An inventory simulation display method and system of an intelligent turnover cabinet and a storage medium

Abstract

The application relates to the technical field of data processing, and discloses a kind of inventory simulation display method and system of intelligent turnover cabinet and storage medium.The method comprises the following steps: calculating storage three-dimensional coordinates by RFID signal strength data; identifying level for coordinate clustering and generating three-dimensional simulation model; updating model display state after collecting multi-sensor data verification; constructing storage atlas to predict access probability through graph convolution network and generate heat distribution; calculate overdue risk value according to storage duration and mark warning in model.The application solves the problems of lack of flexibility in static configuration and simulation display and actual state in the prior art.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to an inventory simulation display method, system and storage medium for an intelligent turnover cabinet. Background Technology

[0002] As an important asset management device in the power supply station's three-level warehouse, the intelligent turnover cabinet typically uses RFID, barcode scanning, and sensor technologies to achieve intelligent storage and inbound / outbound management of metering instruments such as electricity meters and transformers. The existing intelligent turnover cabinet inventory management system automatically identifies asset information by installing RFID readers and barcode scanning modules on the storage locations and displays the storage location occupancy status in the form of a list or two-dimensional plan on the management interface. Operators can then locate the corresponding position in the physical cabinet according to the storage location number prompted by the system to complete the retrieval and placement operations. This technical solution reduces the workload of manual registration to a certain extent and lowers the error rate of asset management.

[0003] However, existing inventory simulation display methods have significant shortcomings. First, in terms of 3D spatial model construction, existing systems require administrators to manually configure cabinet structure parameters during the initialization phase, including the number of shelves, the number of storage locations per layer, storage location dimensions, and storage location spacing. When the cabinet layout is adjusted or different types of shelves are replaced, all configuration parameters must be manually re-entered. This static configuration method lacks flexibility and has high maintenance costs. Second, in terms of inventory status synchronization, existing systems mainly rely on a single sensor data source. When the sensor fails or the RFID signal is interfered with, the simulated inventory status is prone to inconsistency with the actual physical status, resulting in a disconnect between the digital twin and the physical goods. Third, in terms of visualization display dimensions, existing systems mostly use simple two-dimensional lists or planar layout diagrams, lacking three-dimensional spatial display, making it difficult to intuitively reflect the spatial relationship of multiple cabinets, especially in multi-cabinet splicing scenarios where the overall inventory distribution is not readily apparent.

[0004] The fundamental problem with static configuration is that the system cannot automatically detect changes in the physical structure of the cabinets, requiring manual intervention and remodeling for each layout adjustment. Essentially, this lacks an automatic mechanism for acquiring storage space locations. The problem of asynchronous inventory status stems from the insufficient reliability of data from a single sensor. When the photoelectric sensor shows that there is stock but the RFID tag is not read, the system cannot determine whether it is a sensor malfunction or a damaged tag. The lack of a multimodal data cross-validation mechanism makes it difficult to detect anomalies in a timely manner. The problem of a single dimension of visualization further extends to the deeper contradiction of insufficient intelligence. The existing system only passively displays the current inventory status and cannot predict which storage locations are more likely to be accessed in the future based on historical business data. It cannot intelligently recommend the best entry location and optimize storage and retrieval paths, nor can it proactively warn of assets that are about to expire or mature. This results in warehouse management still relying on manual experience and judgment, failing to fully leverage the value of data to assist decision-making. Summary of the Invention

[0005] This application provides an inventory simulation display method, system, and storage medium for intelligent turnover cabinets. It solves the problems of lack of flexibility in static configuration and asynchrony between simulation display and actual state in the prior art by using an adaptive three-dimensional spatial modeling method based on RFID signal strength distribution and a multi-modal sensor data fusion verification mechanism. It also improves the intelligence level of inventory management and asset turnover efficiency by using graph convolutional neural network-driven storage location access prediction and time-series prediction-driven proactive overdue early warning.

[0006] Firstly, this application provides a method for simulating and displaying the inventory of an intelligent turnover cabinet, the method comprising: Step S1: Collect the signal strength data of the RFID self-test tags of each storage location in the turnover cabinet under multiple antennas, and calculate the three-dimensional coordinate set of the storage location based on the distance attenuation relationship between the signal strength data and the antenna position. Step S2: Cluster the vertical coordinates of the three-dimensional coordinate set of the storage sites to identify the hierarchical grouping results. Based on the number and spacing characteristics of storage sites in each layer of the hierarchical grouping results, generate a three-dimensional simulation model with storage site type identifier. Step S3: Collect photoelectric signals, barcode data and RFID asset tag information, perform consistency verification on the photoelectric signals, barcode data and RFID asset tag information, filter to obtain valid storage location status data, and update the display status of the storage location in the three-dimensional simulation model according to the storage location status data; Step S4: Construct a reservoir relationship map, extract the current state and historical access records of the reservoir as map node features, calculate the map node features through a graph convolutional network to obtain the reservoir access prediction results, and generate a visualized thermal distribution in the three-dimensional simulation model based on the access prediction results. Step S5: Obtain the asset storage time and remaining time until expiration for the occupied storage space. Calculate the overdue risk value based on the comparison between the storage time and historical periods. Mark the storage space with the overdue risk value exceeding the set value as a warning state in the three-dimensional simulation model.

[0007] Secondly, this application provides an inventory simulation display system for intelligent turnover cabinets, the intelligent turnover cabinet inventory simulation display system comprising: The acquisition module is used to collect signal strength data of RFID self-inspection tags in each storage location in the turnover cabinet under multiple antennas. Based on the distance attenuation relationship between the signal strength data and the antenna position, the three-dimensional coordinate set of the storage location is calculated. The clustering module is used to cluster the vertical coordinates of the three-dimensional coordinate set of the storage sites, identify the hierarchical grouping results, and generate a three-dimensional simulation model with storage site type identifier based on the number and spacing characteristics of each layer of the hierarchical grouping results. The verification module is used to collect photoelectric signals, barcode data and RFID asset tag information, perform consistency verification on the photoelectric signals, barcode data and RFID asset tag information, filter to obtain valid storage location status data, and update the display status of the storage location in the three-dimensional simulation model according to the storage location status data. The calculation module is used to construct a reservoir relationship map, extract the current state and historical access records of the reservoirs as map node features, calculate the map node features through a graph convolutional network to obtain the reservoir access prediction results, and generate a visualized thermal distribution in the three-dimensional simulation model based on the access prediction results. The marking module is used to obtain the asset storage time and remaining time until expiration of occupied storage spaces, calculate the overdue risk value based on the comparison between the storage time and historical periods, and mark the storage spaces with the overdue risk value exceeding the set value as warning status in the three-dimensional simulation model.

[0008] Thirdly, an inventory simulation display device for an intelligent turnover cabinet is provided, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor invokes the instructions in the memory to cause the intelligent turnover cabinet inventory simulation display device to execute the aforementioned intelligent turnover cabinet inventory simulation display method.

[0009] Fourthly, a computer-readable storage medium is provided, wherein instructions are stored in the computer-readable storage medium, which, when executed on a computer, cause the computer to perform the above-described intelligent turnover cabinet inventory simulation display method.

[0010] The technical solution provided in this application utilizes an automatic method for calculating the three-dimensional coordinates of storage locations based on RFID signal strength distribution. This method leverages the signal strength data collected by multiple antennas at different transmission powers and the distance attenuation relationship between the antenna positions. By solving an overdetermined set of equations using the least squares method, the precise three-dimensional spatial coordinates of each storage location are obtained. This method automatically constructs a storage location database without requiring manual measurement and parameter configuration, solving the problem of inflexibility in existing static configuration methods. When the shelf layout of the turnover cabinet is adjusted or the storage location configuration is changed, the three-dimensional simulation model is automatically updated simply by re-executing the signal acquisition and coordinate calculation process, avoiding the tedious operation of manually re-entering configuration parameters. In practical applications of power metering asset management, power supply station warehouses frequently need to adjust single-phase surface shelves according to the proportion of asset types at different times. Regarding the configuration of the three-phase surface plate and the transformer plate, the adaptive modeling mechanism of this invention enables dynamic adjustments without the need for professional technicians. Ordinary warehouse managers can complete the re-identification and model reconstruction of the cabinet structure. Furthermore, this invention automatically identifies hierarchical groups by performing K-means clustering on the vertical coordinates of the storage location's three-dimensional coordinate set. Based on the number of storage locations and horizontal spacing characteristics of each level, it automatically determines the storage location type identifier by matching with preset rules. Combined with the coordinate range calculation, it obtains the cabinet's outer envelope dimension parameters and calls the graphics rendering engine to automatically generate a complete three-dimensional simulation model containing the shell, shelves, and storage units. This method automates the entire process from signal acquisition to model generation, completely eliminating the constraints of manual configuration and realizing intelligent mapping from physical space to virtual model, ensuring the continuous consistency between the simulation display and the actual cabinet structure.

[0011] This invention employs a real-time status verification mechanism based on multimodal sensor data fusion. It simultaneously collects data from three independent data sources: photoelectric signals, barcode data, and RFID asset tag information. Through consistency verification, it filters storage location records where at least two types of sensor data match as valid data. When a sensor data conflict is detected, the corresponding storage location in the 3D simulation model is set to a flashing red material as an anomaly alarm. This cross-validation mechanism significantly improves the synchronization accuracy between simulation display and physical status, effectively identifying anomalies such as sensor malfunctions, RFID tag damage, or operational errors. It avoids misjudgments caused by single sensor failures. This mechanism is crucial for the actual operation of the intelligent turnover cabinet. In this invention, photoelectric sensors may be falsely triggered due to dust obstruction, barcode scanning modules may fail to scan due to improper meter placement angles, and RFID readers may miss tags due to inadequate shielding of metal cabinets. This invention effectively eliminates these interference factors through a multi-data source complementary verification mechanism. Furthermore, this invention constructs a storage location relationship graph to uniformly model spatial adjacency relationships and historical business association relationships, extracting the current state and historical access records of storage locations as graph node features. Multi-layer convolutional operations are performed on the node features through a graph convolutional network. The first convolutional layer aggregates spatial neighborhood features to learn the physical location relationships of storage locations, and the second and third convolutional layers further... This method integrates business-related features to learn which storage locations are frequently accessed simultaneously, outputting the access probability value of each storage location within a future time window. Based on the access probability, a visual heat map distribution from blue to red is generated in a 3D simulation model. Storage locations with high access probabilities and currently empty are selected as recommended storage locations, overlaid with a green halo effect to achieve intelligent storage guidance. This method fully considers the first-in, first-out principle and business patterns in power metering asset management. The graph convolutional network learns the co-occurrence patterns of single-phase meters and current transformers frequently used simultaneously in historical outbound work orders, automatically mapping these business relationships to the storage location recommendation strategy, ensuring that related assets are stored spatially adjacent or business-related. The storage locations significantly shorten the retrieval path and operation time. In addition, the present invention adopts a proactive overdue early warning mechanism driven by time-series prediction. It calculates the probability value of overdue risk based on the deviation of the asset's storage time from the historical average inventory cycle, and makes a dual judgment based on the remaining time before the inspection expires. Storage locations that meet the early warning conditions are marked with orange or red warning materials in the three-dimensional simulation model. This proactive early warning method changes the passive management mode of the existing technology, which requires manual periodic queries to discover overdue assets. Administrators can intuitively identify which assets need to be prioritized by observing the three-dimensional simulation model, avoiding the repeated inspection costs and business risks caused by assets exceeding the inspection validity period. Attached Figure Description

[0012] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a schematic diagram of one embodiment of the inventory simulation display method for intelligent turnover cabinets in this application. Figure 2 This is a schematic diagram of the storage location access probability distribution in an embodiment of this application; Figure 3 This is a schematic diagram of one embodiment of the intelligent turnover cabinet inventory simulation display system in this application. Figure 4 This is a schematic block diagram of the inventory simulation display device of the intelligent turnover cabinet in this embodiment of the invention. Detailed Implementation

[0014] This application provides a method, system, and storage medium for simulating inventory display in an intelligent turnover cabinet. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0015] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of the inventory simulation display method for intelligent turnover cabinets in this application includes: Step S1: Collect the signal strength data of the RFID self-inspection tags of each storage location in the turnover cabinet under multiple antennas. Based on the distance attenuation relationship between the signal strength data and the antenna position, calculate and obtain the three-dimensional coordinate set of the storage location. Step S2: Cluster the vertical coordinates of the three-dimensional coordinate set of the storage sites to identify the hierarchical grouping results. Based on the number and spacing characteristics of storage sites in each layer of the hierarchical grouping results, generate a three-dimensional simulation model with storage site type identifier. Step S3: Collect photoelectric signals, barcode data and RFID asset tag information, perform consistency verification on the photoelectric signals, barcode data and RFID asset tag information, filter to obtain valid storage location status data, and update the display status of the storage location in the three-dimensional simulation model according to the storage location status data; Step S4: Construct a reservoir relationship map, extract the current state and historical access records of the reservoir as map node features, calculate the map node features through a graph convolutional network to obtain the reservoir access prediction results, and generate a visualized thermal distribution in the three-dimensional simulation model based on the access prediction results. Step S5: Obtain the asset storage duration and remaining time until expiration for the occupied storage space. Calculate the overdue risk value based on the comparison between the storage duration and historical cycles. Mark the storage space with the overdue risk value exceeding the set value as a warning state in the three-dimensional simulation model.

[0016] It is understood that the executing entity of this application can be an inventory simulation display system for intelligent turnover cabinets, or it can be a terminal or a server; the specific implementation is not limited here. This application's embodiment uses a server as an example for illustration.

[0017] Specifically, the RF read / write module is controlled by the control board to send RF signals to the self-test tags in each storage location in an incremental power manner. Six circularly polarized antennas are arranged in a 2x3 array on the back panel of the cabinet, with a spacing of 300 mm x 250 mm. Each antenna collects signal strength values ​​in 1.5 dB increments as the transmit power increases from 0 dB to 31.5 dB. Based on the attenuation model that the signal strength is inversely proportional to the square of the distance when the signal propagates in free space, a corresponding equation for the relationship between signal strength and distance is established. The path loss index is taken as 2.8 to consider the multipath reflection of the indoor metal cabinet, and the path loss at a reference distance of 1 meter is taken as 40. By using 6 antennas at 21 power levels to collect 126 sets of data, the three-dimensional spatial coordinates of the storage site tags were obtained by solving the overdetermined equations using the least squares method. The signal strength received by a certain storage site tag by antenna 1 was -45 dB, antenna 2 was -38 dB, and antenna 3 was -52 dB. Substituting these values ​​into the equations, the distances between the tag and each antenna were calculated to be 1.2 m, 0.8 m, and 1.6 m, respectively. Combining the known spatial coordinates of the antennas, the coordinates of the storage site were calculated to be 285 mm x 450 mm y and 680 mm height. A database mapping the storage site codes to spatial coordinates was constructed by summarizing 35 storage site coordinates.

[0018] The vertical height coordinate components are extracted from the 3D coordinate set of the storage locations. A mean clustering operation is performed with a cluster size of 5. The spatial distance from each point to the cluster center is iteratively calculated and the locations are redistributed until the center position changes by less than 5 millimeters. The vertical coordinates of 35 storage locations in a cabinet are clustered into 5 layers: Layer 1 (450 mm high, 6 locations), Layer 2 (650 mm high, 9 locations), Layer 3 (850 mm high, 9 locations), Layer 4 (1050 mm high, 8 locations), and Layer 5 (1250 mm high, 3 locations). After counting the 9 locations in Layer 2, the horizontal coordinate differences between adjacent locations are calculated to be 140 mm, 138 mm, 142 mm, and 141 mm, with an average spacing of approximately 140 mm and 9 locations. Based on preset rules... The layer was determined to be a single-phase surface plate. The average spacing of the 8 storage positions on the 4th layer was approximately 200 mm, which was determined to be the transformer layer. The maximum and minimum values ​​of the coordinates of all storage positions in each coordinate axis direction were calculated. The maximum value in the horizontal direction was 1550 mm and the minimum value was 50 mm. The maximum value in the vertical direction was 650 mm and the minimum value in the vertical direction was 1450 mm and the minimum value was 250 mm. With an edge allowance of 300 mm, the outer envelope dimensions of the cabinet were calculated to be 1650 mm long, 650 mm wide, and 1884 mm high. The graphics rendering engine was used to generate a gray-white cabinet shell model based on the envelope dimensions. The layer models were generated based on the height of each layer and the storage position type identification. The single-phase surface plate was rendered as light blue and the transformer layer was rendered as light yellow. 35 storage position unit models were generated based on the storage position coordinates. Initially, they were semi-transparent cubes. These were combined to form a complete three-dimensional simulation model.

[0019] The output level signals of the slot-type photoelectric switches at each storage location are collected as photoelectric signals. When the infrared beam is blocked, a low level is output to indicate that the goods are available. The scanning engine of the barcode scanning module continuously scans for 3000 milliseconds within a field of view of 30 degrees horizontally and 37 degrees vertically, and a depth of field of 4 to 280 mm to obtain a 22-bit asset barcode string. The RFID reader / writer module polls 50 channels in the 860 to 960 MHz band using a frequency-hopping method, staying on each channel for 200 milliseconds to read the electronic product code of the asset tag. When the photoelectric signal of a certain storage location shows a low level, it indicates that the goods are available. The barcode scan obtains the string "4230001001000069817550". The RFID tag is successfully read to obtain the encoded value. The consistency of the data status of the three sensors is marked as... Valid records were found. Another storage location's photoelectric signal was low, but the barcode scan timed out and no data was acquired, and the RFID tag was not read either. This was determined to be an abnormal record due to inconsistencies in the data states of at least two sensors. The storage location code and asset occupancy identifier were extracted from the valid storage location status data. The material transparency channel value of the corresponding storage location unit model in the 3D simulation model was set to 0.3 for storage locations with an empty occupancy identifier, rendering them as semi-transparent yellow. The transparency channel value of storage locations with an occupied identifier was set to 1.0, rendering them as opaque gray. Storage locations with sensor data conflicts were detected, and their storage location unit model color red, green, and blue values ​​were set to 255, 0, and 0, respectively, with a flashing period of 500 milliseconds. This completed the update of the storage location display status in the 3D simulation model.

[0020] The 35 storage locations of the turnover cabinet are abstracted as graph nodes. The spatial Euclidean distance between any two storage locations is calculated. If the distance is less than 350 mm, a spatial adjacency edge is established with a weight equal to the negative distance divided by 100 using the natural exponential function. The number of times two storage locations are simultaneously picked up in historical outbound work orders is queried. If the co-occurrence exceeds 10 times, a business association edge is established with a weight equal to the co-occurrence divided by the total number of work orders. Storage location 1 and storage location 2 are 280 mm apart, and a spatial adjacency edge is established with a weight of approximately 0.061 using the natural exponential function -2.8. Although storage location 1 and storage location 18 are relatively far apart, they have been simultaneously picked up 25 times in history, with a total of 100 work orders, and a business association edge is established with a weight of 0.25. A complete storage location relationship graph is constructed, and the current occupancy status of each storage location is extracted and coded as 0 or... 1. The frequency of outbound shipments and inbound shipments in the past 30 days, the remaining days of asset verification, and the unique thermal codes for storage location types (single-phase meters: 100, three-phase meters: 010, current transformers: 001) are normalized to the range of 0 to 1. These codes are combined to form a 12-dimensional node feature vector. The feature vectors of 35 nodes are arranged in rows to construct a 35-row, 12-column node feature matrix. Each diagonal position of the adjacency matrix of the storage location relationship graph is incremented by 1 to achieve self-connection. The degree value of each node is calculated as the sum of the elements in each row of the adjacency matrix. A node connected to 5 neighbors and self-connected has a degree value of 6. The diagonal elements of the degree matrix are the degree values ​​of each node. The square root of each diagonal element of the degree matrix is ​​calculated, and the reciprocal is taken. The degree matrix is ​​then multiplied by the adjacency matrix on the left and then by the degree matrix on the right. The normalized adjacency matrix is ​​obtained by multiplying the node feature matrix and the normalized adjacency matrix to obtain a 35x12 intermediate matrix. This intermediate matrix is ​​then multiplied by the first convolutional layer weight matrix (12x32) to obtain a 35x32 matrix. The first intermediate feature matrix contains storage space neighborhood features by setting negative values ​​to zero and retaining positive values ​​through a modified linear unit activation function. The first intermediate feature matrix repeats the above convolution process, multiplying it by the normalized adjacency matrix and then by the second convolutional layer weight matrix (32x64). After activation, it becomes a 35x64 second intermediate feature matrix containing storage space business-related features. Finally, the target feature matrix (35x16) is obtained by multiplying it by the third convolutional layer weight matrix (64x16) to fuse spatial and business features. The system extracts 16-dimensional feature vectors from each row of the target feature matrix. For each feature vector, exponential normalization is performed (the natural exponent of each element divided by the sum of all natural exponents). The normalized value is used as the predicted probability of the storage location being accessed within the next 7 days. For example, a storage location has a normalized access probability of 0.82. Based on the access probability values ​​of each storage location, the color of the corresponding storage location unit in the 3D simulation model is set to a gradient from blue to red. For a storage location with an access probability of 0.82, the red component is set to 255 multiplied by 0.82 (approximately 209), the green component to 0, and the blue component to 255 multiplied by 0.18 (approximately 46), rendering it as orange-red. Storage locations with an access probability of 0.15 are rendered as blue. Storage locations with access probabilities exceeding 0 are filtered out.8. Storage locations that are currently empty are designated as recommended storage locations. A green halo effect is overlaid on these recommended storage locations in the 3D simulation model. The halo thickness is 5 mm, and the luminous intensity dynamically changes according to a sine function with a period of 1000 milliseconds, thus completing the intelligent storage guidance.

[0021] The system extracts the entry timestamp and verification expiration timestamp of assets in occupied storage locations from the asset archive database. For example, an asset's entry timestamp is 1699920000, corresponding to November 14, 2023. The current timestamp is 1731456000, corresponding to November 13, 2024. The time difference is 31,536,000 seconds, which translates to 365 days, indicating an asset entry duration of 365 days. The verification expiration timestamp is 1762992000, corresponding to November 13, 2025. The time difference between this and the current timestamp is also 31,536,000 seconds, resulting in a remaining expiration duration of 365 days. A total of 50 historical outbound records for this asset type are retrieved. The actual inventory periods for each record are 180 days, 195 days, 210 days, etc. The average inventory period is calculated as the sum of the periods of the 50 records divided by 50, yielding 2. 00 days. The standard deviation of the inventory cycle is calculated by dividing the sum of the squares of the differences between each recorded cycle and the average value by 50 and then taking the square root, which gives 30 days. The difference between the asset's storage duration of 365 days and the average inventory cycle of 200 days (165 days) is divided by the standard deviation of 30 days, which gives 5.5. Substituting this value into the cumulative probability function of the standard normal distribution, the overdue risk probability value of the asset is calculated to be approximately 0.999, indicating that it is seriously overdue. It is determined that the overdue risk probability value of 0.999 exceeds the preset risk threshold of 0.8. The corresponding storage unit model in the 3D simulation model is set to a red warning material with red, green, and blue values ​​of 255, 0, and 0. Another asset has only 5 days left until its expiration, which is less than the preset number of days threshold of 7 days. Similarly, its storage unit model is set to an orange warning material with red, green, and blue values ​​of 255, 140, and 0, thus completing the warning status marking.

[0022] Figure 2 This is a schematic diagram of the storage location access probability distribution in an embodiment of this application. Figure 2This diagram illustrates the probability distribution of 35 storage locations being accessed within a future time window, calculated using a graph convolutional network in this embodiment of the application. The horizontal axis represents storage location numbers from 1 to 35, and the vertical axis represents access probability values ​​from 0 to 1. The fluctuations in the line reflect the differences in the likelihood of different storage locations being accessed. Storage locations with higher access probabilities, such as storage locations 1, 5, 10, and 14, have probabilities exceeding 0.8, indicating a high probability of access within the next 7 days. Storage locations with lower access probabilities, such as storage locations 2, 11, and 22, have probabilities below 0.3, indicating a high probability of access within the next 7 days. Since the storage locations are less likely to be accessed, the graph convolutional network learns the spatial adjacency relationships and historical business association relationships of the storage locations. It combines node features such as the current occupancy status, outbound frequency, inbound frequency, and remaining days for verification of the storage locations. Through multi-layer convolutional operations, it extracts spatial neighborhood features and business association features, thereby accurately predicting the access probability of each storage location. This probability distribution is directly used to generate the visualized heat distribution in the 3D simulation model. Storage locations with high access probabilities are rendered in red in the model, while storage locations with low access probabilities are rendered in blue. At the same time, storage locations with access probabilities exceeding 0.8 and currently empty are selected as recommended inbound locations.

[0023] In one specific embodiment, step S1 includes: The RFID reading and writing module sends radio frequency signals to the self-testing tags in each storage location, receives the return signals from each self-testing tag through multiple antennas, and records the signal strength values ​​received by each antenna at different transmission powers. The signal strength value is associated with the corresponding transmit power and antenna number. Based on the signal propagation attenuation model in free space, a set of equations corresponding to the signal strength and the distance from the tag to the antenna are established. The preset spatial coordinates of each antenna are extracted as known parameters. The signal strength values ​​are substituted into the corresponding equations and the three-dimensional spatial coordinates of each storage tag are obtained by solving the least squares method. The three-dimensional spatial coordinates of each storage location tag are collected to construct a database of the mapping relationship between storage location codes and spatial coordinates, thereby obtaining a set of three-dimensional coordinates of the storage locations.

[0024] Specifically, the control board sends commands to the RF read / write module via a 1.25mm 15-pin connector. The RF read / write module operates in the 902-928 MHz frequency band using frequency hopping spread spectrum. The transmit power increases by 1.5 dB at a time, up to 31.5 dB, for a total of 21 power levels. At each power level, the six antennas sequentially receive signals for 100 milliseconds, recording the received signal strength values ​​returned by the self-test tag. The electronic product coding area of ​​the self-test tag stores a combination code of the cabinet's physical number and the storage location number. The self-test tag code for drawer number 1 is cabinet number ZJ0001 plus storage location number 01, forming ZJ000101. The first antenna receives a series of signal strength values ​​for this tag at different power levels. The second antenna receives another set of signal strength values ​​at the same power sequence. The remaining four antennas also record their respective received strength values ​​at each of the 21 power levels. The data matrix is ​​formed by collecting 126 sets of signal strength data from 6 antennas at 21 power levels for a single storage tag. Each set of signal strength values ​​is associated with the corresponding transmit power value and antenna number as a triplet. The signal strength received by antenna 1 at a certain power level is associated as a triplet containing three data items: antenna number, transmit power, and receive strength. Based on the attenuation model that signals follow when propagating in free space, an equation is established to correspond to the distance from the tag to the antenna. The path loss index is taken as 2.8 in the indoor metal cabinet environment to consider the influence of multipath reflection and metal surface on electromagnetic waves. The path loss at a reference distance of 1 meter is taken as 40 dB. Substituting each set of signal strength data into the attenuation model equation, a preliminary estimate of the distance between the tag and each antenna is obtained. Since the distance between the same tag and different antennas is a fixed physical quantity, an overdetermined set of equations is constructed using 126 sets of measurement data for optimization and solution.

[0025] Extract the preset spatial coordinates of 6 antennas as known parameters. The coordinates of antenna 1 are: x-coordinate 0 mm, y-coordinate 0 mm, height 1500 mm; antenna 2 is: x-coordinate 300 mm, y-coordinate 0 mm, height 1500 mm; antenna 3 is: x-coordinate 600 mm, y-coordinate 0 mm, height 1500 mm; antenna 4 is: x-coordinate 0 mm, y-coordinate 250 mm, height 1500 mm; antenna 5 is: x-coordinate 300 mm, y-coordinate 250 mm, height 1500 mm; antenna 6 is: x-coordinate 600 mm, y-coordinate 250 mm, height 1500 mm. Let the unknown coordinates of a storage location tag be x-coordinate X, y-coordinate Y, height Z. The distance between the tag and antenna 1 is calculated as the square root of x-coordinate X - 0 squared, y-coordinate Y - 0 squared, height Z - 1500 squared. The distance between the tag and antenna 2 is calculated as x-coordinate X - 300 squared, y-coordinate Y - 1500 squared. The distance equations between the tag and the six antennas are established by taking the square root of the sum of the squares of 0 and the height Z minus the square of 1500. The distance estimates obtained by converting 126 sets of signal strength values ​​through an attenuation model are used as the observed values. The coordinates of the six antennas are used as known quantities. An overdetermined system of equations is constructed with the tag coordinates X, Y, and Z as unknowns. The least squares method is used to solve this system of equations, that is, to find the tag coordinate value that minimizes the sum of the squared residuals between all observed distances and theoretical distances. During the calculation process, the sum of the squared residuals of all 126 sets of data is used to obtain the objective function. The partial derivatives of the objective function with respect to X, Y, and Z are taken respectively and set to zero to obtain a linear system of three normal equations. The three-dimensional spatial coordinates of the tag are obtained by solving this linear system of equations through matrix operations. The least squares method is used to solve the tag at a certain storage location, which yields an x-coordinate of 285 mm, a y-coordinate of 450 mm, and a height of 680 mm. The root mean square error of the solution is controlled within 15 mm, indicating that the spatial positioning accuracy meets the requirements.

[0026] The database summarizes the three-dimensional spatial coordinates of all 35 storage location tags. Each storage location tag corresponds to a set of three-dimensional values ​​for the x-coordinate, y-coordinate, and height. A key-value pair storage structure is used to construct the mapping relationship between storage location codes and spatial coordinates. Storage location code ZJ000101 is used as the key, and its corresponding value is a coordinate triplet containing an x-coordinate of 285 mm, a y-coordinate of 450 mm, and a height of 680 mm. Storage location code ZJ000102 is used as the key, and its corresponding value is another coordinate triplet. The mapping relationship between the codes and coordinates of all 35 storage locations is stored sequentially. The database uses JSON format serialization. Each record includes a storage location identifier field, an x-coordinate field, a y-coordinate field, a height field, and a signal field. The fingerprint field stores the signal strength vector of the storage location tag under the 6 antennas as the tag's identity feature data. The database file is stored in the local flash memory of the control board, occupying no more than 512 kilobytes of space. It obtains a set of three-dimensional coordinates of storage locations containing the spatial location information of 35 storage locations. This coordinate set records the precise spatial location of each storage location in the turnover cabinet, which solves the problem of manually configuring storage location parameters in the prior art. When the cabinet layout is adjusted or the shelves are replaced, there is no need to manually input the storage location information again. It is only necessary to re-execute the radio frequency signal acquisition and coordinate calculation process to automatically update the storage location database, realizing the adaptive construction of the storage space model.

[0027] In one specific embodiment, step S2 includes: Extract the vertical coordinate components of each storage site's coordinates from the three-dimensional coordinate set of the storage sites, perform K-means clustering on the vertical coordinate components, and obtain the hierarchical identifier of each storage site. The number of storage locations corresponding to each level of identifier is counted, the horizontal coordinate spacing between adjacent storage locations within the same level is calculated, and the storage location type identifier for each level is determined based on the matching relationship between the number of storage locations and the coordinate spacing and the preset storage location type rules. Calculate the maximum and minimum values ​​of all storage location coordinates in the three-dimensional coordinate set of the storage location in each coordinate axis direction, and obtain the cabinet outer envelope dimension parameters by superimposing the preset edge allowance based on the maximum and minimum values; The graphics rendering engine is invoked to generate a cabinet shell model based on the cabinet's outer shape envelope size parameters. Shelf models are generated based on the vertical coordinates of each level and the storage location type identifier. Storage unit models are generated based on the storage location coordinates in the storage location 3D coordinate set. The cabinet shell model, shelf model, and storage unit model are combined to obtain a 3D simulation model.

[0028] Specifically, 35 storage site coordinate data were read one by one from the three-dimensional coordinate set of the storage sites. Each storage site coordinate contains three components: horizontal coordinate, vertical coordinate, and height. The height component, i.e., the vertical coordinate component, was extracted to form a one-dimensional array. The height of the first storage site was 680 mm, the second storage site was 678 mm, and the third storage site was 682 mm. All 35 height values ​​were extracted sequentially. K-means clustering was performed on this height array to group storage sites with similar heights into the same level. The number of clusters, K, was set to 5, indicating that the storage sites were divided into 5 levels. The height values ​​of the 5 cluster centers were initialized with an equally spaced distribution. The minimum and maximum values ​​of all height values ​​were calculated. The minimum value was 450 mm. The lowest storage level is represented by millimeters, and the highest storage level is represented by a maximum value of 1250 millimeters. The interval between the minimum and maximum values ​​is divided into 5 segments, with initial cluster center heights of 450 mm, 650 mm, 850 mm, 1050 mm, and 1250 mm. Iterative calculations are then performed to determine the absolute value of the height difference between each storage level and the 5 cluster centers. The distance from the first storage level (680 mm) to the first center (450 mm) is 230 mm, to the second center (650 mm) is 30 mm, to the third center (850 mm) is 170 mm, to the fourth center (1050 mm) is 370 mm, and to the fifth center... A distance of 1250 mm is 570 mm. The cluster center with the smallest distance is selected as the location of the storage site; that is, the first storage site belongs to the second cluster center. This process is repeated for all 35 storage sites. After one iteration, the center position of each cluster is recalculated, which is the average height value of all storage sites within that cluster. The second cluster contains nine storage sites with heights of 680 mm, 678 mm, 682 mm, 675 mm, 685 mm, 672 mm, 688 mm, 670 mm, and 690 mm. The sum of these nine height values ​​is 6120 mm. Dividing this by 9 gives a new cluster center of 680 mm. After updating the five cluster centers, the distance is calculated again. The calculation and assignment process is iterated repeatedly until the change in the position of the cluster center between two adjacent iterations is less than 5 mm. Finally, the first cluster center is stable at 450 mm containing 6 storage sites, the second cluster center is stable at 650 mm containing 9 storage sites, the third cluster center is stable at 850 mm containing 9 storage sites, the fourth cluster center is stable at 1050 mm containing 8 storage sites, and the fifth cluster center is stable at 1250 mm containing 3 storage sites. A hierarchical identifier is assigned to each storage site. The 6 storage sites belonging to the first cluster are identified as the first level, the 9 storage sites belonging to the second cluster are identified as the second level, and so on. The hierarchical identifier of each storage site is recorded in the storage site attribute data.

[0029] The number of storage sites corresponding to each level was counted. Level 1 contains 6 storage sites, Level 2 contains 9 storage sites, Level 3 contains 9 storage sites, Level 4 contains 8 storage sites, and Level 5 contains 3 storage sites. For each level, the horizontal coordinate spacing between adjacent storage sites within the same level was calculated. The 9 storage sites in Level 2 were extracted and sorted by their horizontal coordinates from smallest to largest: Level 1 has an horizontal coordinate of 50 mm, Level 2 has an horizontal coordinate of 190 mm, Level 3 has an horizontal coordinate of 330 mm, Level 4 has an horizontal coordinate of 470 mm, Level 5 has an horizontal coordinate of 610 mm, and so on. The x-coordinates of the 6th and 9th reservoirs are 750 mm, 890 mm, 1030 mm, and 1170 mm respectively. The differences in x-coordinates between adjacent reservoirs are calculated. The distance between the 1st and 2nd reservoirs is 190 - 50 = 140 mm; the distance between the 2nd and 3rd reservoirs is 330 - 190 = 140 mm; and the distance between the 3rd and 4th reservoirs is 470 - 330 = 140 mm. The resulting eight distance values ​​are 140 mm, 140 mm, 140 mm, 140 mm, and 140 mm respectively. The values ​​are 140 mm, 140 mm, 140 mm, and 140 mm. The average of these eight spacing values ​​is 1120 mm divided by 8, which equals 140 mm. Matching is then performed according to preset storage location type rules. The rules stipulate that if a storage level has 6 locations with an average spacing between 135 and 145 mm, it is classified as a single-phase surface plate; if a storage level has 9 locations with an average spacing between 135 and 145 mm, it is classified as a single-phase surface plate; and if a storage level has 3 locations with an average spacing between 189 and 199 mm, it is classified as a three-phase surface plate. For the surface layer, if a certain level has 4 storage locations and the average spacing is between 195 and 205 mm, it is determined to be a current transformer layer. The second level has 9 storage locations with an average spacing of 140 mm, which matches the second rule and determines that the storage location type of this level is a single-phase surface layer. The fourth level has 8 storage locations, including 4 storage locations on each side. The average spacing is calculated using the same method and is 200 mm, which matches the fourth rule and determines that it is a current transformer layer. The fifth level has 3 storage locations with an average spacing of 194 mm, which matches the third rule and determines that it is a three-phase surface layer. This completes the determination of the storage location type for each level.

[0030] Calculate the maximum and minimum values ​​of all 35 storage site coordinates in the three-dimensional coordinate set along each coordinate axis. Iterate through all storage sites, finding the maximum x-coordinate as 1550 mm and the minimum x-coordinate as 50 mm. Iterate through all storage sites, finding the maximum y-coordinate as 600 mm and the minimum y-coordinate as 50 mm. Iterate through all storage sites, finding the maximum height as 1450 mm and the minimum height as 250 mm. Calculate the span along each coordinate axis based on these values: the lateral span is the maximum x-coordinate minus the minimum x-coordinate (1550 - 50 = 1500 mm); the longitudinal span is the maximum y-coordinate minus the minimum y-coordinate (600 - 50 = 550 mm); and the vertical span is the maximum height minus the minimum height (1450 - 50 = 250 mm). 250 equals 1200 mm. Adding the preset edge allowance to consider the cabinet shell thickness and installation space, the horizontal direction is based on a 1500 mm span, with an edge allowance of 75 mm added to the left and right sides, resulting in a cabinet length of 1500 + 150 = 1650 mm. The vertical direction is based on a 550 mm span, with an edge allowance of 50 mm added to the front and back, resulting in a cabinet width of 550 + 100 = 650 mm. The vertical direction is based on a 1200 mm span, with an additional 250 mm added to the bottom for the base and casters, and an additional 434 mm added to the top for the access panel and electrical control panel space, resulting in a cabinet height of 1200 + 684 = 1884 mm. The resulting cabinet outer dimensions include a length of 1650 mm, a width of 650 mm, and a height of 1884 mm.

[0031] The WebGL graphics rendering engine is invoked to begin 3D model construction. Based on the cabinet's outer envelope dimensions, a cabinet shell model is generated. A cuboid mesh object is created with a length of 1650 mm, a width of 650 mm, and a height of 1884 mm. Material properties are set for this cuboid object: color is grayish-white, and red, green, and blue values ​​are 240, 240, 240, simulating a sprayed steel plate effect. The material type is diffuse to add ambient light occlusion. Shelf models are generated based on the vertical coordinates and storage type identifiers for each layer. The second layer, with a vertical coordinate of 650 mm, is a single-phase surface shelf. Create a rectangular patch object with a length of 1550 mm, a width of 550 mm, and a thickness of 15 mm. Set the center position of the patch to an x-coordinate of 800 mm, a y-coordinate of 300 mm, and a height of 650 mm. Assign this patch a light blue material with red, green, and blue values ​​of 173, 216, and 230. For the fourth layer, with a vertical coordinate of 1050 mm, create a rectangular patch for the storage location type (transformer layer plate) and set it to a light yellow material with red, green, and blue values ​​of 255, 255, and 224. Create all five layers of the layer plate model sequentially. Generate the storage unit model based on the storage location coordinates in the storage location 3D coordinate set, starting with the first storage location. The coordinates are 285 mm x-coordinate, 450 mm y-coordinate, and 680 mm height. Create a cube mesh object with a side length slightly smaller than the actual size. Set the side length of the single-phase table storage cube to 120 mm, and align the center of the cube with the storage coordinates. Set the initial material of this cube to semi-transparent yellow with red, green, and blue values ​​of 255, 255, and 0, and an opacity channel value of 0.3. Create all 35 storage unit models in sequence. Combine the cabinet shell model, shelf model, and storage unit model to form a complete scene tree structure. The cabinet shell model is the root node, and the shelf model is a child node attached to the shell. Below, the storage unit model is mounted as a grandchild node under the corresponding shelf. The scene camera position and viewing direction are set. The camera position is set 2 meters in front of the cabinet and 1 meter high, looking towards the center of the cabinet. Ambient light and directional light are added to illuminate the scene. The rendering loop is started to refresh the screen at a frequency of 30 frames per second. The resulting 3D simulation model is presented on the display screen as a virtual model that completely corresponds to the physical turnover cabinet space structure. This solves the problem of lack of flexibility in the static configuration method in the existing technology. When the storage layout is adjusted, the clustering recognition and model generation process can be re-executed to automatically adapt to the new cabinet structure.

[0032] In one specific embodiment, step S3 includes: The output level signal of the photoelectric sensor of each storage location is collected as the photoelectric signal, the asset barcode string is obtained through the barcode scanning module as the barcode data, and the EPC code of the asset tag is obtained through the RFID reading and writing module as the RFID asset tag information. Determine whether the storage location occupancy status indicated by the photoelectric signal is consistent with the existence status of the barcode data and RFID asset tag information. Mark the storage location record that matches the status of at least two types of sensor data in the consistency judgment result as a valid record to obtain valid storage location status data. Extract the storage location code and asset occupancy identifier from the valid storage location status data. Set the corresponding storage location unit model in the 3D simulation model to a semi-transparent material for storage locations with an empty asset occupancy identifier, and set the corresponding storage location unit model in the 3D simulation model to an opaque material for storage locations with an asset occupancy identifier that are in stock. If there are conflicting sensor data records in the valid storage space status data, set the corresponding storage space unit model in the 3D simulation model to a flashing red material to update the display status of the storage space in the 3D simulation model.

[0033] Specifically, the output level signal of the ITR9909 slot-type photoelectric switch at each storage location is collected as the photoelectric signal. The photoelectric switch includes an infrared emitting tube and a receiving tube. The emitting tube emits a 940 nm wavelength infrared beam. When an energy meter or current transformer is placed in the storage location, the object blocks the beam, causing the receiving tube to not receive the light signal and output a low level. When the storage location is empty, the beam is unobstructed, and the receiving tube outputs a high level. The control board reads the level status through the S signal pin of the PH2.0-3P interface. A low level indicates goods are present, and a high level indicates no goods are present. The asset barcode string is obtained through an E10Y barcode scanning module. The scanning module has a built-in 120 frames per second image sensor with a field of view of 30 degrees horizontally and 37 degrees vertically. When the photoelectric switch detects goods, it triggers the scanning module to start. The scanning engine continuously scans for 3000 milliseconds within a depth of field range of 4 mm to 280 mm. After the image sensor captures the barcode image, the decoding algorithm identifies the barcode. A two-digit string, such as 4230001001000069817550, is used as barcode data. This barcode string is transmitted to the control board via a TTL serial port. The RFID reader / writer module retrieves the electronic product code of the asset tag. The reader / writer module operates in the 860-960 MHz band using frequency hopping spread spectrum, dividing the band into 50 channels, each with a bandwidth of 2 MHz. The reader / writer module sequentially transmits carrier signals for 200 milliseconds on each channel. After receiving the carrier signal, the asset tag modulates and returns its stored electronic product code via backscattering. The reader / writer module demodulates the received signal and extracts the electronic product code, such as E271000000000000000000021012, as the RFID asset tag information. The first two digits of this code indicate the asset type: E2 for single-phase meters, E3 for three-phase meters, and E4 for current transformers. The middle 18 digits represent the production batch and serial number, and the last two digits are the check digit.

[0034] To determine if the storage location occupancy status indicated by the photoelectric signal is consistent with the presence status of barcode data and RFID tag information, a status comparison table for the three sensor data is established. A low level photoelectric signal corresponds to an in-stock status, while a high level corresponds to an out-of-stock status. Obtaining a 22-bit string from the barcode data corresponds to an in-stock status, while a scan timeout corresponds to an out-of-stock status. Reading the RFID tag information corresponds to an in-stock status, while not reading the code corresponds to an out-of-stock status. For example, if a storage location's photoelectric signal is low, it indicates in-stock status; barcode scanning yields the string "4230001001000069817550", indicating in-stock status; and the RFID tag reads the code "E2710000000000000000000021012", indicating in-stock status. All three sensor data states are in-stock, indicating a consistent occupancy. For another storage location, a low photoelectric signal indicates in-stock status, but a barcode scan timeout indicates an out-of-stock status, and the RFID tag not being read also indicates an out-of-stock status. In this case, the photoelectric signal indicates in-stock status, but the barcode data is out-of-stock. If the data from both the photoelectric sensor and the radio frequency sensor show that the storage location is empty, the following conditions will be met: 1) the photoelectric sensor shows a high level, the barcode scan does not acquire data, the radio frequency tag does not read the code, and the data from all three sensors shows that the storage location is empty. Records with at least two matching sensor data are marked as valid. For the first storage location, all three sensors are consistent, and it is marked as valid. For the second storage location, only the photoelectric sensor shows "in stock," the barcode shows "empty," and the radio frequency tag shows "empty," but these three are inconsistent, yet the barcode and radio frequency tags are consistent, and it is marked as valid. For the third storage location, all three sensors show "empty," and it is marked as valid. If a storage location shows "in stock" on the photoelectric sensor and on the barcode, but the radio frequency tag shows "empty," and the photoelectric sensor and barcode tag are consistent, it is also marked as valid. Only when all three sensor data are inconsistent (e.g., photoelectric sensor shows "in stock," barcode shows "empty," radio frequency tag shows "in stock") is it marked as invalid. Valid storage location status data includes fields such as storage location code, photoelectric sensor status, barcode string, radio frequency tag code, and overall occupancy status.

[0035] Extract the storage location code and asset occupancy identifier from the valid storage location status data. The storage location code is a unique identifier formed by ZJ plus the cabinet number and the storage location serial number. The asset occupancy identifier is determined based on comprehensive sensor data. If all three sensors show that the item is in stock, or at least two sensors show that the item is in stock, the occupancy identifier is set to in stock. If all three sensors show that the item is empty, or at least two sensors show that the item is empty, the occupancy identifier is set to empty. The asset occupancy identifier corresponding to storage location code ZJ000115 is found to be empty. Locate the storage unit model node with storage location code ZJ000115 in the scene tree of the 3D simulation model, obtain the material object reference of this node, and modify the red, green, and blue color attributes of the material object. Set the values ​​to 255, 255, 0 (yellow), and modify the transparency channel value of the material object to 0.3 to make the material semi-transparent. The storage unit model is displayed as a semi-transparent yellow cube on the screen, indicating that the storage unit is empty. Extract the asset occupancy identifier corresponding to the storage unit code ZJ000122, which is in stock. Find the storage unit model node in the scene tree, modify the red, green, and blue values ​​of its material object to 128, 128, 128 (gray), and modify the transparency channel value to 1.0 to make the material completely opaque. The storage unit model is displayed as an opaque gray cube, indicating that the storage unit is in stock. Process the display status update of all 35 storage units in sequence.

[0036] Storage location records with sensor data conflicts are detected within the valid storage location status data. A conflict is defined as a discrepancy between the photoelectric signal and barcode or radio frequency data. For example, a storage location might have a low-level photoelectric signal indicating stock availability, but the barcode scan timed out and the radio frequency failed to read the tag, both indicating empty stock. Although the barcode and radio frequency data match, the record is marked as valid, but a conflict exists between the photoelectric signal and the other two. Therefore, this storage location record is marked as conflicting. In the 3D simulation model, the node representing this storage location unit is located, and its material object's red, green, and blue values ​​are modified to 255, 0, 0 (pure red). A material animation controller is created with a blinking effect. The animation controller periodically switches between a transparency channel value of 1.0 and 0.3, with a switching period of 500 milliseconds, meaning it changes between opaque and semi-transparent every 500 milliseconds. This storage location unit model... The display of a red flashing cube on the screen attracts the operator's attention. Another storage location's photoelectric display shows that the barcode has been successfully acquired, but the radio frequency (RF) cannot read the tag. The photoelectric display and barcode are consistent, but there is a conflict with the RF display, so it is also marked as a conflict and set to flashing red. This completes the update of the display status of all storage locations in the 3D simulation model. By observing the 3D model on the screen, the operator can intuitively identify which storage locations are empty and display semi-transparent yellow, which storage locations have goods and display opaque gray, and which storage locations have abnormal data and display flashing red. This solves the problem of the simulation display not being synchronized with the actual status in the existing technology. Through multimodal sensor data fusion and cross-validation mechanism, the consistency accuracy between the simulation model and the physical state is improved from the original 92% to 98.5%, and the detection rate of abnormal situations is improved to 95%.

[0037] In one specific embodiment, step S4 includes: Each storage location in the turnover cabinet is abstracted as a graph node. Connection edges between nodes are established based on the spatial distance relationship between storage locations and the historical business relationship to obtain the storage location relationship graph. Extract the current occupancy status, outbound frequency, inbound frequency, and remaining days for verification of each storage location, and combine them to form a node feature vector. Input the node feature vector and the adjacency matrix of the storage location relationship graph into a graph convolutional network for convolution operation to obtain the access probability value of each storage location. Based on the access probability value, the corresponding storage unit model of each storage location in the 3D simulation model is set to a gradient color from blue to red. Storage locations with high access probability values ​​are rendered as red, and storage locations with low access probability values ​​are rendered as blue, thus obtaining a visualized thermal distribution. Storage locations whose access probability exceeds a preset probability threshold and are currently empty are selected as recommended storage locations. A green halo effect is then overlaid on the corresponding storage location unit model in the 3D simulation model to complete the intelligent warehousing guidance.

[0038] Specifically, the 35 storage locations of the turnover cabinet are abstracted as graph nodes, each node corresponding to a storage location code. The spatial Euclidean distance between any two storage locations is calculated. Storage location 1 has coordinates of x-coordinate 285, y-coordinate 450, and height 680; storage location 2 has coordinates of x-coordinate 425, y-coordinate 450, and height 680. The horizontal distance between the two storage locations is 425 - 285 = 140 mm, the vertical distance is 0 mm, and the spatial distance is the square root of the sum of the squares of 140 mm and 0. This distance is then checked against a preset threshold of 350 mm. If the distance between storage location 1 and storage location 2 is 140 mm, which is less than 350 mm, a spatial adjacency edge is established between the two nodes. The edge weight is calculated as the negative distance of the natural exponential function divided by 100, which is approximately the negative 1.4 power of the natural constant, equal to 0.247. Although storage location 18 is spatially distant (more than 350 mm), no spatial adjacency edge is established. However, a query of the past 6 months of outbound work order records reveals that storage location 1 and storage location 18 were simultaneously picked up 25 times in the same work order. The total number of work orders is 100. The co-occurrence rate of the two storage locations is 25 divided by 100, which equals 0.25. It is determined that the number of co-occurrences exceeds the preset threshold of 10 times, so a business association edge is established between the two nodes. The edge weight is directly adopted using the co-occurrence rate of 0.25. The spatial distance and business co-occurrence relationship between all 35 storage locations are calculated sequentially, and a 35x35 adjacency matrix is ​​constructed. The spatial adjacency edge weight of 0.247 is filled in the first row and second column of the matrix, and the business association edge weight of 0.25 is filled in the first row and 18th column. The matrix element corresponding to storage locations without connection relationship is filled with 0. The resulting storage location relationship graph contains 35 nodes and several strip weighted connection edges.

[0039] Extract the current occupancy status of each storage location by querying the occupancy field of the storage location table from the database. Occupancy status is coded as 1 for "in stock" and 0 for "empty". Extract the outbound frequency by counting the number of outbound operations for each storage location in the past 30 days from the operation log table. Storage location 1 has had 12 outbound operations in the past 30 days, recorded as a frequency of 12. Extract the inbound frequency by counting the number of inbound operations for each storage location in the past 30 days from the operation log table. Storage location 1 has had 10 inbound operations in the past 30 days, recorded as a frequency of 10. Extract the remaining days for verification by querying the verification expiration timestamps of assets within the storage location from the asset file table. Storage location 1... The asset's inspection expiration timestamp represents a future date, and the current timestamp represents the current date. Subtracting the two timestamps gives the difference in seconds, which is converted to days to obtain the remaining inspection days, for example, 365 days. The remaining inspection days are normalized to the range of 0 to 1. The normalization calculation is the remaining inspection days divided by the total number of days in the inspection cycle. For example, the inspection cycle for an electricity meter is 730 days (2 years), so the normalized value is 365 divided by 730, which equals 0.5. The storage location type code is extracted: single-phase meter storage location code is 100, three-phase meter storage location code is 010, and current transformer storage location code is 001. These values ​​are combined to form a node. The feature vector for storage location 1 is 12-dimensional, containing features such as occupancy status (1), outbound frequency (12), inbound frequency (10), normalized value of remaining days for verification (0.5), one-hot encoding of storage location type (100), normalized values ​​of storage location x-coordinate, y-coordinate, and height, and the number of adjacent empty storage locations. The feature vectors of the 35 storage locations are stacked in rows to form a 35x12 node feature matrix. A self-connection is achieved by adding 1 to each diagonal position of the adjacency matrix. The connection weight between storage location 1 and itself is set to 1. The modified adjacency matrix has all diagonal elements set to 1. 1. Calculate the degree value of each node, which is the sum of the elements in each row of the adjacency matrix. Storage position 1 has an edge weight of 0.247 with storage position 2, an edge weight of 0.25 with storage position 18, and an edge weight of 1 with itself. Its degree value is the sum of 0.247, 0.25, 1, and the weights of other adjacent edges. Assuming that storage position 1 is connected to a total of 5 neighbors and has a self-connection degree of 6, construct a degree matrix as a 35x35 diagonal matrix. The diagonal elements are the degree values ​​of each node. Calculate the negative half power of the degree matrix, which is the square root of each diagonal element and the reciprocal. The negative half power of the degree value of 6 for storage position 1 is the reciprocal of the square root of 6, which is approximately equal to 0.408, yielding a diagonal matrix of the degree matrix to the power of -1 / 2. Multiplying this matrix left-by the adjacency matrix and then right-by the normalized adjacency matrix, we obtain a 35x12 intermediate matrix. Each row of this intermediate matrix is ​​a weighted average of the features of the corresponding storage location and its neighboring nodes. This is then multiplied by the 12x32 weight matrix of the first convolutional layer to obtain a 35x32 matrix. The negative values ​​in the matrix are set to 0, while the positive values ​​are retained, through a modified linear unit activation function. This yields the first-layer output matrix containing the spatial neighborhood features of the storage locations. The first-layer output matrix undergoes the same convolution process, multiplied by the normalized adjacency matrix and then by the 32x64 weight matrix of the second convolutional layer, followed by an activation function to obtain a 35x64 second-layer output matrix. Including the business-related features of the storage locations, the third convolutional layer weight matrix (64 rows, 16 columns) is used to obtain a final output matrix (35 rows, 16 columns) that fuses both spatial and business features. A 16-dimensional vector is extracted from each row of the final output matrix. The natural exponent value is calculated for each of the 16 elements of this vector. The sum of these 16 natural exponent values ​​is used as the normalized denominator. The normalized probability of each element is obtained by dividing its natural exponent value by the normalized denominator. The sum of the 16 normalized probabilities is 1. The normalized probability of the first element is taken as the predicted probability of that storage location being accessed within the next 7 days. The access probability value for storage location 1 is 0.82, indicating an 82% chance of being accessed within the next 7 days. This yields the access probability vectors for all 35 storage locations.

[0040] The display color of each storage location in the 3D simulation model is calculated based on the access probability value. The color transitions linearly from blue to red. The blue, red, green, and blue values ​​are 0, 0, and 255, while the red, red, green, and blue values ​​are 255, 0, and 0. For storage location 1, with an access probability value of 0.82, the red component is approximately 209 (255 multiplied by 0.82), the blue component is approximately 46 (255 multiplied by 1 minus 0.82), and the green component is always 0. Therefore, the red, green, and blue values ​​for this storage location are set to 209, 0, and 46, resulting in an orange-red color. For storage location 2, with an access probability value of 0.15, the red component is approximately 38 (255 multiplied by 0.15), and the blue component is... The color component is 255 multiplied by 0.85, which is approximately 217. The red, green, and blue values ​​are set to 38, 0, and 217, respectively, to produce a dark blue color. In the 3D simulation model scene tree, all storage unit model nodes are traversed. The cube model node corresponding to storage 1 is found, and its material object color attribute is modified to red, green, and blue 209, 0, and 46. The cube model node corresponding to storage 2 is found, and its material object color attribute is modified to red, green, and blue 38, 0, and 217. The colors of all 35 storage unit models are updated in sequence to obtain the visualized thermal distribution of the 3D simulation model. When the operator observes the model, red storage units indicate a high access probability, and blue storage units indicate a low access probability.

[0041] Storage locations with access probability values ​​exceeding a preset probability threshold of 0.8 are filtered. The access probability value vectors of 35 storage locations are traversed. Storage location 1 (probability value 0.82, greater than 0.8), storage location 5 (probability value 0.85, greater than 0.8), and storage location 12 (probability value 0.78, less than 0.8) are not filtered, resulting in a list of storage locations with high access probabilities. Then, storage locations currently empty are filtered from this list. For storage location 1, a value of 1 indicates it has stock and is not filtered; for storage location 5, a value of 0 indicates it is empty and is retained. This yields a recommended list of storage locations that meet both high access probability and current empty status. A green halo effect is overlaid on the recommended storage locations in the 3D simulation model. A circular geometric object is created around the cube model node corresponding to storage location 5, with the inner ring... The outer diameter is 5 mm larger than the inner diameter, and the ring material is set to self-illuminating green with red, green, and blue values ​​of 0, 255, and 0. The luminous intensity parameter is set to dynamically change, and an animation controller is created to make the luminous intensity change periodically between 0 and 1 according to a sine function with a change period of 1000 milliseconds. The ring appears as a green halo on the screen, flashing bright and dark with a 1-second cycle. When the operator sees the green halo on the storage location, they know that the location is the recommended storage location, thus completing the intelligent storage guidance. This solves the problem of insufficient intelligence and inability to predict the optimal storage path in existing technologies. By combining the first-in, first-out requirement of power metering assets with the characteristics of power supply station business processes through graph convolutional networks, the average asset retrieval time is reduced from the original 45 seconds to 28 seconds.

[0042] In one specific embodiment, the node feature vectors and the adjacency matrix of the storage location relation graph are input into a graph convolutional network for convolution operations to obtain the access probability value of each storage location, including: The node feature vectors are arranged in the order of storage locations to construct a node feature matrix. An identity matrix is ​​added to the adjacency matrix of the storage location relationship graph to achieve self-connection. The degree value of each node is calculated to form a degree matrix. The normalized adjacency matrix is ​​obtained by performing matrix multiplication operation with the adjacency matrix based on the negative half power of the degree matrix. Multiply the node feature matrix with the normalized adjacency matrix, then multiply it with the weight matrix of the first convolutional layer, and filter out negative values ​​through the ReLU activation function to obtain the first layer intermediate feature matrix containing the features of the storage space neighborhood. Multiply the first-layer intermediate feature matrix with the normalized adjacency matrix, then multiply it with the weight matrix of the second convolutional layer and activate it with ReLU to obtain the second-layer intermediate feature matrix containing the storage location business-related features. Repeat the above convolution operation on the second-layer intermediate feature matrix and pass it through the third convolutional layer to obtain the target feature matrix that integrates the dual features of space and business. Extract the feature vectors of each storage location in the target feature matrix, perform Softmax normalization on each feature vector, and use the normalized values ​​as the probability prediction values ​​of the corresponding storage location being accessed in the future time window to obtain the access probability value of each storage location.

[0043] Specifically, the node feature vectors of the 35 storage sites are arranged in ascending order of storage site number. The 12-dimensional feature vector of storage site 1 is used as the first row, the 12-dimensional feature vector of storage site 2 as the second row, and so on up to the 12-dimensional feature vector of storage site 35 as the 35th row, constructing a 35-row, 12-column node feature matrix. An identity matrix is ​​added to the adjacency matrix of the storage site relationship graph to achieve self-connection. In the original adjacency matrix, the element in the first row and first column is 0, indicating that storage site 1 is not connected to itself. After adding the identity matrix, the element at that position becomes 1, indicating that storage site 1 has established a connection with itself. The diagonal elements of the adjacency matrix are all changed from 0 to 1. The modified adjacency matrix is ​​denoted as the self-join adjacency matrix. The degree value of each node is calculated to form a degree matrix. Traversing each row of the self-join adjacency matrix, the sum of all elements in that row is calculated to obtain the degree value of the corresponding node. The first row of elements corresponding to storage position 1 contains the edge weight of 0.247 with storage position 2, the edge weight of 0.25 with storage position 18, the connection weight of itself, and the edge weights with other neighbors. The degree value of storage position 1 is summed to obtain a degree value of 6. The degree values ​​of the 35 storage positions are filled into a 35x35 degree matrix. The diagonal positions of the matrix are as follows: the element in the first row and first column of the degree matrix is ​​6, and the rest are 0; the element in the second row and second column is the degree value of storage position 2, and the rest are 0, and so on. The normalized coefficient matrix is ​​calculated using the negative half power of the degree matrix. The negative half power of the element 6 in the first row and first column of the degree matrix is ​​calculated by first taking the square root of 6, which is approximately 2.449, then taking its reciprocal, which is approximately 0.408. Performing the same operation on each element on the diagonal of the degree matrix yields a normalized coefficient matrix that is still a diagonal matrix. The left normalized matrix is ​​obtained by left multiplying the normalized coefficient matrix by the self-joined adjacency matrix. The first element of the left normalized matrix is ​​the corresponding element of the first row of the self-connected adjacency matrix multiplied by 0.408. Then, the left normalized matrix is ​​multiplied by the normalization coefficient matrix to obtain the normalized adjacency matrix. The element in the second column of the first row of the normalized adjacency matrix is ​​calculated as the original edge weight 0.247 multiplied by the normalization coefficient of storage position 1 0.408, and then multiplied by the normalization coefficient of storage position 2, which is assumed to be 0.5, resulting in approximately 0.05. This normalization operation reduces the adjacency weight of nodes with large degree values ​​and amplifies the adjacency weight of nodes with small degree values, thus obtaining the normalized adjacency matrix for subsequent convolution operations.

[0044] A matrix multiplication operation is performed between the 35x12 node feature matrix and the 35x35 normalized adjacency matrix. Each row of the node feature matrix is ​​a 12-dimensional feature vector, and each row of the normalized adjacency matrix is ​​a 35-dimensional weight vector. The first row of the matrix multiplication result is calculated by multiplying the 35 weights of the first row of the normalized adjacency matrix by the corresponding rows of the 35 nodes in the node feature matrix and then summing the results. The new feature of storage location 1 is equal to its own feature multiplied by the self-join weight, plus the feature of storage location 2 multiplied by the adjacency weight, plus the feature of storage location 18 multiplied by the business association weight, plus the weighted sum of the features of other neighbors. This results in a 35x12 aggregated feature matrix, where each row is the weighted average of the features of the corresponding storage location and its neighbors. This is then multiplied by the 12x32 weight matrix of the first convolutional layer. The 12-dimensional vectors of each row of the aggregated feature matrix are multiplied by the 12 rows of the weight matrix. Multiplying 32 columns yields a 32-dimensional output vector. The weight matrix, learned through training, functions to linearly transform the 12-dimensional input features to a 32-dimensional feature space, resulting in a 35x32 convolutional output matrix. A modified linear unit activation function is used to process each element of the convolutional output matrix. The activation function rule is that if the element value is greater than 0, it remains unchanged; if the element value is less than or equal to 0, it is set to 0. For example, an element with a value of -2.3 becomes 0 after the activation function, and an element with a value of -5.6 remains 5.6 after the activation function. The activation function filters out negative values ​​and retains only positive values, introducing non-linearity. This yields a first-layer intermediate feature matrix of 35x32 columns containing features of the storage space neighborhood. The 32-dimensional feature vector in each row of this matrix integrates the original features of the corresponding storage space and its spatially neighboring storage spaces.

[0045] The first-layer intermediate feature matrix (35 rows, 32 columns) is multiplied with the normalized adjacency matrix (35 rows, 35 columns), with the calculation process being the same as the first layer. The 35 weights in the first row of the normalized adjacency matrix are multiplied by the 32-dimensional feature vectors in the 35th row of the first-layer intermediate feature matrix, and then summed to obtain a second aggregated feature matrix (35 rows, 32 columns). This second aggregated feature matrix is ​​then multiplied by the second convolutional layer weight matrix (32 rows, 64 columns). Each row of the second aggregated feature matrix's 32-dimensional vector is multiplied by the weight matrix (32 rows, 64 columns) to obtain a 64-dimensional output vector. The weight matrix maps the 32-dimensional features to a 64-dimensional feature space, resulting in a second-layer convolutional output matrix (35 rows, 64 columns). After processing with a modified linear unit activation function to filter out negative values ​​and retain positive values, a second-layer intermediate feature matrix (35 rows, 64 columns) containing features related to storage location business is obtained. The 64-dimensional feature vector in each row of this matrix is... Based on the spatial neighborhood features, business relationships are further integrated. The second layer features of storage location 1 not only include information about its spatial neighbor storage location 2, but also information about business-related storage location 18. The 35-row, 64-column intermediate feature matrix of the second layer is multiplied by the normalized adjacency matrix and then by the 64-row, 16-column weight matrix of the third convolution layer. The 64-dimensional vector of each row of the second-layer intermediate feature matrix is ​​multiplied by the 64-row, 16-column weight matrix to obtain a 16-dimensional output vector. The weight matrix compresses the 64-dimensional features into a 16-dimensional feature space, resulting in a 35-row, 16-column third-layer convolution output matrix. After activation function processing, a target feature matrix of 35-row, 16-column that integrates both spatial and business features is obtained. The 16-dimensional feature vector of each row of this matrix, extracted by the three-layer graph convolutional network, fully captures the location characteristics and business characteristics of the corresponding storage location in the turnover cabinet map structure.

[0046] Extract the feature vectors corresponding to each storage location in the target feature matrix. The 16-dimensional feature vector corresponding to storage location 1 in the first row is denoted as Vector 1, the 16-dimensional feature vector corresponding to storage location 2 in the second row is denoted as Vector 2, and so on. Perform Softmax normalization on each feature vector. Softmax normalization maps the multiple element values ​​of the feature vector to a probability distribution such that the sum of all element values ​​is 1. Calculate the natural exponent values ​​for each of the 16 elements of Vector 1. Assume the first element value is 2.3, with a natural exponent of approximately 9.974 (2.3 times the natural constant). Assume the second element value is -0.8, with a natural exponent of approximately 0.449. Assume the third element value is 1.5, with a natural exponent of approximately 4.482. Calculate all 16 natural exponent values ​​and sum them up. Assume the sum is 120. Divide the natural exponent value of the first element (9.974) by the sum of 120. The normalized probability of the first element is approximately 0.083. Dividing the natural exponent value of the second element (0.449) by the sum of 120 yields a normalized probability of approximately 0.004. Dividing the natural exponent value of the third element (4.482) by the sum of 120 yields a normalized probability of approximately 0.037. These are used to calculate 16 normalized probabilities, which are then summed to 1. The normalized probability of the first element (0.083) is taken as the predicted probability of storage location 1 being accessed within the next 7-day time window. If the training data indicates that storage location 1 is indeed accessed within the next 7 days, a higher probability indicates a more accurate prediction. The Softmax normalization operation is repeated for all 35 storage locations to extract their respective access probability predictions, resulting in a one-dimensional array containing the access probability values ​​of the 35 storage locations. This array guides subsequent heat map visualization and intelligent storage recommendation, solving the problem of the inability to predict storage location access probabilities in existing technologies.

[0047] In one specific embodiment, step S5 includes: Extract the entry timestamp and verification expiration timestamp of the assets in the occupied storage space from the asset archive database. Calculate the asset entry duration based on the time difference between the current timestamp and the entry timestamp. Calculate the remaining expiration duration based on the time difference between the verification expiration timestamp and the current timestamp. Query historical outbound records of the same type of assets, calculate the actual inventory cycle of each historical record, and obtain the average inventory cycle and the standard deviation of the inventory cycle. Divide the difference between the asset's storage time and the average inventory cycle by the standard deviation of the inventory cycle, and substitute the result into the normal distribution cumulative probability function to obtain the overdue risk probability value of the corresponding asset. Determine whether the probability value of the overdue risk exceeds the preset risk threshold or whether the remaining time to maturity is less than the preset number of days threshold. Set the corresponding storage unit model in the three-dimensional simulation model of the storage location that meets either condition to orange or red warning material to complete the warning status marking.

[0048] Specifically, the system queries the asset archive database for the entry timestamp and verification expiration timestamp fields of assets within occupied storage locations. The timestamps are in Unix format, representing the number of seconds elapsed since 00:00 on January 1, 1970. The entry timestamp for the single-phase electricity meter in storage location 15 is 1699920000, which, when converted to a Gregorian calendar date, corresponds to the entry date. The verification expiration timestamp is 1762992000, which, when converted to a Gregorian calendar date, corresponds to the expiration date. The system then retrieves the current timestamp, 1731456000, corresponding to the current date at the time of the query. Based on the time difference between the current timestamp and the entry timestamp, the asset's storage duration is calculated. The time difference is the current... The difference in seconds between the previous timestamp and the entry timestamp is divided by 86400 (the number of seconds in a day) to get the number of days. The storage duration of an asset in storage location 15 is calculated by subtracting the entry timestamp from the current timestamp and dividing the result by 86400 to get the number of days the asset has been in storage. The remaining time until expiration is calculated based on the time difference between the verification expiration timestamp and the current timestamp. The time difference is calculated by subtracting the current timestamp from the verification expiration timestamp and dividing the result by 86400 to get the number of days. The remaining time until expiration of an asset in storage location 15 is calculated by subtracting the current timestamp from the verification expiration timestamp and dividing the result by 86400 to get the number of days remaining until the verification expiration.

[0049] To query historical outbound records of the same type of asset, retrieve all outbound records of single-phase electricity meters from the operation log database. Each outbound record includes fields such as asset code, inbound timestamp, and outbound timestamp. Calculate the actual inventory period for each historical record by subtracting the inbound timestamp from the outbound timestamp and then dividing by 86400 to obtain the actual number of days the asset was in stock. Assuming 50 historical outbound records of single-phase electricity meters are retrieved, the actual inventory period for the first record is 180 days (outbound time minus inbound time), the second record is 195 days, the third record is 210 days, and so on. Calculate the average inventory period by summing the inventory periods of the 50 records and then dividing by 50. The sum is 180 + 195 + 210 + ... Assuming the total period value of the remaining 47 records is 10,000 days, and the average inventory period is 10,000 divided by 50, which equals 200 days, we calculate the standard deviation of the inventory period to measure the volatility of the inventory period. First, we calculate the difference between the inventory period of each record and the average. The difference for the first record is 180 minus 200, which equals -20 days; the difference for the second record is 195 minus 200, which equals -5 days; and the difference for the third record is 210 minus 200, which equals +10 days. We square the difference for each record: the square of the difference for the first record is -20 squared, which equals 400; the square of the difference for the second record is -5 squared, which equals 25; and the square of the difference for the third record is +10 squared, which equals 100. We sum the squares of the differences for all 50 records and divide by 50 to get the variance. We take the square root of the variance to get the standard deviation, which is assumed to be 30 days.

[0050] The standardized deviation is obtained by dividing the difference between the asset's storage time and the average inventory cycle by the standard deviation of the inventory cycle. Assuming the storage time of an asset within storage space 15 is 365 days, the difference between this and the average inventory cycle of 200 days is 365 minus 200 equals 165 days. Dividing this difference of 165 days by the standard deviation of 30 days gives a standardized deviation of 165 divided by 30 equals 5.5. The standardized deviation indicates that the asset's storage time exceeds the average by 5.5 standard deviations. Substituting the standardized deviation into the cumulative probability function of a normal distribution, we calculate the probability of overdue risk. The cumulative probability function calculates the probability of being less than or equal to a certain standardized value. A standardized deviation of 5.5 corresponds to a cumulative probability close to 1, indicating that the inventory cycle of almost all normal assets is shorter than the current storage time. Subtracting this cumulative probability from 1 gives the overdue probability. Since the cumulative probability is close to 1, the overdue probability is close to 1 minus 1, meaning the overdue risk is extremely high. The overdue risk probability of an asset within storage space 15 is approximately 0.999, indicating that the asset has significantly exceeded the normal inventory cycle range.

[0051] The system determines whether the probability of overdue risk exceeds a preset risk threshold. A preset risk threshold of 0.8 indicates an overdue probability exceeding 80%, classifying it as high risk. The overdue risk probability of assets within storage location 15 is 0.999, greater than the preset risk threshold of 0.8, satisfying the first judgment condition. Next, it determines whether the remaining time until maturity is less than a preset number of days threshold. A preset number of days threshold of 7 indicates that the due date is less than 7 days, classifying it as nearing maturity. Assuming the remaining time until maturity of assets within storage location 20 is 5 days, less than the preset number of days threshold of 7, satisfying the second judgment condition, storage locations meeting either condition are marked as having a warning status in the 3D simulation model and searched in the scene tree. For the storage unit model node corresponding to storage location 15, modify its material object color attribute (red, green, blue) to 255, 0, 0 (pure red) to indicate a serious overdue warning. Find the storage unit model node corresponding to storage location 20 and modify its material object color attribute (red, green, blue) to 255, 140, 0 (orange) to indicate an impending expiration warning. Complete the warning status marking. When the operator observes the 3D simulation model, a red storage location indicates that the asset has seriously exceeded the normal inventory cycle and needs to be handled immediately, while an orange storage location indicates that the asset is about to reach its inspection validity period and needs to be sent for inspection or released for use as soon as possible. This solves the problem of passively querying overdue information in the existing technology and transforms it into an active warning mode.

[0052] The inventory simulation display method of the intelligent turnover cabinet in the embodiments of this application has been described above. The inventory simulation display system of the intelligent turnover cabinet in the embodiments of this application is described below. Please refer to [link / reference]. Figure 3 One embodiment of the intelligent turnover cabinet inventory simulation display system in this application includes: The data acquisition module is used to collect signal strength data of RFID self-inspection tags in each storage location within the turnover cabinet under multiple antennas. Based on the distance attenuation relationship between the signal strength data and the antenna position, the three-dimensional coordinate set of the storage location is calculated. The clustering module is used to cluster the vertical coordinates of the three-dimensional coordinate set of the storage sites, identify the hierarchical grouping results, and generate a three-dimensional simulation model with storage site type identifier based on the number and spacing characteristics of each layer of the hierarchical grouping results. The verification module is used to collect photoelectric signals, barcode data and RFID asset tag information, perform consistency verification on the photoelectric signals, barcode data and RFID asset tag information, filter to obtain valid storage location status data, and update the display status of the storage location in the three-dimensional simulation model according to the storage location status data; The calculation module is used to construct a reservoir relationship map, extract the current state and historical access records of the reservoirs as map node features, calculate the map node features through a graph convolutional network to obtain the reservoir access prediction results, and generate a visualized thermal distribution in the three-dimensional simulation model based on the access prediction results. The marking module is used to obtain the asset storage time and remaining time until expiration of occupied storage spaces, calculate the overdue risk value based on the comparison between the storage time and historical periods, and mark the storage spaces with the overdue risk value exceeding the set value as warning status in the three-dimensional simulation model.

[0053] above Figure 3 The inventory simulation display system of the intelligent turnover cabinet in this embodiment of the invention is described in detail from the perspective of modular functional entities. The inventory simulation display device of the intelligent turnover cabinet in this embodiment of the invention is described in detail from the perspective of hardware processing.

[0054] Reference Figure 4 This invention also provides an inventory simulation display device for an intelligent turnover cabinet. This intelligent turnover cabinet inventory simulation display device can be a server, and its internal structure can be as follows: Figure 4 As shown, the intelligent turnover cabinet's inventory simulation display device includes a processor, memory, display screen, input device, network interface, and database connected via a system bus. The processor, designed as a computer, provides computing and control capabilities. The memory of the intelligent turnover cabinet's inventory simulation display device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the intelligent turnover cabinet's inventory simulation display device stores the data corresponding to this embodiment. The network interface of the intelligent turnover cabinet's inventory simulation display device is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements the above-described method.

[0055] Those skilled in the art will understand that Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the intelligent turnover cabinet inventory simulation display device to which the present invention is applied.

[0056] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when the instructions are executed on a computer, cause the computer to perform the steps of the inventory simulation display method of the intelligent turnover cabinet.

[0057] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0058] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause an intelligent turnover cabinet inventory simulation display device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0059] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for simulating and displaying inventory in an intelligent turnover cabinet, characterized in that, The method includes: Step S1: Collect the signal strength data of the RFID self-test tags of each storage location in the turnover cabinet under multiple antennas, and calculate the three-dimensional coordinate set of the storage location based on the distance attenuation relationship between the signal strength data and the antenna position. Step S2: Cluster the vertical coordinates of the three-dimensional coordinate set of the storage sites to identify the hierarchical grouping results. Based on the number and spacing characteristics of storage sites in each layer of the hierarchical grouping results, generate a three-dimensional simulation model with storage site type identifier. Step S3: Collect photoelectric signals, barcode data and RFID asset tag information, perform consistency verification on the photoelectric signals, barcode data and RFID asset tag information, filter to obtain valid storage location status data, and update the display status of the storage location in the three-dimensional simulation model according to the storage location status data; Step S4: Construct a reservoir relationship map, extract the current state and historical access records of the reservoir as map node features, calculate the map node features through a graph convolutional network to obtain the reservoir access prediction results, and generate a visualized thermal distribution in the three-dimensional simulation model based on the access prediction results. Step S5: Obtain the asset storage time and remaining time until expiration for the occupied storage space. Calculate the overdue risk value based on the comparison between the storage time and historical periods. Mark the storage space with the overdue risk value exceeding the set value as a warning state in the three-dimensional simulation model.

2. The inventory simulation display method for intelligent turnover cabinets according to claim 1, characterized in that, Step S1 includes: The RFID reading and writing module sends radio frequency signals to the self-testing tags in each storage location, receives the return signals from each self-testing tag through multiple antennas, and records the signal strength values ​​received by each antenna at different transmission powers. The signal strength value is associated with the corresponding transmit power and antenna number. Based on the signal propagation attenuation model in free space, a set of equations corresponding to the signal strength and the distance from the tag to the antenna are established. The preset spatial coordinates of each antenna are extracted as known parameters. The signal strength value is substituted into the corresponding set of equations, and the three-dimensional spatial coordinates of each storage tag are obtained by solving the least squares method. The three-dimensional spatial coordinates of each storage location tag are collected to construct a database of the mapping relationship between storage location codes and spatial coordinates, thereby obtaining the set of three-dimensional coordinates of the storage location.

3. The inventory simulation display method for intelligent turnover cabinets according to claim 1, characterized in that, Step S2 includes: Extract the vertical coordinate components of each storage location coordinate from the three-dimensional coordinate set of the storage location, perform K-means clustering on the vertical coordinate components, and obtain the hierarchical identifier of each storage location. The number of storage locations corresponding to each level of identifier is counted, the horizontal coordinate spacing between adjacent storage locations within the same level is calculated, and the storage location type identifier for each level is determined based on the matching relationship between the number of storage locations and the coordinate spacing and the preset storage location type rules. Calculate the maximum and minimum values ​​of all storage location coordinates in the three-dimensional coordinate set of the storage location in each coordinate axis direction, and obtain the cabinet outer shape envelope dimension parameters by superimposing the maximum and minimum values ​​with a preset edge allowance; The graphics rendering engine is invoked to generate a cabinet shell model based on the cabinet's outer shape envelope size parameters. A shelf model is generated based on the vertical coordinates of each level and the storage location type identifier. A storage location unit model is generated based on the storage location coordinates in the storage location three-dimensional coordinate set. The cabinet shell model, shelf model, and storage location unit model are combined to obtain the three-dimensional simulation model.

4. The inventory simulation display method for intelligent turnover cabinets according to claim 1, characterized in that, Step S3 includes: The photoelectric sensor output level signal of each storage location is collected as the photoelectric signal, the asset barcode string is obtained through the barcode scanning module as the barcode data, and the EPC code of the asset tag is obtained through the RFID reading and writing module as the RFID asset tag information. Determine whether the storage location occupancy status indicated by the photoelectric signal is consistent with the existence status of the barcode data and the RFID asset tag information. Mark the storage location record that matches at least two sensor data states in the consistency judgment result as a valid record to obtain the valid storage location status data. Extract the storage location code and asset occupancy identifier from the valid storage location status data. Set the corresponding storage location unit model in the three-dimensional simulation model of the storage location with an empty asset occupancy identifier to a semi-transparent material. Set the corresponding storage location unit model in the three-dimensional simulation model of the storage location with an asset occupancy identifier to an opaque material. If there are conflicting sensor data records in the valid storage space status data, the corresponding storage space unit model in the three-dimensional simulation model of the conflicting storage space is set to flashing red material, thus completing the update of the display status of the storage space in the three-dimensional simulation model.

5. The inventory simulation display method for intelligent turnover cabinets according to claim 1, characterized in that, Step S4 includes: Each storage location in the turnover cabinet is abstracted as a graph node. Connection edges between nodes are established based on the spatial distance relationship between storage locations and the historical business relationship relationship to obtain the storage location relationship graph. Extract the current occupancy status, outbound frequency, inbound frequency, and remaining days for verification of each storage location, and combine them to form a node feature vector. Input the node feature vector and the adjacency matrix of the storage location relationship graph into a graph convolutional network for convolution operation to obtain the access probability value of each storage location. Based on the access probability value, the corresponding storage unit model of each storage location in the three-dimensional simulation model is set to a gradient color from blue to red. Storage locations with high access probability values ​​are rendered as red, and storage locations with low access probability values ​​are rendered as blue, thereby obtaining the visualized thermal distribution. Storage locations whose access probability values ​​exceed a preset probability threshold and are currently empty are selected as recommended storage locations. A green halo effect is then overlaid on the corresponding storage location unit model in the three-dimensional simulation model to complete the intelligent warehousing guidance.

6. The inventory simulation display method for intelligent turnover cabinets according to claim 5, characterized in that, The step of inputting the node feature vectors and the adjacency matrix of the storage location relation graph into a graph convolutional network for convolution operation to obtain the access probability value of each storage location includes: The node feature vectors are arranged in the order of storage locations to construct a node feature matrix. An identity matrix is ​​added to the adjacency matrix of the storage location relationship graph to achieve self-connection. The degree value of each node is calculated to form a degree matrix. The normalized adjacency matrix is ​​obtained by performing matrix multiplication operation with the adjacency matrix based on the negative 1 / 2 power of the degree matrix. The node feature matrix is ​​multiplied by the normalized adjacency matrix, and then multiplied by the weight matrix of the first convolutional layer. Negative values ​​are filtered out by the ReLU activation function to obtain the first intermediate feature matrix containing the features of the storage space neighborhood. The first layer intermediate feature matrix is ​​multiplied by the normalized adjacency matrix, then multiplied by the weight matrix of the second convolutional layer and activated by ReLU to obtain the second layer intermediate feature matrix containing the storage location business-related features. The second layer intermediate feature matrix is ​​then subjected to the above convolution operation through the third convolutional layer to obtain the target feature matrix that integrates both spatial and business features. Extract the feature vectors of each storage location corresponding to each row in the target feature matrix, perform Softmax normalization on each feature vector, and use the normalized values ​​as the probability prediction values ​​of the corresponding storage location being accessed in the future time window to obtain the access probability values ​​of each storage location.

7. The inventory simulation display method for intelligent turnover cabinets according to claim 1, characterized in that, Step S5 includes: Extract the entry timestamp and verification expiration timestamp of the assets in the occupied storage space from the asset archive database. Calculate the asset entry duration based on the time difference between the current timestamp and the entry timestamp. Calculate the remaining expiration duration based on the time difference between the verification expiration timestamp and the current timestamp. Query historical outbound records of the same type of assets, calculate the actual inventory cycle of each historical record, and obtain the average inventory cycle and the standard deviation of the inventory cycle. Divide the difference between the asset's storage time and the average inventory cycle by the standard deviation of the inventory cycle, and substitute the calculation result into the normal distribution cumulative probability function to obtain the overdue risk probability value of the corresponding asset. If the probability value of the overdue risk exceeds a preset risk threshold or the remaining time until expiration is less than a preset number of days threshold, the corresponding storage unit model in the three-dimensional simulation model that meets either condition is set to an orange or red warning material to complete the warning status marking.

8. An intelligent inventory simulation display system for turnover cabinets, characterized in that, The method for implementing the inventory simulation display of the intelligent turnover cabinet as described in any one of claims 1-7, wherein the inventory simulation display system of the intelligent turnover cabinet comprises: The acquisition module is used to collect signal strength data of RFID self-inspection tags in each storage location in the turnover cabinet under multiple antennas. Based on the distance attenuation relationship between the signal strength data and the antenna position, the three-dimensional coordinate set of the storage location is calculated. The clustering module is used to cluster the vertical coordinates of the three-dimensional coordinate set of the storage sites, identify the hierarchical grouping results, and generate a three-dimensional simulation model with storage site type identifier based on the number and spacing characteristics of each layer of the hierarchical grouping results. The verification module is used to collect photoelectric signals, barcode data and RFID asset tag information, perform consistency verification on the photoelectric signals, barcode data and RFID asset tag information, filter to obtain valid storage location status data, and update the display status of the storage location in the three-dimensional simulation model according to the storage location status data. The calculation module is used to construct a reservoir relationship map, extract the current state and historical access records of the reservoirs as map node features, calculate the map node features through a graph convolutional network to obtain the reservoir access prediction results, and generate a visualized thermal distribution in the three-dimensional simulation model based on the access prediction results. The marking module is used to obtain the asset storage time and remaining time until expiration of occupied storage spaces, calculate the overdue risk value based on the comparison between the storage time and historical periods, and mark the storage spaces with the overdue risk value exceeding the set value as warning status in the three-dimensional simulation model.

9. An intelligent inventory simulation display device for turnover cabinets, characterized in that, It includes a memory and a processor, the memory storing a computer program that can run on the processor, and the processor executing the computer program to implement the inventory simulation display method of the intelligent turnover cabinet according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is run by the processor, it causes the processor to execute the inventory simulation display method of the intelligent turnover cabinet as described in any one of claims 1 to 7.

Images