A warehouse global adaptive intelligent task scheduling system and method based on capability tags

Abstract

This invention discloses a warehouse global adaptive intelligent task scheduling system and method based on capability tags, aiming to solve the problems of poor matching between equipment capabilities and tasks, weak prediction of path conflicts, and insufficient dynamic adjustment capabilities. The method includes: collecting multi-dimensional data on equipment, tasks, and the environment, and converging it in a unified backend; preprocessing the data to construct a prediction and risk identification model and generate risk warnings; the global scheduling layer establishes operational capability tags for each piece of equipment based on risk predictions and new tasks, initially screening candidate equipment, optimizing matching and task allocation in conjunction with global status, and dynamically planning work paths to generate scheduling and path instructions; the equipment execution layer sends instructions to the control system to drive equipment operation; and the closed-loop feedback layer continuously collects data and iteratively optimizes model parameters and scheduling rules. This invention achieves accurate matching, improving collaborative efficiency and system stability.

Description

Technical Field

[0001] This invention relates to the field of intelligent warehouse scheduling technology, and in particular to a warehouse global adaptive intelligent task scheduling system and method based on capability tags. Background Technology

[0002] With the booming development of e-commerce logistics and intelligent manufacturing, intelligent warehousing systems have become a core component in improving supply chain efficiency. Traditional warehouse management relies heavily on manual experience for location allocation and task scheduling, resulting in low efficiency, high error rates, and low space utilization. Although some warehouses have introduced warehouse management systems (WMS) to achieve information-based management of storage locations and inventory, location allocation often relies solely on simple rules such as first-in-first-out (FIFO) and fixed zoning, without fully considering key factors such as material turnover rate, storage environment requirements, and inbound / outbound frequency.

[0003] Furthermore, the application of Automated Guided Vehicles (AGVs) in warehousing is becoming increasingly widespread, enabling automated material handling and reducing labor costs. However, in multi-AGV collaborative operation scenarios, existing path planning methods cannot accurately predict path conflicts, leading to frequent collisions and congestion between AGVs, thus reducing overall handling efficiency. This is particularly true in high-density storage warehouses, where the number of AGVs is large, path planning is complex, and system response speed is slow, making it difficult to meet the demands of rapidly changing dynamic environments. While some improvement solutions have been proposed in the industry, they are mostly limited to single-vehicle path planning and obstacle avoidance, lacking multi-vehicle collaborative scheduling, dynamic load-speed mapping, and global deadlock coordination mechanisms. Some systems have attempted scheduling based on cloud computing, but response latency exists, making it difficult to meet millisecond-level response requirements for AGV obstacle avoidance and dynamic traffic management. Therefore, a method is needed that can adjust scheduling strategies based on dynamic changes in tasks and real-time system status to achieve efficient coordination and load balancing of equipment resources, thereby improving the overall operating efficiency and stability of the system. Summary of the Invention

[0004] In view of the shortcomings of the prior art, the purpose of this invention is to provide a warehouse global adaptive intelligent task scheduling system and method based on capability tags, which aims to solve the technical problems of existing warehouse scheduling systems that lack a comprehensive quantitative and refined matching mechanism for equipment capabilities, lack a global collaborative scheduling and path conflict prediction mechanism, and lack real-time risk prediction and dynamic scheduling adjustment capabilities when multiple devices are working together.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: The first aspect of this invention provides a warehouse global adaptive intelligent task scheduling method based on capability tags, comprising the following steps: real-time acquisition of equipment data, task data, and scenario environment data through a multi-dimensional data acquisition layer; data processing and risk prediction through a data preprocessing and risk prediction layer; equipment matching, task allocation, and path planning through a global intelligent scheduling decision layer; driving warehouse equipment to perform operations through an equipment execution layer; and finally, optimizing model parameters through a closed-loop feedback iteration layer.

[0006] Furthermore, the process of matching equipment and allocating tasks based on the material attributes, timeliness requirements, and operational constraints of the tasks to be performed specifically includes: firstly, initial screening of equipment based on capability tags, and then secondary optimization matching based on global equipment load, spatial distance, and path status. This two-stage equipment matching mechanism ensures the accuracy of task allocation and overall efficiency.

[0007] Preferably, before generating scheduling instructions and path instructions, the process further includes: comprehensively considering the urgency of time, material value, storage urgency, and equipment adaptability to calculate task priorities. This multi-dimensional task priority calculation ensures that high-priority tasks are processed first.

[0008] Specifically, after receiving risk prediction results and newly issued tasks through the global intelligent scheduling decision layer, the system automatically performs task splitting, merging of similar tasks, and priority scheduling of urgent orders based on task attributes and system status. This task logic optimization improves the flexibility and efficiency of task processing.

[0009] Furthermore, the time-series prediction model is used to predict the task traffic in each region over a future period, and the congestion risk identification model is used to identify the probability of channel congestion. A risk warning signal is generated when the predicted task traffic exceeds a regional processing capacity threshold or the congestion probability exceeds a set threshold. Through time-series prediction and congestion risk identification, proactive risk identification and early warning are achieved.

[0010] Preferably, the dynamic path planning based on the real-time location of the device specifically includes: planning the optimal path using the A* algorithm by combining the current location of the device, the urgency of the task, the width of the work area passage, and the distribution of obstacles; and using an improved ant colony algorithm for task allocation when the device receives multiple tasks simultaneously, wherein the pheromone update rule is based on the total task completion time and the number of path conflicts. Through the collaborative application of the A* algorithm and the improved ant colony algorithm, single-device path optimization and multi-device global collaboration are achieved.

[0011] Specifically, the iterative optimization of the time-series prediction model, operations research algorithm parameters, and capability label determination threshold minimizes the mean square error between the model's prediction accuracy and the actual operating data. Through iterative optimization, the system's scheduling accuracy and response capability in dynamic environments are improved.

[0012] The second aspect of this invention provides a warehouse global adaptive intelligent task scheduling system based on capability tags, comprising a multi-dimensional data acquisition layer, a data preprocessing and risk prediction layer, a global intelligent scheduling decision layer, an equipment execution layer, and a closed-loop feedback iteration layer. The multi-dimensional data acquisition layer interfaces with external management systems and equipment control systems, collecting equipment data, task data, and scenario environment data in real time, with all data uniformly aggregated to the scheduling system backend. The data preprocessing and risk prediction layer cleans, denoises, and standardizes the collected heterogeneous raw data, building time-series prediction models and congestion risk identification models. It combines historical and real-time operational data to predict task flow, channel congestion probability, and equipment malfunction probability in various areas over a future period, generating risk warning signals. The global intelligent scheduling decision layer receives risk prediction results and newly issued tasks, establishes an operational capability tag system for each warehouse device, matches equipment and allocates tasks based on material attributes, timeliness requirements, and operational constraints, dynamically plans operational paths based on real-time equipment location, and generates scheduling instructions and path instructions. The equipment execution layer sends scheduling and path instructions to the equipment control system, driving the warehousing equipment to perform warehousing operations. The closed-loop feedback iteration layer continuously collects feedback data such as equipment execution status, task completion progress, and real-time changes in the scenario, and iteratively optimizes the time series prediction model, operations research optimization algorithm parameters, and capability label judgment thresholds.

[0013] By adopting a five-layer architecture design, the system realizes a complete scheduling process from data collection, risk prediction, intelligent decision-making, equipment execution to closed-loop feedback, ensuring the efficiency, stability and adaptability of warehousing operations.

[0014] Beneficial Effects: This invention provides a warehouse global adaptive intelligent task scheduling system and method based on capability tags, which has the following beneficial effects: By introducing a four-dimensional capability tagging system and a two-stage device matching mechanism, this invention achieves a comprehensive quantitative expression of device capabilities and precise matching with task requirements, avoiding efficiency losses caused by mismatches between device capabilities and task needs. Compared to existing technologies that allocate tasks based solely on simple rules, this invention comprehensively quantifies device capabilities through a capability tagging system and performs secondary optimization by combining global load, spatial distance, and path status, significantly improving the accuracy of task allocation and overall efficiency.

[0015] By introducing a time-series prediction model and a congestion risk identification model, and combining historical and real-time operational data to extrapolate task traffic, channel congestion probability, and equipment malfunction probability in various regions over a future period, this invention achieves proactive risk identification and early warning. This provides a basis for pre-emptive intervention in scheduling decisions, preventing risks from escalating into actual failures or congestion. Compared to existing technologies that cannot accurately predict path conflicts leading to frequent AGV collisions and congestion, this invention, through time-series prediction and congestion risk identification, provides early warnings and schedule adjustments before risks occur, significantly reducing the number of path conflicts and improving system stability.

[0016] By introducing the A* algorithm and an improved ant colony algorithm, and combining the current location of the equipment, the urgency of the task, the width of the work area passage, and the distribution of obstacles for path planning and task allocation, this invention achieves optimal path for single equipment and global collaboration among multiple equipment, avoiding path conflicts and congestion, and improving the efficiency of multi-equipment collaboration. Compared with existing technologies that are limited to single-vehicle path planning and obstacle avoidance and lack a global collaborative scheduling mechanism, this invention ensures optimal path for single equipment through the A* algorithm and achieves global collaboration among multiple equipment through the improved ant colony algorithm. The pheromone update rule is based on the total task completion time and the number of path conflicts, ensuring the high efficiency of multi-equipment collaborative operation.

[0017] By introducing a closed-loop feedback iteration layer, feedback data such as device execution status, task completion progress, and real-time changes in the scenario are continuously collected. This data is then used to iteratively optimize the time-series prediction model, operations research algorithm parameters, and capability label judgment thresholds, achieving adaptive evolution of the scheduling system and improving its scheduling accuracy and responsiveness in dynamic environments. Compared to existing technologies that cannot adjust scheduling strategies based on dynamic task changes and real-time system status, this invention, through a closed-loop feedback iteration mechanism, enables the scheduling system to continuously learn and optimize, adapting to dynamically changing operating environments and significantly improving the system's adaptability and intelligence. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the overall architecture of the warehouse global adaptive intelligent task scheduling system based on capability tags according to the present invention.

[0019] Figure 2 This is a flowchart illustrating the warehouse global adaptive intelligent task scheduling method based on capability tags according to the present invention. Detailed Implementation

[0020] This invention provides a warehouse-wide adaptive intelligent task scheduling system and method based on capability tags. By constructing a five-layer architecture, it achieves refined management and dynamic optimization of multi-device collaborative scheduling. The invention will be further described in detail below with reference to embodiments. Example 1

[0021] like Figure 1 As shown in the figure, this embodiment provides a warehouse global adaptive intelligent task scheduling system based on capability tags. The system includes a multi-dimensional data acquisition layer, a data preprocessing and risk prediction layer, a global intelligent scheduling decision layer, an equipment execution layer, and a closed-loop feedback iteration layer.

[0022] The multi-dimensional data acquisition layer connects to the WMS warehouse management system, RCS equipment control system, and PLC controller through standard communication protocols.

[0023] Specifically, the multi-dimensional data acquisition layer uses OPC UA or Modbus TCP communication protocols to collect real-time data from the entire domain, including device data, task data, and scene environment data, 24 hours a day.

[0024] The equipment data includes the online status, real-time location, operating speed, remaining battery power, fault warning information, and current number of tasks for equipment such as AGVs, conveyors, and stacker cranes. The real-time location of the AGVs is obtained through GPS positioning coordinates.

[0025] Task data includes order type, task priority, time requirement, material weight, material size, and start and end areas of the operation. Order types include inbound orders, outbound orders, and in-warehouse handling orders.

[0026] The scenario environment data includes task density, channel occupancy rate, and historical task traffic patterns for each work area. All data is uniformly aggregated in JSON format to the scheduling system's backend database.

[0027] The preprocessing and risk assessment layer cleans, denoises, and standardizes the collected heterogeneous raw data.

[0028] Specifically, data cleaning includes removing outliers and missing values, noise reduction uses moving average filtering, and format standardization includes unifying timestamp formats and coordinate system transformation.

[0029] The data preprocessing and risk assessment layer establishes a time-series prediction model and a congestion risk identification model. The time-series prediction model uses an LSTM neural network, taking historical 24-hour task traffic data as input, to predict the task traffic of each work area in the next hour.

[0030] The congestion risk identification model employs a convolutional neural network, taking real-time channel occupancy, device location distribution, and task queue length as input, and outputting the congestion probability for each channel. A risk warning signal is generated when the predicted task traffic exceeds the regional processing capacity threshold or the congestion probability exceeds 80%.

[0031] The data preprocessing and risk prediction layer uses machine learning algorithms to identify abnormal equipment operating status, and combines deep learning algorithms to identify risks such as task queuing and channel congestion. It autonomously predicts potential risks such as congestion, task backlog, and equipment failure, and pushes risk warning signals to the global intelligent scheduling decision layer.

[0032] The global intelligent scheduling decision layer is the core module of this invention, which introduces operations research optimization algorithms and dynamic scheduling algorithms to build a global intelligent scheduling model.

[0033] After receiving risk prediction results and newly issued tasks, the global intelligent scheduling decision layer automatically performs logical optimizations such as task splitting, merging similar tasks, and prioritizing urgent orders. For example, it splits large batches of orders into multiple sub-tasks and merges multiple tasks to the same destination into a single batch.

[0034] The global intelligent scheduling decision-making layer establishes a complete operational capability labeling system for each automated device.

[0035] The operational capability labeling system includes static basic labels, dynamic operational labels, equipment health labels, and operational efficiency labels. Static basic labels include equipment type, rated load, compatible material types, hardware parameters, and the coordinate range of the operable area.

[0036] The dynamic operation tag includes the current load weight, remaining available workload, real-time energy consumption, and operating status. The equipment health tag includes the fault warning level, historical failure rate, and equipment maintenance status.

[0037] The job efficiency tag includes the average processing time for similar tasks and the success rate of path conflict avoidance. The tag data is updated in real time and stored in the scheduling system's backend database.

[0038] The global intelligent scheduling decision layer adopts a two-stage equipment matching mechanism to complete equipment selection and task allocation.

[0039] The first stage is the initial screening of capability tags. After receiving a new task, the scheduling model queries the capability tag database based on the material weight, material size, and start and end areas of the operation, and selects a set of candidate equipment that meet the requirements of rated load, compatible material type, and operable area.

[0040] The second stage is global load optimization and matching. From the candidate device set, the global load index, spatial distance index, and path status index are calculated for each device. The global load index is the current number of tasks being carried divided by the remaining number of tasks that can be carried. The spatial distance index is the straight-line distance between the device's current location and the task's starting point.

[0041] The path status index represents the probability of a conflict between the device's current path and the task path. The device with the lowest weighted sum of the three indices is selected as the final matching device.

[0042] The global intelligent scheduling decision layer determines task priorities through multi-dimensional calculation methods.

[0043] Specifically, the time urgency score is equal to the task timeliness requirement timestamp minus the current timestamp, then divided by the system's average task processing time, normalized to 0 to 100. The material value score is equal to the material weight multiplied by the value per unit weight plus the material volume multiplied by the value per unit volume, then divided by the system's average material value, normalized to 0 to 100.

[0044] The storage urgency score is calculated by multiplying the current warehouse space utilization rate by the material storage urgency coefficient, then dividing by the system's average storage urgency, and normalizing to a score between 0 and 100. The equipment adaptability score is calculated by dividing the number of devices in the candidate equipment set that meet the capability label requirements by the total number of devices in the system, and normalizing to a score between 0 and 100.

[0045] Task priority is equal to 0.4 times the time urgency score plus 0.3 times the material value score plus 0.2 times the storage urgency score plus 0.1 times the equipment adaptability score.

[0046] The global intelligent scheduling decision-making layer dynamically plans the operation path based on the real-time location of the equipment.

[0047] For single-device, single-task scenarios, the A* algorithm is used for path planning. The warehouse map is rasterized, with the device's current location as the starting point and the task's endpoint as the target point. The heuristic function is the Manhattan distance, and the cost function is the path length plus a channel occupancy penalty term to search for the optimal path.

[0048] For single-device, multi-tasking scenarios, an improved ant colony algorithm is used for task allocation. The ant colony is initialized, with each ant representing a specific task execution order. The pheromone update rule is that the pheromone increment equals 1 divided by the total task completion time plus the number of path conflicts, and the pheromone evaporation coefficient is 0.1.

[0049] After 100 iterations, the path with the highest pheromone concentration is selected as the optimal task execution order.

[0050] The global intelligent scheduling decision layer generates scheduling instructions and path instructions. Scheduling instructions include the device ID, task ID, and task priority. Path instructions include the path coordinate sequence and estimated arrival time.

[0051] The equipment execution layer uniformly distributes scheduling instructions and path instructions to the RCS and PLC controllers.

[0052] Specifically, the equipment execution layer sends scheduling and path instructions to the RCS and PLC controllers through standard communication interfaces, driving automated equipment such as AGVs, conveyors, and stacker cranes on site to perform warehousing operations according to the planned scheme.

[0053] The closed-loop feedback iteration layer continuously collects feedback data such as device execution status, task completion progress, and real-time changes in the scene.

[0054] Specifically, the device execution layer collects data on device execution status, task completion progress, and real-time scene changes every 10 seconds during operation and transmits this data back to the scheduling system. Device execution status includes current location, speed, battery level, and fault status. Task completion progress includes the number of completed tasks, the number of tasks currently executing, and the number of waiting tasks.

[0055] The scene changes in real time, including channel occupancy and regional task density. The closed-loop feedback iteration layer triggers model parameter optimization once every 24 hours.

[0056] The weight parameters of the LSTM time-series prediction model are adjusted using gradient descent, the convolution kernel parameters of the congestion risk identification model are adjusted using backpropagation, and the capability label determination thresholds are adjusted using a genetic algorithm. The capability label determination thresholds include the rated load threshold and the fault warning level threshold. The optimization objective is to minimize the mean squared error between the model's prediction accuracy and the actual operating data.

[0057] The system works as follows: After the system starts up, the multi-dimensional data acquisition layer continuously collects operational data from the entire domain and aggregates it into the background database.

[0058] The data preprocessing and risk prediction layer cleans, reduces noise, and standardizes the format of the data, and generates risk warning signals through time series prediction models and congestion risk identification models.

[0059] The global intelligent scheduling decision layer receives newly issued tasks and risk warning signals, completes equipment selection through initial screening with capability tags and global load optimization matching, determines the task execution order through multi-dimensional calculation of task priority, plans equipment paths through A* algorithm and improved ant colony algorithm, and generates scheduling instructions and path instructions.

[0060] The equipment execution layer sends instructions to the RCS and PLC controllers to drive AGVs, conveyors, and stacker cranes to perform warehousing operations.

[0061] The closed-loop feedback iteration layer continuously collects the device execution status, task completion progress, and real-time changes in the scene, and sends them back to the scheduling system, triggering model parameter optimization once every 24 hours.

[0062] By introducing a four-dimensional capability labeling system and a two-stage equipment matching mechanism, the system achieves a comprehensive quantitative expression of equipment capabilities and precise matching with task requirements, avoiding efficiency losses caused by mismatches between equipment capabilities and task needs. Furthermore, by introducing time-series prediction models and congestion risk identification models, the system enables proactive risk identification and early warning, preventing risks from evolving into actual failures or congestion.

[0063] By introducing the A* algorithm and an improved ant colony algorithm, single-device path optimization and multi-device global collaboration are achieved, avoiding path conflicts and congestion. The introduction of a closed-loop feedback iterative layer enables adaptive evolution of the scheduling system, improving scheduling accuracy and responsiveness in dynamic environments. Example 2

[0064] like Figure 2 As shown in the figure, this embodiment provides a warehouse global adaptive intelligent task scheduling method based on capability tags, which includes the following steps.

[0065] Step S1: Collect device data, task data, and scene environment data in real time through a multi-dimensional data acquisition layer, and all data are uniformly aggregated to the scheduling system backend.

[0066] Specifically, equipment data includes the online status, real-time location, operating speed, remaining power, fault warning information, and current number of tasks for equipment such as AGVs, conveyors, and stacker cranes. Task data includes order type, task priority, timeliness requirements, material weight, material size, and start and end areas of the operation. Scene environment data includes task density, channel occupancy rate, and historical operation flow patterns for each work area.

[0067] Step S2 involves cleaning, denoising, and standardizing the collected heterogeneous raw data through data preprocessing and risk prediction layers. A time-series prediction model and a congestion risk identification model are then built. By combining historical operation data and real-time data, the task flow, channel congestion probability, and equipment malfunction probability in each area over a future period are predicted, and a risk warning signal is generated.

[0068] In practice, the collected raw data is cleaned to remove outliers and missing values, and noise is reduced using a moving average filter. The data is then standardized in format, and timestamp format and coordinate system transformation are unified. An LSTM time-series prediction model is built, taking historical 24-hour task traffic data as input, to predict the task traffic for each work area in the next hour.

[0069] A convolutional neural network congestion risk identification model is built. Inputs include real-time channel occupancy rate, device location distribution, and task queue length. The model outputs the congestion probability for each channel. When the predicted task traffic exceeds the regional processing capacity threshold or the congestion probability exceeds 80%, a risk warning signal is generated and pushed to the scheduling decision layer.

[0070] Step S3: The global intelligent scheduling decision layer receives risk prediction results and newly issued tasks, establishes an operational capability tag system for each warehousing equipment, completes equipment matching and task allocation based on the material attributes, time requirements, and operational constraints of the tasks to be executed, dynamically plans the operation path based on the real-time location of the equipment, and generates scheduling instructions and path instructions.

[0071] Specifically, upon receiving risk assessment results and newly issued tasks, the system automatically optimizes logic such as task splitting, merging similar tasks, and prioritizing urgent orders. A complete operational capability tagging system is established for each automated device, including static basic tags, dynamic operational tags, device health tags, and operational efficiency tags.

[0072] Based on the material attributes, time requirements, and operational constraints of the tasks to be executed, initial equipment screening is performed using capability tags. Then, a secondary optimization and matching process is conducted by considering global equipment load, spatial distance, and path status to achieve refined task allocation. Task priorities are calculated by comprehensively considering time urgency, material value, storage urgency, and equipment adaptability.

[0073] Based on the device's current location, task urgency, work area corridor width, and obstacle distribution, the optimal path is planned using the A* algorithm. When the device receives multiple tasks simultaneously, an improved ant colony algorithm is used for task allocation, generating final scheduling and path instructions.

[0074] Step S4: The scheduling instructions and path instructions are sent to the equipment control system through the equipment execution layer to drive the warehousing equipment to perform warehousing operations.

[0075] Specifically, the final scheduling instructions and path instructions generated by the scheduling decision layer are uniformly sent to the RCS and PLC controllers to drive the automated equipment such as AGVs, conveyors, and stacker cranes on site to perform warehousing operations according to the planned scheme.

[0076] Step S5 involves continuously collecting feedback data such as device execution status, task completion progress, and real-time changes in the scene through a closed-loop feedback iteration layer, and iteratively optimizing the time series prediction model, operations research optimization algorithm parameters, and capability label judgment threshold.

[0077] Specifically, during the operation, feedback data such as equipment execution status, task completion progress, and real-time scene changes are collected every 10 seconds and transmitted back to the scheduling system. Model parameter optimization is triggered every 24 hours, adjusting the weight parameters of the LSTM time series prediction model through gradient descent algorithm and adjusting the congestion risk identification model's congestion kernel parameters through backpropagation algorithm.

[0078] By adjusting the threshold for ability label determination using a genetic algorithm, the mean square error between the model's prediction accuracy and the actual operating data is minimized, forming a closed-loop operation mechanism of data acquisition, intelligent decision-making, equipment execution, and feedback optimization. Example 3

[0079] The difference between this embodiment and Embodiment 1 is that the work efficiency label in the four-dimensional capability labeling system is replaced with the historical work record label.

[0080] The historical job record tags include the total number of tasks completed by the equipment in history, the average completion time of the tasks in history, the number of path conflicts in history, and the number of failures in history. The historical job record tags are obtained by querying the equipment's historical job database.

[0081] The database records all job records of the equipment since it was put into use, including task ID, task type, task start time, task end time, task completion time, number of path conflicts, and number of failures. During the initial screening of equipment, the scheduling model calculates the historical average task completion time and the historical number of path conflicts of the equipment by querying historical job record tags.

[0082] Historical operation record tags are suitable for scenarios where equipment has been in use for a long time and sufficient historical data has been accumulated. Operation efficiency tags are suitable for scenarios where equipment has been in use for a short time and real-time evaluation of equipment efficiency is required. Example 4

[0083] This embodiment describes the system's adaptive handling capability in response to emergencies. When equipment malfunctions, order priorities change, or there is a sudden regional congestion, the system updates equipment capability tags and scenario data in real time, dynamically readjusting task allocation schemes and equipment operation paths.

[0084] Specifically, when an AGV malfunctions, the closed-loop feedback iteration layer immediately collects fault information, updates the AGV's equipment health tag, and sets the fault warning level to the highest level. The global intelligent scheduling decision layer receives the updated capability tag, removes the AGV from the candidate equipment set, and reassigns the task being performed by the AGV to other available equipment.

[0085] When an order priority changes, the data preprocessing and risk prediction layer receives the priority change signal and updates the task priorities in the task database. The global intelligent scheduling decision layer recalculates the task priorities, adjusts the task execution order, moves high-priority tasks to the front of the task queue, and generates new scheduling instructions and path instructions.

[0086] When a channel suddenly becomes congested, the congestion risk identification model detects that the congestion probability of that channel exceeds 80% and generates a risk warning signal. After receiving the warning signal, the global intelligent scheduling decision layer replans the device paths through that channel, sets the channel occupancy penalty term of the corresponding grid of the congested channel to the maximum value, takes the current position of the device as the starting point, the original task endpoint as the target point, uses the Manhattan distance as the heuristic function, and the sum of the path length and the channel occupancy penalty term as the cost function, and uses the A* algorithm to search for an alternative optimal path to bypass the congested channel. The path instructions are then updated and sent to the device execution layer to ensure that the device avoids the congested area.

[0087] Through the aforementioned adaptive processing mechanism, the system can quickly resume normal operation in the face of emergencies, reduce downtime caused by equipment failure or other reasons, and ensure the continuity and stability of warehousing operations.

[0088] It is understood that the time series prediction model in the above embodiments is not limited to the LSTM model, and other time series prediction models such as the GRU model, Transformer model, and ARIMA model can also be used. The congestion risk identification model is not limited to convolutional neural networks, and other machine learning models such as recurrent neural networks, support vector machines, and random forests can also be used.

[0089] Clearly, path planning algorithms are not limited to the A* algorithm; other algorithms such as Dijkstra's algorithm, Floyd's algorithm, and dynamic programming can also be used. Similarly, multi-device task allocation algorithms are not limited to the improved ant colony algorithm; other heuristic optimization algorithms such as genetic algorithms, particle swarm optimization, and simulated annealing can also be employed.

[0090] It is understandable that the weighting coefficients for each dimension in task priority calculation are not limited to 0.4, 0.3, 0.2, and 0.1, and can be adjusted according to the actual application scenario. For example, in scenarios with strict time requirements, the weighting coefficient for time urgency can be adjusted to 0.6.

[0091] Clearly, the optimization period for the model parameters in the closed-loop feedback iterative layer is not limited to 24 hours; it can also be dynamically adjusted based on the system's operating status. For example, the optimization period can be extended to 48 hours when the system is running stably, and shortened to 12 hours when the system is experiencing significant fluctuations.

[0092] It is understandable that the equipment execution layer is not limited to driving the equipment through RCS and PLC controllers. It can also directly receive scheduling instructions through the equipment's built-in control system, as long as the issuance of scheduling instructions and precise control of the equipment can be achieved.

Claims

1. A warehouse global adaptive intelligent task scheduling method based on capability tags, executed by a scheduling system, characterized in that, Includes the following steps: The system collects device data, task data, and scene environment data in real time through a multi-dimensional data acquisition layer, and all data is uniformly aggregated to the scheduling system backend. The collected data is preprocessed through a data preprocessing and risk prediction layer to build a prediction model and a risk identification model. By combining historical data and real-time data, the system's operating status is predicted, and risk warning signals are generated. The global intelligent scheduling decision layer receives risk prediction results and newly issued tasks, establishes an operational capability tag system for each storage equipment, conducts initial screening of equipment based on the operational capability tag system to obtain candidate equipment, and then performs optimization matching in combination with global status information to complete task allocation. Combined with the real-time location of the equipment, the operation path is dynamically planned to generate scheduling instructions and path instructions. The equipment execution layer sends scheduling and routing instructions to the equipment control system to drive the warehousing equipment to perform warehousing operations. The system continuously collects feedback data through a closed-loop feedback iteration layer to iteratively optimize the system model parameters and scheduling rules.

2. The warehouse global adaptive intelligent task scheduling method based on capability tags according to claim 1, characterized in that, The process of matching equipment and allocating tasks based on the material attributes, time requirements, and operational constraints of the tasks to be performed includes: firstly, initial screening of equipment based on capability tags, and then secondary optimization matching based on global equipment load, spatial distance, and path status.

3. The warehouse global adaptive intelligent task scheduling method based on capability tags according to claim 1, characterized in that, Before generating scheduling instructions and path instructions, the process also includes: comprehensively considering the urgency of time, the value of materials, the urgency of storage, and the priority of equipment adaptability calculation tasks.

4. The warehouse global adaptive intelligent task scheduling method based on capability tags according to claim 1, characterized in that, After receiving risk prediction results and newly issued tasks through the global intelligent scheduling decision layer, the system automatically performs task splitting, merging of similar tasks, and priority order placement based on task attributes and system status.

5. The warehouse global adaptive intelligent task scheduling method based on capability tags according to claim 1, characterized in that, The dynamic planning of work paths based on the real-time location of the equipment specifically includes: rasterizing the warehouse map by combining the current location of the equipment, the urgency of the task, the width of the work area passage, and the distribution of obstacles; using the current location of the equipment as the starting point and the end point of the task as the target point; using Manhattan distance as the heuristic function and the sum of path length and passage occupancy penalty term as the cost function; and using an improved ant colony algorithm for task allocation when the equipment receives multiple tasks at the same time, wherein the pheromone update rule is based on the total completion time of the task and the number of path conflicts.

6. The warehouse global adaptive intelligent task scheduling method based on capability tags according to claim 1, characterized in that, The equipment data includes the online status, real-time location, operating speed, remaining power, fault warning information, and current number of tasks carried by the warehousing equipment; the task data includes order type, task priority, timeliness requirements, material weight, material size, and start and end areas of the operation; the scene environment data includes task density, channel occupancy rate, and historical operation flow patterns for each operation area.

7. The warehouse global adaptive intelligent task scheduling method based on capability tags according to claim 1, characterized in that, The construction of the prediction model and risk identification model specifically includes: building a time-series prediction model to predict the task traffic in each region over a future period of time; building a congestion risk identification model to identify the probability of channel congestion; and generating a risk warning signal when the predicted task traffic exceeds the regional processing capacity threshold or the congestion probability exceeds a set threshold.

8. The warehouse global adaptive intelligent task scheduling method based on capability tags according to claim 1, characterized in that, The operational capability tagging system includes static basic tags, dynamic operational tags, equipment health tags, and work efficiency tags. The static basic tags include equipment type, rated load, compatible materials, hardware parameters, and workable area. The dynamic operational tags include current load, remaining available task capacity, real-time energy consumption, and operational status. The equipment health tags include fault warning level, historical failure rate, and equipment maintenance status. The work efficiency tags include average completion time for similar tasks and historical path conflict counts.

9. The warehouse global adaptive intelligent task scheduling method based on capability tags according to claim 1, characterized in that, The iterative optimization of system model parameters and scheduling rules specifically includes: adjusting the weight parameters of the time series prediction model through gradient descent algorithm, adjusting the parameters of the risk identification model through backpropagation algorithm, and adjusting the capability label judgment threshold through genetic algorithm, so as to minimize the mean square error between the model prediction accuracy and the actual operating data.

10. A warehouse global adaptive intelligent task scheduling system based on capability tags, characterized in that, include: The multi-dimensional data acquisition layer is used to connect with external management systems and equipment control systems to collect equipment data, task data and scene environment data in real time. All data is uniformly aggregated to the scheduling system backend. The multi-dimensional data acquisition layer is connected to the data preprocessing and risk prediction layer through a data interface. The data preprocessing and risk prediction layer is used to preprocess the collected data, build prediction models and risk identification models, combine historical data and real-time data to predict the system operation status, and generate risk warning signals. The data preprocessing and risk prediction layer is connected to the global intelligent scheduling decision layer through data interfaces and control interfaces. The global intelligent scheduling decision layer is used to receive risk prediction results and newly issued tasks, and to establish an operational capability tag system for each warehouse equipment. The operational capability tag system includes static basic tags, dynamic operation tags, equipment health tags and operation efficiency tags. Based on the material attributes, time requirements and operation constraints of the task to be executed, the system completes equipment matching and task allocation, and dynamically plans the operation path in combination with the real-time location of the equipment, generating scheduling instructions and path instructions. The equipment execution layer is used to send scheduling instructions and path instructions to the equipment control system to drive the warehousing equipment to perform warehousing operations. The closed-loop feedback iteration layer is used to continuously collect feedback data such as device execution status, task completion progress, and real-time changes in the scene, and to iteratively optimize the time series prediction model, operations research optimization algorithm parameters, and capability label judgment thresholds.

Images