Multi-line interpenetration logistics conveying line scheduling algorithm

Abstract

The invention discloses a multi-line interpenetration logistics conveying line scheduling algorithm, and particularly relates to the technical field of logistics conveying line scheduling. The method comprises the following steps: constructing a dynamic graph structure model, dynamically mapping each node and path in a logistics system into a graph structure, carrying out embedded learning on a node state by utilizing a graph neural network, identifying an abnormal propagation area in real time when a fault occurs, pruning a scheduling graph, removing high-risk nodes and paths, and reconstructing a safe and feasible task path. And meanwhile, the scheduling execution accuracy is guaranteed through conflict detection and feasibility verification, and online adaptive optimization of the graph model is realized in combination with a data acquisition and feedback mechanism in an operation process, so that the prediction capability, the disaster recovery scheduling capability and the overall operation robustness of the scheduling system on complex anomalies are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of logistics conveying line scheduling, in particular to a multi-line interpenetration logistics conveying line scheduling algorithm. BACKGROUND

[0002] Multi-line interpenetration logistics conveying line scheduling refers to, in a logistics system, for the complex scene of multiple conveying lines crossing and operating in parallel, coordinating and optimizing the time sequence, path and priority of material transportation through intelligent scheduling algorithm, so as to realize efficient and orderly operation of each conveying line, avoid congestion and conflict, and improve the conveying efficiency and response capability of the whole logistics system.

[0003] The prior art has the following shortcomings:

[0004] In a multi-line interpenetration logistics system, the scheduling system often regards equipment failure such as elevator jam as a local exception, only suspends the directly associated task, and ignores the possible chain reaction it may cause to the whole system. In actual operation, such local failure often transmits upward through path dependence, causing upstream AGV task backlog, path conflict unable to be released in time, and finally causing scheduling logic to be stuck and even the whole system to be paralyzed. SUMMARY

[0005] The purpose of the present application is to provide a multi-line interpenetration logistics conveying line scheduling algorithm to solve the problems in the background art.

[0006] In order to achieve the above purpose, the present application provides the following technical scheme: a multi-line interpenetration logistics conveying line scheduling algorithm, comprising:

[0007] A dynamic graph structure model is constructed for the logistics conveying system, and each conveying unit node and task flow path is defined as a node and an edge in the graph structure;

[0008] Based on the constructed graph structure, a graph neural network model is used to learn the state feature embedding of each node;

[0009] When an abnormality is detected in a node, the current system state is input into the graph neural network model in real time, the affected path area and the propagation risk score are obtained, and the abnormality propagation area is automatically identified;

[0010] According to the predicted abnormality propagation area, the scheduling graph is pruned, the high-risk path nodes are excluded, and the logistics scheduling path is regenerated based on the pruned graph structure;

[0011] The generated new scheduling path is subjected to conflict detection and execution feasibility verification, and the feasible scheduling scheme is issued to the control system for execution;

[0012] The running data is continuously collected during the execution, and the scheduling execution result is fed back for online update training of the graph neural network model.

[0013] Preferably, in the dynamic graph structure model, the nodes include a conveying section segment unit, a junction, a buffer area, a vertical lifting device, a sorting device, an AGV parking point, or a transfer point.

[0014] And for each node, its node type, current running state, material load, congestion level, and historical fault label are maintained in real time as input features of the graph neural network.

[0015] Preferably, after analyzing the node input feature vector, a node average waiting chain length is generated, and the generation method is as follows: a directed graph G=(V,E) is established for the task scheduling system, and the weight w(u,v) of each edge (u,v) ∈E is initialized as: ; in the formula, represents the cumulative processing time of the current waiting task on the node u, represents the average time consumption from u to v;

[0016] A longest path length table is defined for each node v∈V: L(v)=longest path length to node v; all nodes are traversed in topological order, and each edge (u,v) ∈E(u, v) is updated as follows: ; it is indicated that the longest path length of the node v is equal to the maximum path length of all predecessor nodes u plus the transmission weight w(u,v);

[0017] For the target node vtarget, all valid predecessor paths are backtracked, and all path lengths are recorded; a maximum backtracking depth limit is set or only the first K paths are taken to calculate the average as the node average waiting chain length, and the expression is as follows: ; in the formula, is the node average waiting chain length.

[0018] Preferably, after analyzing the edge input feature vector, a path task backflow probability is generated, and the generation method is as follows:

[0019] A fixed time window is selected, and for the path u→v, all task identifiers sent from the node u to v are recorded; these tasks are accumulated to form (u→v); for each task: it is judged whether there is a backflow task, and accumulated to (u→v); the path task backflow probability is calculated, and the expression is as follows: ; in the formula, is the path task backflow probability.

[0020] Preferably, the system inputs the current graph structure state, including the average waiting chain length of nodes or the probability of path task backflow, into the trained graph neural network model every scheduling period;

[0021] The model output is: y^impact(v), representing the risk score of node v being affected by the anomaly at present; y^spread(v), representing the probability of node v continuing to spread the anomaly;

[0022] According to the set threshold: if y^impact(v)>0.6: mark as an affected node; if y^spread(v)>0.5: mark as an anomaly propagation path node;

[0023] All affected nodes and related edges are extracted from the original graph to construct an anomaly propagation subgraph.

[0024] Preferably, when pruning the scheduling graph, the anomaly propagation score threshold is set, the nodes with anomaly impact scores exceeding the preset value and their associated paths are deleted, and the reachability and task integrity of the scheduling graph after pruning are judged, and if there are breakpoint tasks, scheduling agent nodes or jump nodes are introduced to perform path reconstruction.

[0025] Preferably, in the path reconstruction process, a multi-objective path search algorithm is used to generate multiple candidate paths according to the path safety score, task timeliness requirement and resource occupation conflict, and the candidate paths are filtered according to priority, and a resource occupation time slice graph is established for each candidate path to judge whether there is resource conflict, time conflict or device mutual exclusion.

[0026] In the above technical solution, the technical effects and advantages provided by the present application are:

[0027] 1、The multi-thread interleaved logistics conveying line scheduling algorithm provided by the present application breaks through the limitation of traditional systems regarding abnormal events as local events, and through the construction of a dynamic graph structure and the introduction of a graph neural network for anomaly propagation modeling, the system has the ability to perceive and predict the global scheduling ripple effect caused by local faults. The scheduling system can identify potential affected areas through real-time graph state reasoning at the initial stage of fault occurrence, automatically prune high-risk paths and reconstruct the task scheduling scheme, significantly improving the robustness and disaster recovery capability of the scheduling strategy.

[0028] 2、The application realizes online iterative optimization of the graph neural network model by continuously collecting system operation data and constructing a feedback learning mechanism, so that the scheduling system has self-adaptive learning ability and can continuously improve the prediction accuracy and response efficiency as the task load changes and the equipment state evolves. The overall scheme realizes the closed-loop fusion of scheduling graph modeling, risk perception, path reconstruction, conflict control and model learning, is suitable for intelligent logistics scheduling systems in multi-line interweaving, high concurrency and variable scenarios, and can effectively improve the stability, intelligent level and global optimal scheduling effect of the system. BRIEF DESCRIPTION OF DRAWINGS

[0029] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings described below are only some embodiments described in the present application, and other drawings can also be obtained by those skilled in the art based on these drawings.

[0030] Figure 1 The method mind map of the present application. DETAILED DESCRIPTION

[0031] In order to make the purpose, technical scheme and advantages of the embodiments of the present application more clear, the technical scheme in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0032] Embodiment, please refer to Figure 1 The multi-line interweaving logistics conveying line scheduling algorithm described in the present embodiment includes:

[0033] A dynamic graph structure model is constructed for the logistics conveying system, and each conveying unit node and task flow path is defined as a node and an edge in the graph structure.

[0034] Based on the constructed graph structure, a graph neural network model is used to perform state feature embedding learning on each node.

[0035] When an abnormal node is detected, the current system state is input into the graph neural network model in real time to obtain the affected path area and the propagation risk score, and the abnormal influence propagation area is automatically identified.

[0036] According to the predicted abnormal influence propagation area, the scheduling graph is pruned to exclude high-risk path nodes, and the logistics scheduling path is regenerated based on the pruned graph structure.

[0037] The generated new scheduling path is subjected to conflict detection and execution feasibility verification, and the feasible scheduling scheme is issued to the control system for execution.

[0038] During the execution process, running data is continuously collected, and the scheduling execution result is fed back for online update training of the graph neural network model.

[0039] All physical or logical units with control logic, state changes, and resource competition in the logistics system are abstracted as nodes of the graph. They include but are not limited to: conveyor belt segments (each segment can be independently controlled to start and stop), intersection points (where multiple conveyor lines or AGV paths converge), buffer zones or storage shelves, vertical lifting devices, sorting devices or mechanical arm interaction points, and AGV parking points, charging points, and transfer points.

[0040] Each node should dynamically maintain the following attributes during graph initialization and running for the input features of the graph neural network: node type code (such as "01" for intersection point and "02" for elevator), current running state (normal, abnormal, maintenance, shutdown, etc.), real-time load (such as the number of materials currently on the conveyor segment, unit load), congestion level (estimated by accumulation detection or queue length), and fault history label (whether there is a fault record in the recent time window).

[0041] If there is a logistics task transmission relationship between any two logistics nodes, a directed edge is created in the graph to represent the task flow from one node to another. For example, conveyor segment A→intersection B→elevator C will form directed edges A→B and B→C.

[0042] Each edge also has variable attributes reflecting path transmission capacity and running state, including: average transmission time (updated dynamically based on historical or real-time data), traffic capacity (i.e., the number of materials that can pass through per unit time), current occupancy state (whether it is occupied by a certain task), and edge weight (a cost function used for path search or scheduling optimization, such as time, distance, priority).

[0043] If there is a possibility of bidirectional flow in the path, such as AGV bidirectional travel, a bidirectional edge should be established; if a segment is a shared segment (shared by multiple AGVs / tasks), a "mutual exclusion weight" attribute should be set for conflict judgment.

[0044] The graph structure needs to be updated in real time with the changes in the logistics system state, and a state synchronization is performed once every scheduling period (such as every 1 second): node state update: update the node state according to the PLC signal or scheduling feedback; edge state update: calculate the task transmission delay and congestion level between nodes, and dynamically update the edge weight; add / delete nodes or edges: dynamically adjust the graph structure when the system adds new devices, disables paths, or the task structure changes.

[0045] Access to the underlying perception system (such as AGV positioning, RFID, sensors, WMS / MES) maps the data into a graph structure input format: the state of each node / edge is collected and converted into a standard vector representation by the state perception component; the vector can be used as the node input feature hv and edge feature euv of the graph neural network.

[0046] A sparse graph data structure is used to improve storage efficiency and processing speed, which can be represented in the following ways: Adjacency List represents the graph connection relationship; Feature Matrix stores the feature vector of nodes / edges; use a graph database (such as Neo4j) or a graph computing framework (such as DGL, PyTorch Geometric) to manage the graph structure.

[0047] Numerical normalization (such as Min-Max, Z-score) is performed on the attributes of nodes and edges; One-hot or embedding processing is performed on discrete types (such as fault state, type code). The above graph structure is converted into an input format acceptable to the graph neural network, such as (nodes, edges, features) combination; support dynamic graph time series snapshot input (for time series GNN training).

[0048] Building a dynamic graph structure model is not only the core foundation of data modeling for the scheduling system, but also a bridge connecting intelligent prediction and optimal control. By deeply modeling the relationship between nodes and paths in the logistics system and integrating real-time perception information, the subsequent graph neural network can have the ability of "structure perception", "state reasoning" and "dynamic evolution", providing a reliable data structure foundation and learning interface for disaster recovery scheduling, anomaly prediction and path reconstruction.

[0049] Based on the constructed graph structure, a graph neural network model is used to learn the state feature embedding of each node. The graph neural network aggregates the state of the node neighborhood and outputs the abnormal influence prediction value and influence propagation probability of each node, specifically including:

[0050] A feature vector is constructed for each node, including: node type (such as intersection, elevator, buffer zone) → One-hot encoding or embedding encoding; current running state (normal, abnormal, fault) → numerical mapping (such as 0, 1, 2); current load (number of materials or occupancy ratio) → numerical normalization; congestion degree (stacking time, queue number) → real number feature; fault history identifier → binary variable (whether a fault has occurred recently)

[0051] Each directed edge establishes the following feature vector: transmission time (dynamic change) → real value; traffic capacity (throughput) → real number normalization; current path state (free, congestion) → discrete coding; path directionality → used as a graph structure connection index (u→v).

[0052] Continuous numerical features are normalized (e.g., min-max normalization), and discrete variables are one-hot encoded or embedded.

[0053] A neural network model with graph aggregation capability is used, such as GCN (Graph Convolutional Network), GraphSAGE (sampling neighbor aggregation, suitable for large graphs), and GAT (Graph Attention Network with attention mechanism).

[0054] The GNN updates the hidden state of each node by the following formula: , where represents the hidden vector of the k-th layer node v, represents the neighbor node set of node v, AGGREGATE: such as summation, average, max pooling, attention weighting, represents the learnable weight of the k-th layer; σ represents the activation function, such as ReLU.

[0055] Usually 2-3 layers of graph convolution layers are selected to capture the influence of multi-hop neighbors; BatchNorm and Dropout layers are added after each layer to enhance the generalization ability.

[0056] Train the graph neural network to learn the abnormal influence prediction model: collect historical scheduling data, including: system running state graph (node state + edge weight), fault / abnormal node and its subsequent influence area (can be used as a supervision signal), construct two types of training targets for each node: abnormal influence score: the possibility of the node being affected by the abnormality in the current graph structure (0~1), propagation probability: whether the node will spread the abnormality to other nodes (0 or 1 classification); if it is a regression output (influence score): use MSE (mean square error) loss; if it is a categorical output (whether to spread): use cross-entropy loss; support multi-task training (Multi-task loss).

[0057] When any node state is detected to be abnormal (e.g., device failure, long-time retention, data link breakage) during system operation, the scheduling control center immediately triggers the "abnormal identification flag" and marks the fault node vfail, and synchronizes the following information:

[0058] Fault occurrence timestamp, fault type (hardware card block, timeout non-response, communication interruption, etc.), node state snapshot (including peripheral node state); this event as a trigger condition, start real-time GNN inference process.

[0059] The scheduling system collects the status of all nodes and edges from the state-aware components in the current period, and constructs the input graph structure, including:

[0060] Node input feature vector (such as load, status code, fault flag);

[0061] Edge input feature vector (such as current path transmission time, traffic state);

[0062] Graph topology structure (connection relationship between nodes and nodes);

[0063] This graph structure dynamically reflects the running state snapshot of the entire system as input into the graph neural network model.

[0064] Among them, the node average waiting chain length is generated after analyzing the node input feature vector, and the generation method is:

[0065] A directed graph G=(V,E) is established for the task scheduling system, and the weight w(u,v) of each edge (u,v) ∈E is initialized as: ; In the formula, represents the cumulative processing time of the current waiting task on node u, represents the average time consumption from u to v;

[0066] Since the scheduling graph is a DAG, the graph can be topologically sorted to generate a node sequence ; Ensure that the direction of all edges is: if , then i

[0067] Define the longest path length table for each node v ∈V: L(v)=the longest path length to node v; Initially: L(v)=0 if v is the source node (no incoming edge); L(v)=−∞, other nodes;

[0068] According to the topological order, traverse all nodes, and perform update for each edge (u,v) ∈E(u, v): ; It means that the longest path length of node v is equal to the maximum path length of all its predecessor nodes u plus the transmission weight w(u,v).

[0069] For the target node vtarget, backtrack all valid predecessor paths, and record all path lengths ; You can set a maximum backtracking depth limit or only take the top K paths to calculate the average as the node average waiting chain length, the expression is: ; In the formula, The average waiting chain length of the node.

[0070] After analyzing the edge input feature vector, the path task backflow probability is generated, and the generation method is:

[0071] A fixed time window (such as the past 1 hour, the past N scheduling periods) is selected to count the task behavior data, and the short-term reaction sensitivity and long-term trend are balanced. For path u→v: record all task identifiers (such as task ID, AGV number, tray number, etc.) sent from node u to v; accumulate these tasks to form (u→v); for each of the above tasks: judge whether it has the following behaviors in the middle:

[0072] Turned back by the system (such as AGV executing a withdrawal instruction);

[0073] Returned to u or its upstream node (cache failure, task failure);

[0074] Detour but not enter the target node v;

[0075] If one of the above conditions is met, it is recorded as a backflow task and accumulated to (u→v); it can be identified through scheduling system logs, path reconstruction records, and exception labels.

[0076] Calculate the path task backflow probability, the expression is: ; In the formula, is the path task backflow probability, if the result is close to 0: the path is stable; if it is close to or exceeds 0.2~0.3: it indicates that there may be problems with the scheduling strategy or the path logic is chaotic.

[0077] The system will input the current graph structure state, including the average waiting chain length of the node or the path task backflow probability, into the trained graph neural network model every scheduling period;

[0078] The model outputs two results:

[0079] y^impact(v): the risk score of node v being affected by the anomaly (continuous value);

[0080] y^spread(v): whether node v is likely to continue to spread the anomaly (Boolean probability);

[0081] According to the set threshold: if y^impact(v)>0.6: mark it as an affected node; if y^spread(v)>0.5: mark it as an anomaly propagation path node;

[0082] Extract all affected nodes and related edges from the original graph, construct an anomaly propagation subgraph, which is used for the next step of scheduling graph pruning and path reconstruction.

[0083] After completing the graph neural network inference, mark the nodes with y^impact(v)>0.6 as high-risk;

[0084] Remove these nodes from the scheduling graph, including:

[0085] All incoming edges (paths from other nodes to this node);

[0086] All outgoing edges (paths from this node to other nodes);

[0087] Suspend all current scheduled tasks and new task access for this node.

[0088] The removal operation can be directly operated in the scheduling graph data structure, such as updating the adjacency list or node list.

[0089] Scan all path edges for propagation scores;

[0090] If a path score exceeds a certain threshold (e.g., > 0.5), consider the path unsafe;

[0091] Remove the path (directed edge) from the scheduling graph;

[0092] If there is a backup path (such as multi-path switching), mark the main path as unavailable and start the backup path priority usage strategy.

[0093] After pruning, check whether the graph structure still has: connectivity between the starting point and the ending point; whether the key task is isolated; whether there is a situation that the task must pass through the pruned node / path. If there are the above problems, further execute the scheduling reconstruction mechanism or introduce temporary scheduling agent logic (see the following path generation).

[0094] After graph pruning, a "healthy path graph" or "disaster recovery scheduling graph" is obtained, which is used for the next round of scheduling path generation.

[0095] According to the current task list to be scheduled, determine the starting node, target node, time requirement (such as the latest arrival time, priority level), and special path preference (such as only high-speed segments, avoiding elevators) for each task. Establish the mapping requirements between tasks and the remaining available graph.

[0096] Path planning algorithms: A* algorithm: considering the minimum cost function (time, risk); Dijkstra: used for the shortest path priority strategy in the graph; K-shortest path algorithm: generating multiple alternative paths for comparison; constraint graph traversal algorithm: used to handle task windows, node capacity, and other conditions. Algorithm input: current graph, start and end points, feasible constraints, output one or more feasible paths.

[0097] Due to the simultaneous scheduling of multiple tasks, the following needs to be performed: path overlap detection (whether there are shared nodes / paths); time window conflict judgment (whether there is a conflict in the same time period); AGV or conveying device resource occupation check.

[0098] If conflicts are found, the following strategies can be used for optimization:

[0099] Adjust the start time of the task; use alternative paths; reduce the scheduling priority of part of the low-priority tasks; introduce local cache or transit node buffer scheduling.

[0100] The final scheduling path structure includes: the path node sequence of each task; the time window of each node that the task is expected to pass through; the arrangement of key device use in the path. This structure will be issued to the control system (such as WCS, AGV scheduling controller) and connected to the real-time state system for execution.

[0101] For all generated new scheduling paths, check whether there are the following conflict types:

[0102] Node conflict: multiple tasks plan to occupy the same conveying section, intersection, or lifting device at the same time;

[0103] Edge path conflict: multiple tasks share the same path (i.e. edge in the graph) in the same time period;

[0104] Device mutual exclusion conflict: such as multiple AGVs requesting to enter a single AGV channel, narrow channel, etc. Implementation: create a "time slice resource occupation table" for each task path; scan all path time slices to determine whether there is an overlap; use conflict window detection algorithms (such as sliding time window intersection judgment) on the scheduling graph for fast filtering.

[0105] If there is a relationship between tasks as follows, conflicts may also occur:

[0106] A task must wait for another task to complete before starting; a node needs to complete task A first, then task B; multiple batches of single tasks need to share a key cache or loading point.

[0107] Solution: analyze the dependency graph between tasks; set "start waiting conditions" for tasks with dependencies; strengthen scheduling sequence constraint logic.

[0108] Load limits for elevators / sorting equipment; reverse traversal restrictions for AGV paths; uniqueness detection of path usage direction (e.g., only one-way passable paths); buffer capacity limits (whether there are available slots at the target point or transfer point).

[0109] After ensuring there are no logical conflicts, it is also necessary to verify the actual execution capability: determine whether the nodes in each path are in an "executable state" within the expected scheduling period; consider whether the nodes are scheduled for maintenance, power-on / off, or task mode switching; verify the device load capacity: whether the current task exceeds the processing limit.

[0110] Accumulate the latency of each node / edge in the path; ensure that the total duration of the entire path does not exceed the time constraint of the task; ensure that the difference in pace between task batches is controllable, and avoid system jitter or backlog.

[0111] Comprehensive reachability verification: Does the starting point meet the "starting conditions" (is the task ready and resources available)? Is the path completely connected? Does the target location have the capability to receive tasks (space / capability / priority)?

[0112] The validated scheduling path scheme is converted into a data format that the system can recognize, including: the node sequence of each task; the task start time and the estimated entry / exit time of each node; the mapping of equipment control commands on the path (such as start / stop commands and AGV action codes); and conflict avoidance parameters (such as avoidance time and detour commands).

[0113] The scheduling plan is sent to: WCS system via API, message queue or industrial communication protocol (such as OPC-UA, Modbus, MQTT): control conveyor belts, sorting, lifting and other equipment; AGV scheduler: assign specific travel paths and task nodes to AGVs; MES or WMS system: update task status and inventory flow.

[0114] The control system sends back a "successful reception" signal; if a path scheduling fails or the task is suspended, a "rollback mechanism" is triggered: the alternative path is switched first; or the scheduling path generation module is re-entered to execute the next plan.

[0115] The control system periodically reports the task execution status (such as node arrival, equipment execution success, abnormal interruption, etc.); the scheduling system updates the system diagram status based on the feedback and makes the next round of scheduling decisions.

[0116] The exception interruption handling mechanism is triggered again if a new exception (carding, timeout, task conflict) is encountered during execution: graph structure state update; exception prediction module activation; graph pruning → path reconstruction → scheduling recalculation.

[0117] During the execution of scheduling tasks, the system needs to continuously collect the following types of data from devices, control systems, and sensor networks:

[0118] Node-level operational data: Actual task entry and exit time; Node state change log (normal / abnormal / blocking); Device fault flag, maintenance cycle record; Buffer capacity variation and queue length.

[0119] Path edge-level operational data: Actual path travel time; Whether there is a backflow behavior; Path interruption, switching or jumping execution record; AGV travel trajectory or conveying line congestion state.

[0120] Task-level execution result: Scheduling success and failure flag; Success rate statistics (whether completed on time); Scheduling conflict occurrence times and re-scheduling times.

[0121] The collected raw data is usually heterogeneous and time-inconsistent log data. In order to facilitate model use, the following processing is required:

[0122] Time alignment processing: Aligning each node / edge data with the task timestamp as the center;

[0123] Feature extraction: Convert the original signal to a structured feature vector, such as "waiting time fluctuation" and "backflow times";

[0124] Abnormal labeling: Through pre-set rules or manual review, label the abnormal path or node that has occurred (for supervised learning);

[0125] Sample sliding window generation: Construct a training sample set with graph structure + state snapshot + execution label as a unit.

[0126] The system performs the following process every certain time interval or when the update condition is triggered:

[0127] Combine the scheduling execution results and graph structure state of the recent period to generate new training samples; Sample content includes: Node / edge feature vector, graph structure topology, target label (abnormal score, propagation score).

[0128] According to the existing graph neural network model architecture, adopt any of the following strategies:

[0129] Add new samples to the existing training set; Continue training for several rounds (usually fine-tune with small learning rate) based on the original model; Use early stopping to control to avoid overfitting.

[0130] Fix the underlying GNN structure and only update the output layer weight; Preserve the original model structure and quickly adapt to the latest state distribution.

[0131] Weight amplification for samples with large prediction errors in the recent period; Introduce a sample importance factor to improve the system's learning ability for rare abnormal events.

[0132] After online training, run prediction test on simulation environment or spare graph structure first; verify whether the accuracy improvement of abnormal area identification is significant; if it passes the accuracy and robustness threshold detection, the main model can be replaced.

[0133] Support switching of graph neural network model without interrupting system operation; the current model continues to process tasks until the end of the scheduling period; the new model is enabled at the start of the next period.

[0134] The above formulas are dimensionless numerical calculations, and the formulas are obtained by software simulation of a large amount of data to obtain a formula of the latest real situation, and the preset parameters in the formula are set by a person skilled in the art according to the actual situation.

[0135] The above embodiments can be realized wholly or partially by software, hardware, firmware or any other combination. When realized by software, the above embodiments can be realized wholly or partially in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are wholly or partially generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another, for example, the computer instructions can be transferred from one website, computer, server or data center to another website, computer, server or data center through wired or wireless (such as infrared, wireless, microwave, etc.) mode. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center and the like containing one or more available medium sets. The available medium can be a magnetic medium (such as a floppy disk, a hard disk, a magnetic tape), an optical medium (such as a DVD) or a semiconductor medium. The semiconductor medium can be a solid-state disk.

[0136] It should be understood that the term "and / or" in this document is merely used to describe associated objects, and can represent the three conditions of "A", "B", and "A and / or B", i.e., the three conditions of "A alone", "B alone", and "A and B together". In addition, the character " / " in this document generally represents an "or" relationship between the associated objects, but can also represent an "and / or" relationship. Those skilled in the art can realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed in this document can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0137] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application.

Claims

1. A multi-threaded interleaved material flow conveyor line scheduling algorithm, characterized by: The application relates to a logistics transportation system dynamic graph structure model, which defines each transportation unit node and a task flow path as a node and an edge in a graph structure. Based on the constructed graph structure, a graph neural network model is used for state feature embedding learning of each node. When an abnormal node is detected, the current system state is input into the graph neural network model in real time to obtain an affected path area and a propagation risk score, and an abnormal influence propagation area is automatically identified. According to the predicted abnormal influence propagation area, the scheduling graph is pruned, high-risk path nodes are excluded, and a logistics scheduling path is regenerated based on the pruned graph structure. The generated new scheduling path is subjected to conflict detection and execution feasibility verification, and a feasible scheduling scheme is issued to a control system for execution. During the execution process, running data is continuously collected, and the scheduling execution result is fed back for online update training of the graph neural network model. In the dynamic graph structure model, the nodes include a transportation section segmented unit, a junction, a buffer area, a vertical lifting device, a sorting device, an AGV parking point or a transfer point.

2. The multi-threaded in-line flow conveyor scheduling algorithm of claim 1, wherein: The node type, current running state, material load, congestion level and historical fault label of each node are maintained in real time as input features of the graph neural network. After analyzing the edge input feature vector, a path task backflow probability is generated.

3. The multi-threaded in-line flow conveyor scheduling algorithm of claim 1, wherein: After analyzing the node input feature vector, the node average waiting chain length is generated, and the generation method is: a directed graph G=(V, E) is established for the task scheduling system, and the weight w(u, v) of each edge (u, v) ∈E is initialized as: ; In the formula, represents the cumulative processing time of the current waiting task on node u, represents the average time consumption from u to v; Define a longest path length table for each node v e V: L(v) = longest path length to node v; traverse all nodes in topological order and perform an update for each edge (u, v) e E(u, v): ; indicates that the longest path length of node v is equal to the maximum path length of all its predecessor nodes u plus the transmission weight w(u, v); For target node vtarget, trace all valid predecessor paths and record all path lengths ; set maximum trace depth limit or only take the first K paths to calculate the average as the node average waiting chain length, expressed as: ; in which, is the node average waiting chain length.

4. The multi-threaded in-line flow conveyor scheduling algorithm of claim 3, wherein: In each scheduling period, the current graph structure state, including the node average waiting chain length or the path task backflow probability, is input into the trained graph neural network model. Select a fixed time window, for path u→v, record all task identification from node u to v; accumulate these tasks to form (u→v); for each task: judge whether it exists anti-flow task, accumulate to (u→v); calculate path task anti-flow probability, expression is: ; in the formula, path task anti-flow probability.

5. The multi-threaded in-line flow conveyor scheduling algorithm of claim 4, wherein: The model output is y^impact(v), which represents the risk score of the node v being affected by the exception, and y^spread(v), which represents the probability of the node v continuing to propagate the exception. According to the set threshold value, if y^impact(v)>0.6, the node is marked as an affected node; if y^spread(v)>0.5, the node is marked as an abnormal propagation path node. All affected nodes and related edges are extracted from the original graph to construct an abnormal propagation subgraph. When the scheduling graph is pruned, the abnormal propagation score threshold value is set, the nodes with an abnormal influence score higher than the preset value and their associated paths are deleted, and the reachability and task integrity of the scheduling graph after pruning are judged.

6. The multi-threaded in-line flow conveyor line scheduling algorithm of claim 5, wherein: If there is a breakpoint task, a scheduling agent node or a jump node is introduced to perform path reconstruction.

7. The multi-threaded in-line flow conveyor scheduling algorithm of claim 6, wherein: In the path reconstruction process, a multi-objective path search algorithm is used to generate multiple candidate paths based on path safety score, task timeliness requirement and resource occupation conflict, and the candidate paths are screened according to priority, and a resource occupation time slice graph is established for each candidate path to judge whether there is a resource conflict, a time conflict or a device mutual exclusion relationship.

Images