Warehousing intelligent monitoring system based on digital twinning

Abstract

The invention relates to the technical field of intelligent warehousing monitoring, and discloses an intelligent warehousing monitoring system based on digital twinning, which comprises a data acquisition layer, a digital twinning modeling layer, an intelligent analysis layer and an execution control layer, and is characterized in that three-dimensional coordinates and temperature distribution data of goods are acquired through a laser radar array and an infrared thermal imager; constructing a dynamic point cloud data set in combination with RFID positioning; an improved YOLOv7 model is adopted to be embedded into a CBAM attention mechanism to recognize the cargo form risk, and an LSTM time sequence model is fused to predict a temperature anomaly trajectory; an AGV obstacle avoidance path is optimized based on a genetic-ant colony hybrid algorithm, and a temperature control system and a security device are driven to perform linkage execution through an OPC UA protocol. And finally, millisecond-level synchronous mapping of the physical warehouse and the digital twinborn body, real-time deviation correction regulation and control of an equipment running track and cross-system collaborative response of an emergency strategy are realized, and the safety early warning accuracy and dynamic protection robustness of the storage environment are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of warehouse intelligent monitoring, in particular to a warehouse intelligent monitoring system based on digital twinning. BACKGROUND

[0002] Warehouse intelligent monitoring technology is a comprehensive system that integrates Internet of Things, artificial intelligence, sensor networks and data analysis, aiming to realize real-time perception, intelligent analysis and automatic response of warehouse environment, cargo status and operation process. Its core lies in improving the accuracy, safety and efficiency of warehouse management through multi-dimensional data collection and intelligent decision-making. The warehouse intelligent monitoring system based on digital twinning is a comprehensive system integrating Internet of Things perception, three-dimensional modeling and intelligent decision-making, aiming to improve the safety control efficiency of warehouse environment and the precision of equipment collaboration. It is one of the core technical architectures for real-time dynamic monitoring of the system. The system builds a virtual mapping of the physical warehouse through a multi-source sensor group and realizes autonomous risk early warning by combining artificial intelligence algorithms, providing full-link management support for modern smart warehouses.

[0003] Currently, since warehouse operation involves a large number of dynamic change elements, the traditional monitoring system relied upon cannot synchronize the difference state of physical space and virtual model in real time when monitoring the safety of warehouse environment. If the stacking form of goods mutates and the temperature fluctuates abnormally, it may cause risk early warning lag, seriously reducing the response speed of safety hazards.

[0004] Therefore, the present application provides a warehouse intelligent monitoring system based on digital twinning to solve the above problems. SUMMARY

[0005] (I) Technical problems solved In view of the deficiencies of the prior art, the present application provides a warehouse intelligent monitoring system based on digital twinning to solve the problems raised in the background art.

[0006] (II) Technical solutions To achieve the above purpose, the present application provides the following technical solutions: a warehouse intelligent monitoring system based on digital twinning, comprising the following steps: S1, collecting warehouse environment dynamic monitoring data and warehouse equipment static parameter data through a multi-source sensor device group; S2, based on the warehouse environment dynamic monitoring data and warehouse equipment static parameter data, constructing a three-dimensional visual warehouse digital twinning body model, and generating warehouse environment twin point cloud data set; S3, performing multi-dimensional feature extraction on the warehouse environment twin point cloud data set to generate an environment feature vector matrix; S4, inputting the environment feature vector matrix into a pre-trained abnormal behavior detection model for real-time risk analysis and outputting a warehouse safety state evaluation report; S5, when a security risk is detected, a path optimization algorithm is called to generate device obstacle avoidance instruction set and emergency response strategy; S6, the emergency response strategy is executed by executing control layer to drive AGV scheduling system, temperature control equipment and security device; S7, update digital twin model based on Internet of Things feedback data to form a closed-loop optimization system; Among them, the multi-source sensing device group includes laser radar, infrared thermal imager, RFID positioning device and temperature and humidity sensor, the digital twin modeling layer maps the physical warehouse to virtual entity by using point cloud reconstruction algorithm, and the intelligent analysis layer integrates improved YOLOv7 target detection model and LSTM time series prediction model to realize dynamic risk early warning.

[0007] Preferably, S1 includes the following steps: S11, scanning the goods stacking form through the laser radar array on the top of the warehouse to generate three-dimensional point cloud coordinate data; S12, collecting temperature distribution data by deploying infrared thermal imagers in the aisle of the shelf; S13, tracking the position coordinates of AGV and the moving trajectory of goods in real time through RFID positioning device; S14, using distributed temperature and humidity sensor network to obtain warehouse environment parameter time series data.

[0008] Preferably, S2 includes the following steps: S21, importing the three-dimensional point cloud coordinate data into Unity3D engine, and using Delaunay triangulation algorithm to construct goods surface grid model; S22, generating heat map overlay layer by fusing temperature distribution data and RFID positioning data; S23, binding environment parameter time series data to virtual shelf node through space-time alignment algorithm to construct warehouse digital twin model Mtwins; S24, outputting twin point cloud data set Pcloud containing spatial coordinates, temperature gradient and goods displacement vector.

[0009] Preferably, S3 includes the following steps: S31, feature dimension reduction processing is performed on the twin point cloud data set Pcloud: Among them, Pcloud is point cloud data set, unit is meter, which represents three-dimensional coordinates of goods surface, Pcloud is point cloud mean vector, unit is meter, P is principal component analysis projection matrix, dimensionless, For the dimension-reduced point cloud dataset, the unit is meter; S32, extract the shape feature after dimension reduction, temperature anomaly gradient and displacement mutation frequency, and construct an environment feature vector matrix Vfeature.

[0010] Preferably, the S4 comprises the following steps: S41, input Vfeature into an improved YOLOv7 model for real-time target detection, and the improved YOLOv7 model embeds a CBAM attention mechanism in the neck network; S42, analyze the continuity anomaly of temperature and displacement through an LSTM time series prediction module; S43, when detecting the risk of goods collapse and fire hazards, generate a safety state evaluation report Ralert.

[0011] Preferably, the S5 comprises the following steps: S51, establish a warehouse equipment scheduling optimization matrix, and normalize the dimension: the cost unit is yuan / meter, and the distance unit is meter: wherein, is an AGV path decision variable, dimensionless, is a path cost coefficient, with a unit of yuan / meter, is an obstacle avoidance constraint matrix, dimensionless, is an emergency target vector, with a unit of meter, is a regularization coefficient, dimensionless; S52, solve the optimal obstacle avoidance path by using a genetic-ant colony hybrid algorithm; S53, generate an emergency strategy set Splan containing AGV re-routing instructions and sprinkler system activation instructions.

[0012] Preferably, the S6 comprises the following steps: S61, issue the Splan instructions to the AGV controller through the OPC UA protocol; S62, drive the temperature control system to perform cooling operation, and synchronously start the exhaust device; S63, activate the security broadcast system to prompt personnel evacuation.

[0013] Preferably, the system further comprises a digital twin self-optimization module, configured to perform the following operations: S81, collect equipment execution data during emergency response; S82, calculate the deviation δ of the actual response effect and the twin prediction result; S83, when δ> threshold value θ, the threshold value θ=0.05, update the abnormal detection model parameters through a reinforcement learning algorithm, and the value function Q is dimensionless: wherein, is the old model parameter, dimensionless, is the new model parameter, dimensionless, is the learning rate, dimensionless, in the range [0, 1], is the discount factor, dimensionless, in the range [0, 1], is the state-action value function, dimensionless, is the Q function gradient, dimensionless; S84, generating an iterative version of the digital twin model Mtwinsv2S84, generating an iterative version of the digital twin model Mtwinsv2.

[0014] Preferably, the system integrates a blockchain evidence module, for: S91, encrypting and writing the security event data into the Hyperledger Fabric chaincode; S92, realizing verifiable query of the abnormal data by the audit party based on zero-knowledge proof; S93, generating an unalterable security event evidence chain Cchain.

[0015] Preferably, the system deploys an edge computing gateway to realize: S101, running a lightweight digital twin engine on the warehouse local edge node; S102, transmitting critical alarm data to the cloud platform through a time-sensitive network; S103, adopting model segmentation technology to divide the YOLOv7 detection model into an edge-end and cloud-end collaborative execution module.

[0016] (Three) beneficial effects Compared with the prior art, the present application provides a warehouse intelligent monitoring system based on digital twinning, which has the following beneficial effects: 1. In the present application, by setting a multi-source sensing fusion module, when real-time monitoring of the warehouse environment is carried out, the three-dimensional coordinates and temperature distribution of the goods are synchronously collected by laser radar scanning and infrared thermal imaging, and the displacement trajectory of the RFID positioning and tracking equipment is combined to ensure the multidimensional integrity of the physical space data; at the same time, a dynamic mapping mechanism of the point cloud data set and the digital twin model is established, which can detect the risk of sudden changes in the stacking form and abnormal fluctuations in the temperature of the goods in real time, avoid the early warning lag problem caused by data asynchrony in traditional monitoring systems, and improve the accuracy of safety hazard identification and response timeliness. 2. In the present application, by setting the intelligent decision-making center module, when conducting warehouse risk analysis, the improved YOLOv7 target detection model and the LSTM time series prediction model are integrated to analyze the feature vector matrix, and the cargo collapse probability and fire hazard level are calculated in real time; at the same time, the genetic-ant colony hybrid algorithm is introduced to optimize the equipment scheduling path, so that the system can accurately associate the environmental parameter changes and the AGV running track, and dynamically generate obstacle avoidance instructions when the path planning deviates, eliminate the decision-making error caused by the failure of multi-source data fusion, and ensure that the warehouse equipment collaborative operation accuracy meets the industrial safety standards.3. In the present application, by setting the closed-loop execution control module, when conducting emergency response operation, the OPC UA protocol is used to issue instructions to the AGV controller, the temperature control system and the security device, to realize millisecond-level cooperation of spray cooling, exhaust linkage and evacuation broadcast; at the same time, based on the Internet of Things feedback data, the digital twin model parameters are corrected in real time, so that the system can break the execution island between security, logistics and environmental control equipment, solve the response delay problem caused by the disconnection of traditional warehouse cross-system strategies, and improve the overall safety protection efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 It is the overall framework schematic diagram of the warehouse intelligent monitoring system based on digital twin of the present application. DETAILED DESCRIPTION

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

[0019] Please refer to Figure 1 The warehouse intelligent monitoring system based on digital twin comprises the following steps: S1, collecting warehouse environment dynamic monitoring data and warehouse equipment static parameter data through a multi-source sensing device group; S2, based on the warehouse environment dynamic monitoring data and the warehouse equipment static parameter data, constructing a three-dimensional visual warehouse digital twin model, and generating a warehouse environment twin point cloud data set; S3, performing multi-dimensional feature extraction on the warehouse environment twin point cloud data set to generate an environment feature vector matrix; S4, inputting the environment feature vector matrix into a pre-trained abnormal behavior detection model for real-time risk analysis, and outputting a warehouse safety state evaluation report; S5, when detecting a safety risk, calling a path optimization algorithm to generate a device obstacle avoidance instruction set and an emergency response strategy; S6, execute the emergency response strategy by executing the control layer to drive the AGV scheduling system, temperature control equipment and security device; S7, update the digital twin model based on the Internet of Things feedback data to form a closed-loop optimization system; Among them, the multi-source sensing device group includes laser radar, infrared thermal imager, RFID positioning device and temperature and humidity sensor, the digital twin modeling layer uses point cloud reconstruction algorithm to map the physical warehouse to virtual entity, the intelligent analysis layer integrates improved YOLOv7 target detection model and LSTM time series prediction model to realize dynamic risk early warning; S11, scan the goods stacking form through the laser radar array on the top of the warehouse to generate three-dimensional point cloud coordinate data; S12, collect temperature distribution data using the infrared thermal imager deployed in the aisle of the shelf; S13, track the AGV position coordinates and goods movement trajectory in real time through the RFID positioning device; S14, obtain warehouse environment parameter time series data using a distributed temperature and humidity sensor network; S21, import the three-dimensional point cloud coordinate data into the Unity3D engine, and use the Delaunay triangulation algorithm to construct a goods surface grid model; S22, generate a heat map overlay layer by fusing temperature distribution data and RFID positioning data; S23, bind the environment parameter time series data to the virtual shelf node through the spatio-temporal alignment algorithm to construct a warehouse digital twin model Mtwins; S24, output the twin point cloud data set Pcloud containing spatial coordinates, temperature gradient and goods displacement vector; S31, perform feature dimensionality reduction processing on the twin point cloud data set Pcloud: Among them, is the point cloud data set, with units of meters, representing the three-dimensional coordinates of the goods surface, is the point cloud mean vector, with units of meters, is the principal component analysis projection matrix, dimensionless, is the reduced point cloud data set, with units of meters; S32, extract the reduced form feature, temperature anomaly gradient and displacement mutation frequency to construct the environment feature vector matrix Vfeature; S41, input Vfeature into the improved YOLOv7 model for real-time target detection, and the improved YOLOv7 model embeds CBAM attention mechanism in the neck network; S42, analyze the continuity anomaly of temperature and displacement through the LSTM time series prediction module; S43, generating a safety state assessment report Ralert when detecting the risk of goods collapse and fire hazards; S51, establishing a warehouse equipment scheduling optimization matrix, and performing dimension normalization: cost unit is yuan / meter, distance unit is meter: wherein, is an AGV path decision variable, dimensionless, is a path cost coefficient, unit is yuan / meter, is an obstacle avoidance constraint matrix, dimensionless, is an emergency target vector, unit is meter, is a regularization coefficient, dimensionless; S52, solving the optimal obstacle avoidance path by using a genetic-ant colony hybrid algorithm; S53, generating an emergency strategy set Splan containing AGV re-routing instructions and sprinkler system activation instructions; S61, issuing the Splan instructions to the AGV controller through the OPC UA protocol; S62, driving the temperature control system to perform cooling operation, and synchronously starting the exhaust device; S63, activating the security broadcast system to prompt personnel evacuation; S81, collecting equipment execution data during emergency response; S82, calculating the deviation δ of the actual response effect and the twin prediction result; S83, when δ> threshold value θ, threshold value θ=0.05, updating the abnormality detection model parameters by using a reinforcement learning algorithm, and the value function Q is dimensionless: wherein, is an old model parameter, dimensionless, is a new model parameter, dimensionless, is a learning rate, dimensionless, and the value range is [0, 1], is a discount factor, dimensionless, and the value range is [0, 1], is a state-action value function, dimensionless, is a Q function gradient, dimensionless; S84, generating an iterative version Mtwinsv2 of the digital twin model; S91, writing the security event data into the Hyperledger Fabric chain code after encryption; S92, realizing the verifiable query of the abnormal data by the audit party based on zero-knowledge proof; S93, generating an unforgeable security event evidence chain Cchain; S101, running a lightweight digital twin engine at the local edge node of the warehouse; S102, transmitting critical alarm data to the cloud platform through the time-sensitive network; S103, using model segmentation technology to divide the YOLOv7 detection model into an edge-end and a cloud-end collaborative execution module.

[0020] Specific embodiment one, multi-source data acquisition and twin modeling When a certain intelligent logistics warehouse deploys the system, the data acquisition layer scans the cargo stacking profile through the laser radar array installed on the top of the warehouse, generating millimeter-level precision three-dimensional point cloud coordinate data. At the same time, the infrared thermal imager in the goods shelf channel captures the surface temperature distribution of the goods in real time, the RFID tag integrated in the cargo box and the AGV vehicle-mounted reader continuously interact with the position information, and the distributed temperature and humidity sensor network collects environmental parameter changes. The digital twin modeling layer inputs the above data into the central processing unit, uses point cloud reconstruction technology in the Unity3D engine to build a virtual warehouse model, generates a dynamic mesh on the goods surface through the Delaunay triangulation algorithm, and maps the temperature data as a thermal map overlay layer. When a batch of flammable goods abnormally rises in temperature, the system automatically superimposes the RFID positioning track and the thermal map, marks the high-temperature risk area coordinates in the virtual model, and realizes real-time mirror mapping between the physical environment and the digital space.

[0021] Specific embodiment two, dynamic risk identification and decision optimization During the peak period of warehouse operations, the intelligent analysis layer performs multi-dimensional analysis on the point cloud data set output by the twin model, extracts the cargo stacking shape feature vector and temperature gradient change time sequence, and improves the YOLOv7 target detection model embedded with the CBAM attention mechanism to identify the risk of cargo pile inclination angle exceeding the limit. At the same time, the LSTM time sequence model analyzes the abnormal rate of temperature rise in a certain area, and the system determines that there is a risk of cargo collapse and fire hazard, generates a safety warning report and triggers the path optimization module, plans an AGV obstacle avoidance path in the virtual model through a hybrid solution of genetic algorithm and ant colony algorithm, and avoids the inclined cargo pile and high-temperature area. When a certain AGV deviates from the planned route due to cargo obstruction, the system recalculates the path in real time and issues new navigation coordinates to ensure that the equipment running track always meets the safety distance threshold.

[0022] Specific embodiment three, cross-device collaboration and closed-loop response When the system detects a fire risk in a certain cargo area, the control layer sends an emergency avoidance instruction to the AGV controller through the OPC UA protocol, drives the equipment to bypass to a safe area, synchronously activates the temperature control system to start the sprinkler device to directional cooling of high-temperature cargo piles, and links the exhaust device to accelerate smoke dissipation. The security broadcast system automatically plays the evacuation prompt to guide personnel to evacuate the risk area. All execution data are returned to the digital twin modeling layer through the Internet of Things, and the cargo pile state and equipment position in the virtual model are dynamically updated. When there is a deviation between the actual sprinkling range and the model prediction, the system corrects the heat diffusion parameters based on the reinforcement learning algorithm to generate an optimized twin model version to guide subsequent response, forming a closed-loop management and control from risk identification to strategy iteration.

[0023] Specific embodiment four, self-optimization process of twin model During the night inspection process of a certain warehouse center, the system detects abnormal temperature rise in the cargo area but does not reach the fire threshold. The digital twin self-optimization module continuously collects temperature control equipment sprinkling range and exhaust efficiency data, compares the heat diffusion track in the twin body prediction model, and when the actual water mist coverage area deviates from the virtual model by more than the critical value, the system automatically triggers the reinforcement learning mechanism, updates the LSTM time series prediction model weight by adjusting the thermodynamic conduction parameters, and the new model version increases the warehouse air circulation factor calculation in the subsequent high-temperature event prediction, so that the sprinkling instruction is issued in advance. At the same time, the heat map coverage layer of the virtual twin body is updated, and the corrected risk spread boundary is dynamically displayed to guide the security broadcast system to accurately narrow the evacuation range. The whole process realizes the self-repair of model prediction deviation and the accurate iteration of response strategy.

[0024] Specific embodiment five, edge-cloud collaborative risk response In the joint monitoring scene of cross-regional warehouse clusters, the edge computing gateway deploys a lightweight digital twin engine. When the laser radar of a certain sub-warehouse scans a severely tilted cargo pile, the edge node immediately runs the improved YOLOv7 model to intercept the neck network feature layer, determines that the collapse probability is over-limit locally, and issues a preliminary alarm. At the same time, the key point cloud data is compressed and transmitted to the cloud through the time-sensitive network. The cloud platform calls the complete LSTM model to analyze the historical temperature mutation records of the region, confirms the composite risk, generates a global emergency strategy, and the edge gateway receives the instruction to preferentially schedule the AGV of the warehouse to transfer high-risk goods. Synchronously activate the exhaust device of the adjacent warehouse to support across regions, and perform real-time obstacle avoidance path re-planning on the edge side throughout the process, and complete resource collaborative scheduling in the cloud to ensure that the key response delay meets the industrial safety time efficiency standard.

[0025] Specific embodiment six, cold chain medicine safety monitoring and blockchain traceability A certain biological medicine storage center deploys the system to monitor the storage and transportation process of high-value vaccines. When the refrigerated truck compartment enters the warehouse, the data acquisition layer captures the temperature and humidity parameters in real time through the compartment door sensor group: the laser radar scans the vaccine packaging stack form to generate three-dimensional point cloud coordinates, the infrared thermal imager detects the uniformity of the refrigerant distribution on the packaging surface, the RFID chip records the batch information and historical temperature control records of each box of vaccine, the digital twin modeling layer reconstructs the micro-environment model of the refrigerated truck compartment in the virtual environment, superimposes the point cloud coordinates and temperature field to generate a cold chain safety thermal map, and binds the RFID temperature change history of each box of vaccine to the virtual packaging node through the spatio-temporal alignment algorithm, and the intelligent analysis layer starts multi-dimensional risk calculation: the improved YOLOv7 model embedded with CBAM attention mechanism identifies the packaging box stack angle, combined with the LSTM time series model to analyze the temperature fluctuation curve, when it is detected that the vaccine packaging in a certain area is tilted and the temperature exceeds the standard for two consecutive hours, the system determines that the batch is deteriorated, generates a safety alarm report to trigger a three-way response: the execution control layer closes the refrigeration valve in the problem area through the OPC UA protocol, dispatches the AGV to transport the high-risk vaccine to the isolation area, and the digital twin self-optimization module collects the temperature change during the transportation process, and the reinforcement learning algorithm corrects the heat conduction parameters and updates the twin model thermal map boundary.

[0026] It should be noted that the relational terms herein such as first and second and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any such actual relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element preceded by "comprises a" does not, without more constraints, foreclose the existence of additional identical elements in the process, method, article, or apparatus that comprises the element.

[0027] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and changes can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A warehouse intelligent monitoring system based on digital twins, characterized in that: The system includes: a data acquisition layer, a digital twin modeling layer, an intelligent analysis layer, and an execution control layer; The data acquisition layer collects dynamic monitoring data of the warehouse environment and static parameter data of the equipment through a multi-source sensor group. Each device is synchronized to the central processing unit via the Internet of Things protocol. The digital twin modeling layer constructs a three-dimensional visualized warehouse digital twin model based on the collected data. It binds environmental parameters to virtual shelf nodes through a spatiotemporal alignment algorithm and outputs a twin point cloud dataset containing spatial coordinates, temperature gradients, and cargo displacement vectors. The intelligent analysis layer extracts multi-dimensional features from the twin point cloud dataset and inputs them into the abnormal behavior detection model. The model integrates an improved YOLOv7 target detection module with a neck network embedded with the CBAM attention mechanism and an LSTM time series prediction module. It performs joint analysis of morphological features, temperature anomaly gradients, and displacement mutation frequencies. When detecting cargo collapse risks and fire hazards, it generates a safety status assessment report. Based on the analysis results, the execution control layer calls the path optimization algorithm to generate equipment obstacle avoidance instructions and emergency strategies. It drives the AGV controller to perform rerouting, the temperature control system to perform cooling and ventilation linkage, and the security broadcast device to activate evacuation prompts through the OPCUA protocol. The four layers interact in real time via a data bus using the MQTT protocol, forming a closed-loop operation process of "perception-modeling-analysis-execution".

2. A warehouse intelligent monitoring system based on digital twins, characterized in that: Includes the following steps: S1. Collect dynamic monitoring data of the storage environment and static parameter data of the storage equipment through a multi-source sensor group; S2. Based on the dynamic monitoring data of the warehousing environment and the static parameter data of the warehousing equipment, construct a three-dimensional visualized digital twin model of the warehousing environment and generate a warehousing environment twin point cloud dataset. S3. Perform multi-dimensional feature extraction on the twin point cloud dataset of the warehouse environment to generate an environmental feature vector matrix; S4. Input the environmental feature vector matrix into the pre-trained abnormal behavior detection model for real-time risk analysis and output a warehouse safety status assessment report. S5. When a safety risk is detected, the path optimization algorithm is invoked to generate a set of device obstacle avoidance instructions and an emergency response strategy. S6. The emergency response strategy is executed by driving the AGV scheduling system, temperature control equipment, and safety devices through the control layer. S7. Update the digital twin model based on IoT feedback data to form a closed-loop optimization system; The multi-source sensing device group includes lidar, infrared thermal imager, RFID positioning device and temperature and humidity sensor. The digital twin modeling layer uses point cloud reconstruction algorithm to map the physical warehouse into a virtual entity. The intelligent analysis layer integrates the improved YOLOv7 target detection model and LSTM time series prediction model to realize dynamic risk warning.

3. The intelligent warehouse monitoring system based on digital twins according to claim 2, characterized in that: S1 includes the following steps: S11. Scan the stacking pattern of goods using the lidar array on the top of the warehouse to generate three-dimensional point cloud coordinate data; S12. Collect temperature distribution data using infrared thermal imagers deployed in the shelf aisles; S13. Track the AGV's location coordinates and cargo movement trajectory in real time using an RFID positioning device; S14. Use a distributed temperature and humidity sensor network to acquire time-series data of warehouse environmental parameters.

4. The warehouse intelligent monitoring system based on digital twins according to claim 2, characterized in that: S2 includes the following steps: S21. Import the three-dimensional point cloud coordinate data into the Unity3D engine and use the Delaunay triangulation algorithm to construct a mesh model of the cargo surface. S22. A heat map overlay layer is generated by fusing temperature distribution data and RFID positioning data; S23. By using a spatiotemporal alignment algorithm, environmental parameter time-series data are bound to virtual shelf nodes to construct a warehouse digital twin model, Mtwins. S24. Output a twin point cloud dataset Pcloud containing spatial coordinates, temperature gradient, and cargo displacement vector.

5. The warehouse intelligent monitoring system based on digital twins according to claim 2, characterized in that: S3 includes the following steps: S31. Perform feature dimensionality reduction processing on the twin point cloud dataset Pcloud: in, This is a point cloud dataset, with units of meters, representing the three-dimensional coordinates of the cargo surface. This is the mean vector of the point cloud, in meters. Principal component analysis projection matrix, dimensionless. This is the dimensionality-reduced point cloud dataset, in meters; S32. Extract the morphological features, temperature anomaly gradient, and displacement mutation frequency after dimensionality reduction, and construct the environmental feature vector matrix Vfeature.

6. The warehouse intelligent monitoring system based on digital twins according to claim 2, characterized in that: S4 includes the following steps: S41. Input the Vfeature into the improved YOLOv7 model for real-time target detection. The improved YOLOv7 model embeds the CBAM attention mechanism in the neck network. S42. Analyze the continuity anomalies of temperature and displacement using the LSTM time series prediction module; S43. When a risk of cargo collapse or fire hazard is detected, a safety status assessment report (Ralert) is generated.

7. The warehouse intelligent monitoring system based on digital twins according to claim 2, characterized in that: S5 includes the following steps: S51. Establish a warehouse equipment scheduling optimization matrix, with normalized dimensions: cost unit is yuan / meter, distance unit is meters. in, For AGV path decision variables, dimensionless. This is the path cost coefficient, expressed in yuan per meter. The obstacle avoidance constraint matrix is ​​dimensionless. This is the emergency target vector, in meters. The regularization coefficient is dimensionless. S52. Use a genetic-ant colony hybrid algorithm to solve for the optimal obstacle avoidance path; S53. Generate an emergency strategy set Splan that includes AGV rerouting instructions and sprinkler system activation instructions.

8. The warehouse intelligent monitoring system based on digital twins according to claim 2, characterized in that: S6 includes the following steps: S61. Send the Splan command to the AGV controller via the OPC UA protocol; S62, Drive the temperature control system to perform cooling operation and simultaneously start the exhaust device; S63. Activate the security broadcast system to provide personnel evacuation notices.

9. The intelligent warehouse monitoring system based on digital twins according to claim 2, characterized in that: The system also includes a digital twin self-optimization module, used to perform the following operations: S81. Collect equipment execution data during emergency response; S82. Calculate the deviation δ between the actual response and the twin prediction results; S83. When δ > threshold θ, threshold θ = 0.

05. The anomaly detection model parameters are updated using a reinforcement learning algorithm. The value function Q is dimensionless. in, These are parameters from the old model, dimensionless. The parameters for the new model are dimensionless. The learning rate is dimensionless and ranges from [0,1]. The discount factor is dimensionless and ranges from [0,1]. The state-action value function is dimensionless. The gradient of the Q-function is dimensionless. S84, Generate Digital Twin Model Iterative Version Mtwinsv2 10. The intelligent warehouse monitoring system based on digital twins according to claim 2, characterized in that: The system integrates a blockchain-based evidence storage module, used for: S91. Encrypt the security event data and write it into the Hyperledger Fabric chaincode; S92. Enable auditors to verify abnormal data based on zero-knowledge proofs; S93. Generate an immutable secure event evidence chain, Cchain.

Images