A liquid barrel intelligent storage method based on dynamic storage location optimization of a hazardous chemical barrel

Abstract

The application discloses a kind of liquid barrel intelligent warehousing methods based on dynamic library location optimization of dangerous chemical barrel, belong to dangerous chemical management technical field.The method is by digital twin platform to build synchronous virtual warehouse, based on library location entropy value model to minimize system total entropy as target dynamically calculates optimal complement and the strategy of warehouse in and out;Three-dimensional grid map and improved path planning algorithm are used to generate global optimal path, fusion multi-sensor constructs dynamic obstacle probability graph to realize real-time obstacle avoidance;Distributed leakage detection network is deployed, combined with chemical pollution diffusion model to generate reverse gradient evacuation path, linkage mechanical isolation and explosion suppression system;Through material identification sign and digital twin, the whole process is traced, and AI inventory early warning is realized using LSTM prediction model.The application changes warehousing management from static passive mode to dynamic optimization, active safety, digital new mode of intelligent decision-making, significantly improves the safety and efficiency of dangerous chemical warehousing.

Description

Technical Field

[0001] This invention belongs to the field of hazardous chemical management technology, specifically relating to an intelligent warehousing method for liquid drums based on dynamic storage location optimization of hazardous chemical drums. Background Technology

[0002] Currently, domestic material handling station technologies have the following shortcomings and inapplicability regarding the barrelled raw materials (which possess corrosive, toxic, volatile, flammable, and explosive properties) used in semiconductor manufacturing processes and their storage in flat warehouses. Firstly, current technology cannot monitor the sealing of barrelled raw materials in real time. If a leak occurs, it cannot be detected promptly, leading to the spread of hazardous liquids. Secondly, inspections are mostly manual, which cannot achieve 24-hour real-time monitoring, is not only time-consuming and labor-intensive but also poses certain risks. Thirdly, there is no timely response and handling for leaks. Fourthly, automated loading and unloading capabilities are insufficient. Existing automated loading and unloading equipment in flat warehouses cannot adapt to the special shape and weight of barrelled raw materials. The loading and unloading process requires a large amount of manual intervention, resulting in low efficiency and safety risks. Finally, the warehouse layout is unreasonable. The existing warehouse layout does not fully consider the storage needs of barrelled raw materials, resulting in excessively long logistics paths, increasing handling time and costs. The use of corrosion-resistant materials is insufficient. Existing warehouse shelves, floors, walls, guided carts, and loading / unloading equipment are typically not specially designed for highly corrosive chemicals. Their materials cannot withstand the erosion of acids, alkalis, and other corrosive substances, easily leading to equipment damage and leakage risks, affecting service life and safety. Intelligent scheduling and path planning are also lacking. For the handling and storage of drummed raw materials, the existing intelligent storage stations' scheduling and path planning systems are not flexible and efficient enough, unable to optimize based on the hazardous characteristics and usage frequency of the raw materials. Explosion-proof measures are inadequate. For flammable liquids, existing intelligent storage stations do not adequately consider explosion-proof design, posing an explosion risk. The intelligent monitoring and management system is incomplete, lacking real-time data integration. Existing intelligent storage station monitoring systems cannot integrate with other systems in the semiconductor manufacturing process (such as MES and ERP) in real time, resulting in information silos and preventing intelligent management of the entire process. Predictive maintenance capabilities are weak. Existing intelligent storage station management systems lack sufficient data analysis and machine learning capabilities for predictive maintenance, unable to accurately predict equipment failures and raw material needs, leading to production interruptions and increased maintenance costs. The remote operation function is limited; the existing intelligent storage stations lack sufficient remote operation capabilities to meet the operational needs of high-risk environments. To address these issues, existing storage station technology needs to be optimized and upgraded to improve the safety, efficiency, and intelligence of barrelled raw material storage and management in semiconductor manufacturing processes. Summary of the Invention

[0003] To address the aforementioned problems in the existing technology, this invention provides an intelligent storage and retrieval method for liquid tanks. The objective of this invention can be achieved through the following technical solution: A smart warehousing method for liquid drums based on dynamic location optimization of hazardous chemical drums includes: S1: Construct a virtual warehouse model that is fully synchronized with the physical warehouse through a digital twin platform, and establish a material identification system for full-process identification and status tracking; S2: Monitor the usage status of storage locations in real time through storage location data collectors; based on the storage location entropy model, with the goal of minimizing the total entropy of the system, dynamically calculate the optimal replenishment scheme and inbound / outbound strategy, and generate the global optimal scheduling decision, determining at least two of the following: storage location occupancy rate, raw material turnover frequency, and distance to high-frequency raw material areas. S3: Execute the transport operation instruction according to the global optimal scheduling decision; generate the global optimal driving path based on map and path planning algorithms; perform real-time obstacle avoidance and path replanning by fusing multiple sensors; S4: Deploy a distributed leak detection network. When a leak alarm is triggered, generate a concentration gradient field and isolation zone based on a chemical pollution diffusion model, dynamically generate a reverse gradient evacuation path, and link the mechanical isolation subsystem and explosion suppression protection system to perform physical isolation and emergency response. S5: Through the material identification tag and digital twin platform, the entire process of traceability and status tracking from warehousing, storage, outbound to empty barrel recycling is realized.

[0004] S6: Through the multi-source heterogeneous data access layer, the analysis module is trained based on a standardized dataset. During the model inference stage, the predicted demand values ​​for each raw material are calculated, and the predicted raw material shortage is calculated, completing the data processing and early warning triggering of the AI ​​early warning system. Specifically, the material identification tag adopts a multi-level coding structure for raw material classification, batch management, attribute association, and full-process traceability query; the storage tag is embedded in the center of the pallet and is used to store the pallet ID and its associated bucket ID group.

[0005] Specifically, the warehouse location data acquisition device includes, but is not limited to, weighing sensors and photoelectric sensors; wherein, the real-time data from the weighing sensors determines whether the warehouse location is vacant, and automatically sends an vacancy signal to the warehouse management system when it is vacant.

[0006] Specifically, the objective function of the storage location entropy model is to minimize the total entropy of the system. The total entropy is calculated by weighting three components: storage location occupancy rate entropy, raw material turnover frequency entropy, and distance entropy from the high-frequency raw material area.

[0007] Specifically, the calculation process for the optimal replacement scheme and the inbound / outbound strategy includes: With entropy reduction rate as the objective, the same type of raw material barrels are searched for as candidate replacement barrels within a preset radius according to the nearest priority principle; In the digital twin platform, electronic fences are set up for high-risk raw materials and marked as restricted areas, excluding paths that pass through these areas from the candidate paths; Taking into account the probability of storage location status, path length, and safety rules, the optimal handling path from the candidate bucket location to the target vacant storage location is calculated; Output the final replacement command to control the automated handling equipment to perform handling and precise placement operations.

[0008] Specifically, the calculation process of the globally optimal driving path includes: Based on the factory's 3D grid map, the warehouse space is discretized into passable nodes and obstacle nodes, and a weighted directed topology network is constructed. The node weights comprehensively reflect the constraints of path length, ground flatness, and equipment turning radius. The search is performed in the topology network, and the static optimal basic path from the starting point to the target point is output. During the operation of automated handling equipment, a dynamic obstacle probability map is constructed by fusing LiDAR and depth camera, and the obstacle existence probability of grid nodes is updated in real time. When the node probability exceeds the threshold, the network edge weights are dynamically adjusted to trigger local obstacle avoidance or global path replanning.

[0009] Specifically, the process of constructing and updating the dynamic obstacle probability map includes: Point cloud data of the warehouse environment is acquired by solid-state LiDAR. Ground segmentation and statistical filtering algorithms are applied to remove ground point cloud and noise interference, retain the effective points of obstacles, and extract the centroid coordinates, contour size and motion velocity vector of each obstacle. Simultaneously, image streams are acquired by depth cameras, obstacle categories are identified based on convolutional neural networks, and obstacle bounding boxes and confidence scores are calculated. During the operation of automated handling equipment, the fused obstacle status is periodically input into the spatiotemporal probability graph model at fixed time windows to update the obstacle existence probability of each grid node in real time. When the probability value of a node exceeds a preset threshold, the node is marked as a high collision risk area, and the passage weight of the corresponding edge in the topology network is dynamically adjusted to a high penalty value, triggering the execution of local obstacle avoidance strategy or global path replanning.

[0010] Specifically, the chemical pollution diffusion model integrates a chemical property database and environmental sensor data, collects data from an instantaneous leakage model to solve partial differential equations, calculates the spatiotemporal distribution and diffusion range of the leaked pollutants in real time, delineates isolation zones of different hazard levels based on the calculation results, and generates dynamic electronic fences in a digital twin platform.

[0011] Specifically, the distributed leak detection network adopts a multi-level distributed architecture to achieve real-time monitoring, including: The distributed leak detection network adopts a multi-level distributed architecture, including: Point gas detectors, linear infrared beam detectors, and image-type infrared thermal imagers are distributed and deployed within the storage facilities to form a grid-like monitoring network. When any sensor detects a value exceeding a preset alarm threshold, the system automatically generates a leak source coordinate marker in the digital twin platform and activates a three-level alarm mechanism. The three-level alarm mechanism includes: Level 1 alarm triggers local audible and visual warnings; Level 2 alarm pushes information on the location and hazard level of the leak source to the warehouse management system; Level 3 alarm links the entire explosion suppression and protection system and triggers emergency braking and reverse gradient evacuation of all automated handling equipment.

[0012] Specifically, the mechanical isolation subsystem is used to automatically perform physical isolation operations when a leakage alarm is triggered, including but not limited to: Control the closure of automatic fireproof isolation doors or explosion-proof barriers deployed at the boundaries of warehouse zones; Activate the negative pressure exhaust system to directionally exhaust the leaked gas and close the air supply duct in the contaminated area; The inert gas explosion suppression device and mechanical covering mechanism work together to initially seal and suppress the leak point; After the central control system remote operators confirm that the area is in a safe state and authorize it, explosion-proof automated handling equipment is dispatched to carry emergency supplies to the leak point to assist in the leak plugging or relocation tasks.

[0013] Specifically, the data processing and early warning triggering process of the AI ​​early warning system includes: Through a multi-source heterogeneous data access layer, production planning and scheduling data from the production execution system, real-time order data from the order management system, current inventory data from the warehouse management system, and time-series consumption records of each raw material in the historical consumption database are captured in real time. The raw data is preprocessed by format normalization, outlier removal, and time alignment to form a standardized dataset. The time series prediction analysis module is trained based on a standardized dataset, and a raw material consumption prediction model is constructed using a long short-term memory network. The input features of the model include product type, output, and process route in the production plan, delivery cycle and priority in the order data, and seasonal fluctuations and trend changes in historical consumption data. The model output is the predicted demand for each raw material within the time window. During the model inference phase, real-time production plans and order data are input into the prediction model to calculate the predicted demand for each raw material in each time window. Simultaneously, the current available inventory and in-transit inventory are obtained from the warehouse management system to calculate the predicted shortage.

[0014] Furthermore, the raw material consumption prediction model continuously collects the deviation between actual consumption data and predicted values ​​to construct a prediction error time series; the early warning threshold is dynamically adjusted based on the supplier's on-time delivery rate and the frequency of production plan changes; an early warning feedback closed-loop mechanism is established, and when early warning information is pushed, the response time, handling measures, and final results of the receiving personnel are recorded.

[0015] Furthermore, the inbound / outbound strategy also includes dynamic location scheduling based on the first-in, first-out (FIFO) principle or a preset expiration date management principle: For each pallet / box of materials, the entry time is recorded and the storage time is continuously monitored. The storage time is automatically identified, and the turnover priority is adaptively marked according to the storage time and the storage location of the materials. The handling path and retrieval time of the automated handling equipment for outbound operations are updated based on the turnover priority.

[0016] Furthermore, the inbound / outbound strategy also includes a pre-defined scheduling strategy that coordinates with the upper-level production management system: Receive and parse production plans or outbound requirements from ERP or MES systems, and intelligently schedule the required materials from various storage locations to the reserve storage location or buffer area near the outbound outlet in advance and in batches; when the formal outbound instruction is issued, the materials are already in the ready-to-go state.

[0017] The beneficial effects of this invention are as follows: By integrating automated equipment, intelligent algorithms, and digital twin technology, a high degree of intelligence, safety, and efficiency in liquid drum storage management has been achieved. By minimizing the total system entropy and comprehensively considering factors such as storage space occupancy, raw material turnover frequency, and distance from high-frequency usage areas, the system automatically calculates the optimal replenishment plan. This reduces material handling paths and time, avoids storage space chaos, and improves shelf space utilization and access efficiency.

[0018] Based on a 3D grid map of the factory and an improved Dijkstra algorithm, a static optimal basic path is generated, and real-time obstacle avoidance is performed by combining a dynamic obstacle probability map. This ensures the high efficiency and flexibility of AGV operation, reduces waiting and congestion time, and improves the overall material flow speed.

[0019] A multi-level distributed leak detection system (including point-type combustible gas detectors, infrared beam detectors, and thermal imagers) monitors the environment in real time and triggers an alarm immediately upon detecting a leak. A chemical pollution diffusion model, based on physical properties and environmental data, rapidly generates a concentration gradient field and isolation zone, and automatically generates a reverse gradient evacuation path to guide AGVs safely away from the contaminated area, effectively preventing personnel casualties and equipment damage.

[0020] The storage area is divided into independent explosion suppression units, linked to ventilation control and inert gas explosion suppression devices. Upon triggering a leak alarm, the explosion suppression procedure is automatically activated to suppress the risk of explosion and is also linked to the fire suppression system, significantly reducing the probability of major safety accidents.

[0021] Electronic fences are set up for high-risk raw materials (such as hydrofluoric acid) in the digital twin platform to eliminate dangerous paths, prevent AGVs from accidentally entering restricted areas, and improve operational safety.

[0022] Each raw material barrel has a laser-etched marking on its outer wall, and a corrosion-resistant label is embedded in the pallet, linking it to the barrel's ID information. In the event of a leak or anomaly, the information is quickly read using a handheld scanner, highlighting the abnormal barrel on the digital twin warehouse map and generating a traceability report. This enables full lifecycle tracking of raw materials from inbound to outbound, improving quality control and problem-solving efficiency.

[0023] When raw material barrels are removed or empty barrels are recycled, the system automatically updates the data and removes the ID association, ensuring the real-time accuracy and consistency of inventory information and reducing errors from manual recording.

[0024] From AGV material handling to roller conveyor conveying controlled by pneumatic cylinders, the entire automated process is triggered by photoelectric sensors, eliminating the need for manual intervention. This reduces manpower requirements and minimizes errors and delays caused by human operation.

[0025] An LSTM model trained on historical consumption data predicts future raw material demand and compares it with real-time inventory to automatically generate early warning information. This avoids inventory shortages or surpluses, optimizes procurement and production plans, and reduces the risk of emergency restocking and production interruptions. AGVs use solid-state LiDAR and depth cameras to construct dynamic obstacle probability maps in real time, automatically detouring or replanning paths when encountering obstacles, ensuring reliable operation in complex environments. In emergency situations, an optimal evacuation path is generated based on a concentration gradient vector field and multi-objective optimization algorithms, ensuring the AGV evacuates at the maximum safe speed and minimizing risk. The AI ​​early warning system dynamically adjusts warning thresholds based on actual consumption errors, supplier on-time delivery rates, and the frequency of production plan changes, improving the accuracy and adaptability of warnings.

[0026] The entire warehousing system incorporates both time-based, routine optimization and demand-based, proactive scheduling, constructing a highly adaptive, low-latency physical logistics system. This not only ensures the activity and visibility of inventory materials but also compresses the outbound operation cycle from both spatial and temporal dimensions, providing solid warehousing support for continuous production and agile supply chains. Ultimately, it achieves a synergistic improvement in warehouse space utilization, outbound speed, and operational flexibility. The entire warehousing process is mapped in real-time within a digital twin platform, allowing for visual monitoring, simulation, and optimization. Warehouse location status, route planning, and safety rules are all tested and executed in a virtual environment, enhancing the scientific nature of decision-making and system controllability. Through multi-source heterogeneous data access and time-series predictive analysis, the system continuously learns and adjusts, achieving continuous improvement in warehouse management.

[0027] The specific mathematical model formula for the chemical pollution diffusion model (such as the Gaussian plume model) is used as a specific implementation method in this embodiment. The chemical pollution diffusion model adopts an indoor leakage diffusion algorithm based on the Gaussian plume model, and the calculation formula for its concentration spatiotemporal distribution is as follows:

[0028] Where: C(x,y,z,t) is the pollutant concentration at spatial point (x,y,z) at time t (unit: mg / m³). 3 ); Q is the leakage source intensity (unit: mg / s), which is obtained by integrating the leakage rate provided by the leakage detection system over time. u represents the average velocity (in m / s) of the airflow in the x-direction (prevailing wind direction) within the warehouse, which is collected in real time by wind speed sensors deployed within the warehouse; σ y ,σ z These are the diffusion coefficients (unit: m) in the horizontal and vertical directions, respectively. Based on the turbulence intensity and atmospheric stability level of the storage environment, an empirical formula for indoor diffusion coefficients is used: σ y =a*x b ,σ z =c*x d The calculation involves coefficients a, b, c, and d pre-stored in a chemical property database and dynamically corrected based on temperature and humidity sensor data; H represents the effective leak source height (in meters), taken as the actual height of the leak point above the ground; f(t) is the time accumulation factor, applied during continuous leakage before reaching a steady state. This is used to describe the propagation process of a leaking gas front.

[0029] The solution and output process is as follows: Parameter initialization: Once the leak detection system triggers an alarm, the model immediately retrieves parameters such as molecular weight, density, saturated vapor pressure, and acute toxicity exposure limits (such as IDLH and ERPG-2) of the leaked raw material from the chemical property database, and obtains the current temperature, humidity, air pressure, and ventilation airflow velocity vector from the environmental sensor network.

[0030] Concentration field calculation: Taking the leakage source as the origin, the warehouse space is discretized into three-dimensional grid nodes. The Gaussian plume formula mentioned above is applied to each node to calculate the concentration change curve over time within a future time window (e.g., 0~300 seconds).

[0031] Isolation zone delineation: Nodes with calculated concentrations exceeding the "severe risk zone threshold" (e.g., 50% of IDLH value) are marked as severe risk zones, those exceeding the "medium risk zone threshold" (e.g., 10% of IDLH value) are marked as medium risk zones, and those exceeding the "light risk zone threshold" (e.g., occupational exposure limit) are marked as light risk zones. In the digital twin platform, severe risk zones are rendered as red restricted areas, medium risk zones as yellow warning zones, and light risk zones as blue observation zones, with corresponding electronic fences dynamically generated.

[0032] Reverse gradient evacuation path generation: Calculate the concentration gradient vector field centered on the leakage source. its opposite direction This is the direction in which the concentration decreases the fastest. Starting from the current position of the AGV, the A* algorithm is used to search for the shortest safe path on the grid map that is in the opposite direction of the concentration gradient and avoids nodes in the most dangerous areas. This path is then output as the evacuation path.

[0033] Model adaptive update mechanism: This model continuously receives data on changes in leak source intensity and environmental parameters every 1 second, and recalculates the concentration field distribution in real time. When a pollution cloud is detected spreading to fire lanes, emergency exits, or areas where personnel are frequently present, the highest level alarm is automatically triggered across the entire area, linking the explosion suppression and fire protection systems. Attached Figure Description

[0034] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0035] Figure 1 This is a schematic diagram of the main process of warehousing operations according to the present invention; Figure 2 This is a block diagram of the system layered architecture in this invention; Figure 3 This is a schematic diagram of the physical layout and sensing deployment of the hazardous chemical warehouse in this invention. Detailed Implementation

[0036] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided to make this disclosure more comprehensive and complete, and to fully convey the concept of example embodiments to those skilled in the art. Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a full understanding of embodiments of this disclosure. However, those skilled in the art will recognize that the technical solutions of this disclosure can be practiced without one or more of the specific details, or other methods, components, apparatuses, steps, etc., can be employed. In other instances, well-known methods, apparatuses, implementations, or operations are not shown or described in detail to avoid obscuring aspects of this disclosure. The blocks shown in the drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices. The flowcharts shown in the drawings are merely illustrative and do not necessarily include all contents and operations / steps, nor do they necessarily have to be performed in the order described. For example, some operations / steps can be broken down, while others can be combined or partially combined. Therefore, the actual execution order may change depending on the actual situation.

[0037] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided.

[0038] Please see Figures 1-3 A smart warehousing method for liquid drums based on dynamic location optimization of hazardous chemical drums, comprising: S1: Construct a virtual warehouse model that is fully synchronized with the physical warehouse through a digital twin platform, and establish a material identification system for full-process identification and status tracking; S2: Monitor the usage status of storage locations in real time through storage location data collectors; based on the storage location entropy model, with the goal of minimizing the total entropy of the system, dynamically calculate the optimal replenishment scheme and inbound / outbound strategy, and generate the global optimal scheduling decision, determining at least two of the following: storage location occupancy rate, raw material turnover frequency, and distance to high-frequency raw material areas. S3: Execute the transport operation instruction according to the global optimal scheduling decision; generate the global optimal driving path based on map and path planning algorithms; perform real-time obstacle avoidance and path replanning by fusing multiple sensors; S4: Deploy a distributed leak detection network. When a leak alarm is triggered, generate a concentration gradient field and isolation zone based on a chemical pollution diffusion model, dynamically generate a reverse gradient evacuation path, and link the mechanical isolation subsystem and explosion suppression protection system to perform physical isolation and emergency response. S5: Through the material identification tag and digital twin platform, the entire process of traceability and status tracking from warehousing, storage, outbound to empty barrel recycling is realized; S6: Through the multi-source heterogeneous data access layer, the analysis module is trained based on a standardized dataset. During the model inference stage, the predicted demand value of each raw material is calculated, the predicted raw material shortage is calculated, and the data processing and early warning triggering of the AI ​​early warning system are completed.

[0039] The AI ​​early warning system accesses production plan data and real-time order data from the production execution system or order management system in real time. Based on the analysis module trained on historical consumption data, it calculates the predicted demand for each raw material in a specific future time period. The predicted demand is compared with the current available inventory in real time. When the shortage exceeds the preset early warning threshold, an early warning message is automatically generated and pushed through an audible and visual alarm, a pop-up window on the central control room screen, or a mobile terminal.

[0040] The material identification tags include, but are not limited to, one or more of barcodes, QR codes, and RFID tags, supporting multi-level coding structures for raw material classification, batch management, and traceability; and based on an AI early warning system, they enable inventory shortage warnings, leakage risk warnings, and equipment health warnings.

[0041] Specifically, the objective function of the storage location entropy model is to minimize the total entropy of the system, and the storage location status is quantitatively evaluated by storage location occupancy rate, raw material turnover frequency, and distance from the high-frequency raw material area.

[0042] Specifically, the calculation process of the optimal replacement scheme includes: With entropy reduction as the objective, when a warehouse vacancy signal is received, the model first searches for similar raw material barrels as candidate replacement barrels within a preset radius according to the principle of proximity priority. Apply security isolation rules to set up electronic fences for high-risk hydrofluoric acid raw materials in the digital twin platform and mark them as restricted areas, and exclude the paths of guide vehicles that accidentally enter the area from the candidate paths; Considering the probability of storage location status, path length, and safety rules, calculate the shortest handling path from the candidate bucket location to the vacant storage location; Output the final replacement command to control the guide vehicle to perform handling and precise placement operations.

[0043] Specifically, the calculation process of the optimal basic path includes: Based on the factory's 3D grid map, the warehouse space is discretized into passable nodes and obstacle nodes, and a weighted directed topology network is constructed. The node weights comprehensively reflect the constraints of path length, ground flatness, and equipment turning radius. An improved Dijkstra algorithm is used to search in the topology network and output a static optimal basic path from the starting point to the target point, which serves as the global driving basis for the automated guided vehicle.

[0044] In this embodiment, the objective function for minimizing the total entropy of the system is defined as: , Where w1+w2+w3=1, and N is the total number of warehouses; Regarding the calculation of entropy components, taking a single storage location i as an example, the storage location occupancy rate entropy H occ (i):

[0045] Where, p occ (i) represents the occupancy probability, calculated from real-time data from the weighing sensor, where α is the data confidence constant. , Raw material turnover frequency entropy H freq (i): , Where K represents the total number of raw material types, p k (i) represents the historical turnover probability of the k-th type of raw material at storage location i: , Distance Entropy : , Where d(i,high-freq-zone) is the Manhattan distance from storage location i to the high-frequency use zone, max(d) is the maximum diagonal distance of the warehouse, and the high-frequency use zone is defined as the set of coordinates of areas with more than 50 outbound shipments in the past 30 days.

[0046] The scenario for calculating the optimal replacement scheme is set as follows: Triggering event: The hydrofluoric acid barrel at storage location A (coordinates 8,12,1) is removed, the load cell returns to zero, and an empty signal is sent. System status: The same type of raw material (hydrofluoric acid) is in stock at storage locations D (20,5,1) and E (15,18,2); Safety restrictions: The area within a 5-meter radius of the hydrofluoric acid electronic fence is a no-entry zone; The calculation process is as follows: Candidate replacement buckets are searched using a proximity-based approach. Within a preset radius, similar raw material buckets are searched to obtain the selected inventory E. In the digital twin platform, a marked electronic fence is generated with storage location A as the center and a safety constraint radius of 5 meters. Path collision detection: All possible paths from storage location E to storage location A are checked. If a path passes through the electronic fence, the path is excluded. After path verification, a safe detour path is generated. Entropy reduction rate calculation and optimal path planning: Definition of entropy reduction rate: , Calculate the expected entropy value after padding: Storage location A (vacant → occupied): Entropy value changes from 0 (empty) to (expected calculation); Storage location E (occupied → vacant): Entropy value changes from the occupied calculation result to 0 (temporarily vacant); The generated path is scored using multiple objectives by calculating the total entropy change and entropy reduction rate. , Where, d path The path length is represented by _Score, which is the safety compliance detection value. The optimal path is selected based on the generated candidate paths and the score.

[0047] Specifically, the process of constructing and updating the dynamic obstacle probability map includes: Point cloud data of the warehouse environment is acquired by solid-state LiDAR. Ground segmentation and statistical filtering algorithms are applied to remove ground point cloud and noise interference, retain the effective points of obstacles, and extract the centroid coordinates, contour size and motion velocity vector of each obstacle. Simultaneously, image streams are acquired by depth cameras, obstacle categories are identified based on convolutional neural networks, and obstacle bounding boxes and confidence scores are calculated. During the operation of the automated guided vehicle, the fused obstacle status is periodically input into the spatiotemporal probability graph model at fixed time windows to update the obstacle existence probability of each grid node in real time. When the probability value of a node exceeds a preset threshold, the node is marked as a high collision risk area, and the passage weight of the corresponding edge in the topology network is dynamically adjusted to a high penalty value, triggering the automated guided vehicle to execute a local obstacle avoidance strategy or a global path replanning.

[0048] Specifically, the process of constructing and applying the chemical pollution diffusion model includes: It integrates a chemical property database and automatically retrieves molecular weight, density, saturated vapor pressure, and toxicity threshold parameters based on the type of leaked material; it connects to the alarm signals of the leak detection system in real time to obtain the location coordinates of the leak source, the leak rate, and the duration; and it simultaneously collects storage environment parameters, including temperature, humidity, air pressure, and ventilation airflow velocity vector. The spatiotemporal distribution of pollutant concentration was calculated using a transient leakage model. Partial differential equations were solved to obtain the concentration gradient fields in the downwind, crosswind, and vertical directions. Based on the acute toxicity exposure limits of the raw materials, three levels of isolation zones—light, medium, and heavy—were delineated. In the digital twin platform, the heavy-risk zone was rendered as a restricted area and a dynamic electronic fence was generated. The area blockade command was broadcast to all automated guided vehicles. Centered on the leak source, a safe evacuation vector field is calculated along the opposite direction of the concentration gradient, generating multiple candidate evacuation paths. The Dijkstra algorithm is then used to select the optimal evacuation path with the shortest length and fewest turns. An in-vehicle audible and visual alarm device guides the AGV to leave the contaminated area at the maximum safe speed. Simultaneously, the model continuously updates the diffusion range. When the contamination cloud is detected to spread to fire lanes or densely populated areas, the full-domain explosion suppression protection system is automatically triggered, activating the inert gas explosion suppression device and linking the fire sprinkler system.

[0049] Specifically, the generation and execution process of the reverse gradient evacuation path includes: Based on the three-dimensional concentration field data output by the chemical pollution diffusion model, the concentration gradient vector field at each location around the leak point is calculated; taking the current position of the automated guided vehicle as the starting point, the direction of the fastest decrease in concentration gradient is determined as the initial evacuation direction; Based on the initial evacuation direction, a multi-objective optimization algorithm is used in the digital twin platform to generate candidate evacuation paths.

[0050] Specifically, the leak detection system adopts a multi-level distributed architecture to achieve real-time monitoring, including: Point-type combustible gas detectors are deployed in a distributed manner at the bottom of the shelves, the top of the liquid tank, and both sides of the roller conveyor to form a grid monitoring network; A linear infrared beam detector array is deployed around the perimeter of the storage area to create a grating detection barrier. An alarm is triggered when the light path is blocked by leaked liquid vapor. Deploy image-type infrared thermal imagers at monitoring nodes to detect local temperature drops caused by liquid evaporation or temperature rises caused by exothermic chemical reactions using temperature field anomaly identification algorithms; When any sensor detects a value exceeding a preset alarm threshold, the system automatically generates a leak source coordinate marker in the digital twin platform and activates a three-level alarm mechanism: Level 1 alarm triggers local audible and visual warnings; Level 2 alarm pushes the leak source location and hazard level information to the warehouse management system; and Level 3 alarm links the entire explosion suppression protection system and triggers emergency braking and reverse gradient evacuation of all automated guided vehicles.

[0051] Specifically, the comprehensive explosion suppression protection system divides the storage area into multiple explosion suppression protection units. Each unit is equipped with an independent combustible gas monitoring subnet, a ventilation control subsystem, and an inert gas explosion suppression device. The system receives three-level alarm signals from the leak detection system in real time via an industrial bus. When a level two or higher alarm is triggered, the system automatically starts the preset explosion suppression program of the corresponding protection unit.

[0052] The distributed leak detection network also includes a mechanical linkage isolation subsystem, used to automatically perform physical isolation operations when a leak alarm is triggered, including: Control the closure of automatic fireproof isolation doors or explosion-proof barriers deployed at the boundaries of warehouse zones to block the path of hazard spread; Activate the negative pressure exhaust system to directionally exhaust the leaked gas and close the air supply duct in the contaminated area; The inert gas explosion suppression device and mechanical covering mechanism work together to initially seal and suppress the leak point; Dispatch explosion-proof automated handling equipment to carry emergency supplies to the leak point to assist in sealing or relocating the leak.

[0053] The automated storage and retrieval system records the entry time of each pallet / box of materials and continuously monitors their storage time. Based on the "First-In, First-Out" (FIFO) principle or preset expiration date management, the system automatically performs dynamic scheduling of materials within the warehouse: the system automatically identifies materials with longer storage times and those nearing their outbound cycle, marking them as "high-turnover priority." Through automated equipment such as stacker cranes and shuttle cars, the system gradually and systematically transfers these materials from deeper, inconveniently located storage areas to "fast-turnover areas" or easily accessible mid-level locations near the exit. This avoids stagnation and optimizes pathways, reducing the risk of stagnation due to long-term storage. More importantly, it significantly shortens the walking and retrieval time for future outbound shipments, minimizing outbound preparation time.

[0054] To address the demands of high-efficiency production cycles or urgent orders, the automated warehouse (WMS) can deeply collaborate with upper-level ERP / MES systems. When the upper-level system issues production plans or outbound requests for a future period, the WMS receives and parses the required material list and time nodes in real time. The system intelligently schedules the planned materials from various storage locations to the nearest reserve storage location or buffer area to the outbound outlet in advance and in batches. This achieves "goods-to-person" or "just-in-time" delivery. Through this pre-configured operation, materials are already in the optimal ready-to-go state when the formal outbound instruction is issued, achieving a response time of seconds and greatly improving the overall throughput efficiency during peak periods or continuous outbound operations.

[0055] Specifically, the execution methods for triggering the signal and the process control include: It can be executed using wireless radio frequency communication and infrared optical communication. Wireless radio frequency communication acts like a "mobile cellular network" covering the entire warehouse, providing a high-speed data highway for mobile devices such as stacker cranes and AGVs, carrying the main control and information flow. Infrared communication uses "precise landmarks" and "safety switches" distributed at key nodes to provide absolute position calibration and reliable area trigger signals, ensuring long-term accuracy of mechanical positioning and safety in specific areas.

[0056] Photoelectric sensors are installed at the entrance of the automated guided vehicle (AGV) picking and roller conveyor. Through-beam photoelectric sensors are installed on the side or below the designated picking position on the shelf. When the AGV moves to a predetermined distance in front of the target location, the AGV body or picking mechanism blocks the beam emitted by the sensor. When the sensor detects that the light beam is blocked, it immediately sends a positioning signal to the automated guided vehicle (AGV) control system and the material station main control system. This signal triggers the AGV to stop moving and start a precise material picking action. A through-beam photoelectric sensor is installed at the entrance of the conveyor roller line, spanning the conveyor belt. When the raw material barrel is placed at the roller line entrance by the AGV and begins to enter the conveying area, the barrel body blocks the sensor's light beam. The sensor detects that the light beam is blocked and sends a signal to the conveyor line control system and the material station main control system, marking that the raw material barrel has officially entered the conveying process. The system records this moment and associates it with the barrel identification information, while updating the inventory status.

[0057] Specifically, the data processing and early warning triggering process of the AI ​​early warning system includes: A multi-source heterogeneous data access layer is constructed to capture production planning and scheduling data from the production execution system, real-time order data from the order management system, current inventory data from the warehouse management system, and time-series consumption records of each raw material from the historical consumption database in real time; the raw data is preprocessed by format normalization, outlier removal, and time alignment to form a standardized dataset. The time series prediction analysis module is trained based on a standardized dataset, and a raw material consumption prediction model is constructed using a long short-term memory network. The input features of the model include product type, output, and process route in the production plan, delivery cycle and priority in the order data, and seasonal fluctuations and trend changes in historical consumption data. The model output is the predicted demand for each raw material within the time window. During the model inference phase, real-time production plans and order data are input into the prediction model to calculate the predicted demand for each raw material in each time window. Simultaneously, the current available inventory and in-transit inventory are obtained from the warehouse management system to calculate the predicted shortage.

[0058] Specifically, the raw material consumption prediction model continuously collects the deviation between actual consumption data and predicted values ​​to construct a prediction error time series; the early warning threshold is dynamically adjusted based on the supplier's on-time delivery rate and the frequency of production plan changes; an early warning feedback closed-loop mechanism is established, and when the early warning information is pushed, the response time, handling measures, and final results of the receiving personnel are recorded.

[0059] The specific mathematical model formula for the chemical pollution diffusion model (such as the Gaussian plume model) is used as a specific implementation method in this embodiment. The chemical pollution diffusion model adopts an indoor leakage diffusion algorithm based on the Gaussian plume model, and the calculation formula for its concentration spatiotemporal distribution is as follows:

[0060] Where: C(x,y,z,t) is the pollutant concentration at spatial point (x,y,z) at time t (unit: mg / m³). 3 Q represents the leakage source intensity (unit: mg / s), obtained by integrating the leakage rate provided by the leakage detection system over time; u represents the average velocity of the ventilation airflow in the x-direction (main wind direction) within the warehouse (unit: m / s), collected in real time by wind speed sensors deployed within the warehouse; σ y ,σ z These represent the diffusion coefficients (in meters) in the horizontal and vertical directions, respectively. Based on the turbulence intensity and atmospheric stability level of the storage environment, the empirical formula σ for indoor diffusion coefficients is used. y =a*x b ,σ z =c*x d The calculation involves coefficients a, b, c, and d pre-stored in a chemical property database and dynamically corrected based on temperature and humidity sensor data; H represents the effective leak source height (in meters), taken as the actual height of the leak point above the ground; f(t) is the time accumulation factor, applied during continuous leakage before reaching a steady state. This is used to describe the propagation process of a leaking gas front.

[0061] The solution and output process is as follows: Parameter initialization: Once the leak detection system triggers an alarm, the model immediately retrieves parameters such as molecular weight, density, saturated vapor pressure, and acute toxicity exposure limits (such as IDLH and ERPG-2) of the leaked raw material from the chemical property database, and obtains the current temperature, humidity, air pressure, and ventilation airflow velocity vector from the environmental sensor network.

[0062] Concentration field calculation: Taking the leakage source as the origin, the warehouse space is discretized into three-dimensional grid nodes. The Gaussian plume formula mentioned above is applied to each node to calculate the concentration change curve over time within a future time window (e.g., 0~300 seconds).

[0063] Isolation zone delineation: Nodes with calculated concentrations exceeding the "severe risk zone threshold" (e.g., 50% of IDLH value) are marked as severe risk zones, those exceeding the "medium risk zone threshold" (e.g., 10% of IDLH value) are marked as medium risk zones, and those exceeding the "light risk zone threshold" (e.g., occupational exposure limit) are marked as light risk zones. In the digital twin platform, severe risk zones are rendered as red restricted areas, medium risk zones as yellow warning zones, and light risk zones as blue observation zones, with corresponding electronic fences dynamically generated.

[0064] Reverse gradient evacuation path generation: Calculate the concentration gradient vector field centered on the leakage source. its opposite direction This is the direction in which the concentration decreases the fastest. Starting from the current position of the AGV, the A* algorithm is used to search for the shortest safe path on the grid map that is in the opposite direction of the concentration gradient and avoids nodes in the most dangerous areas. This path is then output as the evacuation path.

[0065] Model adaptive update mechanism: This model continuously receives data on changes in leak source intensity and environmental parameters every 1 second, and recalculates the concentration field distribution in real time. When a pollution cloud is detected spreading to fire lanes, emergency exits, or areas where personnel are frequently present, the highest level alarm is automatically triggered across the entire area, linking the explosion suppression and fire protection systems.

[0066] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for intelligent warehousing of liquid containers based on dynamic location optimization of hazardous chemical containers, characterized in that, include: S1: Construct a virtual warehouse model that is fully synchronized with the physical warehouse through a digital twin platform, and establish a material identification system for full-process identification and status tracking; and set material identification tags and storage tags on raw material barrels and pallets for full-process identification and status tracking. S2: Monitor the usage status of storage locations in real time through storage location data collectors; based on the storage location entropy model, with the goal of minimizing the total entropy of the system, dynamically calculate the optimal replenishment scheme and inbound / outbound strategy to generate the global optimal scheduling decision; at least two of the following are determined: storage location occupancy rate, raw material turnover frequency, and distance to high-frequency raw material areas; S3: Execute the transport operation instruction according to the global optimal scheduling decision; generate the global optimal driving path based on map and path planning algorithms; perform real-time obstacle avoidance and path replanning by fusing multiple sensors; S4: Deploy a distributed leak detection network. When a leak alarm is triggered, generate a concentration gradient field and isolation area based on a chemical pollution diffusion model, dynamically generate a reverse gradient evacuation path, and link physical isolation and emergency response. S5: Through the material identification tag and digital twin platform, the entire process of traceability and status tracking from warehousing, storage, outbound to empty barrel recycling is realized; S6: Through the multi-source heterogeneous data access layer, the analysis module is trained based on a standardized dataset. During the model inference stage, the predicted demand value of each raw material is calculated, the predicted raw material shortage is calculated, and the data processing and early warning triggering of the AI ​​early warning system are completed.

2. The method according to claim 1, characterized in that, The material identification tag adopts a multi-level coding structure for raw material classification, batch management, attribute association, and full-process traceability query; the storage tag is embedded in the center of the pallet and is used to store the pallet ID and its associated bucket ID group.

3. The method according to claim 1, characterized in that, The warehouse location data acquisition device includes, but is not limited to, weighing sensors and photoelectric sensors; wherein, the real-time data from the weighing sensors determines whether the warehouse location is vacant, and automatically sends an vacancy signal to the warehouse management system when it is vacant.

4. The method according to claim 1, characterized in that, The objective function of the storage location entropy model is to minimize the total entropy of the system. The total entropy is calculated by weighting three components: storage location occupancy rate entropy, raw material turnover frequency entropy, and distance entropy from the high-frequency raw material area.

5. The method according to claim 1, characterized in that, The calculation process for the optimal replacement scheme and the inbound / outbound strategy includes: With entropy reduction rate as the objective, the same type of raw material barrels are searched for as candidate replacement barrels within a preset radius according to the nearest priority principle; In the digital twin platform, electronic fences are set up for high-risk raw materials and marked as restricted areas, excluding paths that pass through these areas from the candidate paths; Taking into account the probability of storage location status, path length, and safety rules, the optimal handling path from the candidate bucket location to the target vacant storage location is calculated; Output the final replacement command to control the automated handling equipment to perform handling and precise placement operations.

6. The method according to claim 1, characterized in that, The calculation process for the globally optimal driving path includes: Based on the factory's 3D grid map, the warehouse space is discretized into passable nodes and obstacle nodes, and a weighted directed topology network is constructed. The node weights comprehensively reflect the constraints of path length, ground flatness, and equipment turning radius. The search is performed in the topology network, and the static optimal basic path from the starting point to the target point is output. During the operation of automated handling equipment, a dynamic obstacle probability map is constructed by fusing LiDAR and depth camera, and the obstacle existence probability of grid nodes is updated in real time. When the node probability exceeds the threshold, the network edge weights are dynamically adjusted to trigger local obstacle avoidance or global path replanning.

7. The method according to claim 1, characterized in that, The process of constructing and updating the dynamic obstacle probability map includes: Point cloud data of the warehouse environment is acquired by solid-state LiDAR. Ground segmentation and statistical filtering algorithms are applied to remove ground point cloud and noise interference, retain the effective points of obstacles, and extract the centroid coordinates, contour size and motion velocity vector of each obstacle. Simultaneously, image streams are acquired by depth cameras, obstacle categories are identified based on convolutional neural networks, and obstacle bounding boxes and confidence scores are calculated. During the operation of automated handling equipment, the fused obstacle status is periodically input into the spatiotemporal probability graph model at fixed time windows to update the obstacle existence probability of each grid node in real time. When the probability value of a node exceeds a preset threshold, the node is marked as a high collision risk area, and the passage weight of the corresponding edge in the topology network is dynamically adjusted to a high penalty value, triggering the execution of local obstacle avoidance strategy or global path replanning.

8. The method according to claim 1, characterized in that, The chemical pollution diffusion model integrates chemical property databases and environmental sensor data, collects instantaneous leakage model data to solve partial differential equations, calculates the spatiotemporal distribution and diffusion range of leaked pollutants in real time, and delineates isolation zones of different hazard levels based on the calculation results, generating dynamic electronic fences in the digital twin platform.

9. The method according to claim 1, characterized in that, The distributed leak detection network employs a multi-level distributed architecture to achieve real-time monitoring, specifically including: The distributed leak detection network adopts a multi-level distributed architecture, including: Point gas detectors, linear infrared beam detectors, and image-type infrared thermal imagers are distributed and deployed within the storage facilities to form a grid-like monitoring network. When any sensor detects a value exceeding a preset alarm threshold, the system automatically generates a leak source coordinate marker in the digital twin platform and activates a three-level alarm mechanism. The three-level alarm mechanism includes: Level 1 alarm triggers local audible and visual warnings; Level 2 alarm pushes information on the location and hazard level of the leak source to the warehouse management system; Level 3 alarm links the entire explosion suppression and protection system and triggers emergency braking and reverse gradient evacuation of all automated handling equipment.

10. The method according to claim 1, characterized in that, The mechanical isolation subsystem is used to automatically perform physical isolation operations when a leakage alarm is triggered, including but not limited to: Control the closure of automatic fireproof isolation doors or explosion-proof barriers deployed at the boundaries of warehouse zones; Activate the negative pressure exhaust system to directionally exhaust the leaked gas and close the air supply duct in the contaminated area; The inert gas explosion suppression device and mechanical covering mechanism work together to initially seal and suppress the leak point; After the central control system remote operators confirm that the area is in a safe state and authorize it, explosion-proof automated handling equipment is dispatched to carry emergency supplies to the leak point to assist in the leak plugging or relocation tasks.

11. The method according to claim 1, characterized in that, The data processing and early warning triggering process of the AI ​​early warning system includes: Through a multi-source heterogeneous data access layer, production planning and scheduling data from the production execution system, real-time order data from the order management system, current inventory data from the warehouse management system, and time-series consumption records of each raw material in the historical consumption database are captured in real time. The raw data is preprocessed by format normalization, outlier removal, and time alignment to form a standardized dataset. The time series prediction analysis module is trained based on a standardized dataset, and a raw material consumption prediction model is constructed using a long short-term memory network. The input features of the model include product type, output, and process route in the production plan, delivery cycle and priority in the order data, and seasonal fluctuations and trend changes in historical consumption data. The model output is the predicted demand for each raw material within the time window. During the model inference phase, real-time production plans and order data are input into the prediction model to calculate the predicted demand for each raw material in each time window. Simultaneously, the current available inventory and in-transit inventory are obtained from the warehouse management system to calculate the predicted shortage.

12. The method according to claim 11, characterized in that, The raw material consumption prediction model continuously collects the deviation between actual consumption data and predicted values ​​to construct a prediction error time series; the early warning threshold is dynamically adjusted based on the supplier's on-time delivery rate and the frequency of production plan changes. Establish a closed-loop mechanism for early warning feedback. When an early warning message is sent, record the response time, handling measures, and final result of the recipient.

13. The method according to claim 5, characterized in that, The inbound and outbound strategy also includes dynamic location scheduling based on the first-in-first-out principle or the preset expiration date management principle: For each pallet / box of materials, the entry time is recorded and the storage time is continuously monitored. The storage time is automatically identified, and the turnover priority is adaptively marked according to the storage time and the storage location of the materials. The handling path and retrieval time of the automated handling equipment for outbound operations are updated based on the turnover priority.

14. The method according to claim 5, characterized in that, The inbound and outbound strategies also include pre-defined scheduling strategies that coordinate with the upper-level production management system. Receive and parse production plans or outbound requirements from ERP or MES systems, and intelligently schedule the required materials from various storage locations to the reserve storage location or buffer area near the outbound outlet in advance and in batches; when the formal outbound instruction is issued, the materials are already in the ready-to-go state.

Images