Automatic warehousing system and sorting method thereof

Abstract

The invention discloses an automatic warehousing system and a sorting method thereof, and the system comprises an execution module, a sensing module and a digital twin management module: the execution module is composed of a vertical lifting container, an autonomous mobile robot and a sorting machine; the sensing module collects cargo full life cycle data through a visual sensor, an RFID tag and a reader-writer. The digital twinborn management module constructs twinborn bodies of the cargos and the equipment, dynamically classifies the cargos based on real-time data, automatically responds to inventory abnormity and simulates and generates an optimal warehousing and sorting route. According to the sorting method, data are collected through the Internet of Things and the RFID technology, goods are classified in real time through a fuzzy clustering algorithm, the digital twinborn body serves as a training environment, a multi-device collaborative optimal sorting path is generated through a reinforcement learning model, and dynamic adjustment is conducted according to the real-time state. According to the scheme, intelligent management and efficient sorting of the warehousing system are achieved, the inventory management precision and the equipment cooperation efficiency are improved, and the system is suitable for high-frequency and high-flexibility warehousing scenes.

Description

Technical Field

[0001] The present invention relates to the field of warehousing, and in particular to an automated warehousing system and a sorting method thereof. Background Art

[0002] Automated warehousing systems, as the core infrastructure of modern logistics, are currently widely used in e-commerce, manufacturing, and other fields. Existing technologies primarily utilize warehouse management systems (WMS) combined with automated equipment (such as AGVs and stacker cranes) to achieve cargo storage and sorting. For example, traditional systems use RFID tags or barcodes to identify goods, utilize fixed-path algorithms (such as the Dijkstra algorithm) to plan equipment movement routes, and allocate storage locations based on preset rules (such as the ABC classification method). However, these systems generally rely on static rules and historical experience, lacking the ability to dynamically respond to real-time data, and equipment collaborative scheduling primarily relies on manually preset logic, making it difficult to cope with complex scenarios such as order fluctuations and equipment failures.

[0003] Existing automated warehousing systems have exposed the following core issues in practical applications: First, inventory management is insufficiently refined: Traditional sorting methods (such as fixed ABC classification) cannot respond in real time to changes in goods turnover, resulting in the mixing of high-frequency and low-frequency goods. This leads to cross-regional sorting, resulting in low efficiency. Second, route planning lacks flexibility: Sorting routes generated by fixed algorithms fail to incorporate real-time equipment status (such as AGV battery level and sorter load), leading to equipment congestion or idle runs. This is particularly true during peak order periods, such as sales promotions, where sorting times fluctuate significantly. Third, exception handling is not automated enough: Inventory anomalies (such as out-of-stocks and slow sales) rely on manual inspections, while equipment failures (such as sudden downtime of stacker cranes) require manual intervention, leading to delayed or even interrupted order fulfillment. Fourth, multi-device collaboration is limited: Different devices (such as AGVs and sorters) lack dynamic linkage mechanisms, and task allocation relies on fixed priority rules, making it impossible to achieve globally optimal scheduling. These issues make existing systems unable to meet the flexible and intelligent requirements of modern warehousing in terms of efficiency, stability, and cost control. Summary of the Invention

[0004] The main purpose of the present invention is to provide an automated warehousing system and a sorting method thereof, aiming to.

[0005] The technical solutions of the present invention are as follows: Automated warehousing system, including: An execution module, which carries out the handling and sorting of goods according to instructions, includes a vertical lift container, an autonomous mobile robot, and a sorter. The vertical lift container is used to store one or more types of goods, the autonomous mobile robot is used to realize the automated handling of goods, and the sorter is used to sort and transport goods. The perception module collects data on the entire life cycle of goods and includes visual sensors, RFID tags, fixed RFID readers, and mobile RFID readers. The RFID tags are placed on the goods, the visual sensors and fixed RFID readers are placed at different heights of the vertical lift container, and the mobile RFID reader is placed on the automatic mobile robot. The digital twin management module constructs digital twins of goods and equipment, simulates and generates optimal warehousing and sorting routes, and transmits instructions to the execution module. The digital twin includes at least a dynamic portrait unit of goods and an inventory self-healing unit. The dynamic portrait unit classifies goods in real time based on the data of the entire life cycle of goods, and the inventory self-healing unit automatically responds to inventory anomalies based on preset rules.

[0006] In one possible implementation, the cargo lifecycle data includes basic attributes and real-time status; The basic attributes include size, weight, shelf life, and turnover rate; The real-time status includes storage location, in and out time, and adjacent goods.

[0007] In a possible implementation, in the dynamic portrait unit, the categories of real-time classification of goods include: High-frequency dynamic items are preferentially stored on the bottom floor of vertical lift containers, with automatic mobile robots configuring high-frequency access paths; Medium frequency buffers are stored in the middle layer of vertical lift containers, taking into account both efficiency and space utilization; Low-frequency static items are stored on the upper floors of vertical lift containers, freeing up sorting channel resources. The classification results are automatically updated every 24 hours, and specific classifications are made based on promotions, seasonal changes, and other needs.

[0008] In one possible embodiment, the dynamic portrait unit includes a thermal migration mechanism, which is triggered when the turnover rate of goods exceeds a set threshold for three consecutive days. The thermal migration mechanism will vertically lift the bottom layer of the container with the cargo migration value that triggers the threshold during the off-peak period at night, and update the storage location information in the digital twin model at the same time.

[0009] In one possible implementation, the digital twin management module generates an optimal sorting path by running a reinforcement learning model simulation. The reinforcement learning model is trained based on a comprehensive state space consisting of the cargo status mapped by the cargo digital twin, the real-time operating data fed back by the equipment twin, and the order demand characteristics, to achieve global optimization of multi-device collaborative paths.

[0010] In one possible implementation, the reinforcement learning model automatically reviews the sorting data of the previous day at a fixed time point every day, calculates the deviation between the reinforcement learning model planned path and the actual execution path, and updates the parameters.

[0011] In a possible implementation, the inventory self-healing module includes: Inventory health calculation unit generates warnings based on safety stock warning lines and turnover rate abnormality thresholds; The risk simulation and deduction unit simulates out-of-stock or slow-selling conditions for the warning SKU and triggers at least one self-healing action, which includes internal inventory transfer, automatic generation of replenishment orders, location status marking, and task interception. The SKU is the smallest unit of goods with a unique electronic tag.

[0012] In one possible implementation, the safety stock warning line is the average daily sales volume of the corresponding SKU multiplied by the preset number of days. If the corresponding SKU inventory is less than or equal to the safety stock warning line, a yellow warning is triggered. The turnover rate abnormality threshold is that when the current turnover rate of the corresponding SKU drops by more than the preset value compared with the turnover rate of the previous week, a blue warning is triggered.

[0013] In one possible implementation, for SKUs that trigger yellow alerts, a stock-out simulation is performed to predict the number of days in the future. If inventory is consumed based on historical sales averages, the remaining inventory is calculated. If the remaining inventory is less than the preset value, a self-healing process is triggered. The self-healing process prioritizes internal inventory transfers. If there is no internal inventory, a replenishment order is automatically generated and the sorting plan is updated simultaneously. For SKUs that trigger a blue alert, a slow-moving simulation is performed to predict sales trends for a preset number of days in the future. If the sales trend continues to be low, the product will be automatically moved to a higher level of a vertical lift container.

[0014] The sorting method of the automated warehousing system as described above comprises the following steps: The IoT sensors and RFID technology are used to collect data on the entire life cycle of goods, including their attributes, real-time status, and historical turnover rates. Fuzzy clustering algorithms are then used to classify goods in real time based on this data. Build a digital twin of the warehouse environment that includes cargo location, equipment status, and order requirements. The digital twin maps the physical warehouse system in real time. The digital twin is used as a training environment for a reinforcement learning model. Parameters such as the order task queue and equipment load status are input to generate an optimal sorting path that includes multi-device collaboration. The reward function of the reinforcement learning model comprehensively considers order completion time, equipment utilization, energy consumption cost, and sorting error rate. The autonomous mobile robot is dispatched to perform the picking task according to the optimal sorting path, and the sorting progress is monitored in real time. When the equipment fails or the order priority changes, the incremental training of the reinforcement learning model is triggered to dynamically update the sorting path.

[0015] The working principle and beneficial effects of the present invention are: 1. The technical solution of the present invention uses the perception module to collect real-time data on the entire life cycle of goods (including location, turnover rate, and equipment status), and constructs dynamic cargo portraits and equipment twins through the digital twin management module, realizing real-time classification of goods (such as high-frequency dynamic categories and low-frequency static categories) and automatic adjustment of storage locations, solving the problem of low cross-region sorting efficiency caused by traditional fixed classification.

[0016] 2. The technical solution of the present invention uses a digital twin management module to simulate and generate optimal warehousing and sorting routes based on a reinforcement learning algorithm, dynamically optimizes the path by integrating real-time order demand, equipment load and other parameters, and schedules execution modules (autonomous mobile robots, sorting machines) to work together to achieve global optimal planning and dynamic obstacle avoidance of multiple equipment paths, solving the problems of insufficient flexibility in fixed algorithm path planning and empty running due to equipment congestion.

[0017] 3. The inventory self-healing unit automatically triggers inventory anomaly responses (such as internal transfers and replenishment order generation) based on preset rules (such as safety inventory warning lines and turnover rate thresholds), and combines with the equipment twin to monitor faults in real time (such as vertical lift container anomalies). The linkage execution module automatically adjusts task allocation to achieve unmanned processing of inventory anomalies and equipment failures, solving the problems of delayed manual inspections and untimely abnormal responses.

[0018] 4. Through the digital twin management module, the various devices in the execution module (vertical lift containers, autonomous mobile robots, and sorting machines) are uniformly dispatched, tasks are dynamically matched based on real-time data (such as prioritizing high-frequency goods to mobile robots), and the device collaboration strategy is continuously optimized through the reinforcement learning model to achieve intelligent multi-device task allocation and load balancing, solving the problem of low collaboration efficiency caused by traditional fixed priority rules. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying any creative work.

[0020] Figure 1 This is a structural block diagram of the automated warehousing system in Example 1; Figure 2Flowchart of the sorting method in Example 2.

[0021] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solutions and advantages of this application more clearly understood, the present application is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application. Example

[0023] like Figure 1 As shown, this embodiment proposes an automated warehousing system, including: The execution module carries out the handling and sorting of goods according to instructions, including vertical lifting containers, autonomous mobile robots, and sorting machines. The vertical lifting containers are used to store one or more types of goods, the autonomous mobile robots are used to realize the automated handling of goods, and the sorting machines are used to sort and transport goods.

[0024] The execution module, as the physical execution unit of the automated warehousing system, receives instructions from the digital twin management module and completes the entire process from storage to sorting. Vertical lift containers provide high-density storage space, supporting the compartmentalized storage of multiple categories of goods; autonomous mobile robots are responsible for the dynamic transportation of goods between storage and sorting areas; and sorting machines accurately classify goods and transport them to designated shipping ports. The execution module utilizes the high-density storage of vertical lift containers to improve storage space utilization, optimizes transportation paths through the dynamic navigation of autonomous mobile robots, and improves sorting efficiency through the precise classification of sorting machines. The three functions work together to greatly improve sorting efficiency, while also supporting flexible operations to adapt to order fluctuations, reducing reliance on manual labor, and improving safety in high-risk environments.

[0025] The perception module collects data on the entire life cycle of goods, including visual sensors, RFID tags, fixed RFID readers and mobile RFID readers. RFID tags are set on the goods, visual sensors and fixed RFID readers are set at different heights of the vertical lifting container, and mobile RFID readers are set on the automatic mobile robot.

[0026] The perception module connects the physical and digital worlds, collecting real-time data on cargo location, status, turnover, and equipment operating parameters through multiple sensor types. RFID tags uniquely identify cargo and, in conjunction with fixed readers (on each container level) and mobile readers (on robots), enable full-process cargo tracking. Visual sensors (such as high-definition cameras and 3D vision) monitor anomalies such as cargo stacking within containers and material jams in sorting machines. The perception module accurately tracks and synchronizes data on cargo throughout its lifecycle, providing real-time data support for dynamic inventory management and effectively improving inventory accuracy. Visual monitoring can detect anomalies such as cargo tipping and equipment jams in advance, providing early warning compared to manual inspections. Synchronized equipment status data forms a health profile, supporting predictive maintenance and effectively reducing equipment failure rates.

[0027] The digital twin management module builds digital twins of goods and equipment, simulates and generates optimal warehousing and sorting routes, and passes instructions to the execution module. The digital twin includes at least a dynamic portrait unit of goods and an inventory self-healing unit. The dynamic portrait unit classifies goods in real time based on the data of the entire life cycle of goods, and the inventory self-healing unit automatically responds to inventory anomalies based on preset rules.

[0028] The digital twin management module is used to construct digital twins of goods (including dynamic profiling units and inventory self-healing units) and equipment based on sensory data, optimizing physical-world operations through digital spatial simulation. The dynamic profiling unit automatically categorizes goods based on real-time data (e.g., high-frequency dynamic, medium-frequency buffer, and low-frequency static). The inventory self-healing unit pre-sets rules and generates optimal warehouse layouts and sorting routes using reinforcement learning algorithms. This has the beneficial effects of dynamically classifying goods and automatically adjusting storage locations, significantly shortening picking routes; monitoring inventory anomalies in real time and triggering automatic responses, significantly improving replenishment efficiency; optimizing multi-device collaborative routes through reinforcement learning, reducing robot idleness and energy costs; and simulating equipment operating status to proactively avoid failures, improving system stability and risk management capabilities.

[0029] In this embodiment, the cargo lifecycle data includes basic attributes and real-time status; Basic attributes include size, weight, shelf life, and turnover rate.

[0030] Basic attributes, as inherent characteristic data of goods, are used to support the underlying planning and strategy formulation of the warehousing system. Dimensions and weight are used to match container space, plan stacking methods, and select handling equipment specifications; expiration dates are used to implement first-in-first-out (FIFO) strategies and inventory alerts; and turnover rates are used to differentiate between high-frequency and low-frequency goods and optimize warehouse allocation (such as placing high-frequency goods in easily accessible areas). Through basic attribute data, the system can achieve refined allocation of warehousing resources, namely, adapting container specifications to the size and weight of the goods to avoid wasted space or overloading equipment; automatically triggering near-expiry alerts and prioritizing outbound shipments based on expiration dates to reduce inventory losses; and dynamically adjusting warehouse layouts based on turnover rates to shorten picking routes, improve inbound and outbound efficiency, and reduce the error rate of manual sorting.

[0031] Real-time status includes storage location, in and out time, and adjacent goods.

[0032] Real-time status data reflects the dynamic state of goods in the warehousing process, supporting the system's real-time monitoring and intelligent scheduling. Storage locations are used to accurately locate goods, enabling rapid sorting and inventory. Inbound and outbound times are used to calculate goods turnover efficiency and analyze traffic patterns. Adjacent goods data is used to avoid conflicts in stacking goods (such as avoiding odor cross-contamination and the risk of squeezing) and optimize space utilization. Using real-time status data, the system can construct a dynamic inventory map, making goods locations transparent, improving inventory efficiency and reducing manual errors. Traffic peaks are analyzed based on inbound and outbound times, allowing equipment and manpower to be dispatched in advance to address order fluctuations. Monitoring the status of adjacent goods automatically avoids category conflicts and ensures storage safety. Intelligent stacking algorithms can also be used to improve shelf space utilization and reduce warehousing operating costs.

[0033] In this embodiment, in the dynamic portrait unit, the categories of real-time classification of goods include: High-frequency dynamic items are preferentially stored on the bottom floor of vertical lift containers and are equipped with high-frequency access paths for automatic mobile robots.

[0034] Frequently moving goods enter and exit the warehouse frequently. By prioritizing storage on the bottom floor of vertical lift containers and matching them with the high-frequency access paths of autonomous mobile robots, this effectively shortens picking and handling distances. This reduces ineffective movement of goods back and forth between upper floors, significantly improving the efficiency of high-frequency goods entering and exiting the warehouse, and reducing equipment energy consumption. The rapid response capabilities of the bottom-level storage locations directly shorten order fulfillment times, making them particularly suitable for high-frequency scenarios such as e-commerce promotions and instant delivery.

[0035] Medium frequency buffer type is stored in the middle layer of the vertical lifting container, taking into account both efficiency and space utilization.

[0036] Medium-frequency buffer cargo, with moderate inbound and outbound frequency, is stored in the middle layer of the container (the "golden area" that balances efficiency and space), balancing operational efficiency and storage density. This design prevents the middle layer from being idle or occupied by low-frequency cargo, improving overall storage utilization. Cargo stored at this height requires less frequent lifting, reducing mechanical wear and extending the life of equipment. It also provides flexibility for temporary, high-frequency orders, such as pre-sales stocking.

[0037] Low-frequency static items are stored on the upper floors of vertical lift containers to free up sorting channel resources.

[0038] Low-frequency, static goods (such as seasonal products and long-tail inventory) are stored in the upper levels of the container, leveraging the low density of the upper levels to free up the lower-level sorting aisles. This frees up the lower aisles for high-frequency goods, thus avoiding congestion in the hardest-hit areas.

[0039] The classification results are automatically updated every 24 hours, and specific classifications are made based on promotions, seasonal changes, and other needs.

[0040] Specialty classification breaks conventional classification logic based on external needs (such as holiday promotions and seasonal product changes) and temporarily adjusts the priority of specific product categories. For example, promotional items can be temporarily reclassified as "high-frequency dynamic" or seasonal items can be preemptively reclassified from "low-frequency static" to the middle or bottom tier. Setting specialty classification allows for agile response to market changes, reduces temporary operating costs, and enables dynamic resource reuse.

[0041] In this embodiment, the dynamic portrait unit includes a thermal migration mechanism, which is triggered when the turnover rate of goods exceeds the set threshold for three consecutive days. The thermal migration mechanism will vertically lift the bottom layer of the container with the cargo migration value that triggers the threshold in the scheduling execution module during the low-peak period at night, and update the storage location information in the digital twin model at the same time.

[0042] The hot migration mechanism monitors changes in cargo turnover in real time and automatically triggers when the turnover of a specific item exceeds a set threshold for three consecutive days. Its core function is to dynamically respond to changes in cargo popularity. During the nighttime off-peak period, the scheduling execution module migrates high-frequency goods to the lower levels of the vertical lift container (a high-efficiency operation area), while simultaneously updating the storage location information in the digital twin model. This enables dynamic reallocation of warehouse resources and avoids the lag of manual intervention. The hot migration mechanism significantly improves the sorting efficiency of high-frequency goods through automated location adjustments. The convenience of the lower levels shortens the picking paths of autonomous mobile robots, reducing equipment congestion during peak daytime hours. Executing migration operations during the nighttime off-peak period avoids disrupting normal order processing and ensures operational continuity. Real-time interaction with the digital twin model ensures synchronized inventory data updates, avoiding manual inventory errors. The mechanism also proactively responds to market demand changes (such as sudden hot items), improving the agility of the warehouse system through predictive adjustments. Ultimately, this mechanism achieves a comprehensive optimization of sorting time, reduced equipment energy consumption, and inventory management accuracy.

[0043] In this embodiment, the digital twin management module generates the optimal sorting path by running a reinforcement learning model simulation. The reinforcement learning model is trained based on a comprehensive state space composed of the cargo status mapped by the cargo digital twin, the real-time operation data fed back by the equipment twin, and the order demand characteristics, to achieve global optimization of the collaborative path of multiple devices.

[0044] By constructing a comprehensive state space encompassing cargo status (e.g., classification, storage location), equipment operating data (e.g., robot position, sorter load), and order requirements (e.g., priority, batch size), the reinforcement learning model simulates and generates optimal sorting paths for multi-device collaboration, aiming for global optimization. This replaces traditional rule-based scheduling logic, enabling an upgrade from local optimization of a single device to dynamic collaboration across the entire system. By continuously learning from historical data and providing real-time feedback, the reinforcement learning model dynamically adapts to complex scenarios such as order fluctuations and equipment status changes, significantly reducing robot idle runs and the probability of path conflicts, thereby improving sorting efficiency. Furthermore, path planning based on a global perspective balances equipment loads, reduces wear on mechanical components, and extends equipment life, ultimately achieving multi-dimensional optimization of the warehousing system's energy consumption, efficiency, and stability.

[0045] The global optimization of multi-device collaborative paths includes the following steps: Data collection and modeling integrate digital twin data for goods (e.g., real-time classification, shelf life, and turnover rate), equipment twin data (e.g., robot battery status and sorter fault warnings), and order data (e.g., SKU combinations and delivery time requirements) to construct a multidimensional state space. Data is standardized to eliminate dimensional differences (e.g., converting cargo volume and robot speed into normalized metrics) to ensure consistency of model inputs.

[0046] Reward function design defines core optimization objectives (e.g., minimizing total sorting time and maximizing equipment load balancing). A reward mechanism is designed: efficient sorting route planning is rewarded positively, while idle runs, congestion, and equipment overload are penalized negatively. Dynamic weighting factors are introduced to adjust optimization priorities based on business needs (e.g., prioritizing sorting efficiency during promotional events and prioritizing equipment wear and tear control during daily operations).

[0047] Model training and simulation utilizes deep reinforcement learning algorithms (such as PPO and DQN) to simulate equipment collaboration in different order scenarios within a digital twin environment. Path strategies are optimized through a "trial-error-feedback-iteration" process. Offline training utilizes historical order data, followed by online fine-tuning using real-time data feeds to ensure the model adapts to dynamic warehouse environments.

[0048] Strategy verification and deployment: Virtually verify the generated sorting paths in the digital twin management module to test robustness in extreme scenarios such as multi-device obstacle avoidance and task preemption. Small-scale pilot runs collect actual operational data, compare model predictions with actual efficiency, and then gradually roll out the system-wide implementation after parameter adjustments.

[0049] Continuous optimization and feedback loops are established, with a real-time monitoring mechanism established. This collects actual sorting route execution data (such as completion time and equipment failure rate) and compares it with the model's predicted values ​​to generate error feedback. Regularly updating the training dataset (such as incorporating new SKU features and equipment upgrade parameters) triggers model retraining to ensure optimal scheduling capabilities over the long term.

[0050] In this embodiment, the reinforcement learning model automatically reviews the sorting data of the previous day at a fixed time point every day, calculates the deviation between the planned path of the reinforcement learning model and the actual execution path, and updates the parameters.

[0051] The reinforcement learning model automatically reviews the previous day's sorting data at a fixed time each day to calculate the deviation between the model's planned path and the actual execution path (such as path adjustments caused by temporary equipment failures or changes in order priorities). Its core function is to calibrate the model parameters using real-world operational data, forming a closed loop of "training-execution-feedback-optimization." This ensures that the model strategy continuously adapts to the dynamic changes in the warehouse site and avoids the accumulation of planning errors due to environmental differences. By quantifying deviation data (such as path length errors and task completion time errors), the automatic review mechanism can specifically optimize the model's reward function weights and state space parameters, making the subsequently planned sorting paths more closely aligned with actual operational scenarios and gradually improving sorting efficiency.

[0052] In this embodiment, the inventory self-healing module includes: The inventory health calculation unit generates early warnings based on the safety stock warning line and turnover rate abnormality threshold.

[0053] The Inventory Health Calculation Unit monitors SKU inventory counts and turnover rates in real time, comparing them to pre-set safety stock warning lines and turnover rate anomaly thresholds (e.g., 50% below the historical average). This automatically identifies potential out-of-stock and slow-moving risks and generates early warnings, providing a decision-making basis for inventory self-healing and replacing the traditional passive monitoring model of periodic manual inspections. This unit enables real-time perception and proactive early warning of inventory risks, shortening the early warning response time from "once a day" for manual inspections to "minutes," thus avoiding order shortages or backlogs caused by delayed detection of inventory anomalies. Standardized threshold settings and automated calculations reduce manual judgment errors, improve the accuracy of inventory health assessments, and provide a reliable foundation for subsequent self-healing actions.

[0054] The risk simulation and deduction unit simulates out-of-stock or slow-moving conditions for the warning SKUs and triggers at least one self-healing action, which includes internal inventory transfer, automatic generation of replenishment orders, location status marking, and task interception. SKU is the smallest unit of goods with a unique electronic tag.

[0055] The risk simulation unit uses historical sales data and warehousing rules to simulate the impact of out-of-stock or slow-moving SKUs generated by the inventory health calculation unit (such as order delays caused by out-of-stock conditions and increased warehousing costs due to slow-moving items). Based on the simulation results, the system automatically triggers self-healing actions such as internal inventory transfers, replenishment order generation, location status marking (such as freezing locations for slow-moving items), and task interception (suspending the allocation of new warehousing tasks to locations for slow-moving items), automating the entire process from risk warning to resolution. By pre-simulating the evolution paths of different risk scenarios, the system can select the optimal self-healing strategy (such as prioritizing the allocation of inventory from nearby warehouses rather than emergency purchases), effectively reducing the cost of handling exceptions.

[0056] In this embodiment, the safety stock warning line is the average daily sales volume of the corresponding SKU multiplied by the preset number of days. If the corresponding SKU inventory is less than or equal to the safety stock warning line, a yellow warning is triggered.

[0057] The safety stock warning line uses a quantitative formula of "average daily sales x preset number of days" to set a dynamic inventory safety boundary for each SKU. When the inventory level falls below or equals the warning line, a yellow alert is automatically triggered. Its core function is to identify potential out-of-stock risks in advance, reserve buffer time for replenishment decisions, and avoid order fulfillment interruptions caused by inventory depletion. The standardized warning line calculation method shifts inventory management from "empirical judgment" to "data-driven", improving the accuracy of warnings. Early triggering of yellow alerts can shorten replenishment cycles, ensure that in-transit inventory gaps are filled in a timely manner, and reduce out-of-stock rates compared to traditional manual monitoring models. Dynamic adaptation to the sales characteristics of different SKUs (for example, the preset number of days can be adjusted by category) avoids out-of-stocks of high-frequency products or backlogs of low-frequency products caused by universal warning lines, thereby improving overall inventory turnover efficiency.

[0058] The turnover rate abnormality threshold is that when the current turnover rate of the corresponding SKU drops by more than the preset value compared with the turnover rate of the previous week, a blue warning is triggered.

[0059] The turnover rate anomaly threshold compares the current SKU turnover rate with the previous week's data. A blue alert is triggered when the decline exceeds a preset value. This identifies SKUs at risk of slowing sales due to a sudden drop in sales. Its core function is to promptly capture changes in market demand or fluctuations in product competitiveness, providing early signals for inventory strategy adjustments. Real-time monitoring of turnover rate fluctuations can proactively identify slow-selling trends, shortening the warning cycle compared to traditional monthly sales analysis. A triggered blue alert automatically triggers the system to initiate slow-selling processing (such as prioritizing allocation to promotional areas or adjusting inventory locations to higher levels), shortening the turnover period for slow-selling inventory. By calibrating preset thresholds based on historical data (e.g., allowing greater fluctuations for seasonal products), the system can reduce false alarms, prioritize inventory resources toward high-turnover categories, and improve warehouse space utilization.

[0060] In this embodiment, for SKUs that trigger a yellow alert, an out-of-stock simulation is performed to predict the number of days in the future if the inventory is consumed according to the historical sales average. The number of days that the remaining inventory can support is calculated. If the number of days that can be supported is less than the preset value, the self-healing process is triggered. The self-healing process gives priority to internal inventory transfer. If there is no inventory internally, a replenishment order is automatically generated and the sorting plan is updated simultaneously.

[0061] For SKUs that trigger yellow alerts, out-of-stock simulations are used to predict the number of days inventory can last. If the number is lower than a preset value, internal inventory transfers (such as transferring from other storage areas or warehouses) are prioritized. When there is no internal inventory, replenishment orders are automatically generated, and the sorting plan is adjusted simultaneously (such as prioritizing the consumption of inventory that is about to expire). Its core role is to respond to out-of-stock risks in a layered manner, prioritize solving problems through low-cost internal resource allocation, avoid inventory backlogs caused by blind purchases, and ensure coordination between sorting operations and replenishment rhythms. Out-of-stock simulations enable forward-looking quantitative assessments of inventory risks, shifting replenishment decisions from "ordering based on experience" to "driven by data predictions," improving replenishment accuracy. The priority internal transfer mechanism reduces reliance on external procurement, reduces logistics costs, and responds faster to transfers than external procurement. Dynamic adjustments to the sorting plan avoid out-of-stock interruptions during picking, improving order fulfillment efficiency and significantly increasing customer satisfaction.

[0062] For SKUs that trigger a blue alert, a slow-moving simulation is performed to predict sales trends for a preset number of days in the future. If the sales trend continues to be low, the product will be automatically moved to a higher level of a vertical lift container.

[0063] For SKUs that trigger a blue alert, future sales trends are analyzed through slow-moving simulations. If sales continue to be sluggish, the goods are automatically moved to the upper levels of vertical lift containers. Its core function is to promptly release high-efficiency storage resources on the lower levels, reduce the occupation of high-frequency operation areas by slow-moving products, and at the same time, reduce ineffective operations during sorting through physical isolation, freeing up storage space for potential high-frequency goods. Slow-moving simulations identify long-term low-turnover goods in advance, preventing them from occupying prime storage locations for a long time and improving storage utilization. The automated operation of moving goods to the upper levels does not require human intervention, improving processing efficiency and reducing the risk of human error. The centralized management of high-level storage locations facilitates the subsequent unified execution of promotions, returns, and other batch processing, shortening the turnover cycle of slow-moving products and reducing warehouse operating costs. Example

[0064] like Figure 2 As shown, this embodiment proposes a sorting method for an automated warehousing system, comprising the following steps: The entire life cycle data of goods is collected through IoT sensors and RFID technology. The data includes goods attributes, real-time status and historical turnover rate. Based on the fuzzy clustering algorithm, goods are classified in real time according to the entire life cycle data.

[0065] The collection of data throughout the entire life cycle enables the system to accurately perceive the status of goods, and real-time classification ensures that high-frequency goods are always in the efficient operation area, shortening the search and transportation distance during sorting, and improving sorting efficiency; dynamic classification avoids the waste of storage resources caused by traditional fixed classification, making the warehouse layout more in line with actual business needs and reducing the frequency and cost of manual adjustments.

[0066] Build a digital twin of the warehouse environment that includes cargo location, equipment status, and order requirements. The digital twin maps the physical warehouse system in real time.

[0067] The real-time mapping capability of the digital twin enables the system to comprehensively monitor the warehouse's operating status, promptly detect potential anomalies (such as impending equipment failure and storage congestion), and improve the timeliness of risk warnings; the virtual environment provides a low-cost trial-and-error space for subsequent path optimization, avoiding the efficiency loss caused by direct verification in the physical system and enhancing the reliability of decision-making.

[0068] The digital twin is used as the training environment for the reinforcement learning model. Parameters such as the order task queue and equipment load status are input to generate an optimal sorting path that involves multi-device collaboration. The reward function of the reinforcement learning model comprehensively considers order completion time, equipment utilization, energy consumption cost, and sorting error rate.

[0069] The global optimization capability of the reinforcement learning model can comprehensively balance order timeliness, equipment energy consumption, and sorting accuracy, reduce equipment idleness and path conflicts, and improve equipment utilization; multi-device collaborative path planning makes the sorting process smoother, reduces operation interruptions caused by unreasonable scheduling, and improves the overall stability and execution efficiency of the warehousing system.

[0070] Dispatch autonomous mobile robots to perform pickup tasks according to the optimal sorting path, monitor the sorting progress in real time, and trigger incremental training of the reinforcement learning model when equipment fails or order priority changes, and dynamically update the sorting path.

[0071] Automated execution reduces manual intervention and improves sorting efficiency and consistency; real-time monitoring and dynamic path update mechanisms can quickly respond to abnormal situations such as equipment failures and order changes, avoiding task stalls caused by sudden problems and ensuring the continuity of the sorting process; incremental training allows the model to adapt to new scenarios without full retraining, shortening adjustment time and enhancing the system's agility and adaptability.

[0072] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, techniques, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results.

[0073] Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interaction systems, and mechatronics. AI software technologies primarily encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0074] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing related hardware using computer-readable instructions. The computer-readable instructions can be stored in a computer-readable storage medium. When executed, the program can include the processes in the above-described method embodiments. The aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0075] It should be understood that although the steps in the flowcharts of the accompanying drawings are shown in sequence as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some of the steps in the flowcharts of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a portion of the sub-steps or stages of other steps.

[0076] The above are only preferred embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent replacements, and improvements made within the spirit and principles of the present application should be included in the scope of protection of the present application.

Claims

1. Automated warehousing system, characterized by: include: An execution module, which carries out the handling and sorting of goods according to instructions, includes a vertical lift container, an autonomous mobile robot, and a sorter. The vertical lift container is used to store one or more types of goods, the autonomous mobile robot is used to realize the automated handling of goods, and the sorter is used to sort and transport goods. The perception module collects data on the entire life cycle of goods and includes visual sensors, RFID tags, fixed RFID readers, and mobile RFID readers. The RFID tags are placed on the goods, the visual sensors and fixed RFID readers are placed at different heights of the vertical lift container, and the mobile RFID reader is placed on the automatic mobile robot. The digital twin management module constructs digital twins of goods and equipment, simulates and generates optimal warehousing and sorting routes, and transmits instructions to the execution module. The digital twin includes at least a dynamic portrait unit of goods and an inventory self-healing unit. The dynamic portrait unit classifies goods in real time based on the data of the entire life cycle of goods, and the inventory self-healing unit automatically responds to inventory anomalies based on preset rules.

2. The automated warehousing system according to claim 1, characterized in that: The cargo life cycle data includes basic attributes and real-time status; The basic attributes include size, weight, shelf life, and turnover rate; The real-time status includes storage location, in and out time, and adjacent goods.

3. The automated warehousing system according to claim 1, characterized in that: In the dynamic portrait unit, the categories of real-time classification of goods include: High-frequency dynamic items are preferentially stored on the bottom floor of vertical lift containers, with automatic mobile robots configuring high-frequency access paths; Medium frequency buffers are stored in the middle layer of vertical lift containers, taking into account both efficiency and space utilization; Low-frequency static items are stored on the upper floors of vertical lift containers, freeing up sorting channel resources. The classification results are automatically updated every 24 hours, and specific classifications are made based on promotions, seasonal changes, and other needs.

4. The automated warehousing system according to claim 1 or 3, characterized in that: The dynamic portrait unit includes a thermal migration mechanism, which is triggered when the cargo turnover rate exceeds a set threshold for three consecutive days. The thermal migration mechanism will vertically lift the cargo migration value that triggers the threshold in the scheduling execution module during the low-peak period at night, and update the storage location information in the digital twin model at the same time.

5. The automated warehousing system according to claim 1, characterized in that: The digital twin management module generates the optimal sorting path by running a reinforcement learning model simulation. The reinforcement learning model is trained based on a comprehensive state space composed of the cargo status mapped by the cargo digital twin, the real-time operation data fed back by the equipment twin, and the order demand characteristics, to achieve global optimization of the multi-device collaborative path.

6. The automated warehousing system according to claim 5, characterized in that: The reinforcement learning model automatically reviews the sorting data of the previous day at a fixed time point every day, calculates the deviation between the planned path of the reinforcement learning model and the actual execution path, and updates the parameters.

7. The automated warehousing system according to claim 1, characterized in that: The inventory self-healing module includes: Inventory health calculation unit generates warnings based on safety stock warning lines and turnover rate abnormality thresholds; The risk simulation and deduction unit simulates out-of-stock or slow-selling conditions for the warning SKU and triggers at least one self-healing action, which includes internal inventory transfer, automatic generation of replenishment orders, location status marking, and task interception. The SKU is the smallest unit of goods with a unique electronic tag.

8. The automated warehousing system according to claim 7, characterized in that: The safety stock warning line is the average daily sales volume of the corresponding SKU multiplied by the preset number of days. If the corresponding SKU inventory is less than or equal to the safety stock warning line, a yellow warning is triggered; The turnover rate abnormality threshold is that when the current turnover rate of the corresponding SKU drops by more than the preset value compared with the turnover rate of the previous week, a blue warning is triggered.

9. The automated warehousing system according to claim 8, characterized in that: For SKUs that trigger yellow alerts, out-of-stock simulation is performed to predict the number of days in the future. If inventory is consumed based on the historical sales average, the remaining inventory is calculated. If the remaining inventory is less than the preset value, the self-healing process is triggered. The self-healing process prioritizes internal inventory transfer. If there is no internal inventory, a replenishment order is automatically generated and the sorting plan is updated simultaneously. For SKUs that trigger a blue alert, a slow-moving simulation is performed to predict sales trends for a preset number of days in the future. If the sales trend continues to be low, the product will be automatically moved to a higher level of a vertical lift container.

10. The sorting method for an automated warehousing system according to any one of claims 1 to 9, characterized in that: The steps include: The IoT sensors and RFID technology are used to collect data on the entire life cycle of goods, including their attributes, real-time status, and historical turnover rates. Fuzzy clustering algorithms are then used to classify goods in real time based on this data. Build a digital twin of the warehouse environment that includes cargo location, equipment status, and order requirements. The digital twin maps the physical warehouse system in real time. The digital twin is used as a training environment for a reinforcement learning model. Parameters such as the order task queue and equipment load status are input to generate an optimal sorting path that includes multi-device collaboration. The reward function of the reinforcement learning model comprehensively considers order completion time, equipment utilization, energy consumption cost, and sorting error rate. The autonomous mobile robot is dispatched to perform the picking task according to the optimal sorting path, and the sorting progress is monitored in real time. When the equipment fails or the order priority changes, the incremental training of the reinforcement learning model is triggered to dynamically update the sorting path.

Images