Coordinated scheduling system of warehouse AGV and buffer rack based on real-time state monitoring

Abstract

The application discloses a warehouse AGV and buffer rack cooperative scheduling system based on real-time state monitoring and relates to the technical field of warehouse logistics automation and intelligent scheduling; through a multi-source state sensing unit, real-time operation data of the AGV and the lifting type buffer rack are collected, full-element digital twin body synchronous operation is driven after edge side cleaning and space-time alignment are completed, joint full-process pre-performance on a scheduling task is carried out by relying on a simulation deduction verification unit, space conflict, time sequence matching and time efficiency standard verification are completed, equipment instructions are issued by a scheduling execution unit according to an optimal time sequence, finally, model parameter and scheduling strategy are iteratively corrected based on virtual-real deviation by a closed-loop optimization unit, and complete closed-loop cooperative scheduling of perception, simulation, optimization, execution and feedback is realized.

Description

Technical Field

[0001] This invention relates to the field of warehouse logistics automation and intelligent scheduling technology, specifically a collaborative scheduling system for warehouse AGVs and buffer racks based on real-time status monitoring. Background Technology

[0002] In the warehousing operations of e-commerce retail and manufacturing enterprises, the automated operation mode combining lifting buffer racks and AGVs has been widely adopted. Enterprises typically issue inbound tasks through a warehouse management system, with independent AGV control systems and buffer rack electrical control systems respectively executing path planning and rack operation control. Some enterprises deploy digital twin systems for visualizing equipment status and conducting post-operation maintenance analysis. In daily operations, buffer racks serve as temporary storage, sorting buffers, and goods collection and turnover mechanisms, while AGVs handle cross-regional transfers and forklift retrieval. The efficiency of their collaboration directly determines the overall operational capacity of the warehousing process.

[0003] Under the existing independent scheduling model, AGV path planning and buffer rack operation control lack a joint verification process. The scheduling scheme is directly issued and executed, which easily leads to problems such as AGV arrival waiting, channel space interference, and operation timing misalignment during peak task periods. These issues require manual intervention and adjustments, significantly reducing operational efficiency. At the same time, the digital twin model uses fixed factory parameters, which will continuously generate virtual-real discrepancies with equipment wear, ground subsidence, and load changes. The reliability of simulation results gradually decreases, and adjustments to new scheduling rules and operation processes must rely on on-site testing and debugging. This not only has a long cycle and high trial-and-error costs, but also occupies normal operation capacity and cannot quickly generate feasible fault-tolerant scheduling schemes under abnormal operating conditions.

[0004] In summary, existing technologies cannot simultaneously achieve collaborative scheduling pre-verification of AGVs and buffer racks, adaptive iteration of twin models, and low-cost verification of scheduling strategies. They cannot meet the actual needs of efficient and stable operation of warehousing systems under order fluctuations, and have technical bottlenecks such as high scheduling conflict rate, rapid decay of model accuracy, and high cost of strategy iteration. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a collaborative scheduling system for warehouse AGVs and buffer racks based on real-time status monitoring. It can collect real-time operating data of AGVs and lifting buffer racks through multi-source status sensing units, and drive the synchronous operation of all-element digital twins after completing edge-side cleaning and spatiotemporal alignment. Relying on the simulation and verification unit, it conducts joint full-process pre-rehearsal of scheduling tasks, and completes the verification of spatial conflicts, timing matching and timeliness compliance. The scheduling execution unit issues equipment instructions according to the optimal timing. Finally, the closed-loop optimization unit iteratively corrects the model parameters and scheduling strategy based on the virtual-real deviation, realizing a complete closed-loop collaborative scheduling of perception, simulation, optimization, execution and feedback.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a collaborative scheduling system for warehouse AGVs and buffer racks based on real-time status monitoring, the system comprising the following components: Multi-source status perception unit: Deployed in the physical site of the warehouse, it collects real-time operating status data from the AGV end and the lifting buffer rack end respectively, and transmits the edge-side data to the digital twin modeling unit after completing edge-side data cleaning and spatiotemporal alignment. Digital twin modeling unit: Constructs a digital twin of all elements of the warehousing scenario, receives real-time status data to drive the twin to run synchronously with the physical system, and maintains a two-way mapping relationship at the geometric, physical, behavioral and rule levels; Simulation and verification unit: Receives the scheduling task to be executed, and simultaneously conducts a joint full-process pre-simulation of AGV driving path and buffer rack lifting operation in the digital twin. It completes the verification of spatial conflict, timing matching and timeliness compliance according to the set verification rules, and outputs the verified collaborative scheduling scheme. The scheduling execution unit breaks down the verified scheduling scheme into AGV motion instructions and buffer rack operation instructions, and sends them to the corresponding physical devices for execution according to the matching time sequence. Closed-loop optimization unit: Collects the deviation between the actual execution data of physical equipment and the simulation data, and iteratively optimizes the twin model parameters and scheduling algorithm rules in reverse to form a complete operation closed loop of perception-simulation-optimization-execution-feedback.

[0007] Furthermore, the multi-source state perception unit includes an AGV on-board perception module and a buffer rack in-situ perception module. The AGV on-board perception module integrates a laser SLAM positioning module, a six-axis attitude sensor, a power detection module, and a fork pressure sensor, which respectively collects the AGV's planar coordinates, driving speed, remaining battery power, fork height, and load weight data, with a collection frequency set to 20Hz. The buffer rack in-situ perception module deploys photoelectric occupancy sensors at each layer of the storage location, strain gauge deformation sensors at the column and beam positions, and an absolute encoder at the lifting mechanism, which respectively collects the storage location occupancy status, structural deformation, and lifting platform position data, with a collection frequency set to 10Hz. All perception data are connected to the edge computing gateway at the warehouse site. The gateway uses the industrial Ethernet protocol to complete data aggregation, supports parallel access of multiple devices, and the data transmission packet loss rate is less than 0.01%.

[0008] Furthermore, the edge computing gateway of the multi-source state perception unit performs data preprocessing and time synchronization. First, it removes abnormal data that exceeds 120% of the device's rated range. Then, it uses linear interpolation between consecutive frames to complete data that is missing for no more than 3 consecutive frames. Finally, based on the unified timestamp of BeiDou time synchronization and combined with the PTP precise time protocol, it interpolates and aligns the data of the two types of devices in 10ms time slices, controlling the spatiotemporal synchronization error within 8ms. After completing the format standardization processing, it is transmitted to the digital twin modeling unit.

[0009] Furthermore, the digital twin modeling unit adopts a layered architecture to construct a full-element twin. The geometric layer constructs a 1:1 scene model based on warehouse CAD drawings and on-site 3D scanning data, restoring the spatial dimensions and initial positions of site passages, columns, AGV bodies, and buffer racks. The spatial modeling error does not exceed ±5mm. The physical layer inputs the equipment's factory calibration parameters and corrects them by combining 100 sets of measured data under both no-load and full-load conditions. It constructs AGV driving dynamics models and buffer rack lifting dynamics models respectively. The behavior layer encapsulates the action logic of AGV straight movement, turning, forklift picking, and loading, as well as the operation logic of buffer rack lifting, location locking and unlocking, and location sorting. The action execution parameters correspond one-to-one with the physical equipment control parameters. The rules layer embeds task priority rules, equipment safety distance constraints, passage rules, and abnormal handling rules for warehouse operations. The layered architecture synchronously receives real-time status data to drive updates, and the overall model update frequency is not less than 10Hz, maintaining a real-time bidirectional mapping with the physical system.

[0010] Furthermore, the spatial conflict verification of the simulation and deduction verification unit adopts a spatiotemporal conflict risk quantification method, and calculates the conflict risk value between a single AGV and its corresponding buffer rack using the following formula. : ,in This is the real-time straight-line distance between the AGV and the buffer rack's working boundary, in meters. The equivalent safety radius of the AGV is the value of the equivalent radius after expanding the outline of the equipment body by 0.1m for safety margin, in meters. The equivalent safety radius of the buffer rack operating area is taken as the equivalent radius value after expanding the rack operating outline by 0.1m for safety margin, in meters. This represents the real-time driving speed of the AGV, expressed in m / s. The real-time movement speed of the buffer rack lifting mechanism is expressed in m / s. This refers to the maximum rated speed of equipment within the warehouse, expressed in m / s. For distance factor weights, The two weights, representing speed factors, are based on 1200 sets of conflict condition samples collected on-site at different distances and speeds. The target is defined by the risk value corresponding to the actual collision critical state, and the results are obtained through multiple linear regression analysis. Take 0.6, Take 0.4, when the calculated value is... When the value is ≥1, it is determined that there is a risk of spatial conflict. The system automatically adjusts the AGV's travel path and the timing of the buffer rack's operation, and recalculates the risk value until it reaches ≥1. <1. The timing matching verification extracts the estimated arrival time of the AGV and the estimated ready time of the buffer rack, calculates the time deviation value, sets the time window for ordinary inbound and outbound tasks to ±1.5s, and sets the time window for high-priority urgent orders to ±0.8s. If the deviation value is within the corresponding time window range, the timing matching is considered to be successful. If it exceeds the range, the AGV driving speed and acceleration / deceleration parameters are fine-tuned first, and the adjustment range is controlled within 10% of the rated parameters. If the second verification still fails, the start time of the buffer rack operation is adjusted. The number of timing adjustments for a single task shall not exceed 3 times. If the upper limit is exceeded, the task reassignment process is triggered.

[0011] Furthermore, the simulation and verification unit supports abnormal working condition pre-simulation. The multi-source state perception unit collects equipment operation data in real time and classifies it into three levels according to the degree of abnormality: minor, moderate, and severe. Minor abnormalities include AGV remaining power in the range of 15%-20% and buffer rack deformation of 80%-90% of the rated limit. Moderate abnormalities include AGV remaining power in the range of 10%-15% and buffer rack deformation of 90%-95% of the rated limit and AGV driving deviation exceeding twice the rated value. Severe abnormalities include AGV remaining power below 10%, buffer rack deformation exceeding the rated limit, and equipment shutdown alarm. After detecting an abnormal warning signal, the system injects the corresponding abnormal parameters into the digital twin and starts the abnormal diffusion simulation and deduction to simulate the impact range and duration of the abnormality on the surrounding operation links. Simultaneously, it generates a fault-tolerant scheduling scheme of the corresponding level, including task transfer path, backup equipment replacement logic, and fault area isolation rules. After the scheme passes the simulation verification, it is stored as a backup scheme and is directly called and executed when the corresponding abnormality is actually triggered. The scheduling response time under abnormal working conditions is controlled within 3 seconds.

[0012] Furthermore, the simulation and verification unit supports offline verification of scheduling strategies. When adding new scheduling rules, adjusting storage location allocation rules, or expanding the number of devices, the corresponding parameters and scenario conditions are directly configured in the digital twin. A test set is constructed, including three typical scenarios: off-peak operation, peak operation, and abnormal working conditions. Each scenario has no less than 1,000 sets of standard test tasks. A full-cycle simulation test is launched for 7 consecutive days, 24 hours a day. The system throughput, equipment conflict rate, equipment utilization rate, and average duration of a single task are automatically collected throughout the process. After the test is completed, the operation indicators are compared with those of the existing strategy. When the system throughput of the new strategy increases by more than 5% and the equipment conflict rate decreases by more than 10%, it is considered to have passed the verification. The verified strategy is deployed in a gray-scale manner, first running for 72 hours in a quarter of the warehouse area. After the operation is stable and there are no abnormalities, it is gradually expanded to the entire area. The verification cycle of strategy adjustment is shortened by more than 80% compared with the on-site test and debugging method.

[0013] Furthermore, the scheduling execution unit is equipped with an instruction timing matching module. After receiving the verified scheduling scheme, the module decomposes the scheme into independent AGV movement instructions and buffer rack operation instructions. Based on the optimal time node obtained from simulation, the module sends lifting and positioning instructions and cargo unlocking instructions to the target buffer rack 2 seconds in advance. It receives the status feedback signal of the buffer rack in real time. After confirming that the lifting platform is in place and the cargo locking mechanism has been unlocked, it sends driving and forklift operation instructions to the corresponding AGV. During the instruction issuance process, the operating status of the equipment is continuously monitored. If the buffer rack experiences a temporary operation delay, a deceleration and waiting instruction is simultaneously sent to the AGV. If the AGV experiences a path delay, a standby hold instruction is simultaneously sent to the buffer rack to ensure that when the AGV arrives at the target cargo location, the buffer rack has completed all operation preparations. The connection waiting time between the two is controlled within 1 second. The instruction transmission adopts the industrial Ethernet protocol, and the single instruction issuance delay is controlled within 100ms.

[0014] Furthermore, the closed-loop optimization unit is equipped with a twin model parameter adaptive correction mechanism, which calculates the corrected dynamic parameters using the following formula. : ,in To correct the current dynamic parameters of the pre-twin model, The virtual-to-real deviation rate for the current single task is calculated by dividing the absolute value of the difference between the actual operating value of the physical equipment and the simulation value by the simulation value. This is the gain coefficient for single-shot deviation correction. This represents the average virtual-to-real deviation rate from the actual data in the last 10 similar tasks. This is a historical bias smoothing factor. and Based on three consecutive months of operational data from the field, with a sample size of no less than 500 groups for each task type, and aiming to minimize the global average virtual-to-real deviation, calibration was performed through iterative fitting using gradient descent. Take 0.72, The parameter is set to 0.28. When the virtual-to-real deviation rate of three consecutive tasks of the same type exceeds 5%, the parameter correction process is automatically triggered. The corrected parameters are directly updated to the physical layer of the digital twin modeling unit for use in all subsequent simulation and deduction processes. The parameter correction process is completed automatically during off-peak business hours without downtime, gradually reducing the virtual-to-real mapping error and improving the simulation accuracy of the twin model.

[0015] Furthermore, the closed-loop optimization unit is embedded in a deep reinforcement learning scheduling policy network, employing the following multi-objective comprehensive reward function. Guiding strategy iteration: ,in The average completion time for a single task, in seconds. The baseline task completion time is set, and the average completion time of similar tasks over the past 30 days is taken as the benchmark. This represents the total energy consumption for a single task, expressed in J. The preset baseline single-task energy consumption is the average energy consumption of similar tasks over the past 30 days. The number of device conflicts within a single decision step. This represents the maximum number of allowed conflicts within a single decision step. As a weight for timeliness indicators, As the weight of energy efficiency indicators, As for the conflict penalty weights, a third-order judgment matrix is ​​constructed by combining the three types of weights with the requirement of prioritizing warehousing operation timeliness. This matrix is ​​calculated using the analytic hierarchy process and determined after consistency testing. Take 0.5, Take 0.3, Taking 0.2, all three sub-items are mapped to the dimensionless interval of 0-1. During the daily off-peak period, the system calls the historical scheduling data and simulation deviation data of the day to train the strategy network. The new strategy after training is first verified by full simulation in the digital twin. If the performance indicators meet the standards, the existing strategy is automatically replaced and put online. No downtime is required throughout the process. The scheduling strategy is continuously iterated and optimized to adapt to the dynamic changes in the warehousing business.

[0016] Compared with existing technologies, this warehouse AGV and buffer rack collaborative scheduling system based on real-time status monitoring has the following advantages: I. This system uses a digital twin to conduct joint simulation and pre-run of AGVs and lifting buffer racks. Before issuing instructions, it completes spatial conflict and timing matching verification, avoiding equipment interference and operation waiting problems from the source. This can significantly reduce the occurrence rate of scheduling conflicts, shorten the equipment connection waiting time, and improve the overall throughput efficiency of warehouse inbound and outbound operations.

[0017] Second, this system constructs a closed-loop optimization mechanism for virtual and real data, which adaptively corrects the dynamic parameters of the twin model based on actual operating data. It supports offline simulation verification and gray-scale deployment of scheduling strategies, which improves the accuracy of twin simulation, significantly shortens the strategy iteration cycle, reduces on-site debugging costs, and enhances the system's adaptability to business fluctuations.

[0018] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is an overall flowchart of a collaborative scheduling system for warehouse AGVs and buffer racks based on real-time status monitoring. Figure 2 This is a flowchart of the joint pre-simulation and verification process for a collaborative scheduling system of warehouse AGVs and buffer racks based on real-time status monitoring. Figure 3 This is a flowchart illustrating the parameter correction and deep reinforcement learning scheduling strategy iteration of the closed-loop optimization unit model for a collaborative scheduling system of warehouse AGVs and buffer racks based on real-time status monitoring. Detailed Implementation

[0021] 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 below.

[0022] This embodiment is applied to a piecemeal picking buffer storage scenario in an e-commerce regional fulfillment center. The total area of ​​this buffer area is approximately 6000㎡, equipped with 28 stealth forklift AGVs and 18 sets of six-layer lifting buffer racks. Its main functions include temporary storage, consolidation, and outbound transfer of picked goods. During daily operations, there are significant differences between order peaks and troughs, with a high density of inbound and outbound tasks during peak periods. Under the original independent scheduling mode, equipment conflicts were frequent, and connection waiting times were high, making it difficult to meet fulfillment timeliness requirements. This embodiment deploys a collaborative scheduling system for warehouse AGVs and lifting buffer racks based on real-time status monitoring. This achieves full-domain awareness of equipment status, twin simulation pre-verification, and virtual-real closed-loop optimization, improving the operational efficiency and stability of the buffer area.

[0023] The multi-source state perception unit is deployed and data collected in two ways: AGV vehicle-mounted and in-situ buffer rack. A laser SLAM positioning module is installed on the top of each AGV, outputting planar coordinates and heading angle data at a frequency of 20Hz, with a positioning accuracy of ±10mm. A six-axis attitude sensor is installed at the vehicle's center of gravity to collect data on driving acceleration, yaw angle, and vibration amplitude. A battery circuit is connected to a power detection module to read the remaining power, output voltage, and current data in real time. Strain gauge pressure sensors are installed at the root of the forks to collect the total load weight and load distribution on the left and right forks, identifying off-center loading. A through-beam photoelectric occupancy sensor with a response time ≤5ms is installed at the front of each layer of each lifting buffer rack to detect the presence and movement of goods. A strain gauge deformation sensor with a resolution of 1με is attached to the middle of the column and the lower surface of the crossbeam to collect real-time data on the vertical deformation of the column and the flexural deformation of the crossbeam. The drive motor of the lifting mechanism is equipped with a multi-turn absolute encoder with a position accuracy of ±1mm, providing real-time feedback on the height and running speed of the lifting platform.

[0024] All sensing devices are connected to the on-site edge computing gateway. One edge gateway is configured for every six buffer racks. The gateway is connected to an industrial Ethernet ring network with a transmission bandwidth of 100Mbps. The gateway first performs data preprocessing, removing abnormal data exceeding 120% of the device's rated range. For example, if the AGV's rated travel speed is 2m / s, speed data exceeding 2.4m / s is directly identified as abnormal and removed. For data missing no more than three consecutive frames, linear interpolation between preceding and following frames is used to complete the data, calculating the corresponding value for the missing time based on the time proportion. After preprocessing, the gateway uses the second pulse signal output by the BeiDou timing module, combined with the PTP precise time protocol, to synchronize the clocks of all devices. Then, it interpolates and resamples data from different acquisition frequencies in fixed 10ms time slices to generate a unified timestamp state dataset. The measured spatiotemporal synchronization error is controlled within 8ms. The standardized data is transmitted to the digital twin modeling unit via the MQTT protocol.

[0025] The digital twin modeling unit is deployed on the twin server in the warehouse control center, employing a layered architecture to construct a full-element digital twin. The geometric layer first imports the warehouse CAD floor plan, marking the aisle boundaries, column coordinates, and buffer rack installation reference points. Then, it combines this with point cloud data collected from on-site 3D laser scanning to recreate the 3D scene of the site at a 1:1 scale. The 3D models of the AGVs and buffer racks are imported from the original equipment manufacturer's models, aligned according to on-site reference points, with an overall spatial modeling error not exceeding ±5mm. The physical layer first inputs the equipment's factory calibration parameters, including the AGV's total weight of 200kg, rated load of 500kg, initial wheel friction coefficient of 0.6, and the buffer rack's rated load per location of 100kg, rated speed of the lifting mechanism of 0.5m / s, and structural elastic modulus of 206GPa. After parameter input, on-site calibration is performed, collecting 100 sets of AGV acceleration and deceleration operation data under both no-load and full-load conditions. The actual ground friction coefficient, acceleration and deceleration response time, and other parameters are fitted to correct the AGV's driving dynamics model. Similarly, 100 sets of buffer rack lifting operation data were collected under both no-load and full-load conditions to correct the start-stop acceleration and deceleration curves and load response parameters of the lifting mechanism.

[0026] The behavior layer breaks down the AGV's straight-moving, turning, picking up, and placing actions into independent action units. The input parameters of each action unit correspond one-to-one with the control commands of the AGV controller. For example, straight-moving actions correspond to target speed and travel distance parameters, while picking up actions correspond to fork rise height and forward depth parameters. The lifting, locking / unlocking, and sorting actions of the buffer rack are also broken down into corresponding action units, and the action execution logic is completely matched with the operating logic of the buffer rack's electronic control program. The rules layer embeds task priority rules, equipment safety distance constraints, passageway rules, and exception handling rules for warehouse operations. Task priority decreases progressively in the order of outbound orders, inbound replenishment, and inventory sorting, with urgent orders set as the highest priority. The four-layer architecture synchronously receives real-time status data to drive updates, with the overall model update frequency not less than 10Hz, maintaining a real-time bidirectional mapping with the physical system.

[0027] The simulation and verification unit runs on a digital twin server. After receiving the task set from the warehouse management system, it breaks down a single batch of tasks into AGV transportation sub-tasks and buffer rack operation sub-tasks, imports them into the digital twin, and starts discrete event simulation with a simulation step size of 100ms. In the spatial conflict verification stage, the spatiotemporal conflict risk quantification method is used to calculate the conflict risk value K between a single AGV and its corresponding buffer rack. The equivalent safety radius of the AGV is taken as a safety margin of 0.1m beyond the radius of the circumscribed circle of its body contour, which is 0.7m in this embodiment. The equivalent safety radius of the buffer rack operation area is taken as a safety margin of 0.1m beyond the radius of the rack operation contour, which is 1.3m in this embodiment. The distance factor weight α and the speed factor weight β are obtained through multiple linear regression analysis based on 1200 sets of conflict and near-conflict condition samples collected on-site at different distances and movement speeds, with the collision critical state corresponding to K=1 as the calibration target. α is set to 0.6, and β to 0.4. When the calculated K≥1, it is determined that there is a risk of spatial conflict. The system first adjusts the AGV travel path to increase the straight distance between devices. After adjustment, the risk value is recalculated until K<1.

[0028] During the timing matching verification phase, the estimated arrival time after AGV path planning and the estimated readiness time of the buffer rack operation are extracted synchronously during simulation, and the time deviation between the two is calculated. The time window width is dynamically set according to the task level. The operation fluctuation range of ordinary inbound and outbound tasks is approximately ±1.5s, and the corresponding time window is set to ±1.5s. High-priority urgent orders have higher requirements for timeliness accuracy, and the time window is narrowed to ±0.8s. When the time deviation value is within the corresponding time window range, the timing matching is considered successful. If it exceeds the range, the AGV travel speed and acceleration / deceleration parameters are fine-tuned first, and the adjustment range is controlled within 10% of the rated parameters to avoid interfering with other tasks. If the speed adjustment still fails to match, the start timing of the buffer rack operation is adjusted accordingly. The number of timing adjustments for a single task does not exceed 3 times. If the limit is exceeded, the task reassignment process is triggered to avoid invalid loops.

[0029] The abnormal operating condition simulation function is triggered by status monitoring data. The system classifies equipment status into three levels: minor, moderate, and severe, based on the degree of abnormality. A minor abnormality is defined as AGV remaining power between 15% and 20% and buffer rack deformation between 80% and 90% of the rated limit. A moderate abnormality is defined as AGV remaining power between 10% and 15%, buffer rack deformation between 90% and 95% of the rated limit, and AGV travel deviation exceeding twice the rated value. A severe abnormality is defined as AGV remaining power below 10%, buffer rack deformation exceeding the rated limit, and equipment triggering a shutdown alarm. After detecting an abnormality warning signal, the system injects the corresponding abnormality parameters into the digital twin, initiates an abnormality propagation simulation, simulates the impact range and duration of the abnormality on surrounding work links, and simultaneously generates a corresponding level of fault-tolerant scheduling scheme, including task transfer paths, backup equipment replacement logic, and fault area isolation rules. After the scheme passes simulation verification, it is stored as a backup scheme and directly invoked when the corresponding abnormality is actually triggered. The scheduling response time under abnormal operating conditions is controlled within 3 seconds.

[0030] The scheduling execution unit is deployed on the on-site industrial control computer, connecting to both the AGV control system and the buffer rack electrical control system. The unit includes an instruction timing matching module. After receiving a verified scheduling plan, the module breaks it down into independent AGV movement instructions and buffer rack operation instructions. Based on the optimal time node obtained from simulation, the module issues lifting and positioning instructions and location unlocking instructions to the target buffer rack 2 seconds in advance. This lead time is determined based on the average lifting and positioning time of 1.5 seconds for the buffer rack, with a 0.5-second margin. The module receives real-time status feedback signals from the buffer rack. After confirming that the lifting platform is in place and the location locking mechanism has unlocked, it issues driving and forklift operation instructions to the corresponding AGV. During instruction issuance, the system continuously monitors the equipment's operating status. If the buffer rack experiences a temporary operational delay, a deceleration and waiting instruction is simultaneously issued to the AGV, controlling it to remain in a safe, low-speed standby area. If the AGV encounters a path obstacle causing a delay in arrival, a standby hold instruction is simultaneously issued to the buffer rack to maintain the lifting platform in position. In this embodiment, the waiting time for the AGV to connect with the buffer rack is controlled within 1 second, and the transmission of a single instruction adopts the industrial Ethernet protocol, with the issuance delay controlled within 100ms.

[0031] The closed-loop optimization unit employs an adaptive correction mechanism for the twin model parameters. After a single batch of tasks is completed, the unit collects actual operating data from the physical equipment, including the actual AGV travel time, actual acceleration and deceleration parameters, and actual lifting and lowering time of the buffer rack. This data is compared point-by-point with the corresponding data from the simulation to calculate the virtual-to-real deviation rate Δe for a single task. To avoid incorrect parameter correction due to occasional fluctuations, the parameter correction process is automatically triggered only when the virtual-to-real deviation rate for three consecutive tasks of the same type exceeds 5%. The corrected dynamic parameters P' are calculated using an adaptive correction formula. The single-time deviation correction gain coefficient γ and the historical deviation smoothing factor δ are calibrated based on three consecutive months of on-site operating data. The effective sample size for each type of task is no less than 500 sets. The optimization objective is to minimize the global average virtual-to-real deviation, and the parameters are obtained through gradient descent iterative fitting, with γ set to 0.72 and δ set to 0.28. The corrected parameters are directly updated to the physical layer of the digital twin modeling unit for use in all subsequent simulation processes. The parameter correction process is completed automatically during the daily off-peak hours without downtime intervention, which can gradually reduce the virtual-real mapping error and improve the simulation accuracy of the twin model.

[0032] A deep reinforcement learning scheduling strategy network is embedded within the unit, employing a multi-objective comprehensive reward function to guide strategy iteration. All three components of the reward function are normalized using benchmark values: the benchmark task completion time (t_ref) is the average completion time of similar tasks over the past 30 days; the benchmark single-task energy consumption (e_ref) is the average energy consumption of similar tasks over the past 30 days; and the maximum allowed number of conflicts per step decision (n_max) is set to 2, mapping all three components to a dimensionless interval of 0 to 1. A three-order judgment matrix is ​​constructed based on the timeliness priority requirements of warehousing operations (λ1, λ2, and λ3), calculated using the analytic hierarchy process and determined after consistency verification. λ1 is set to 0.5, λ2 to 0.3, and λ3 to 0.2. The system trains the strategy network daily during off-peak hours using historical scheduling data and simulation deviation data. New strategies are first fully simulated and verified in a digital twin. Once performance indicators meet the standards, the existing strategy is automatically replaced and deployed, without any downtime, enabling continuous iterative optimization of the scheduling strategy.

[0033] The offline verification function for scheduling strategies is used for testing and verifying newly added rules. When adding scheduling rules, adjusting storage location allocation rules, or expanding the number of devices, the corresponding parameters and scenario conditions are directly configured in the digital twin. A test set is constructed, including three typical scenarios: off-peak operations, peak operations, and abnormal operating conditions. Each scenario has no fewer than 1,000 standard test tasks. A full-cycle simulation test is launched continuously for 7 days, 24 hours a day, automatically collecting four core indicators: system throughput, device conflict rate, device utilization rate, and average duration of a single task. After the test, the operating indicators are compared with those of the existing strategy. When the system throughput of the new strategy increases by more than 5% and the device conflict rate decreases by more than 10%, it is considered to have passed the verification. The verified strategy is deployed in a gray-scale manner, first running for 72 hours in a quarter of the warehouse area. After stable operation without any abnormalities, it is gradually expanded to the entire area.

[0034] In summary, this embodiment fully realizes the collaborative scheduling of AGVs and lifting buffer racks by deploying multi-source sensing hardware, constructing a layered digital twin, performing joint simulation verification, and matching instruction timing with a closed-loop optimization mechanism within an e-commerce order picking and caching warehousing scenario. The system can complete dual verification of spatial conflicts and timing matching before instruction issuance, reducing the occurrence rate of scheduling conflicts and shortening equipment connection waiting time from the source. Simultaneously, relying on a closed-loop system of virtual and real data to continuously correct model parameters and iterate scheduling strategies, it supports offline simulation verification and canary deployment, improving operational efficiency while reducing strategy adjustment costs. It can effectively adapt to warehousing operation needs under order fluctuations and has a clear engineering implementation path and practical application value.

[0035] 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 warehouse AGV and buffer rack collaborative scheduling system based on real-time status monitoring, characterized in that, The system includes the following components: Multi-source status perception unit: Deployed in the physical site of the warehouse, it collects real-time operating status data from the AGV end and the lifting buffer rack end respectively, and transmits the edge-side data to the digital twin modeling unit after completing edge-side data cleaning and spatiotemporal alignment. Digital twin modeling unit: Constructs a digital twin of all elements of the warehousing scenario, receives real-time status data to drive the twin to run synchronously with the physical system, and maintains a two-way mapping relationship at the geometric, physical, behavioral and rule levels; Simulation and verification unit: Receives the scheduling task to be executed, and simultaneously conducts a joint full-process pre-simulation of AGV driving path and buffer rack lifting operation in the digital twin. It completes the verification of spatial conflict, timing matching and timeliness compliance according to the set verification rules, and outputs the verified collaborative scheduling scheme. The scheduling execution unit breaks down the verified scheduling scheme into AGV motion instructions and buffer rack operation instructions, and sends them to the corresponding physical devices for execution according to the matching time sequence. Closed-loop optimization unit: Collects the deviation between the actual execution data of physical equipment and the simulation data, and iteratively optimizes the twin model parameters and scheduling algorithm rules in reverse.

2. The warehouse AGV and buffer rack collaborative scheduling system based on real-time status monitoring according to claim 1, characterized in that, The multi-source state perception unit includes an AGV on-board perception module and a buffer rack in-situ perception module. The AGV on-board perception module integrates a laser SLAM positioning module, a six-axis attitude sensor, a power detection module, and a fork pressure sensor, which respectively collect data on the AGV's planar coordinates, travel speed, remaining battery power, fork height, and load weight. The buffer rack in-situ perception module deploys photoelectric occupancy sensors at each layer of the storage space, strain gauge deformation sensors at the column and beam positions, and an absolute encoder at the lifting mechanism, which respectively collect data on the storage space occupancy status, structural deformation, and lifting platform position. All perception data are connected to the edge computing gateway at the warehouse site, and the gateway uses the industrial Ethernet protocol to complete data aggregation.

3. The warehouse AGV and buffer rack collaborative scheduling system based on real-time status monitoring according to claim 1, characterized in that, The edge computing gateway of the multi-source state perception unit performs data preprocessing and time synchronization. First, it removes abnormal data that exceeds 120% of the device's rated range. Then, it uses linear interpolation between consecutive frames to complete data that is missing for no more than 3 consecutive frames. Based on the unified timestamp of BeiDou time synchronization and combined with the PTP precise time protocol, it interpolates and aligns the data of the two types of devices in 10ms time slices, controlling the spatiotemporal synchronization error within 8ms. After completing the format standardization processing, it is transmitted to the digital twin modeling unit.

4. The warehouse AGV and buffer rack collaborative scheduling system based on real-time status monitoring according to claim 1, characterized in that, The digital twin modeling unit adopts a layered architecture to construct a full-element twin. The geometric layer constructs a 1:1 scene model based on warehouse CAD drawings and on-site 3D scanning data, restoring the spatial dimensions and initial positions of site passages, columns, AGV bodies, and buffer racks. The physical layer inputs the equipment's factory calibration parameters and performs corrections by combining 100 sets of measured data under both no-load and full-load conditions on-site, respectively constructing AGV driving dynamics models and buffer rack lifting dynamics models. The behavior layer encapsulates the action logic of AGV straight movement, turning, forklift picking, and goods placement, as well as the operation logic of buffer rack lifting, cargo location locking and unlocking, and cargo location organization. The action execution parameters correspond one-to-one with the physical equipment control parameters. The rules layer embeds task priority rules for warehouse operations, equipment safety distance constraints, passage rules, and abnormal handling rules. The layered architecture synchronously receives real-time status data to drive updates.

5. The warehouse AGV and buffer rack collaborative scheduling system based on real-time status monitoring according to claim 1, characterized in that, The spatial conflict verification of the simulation and deduction verification unit adopts the spatiotemporal conflict risk quantification method, and calculates the conflict risk value between a single AGV and its corresponding buffer rack using the following formula. : ,in This represents the real-time straight-line distance between the AGV and the buffer rack's working boundary. The equivalent safety radius of the AGV. The equivalent safety radius of the buffer rack operating area. This refers to the real-time driving speed of the AGV. For the real-time movement speed of the buffer rack lifting mechanism, This refers to the maximum rated speed of the equipment within the warehouse. For distance factor weights, The weights are for the velocity factor.

6. The warehouse AGV and buffer rack collaborative scheduling system based on real-time status monitoring according to claim 1, characterized in that, The simulation and verification unit supports abnormal working condition pre-simulation function. The multi-source state perception unit collects equipment operation data in real time and classifies the abnormality into three levels: slight, moderate and severe. After detecting an abnormality warning signal, the system injects the corresponding abnormal parameters into the digital twin and starts the abnormality diffusion simulation. It simulates the impact range and duration of the abnormality on the surrounding operation links and generates a corresponding level of fault-tolerant scheduling scheme, including task transfer path, backup equipment replacement logic and fault area isolation rules. After the scheme passes the simulation verification, it is stored as a backup scheme and is directly called and executed when the corresponding abnormality is actually triggered.

7. The warehouse AGV and buffer rack collaborative scheduling system based on real-time status monitoring according to claim 1, characterized in that, The simulation and verification unit supports offline verification of scheduling strategies. When adding new scheduling rules, adjusting the location allocation rules, or expanding the number of equipment, the corresponding parameters and scenario conditions are directly configured in the digital twin. A test set containing three typical scenarios—off-peak operation, peak operation, and abnormal working conditions—is constructed. A full-cycle simulation test is launched, lasting 24 hours a day for 7 consecutive days. Four core indicators—system throughput, equipment conflict rate, equipment utilization rate, and average duration of a single task—are automatically collected throughout the process. After the test is completed, the operating indicators are compared with those of the existing strategy. When the system throughput of the new strategy increases by more than 5% and the equipment conflict rate decreases by more than 10%, it is considered to have passed the verification. The verified strategy is deployed in a gray-scale manner, first running for 72 hours in a quarter of the warehouse area. After stable operation without any abnormalities, it is gradually expanded to the entire area.

8. The warehouse AGV and buffer rack collaborative scheduling system based on real-time status monitoring according to claim 1, characterized in that, The scheduling execution unit is equipped with an instruction timing matching module. After receiving the verified scheduling scheme, the module decomposes the scheme into independent AGV movement instructions and buffer rack operation instructions. Based on the optimal time node obtained from simulation, the module sends lifting and positioning instructions and cargo unlocking instructions to the target buffer rack 2 seconds in advance. It receives the status feedback signal of the buffer rack in real time. After confirming that the lifting platform is in place and the cargo locking mechanism has been unlocked, it sends driving and forklift operation instructions to the corresponding AGV. During the instruction issuance process, the operating status of the equipment is continuously monitored. If the buffer rack experiences a temporary operation delay, a deceleration and waiting instruction is sent to the AGV simultaneously. If the AGV experiences a path delay, a standby hold instruction is sent to the buffer rack simultaneously. The instruction transmission adopts the industrial Ethernet protocol.

9. The warehouse AGV and buffer rack collaborative scheduling system based on real-time status monitoring according to claim 1, characterized in that, The closed-loop optimization unit is equipped with a twin model parameter adaptive correction mechanism, and the corrected dynamic parameters are calculated using the following formula. : ,in To correct the current dynamic parameters of the pre-twin model, This represents the virtual-to-real deviation rate for the current single task. This is the gain coefficient for single-shot deviation correction. This represents the average virtual-to-real deviation rate from the actual data in the last 10 similar tasks. This is the historical bias smoothing factor.

10. The warehouse AGV and buffer rack collaborative scheduling system based on real-time status monitoring according to claim 1, characterized in that, The closed-loop optimization unit is embedded in a deep reinforcement learning scheduling policy network and employs the following multi-objective comprehensive reward function. Guiding strategy iteration: ,in This represents the average completion time for a single task. The preset baseline task completion time, This refers to the overall energy consumption for a single task. The preset baseline single-task energy consumption, The number of device conflicts within a single decision step. This represents the maximum number of allowed conflicts within a single decision step. As a weight for timeliness indicators, As the weight of energy efficiency indicators, Conflict penalty weights.

Images