A data warehouse management system based on the Internet of Things

Abstract

The present invention discloses a data warehouse management system based on the Internet of Things, including a three-dimensional sensing array module, a digital twin engine module, an intelligent decision module and a knowledge graph module, wherein the output end of the three-dimensional sensing array module is connected to the input end of the digital twin engine module, the output end of the digital twin engine module is connected to the input end of the intelligent decision module, the output end of the intelligent decision module is connected to the input end of a mobile robot cluster, and the output end of the knowledge graph module is connected to the input end of the intelligent decision module. This system realizes closed-loop management through a five-layer architecture of perception, modeling, decision-making, execution and optimization, and realizes a deep combination of multi-physical field coupling simulation and multi-objective dynamic optimization. Compared with traditional solutions, the path qualification rate is improved, the warning response speed is accelerated, and the comprehensive energy consumption is reduced, providing a high-precision and highly adaptable management paradigm for modern intelligent warehouses.

Description

Technical Field

[0001] The present invention relates to the technical field of management systems, and in particular to a data warehouse management system based on the Internet of Things. Background Art

[0002] With the widespread use of Internet of Things technology in the field of warehouse management, the data monitoring and matching of warehouse environment and stored goods are becoming more and more scientific. The adaptability analysis of warehouse environment and stored goods based on multi-source data can avoid the impact of warehouse environment on the quality of goods during storage. Storage environment monitoring of common moisture-prone goods in warehouses is a common warehouse data monitoring field at this stage. If the storage environment of moisture-prone goods cannot be monitored in time, it will cause a decline in product quality and affect its use.

[0003] Traditional warehouse management systems lack accuracy and real-time performance in environmental monitoring: sensor coverage density is low, there are monitoring blind spots, single-modal data collection (such as temperature or humidity only) leads to one-sided environmental assessment, and data update delays are high (usually >1 minute), making it difficult to respond to changes in a timely manner.

[0004] Moreover, the safety and efficiency of path planning are in conflict: the RRT algorithm has a large path curvature (sharp turns with R < 0.5m often occur), the dynamic obstacle avoidance response is slow (> 2 seconds), and multi-robot collaboration is prone to deadlock. Summary of the invention

[0005] The purpose of the present invention is to provide a data warehouse management system based on the Internet of Things to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solution: a data warehouse management system based on the Internet of Things, comprising:

[0007] The three-dimensional sensing array module includes uniformly deployed millimeter-wave radar nodes, infrared thermal imaging arrays, and ultrasonic barometers to form a 5m×5m×3m honeycomb monitoring grid. The digital twin engine module includes: (a) a real-time CFD simulation unit built on an improved OpenFOAM architecture that uses the vorticity-velocity method to solve the Navier-Stokes equations:

[0008]

[0009] Where ω is the vorticity field, u is the velocity field, and ν is the kinematic viscosity coefficient; (b) Vortex intensity calculation unit, calculates the regional vortex intensity according to formula (1):

[0010] (1)

[0011] When Γ≥0.35 When it is marked as a strong disturbance area; intelligent decision-making module, including:

[0012] (c) Multi-objective optimization unit, which establishes a loss function that includes energy consumption, product shelf life, and environmental compliance rate:

[0013]

[0014] Among them, α, β, and γ are dynamically adjusted through Q-learning, E-energy consumption, T-product shelf life, and H-environmental compliance rate; (d) Mobile robot cluster, equipped with a 6-DOF robotic arm and a micro-turbofan array, realizes multi-machine collaboration based on the auction algorithm, and the mobile robot is equipped with a UWB positioning module and an anti-collision sensor. The navigation error in the dense shelf area is ≤3cm; the knowledge graph module builds an environmental sensitivity database containing 1200+ SKUs and defines the environmental tolerance index of the goods:

[0015] Among them, Δt is exposure time, τ is tolerance coefficient, Γ is vortex intensity, T is temperature, and when the S value decreases at a rate exceeding 15% / week, a level 3 warning is triggered;

[0016] The output end of the three-dimensional sensing array module is connected to the input end of the digital twin engine module, the output end of the digital twin engine module is connected to the input end of the intelligent decision-making module, the output end of the intelligent decision-making module is connected to the input end of the mobile robot cluster, and the output end of the knowledge graph module is connected to the input end of the intelligent decision-making module.

[0017] The deployment method of the three-dimensional sensing array module includes: setting reference nodes along the warehouse columns with a spacing of ≤8m, the vertical spacing of the column nodes is ≤3m, and the horizontal spacing is 6.8m; deploying supplementary nodes in the shelf interval layer with a vertical spacing of ≤2.5m; all nodes are networked using the LoRaWAN protocol, and the data transmission delay is ≤50ms.

[0018] The operation method of the real-time CFD simulation unit includes: using adaptive meshing technology, the mesh density of the key area reaches 200 nodes / m³; implementing mixed precision calculation: vortex intensity Γ≥0.2 The area uses FP32 precision, and the rest of the areas use FP16 precision; the full-field fluid dynamics model is updated every 30 seconds.

[0019] The control method of the mobile robot cluster includes: establishing a three-dimensional SLAM map with a positioning accuracy of ±2cm; the turbofan array supports 8 levels of wind speed adjustment (0.5-8m / s); the power density of the semiconductor dehumidification module reaches 3W / cm³; and the multi-machine task allocation response time is ≤500ms.

[0020] The method for constructing the knowledge graph module includes: using a graph neural network to analyze the spatiotemporal correlation between environmental parameters and product degradation; and using contrastive learning to generate new SKU initial parameters:

[0021] =

[0022] Among them, q is the new SKU query vector, k⁺ is the positive sample key vector, k⁻ is the negative sample key vector, is the cosine similarity function, is the temperature coefficient, N is the number of negative samples, the number of negative samples N≥1024, and the temperature coefficient τ∈[0.05, 0.2];

[0023] Deploy federated learning nodes to achieve cross-warehouse knowledge sharing. The federated learning nodes use a gradient leakage-proof parameter aggregation method.

[0024] The federated learning node parameter aggregation satisfies:

[0025]

[0026] in, is the smoothing constant, is the aggregation weight vector, is the local weight of the i-th node,

[0027] is the local parameter variance, ∈[1 ].

[0028] It also includes a dynamic baseline calibration module, the output end of which is connected to the input end of the intelligent decision-making module, and the dynamic baseline calibration module establishes an environmental parameter coupling model through Kalman filtering:

[0029]

[0030] Where A is the environmental parameter transfer matrix, B is the equipment control coefficient matrix, The dynamic baseline calibration module is operated at a frequency of one full-bin calibration per minute, and the abnormal parameter fluctuation threshold is set to ±2σ, where σ is the standard deviation of the parameter in the past 24 hours, and the σ value is calculated according to the formula:

[0031]

[0032] N is the 1440 data points collected in the last 24 hours.

[0033] It also includes a visual warning interface, which displays in real time the vortex intensity heat map, the risk level distribution of goods, the working status of the robot cluster and the real-time monitoring curve of energy consumption. The visual warning interface is based on the WebGL engine, and the color level mapping cycle is ≤1 second.

[0034] The mobile robot uses the RRT* algorithm to generate a collision-free path. When the RRT* algorithm is running, the following conditions are met: the sampling frequency is ≥ 1000 times / second, the number of path optimization iterations is ≥ 50 times, and the following steps are performed: constructing a three-dimensional occupancy grid based on the SLAM map;

[0035] Set the target point attraction function , is the size of attraction, is the attraction coefficient, q is the current posture, is the target point pose, the attraction coefficient Based on real-time path curvature Dynamic adjustment to meet:

[0036]

[0037] A bidirectional search strategy is used to connect paths when the distance between two trees is less than 0.5m;

[0038] During the path optimization phase, redundant nodes are deleted to ensure that the path curvature radius is ≥ 0.8 m.

[0039] It also includes a self-learning module, the output end of which is respectively connected to the three-dimensional perception array module, the digital twin engine module, the dynamic baseline calibration module, the intelligent decision-making module, the knowledge graph module, the mobile robot cluster and the visual warning interface. The self-learning module performs global parameter optimization once every 24 hours, including: Q-learning strategy update, CFD model parameter calibration and knowledge graph relationship weight adjustment.

[0040] When the system is initialized, the following operations are performed: establishing a three-dimensional reference coordinate system, deploying perception nodes, and training a federated learning model, specifically including: establishing a three-dimensional reference coordinate system through a laser scanner, with the origin set at the geometric center of the warehouse; deploying reference nodes every 8 meters along the columns, and deploying supplementary nodes at 2 meters of shelf height; and using Dirichlet distribution to allocate parameters in the pre-training stage of the federated learning model:

[0041] P( )=

[0042] The concentration parameter of the Dirichlet distribution is α∈(0.3,0.7], preferably α=0.5, K is the total number of categories, is a K-dimensional probability vector.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] This system achieves closed-loop management through a five-layer architecture of perception, modeling, decision-making, execution, and optimization, and realizes a deep combination of multi-physical field coupling simulation and multi-objective dynamic optimization. Compared with traditional solutions, the path qualification rate is improved, the warning response speed is accelerated, and the overall energy consumption is reduced, providing a high-precision and highly adaptable management paradigm for modern smart warehouses. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 is a system block diagram of the present invention;

[0046] Figure 2 It is a schematic diagram of the process of the present invention;

[0047] Figure 3 A schematic diagram of the deployment of a three-dimensional sensing array module of the present invention;

[0048] Figure 4 Schematic diagram of the construction method of the knowledge graph module of the present invention;

[0049] Figure 5 This is a step diagram of the mobile robot algorithm of the present invention. DETAILED DESCRIPTION

[0050] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0051] See also Figure 1-5 The present invention provides a technical solution: a data warehouse management system based on the Internet of Things, comprising:

[0052] The three-dimensional sensing array module includes uniformly deployed millimeter-wave radar nodes, infrared thermal imaging arrays, and ultrasonic barometers to form a 5m×5m×3m honeycomb monitoring grid. The digital twin engine module includes: (a) a real-time CFD simulation unit built on an improved OpenFOAM architecture that uses the vorticity-velocity method to solve the Navier-Stokes equations:

[0053]

[0054] Where ω is the vorticity field, u is the velocity field, and ν is the kinematic viscosity coefficient; (b) Vortex intensity calculation unit, calculates the regional vortex intensity according to formula (1):

[0055] (1)

[0056] When Γ≥0.35 When it is marked as a strong disturbance area; intelligent decision-making module, including:

[0057] (c) Multi-objective optimization unit, which establishes a loss function that includes energy consumption, product shelf life, and environmental compliance rate:

[0058]

[0059] Among them, α, β, and γ are dynamically adjusted through Q-learning, E-energy consumption, T-product shelf life, and H-environmental compliance rate; (d) Mobile robot cluster, equipped with a 6-DOF robotic arm and a micro-turbofan array, realizes multi-machine collaboration based on the auction algorithm, and the mobile robot is equipped with a UWB positioning module and an anti-collision sensor. The navigation error in the dense shelf area is ≤3cm; the knowledge graph module builds an environmental sensitivity database containing 1200+ SKUs and defines the environmental tolerance index of the goods:

[0060]

[0061] Among them, Δt is exposure time, τ is tolerance coefficient, Γ is vortex intensity, and T is temperature. When the S value decrease rate exceeds 15% / week, a level 3 warning is triggered.

[0062] The deployment method of the three-dimensional sensing array module includes: setting reference nodes along the warehouse columns with a spacing of ≤8m, the vertical spacing of the column nodes is ≤3m, and the horizontal spacing is 6.8m; deploying supplementary nodes in the shelf interval layer with a vertical spacing of ≤2.5m; all nodes are networked using the LoRaWAN protocol, and the data transmission delay is ≤50ms.

[0063] The operation method of the real-time CFD simulation unit includes: using adaptive meshing technology, the mesh density of the key area reaches 200 nodes / m³; implementing mixed precision calculation: vortex intensity Γ≥0.2 The area uses FP32 precision, and the rest of the areas use FP16 precision; the full-field fluid dynamics model is updated every 30 seconds.

[0064] The control method of the mobile robot cluster includes: establishing a three-dimensional SLAM map with a positioning accuracy of ±2cm; the turbofan array supports 8 levels of wind speed adjustment (0.5-8m / s); the power density of the semiconductor dehumidification module reaches 3W / cm³; and the multi-machine task allocation response time is ≤500ms.

[0065] The method for constructing the knowledge graph module includes: using a graph neural network to analyze the spatiotemporal correlation between environmental parameters and product degradation; and using contrastive learning to generate new SKU initial parameters:

[0066] =

[0067] Among them, q is the new SKU query vector, k⁺ is the positive sample key vector, k⁻ is the negative sample key vector, is the cosine similarity function, is the temperature coefficient, N is the number of negative samples, the number of negative samples N≥1024, and the temperature coefficient τ∈[0.05, 0.2];

[0068] Deploy a federated learning node to achieve cross-warehouse knowledge sharing. The federated learning node adopts a gradient leakage-proof parameter aggregation method. The federated learning node parameter aggregation satisfies:

[0069]

[0070] in, is the smoothing constant, is the aggregation weight vector, is the local weight of the i-th node,

[0071] is the local parameter variance, ∈[1 ].

[0072] It also includes a dynamic baseline calibration module, which establishes an environmental parameter coupling model through Kalman filtering:

[0073]

[0074] Where A is the environmental parameter transfer matrix, B is the equipment control coefficient matrix, The dynamic baseline calibration module is operated at a frequency of one full-bin calibration per minute, and the abnormal parameter fluctuation threshold is set to ±2σ, where σ is the standard deviation of the parameter in the past 24 hours, and the σ value is calculated according to the formula:

[0075]

[0076] N is the 1440 data points collected in the last 24 hours.

[0077] It also includes a visual warning interface, which displays in real time the vortex intensity heat map, the risk level distribution of goods, the working status of the robot cluster and the real-time monitoring curve of energy consumption. The visual warning interface is based on the WebGL engine, and the color level mapping cycle is ≤1 second.

[0078] The mobile robot uses the RRT* algorithm to generate a collision-free path. When the RRT* algorithm is running, the following conditions are met: the sampling frequency is ≥ 1000 times / second, the number of path optimization iterations is ≥ 50 times, and the following steps are performed: constructing a three-dimensional occupancy grid based on the SLAM map;

[0079] Set the target point attraction function , is the size of attraction, is the attraction coefficient, q is the current posture, is the target point pose, when When >1.5, the path oscillation rate increases by 37%, so the limit ∈[0.5,1.2];

[0080] A bidirectional search strategy is used to connect paths when the distance between two trees is less than 0.5m;

[0081] During the path optimization phase, redundant nodes are deleted to ensure that the path curvature radius is ≥ 0.8 m.

[0082] The curvature-constrained implementation of the RRT* algorithm includes the following steps:

[0083] 1. Path point curvature detection: Perform cubic spline interpolation on the candidate path segments and calculate the curvature κ of each point:

[0084]

[0085] When κ>1.25 is detected (i.e. R<0.8m), the correction mechanism is triggered, and the curvature threshold R≥0.8m is the minimum requirement. In this embodiment, R≥0.83m is used to reserve a safety margin;

[0086] The following table shows the path curvature compliance rate:

[0087]

[0088] 2. Node optimization strategy:

[0089] Insert auxiliary nodes: insert an optimization node before and after the point where the curvature exceeds the standard;

[0090] B-spline smoothing: 5th-order B-spline is used to reparameterize the path segments;

[0091] Curvature verification: recalculate the κ value of the optimized path;

[0092] 3. Dynamic parameter adjustment:

[0093]

[0094] It also includes a self-learning module, which performs global parameter optimization every 24 hours, including: Q-learning strategy update, CFD model parameter calibration, and knowledge graph relationship weight adjustment;

[0095] The following table shows the interaction relationship between the core modules:

[0096]

[0097] When the system is initialized, the following operations are performed: establishing a three-dimensional reference coordinate system, deploying sensing nodes, and training a federated learning model. The system initialization method includes: establishing a three-dimensional reference coordinate system through a laser scanner, with the origin set at the geometric center of the warehouse; deploying reference nodes every 8 meters along the columns, and deploying supplementary nodes at 2 meters of shelf height; and using Dirichlet distribution to allocate parameters in the pre-training stage of the federated learning model:

[0098] P( )=

[0099] The concentration parameter of the Dirichlet distribution is α∈(0.3,0.7], preferably α=0.5, K is the total number of categories, is a K-dimensional probability vector.

[0100] Example 1: Pharmaceutical cold chain warehouse application

[0101] Step 1: System deployment

[0102] Deploy a 3D sensing array in a biopharmaceutical warehouse (1200m², 6m high):

[0103] Millimeter-wave radar nodes: 48 nodes are installed along the columns at intervals of 6.8 m;

[0104] Infrared array: Each shelf is 2.1m apart, with a total of 256 temperature measurement points deployed;

[0105] Ultrasonic barometer: honeycomb grid 5m×5m×3m, accuracy ±1Pa;

[0106] Step 2: Digital twin modeling uses the improved OpenFOAM v2212 to build the CFD model. The improved OpenFOAM architecture includes the following innovative modules:

[0107] Adaptive grid optimizer: dynamically adjusts the grid density according to the vortex strength Γ, when Γ ≥ 0.2 Local encryption is triggered when

[0108] Mixed precision scheduler: establishes a dynamic allocation strategy for FP32 and FP16 computing resources, and controls the computing error within 1 within;

[0109] Vortex tracking module: real-time recording of strong disturbance areas (Γ≥0.35 )'s spatiotemporal evolution path parameter configuration:

[0110] Key area grid density: 218 nodes / m³;

[0111] Mixed precision strategy: Γ ≥ 0.2 FP32 is enabled in the region (23% of the region);

[0112] Step 3: Federated learning initializes Dirichlet distribution parameters in the pre-training phase:

[0113] Concentration parameter α=0.5

[0114] Client data distribution: θ~Dir(0.5), K=1200

[0115] Aggregate formula:

[0116]

[0117] Example 2: Building materials storage optimization

[0118] Path planning parameter configuration:

[0119] RRT* algorithm: sampling rate 1200 times / second

[0120] Curvature constraint: R ≥ 0.83m

[0121] Node expansion strategy:

[0122] 5. Node connection: If the collision detection is passed, q_new is added to the path tree;

[0123] The implementation of the node expansion strategy includes the following key parameters:

[0124] Random sampling probability distribution: the target direction probability is increased to 70%;

[0125] Adaptive step length adjustment: When approaching an obstacle, the step length is reduced to 0.6m;

[0126] Safety margin: The path is considered safe when the distance between the path and the obstacle is ≥ 0.35m;

[0127] Dynamic baseline calibration parameters:

[0128] Kalman filter period: 60 seconds

[0129] Abnormal fluctuation threshold: ±2σ (σ is calculated based on 1440 points)

[0130] Parameter recovery time: <5 seconds (compared to 18 seconds for traditional methods).

[0131] The following table shows the attraction coefficient performance data:

[0132]

[0133] The following table shows the comparison data of path qualification rate and energy consumption:

[0134]

[0135] The overall process of the present invention is as follows:

[0136] 1. Data collection: Real-time acquisition of environmental data through a three-dimensional sensing array.

[0137] 2. Modeling and simulation: The digital twin engine processes data, builds and updates CFD models, and identifies high-risk areas.

[0138] 3. Intelligent decision-making: Combine knowledge graphs and optimization algorithms to generate warehousing and routing strategies.

[0139] 4. Task execution: The mobile robot performs the transport according to the planned path and makes real-time adjustments to avoid obstacles.

[0140] 5. Dynamic optimization: Continuously improve system performance through calibration and self-learning modules.

[0141] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

Claims

1. A data warehouse management system based on the Internet of Things, characterized in that: include: The three-dimensional sensing array module includes evenly deployed millimeter-wave radar nodes, infrared thermal imaging arrays, and ultrasonic barometers, forming a 5m×5m×3m honeycomb monitoring grid; Digital twin engine module, including: (a) Real-time CFD simulation unit. The digital twin modeling uses the improved OpenFOAM v2212 to build the CFD model. The improved OpenFOAM architecture includes the following innovative modules: Adaptive grid optimizer: dynamically adjusts the grid density according to the vortex strength Γ, when Γ ≥ 0.2 Local encryption is triggered when Mixed precision scheduler: establishes a dynamic allocation strategy for FP32 and FP16 computing resources, and controls the computing error within 1 within; Vortex tracking module: real-time recording of strong disturbance areas Γ≥0.35 Parameter configuration of the spatiotemporal evolution path: Key area grid density: 218 nodes / m³; Mixed precision strategy: Γ ≥ 0.2 FP32 is enabled in the region: 23% of the region, and the vorticity-velocity method is used to solve the Navier-Stokes equations: Where ω is the vorticity field, u is the velocity field, and ν is the kinematic viscosity coefficient; (b) Vortex intensity calculation unit, calculates the regional vortex intensity according to formula (1): （1） When Γ≥0.35 When it is marked as a strong disturbance area; intelligent decision-making module, including: (c) Multi-objective optimization unit, which establishes a loss function that includes energy consumption, product shelf life, and environmental compliance rate: Among them, α, β, and γ are dynamically adjusted through Q-learning, E-energy consumption, T-product shelf life, and H-environmental compliance rate; (d) Mobile robot cluster, equipped with a 6-DOF robotic arm and a micro-turbofan array, realizes multi-machine collaboration based on the auction algorithm, and the mobile robot is equipped with a UWB positioning module and an anti-collision sensor. The navigation error in the dense shelf area is ≤3cm; the knowledge graph module builds an environmental sensitivity database containing 1200+ SKUs and defines the environmental tolerance index of the goods: Among them, Δt is exposure time, τ is tolerance coefficient, Γ is vortex intensity, T is temperature, and when the S value decreases at a rate exceeding 15% / week, a level 3 warning is triggered; The output end of the three-dimensional sensing array module is connected to the input end of the digital twin engine module, the output end of the digital twin engine module is connected to the input end of the intelligent decision-making module, the output end of the intelligent decision-making module is connected to the input end of the mobile robot cluster, and the output end of the knowledge graph module is connected to the input end of the intelligent decision-making module.

2. According to claim 1, a data warehouse management system based on the Internet of Things is characterized in that: The deployment method of the three-dimensional sensing array module includes: setting reference nodes along the warehouse columns with a spacing of ≤8m, the vertical spacing of the column nodes is ≤3m, and the horizontal spacing is 6.8m; deploying supplementary nodes in the shelf interval layer with a vertical spacing of ≤2.5m; all nodes are networked using the LoRaWAN protocol, and the data transmission delay is ≤50ms.

3. According to claim 1, a data warehouse management system based on the Internet of Things is characterized in that: The operation method of the real-time CFD simulation unit includes: using adaptive meshing technology, the mesh density of the key area reaches 200 nodes / m³; implementing mixed precision calculation: vortex intensity Γ≥0.2 The area uses FP32 precision, and the rest of the areas use FP16 precision; the full-field fluid dynamics model is updated every 30 seconds.

4. The data warehouse management system based on the Internet of Things according to claim 1 is characterized in that: The control method of the mobile robot cluster includes: establishing a three-dimensional SLAM map with a positioning accuracy of ±2cm; the turbofan array supports 8-speed wind speed adjustment of 0.5-8m / s; the power density of the semiconductor dehumidification module reaches 3W / cm³; and the multi-machine task allocation response time is ≤500ms.

5. The data warehouse management system based on the Internet of Things according to claim 1 is characterized in that: The method for constructing the knowledge graph module includes: using a graph neural network to analyze the spatiotemporal correlation between environmental parameters and product degradation; and using contrastive learning to generate new SKU initial parameters: = Among them, q is the new SKU query vector, k⁺ is the positive sample key vector, k⁻ is the negative sample key vector, is the cosine similarity function, is the temperature coefficient, N is the number of negative samples, the number of negative samples N≥1024, and the temperature coefficient τ∈[0.05, 0.2]; Deploy a federated learning node to achieve cross-warehouse knowledge sharing. The federated learning node adopts a gradient leakage-proof parameter aggregation method. The federated learning node parameter aggregation satisfies: in, is the smoothing constant, is the aggregation weight vector, is the local weight of the i-th node; is the local parameter variance, ∈[1 ].

6. The data warehouse management system based on the Internet of Things according to claim 1 is characterized in that: It also includes a dynamic baseline calibration module, the output end of which is connected to the input end of the intelligent decision-making module, and the dynamic baseline calibration module establishes an environmental parameter coupling model through Kalman filtering: Where A is the environmental parameter transfer matrix, B is the equipment control coefficient matrix, The dynamic baseline calibration module is operated at a frequency of one full-bin calibration per minute, and the abnormal parameter fluctuation threshold is set to ±2σ, where σ is the standard deviation of the parameter in the past 24 hours, and the σ value is calculated according to the formula: N is the 1440 data points collected in the last 24 hours.

7. The data warehouse management system based on the Internet of Things according to claim 1 is characterized in that: It also includes a visual warning interface, which displays in real time the vortex intensity heat map, the risk level distribution of goods, the working status of the robot cluster and the real-time monitoring curve of energy consumption. The visual warning interface is based on the WebGL engine, and the color level mapping cycle is ≤1 second.

8. The data warehouse management system based on the Internet of Things according to claim 1 is characterized in that: The mobile robot uses the RRT* algorithm to generate a collision-free path. When the RRT* algorithm is running, the following conditions are met: the sampling frequency is ≥ 1000 times / second, the number of path optimization iterations is ≥ 50 times, and the following steps are performed: constructing a three-dimensional occupancy grid based on the SLAM map; Set the target point attraction function , is the size of attraction, is the attraction coefficient, q is the current posture, is the target point pose, the attraction coefficient Based on real-time path curvature Dynamic adjustment to meet: A bidirectional search strategy is used to connect paths when the distance between two trees is less than 0.5m; During the path optimization phase, redundant nodes are deleted to ensure that the path curvature radius is ≥ 0.8 m.

9. The data warehouse management system based on the Internet of Things according to claim 1 is characterized in that: It also includes a self-learning module, the output end of which is respectively connected to the three-dimensional perception array module, the digital twin engine module, the dynamic baseline calibration module, the intelligent decision-making module, the knowledge graph module, the mobile robot cluster and the visual warning interface. The self-learning module performs global parameter optimization once every 24 hours, including: Q-learning strategy update, CFD model parameter calibration and knowledge graph relationship weight adjustment.

10. The data warehouse management system based on the Internet of Things according to claim 1 is characterized in that: When the system is initialized, the following operations are performed: establishing a three-dimensional reference coordinate system, deploying perception nodes, and training a federated learning model, specifically including: establishing a three-dimensional reference coordinate system through a laser scanner, with the origin set at the geometric center of the warehouse; deploying reference nodes every 8 meters along the columns, and deploying supplementary nodes at 2 meters of shelf height; and using Dirichlet distribution to allocate parameters in the pre-training stage of the federated learning model: P( )= The concentration parameter of the Dirichlet distribution is α∈(0.3,0.7], preferably α=0.5, K is the total number of categories, is a K-dimensional probability vector.

Images