Warehouse system full-link processing method

Abstract

The present application relates to the technical field of warehouse intelligent scheduling, in particular, the present application relates to a warehouse system full-link processing method, the present application constructs a full-link directed task dependency graph by taking the task identifier and the expected recovery duration of the abnormal process node as the node, and taking the material flow direction and the resource occupation relationship as the directed edge, determines the main link by using the weighted shortest path algorithm and calculates the time sequence offset of each node, determines the tolerance offset threshold by combining the real-time utilization rate of the equipment and the time efficiency grade of the downstream task, marks the nodes exceeding the threshold, counts the number of nodes exceeding the threshold and the directed edge hop number of the most downstream node exceeding the threshold, and schedules hierarchically accordingly, maintains the original sequence if there is no node exceeding the threshold, only adjusts locally if there are few nodes exceeding the threshold and the hop number is compliant, otherwise triggers full-link rescheduling, realizes accurate quantification of abnormal influence and flexible adaptive scheduling, and improves the efficiency of warehouse operation and the stability of delivery performance.

Description

Technical Field

[0001] This invention relates to the field of intelligent warehouse scheduling technology, and more specifically, to a full-chain processing method for warehouse systems. Background Technology

[0002] Intelligent warehouse scheduling is an important technology, specifically applied to the task scheduling optimization stage after equipment failure at warehouse process nodes. It achieves adaptive control of failures through directed task dependency graphs and hierarchical scheduling. There are strong dependencies between warehouse operation tasks in terms of material flow and resource consumption. Equipment failures will propagate along the task chain, affecting the execution sequence of downstream tasks. Since the timing offset of the failure impact cannot be accurately quantified by combining equipment status, queue load and task timeliness, and the scheduling level cannot be distinguished according to the scope and depth of impact, task scheduling will either be over-adjusted, resulting in resource waste, or under-adjusted, causing delivery delays. As a result, it is difficult to ensure a stable balance between the efficiency of the entire warehouse operation chain and the delivery timeliness. To solve this technical problem, we provide a whole-chain processing method for warehouse systems. Summary of the Invention

[0003] The purpose of this invention is to provide a full-chain processing method for warehousing systems to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, one objective of this invention is to provide a full-chain processing method for a warehousing system, comprising: S1. Obtain the task identifier and estimated recovery time of the abnormal process node device; S2. Construct a directed task dependency graph for the entire warehousing chain, specifically including: taking the currently unexecuted task as a node, taking the material flow direction and resource occupation relationship between tasks as directed edges, using a weighted shortest path algorithm based on equipment status and task queue length to determine the longest dependency path between nodes as the backbone link, taking the abnormal process node as the starting point and traversing along the backbone link in the forward direction, summing the cumulative expected waiting time of all traversed edges between each affected node and the abnormal process node with the expected recovery time, and using the result as the offset of the original task execution time window of the affected node; S3. The real-time utilization rate of the associated device of the affected node is weighted and calculated with the timeliness level value of the adjacent downstream task. The result is normalized and used as the tolerance offset threshold of the node. The offset is compared with the tolerance offset threshold, and the node with the offset greater than the tolerance offset threshold is marked as the threshold node. S4. Count the total number of nodes exceeding the threshold, and record the number of directed edge hops between the downstream node exceeding the threshold and the abnormal process node. S5. When the total number of nodes exceeding the threshold is zero, maintain the current task sequence and task identifier. When the total number of nodes exceeding the threshold is not zero and the number of hops of the directed edges does not exceed the preset hop count threshold, only perform local timing adjustments on the nodes exceeding the threshold and their immediate downstream nodes. Otherwise, trigger full-link task rescheduling.

[0005] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention constructs a directed dependency graph of tasks across the entire warehousing chain, accurately quantifies the time-series offset caused by anomaly propagation, and precisely determines the scope and depth of anomaly impact. This effectively avoids resource waste caused by excessive scheduling adjustments or delivery delays caused by insufficient adjustments. By dynamically dividing scheduling levels based on real-time equipment status, queue load, and task timeliness levels, it achieves flexible adaptive scheduling in abnormal scenarios, significantly reducing the propagation impact of equipment anomalies on subsequent task execution, ensuring the stability of the entire chain's operation sequence, optimizing the execution process based on task dependencies and hierarchical scheduling rules, reducing invalid scheduling operations and resource idleness losses, and improving the utilization rate of warehousing equipment and the efficiency of operation connection at each stage. Attached Figure Description

[0006] Figure 1 This is a flowchart illustrating the overall workflow of the present invention; Figure 2 This is a comparison diagram of the timing offset of the backbone link nodes of this invention; Figure 3 This is a graph showing the cumulative effect of abnormal conduction in this invention. Detailed Implementation

[0007] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0008] Please see Figure 1 As shown, this embodiment provides a full-chain processing method for a warehousing system, including: S1. Obtain the task identifier and estimated recovery time of the abnormal process node device; S2. Construct a directed task dependency graph for the entire warehousing chain, specifically including: taking the currently unexecuted task as a node, taking the material flow direction and resource occupation relationship between tasks as directed edges, using a weighted shortest path algorithm based on equipment status and task queue length to determine the longest dependency path between nodes as the main link, starting from the abnormal process node and traversing along the main link in the forward direction, summing the cumulative expected waiting time and expected recovery time of all traversed edges between each affected node and the abnormal process node, and using the result as the offset of the original task execution time window of the affected node; S3. The real-time utilization rate of the associated equipment of the affected node is weighted and calculated with the timeliness level value of the adjacent downstream task. The result is normalized and used as the tolerance offset threshold of the node. The offset is compared with the tolerance offset threshold, and the node with the offset greater than the tolerance offset threshold is marked as the threshold node. S4. Count the total number of nodes that exceed the threshold, and record the number of directed edge hops between the downstream node that exceeds the threshold and the abnormal process node. S5. When the total number of nodes exceeding the threshold is zero, maintain the current task sequence and task identifier. When the total number of nodes exceeding the threshold is not zero and the number of hops of directed edges does not exceed the preset hop threshold, only perform local timing adjustments on the nodes exceeding the threshold and their immediate downstream nodes. Otherwise, trigger full-link task rescheduling.

[0009] Further explanation is needed. After completing the initial construction of the directed task dependency graph of the entire warehousing chain, in order to accurately locate the critical constraint path most affected by equipment status and task queues in the entire chain, it is necessary to further adopt a weighted shortest path algorithm based on equipment status and task queue length to determine the longest dependency path between nodes as the backbone link. The core of this algorithm is to quantify the equipment operating status and the real-time queue load of nodes together as path weights, and then select the longest dependency path with the largest weight. In specific implementation, firstly, two core quantitative parameters are extracted from the equipment status that has been collected in real time. Equipment status is a comprehensive indicator that reflects the real-time processing efficiency and long-term operational reliability of various physical equipment such as sorting machines, conveyors, and stacker cranes in the warehousing system. The two extracted core parameters are the real-time processing speed of the equipment and the historical average failure interval of the equipment. Among them, the real-time processing speed of the equipment refers to the number of standard tasks that the equipment can complete per unit time, which directly reflects the current real-time operating efficiency of the equipment. The larger the value, the higher the equipment processing efficiency and the lower the generation. The shorter the waiting delay, the better. The historical mean time between failures (MTBF) of the equipment refers to the average duration of continuous, fault-free operation of the equipment over a past period, extracted from the equipment maintenance log database. It is a core indicator for measuring the inherent reliability and long-term operational stability of the equipment. A higher value indicates more stable equipment operation and lower failure risk. After obtaining these two core parameters, they first need to be normalized. Normalization maps parameters with different dimensions and numerical ranges to a standard range of 0 to 1, eliminating the impact of differences in dimensions and numerical magnitudes between parameters on subsequent weight calculations, ensuring fairness and comparability when each parameter participates in the weight calculation. After normalization, since the equipment status weight coefficient needs to reflect the equipment's latency and congestion level (i.e., the slower the real-time processing speed and the shorter the historical mean time between failures, the longer the equipment's waiting time), it is necessary to take the reciprocals of the normalized real-time processing speed and the historical mean time between failures. The reciprocal of the normalized real-time processing speed is denoted as... The reciprocal of the normalized historical mean time between failures is denoted as... Then, a weighted summation is performed on the two reciprocals. Weighted summation involves assigning corresponding weight coefficients to each parameter based on their relative importance to the equipment status, and then summing the results. The weighted summation result is the equipment status weight coefficient. The specific calculation formula is as follows: ,in The final calculated equipment state weight coefficients, The weighting coefficient is the reciprocal of the device's real-time processing speed. The weighting coefficient corresponds to the reciprocal of the historical mean time between failures (MTBF) of the equipment. and The value is pre-calibrated based on the type of warehousing equipment and the actual operating scenario, and strictly meets the following requirements. This is to ensure the normalization and rationality of the weight calculation.

[0010] While calculating the device status weight coefficients, it is necessary to simultaneously obtain the total number of tasks waiting to be executed on each task node as the task queue length. The task queue length refers to the total number of unexecuted tasks waiting to be processed at the current node, directly reflecting the real-time load and congestion level of the node. The longer the queue length, the longer the node's waiting time and the higher the congestion level. Subsequently, the task queue length is normalized in the same way to eliminate the calculation bias caused by the difference in numerical magnitude, and the queue length weight is obtained. The normalization formula is: ,in The normalized queue length weight. This is the actual length of the task queue for the current node. This represents the node's historical minimum task queue length. The maximum historical task queue length for this node is used as the starting point. Finally, in the constructed directed task dependency graph with unexecuted tasks as nodes and material flow and resource usage relationships between tasks as directed edges, the path weight of the directed edge connecting any two adjacent task nodes is defined as the sum of the device state weight coefficient of the device associated with the starting node of the directed edge and the queue length weight of that node. The specific calculation formula is as follows: ,in This represents the path weight of a single directed edge. The weight is a dimensionless value between 0 and 2, comprehensively reflecting the combined impact of device status and node congestion on the task execution stage corresponding to that directed edge. This path weight quantifies the combined influence of device operating status and node real-time queue load on task execution latency. A larger path weight indicates a longer latency for that task segment and a greater constraint on the overall task chain. After assigning path weights to all directed edges in the entire chain, an improved Dijkstra algorithm is used to traverse all task node paths to find the path weight between the starting task node and the ending task node. The path with the largest sum of directed edge path weights is the longest dependency path. The longest dependency path represents the critical path in the entire warehouse task execution that is most affected by equipment status and queue load, has the highest delay risk, and directly restricts the overall task progress. Finally, the longest dependency path is directly defined as the backbone of the entire warehouse task. As the core constraint path of the overall task scheduling, the backbone provides the core path basis and data support for subsequent traversal of affected nodes from abnormal process nodes, calculation of task execution time offset of each affected node, determination of threshold nodes, and execution of local timing adjustments or full-link rescheduling.

[0011] After calculating the device status weight coefficients, in order to further incorporate the real-time load of nodes into the comprehensive quantitative system of path weights, it is necessary to synchronously collect and process the queuing task data of each task node. Specifically, the system will scan and obtain the total number of tasks currently in the queue and not yet started on each node in real time, and define this number as the task queue length. The task queue length is a core indicator that directly reflects the current processing congestion and task backlog of a node. Its value is positively correlated with the node's waiting delay and processing congestion. That is, the larger the task queue length, the more tasks the node is waiting to process, the higher the node load, and the longer the waiting delay for task execution. In order to eliminate the differences in the numerical magnitude and dimension of the task queue length of different nodes and ensure that the queue length indicators of all nodes can participate in the weight calculation under a unified standard, it is necessary to normalize the task queue length. Through normalization, the original task queue length of any node is standardized to a queue length weight in the range of 0 to 1. The closer the value is to 1, the higher the node congestion. The closer the value is to 0, the lower the node congestion.

[0012] After standardizing the queue length weight calculation, the system performs a quantitative calculation of path weights based on a pre-constructed directed task dependency graph. This graph is a directed acyclic graph structure built using all unexecuted tasks in the warehousing system as nodes and the material flow order and resource occupation dependency relationships between tasks as directed edges. Each node represents an independent unexecuted task, and the directed edges represent the execution order, material supply dependency, and equipment resource occupation relationships between tasks. The direction of the edges strictly follows the execution order logic and material flow direction, with no circular dependencies or reverse pointing, fully conforming to the actual execution flow of warehousing tasks. In this directed task dependency graph, the weight of each directed edge connecting two adjacent task nodes is defined as the weight of the starting node of that directed edge. The calculation method is the sum of the device status weight coefficient of the associated device and the queue length weight of the starting node. This method can comprehensively quantify the operational reliability of the device itself and the current real-time load of the node, and transform them into the path weight of the directed edge. Through the above calculation method, the system can quantify the two core influencing factors, device reliability and node real-time load, into the path weight of each directed edge in the directed task dependency graph. The value of the path weight directly corresponds to the degree of delay risk in the task execution stage. That is, the larger the path weight, the more significant the impact of device status and node congestion on the task execution stage, the higher the delay risk of task execution, and the greater the constraint on the overall warehouse task scheduling, ensuring that the overall scheduling strategy can accurately match the real-time changes in device status and node load.

[0013] Building upon this, to further dynamically correct task execution delays by incorporating real-time network flow data from the warehousing system, we multiply the path weight of each directed edge by an additional delay factor dynamically predicted through a machine learning model. This machine learning model is a multilayer perceptron model specifically trained for warehousing task delay prediction. The model uses the current time, the device state weight coefficients of the devices associated with the starting node of the directed edge, the device state weight coefficients of the devices associated with the ending node of the directed edge, and the global average task arrival rate of the warehousing system as input features. The current time refers to the timestamp of the task scheduling moment collected in real-time by the system, primarily used to distinguish differences in task traffic and equipment operating conditions across different time periods. The device state weight coefficients of the devices associated with the starting node of the directed edge are quantitative indicators representing the real-time operational reliability and processing efficiency of the warehousing equipment bound to the task at the starting point of the directed edge, directly reflecting the basic delay impact of the starting device on task execution. The device state weight coefficients of the devices associated with the ending node of the directed edge are also included. The standby state weight coefficient is a quantitative indicator representing the real-time receiving capability and processing efficiency of the warehouse operation equipment bound to the task corresponding to the endpoint of the directed edge. It reflects the delay constraint effect of the endpoint equipment on task flow. The global average task arrival rate of the warehouse system is the total number of newly added tasks to be scheduled in the warehouse system per unit time, used to objectively reflect the global task congestion and overall load level. This machine learning model extracts features from the above four types of input features layer by layer through the built-in fully connected layer, and completes feature mapping using a nonlinear activation function. Finally, it outputs a scaling factor in the range of 0 to 1. This scaling factor is the quantitative value of the additional congestion generated by the directed edge under the current network flow conditions, which can accurately reflect the additional superimposed impact of the real-time network flow on task execution delay. After multiplying this scaling factor with the directed edge path weight calculated above, the product result is the final expected waiting time corresponding to the directed edge segment. The formula for calculating the final expected waiting time can be explicitly expressed as follows: , in the formula Representing the final estimated wait time for a single directed edge, it is a precise latency quantification value that takes into account device status, node load, and global network flow. The additional latency factor, representing the output of the machine learning model, is a key coefficient for dynamically correcting the impact of real-time congestion. After identifying the abnormal process node in the warehousing system, starting from the abnormal process node, all affected task nodes are traversed sequentially along the directed edges of the pre-constructed directed task dependency graph. The final estimated waiting time of all directed edges traversed from the abnormal process node to each affected node is accumulated segment by segment. The sum obtained is the cumulative estimated waiting time from the abnormal process node to the affected node. This cumulative estimated waiting time can comprehensively and accurately reflect the total latency impact of the abnormal process on downstream task nodes, providing a direct and reliable quantitative basis for subsequent task scheduling optimization and time-series replanning of the warehousing system.

[0014] After completing the quantitative calculation of the equipment status weight coefficients and node task queue length indicators of the warehousing system, in order to further accurately characterize the dependency constraints and resource competition logic between tasks, it is necessary to construct a complete task dependency topology based on all unexecuted and executing tasks. The specific implementation method is as follows: Using tasks identified as either not yet started or issued but not yet completed in the warehouse management system as nodes, each node represents a single task unit in the warehouse system that is either inactive or in progress. These nodes form the basic units constituting the entire task dependency topology. The material handling list and resource scheduling table determine the order of material delivery between tasks. The material handling list records the sequence of material flow and delivery from upstream task nodes to downstream task nodes in the warehouse system, clarifying the sequential constraints between upstream initiating nodes and downstream receiving nodes. The resource scheduling table is the core scheduling table that records the allocation, occupancy, and release sequence of various equipment resources in the warehouse system. It accurately clarifies the mutual exclusion relationships of shared equipment resources between tasks. The mutual exclusion of shared equipment resources... The relationship refers to the exclusive constraint that when multiple task nodes need to compete for the same equipment resource, only one task is allowed to occupy the equipment at the same time, and the other tasks must wait for the release. This material delivery order and the mutual exclusion relationship of shared equipment resources are used as the basis for constructing directed edges. Directed edges refer to directional connecting edges that represent the dependency constraints and resource competition relationships between tasks. The direction strictly follows the order of material delivery and the order of shared equipment occupation. In this way, nodes with material delivery associations and shared equipment resource competition associations are directionally connected, and finally a directed acyclic graph structure is formed. A directed acyclic graph refers to a topology structure that does not have closed loops and only has unidirectional dependency constraints, which can effectively avoid scheduling deadlock problems caused by task dependency cycles.

[0015] After constructing the directed acyclic graph (DAG), weights need to be assigned to the directed edges connecting all nodes in the graph. The weight assignment must comprehensively consider factors such as material delivery delays between tasks, resource contention delays for shared equipment, equipment state weight coefficients, and node queue lengths. The specific weight calculation formula is as follows: ,in Refers to a single directed edge The final weighting result is a core quantitative indicator characterizing the strength of inter-task dependency constraints and resource competition delay. The material delivery delay weighting coefficient is a weighting parameter that characterizes the degree to which material delivery timing constraints affect task delays. Its value ranges from 0 to 1. Refers to a single directed edge The corresponding quantitative value for material delivery delay is calculated by the difference between the standard delivery time of the material recorded in the material handling list and the actual estimated delivery time. The shared device resource contention delay weighting coefficient is a weighting parameter that characterizes the degree to which the mutual exclusion of shared devices affects task latency. Its value also ranges from 0 to 1, and it satisfies the following conditions: To ensure the normalization and rationality of weight assignment, Refers to a single directed edge The corresponding shared device resource contention delay quantification value is calculated by the difference between the standard device occupancy time recorded in the resource scheduling table and the actual expected occupancy time.

[0016] After assigning weights to all directed edges, an improved Dijkstra's algorithm is used to find the path from the starting node to the path with the maximum sum of weights of all edges. This improved Dijkstra's algorithm optimizes the core logic of finding the shortest path into a dedicated algorithm for finding the longest cumulative weight path, based on the traditional Dijkstra's algorithm. The traditional Dijkstra's algorithm is used to find the shortest path with the minimum sum of weights, while this improved algorithm, by inversely optimizing the weight accumulation logic, can accurately find the longest path with the maximum sum of weights. The starting node refers to the initial task node in the directed acyclic graph that has no upstream dependent tasks and only downstream output tasks. The path with the maximum sum of weights of all edges is denoted as the maximum cumulative weight path, and the formula for calculating the cumulative weight of the maximum cumulative weight path is as follows: ,in Refers to the cumulative weight of the entire path. The path refers to the first A directed edge.

[0017] After obtaining the path with the maximum cumulative weight, the longest dependency path between nodes is determined based on this path. The longest dependency path refers to the path in the directed acyclic graph from the starting task node to the ending task node that has the most task dependency constraints, the most intense resource contention, and the longest overall latency. Its overall latency is calculated using the following formula: ,in Refers to the overall estimated latency of the longest dependency path. Refers to the standard execution time of the starting task node. The path refers to the first The task delay time corresponding to each directed edge is determined, and the longest dependency path is defined as the backbone link. The backbone link represents the key constraint channel between the starting task and the ending task that is most affected by the device status and queue load. Here, the key constraint channel refers to the constraint path in the entire task dependency topology where the fluctuation of device status and the change of node queue load have the most significant impact on the task execution delay. It is the core constraint link that must be given priority in the process of task scheduling optimization and time sequence replanning of the warehouse system. Its overall delay time directly determines the upper limit of the total task execution time of the entire warehouse system, and is also the core basis for subsequent task scheduling time sequence adjustment and resource reallocation.

[0018] After accurately locating the main backbone of the warehousing task and standardizing the estimated waiting time for each directed edge, when an abnormal process node appears in the warehousing system—that is, a node where equipment failure, task congestion, or scheduling anomalies prevent the task from executing as planned—the system will automatically initiate a standardized calculation process for the task timing offset starting from that abnormal process node. First, the system will sequentially visit each downstream node on the main backbone, following the previously determined direction of the main backbone, from the abnormal process node towards the direction the task has not yet been executed. Here, a downstream node refers to a node located at the abnormal process node in the main backbone topology. The material delivery direction is downstream, and task nodes that need to wait for upstream nodes to complete their tasks or for equipment to recover before execution can proceed. During the sequential access to each downstream node, the system continuously accumulates the final estimated waiting time for each directed edge traversed from the abnormal process node to the currently accessed node. Here, a directed edge refers to a directional connection edge in the main link that connects two adjacent task nodes, representing the task dependency relationship and material delivery sequence. The standardized final estimated waiting time for each directed edge is pre-calculated using equipment state weight, queue load weight, and task dependency strength. The formula for calculating this final estimated waiting time is as follows: ,in This represents the final estimated waiting time for a single directed edge. This represents the equipment status weighting coefficient, used to characterize the degree of impact of equipment failure or anomaly on waiting time. This indicates the base waiting time caused by equipment malfunction. This represents the node queue load weighting coefficient, used to characterize the impact of node task congestion on waiting time. This indicates the waiting time caused by congestion in the node queue. This represents the task dependency strength weighting coefficient, used to characterize the degree of influence of upstream and downstream task dependencies on waiting time. This represents the waiting time caused by task dependencies. By cumulatively adding the final estimated waiting time of each traversed directed edge, the system obtains the cumulative estimated waiting time value from the abnormal process node to the current node. The formula for calculating this cumulative estimated waiting time value is as follows: ,in This represents the cumulative estimated waiting time. This represents all nodes traversed from the abnormal process node to the current node. The final estimated waiting time for each directed edge is accumulated segment by segment. This indicates the th node traversed from the abnormal process node to the current node. The final estimated waiting time for a directed edge.

[0019] After obtaining the cumulative estimated waiting time, the system adds this cumulative estimated waiting time to the estimated recovery time of the abnormal process node obtained in advance. The estimated recovery time here refers to the standardized time required for the abnormal process node to fully recover from an abnormal state and be able to re-execute the task. This standardized time is obtained through machine learning training using historical equipment fault recovery data and task scheduling history data. The sum is directly used as the amount of time to postpone the original planned task execution time window of the current node, which is the timing offset of the current node. The formula for calculating this offset is... ,in This represents the time offset of the current node, which is the specific amount of time that the originally planned task execution time window needs to be postponed. This represents the cumulative estimated waiting time from the abnormal process node to the current node. The system indicates the estimated recovery time of the abnormal process node. The system will write this offset into the task scheduling time sequence table of the current node in real time, and standardize and postpone the original planned task execution time window of the current node. The original planned task execution time window refers to the standardized task execution time period planned in advance for each task node when the warehouse system does not have an abnormal state. It includes the task start execution time and the task end execution time. Through this node-by-node access, segment-by-segment accumulation and standardized calculation method, the system can accurately calculate the time sequence offset required to postpone all downstream nodes on the backbone due to the abnormal process node. This ensures that the adjustment of the entire task scheduling time sequence is accurate and standardized, and also ensures that those skilled in the art can completely reproduce the entire calculation process according to this specific implementation method.

[0020] After completing the standardized calculation of the timing offset of the affected nodes, in order to further accurately determine the upper limit of the tolerable timing offset of the nodes and provide a quantitative basis for subsequent task scheduling timing adjustments, it is necessary to calculate the tolerable offset threshold of the nodes according to the preset full-process quantitative rules. First, the ratio of the actual working time of the storage equipment associated with the affected nodes to the total time of the statistical period in the previous statistical period is obtained from the equipment monitoring system in real time. The equipment monitoring system is a dedicated monitoring system used to collect full-dimensional operating data such as the operating status, working time, and fault information of the storage equipment in real time. The statistical period refers to the pre-set standardized time used to statistically analyze the operating status of the equipment. The actual working time refers to the effective operating time of warehousing equipment within the statistical period during which it truly performs warehousing operations. This ratio is used as the real-time utilization rate of the equipment. The real-time utilization rate is a core quantitative indicator used to characterize the actual operating efficiency of warehousing equipment within the statistical period; a higher value indicates higher equipment operating efficiency. Simultaneously, the timeliness level codes of the downstream tasks immediately adjacent to the affected node are obtained from the task management system. The task management system is a dedicated management system used to coordinate the entire process of warehousing task generation, allocation, execution, and timeliness level classification. Downstream tasks immediately adjacent to the affected node refer to tasks located downstream of the affected node in the task execution topology and having a direct task dependency relationship with the affected node. The next level of tasks uses a timeliness level code, which is a standardized code pre-defined based on the urgency of the promised delivery time of the task order. The promised delivery time refers to the time the warehousing system promises to the task commissioner to complete the task. The higher the urgency, the higher the timeliness level. The system pre-maps different timeliness levels to different level values. The level value is a standardized value used to quantify the urgency of the task timeliness; the higher the timeliness level, the larger the corresponding level value. Subsequently, the real-time utilization rate and level values ​​calculated above are normalized to eliminate the impact of differences in dimensions and numerical magnitudes between indicators on subsequent weighted calculations, thus improving the normalized real-time utilization rate. The utilization rate is assigned a first weighting coefficient, which is a pre-defined standardized coefficient representing the weighting proportion of the real-time utilization rate in the tolerance offset threshold calculation. A second weighting coefficient is assigned to the normalized grade values, which is also a pre-defined standardized coefficient representing the weighting proportion of the grade values ​​in the tolerance offset threshold calculation. The sum of the first and second weighting coefficients is equal to one to ensure the rationality and balance of the weighted calculation. Then, the normalized real-time utilization rate value corresponding to the first weighting coefficient and the normalized grade value corresponding to the second weighting coefficient are weighted and summed to obtain the weighted sum intermediate value. The formula for the weighted sum is as follows: ,in This represents the median value of the weighted sum. Represents the first weighting coefficient. This represents the normalized real-time utilization rate. Represents the second weighting coefficient. The normalized rank values ​​are then used to calculate the intermediate values ​​through weighted summation. These intermediate values ​​are then input into a preset sigmoid function for mapping. The preset sigmoid function is a pre-defined nonlinear activation function used to map any value to a normalized range of zero to one. Its function expression is: ,in Represents the output value of the S-shaped function. This represents the intermediate value of the weighted summation of the input sigmoid function. Representing a natural constant, this nonlinear mapping process ensures the output value remains stable between zero and one. Finally, the standardized value output by the S-shaped function is multiplied by a preset baseline time threshold. This baseline time threshold is a pre-defined standardized time value used to calculate the tolerance offset threshold. It is determined by combining factors such as historical execution data of warehousing tasks, equipment operating characteristics, and task timeliness requirements. The resulting product is the tolerance offset threshold for the normalized node. The tolerance offset threshold is the maximum tolerable time offset of the affected node during task scheduling timeline adjustments. It is a core quantitative indicator used to determine whether the node's timeline offset exceeds the tolerable range. Through the above-mentioned standardized calculation throughout the entire process, the tolerance offset threshold of the affected node can be accurately obtained, providing a precise quantitative basis for subsequent node timeline offset adjustments and task scheduling optimization.

[0021] After calculating the offsets of each affected node and normalizing the corresponding tolerance offset thresholds, to further distinguish whether the node offsets are within a reasonable and controllable range, a one-to-one comparison of the offset and tolerance offset threshold for the same node is necessary. The offset refers to the actual time delay caused by equipment failure, task congestion, or dependency latency, and is a core quantitative indicator of node scheduling latency. The tolerance offset threshold is the maximum reasonable delay time a node can tolerate during task scheduling. It is a standardized threshold obtained by normalizing and weighting the node's real-time utilization rate and downstream task timeliness levels, and mapping it using a S-shaped function. During the comparison, if the offset calculated for a node is greater than its corresponding tolerance offset threshold, the node is marked as an over-threshold node. An over-threshold node is an abnormal node whose actual offset time exceeds the maximum tolerable offset time limit and requires priority scheduling adjustment. In practice, the system directly accesses the task scheduling view database of the warehouse system, which stores all task node scheduling information. The system uses a dedicated database for node status identifiers, offset values, and tolerance offset threshold values. Upon locating a node whose offset exceeds the tolerance offset threshold, the system automatically locates the corresponding status identifier field in the database. This status identifier field is a dedicated character field in the task scheduling view database used to mark different node states such as normal, delayed, and exceeding the threshold. The system automatically updates the original value of this field to the exclusive identifier value representing the exceeding threshold state, thus updating the node status to the exceeding threshold node state. Simultaneously, the system automatically extracts and records the specific offset value obtained by the exceeding threshold node in this calculation, as well as the specific tolerance offset threshold value used for numerical comparison. These two specific values ​​are then synchronously written into the offset record field and tolerance offset threshold record field corresponding to the node in the task scheduling view database. These fields are dedicated to storing the specific values ​​of node offsets and tolerance offset thresholds in the task scheduling view database. This completes the entire process of updating the status, marking the identifier, and recording the values ​​of the exceeding threshold node, providing accurate data support for subsequent task scheduling adjustments and abnormal node handling.

[0022] After completing the threshold state determination and marking operations for all warehousing task nodes, to further quantify the impact range and depth of the abnormal process on the overall task scheduling, it is necessary to sequentially perform the operations of counting the total number of threshold nodes and recording the directed edge hop count corresponding to the downstream threshold node. Specifically, counting the total number of threshold nodes involves using a breadth-first traversal of a pre-constructed warehousing end-to-end directed task dependency graph—a data structure with tasks at each stage of warehousing as nodes and execution dependencies and material flow relationships between tasks as directed edges—to completely traverse all nodes in the data structure, verifying the status identifier field of each node, filtering and accumulating the number of nodes marked as threshold nodes. The final accumulated result is the total number of threshold nodes. Recording the directed edge hop count from the downstream threshold node to the abnormal process node involves using the abnormal process node that triggered the current scheduling anomaly as the starting point for traversal, following the warehousing end-to-end directed task dependency graph... The pre-defined directed edge pointing rules for material flow in the diagram are used. A depth-first search algorithm is employed to continuously search downstream along the directed edges for all threshold nodes. After locating the downstream threshold node at the end of the task execution, the algorithm backtracks and counts the minimum number of directed edges traversed from the abnormal process node to this downstream threshold node. This directed edge refers to a unidirectional connection edge connecting two adjacent task nodes in the warehousing end-to-end directed task dependency graph, representing the order of task execution and the direction of material transfer. This minimum number is the target directed edge hop count. This statistical process starts from the abnormal process node, traverses layer by layer along the material flow, and records the number of directed edges for each reachable path. Finally, the minimum value of the number of directed edges in all feasible paths is taken as the final record. Through the above traversal and search method, the total number of threshold nodes and the corresponding directed edge hop count can be accurately counted, providing accurate data support for the subsequent abnormal assessment of warehousing task scheduling.

[0023] After completing the statistics on the total number of threshold nodes in the entire warehouse task scheduling system and calculating the number of directed edge hops for each threshold node, the scheduling system will execute a hierarchical task scheduling adjustment strategy based on the numerical relationship between the total number of threshold nodes, the number of directed edge hops for each threshold node, and the preset hop count threshold. Threshold nodes refer to task nodes whose actual execution time window exceeds the system's preset allowable time adjustment range during warehouse task scheduling. The number of directed edge hops refers to the minimum number of directed edges traversed from the abnormal process node that triggered the scheduling anomaly, along the directed edge corresponding to the material flow direction in the directed task dependency graph of the entire warehouse chain, to reach the corresponding threshold node. The preset hop count threshold refers to the number of directed edges that the scheduling system pre-sets. A predetermined upper limit for the number of directed edge hops is used to determine the boundary between local and global scheduling adjustments. When the total number of nodes exceeding the threshold count count is zero, the scheduling system will maintain the current task sequence and task identifiers unchanged. Specifically, if the total number of nodes exceeding the threshold count count is zero after traversal and statistics, meaning there are no nodes in the warehouse end-to-end task scheduling system whose task execution time windows exceed the preset allowable range, the scheduling system will not perform any task order adjustment or task identifier re-identification operations, maintaining the stability and continuity of the currently planned task execution sequence and the corresponding task identifiers. When the total number of nodes exceeding the threshold count count count is not zero and the maximum value of the directed edge hop count count corresponding to all nodes exceeding the threshold count is less than or equal to... When the preset hop count threshold is set, the scheduling system only performs local timing adjustments on nodes exceeding the threshold and their immediate downstream nodes. Specifically, if the total number of nodes exceeding the threshold is not zero (meaning there are nodes in the warehouse end-to-end task scheduling system whose task execution time windows exceed the preset allowable range, and the maximum value of the directed edge hop count for all nodes exceeding the threshold is less than or equal to the preset hop count threshold, meaning the influence range of the nodes exceeding the threshold does not exceed the system's preset local adjustment boundary), the scheduling system will activate a pre-built local rescheduling algorithm. This local rescheduling algorithm is a dedicated scheduling algorithm that only adjusts the timing of marked nodes exceeding the threshold and their immediate downstream nodes. The adjustment range of this algorithm is limited to the marked nodes exceeding the threshold and the next hop downstream nodes immediately adjacent to each node exceeding the threshold. In the warehousing end-to-end directed task dependency graph, the first task node directly connected to the threshold node via a directed edge and located downstream of the threshold node's material flow is considered the first task node in the downstream direction of the threshold node. During the adjustment process, the scheduling system will add a buffer time to the original time window of the threshold node. The buffer time is a time compensation value preset by the scheduling system to meet the task timing constraints. The specific calculation method for the buffer time is a preset unit buffer time multiplied by the number of hops of the directed edge corresponding to the threshold node. The preset unit buffer time is the time compensation benchmark value corresponding to a single directed edge preset by the scheduling system. By adding the buffer time, the task execution time window of the threshold node and its immediate downstream nodes is made to meet the system timing constraints.When the total number of out-of-threshold nodes counted by the scheduling system is not zero and the number of hops on the directed edge corresponding to any out-of-threshold node exceeds the preset hop count threshold, the scheduling system will trigger a full-link task rescheduling process. Specifically, if the total number of out-of-threshold nodes is not zero and the number of hops on the directed edge corresponding to any out-of-threshold node exceeds the preset hop count threshold, meaning the influence range of the out-of-threshold node has exceeded the system's preset local adjustment boundary, the scheduling system will initiate a global rescheduling process. This global rescheduling process refers to the scheduling system recalculating the task execution sequence for all unexecuted tasks in the entire warehousing chain based on the latest collected data on the operating status of warehousing equipment, the timing constraints of each task node, and material flow rules. This generates a new task execution sequence and completes the task time planning. This process traverses all unexecuted task nodes in the directed task dependency graph of the entire warehousing chain, combines the real-time operating status of the corresponding equipment, the timing constraints of the tasks, and the material flow rules, replans the execution time windows of each task, and generates a new task execution sequence, thereby achieving stability and rationality in the scheduling of tasks across the entire warehousing chain.

[0024] To verify the effectiveness of the offset calculation model (S2) and tolerance offset threshold judgment rule (S3) proposed in this invention in a real warehousing scenario, we constructed a typical equipment failure event using the "goods-to-person" picking operation area of ​​a large e-commerce automated distribution center as the experimental object, and collected time-series data of each downstream node along the main link.

[0025] Experimental setup: The stacker crane with the serial number RACK-03 was selected as the abnormal process node. This crane experienced a drive motor overload fault at time t=0. The system retrieved its task identifier as "T-SKU-0823" and predicted its estimated recovery time based on the historical fault database and the equipment health model. .

[0026] Using currently unexecuted tasks as nodes, and material flow (e.g., pallets from stacker crane → conveyor line → sorting station) and resource occupancy relationships as directed edges, a directed task dependency graph is constructed for the entire warehousing chain. A weighted shortest path algorithm based on equipment status (real-time processing speed, historical average failure interval) and task queue length is used to determine the longest dependency path from the faulty node to the final outbound point, which serves as the backbone link. This backbone link contains 8 downstream task nodes (numbered N1 to N8).

[0027] For each directed edge, the final estimated waiting time (in minutes) of each edge is calculated by combining the device state weight coefficient, the node queue length weight, and the additional delay factor predicted by the machine learning model: [4.2, 3.8, 5.1, 4.6, 6.3, 5.7, 4.9, 5.4].

[0028] According to step S2 of this invention, starting from the abnormal node, traverse the main link in the forward direction, accumulate the final estimated waiting time of the edges traversed to obtain the cumulative estimated waiting time, and then sum it with the estimated recovery time of 12.0 min to obtain the time offset of each node. .

[0029] Simultaneously, based on the real-time utilization rate of the associated devices at each node (range 0.42–0.89) and the timeliness level of the adjacent downstream tasks (levels 1–5, with higher levels indicating greater urgency), the tolerance offset threshold for each node is calculated after normalization, weighted summation (weighting coefficients: utilization rate 0.55, timeliness level 0.45), Sigmoid mapping, and multiplication by a baseline time threshold of 28 minutes. .

[0030] Experimental results (combined) Figure 2 ): Figure 2 The time offset of the eight nodes N1 to N8 is compared with the tolerance offset threshold.

[0031] For nodes N1 to N4, the offsets are 16.2 min, 20.0 min, 25.1 min, and 29.7 min, respectively, all of which are less than their respective tolerance thresholds (19.6 min, 22.3 min, 26.1 min, and 28.5 min). Therefore, these nodes are determined to be non-threshold nodes and do not require scheduling intervention.

[0032] For node N5, the offset of 35.3 min is greater than its tolerance threshold of 31.0 min; for node N6, the offset of 41.0 min is greater than the threshold of 34.2 min. N5 and N6 are marked as nodes exceeding the threshold.

[0033] Although the offsets of N7 and N8 increased further (45.9 min and 51.3 min), the tolerance threshold also increased accordingly (due to the increase in downstream efficiency level and changes in equipment utilization), and did not exceed the threshold.

[0034] The experiment demonstrates that the offset accumulation model proposed in this invention can accurately reflect the amplification effect of abnormal links, while the dynamic tolerance threshold mechanism can distinguish between "absorbable delay" and "delay that must be intervened in", avoiding ineffective scheduling that treats all delay nodes in the same way.

[0035] Experimental results (combined) Figure 3 ): Figure 3 The stepped bar chart (final estimated waiting time on each side) and the cumulative curve (cumulative estimated waiting time) in the figure visually demonstrate the cumulative effect of anomaly propagation.

[0036] The edge waiting times for the first two hops (N1 to N2) are relatively short (4.2 min and 3.8 min, respectively), and the cumulative waiting time gradually increases to 8.0 min. Starting from N3, due to the increased length of the device queue and fluctuations in the global task arrival rate, the edge waiting time rises to over 5.1 min, and the slope of the cumulative curve increases significantly. By N6, the cumulative estimated waiting time has reached 29.0 min (corresponding to an offset of 41.0 min), exceeding the tolerance threshold of this node and triggering an over-threshold flag.

[0037] The cumulative curve finally reached 39.3 minutes at N8, fully demonstrating the process of the abnormal impact gradually amplifying as the link depth increases. This result verifies the rationality of the offset calculation method of "accumulating the weight of directed edges and summing the recovery time" adopted in this invention. It also provides a reliable quantitative basis for subsequent hierarchical scheduling (S5) based on the number of nodes exceeding the threshold and the number of hops of the directed edge of the downstream node exceeding the threshold.

[0038] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A full-chain processing method for a warehousing system, characterized by: Including the following: S1. Obtain the task identifier and estimated recovery time of the abnormal process node device; S2. Construct a directed task dependency graph for the entire warehousing chain, specifically including: taking the currently unexecuted task as a node, taking the material flow direction and resource occupation relationship between tasks as directed edges, using a weighted shortest path algorithm based on equipment status and task queue length to determine the longest dependency path between nodes as the backbone link, taking the abnormal process node as the starting point and traversing along the backbone link in the forward direction, summing the cumulative expected waiting time of all traversed edges between each affected node and the abnormal process node with the expected recovery time, and using the result as the offset of the original task execution time window of the affected node; S3. The real-time utilization rate of the associated device of the affected node is weighted and calculated with the timeliness level value of the adjacent downstream task. The result is normalized and used as the tolerance offset threshold of the node. The offset is compared with the tolerance offset threshold, and the node with the offset greater than the tolerance offset threshold is marked as the threshold node. S4. Count the total number of nodes exceeding the threshold, and record the number of directed edge hops between the downstream node exceeding the threshold and the abnormal process node. S5. When the total number of nodes exceeding the threshold is zero, maintain the current task sequence and task identifier. When the total number of nodes exceeding the threshold is not zero and the number of hops of the directed edges does not exceed the preset hop count threshold, only perform local timing adjustments on the nodes exceeding the threshold and their immediate downstream nodes. Otherwise, trigger full-link task rescheduling.

2. The end-to-end processing method for a warehousing system according to claim 1, characterized in that: In S2, the equipment status is obtained by real-time collection of multi-dimensional operating parameters of each physical device in the warehousing system. The multi-dimensional operating parameters include at least the device's task identifier, the device's real-time processing speed, and the device's historical average failure interval time. The device's real-time processing speed is calculated by counting the number of tasks completed by the device within a preset time window and combining it with the standard task time. The device's historical average failure interval time is obtained by extracting the time period of continuous fault-free operation of the device in the past period from the device maintenance log database and calculating the average value.

3. The end-to-end processing method for a warehousing system according to claim 2, characterized in that: When using a weighted shortest path algorithm based on device status and task queue length to determine the longest dependency path between nodes as the backbone link, the real-time processing speed of the device and the historical average failure interval time of the device extracted from the device status are first normalized. Then, the reciprocal of the normalized real-time processing speed of the device and the reciprocal of the normalized historical average failure interval time of the device are weighted and summed. The result is used as the device status weight coefficient. At the same time, the number of tasks waiting to be executed on each node is obtained as the task queue length, and the task queue length is normalized to obtain the queue length weight. Finally, in the constructed directed task dependency graph with unexecuted tasks as nodes, the weight of the directed edge connecting two nodes is defined as the sum of the device state weight coefficient of the device associated with the starting node of the edge and the queue length weight of the node. Device reliability and current load status are quantified together as path weights.

4. The end-to-end processing method for a warehousing system according to claim 3, characterized in that: The path weight is used as the base value for the estimated waiting time. The specific calculation method for the estimated waiting time of each directed edge is as follows: The weight of the directed edge is multiplied by an additional delay factor dynamically predicted by a machine learning model. The machine learning model takes the current time, the device state weight coefficient of the device associated with the starting node of the directed edge, the device state weight coefficient of the device associated with the ending node of the directed edge, and the global average task arrival rate of the warehouse system as input features, and outputs a proportional factor representing the degree of additional congestion generated by the directed edge under the current network flow conditions. The product obtained is the final estimated waiting time of the directed edge segment. The sum of the final estimated waiting times of all directed edges on the path from the abnormal process node to each affected node is the cumulative estimated waiting time.

5. The end-to-end processing method for a warehousing system according to claim 4, characterized in that: The construction of the directed task dependency graph across the entire warehousing chain specifically includes: Using all tasks in the warehouse management system that are marked as not started or issued but not completed as nodes, the material delivery order between tasks and the mutual exclusion relationship of shared equipment resources are determined according to the material handling list and resource scheduling table. These are used as directed edges to connect the relevant nodes and indicate the direction, forming a directed acyclic graph structure. In the directed acyclic graph structure, all directed edges are assigned weights, and an improved Dijkstra algorithm is used to find the path with the largest sum of weights from the starting node to all edges on the path, which is denoted as the maximum weight cumulative path. The longest dependency path between nodes is determined based on the maximum weight cumulative path, and the longest dependency path is defined as the backbone link. The backbone link represents the key constraint channel between the starting task and the ending task that is most affected by the device state and queue load.

6. The end-to-end processing method for a warehousing system according to claim 5, characterized in that: Starting from the abnormal process node, each downstream node is visited sequentially in the direction of the main link. For each visited node, the final estimated waiting time of each directed edge traversed from the abnormal process node to the current node is accumulated to obtain the cumulative estimated waiting time value. The cumulative estimated waiting time value is added to the estimated recovery time, and the sum is used as the amount of time to postpone the original planned task execution time window of the current node, i.e., the offset.

7. The end-to-end processing method for a warehousing system according to claim 1, characterized in that: The equipment monitoring system obtains in real time the ratio of the actual working time of the warehouse equipment associated with the affected node in the previous statistical period to the total time of the statistical period, as the real-time utilization rate of the equipment. At the same time, the timeliness level codes of the downstream tasks adjacent to the affected nodes are obtained from the task management system. The timeliness level codes are pre-divided into multiple levels according to the urgency of the promised delivery time of the task order, and different levels are mapped to different level values. The real-time utilization rate and the level value are normalized respectively. The normalized real-time utilization rate is assigned a first weight coefficient, and the normalized level value is assigned a second weight coefficient. The first weight coefficient and the second weight coefficient are weighted and summed. The result of the weighted sum is then input into a preset S-shaped function for mapping processing, so that the output value falls between zero and one. Finally, the output value is multiplied by a preset reference time threshold. The resulting product is the tolerance offset threshold of the node after normalization.

8. The end-to-end processing method for a warehousing system according to claim 7, characterized in that: The offset is compared with the tolerance offset threshold of the same node. Nodes whose offset is greater than the tolerance offset threshold are marked as threshold-exceeding nodes. Specifically, in the task scheduling view database of the warehouse system, the status identifier field of nodes whose offset is greater than the tolerance offset threshold is updated and marked as threshold-exceeding nodes. At the same time, the specific value of the offset calculated by the threshold-exceeding node and the specific value of the tolerance offset threshold used for comparison are recorded.

9. The end-to-end processing method for a warehousing system according to claim 8, characterized in that: The total number of nodes exceeding the threshold refers to traversing all nodes marked as exceeding the threshold state and counting their total number in the data structure of the directed task dependency graph of the entire warehouse chain. The number of directed edge hops between the downstream threshold node and the abnormal process node refers to starting from the abnormal process node, searching downstream along the material flow direction in the directed task dependency graph of the entire warehousing chain until the downstream threshold node is found, and recording the minimum number of directed edges traversed from the abnormal process node to the downstream threshold node, which is denoted as the number of directed edge hops.

10. The end-to-end processing method for a warehousing system according to claim 9, characterized in that: The statement that the current task sequence and task identifier are maintained when the total number of out-of-threshold nodes is zero means that if the total number of out-of-threshold nodes counted is zero, the scheduling system will not perform any task order adjustment or re-identification operation. The phrase "when the total number of out-of-threshold nodes is not zero and the number of directed edge hops does not exceed the preset hop count threshold, only local timing adjustments are made to the out-of-threshold nodes and their immediate downstream nodes" means that if the total number of out-of-threshold nodes is not zero and the maximum value of the directed edge hop count among all out-of-threshold nodes is less than or equal to the preset upper limit of the hop count, then the scheduling system only starts the local rescheduling algorithm. The adjustment range of the local rescheduling algorithm is limited to the marked out-of-threshold nodes and the next hop downstream nodes immediately adjacent to each out-of-threshold node. The constraints are satisfied by adding a buffer time to the original time window of the out-of-threshold nodes. Otherwise, triggering full-link task rescheduling means that if the total number of nodes exceeding the threshold is not zero and the number of hops on a directed edge exceeds the preset hop threshold, the scheduling system will trigger a global rescheduling process, recalculating and generating task execution sequences and time plans for all unexecuted tasks based on the latest device status and constraints.

Images