed-loop warehouse digital twin modeling system.

[0091] 1. Environment initialization module:

[0092] The environment initialization module is one of the most critical structural modules in the early stages of system operation. Its main task is to collect initial storage environment parameters and establish coordinate mapping relationships between these parameters and the actual physical scene.

[0093] This module includes a spatial structure recognition unit, a shelf calibration unit, and an equipment static positioning unit. Upon initial operation, the system uses LiDAR and 3D camera equipment deployed inside the warehouse to perform a global scan of the warehouse space, acquiring spatial coordinate data of structural boundaries such as the ceiling, walls, supporting columns, and floor. Subsequently, the module performs data matching based on a pre-set BIM model to generate a structural mapping diagram and calibrate the spatial boundary coordinate system.

[0094] For shelf numbers, the system uses visual recognition and code matching to retrieve the shelf code table from the warehouse management system (WMS) and uses a QR code recognition camera to mark its location. Equipment coordinates are automatically obtained through its power-on initialization status and static location point acquisition module, and the equipment number is bound to its spatial location according to the set identifier.

[0095] The entire initialization process is completed with millisecond-level time precision and outputs a complete initial storage space status description table, which will serve as a reference for subsequent modeling and error judgment.

[0096] 2. Data Acquisition Module:

[0097] The data acquisition module, as the input layer in the operation of this system, is mainly responsible for the comprehensive and unified collection and preprocessing of various real-time data in the dynamic process of warehousing.

[0098] Its core structure includes a data access unit, a data filtering unit, and a data deduplication unit. The system collects the following data in parallel through various acquisition terminals, including sensor nodes (temperature, humidity, smoke, vibration, etc.), RFID readers, automated equipment feedback ports, and work execution logs:

[0099] 1) The status of goods flow, including the time and location of actions such as warehousing, outbound, transfer, and inventory counting;

[0100] 2) Equipment scheduling status, including scheduling instructions and feedback status of stacker cranes, AGVs, and robotic arms;

[0101] 3) Environmental condition data, including light, humidity, temperature, vibration levels, etc. in different areas;

[0102] 4) System event information, such as task generation, abnormal alarms, path blocking, etc.

[0103] All data is uniformly accessed through the acquisition queue and categorized according to source identifier and time stamp. The data filtering unit is responsible for removing redundant task signals and duplicate sampled data to ensure the uniqueness and continuity of the data. After initial processing, the data is packaged and stored in a buffer, ready for the time synchronization processing flow.

[0104] 3. Time synchronization module:

[0105] To ensure consistency of state during subsequent modeling, the time synchronization module performs unified time alignment processing on all data sources.

[0106] This module introduces a global time synchronization mechanism, based on the deployed NTP master clock service, to provide a standard time source for all terminals and modules. All types of data are accompanied by a local timestamp during acquisition; the time synchronization module converts this to a unified format and completes time alignment through a three-stage mechanism:

[0107] In the first stage, data acquisition terminal type is sorted by coarse-grained time.

[0108] In the second stage, a data acquisition delay model is introduced to estimate the time offset of terminals with communication lag or processing bottlenecks.

[0109] In the third stage, linear interpolation or exponential smoothing algorithms are used to smooth the blank intervals or abrupt changes on the time axis, ensuring that each data point accurately corresponds to a unique time position.

[0110] Finally, the module outputs a set of data packets with unified time annotation, which have consistent time dimensions, clear sources, and complete content, making it easy for subsequent modeling modules to call them efficiently.

[0111] 4. 3D Modeling Module:

[0112] The 3D modeling module is the core computing module of this system. Its task is to receive time-series data and generate a real-time 3D warehouse status model through preset modeling rules.

[0113] This module includes a data aggregation unit, a modeling and calculation unit, and a model output unit. In the data aggregation unit, the system performs object-level fusion on the data within each time slice, that is, it summarizes RFID location, point cloud morphology, scheduling trajectory, and environmental status according to specific objects (such as pallets, AGVs, and shelf positions) to form the data entities required for modeling.

[0114] In the modeling and computation unit, the system employs an incremental modeling mechanism, constructing the spatial state (coordinates, dimensions, pose, status code) of objects based on the changes in each time slice, forming nodes in the 3D spatial model. The spatial construction engine draws the model boundaries based on point cloud and pose data, while preserving the association chain between the object and its behavioral trajectory, so that the model not only represents the current state but also records its temporal evolution logic.

[0115] The model output unit visualizes the modeling results in real time and stores them in the cache as snapshots for the error measurement module to access.

[0116] 5. Error Measurement Module:

[0117] The error measurement module is mainly used to quantitatively evaluate the difference between the current modeling results and the actual physical state, serving as a crucial basis for system closed-loop feedback. The core of this module consists of an error index set, an error function generator, and a goodness-of-fit criterion.

[0118] In this embodiment, the system uses three types of error parameters to form an error index set:

[0119] 1) Spatial position deviation That is, the three-dimensional difference between the coordinates of the modeled object and the measured coordinates;

[0120] 2) Time delay That is, the difference between the modeled state timestamp and the actual time of the event;

[0121] 3) Labeling error This refers to the inconsistency between the modeled object label and the physical object identification code.

[0122] The system uses a weighted linear combination of three types of error parameters to form a modeling error function. The calculation method is as follows:

[0123]

[0124] Where α, β, and γ are weight parameters preset by the system based on error sensitivity. The parameter number. The system... A threshold comparison is performed. If the error of all objects is lower than the set value, the current model is considered valid; otherwise, an error peak distribution map and a residual report will be output for the correction module to use.

[0125] 6. Calibration Module:

[0126] The correction module is the core mechanism for the system to quickly adjust abnormal modeling states in a closed loop. This module consists of three parts: an error localization unit, a data backtracking unit, and a residual optimization unit.

[0127] The error localization unit obtains the residual map from the error measurement module, identifies the key nodes with the most concentrated error distribution (such as abnormal trajectories, mismatched objects, or drift points), constructs a list of correction targets, and initiates corresponding backtracking data requests to the cache database.

[0128] After receiving the request, the data backtracking unit extracts the original input data and modeling snapshot of the target node within a set time period (such as the past 180 seconds) from the cache database, and constructs a complete time series mapping and state trajectory comparison diagram.

[0129] After receiving input, the residual optimization unit constructs an optimization model with the objective function of minimizing the fitting error. Based on perturbation iteration or gradient adjustment strategies, it fine-tunes the position, state, and attitude parameters of the target node. After optimization, the correction vector is fed back to the modeling module and triggers structural reconstruction, achieving a closed-loop correction of the modeling process. If the error still does not meet the standard after correction, the system will automatically iterate a second correction until the error meets the requirements or enters the manual review channel.

[0130] 7. Cache the database:

[0131] The cache database serves as the state storage and data backtracking function within this system, providing historical support for modeling and analysis. The database contains three core datasets:

[0132] 1) Real-time dataset: Contains multi-source data streams that have undergone time synchronization processing, with timestamps and object indexes;

[0133] 2) Modeling state set: contains snapshots of the 3D model generated in each round of modeling, recording the modeling version number, object structure, spatial state, and behavior sequence;

[0134] 3) Calibration log collection: records each error location, optimization parameter and result, supporting backtracking analysis of later model evolution.

[0135] The database design adopts a time-series data architecture, with an index structure built based on a time axis, supporting functions such as high-frequency reads, regional loading, and delayed callbacks. Through this database, the system achieves efficient mapping from the current modeling state to any historical state, and also provides basic data support for error residual analysis and predictive maintenance.

[0136] In summary, the 3D real-time modeling digital twin warehouse management system provided in this embodiment can fuse and model multi-dimensional dynamic data during the warehousing process, and maintain a high degree of synchronization between the digital twin state and the actual warehouse state through error measurement and automatic correction mechanisms. The logic between each functional module is clear and the boundaries are well-defined, enabling full-process control from perception, modeling, judgment, feedback to closed-loop updates. It is a highly practical, intelligent, and scalable warehouse modeling system solution.

[0137] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for real-time 3D modeling of digital twin warehouse management, characterized in that, Includes the following steps: S1. Initialize warehouse environment parameters: including the three-dimensional position of the warehouse entity, shelf number, equipment coordinates, entrance and exit coordinates, and loading and unloading operation area, and establish an association mapping between the initial state information and the actual warehouse environment; S2. Collect and access multiple types of real-time data, including data on goods entering and leaving the warehouse, handling, storage, unloading, and equipment operating status, and align and correct the collected timestamps through a synchronization mechanism based on a unified time source. S3. Construct a 3D real-time state model: The model is based on point cloud and vector data, and combines a dynamic data set that has been processed by historical state restoration to fit the current modeling state and form spatial structure constraints; The 3D real-time state model includes at least a point cloud dataset and corresponding time series identifiers. Each point cloud dataset includes spatial location, volume information and timestamp to realize the dynamic visualization construction of the 3D modeling scene. S4. Modeling Error Assessment and Fit Determination: After modeling is completed, the modeling error of the 3D state model is calculated using an error measurement method, and the modeling fit is determined based on a preset threshold. S4.1 If the modeling error is within the set threshold, the mapping data cache is updated and synchronously written to the modeling database as a snapshot of the digital model at the current moment. Updating the mapping data cache also includes establishing a structured data set on the modeling timeline and writing it to the modeling database with a timestamp index for tracing the historical state of the fitting correction process. S4.2 If the modeling error exceeds the set threshold, the distribution function of the modeling error under different data sources is statistically analyzed, and the two source regions with the largest error are identified as key regions for reverse correction. The parameters of the distribution function include the comprehensive calculation results of spatial coordinate difference, time delay and data fitting residual, and are sorted according to regional weight to lock the key regions with the greatest impact on the error. S5. Perform inverse fitting and historical residual calibration: Extract historical input data and modeling status records of key areas within a set time window, use the residual between the current modeling result and the historical record, adjust the position and attribute information of the model nodes through the minimum residual, generate the corrected modeling result and output it. S6. Output the correction model and perform error verification iteration: Use the modeling result optimized by minimizing the residual as the latest 3D model at the current moment, and compare and verify it with the current state of the real warehousing system again in terms of spatial location, timestamp and item identification 3D indicators.

2. The three-dimensional real-time modeling digital twin warehouse management method according to claim 1, characterized in that: In step S1, the initialization operation is triggered by the start signal of the warehousing system and includes automatic loading and identification of equipment coordinates, static mapping and marking of shelf numbers, and system preset correction of entrance and exit coordinates and loading and unloading operation areas.

3. The digital twin warehouse management method with real-time three-dimensional modeling according to claim 1, characterized in that: In step S2, the multi-type real-time data includes at least one set of data from sensor data, RFID read / write data, and automated handling equipment control command data, and is written to the middleware cache interface in a unified format.

4. The three-dimensional real-time modeling digital twin warehouse management method according to claim 1, characterized in that: In step S6, if all indicators meet the preset error threshold requirements, the status snapshot is updated and the next round of data access and modeling process is entered to realize the dynamic real-time iteration of the digital twin system.

5. The three-dimensional real-time modeling digital twin warehouse management method according to claim 1, characterized in that: In step S5, the reverse fitting and historical residual calibration are performed based on a set fixed time interval, the time interval being 3 minutes in length. The process is carried out in a preset step order, and the modeling residuals of the modeling nodes are recalculated in each interval to form a comparison table of residuals before and after optimization.

6. A three-dimensional real-time modeling digital twin warehouse management system, used to implement the method as described in any one of claims 1 to 5, characterized in that, include: The environment initialization module is used to collect initial warehouse environment parameters and establish a correspondence with the actual scenario; The data acquisition module is used to collect data, including unified collection, screening, and deduplication of data on cargo flow, equipment scheduling, environmental parameters, sensor sampling, and system events; The time synchronization module is used to establish time series, align the time of all connected data sources, and perform synchronization processing and data association through a unified timestamp dimension; The 3D modeling module is used to receive time-series data and generate real-time 3D modeling scenes through calculation rules. The error measurement module is used to calculate the time deviation, spatial deviation, and object deviation between the modeling model and the actual scene, and output the residual error to provide feedback on the modeling fit. The correction module is used to compare the accuracy of key areas in the modeling model based on the error measurement results, and to dynamically correct areas where the modeling error exceeds a preset threshold. The cache database is used to store the access data set, the corrected modeling state set, and bind historical state information associated with timestamps.

Images