Warehouse visual management method based on digital twinning

Abstract

The invention relates to the field of intelligent warehousing, and discloses a digital twinning-based warehouse visual management method, which comprises the steps of deploying an AOA-RFID reader-writer array on a warehousing shelf, and setting an RFID tag on the surface of a goods; the method comprises the following steps: acquiring initial positioning data of goods through an AOA-RFID reader-writer array; performing label signal intensity compensation on the initial positioning data to obtain label signal intensity compensation data; deploying a visual auxiliary correction system based on the label signal intensity compensation data, and identifying the actual position of the RFID label in the goods shelf through a camera to obtain visual positioning data; and performing positioning deviation compensation on the initial positioning data of the goods according to the AOA positioning error data to obtain positioning data after visual correction. The RFID signal attenuation is dynamically corrected through a metal reflection characteristic analysis and temperature correlation model, and the positioning error is reduced to a sub-meter level in combination with visual characteristic matching and illumination adaptive adjustment; and the goods checking and sorting efficiency is obviously improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent warehousing technology, specifically to a warehouse visualization management method based on digital twins. Background Technology

[0002] With the deep integration of IoT technology and warehouse management, RFID-based positioning systems have become a core component of smart warehouses. However, traditional AOA-RFID positioning technology, which combines angle of arrival (AOA) with radio frequency identification (RFID), analyzes the angle of arrival of RFID tag signals and uses positioning algorithms to calculate the spatial position of the tags, achieving high-precision tracking of objects or people. In practical applications, it has significant limitations: multipath effects from metal shelving lead to signal attenuation and positioning errors; dynamic changes in environmental obstructions (such as forklifts and cargo containers) further exacerbate positioning errors; and relying solely on RFID or visual positioning solutions makes it difficult to balance accuracy and cost.

[0003] Existing equipment has the following drawbacks: traditional compensation methods do not consider the impact of dynamic environmental factors such as temperature and light on the reflective properties of metals, resulting in fluctuations in compensation effects; the weight allocation of RFID and visual positioning data relies on static presets and cannot respond in real time to changes in the degree of occlusion; 3D warehouse models usually only display static layouts and lack deep integration with real-time positioning data and dynamic weight strategies; existing systems focus on data collection and display and lack proactive decision-making capabilities based on trajectory prediction and spatial optimization.

[0004] Therefore, this application proposes a warehouse visualization management method based on digital twins to solve the above-mentioned problems. Summary of the Invention

[0005] The purpose of this invention is to provide a warehouse visualization management method based on digital twins, in order to solve the problems of the above-mentioned traditional compensation methods not considering the influence of dynamic environmental factors such as temperature and light on the metal reflection characteristics, resulting in fluctuations in the compensation effect; and the weight allocation of RFID and visual positioning data relying on static presets, which cannot respond to changes in the degree of occlusion in real time.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a warehouse visualization management method based on digital twins, comprising: An AOA-RFID reader array is deployed on the warehouse rack, and RFID tags are affixed to the surface of the goods. Initial positioning data of the goods is collected through the AOA-RFID reader array. Tag signal strength compensation is performed on the initial positioning data to obtain tag signal strength compensation data. A visual-assisted correction system is deployed based on tag signal strength compensation data. The system uses a camera to identify the actual position of RFID tags on the shelf to obtain visual positioning data. The visual positioning data is then compared with the initial positioning data of the goods to obtain AOA positioning error data. The initial positioning data of the goods is then compensated for the positioning deviation based on the AOA positioning error data to obtain the visually corrected positioning data. A dynamic weight allocation model is deployed based on the visually corrected positioning data. Real-time occlusion data of the warehouse is collected by environmental sensors. An occlusion level threshold is constructed based on the real-time occlusion data of the warehouse to obtain occlusion level data. Based on the occlusion level data, the weight coefficients of AOA positioning data and visual positioning data are calculated to obtain dynamic weight allocation data. A digital twin warehouse model is constructed based on dynamic weight allocation data. The visually corrected positioning data and dynamic weight allocation data are then integrated into the digital twin warehouse model to obtain three-dimensional visualized warehouse data. The three-dimensional visualized warehouse data is used to predict the trajectory of goods and analyze the space utilization rate through a cloud platform to obtain warehouse management decision data. The warehouse management decision data is then sent to the warehouse management system to execute visualized warehouse management.

[0007] Specifically, tag signal strength compensation is performed on the initial positioning data to obtain tag signal strength compensated data, including: Environmental metal reflection characteristics were analyzed on the initial positioning data to obtain metal reflection interference data; a multipath effect simulation model was constructed based on the metal reflection interference data to obtain multipath effect simulation data. A dynamic signal attenuation compensation strategy is deployed based on multipath effect simulation data. Tag signal strength fluctuation data is collected through an AOA-RFID reader array. The tag signal strength fluctuation data is correlated with the multipath effect simulation data to obtain signal attenuation correlation data. Based on the signal attenuation correlation data, a tag orientation adaptive compensation table is constructed to obtain the orientation adaptive compensation table data. A vision-assisted positioning system is deployed based on the orientation adaptive compensation table data to obtain vision correction compensation coefficient data; the signal strength in the initial positioning data is dynamically weighted according to the vision correction compensation coefficient data to obtain the signal strength data after vision assistance compensation. An environment-adaptive learning model is deployed based on visually assisted compensation signal strength data. Temperature data of the shelf area is collected by a temperature sensor. The temperature data is correlated with the visually assisted compensation signal strength data to obtain temperature-related reflection characteristic data. The direction-adaptive compensation table is dynamically updated based on the temperature-related reflection characteristic data to generate label signal strength compensation data.

[0008] Among them, a visual-assisted correction system is deployed based on tag signal strength compensation data. This system uses cameras to identify the actual position of RFID tags on the shelf, obtaining visual positioning data, including: Based on the tag signal strength compensation data, deploy an array of cameras to collect RGB-D depth images of the shelf panels, and obtain shelf panel depth image data; perform product region segmentation on the shelf panel depth image data to obtain product outline positioning data; Based on the cargo outline positioning data, a tag feature matching algorithm is deployed to extract the visual feature points of the RFID tags through the camera module; the visual feature points are associated and mapped with the tag ID in the tag signal strength compensation data to obtain tag feature association data; the camera module is automatically focused based on the tag feature association data to obtain visual focus optimization data. Based on visual focus optimization data, a 3D coordinate transformation model is deployed to convert the pixel coordinates in the shelf depth image data into physical coordinates. Combined with the AOA positioning results in the tag signal strength compensation data, multi-source data fusion is performed on the physical coordinates to obtain visual AOA fused positioning data. The visual AOA fused positioning data is then smoothed by Kalman filtering to obtain visually corrected positioning data. An environment-adaptive learning model is deployed based on the visually corrected positioning data. Light intensity data of the shelf area is collected by a light sensor. The real-time light intensity data and the visually corrected positioning data are correlated and analyzed to obtain the light correlation attenuation law data. The exposure parameters of the camera module are dynamically adjusted based on the light correlation attenuation law data to obtain visual positioning data.

[0009] Among them, error mapping analysis is performed between visual positioning data and initial product positioning data to obtain AOA positioning error data, including: Visual positioning data and initial product positioning data are input into the coordinate alignment module. The polar coordinate positioning result of AOA is converted into the Cartesian coordinate system of the vision system through a spatial coordinate transformation algorithm to obtain cross-modal coordinate alignment data. The cross-modal coordinate alignment data is then processed with timestamp synchronization to obtain spatiotemporal synchronized positioning data. Based on the deployment of spatiotemporal synchronous positioning data, an error calculation engine is used to calculate the deviation between the actual coordinates of the same RFID tag in visual positioning and the coordinates in AOA positioning through the Euclidean distance algorithm, thereby obtaining tag-level error distribution data. Based on the label-level error distribution data, an environmental feature association model is deployed, and a clustering algorithm is used to identify the association between high-error areas and shelf structures to obtain environmental association error pattern data. An error mapping learning network is deployed based on environmental correlation error pattern data. An error prediction model is constructed using the random forest algorithm. The predicted error value and the actual measurement error are iteratively trained to obtain AOA positioning error data.

[0010] Specifically, the initial positioning data of the goods is compensated for based on the AOA positioning error data to obtain the visually corrected positioning data, including: An error compensation model is constructed based on AOA positioning error data and initial positioning data of goods. Environmental interference characteristic data are obtained by collecting data on metal reflection intensity and distribution of obstructions in the shelf area through environmental sensors. A dynamic weight allocation engine is deployed based on environmental interference feature data, and a fuzzy logic algorithm is used to calculate the credibility weights of AOA positioning data and visual positioning data. Dynamic weight allocation data is obtained by comparing environmental interference feature data with historical error pattern library in real time. The initial positioning data of goods is weighted and fused based on dynamic weight allocation data. The AOA positioning result is multiplied by the corresponding weight coefficient and then superimposed with the visual positioning data. The superimposed positioning result is smoothed by Kalman filtering to obtain the visually corrected positioning data. The visually corrected positioning data is fed back to the error compensation model, and the environmental interference feature vector and weight allocation strategy are updated through reinforcement learning algorithm. Combined with the real-time positioning accuracy evaluation results, the fuzzy logic rule base is dynamically optimized to obtain positioning deviation compensation data.

[0011] Specifically, based on real-time warehouse occlusion data, occlusion severity grading thresholds are constructed to obtain occlusion severity grading data, including: The visually corrected positioning data and the raw occlusion data collected by the environmental sensors are input into the modal fusion module. The spatial correlation algorithm is used to identify the types of occlusions around the goods and obtain environmental occlusion feature data. A dynamic clustering engine is deployed based on statistical feature data of occlusion degree. The fuzzy C-means clustering algorithm is used to perform soft classification of occlusion degree to obtain a fuzzy membership matrix. Combined with warehouse shelf density and aisle width parameters, the fuzzy membership matrix is ​​adaptively corrected to obtain preliminary occlusion level classification data. Based on the initial classification of occlusion levels, a threshold verification module is deployed. The AOA visual fusion positioning accuracy corresponding to the occlusion level is extracted from the historical positioning error database. The fuzzy clustering center is adjusted using the error backpropagation mechanism to obtain optimized occlusion level classification threshold data.

[0012] Among them, based on the occlusion level classification data, the weight coefficients of AOA positioning data and visual positioning data are calculated to obtain dynamic weight allocation data, including: A weight mapping table is constructed based on the occlusion level classification data, and the AOA and visual weight initial values ​​corresponding to the occlusion label are defined. The accuracy performance of the two positioning technologies under each occlusion level is extracted from the historical positioning error database to obtain the initial weight configuration data. Deploy a weight calculation engine based on the initial weight configuration data, and collect data on the current warehouse occlusion level through environmental sensors to obtain spatially adaptive weight data; A dynamic feedback adjustment module is deployed based on spatial adaptive weight data. AOA and visual positioning data are input into the error evaluation model to obtain the weight data after feedback optimization. The optimized weight data is synchronized to the positioning fusion module, and dynamic weight allocation data is obtained through a weighted average algorithm.

[0013] Specifically, a digital twin warehouse model is constructed based on dynamic weight allocation data. Visually corrected positioning data is then integrated with the dynamic weight allocation data into the digital twin warehouse model to obtain three-dimensional visualized warehouse data, including: The basic framework of a digital twin warehouse is constructed based on dynamic weight allocation data. The virtualization priority of shelf and aisle elements is defined by weight data. Lightweight modeling technology is used to generate an interactive 3D warehouse skeleton, resulting in weight-driven basic model data. The visually corrected positioning data is mapped to a weight-driven base model, and the actual product location is bound to the virtual model coordinate system through a coordinate alignment algorithm to obtain hierarchical positioning mapping data. Deploy a real-time rendering engine based on hierarchical positioning mapping data, control the model update frequency through weight data, and generate rendering strategy optimization data. By integrating rendering strategy optimization data with visually corrected positioning data, a browser-side 3D visualization scene is constructed using WebGL technology; dynamic weight allocation data is embedded into scene element attributes to obtain 3D visualization repository data.

[0014] Among these methods, the cloud platform is used to predict the trajectory of goods and analyze space utilization of 3D visualized warehouse data, resulting in warehouse management decision-making data, including: The structured analysis of the 3D visualized warehouse data extracts the temporal data of the goods location and the spatial coordinate information of the shelves; the 3D coordinates are converted into a tensor format that can be recognized by the trajectory prediction model through a spatiotemporal feature encoder to obtain structured warehouse spatiotemporal data. A hybrid prediction engine is deployed based on structured warehouse spatiotemporal data, and a cargo trajectory prediction model is constructed by combining long short-term memory networks and reinforcement learning algorithms; historical location data and dynamic weight allocation data are input to obtain trajectory prediction results. Based on the trajectory prediction results, a space utilization analysis module is deployed to calculate the occupancy rate between shelf layers and the passage efficiency of aisles using a three-dimensional geometric algorithm; combined with the occupancy level classification data, inefficient storage areas are identified to obtain space analysis decision data; The trajectory prediction results and spatial analysis decision data are input into a multi-objective optimization solver to balance the frequency of goods retrieval, shelf load-bearing limits, and aisle flow requirements; a genetic algorithm is used to generate dynamic location adjustment schemes and picking path planning strategies to obtain warehouse management decision data.

[0015] The calculation of the direction adaptive compensation table data is as follows: in, Data representing the correlation patterns of signal attenuation. This indicates the fluctuation data of the tag signal strength over time. The value of , This represents the mean of the tag signal strength fluctuation data. This indicates that the simulation data of multipath effects is in time. The value of , This represents the mean of the multipath effect simulation data. This indicates the total amount of data.

[0016] Compared with the prior art, the beneficial effects of the present invention are: (1) This invention uses metal reflection characteristics analysis and temperature correlation model to dynamically correct RFID signal attenuation, and combines visual feature matching and adaptive lighting adjustment to reduce positioning error to sub-meter level, significantly improving the efficiency of inventory counting and picking. (2) This invention adjusts the fusion ratio of AOA and visual data in real time through the occlusion degree classification threshold and dynamic weight engine, automatically enhances the visual positioning weight in high occlusion scenarios, and prioritizes the use of RFID data in open areas, balancing accuracy and computational resource consumption. (3) This invention achieves millisecond-level synchronization between location data and virtual scene by constructing a weight-driven three-dimensional warehouse model, and displays the degree of occlusion and positioning accuracy through heat map overlay, enabling managers to intuitively identify inefficient storage areas; (4) This invention integrates trajectory prediction and multi-objective optimization algorithms through a cloud platform, combines historical movement data of goods with real-time positioning information, automatically generates dynamic storage location adjustment schemes, improves shelf space utilization, forms a closed loop from signal compensation, multimodal fusion to decision execution, continuously optimizes compensation models and weight strategies through reinforcement learning, adapts to dynamic changes in the storage environment, reduces the need for manual intervention, and achieves truly unmanned intelligent management. Attached Figure Description

[0017] Figure 1This is a schematic diagram of the overall method's steps in one embodiment of the present invention; Figure 2 This is a schematic diagram of the process structure for generating tag signal strength compensation data in one embodiment of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Please see Figure 1-2 This invention provides a technical solution: a warehouse visualization management method based on digital twins, comprising: S1. Deploy an AOA-RFID reader array on the warehouse rack and set RFID tags on the surface of the goods; collect the initial positioning data of the goods through the AOA-RFID reader array; perform tag signal strength compensation on the initial positioning data to obtain tag signal strength compensation data; S2. Deploy a visual auxiliary correction system based on tag signal strength compensation data. Use a camera to identify the actual position of RFID tags on the shelf to obtain visual positioning data. Perform error mapping analysis between the visual positioning data and the initial positioning data of the goods to obtain AOA positioning error data. Compensate for the positioning deviation of the initial positioning data of the goods based on the AOA positioning error data to obtain the visually corrected positioning data. S3. Deploy a dynamic weight allocation model based on the visually corrected positioning data. Collect real-time occlusion data of the warehouse through environmental sensors. Construct occlusion level thresholds based on the real-time occlusion data of the warehouse to obtain occlusion level data. Calculate the weight coefficients of AOA positioning data and visual positioning data based on the occlusion level data to obtain dynamic weight allocation data. S4. Construct a digital twin warehouse model based on dynamic weight allocation data, and integrate the visually corrected positioning data with the dynamic weight allocation data into the digital twin warehouse model to obtain three-dimensional visualized warehouse data; perform goods trajectory prediction and space utilization analysis on the three-dimensional visualized warehouse data through a cloud platform to obtain warehouse management decision data; send the warehouse management decision data to the warehouse management system to execute warehouse visualization management.

[0020] It's important to note that during operation, AOA (Angle of Arrival) is a technology used to measure the direction of arrival of wireless signals. AOA technology determines the direction of signal propagation by the phase difference between receiving devices. RFID (Radio Frequency Identification) technology uses radio frequency signals to automatically identify objects and collect data. An RFID system includes tags, readers, and a data processing system. An AOA-RFID reader array combines AOA and RFID technologies, using multiple reader arrays for high-precision identification and location of product tags. An RFID tag is a small electronic device used for object identification, typically containing a microchip and an antenna. It communicates with the RFID reader via wireless signals, transmitting the item's unique identification information. Tag signal strength compensation addresses the potential interference from the surrounding environment with RFID tag signal strength, using algorithms to adjust for signal strength variations and ensure accurate positioning data. Visual-assisted correction corrects RFID positioning data using visual technologies such as cameras and image recognition. In warehouses, cameras identify the location of RFID tags and correct initial positioning data for errors. Error mapping analysis analyzes errors between different positioning systems, identifies error sources, and corrects them. In AOA technology, errors typically arise from factors such as multipath propagation and device accuracy. Positioning error data refers to the difference between the location calculated by the AOA positioning system and the actual location. Positioning deviation compensation: corrects positioning errors to make the positioning results more accurate. Different algorithms and sensor data are typically used to adjust the initial positioning result. Dynamic weight allocation model: adjusts the weights of data sources based on different factors to obtain optimal positioning results in different environments. Environmental sensors: devices used to collect environmental data, such as temperature, humidity, occlusion level, and lighting intensity. In warehouse management, environmental sensors help monitor the degree of occlusion between goods and sensors, thus affecting positioning accuracy. Occlusion level classification threshold: classifies different levels of occlusion based on occlusion data, such as complete occlusion, partial occlusion, and no occlusion, to adjust the weights of positioning technologies under different conditions. Digital twin: a digital copy of a physical entity. It reflects the dynamic changes of the real world through real-time data and virtual modeling technology. A digital twin warehouse model is a virtual replica of the warehouse, combining information from RFID, visual data, and environmental sensors to achieve more efficient warehouse management and optimization. 3D visualization uses computer graphics technology to transform warehouse space and product location data into 3D images for a more intuitive understanding of warehouse layout and product location. Product trajectory prediction predicts the movement path or future location of goods based on existing data. Space utilization analysis analyzes the efficiency of space use within the warehouse, such as shelf stacking density and aisle space utilization, thereby optimizing warehouse layout and resource use. A warehouse management system is a software system used to manage the flow, storage, and distribution of goods in the warehouse.

[0021] In one embodiment, tag signal strength compensation is performed on the initial positioning data to obtain tag signal strength compensated data, including: A1. Analyze the environmental metal reflection characteristics of the initial positioning data to obtain metal reflection interference data; construct a multipath effect simulation model based on the metal reflection interference data to obtain multipath effect simulation data. A2. Deploy a dynamic signal attenuation compensation strategy based on multipath effect simulation data. Collect tag signal strength fluctuation data through an AOA-RFID reader array. Perform correlation matching between the tag signal strength fluctuation data and multipath effect simulation data to obtain signal attenuation correlation data. Construct a tag direction adaptive compensation table based on the signal attenuation correlation data to obtain the direction adaptive compensation table data. A3. Deploy a vision-assisted positioning system based on the orientation adaptive compensation table data to obtain vision correction compensation coefficient data; dynamically weight the signal strength in the initial positioning data according to the vision correction compensation coefficient data to obtain the signal strength data after vision-assisted compensation. A4. Deploy an environment-adaptive learning model based on visually assisted compensation signal strength data, collect temperature data of the shelf area through temperature sensors, perform correlation analysis between temperature data and visually assisted compensation signal strength data to obtain temperature-related reflection characteristic data, and dynamically update the direction-adaptive compensation table based on temperature-related reflection characteristic data to generate label signal strength compensation data.

[0022] This design first analyzes the metal reflection characteristics in the environment, identifies metal reflection interference, and establishes a multipath effect simulation model based on this. Since signal propagation along different paths may cause interference, simulating these effects helps predict signal distortion. Next, multipath effect simulation data is used to deploy a signal attenuation compensation strategy. Tag signal fluctuation data is collected using an AOA-RFID array and matched with multipath effect data to obtain signal attenuation patterns, thereby constructing a direction-adaptive compensation table. This step compensates for fluctuations by adjusting signal strength in different directions, improving positioning accuracy. A vision-assisted positioning system is then used to further compensate the positioning signal. By calculating visual correction compensation coefficients, the signal strength in the initial positioning data is dynamically weighted to make the signal more accurate. Finally, using an environment-adaptive learning model, temperature data is collected via a temperature sensor and correlated with the compensated signal strength to discover the impact of temperature on signal strength. Based on the correlation between temperature and signal strength, the direction-adaptive compensation table is dynamically updated, ultimately generating more accurate tag signal strength compensation data.

[0023] In one embodiment, a visual-assisted correction system is deployed based on tag signal strength compensation data. This system uses a camera to identify the actual position of RFID tags on a shelf, obtaining visual positioning data, including: Based on the tag signal strength compensation data, deploy an array of cameras to collect RGB-D depth images of the shelf panels, and obtain shelf panel depth image data; perform product region segmentation on the shelf panel depth image data to obtain product outline positioning data; Based on the cargo outline positioning data, a tag feature matching algorithm is deployed to extract the visual feature points of the RFID tags through the camera module; the visual feature points are associated and mapped with the tag ID in the tag signal strength compensation data to obtain tag feature association data; the camera module is automatically focused based on the tag feature association data to obtain visual focus optimization data. Based on visual focus optimization data, a 3D coordinate transformation model is deployed to convert the pixel coordinates in the shelf depth image data into physical coordinates. Combined with the AOA positioning results in the tag signal strength compensation data, multi-source data fusion is performed on the physical coordinates to obtain visual AOA fused positioning data. The visual AOA fused positioning data is then smoothed by Kalman filtering to obtain visually corrected positioning data. An environment-adaptive learning model is deployed based on the visually corrected positioning data. Light intensity data of the shelf area is collected by a light sensor. The real-time light intensity data and the visually corrected positioning data are correlated and analyzed to obtain the light correlation attenuation law data. The exposure parameters of the camera module are dynamically adjusted based on the light correlation attenuation law data to obtain visual positioning data.

[0024] This design involves several steps. First, based on tag signal strength compensation data, an array of cameras is deployed to capture RGB-D depth images of the shelf shelves, generating shelf depth image data. Next, product area segmentation is performed to obtain product outline positioning data, providing a foundation for subsequent tag positioning. Then, a tag feature matching algorithm maps the visual feature points of the RFID tags extracted by the camera to the tag IDs in the tag signal strength compensation data, obtaining tag feature association data. This helps ensure accurate focusing, allowing for automatic camera focus adjustment, optimizing visual focus data, and improving image clarity. Next, a 3D coordinate transformation model converts the pixel coordinates in the depth image to physical coordinates, and combines this with the AOA positioning results from the tag signal strength compensation data to obtain visual AOA fusion positioning data through multi-source data fusion. This data is then smoothed using Kalman filtering to further improve positioning accuracy. Finally, an environment adaptive learning model is used to collect real-time illumination intensity data through a light sensor, and this data is correlated with the positioning data to obtain illumination attenuation data. Based on this data, the exposure parameters of the camera module are dynamically adjusted to compensate for the impact of illumination changes on positioning accuracy, ensuring accurate visual positioning data is ultimately obtained.

[0025] In one embodiment, visual positioning data is compared with initial product positioning data to obtain AOA positioning error data, including: Visual positioning data and initial product positioning data are input into the coordinate alignment module. The polar coordinate positioning result of AOA is converted into the Cartesian coordinate system of the vision system through a spatial coordinate transformation algorithm to obtain cross-modal coordinate alignment data. The cross-modal coordinate alignment data is then processed with timestamp synchronization to obtain spatiotemporal synchronized positioning data. Based on the deployment of spatiotemporal synchronous positioning data, an error calculation engine is used to calculate the deviation between the actual coordinates of the same RFID tag in visual positioning and the coordinates in AOA positioning through the Euclidean distance algorithm, thereby obtaining tag-level error distribution data. Based on the label-level error distribution data, an environmental feature association model is deployed, and a clustering algorithm is used to identify the association between high-error areas and shelf structures to obtain environmental association error pattern data. An error mapping learning network is deployed based on environmental correlation error pattern data. An error prediction model is constructed using the random forest algorithm. The predicted error value and the actual measurement error are iteratively trained to obtain AOA positioning error data.

[0026] This design involves several steps. First, the coordinate alignment module uses a spatial coordinate transformation algorithm to convert the polar coordinate positioning results of the AOA (Optical Aspect-Oriented Alignment) system into the Cartesian coordinate system of the vision system, achieving cross-modal coordinate alignment. This transformation ensures that data from two different positioning methods can be compared within the same coordinate system. Next, timestamp synchronization is performed to ensure temporal consistency between visual and AOA positioning data, resulting in spatiotemporally synchronized positioning data. Then, the error calculation engine uses the Euclidean distance algorithm to calculate the coordinate deviation of the same RFID tag in visual and AOA positioning, obtaining tag-level error distribution data. This process helps quantify the error differences between visual and AOA positioning. Based on the error distribution data, an environmental feature association model is deployed, and a clustering algorithm is used to identify the correlation between high-error areas and shelf structures, generating environmental association error pattern data. This helps reveal the impact of environmental factors on positioning errors. Finally, an error prediction model is constructed using an error mapping learning network and a random forest algorithm, and iterative error training is performed to optimize the model's prediction accuracy. Through repeated training, the model can provide more accurate AOA positioning error data, thereby effectively improving the accuracy of the positioning system.

[0027] In one embodiment, positioning deviation compensation is performed on the initial positioning data of the goods based on AOA positioning error data to obtain visually corrected positioning data, including: An error compensation model is constructed based on AOA positioning error data and initial positioning data of goods. Environmental interference characteristic data are obtained by collecting data on metal reflection intensity and distribution of obstructions in the shelf area through environmental sensors. A dynamic weight allocation engine is deployed based on environmental interference feature data, and a fuzzy logic algorithm is used to calculate the credibility weights of AOA positioning data and visual positioning data. Dynamic weight allocation data is obtained by comparing environmental interference feature data with historical error pattern library in real time. The initial positioning data of goods is weighted and fused based on dynamic weight allocation data. The AOA positioning result is multiplied by the corresponding weight coefficient and then superimposed with the visual positioning data. The superimposed positioning result is smoothed by Kalman filtering to obtain the visually corrected positioning data. The visually corrected positioning data is fed back to the error compensation model, and the environmental interference feature vector and weight allocation strategy are updated through reinforcement learning algorithm. Combined with the real-time positioning accuracy evaluation results, the fuzzy logic rule base is dynamically optimized to obtain positioning deviation compensation data.

[0028] This design involves several steps. First, an error compensation model combines AOA positioning error data with initial product positioning data to construct a compensation model. Environmental sensors then collect data on the metal reflection intensity and obstruction distribution in the shelf area to obtain environmental interference characteristic data. These environmental factors can affect positioning accuracy, thus requiring dynamic compensation. Next, a dynamic weight allocation engine uses a fuzzy logic algorithm to calculate the credibility weights of the AOA positioning data and the visual positioning data. By comparing the environmental interference characteristic data with a historical error pattern library in real time, dynamic weight allocation data is generated, ensuring that positioning data is appropriately weighted according to credibility under different environmental conditions. Based on the weight allocation data, the weighted fusion process multiplies the AOA positioning result by the corresponding weight coefficient and superimposes it with the visual positioning data. This weighted fusion process improves the accuracy of multi-source data, and Kalman filtering is used to smooth the superimposed data, ultimately yielding visually corrected positioning data. Finally, a feedback mechanism continuously optimizes the environmental interference feature vector and weight allocation strategy through a reinforcement learning algorithm, and dynamically adjusts the fuzzy logic rule base based on real-time positioning accuracy evaluation results to continuously improve the accuracy of the positioning deviation compensation data.

[0029] In one embodiment, occlusion severity grading thresholds are constructed based on real-time warehouse occlusion severity data to obtain occlusion severity grading data, including: The visually corrected positioning data and the raw occlusion data collected by the environmental sensors are input into the modal fusion module. The spatial correlation algorithm is used to identify the types of occlusions around the goods and obtain environmental occlusion feature data. A dynamic clustering engine is deployed based on statistical feature data of occlusion degree. The fuzzy C-means clustering algorithm is used to perform soft classification of occlusion degree to obtain a fuzzy membership matrix. Combined with warehouse shelf density and aisle width parameters, the fuzzy membership matrix is ​​adaptively corrected to obtain preliminary occlusion level classification data. Based on the initial classification of occlusion levels, a threshold verification module is deployed. The AOA visual fusion positioning accuracy corresponding to the occlusion level is extracted from the historical positioning error database. The fuzzy clustering center is adjusted using the error backpropagation mechanism to obtain optimized occlusion level classification threshold data.

[0030] This design involves several steps. First, the modal fusion module integrates the visually corrected positioning data with the raw occlusion data collected by environmental sensors. A spatial association algorithm identifies the types of occlusions around the goods, yielding environmental occlusion feature data. These features reflect the impact of occlusions on the positioning system, providing a foundation for subsequent analysis. Next, the dynamic clustering engine uses a fuzzy C-means clustering algorithm to softly classify occlusion levels based on statistical feature data of occlusion severity, generating a fuzzy membership matrix. This process helps to subdivide different levels of occlusion and accurately analyze occlusion phenomena in the warehouse environment. Simultaneously, by incorporating warehouse shelf density and aisle width parameters, the membership matrix is ​​adaptively corrected to reflect the actual occlusion situation in the warehouse environment, resulting in preliminary occlusion level classification data. Then, the threshold verification module extracts the AOA and visual fusion positioning accuracy corresponding to different occlusion levels from a historical positioning error database. This verification step helps evaluate the performance of the positioning system under different occlusion conditions. Through an error backpropagation mechanism, the fuzzy cluster centers are optimized, ultimately yielding optimized occlusion severity classification threshold data.

[0031] In one embodiment, weighting coefficients are calculated for AOA positioning data and visual positioning data based on occlusion level classification data to obtain dynamic weight allocation data, including: A weight mapping table is constructed based on the occlusion level classification data, and the AOA and visual weight initial values ​​corresponding to the occlusion label are defined. The accuracy performance of the two positioning technologies under each occlusion level is extracted from the historical positioning error database to obtain the initial weight configuration data. Deploy a weight calculation engine based on the initial weight configuration data, and collect data on the current warehouse occlusion level through environmental sensors to obtain spatially adaptive weight data; A dynamic feedback adjustment module is deployed based on spatial adaptive weight data. AOA and visual positioning data are input into the error evaluation model to obtain the weight data after feedback optimization. The optimized weight data is synchronized to the positioning fusion module, and dynamic weight allocation data is obtained through a weighted average algorithm.

[0032] This design involves several steps. First, a weight mapping table is constructed based on occlusion level classification data, defining initial weight values ​​for AOA and visual positioning technologies corresponding to different occlusion levels. Initial weight configuration data is obtained by extracting the accuracy performance of the two positioning technologies under different occlusion levels from a historical positioning error database, laying the foundation for subsequent dynamic weight calculations. Next, the weight calculation engine collects occlusion level data from the current warehouse and combines it with environmental sensor information to calculate spatially adaptive weight data. This data reflects the impact of occlusion factors on positioning accuracy in the current warehouse environment, providing a dynamic adjustment basis for the weighted fusion of positioning data. Subsequently, the dynamic feedback adjustment module inputs AOA and visual positioning data into the error evaluation model based on the spatially adaptive weight data, evaluating the accuracy of the current weight configuration in real time and optimizing the weights to obtain optimized feedback weight data. This process ensures that the weight configuration can adaptively adjust with environmental changes, improving positioning accuracy. Finally, the optimized feedback weight data is synchronized to the positioning fusion module and fused using a weighted average algorithm to obtain dynamic weight allocation data. This process effectively improves the accuracy of multi-source data fusion, ensuring the system can adapt to complex warehouse environment changes.

[0033] In one embodiment, a digital twin warehouse model is constructed based on dynamic weight allocation data. Visually corrected positioning data is then fused with the dynamic weight allocation data into the digital twin warehouse model to obtain three-dimensional visualized warehouse data, including: The basic framework of a digital twin warehouse is constructed based on dynamic weight allocation data. The virtualization priority of shelf and aisle elements is defined by weight data. Lightweight modeling technology is used to generate an interactive 3D warehouse skeleton, resulting in weight-driven basic model data. The visually corrected positioning data is mapped to a weight-driven base model, and the actual product location is bound to the virtual model coordinate system through a coordinate alignment algorithm to obtain hierarchical positioning mapping data. Deploy a real-time rendering engine based on hierarchical positioning mapping data, control the model update frequency through weight data, and generate rendering strategy optimization data. By integrating rendering strategy optimization data with visually corrected positioning data, a browser-side 3D visualization scene is constructed using WebGL technology; dynamic weight allocation data is embedded into scene element attributes to obtain 3D visualization repository data.

[0034] This design, firstly, constructs the basic framework of the digital twin warehouse through dynamically weighted data. Weight data defines the priority of shelf and aisle elements within the virtual model, laying the foundation for subsequent 3D modeling. The 3D skeleton generated using lightweight modeling techniques enables the warehouse's virtual model to be efficient and highly interactive, resulting in simplified weight-driven basic model data. Next, visually corrected positioning data is mapped onto the weight-driven basic model. Through coordinate alignment algorithms, the actual locations of goods are bound to the coordinate system of the virtual model, forming hierarchical positioning mapping data. This mapping ensures consistency in position between real-world items and those in the virtual model, providing accurate basic data for further warehouse management. Based on this, a real-time rendering engine is deployed, controlling the model update frequency according to the weight data to ensure the virtual warehouse model reflects changes in the actual environment in real time. Simultaneously, data optimization through rendering strategies further improves the efficiency and accuracy of model rendering. Finally, using WebGL technology, the optimized rendering data is combined with the visually corrected positioning data to construct a browser-based 3D visualization scene. By embedding dynamic weight allocation data into scene element attributes, the 3D warehouse data not only has real-time performance, but also provides efficient and accurate decision support for warehouse management through dynamic updates.

[0035] In one embodiment, a cloud platform is used to predict the trajectory of goods and analyze space utilization of 3D visualized warehouse data to obtain warehouse management decision data, including: The structured analysis of the 3D visualized warehouse data extracts the temporal data of the goods location and the spatial coordinate information of the shelves; the 3D coordinates are converted into a tensor format that can be recognized by the trajectory prediction model through a spatiotemporal feature encoder to obtain structured warehouse spatiotemporal data. A hybrid prediction engine is deployed based on structured warehouse spatiotemporal data, and a cargo trajectory prediction model is constructed by combining long short-term memory networks and reinforcement learning algorithms; historical location data and dynamic weight allocation data are input to obtain trajectory prediction results. Based on the trajectory prediction results, a space utilization analysis module is deployed to calculate the occupancy rate between shelf layers and the passage efficiency of aisles using a three-dimensional geometric algorithm; combined with the occupancy level classification data, inefficient storage areas are identified to obtain space analysis decision data; The trajectory prediction results and spatial analysis decision data are input into a multi-objective optimization solver to balance the frequency of goods retrieval, shelf load-bearing limits, and aisle flow requirements; a genetic algorithm is used to generate dynamic location adjustment schemes and picking path planning strategies to obtain warehouse management decision data.

[0036] This design first involves structured parsing of the 3D visualized warehouse data, extracting temporal data on product locations and shelf space coordinates. Through a spatiotemporal feature encoder, the 3D coordinates are converted into a tensor format recognizable by the trajectory prediction model, forming structured warehouse spatiotemporal data, providing high-quality data support for subsequent prediction and analysis. Next, based on this structured data, a hybrid prediction engine is deployed, combining a long short-term memory network and reinforcement learning algorithms to construct a product trajectory prediction model. This model, by inputting historical location data and dynamic weight allocation data, generates trajectory prediction results, predicting the future location and trajectory of products. Subsequently, a space utilization analysis module is deployed, using 3D geometric algorithms to analyze the occupancy rate between shelf layers and the passage efficiency of aisles. Simultaneously, combined with occlusion level classification data, inefficient storage areas are identified, generating space analysis decision data to provide suggestions for warehouse space optimization. Finally, the trajectory prediction results and spatial analysis decision data are input into a multi-objective optimization solver. A genetic algorithm is used to balance factors such as the frequency of goods retrieval, shelf load-bearing limits, and aisle flow to generate dynamic location adjustment schemes and picking path planning strategies, thereby obtaining the final warehouse management decision data and improving the efficiency and accuracy of warehouse operations.

[0037] In one embodiment, the direction adaptive compensation table data is calculated as follows: in, Data representing the correlation patterns of signal attenuation. This indicates the fluctuation data of the tag signal strength over time. The value of , This represents the mean of the tag signal strength fluctuation data. This indicates that the simulation data of multipath effects is in time. The value of , This represents the mean of the multipath effect simulation data. This indicates the total amount of data.

[0038] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A warehouse visualization management method based on digital twins, characterized in that, include: Deploy AOA-RFID reader arrays on warehouse shelves and set RFID tags on the surface of goods; Initial positioning data of goods is collected using an AOA-RFID reader array; The initial positioning data is compensated for with tag signal strength to obtain tag signal strength compensated data; A visual-assisted correction system is deployed based on tag signal strength compensation data. The system uses a camera to identify the actual position of RFID tags on the shelf to obtain visual positioning data. The visual positioning data is then compared with the initial positioning data of the goods to obtain AOA positioning error data. Based on the AOA positioning error data, the initial positioning data of the goods is compensated for the positioning deviation to obtain the visually corrected positioning data. A dynamic weight allocation model is deployed based on the visually corrected positioning data, and real-time occlusion data of the warehouse is collected through environmental sensors. Based on real-time warehouse occlusion data, occlusion grading thresholds are constructed to obtain occlusion grading data; Based on the occlusion level classification data, the weight coefficients of AOA positioning data and visual positioning data are calculated to obtain dynamic weight allocation data; A digital twin warehouse model is constructed based on dynamic weight allocation data. The visually corrected positioning data and dynamic weight allocation data are then integrated into the digital twin warehouse model to obtain three-dimensional visualized warehouse data. The three-dimensional visualized warehouse data is used to predict the trajectory of goods and analyze the space utilization rate through a cloud platform to obtain warehouse management decision data. The warehouse management decision data is then sent to the warehouse management system to execute visualized warehouse management.

2. The warehouse visualization management method based on digital twins according to claim 1, characterized in that, Tag signal strength compensation is performed on the initial positioning data to obtain tag signal strength compensation data, including: Environmental metal reflection characteristics were analyzed on the initial positioning data to obtain metal reflection interference data; a multipath effect simulation model was constructed based on the metal reflection interference data to obtain multipath effect simulation data. A dynamic signal attenuation compensation strategy is deployed based on multipath effect simulation data. Tag signal strength fluctuation data is collected through an AOA-RFID reader array. The tag signal strength fluctuation data is correlated with the multipath effect simulation data to obtain signal attenuation correlation data. Based on the signal attenuation correlation data, a tag orientation adaptive compensation table is constructed to obtain the orientation adaptive compensation table data. A vision-assisted positioning system is deployed based on the orientation adaptive compensation table data to obtain vision correction compensation coefficient data; the signal strength in the initial positioning data is dynamically weighted according to the vision correction compensation coefficient data to obtain the signal strength data after vision assistance compensation. An environment-adaptive learning model is deployed based on visually assisted compensation signal strength data. Temperature data of the shelf area is collected by a temperature sensor. The temperature data is correlated with the visually assisted compensation signal strength data to obtain temperature-related reflection characteristic data. The direction-adaptive compensation table is dynamically updated based on the temperature-related reflection characteristic data to generate label signal strength compensation data.

3. The warehouse visualization management method based on digital twins according to claim 2, characterized in that, A vision-assisted correction system is deployed based on tag signal strength compensation data. This system uses cameras to identify the actual location of RFID tags on shelves, obtaining visual positioning data, including: Based on the tag signal strength compensation data, deploy an array of cameras to collect RGB-D depth images of the shelf panels, and obtain shelf panel depth image data; perform product region segmentation on the shelf panel depth image data to obtain product outline positioning data; Based on the cargo outline positioning data, a tag feature matching algorithm is deployed to extract the visual feature points of the RFID tags through the camera module; the visual feature points are associated and mapped with the tag ID in the tag signal strength compensation data to obtain tag feature association data; the camera module is automatically focused based on the tag feature association data to obtain visual focus optimization data. Based on visual focus optimization data, a 3D coordinate transformation model is deployed to convert the pixel coordinates in the shelf depth image data into physical coordinates. Combined with the AOA positioning results in the tag signal strength compensation data, multi-source data fusion is performed on the physical coordinates to obtain visual AOA fused positioning data. The visual AOA fused positioning data is then smoothed by Kalman filtering to obtain visually corrected positioning data. An environment-adaptive learning model is deployed based on the visually corrected positioning data. Light intensity data of the shelf area is collected by a light sensor. The real-time light intensity data and the visually corrected positioning data are correlated and analyzed to obtain the light correlation attenuation law data. The exposure parameters of the camera module are dynamically adjusted based on the light correlation attenuation law data to obtain visual positioning data.

4. The warehouse visualization management method based on digital twins according to claim 3, characterized in that, Error mapping analysis is performed between visual positioning data and initial product positioning data to obtain AOA positioning error data, including: Visual positioning data and initial product positioning data are input into the coordinate alignment module. The polar coordinate positioning result of AOA is converted into the Cartesian coordinate system of the vision system through a spatial coordinate transformation algorithm to obtain cross-modal coordinate alignment data. The cross-modal coordinate alignment data is then processed with timestamp synchronization to obtain spatiotemporal synchronized positioning data. Based on the deployment of spatiotemporal synchronous positioning data, an error calculation engine is used to calculate the deviation between the actual coordinates of the same RFID tag in visual positioning and the coordinates in AOA positioning through the Euclidean distance algorithm, thereby obtaining tag-level error distribution data. Based on the label-level error distribution data, an environmental feature association model is deployed, and a clustering algorithm is used to identify the association between high-error areas and shelf structures to obtain environmental association error pattern data. An error mapping learning network is deployed based on environmental correlation error pattern data. An error prediction model is constructed using the random forest algorithm. The predicted error value and the actual measurement error are iteratively trained to obtain AOA positioning error data.

5. The warehouse visualization management method based on digital twins according to claim 4, characterized in that, Based on the AOA positioning error data, the initial positioning data of the goods is compensated for the positioning deviation to obtain the visually corrected positioning data, including: An error compensation model is constructed based on AOA positioning error data and initial positioning data of goods. Environmental interference characteristic data are obtained by collecting data on metal reflection intensity and distribution of obstructions in the shelf area through environmental sensors. A dynamic weight allocation engine is deployed based on environmental interference feature data, and a fuzzy logic algorithm is used to calculate the credibility weights of AOA positioning data and visual positioning data. Dynamic weight allocation data is obtained by comparing environmental interference feature data with historical error pattern library in real time. The initial positioning data of goods is weighted and fused based on dynamic weight allocation data. The AOA positioning result is multiplied by the corresponding weight coefficient and then superimposed with the visual positioning data. The superimposed positioning result is smoothed by Kalman filtering to obtain the visually corrected positioning data. The visually corrected positioning data is fed back to the error compensation model, and the environmental interference feature vector and weight allocation strategy are updated through reinforcement learning algorithm. Combined with the real-time positioning accuracy evaluation results, the fuzzy logic rule base is dynamically optimized to obtain positioning deviation compensation data.

6. The warehouse visualization management method based on digital twins according to claim 5, characterized in that, Based on real-time warehouse occlusion data, occlusion severity grading thresholds are constructed to obtain occlusion severity grading data, including: The visually corrected positioning data and the raw occlusion data collected by the environmental sensors are input into the modal fusion module. The spatial correlation algorithm is used to identify the types of occlusions around the goods and obtain environmental occlusion feature data. A dynamic clustering engine is deployed based on statistical feature data of occlusion degree. The fuzzy C-means clustering algorithm is used to perform soft classification of occlusion degree to obtain a fuzzy membership matrix. Combined with warehouse shelf density and aisle width parameters, the fuzzy membership matrix is ​​adaptively corrected to obtain preliminary occlusion level classification data. Based on the initial classification of occlusion levels, a threshold verification module is deployed. The AOA visual fusion positioning accuracy corresponding to the occlusion level is extracted from the historical positioning error database. The fuzzy clustering center is adjusted using the error backpropagation mechanism to obtain optimized occlusion level classification threshold data.

7. A warehouse visualization management method based on digital twins according to claim 6, characterized in that, Based on the occlusion level classification data, weight coefficients are calculated for AOA positioning data and visual positioning data to obtain dynamic weight allocation data, including: A weight mapping table is constructed based on the occlusion level classification data, and the AOA and visual weight initial values ​​corresponding to the occlusion label are defined. The accuracy performance of the two positioning technologies under each occlusion level is extracted from the historical positioning error database to obtain the initial weight configuration data. Deploy a weight calculation engine based on the initial weight configuration data, and collect data on the current warehouse occlusion level through environmental sensors to obtain spatially adaptive weight data; A dynamic feedback adjustment module is deployed based on spatial adaptive weight data. AOA and visual positioning data are input into the error evaluation model to obtain the weight data after feedback optimization. The optimized weight data is synchronized to the positioning fusion module, and dynamic weight allocation data is obtained through a weighted average algorithm.

8. The warehouse visualization management method based on digital twins according to claim 7, characterized in that, A digital twin warehouse model is constructed based on dynamically weighted data. Visually corrected positioning data is then integrated with the dynamically weighted data into the digital twin warehouse model to obtain 3D visualized warehouse data, including: The basic framework of a digital twin warehouse is constructed based on dynamic weight allocation data. The virtualization priority of shelf and aisle elements is defined by weight data. Lightweight modeling technology is used to generate an interactive 3D warehouse skeleton, resulting in weight-driven basic model data. The visually corrected positioning data is mapped to a weight-driven base model, and the actual product location is bound to the virtual model coordinate system through a coordinate alignment algorithm to obtain hierarchical positioning mapping data. Deploy a real-time rendering engine based on hierarchical positioning mapping data, control the model update frequency through weight data, and generate rendering strategy optimization data. By integrating rendering strategy optimization data with visually corrected positioning data, a browser-side 3D visualization scene is constructed using WebGL technology; dynamic weight allocation data is embedded into scene element attributes to obtain 3D visualization repository data.

9. A warehouse visualization management method based on digital twins according to claim 8, characterized in that, By using a cloud platform to predict the trajectory of goods and analyze space utilization from 3D visualized warehouse data, warehouse management decision-making data is obtained, including: The structured analysis of 3D visualized warehouse data extracts the temporal data of goods location and shelf space coordinate information; the 3D coordinates are converted into tensor format recognizable by the trajectory prediction model through a spatiotemporal feature encoder to obtain structured warehouse spatiotemporal data. A hybrid prediction engine is deployed based on structured warehouse spatiotemporal data, and a cargo trajectory prediction model is constructed by combining long short-term memory networks and reinforcement learning algorithms; historical location data and dynamic weight allocation data are input to obtain trajectory prediction results. Based on the trajectory prediction results, a space utilization analysis module is deployed to calculate the occupancy rate between shelf layers and the passage efficiency of aisles using a three-dimensional geometric algorithm; combined with the occupancy level classification data, inefficient storage areas are identified to obtain space analysis decision data; The trajectory prediction results and spatial analysis decision data are input into a multi-objective optimization solver to balance the frequency of goods retrieval, shelf load-bearing limits, and aisle flow requirements. A genetic algorithm is used to generate dynamic location adjustment schemes and picking path planning strategies to obtain warehouse management decision data.

10. A warehouse visualization management method based on digital twins according to claim 2, characterized in that, The calculation of the direction adaptive compensation table data is as follows: in, Data representing the correlation patterns of signal attenuation. This represents the value of the tag signal strength fluctuation data at time t. This represents the mean of the tag signal strength fluctuation data. This represents the value of the multipath effect simulation data at time t. denoted by , where represents the mean of the multipath effect simulation data, and T represents the total amount of data.

Images