A flexible production line scheduling method and system based on multi-agv cooperation

Abstract

The application discloses a kind of based on the flexible production line scheduling method and system of multiple AGV cooperation, specifically relates to flexible manufacturing system resource scheduling technical field, for solving the defect that current dynamic priority mechanism causes AGV resource lock when high priority task is concurrent;By obtaining production line task priority data;Estimate high priority task coupling correlation strength;Risk task group is identified based on intensity threshold value;Extract group task resource competition degree and path transmission efficiency, generate task chain transmission impedance factor by graph convolutional neural network;Through the processing impedance factor output collaborative allocation weight by gated recurrent unit model;According to weight integerization distribution AGV resource and generates scheduling instruction;Realize task conflict transmission modeling and resource collaborative allocation, significantly improve production line resource utilization and high urgency task response efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of flexible manufacturing system resource scheduling, and more particularly to a flexible production line scheduling method and system based on multi-AGV cooperation. BACKGROUND

[0002] In modern flexible manufacturing systems, multi-AGV cooperative scheduling technology is a key means to improve production line efficiency. The existing technology generally adopts a dynamic priority scheduling mechanism, that is, AGV resources are allocated in real time according to the urgency of tasks (such as equipment fault repair and single production instruction). This mechanism responds to sudden tasks through a central control system and realizes local task optimization according to preset rules (such as shortest path and time window constraints).

[0003] However, the existing technology has the following defects: when multiple high-priority tasks are triggered concurrently in the time and space dimensions, the dynamic priority mechanism will cause resource locking effect - the AGV cluster occupied by local high-priority tasks cannot be released due to path blockage or task delay, resulting in the stagnation of tasks in other areas due to resource exhaustion, which seriously restricts the overall response efficiency and robustness of the flexible production line. SUMMARY

[0004] In order to overcome the above defects of the prior art, the embodiments of the present application provide a flexible production line scheduling method and system based on multi-AGV cooperation to solve the problems raised in the background art.

[0005] To achieve the above purpose, the present application provides the following technical solutions:

[0006] A flexible production line scheduling method based on multi-AGV cooperation, comprising the following steps:

[0007] S1, obtaining the task priority of all tasks to be executed in the flexible production line;

[0008] S2, evaluating the task coupling correlation strength between any two high-priority tasks and outputting the task coupling evaluation result;

[0009] S3, identifying a risk task group based on the task coupling evaluation result, the risk task group being composed of tasks whose task coupling correlation strength exceeds a preset coupling threshold;

[0010] S4, extracting the resource competition degree and path transmission efficiency of each task in the risk task group, and generating a task chain transmission impedance factor through a graph convolutional neural network;

[0011] S5, inputting the task chain transmission impedance factor into a gated recurrent unit model to output an AGV cooperative allocation weight of the risk task group;

[0012] S6, reassigning execution sequences of all tasks in the risk task group according to the AGV coordination weight and generating AGV scheduling instructions.

[0013] In a preferred embodiment, the task priorities of all tasks to be executed in the flexible production line are obtained, including:

[0014] Real-time collection of task priority data of each task to be executed from the flexible production line control system;

[0015] Wherein the task priority data is dynamically generated by the flexible production line control system according to the urgency of the task triggering event;

[0016] The task triggering event includes equipment fault repair instruction and single insertion production instruction.

[0017] In a preferred embodiment, the task coupling correlation strength between any two high-priority tasks is evaluated and the task coupling evaluation result is output, including:

[0018] Obtain the execution path coordinate set of each high-priority task in the current layout diagram of the flexible production line;

[0019] Calculate the path overlap rate between the execution path coordinate sets of the two high-priority tasks;

[0020] Obtain the task execution time overlap rate of the two high-priority tasks within the scheduling time window;

[0021] Output the product of the path overlap rate and the task execution time overlap rate as the task coupling correlation strength value of the two high-priority tasks and output as the task coupling evaluation result.

[0022] In a preferred embodiment, the path overlap rate is equal to the proportion of the sum of the Euclidean distances of the path overlap segments to the total length of the two task paths; the task execution time overlap rate is equal to the proportion of the length of the overlapping period of the two task execution periods to the total length of the two tasks.

[0023] In a preferred embodiment, a risk task group is identified based on the task coupling evaluation result, and the risk task group is composed of tasks whose task coupling correlation strength exceeds a preset coupling threshold, including:

[0024] Obtain the task coupling correlation strength value of all pairs of high-priority tasks;

[0025] Compare the task coupling correlation strength value with the preset coupling threshold;

[0026] Screen high-priority task pairs whose task coupling correlation strength value exceeds the preset coupling threshold;

[0027] Construct a risk task group based on the topological connected relationship of the screened high-priority task pairs;

[0028] The output risk task group contains all high-priority task sets as the identification result.

[0029] In a preferred embodiment, the resource competition degree and path conduction efficiency of each task in the risk task group are extracted, and the task chain conduction impedance factor is generated by a graph convolutional neural network, including:

[0030] For each task in the risk task group, the real-time AGV demand value and the available value of the area where the task is located are obtained, and the resource competition degree is calculated as the ratio of the real-time AGV demand value to the available value;

[0031] For the same task, the historical path running log data of the task is obtained, and the path conduction efficiency is calculated as the inverse of the path smoothness period proportion value;

[0032] The resource competition degree value and the path conduction efficiency value are combined into a binary feature vector and associated with the task node corresponding to the risk task group;

[0033] Based on the resource dependency relationship data between tasks in the risk task group, a task resource dependency topology graph is constructed, in which the nodes represent tasks and the edges represent dependency relationships;

[0034] The task resource dependency topology graph is input into a graph convolutional neural network to perform a neighborhood feature aggregation operation, and the task chain conduction impedance factor value of each task node is output.

[0035] In a preferred embodiment, the task chain conduction impedance factor is input into a gated recurrent unit model, and the AGV collaborative allocation weight of the risk task group is output, including:

[0036] The task chain conduction impedance factor value sequence of all tasks in the risk task group is obtained;

[0037] The task chain conduction impedance factor value sequence is arranged in time sequence according to task dependency to form an impedance factor time sequence input vector;

[0038] The impedance factor time sequence input vector is input into a pre-trained gated recurrent unit model to perform time sequence association fusion operation;

[0039] The task weight parameter of the task in the risk task group is generated through the output layer of the gated recurrent unit model;

[0040] The task weight parameters of all tasks in the risk task group are normalized to obtain the AGV collaborative allocation weight value of the risk task group.

[0041] In a preferred embodiment, the execution order of all tasks in the risk task group is redistributed according to the AGV collaborative allocation weight, and an AGV scheduling instruction is generated, including:

[0042] Obtaining AGV collaborative allocation weight values corresponding to all tasks in the risk task group;

[0043] Arranging execution priority orders of all tasks in the risk task group in descending order according to the AGV collaborative allocation weight values;

[0044] Allocating AGV cluster resource quantities to each task according to the execution priority order;

[0045] Generating a task scheduling time sequence queue based on the execution priority order and the allocated AGV cluster resource quantities;

[0046] Converting the task scheduling time sequence queue into an executable scheduling instruction set of the AGV control system.

[0047] In a preferred embodiment, the task allocation AGV cluster resource quantity is equal to the task basic resource requirement quantity multiplied by an integerization coefficient of the AGV collaborative allocation weight value.

[0048] In another aspect, the present application provides a flexible production line scheduling system based on multi-AGV collaboration, comprising the following modules:

[0049] A data acquisition module for obtaining task priorities of all tasks to be executed in the flexible production line;

[0050] A conflict detection module for evaluating task coupling correlation strength between any two high-priority tasks and outputting a task coupling evaluation result;

[0051] A risk identification module for identifying a risk task group based on the task coupling evaluation result, the risk task group being composed of tasks with task coupling correlation strength exceeding a preset coupling threshold;

[0052] A resource modeling module for extracting resource competition degree and path conduction efficiency of each task in the risk task group, and generating a task chain conduction impedance factor through a graph convolutional neural network;

[0053] A collaborative decision module for inputting the task chain conduction impedance factor into a gated recurrent unit model and outputting an AGV collaborative allocation weight of the risk task group;

[0054] An instruction generation module for reallocating execution orders of all tasks in the risk task group according to the AGV collaborative allocation weight and generating AGV scheduling instructions.

[0055] Compared with the prior art, the present application has the following beneficial effects:

[0056] The time and space conflict conduction chain between high-priority tasks is accurately identified through a task coupling correlation strength evaluation mechanism, and the limitation of traditional dynamic priority scheduling focusing on single task optimization is broken. A risk task group is constructed based on topological connectivity, and the implicit resource competition effect caused by indirect dependence is effectively captured, thereby avoiding the AGV resource lock risk caused by multi-task concurrency from the source. The resource competition degree and path conduction efficiency are fused through double-factor graph convolution, the resource blocking diffusion characteristics of the task chain are quantified, and the resource allocation model has both spatial competition situation and time conduction characteristics. The evolution law of resource conflict between tasks is dynamically analyzed through the time sequence modeling of the task chain conduction impedance factor through the gating cycle unit, and the weight allocation decision is ensured to meet the causal correlation demand of the production line dynamic scheduling. Finally, through the integer coefficient resource allocation mechanism, the collaborative weight is converted into discrete AGV cluster scheduling instructions, and the resource collaborative optimization of the high-priority task group is realized. BRIEF DESCRIPTION OF DRAWINGS

[0057] Figure 1 A flowchart of a flexible production line scheduling method based on multi-AGV collaboration of the present application;

[0058] Figure 2 A structural schematic diagram of a flexible production line scheduling system based on multi-AGV collaboration of the present application. DETAILED DESCRIPTION

[0059] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0060] Embodiment 1: Figure 1 A flexible production line scheduling method based on multi-AGV collaboration of the present application is given, which comprises the following steps:

[0061] S1, obtaining the task priority of all tasks to be executed in the flexible production line;

[0062] S2, evaluating the task coupling correlation strength between any two high-priority tasks and outputting the task coupling evaluation result;

[0063] S3, identifying a risk task group based on the task coupling evaluation result, the risk task group being composed of tasks with a task coupling correlation strength exceeding a preset coupling threshold;

[0064] S4, extracting the resource competition degree and path conduction efficiency of each task in the risk task group, and generating a task chain conduction impedance factor through a graph convolutional neural network;

[0065] S5, input the task chain transmission impedance factor into the gated recurrent unit model, and output the AGV cooperative allocation weight of the risk task group;

[0066] S6, reassign the execution order of all tasks in the risk task group according to the AGV cooperative allocation weight and generate AGV scheduling instructions.

[0067] S1, obtain the task priority of all tasks to be executed in the flexible production line, which includes the following steps:

[0068] The central processing unit of the flexible production line control system monitors the trigger events associated with production tasks in real time. When a device fault repair signal or an insert order production instruction is detected, the central processing unit calls the built-in event type priority mapping rule: the device fault repair class event is fixedly assigned the highest priority level mark (for example, marked as the value 9), the insert order production class event is fixedly assigned the next highest priority level mark (for example, marked as the value 7), and the regular production task is dynamically assigned a priority level according to the remaining time of order delivery - the specific rule is: obtain the delivery deadline of the task in the order database, calculate the difference between the current time and the deadline (defined as the remaining time), and the remaining time falls into a preset time interval corresponding to a specific priority level, wherein the shorter the remaining time interval corresponds to the higher priority level (for example, the remaining time is less than or equal to 2 hours, level 6 is assigned, less than or equal to 4 hours, level 5 is assigned, and greater than 4 hours, level 4 is assigned). The priority calculation unit encapsulates the generated priority data into a data unit containing event type code, priority level value, and precise timestamp.

[0069] The control system transmits the data unit to the AGV scheduling server through the industrial real-time Ethernet protocol (for example, the PROFINET protocol). The scheduling server performs data integrity verification: first, it verifies the data unit integrity using the cyclic redundancy check algorithm, which uses an international standard to generate a polynomial; second, it checks the timestamp continuity, and if the interval between the current data unit timestamp and the previous unit timestamp exceeds the maximum allowed delay threshold of the system (for example, 500 milliseconds), it is determined as abnormal data. For data units that fail the verification, a retransmission mechanism is started: the first retransmission waiting time is an integer multiple of the system reference period (for example, if the reference period is 50 milliseconds, the retransmission waiting time is 100 milliseconds), and if it fails again, the waiting time is exponentially increased (for example, the second time is 200 milliseconds, and the third time is 400 milliseconds). When the number of retransmissions reaches the upper limit (for example, 3 times), a historical data recovery strategy is used: the priority level value of the same type of task in the recent valid period (for example, within the last 1 minute) is obtained from the historical database, and the arithmetic mean is calculated to replace the priority level value in the abnormal data unit.

[0070] For multi-task conflict scenarios, the system performs priority arbitration: when multiple equipment breakdown repair events occur simultaneously, sort according to the distance between the equipment physical location coordinates and the key workstations (preferential processing for distances less than a threshold value, for example 5 meters); when multiple order production events occur simultaneously, sort according to the order value per unit time indicator in the order management system (preferential processing for high order value per hour, for example). The arbitration result is generated by a special hardware comparator, which uses a two-level comparison architecture: the first level comparator processes the event type code, and the second level comparator processes the distance value or order value indicator value. The arbitration result is written to the special flag bit field of the data unit.

[0071] S2, evaluate the task coupling correlation strength between any two high-priority tasks and output the task coupling evaluation result, the specific implementation includes the following steps:

[0072] The path analysis of the flexible production line control system is performed by reading the equipment location coordinate model in the layout database. For each high-priority task, the coordinate point sequence of its preset movement path is extracted from the process database (for example, task P001 path contains 3 positioning points: starting point location coordinates, intermediate point location coordinates, and target point location coordinates). The path coincidence degree calculation engine performs: continuously compares the coordinate point sequences of the two tasks, and identifies continuous 3 or more reference points that are completely matched in position (for example, the 2nd reference point and the 3rd reference point of task P001 coincide with the 1st reference point and the 2nd reference point of task P002, respectively). For each coincident reference point pair, calculate the physical distance value: the value is the arithmetic square root of the sum of the square of the horizontal position difference and the square of the vertical position difference (for example: the horizontal difference of a pair of coincident points is 3 units, and the vertical difference is 2 units, the distance is about 3.61 units). Divide the total sum of the distances of all coincident points by the total length value of the two task paths (the total length value is the sum of the distances between each adjacent point of each task), to get the path overlap rate (for example: the total sum is 3.6 units, the total length of the two tasks is 14.4 units, the overlap rate is 25%). If no continuous 3 coincident points are found, the path overlap rate is set to 0.

[0073] The time series analysis of the scheduling system retrieves the execution time intervals of the two tasks in the job plan (for example: task P001 plan period 09:00-09:30, task P002 plan period 09:15-09:45). The time overlap degree calculator performs: determines the overlapping sub-interval of the two time intervals (in this example, the coincident period is 09:15-09:30), measures the duration of the sub-interval (15 minutes), and divides this duration by the total execution time of the two tasks (30 minutes for task P001 plus 30 minutes for task P002, equal to 60 minutes), to obtain the task execution time overlap rate (15 / 60=0.25). The calculation result is stored in the data buffer.

[0074] The numerical operation unit of the central processing unit calls the buffer data: extracts the path overlap rate value and the task execution time overlap rate value, performs a floating-point multiplication operation (for example: 0.25 x 0.25 = 0.0625), and the product is defined as the task coupling correlation strength value. The structured data object is constructed: the paired task number is written into the task pair identifier field, the product value is stored in the correlation strength value field, the current system time is recorded in the timestamp field, and the status flag bit is 0 by default. When the path overlap rate is 0 or the time overlap rate is 0, the product is 0, and the status flag bit is 1, indicating an abnormal correlation.

[0075] Hardware requirements: the geometry coprocessor needs to support the vector module length calculation instruction, and the system timing unit needs 1 millisecond precision. Abnormal data processing: when the path coordinates are missing, the average coordinate sequence of the last 3 historical paths of the task is used; when the planned time is not set, the planned interval is inferred according to the standard process time parameter of the work order.

[0076] S3, based on the task coupling evaluation result, a risk task group is identified, the risk task group is composed of tasks whose coupling correlation strength exceeds the preset coupling threshold, and the specific implementation includes the following steps:

[0077] The task coupling correlation strength value data column of all high-priority task pairs is read from the task coupling evaluation result database in pages (for example, the first record: the strength value of task pair P001-P002 is 0.0625). The system accesses the parameter configuration library to retrieve the preset coupling threshold (the threshold is dynamically set according to the historical running data of the production line: the high-priority task conflict frequency in the past 90 days is counted, and when the average conflict frequency per day exceeds 5 times, the threshold is set to 0.05, otherwise it is set to 0.1; the current example threshold is 0.05). The numerical comparison unit scans the task coupling correlation strength value line by line: if the task coupling correlation strength value is greater than the preset coupling threshold (for example, 0.0625 > 0.05), set the valid flag bit (binary value 1) in the task pair state register; if it is less than or equal to, set the invalid flag bit (0). After completing the full table scan, an effective task pair index table is generated (for example, record the effective task pair numbers P001-P002, P002-P003, etc.).

[0078] The topology construction engine loads the effective task pair index table to initialize the task node relationship graph: each unique task number generates an independent node, and the two nodes associated with each other in the effective task pair create a non-directional connection edge (for example, nodes P001-P002 establish an edge, and nodes P002-P003 establish an edge). The connected domain identification process performs a depth-first traversal rule: randomly select an unvisited node as the starting point, recursively visit its adjacent nodes and mark the visit state, until all nodes in the connected domain are traversed (for example, P001 visits adjacent node P002, and P002 visits P003 to form a node set {P001, P002, P003}); after traversal, an independent risk task group number (group ID: G001) is generated. The processing rule for isolated nodes: nodes without adjacent edges form a risk task group (for example, a single node P004 generates group ID: G002). When a ring topology structure is detected (for example, P001-P002-P003-P001 closed loop), it is automatically merged into the same risk task group (group ID: G003).

[0079] The risk task group data structure construction unit performs: assigns a hexadecimal format identifier to each group (for example, G001), and arranges the task number list in the group in ascending order of dictionary sequence (for example, P001, P002, P003). The output thread writes the risk task group data set to the shared memory buffer, and the data packet structure includes three fields: group identifier (fixed length of 4 bytes), task number value (integer), and task number array (variable length string). When the system detects a preset coupling threshold adjustment event (for example, the maintenance period threshold is temporarily adjusted to 0.1), a data snapshot is automatically created to prevent overwriting.

[0080] Hardware requirements: 500 nodes require allocation of 64 GB of memory, and the stack depth is set to an upper limit of 10,000 layers. Abnormal data processing mechanism: when the task coupling correlation strength value is missing, take the average of the last 3 historical values (for example, 08:00, 08:10, 08:20 values are averaged); if the threshold configuration table is damaged, restore the default value 0.05. Ring structure processing optimization: when the number of nodes in the ring exceeds 50, a hierarchical traversal strategy is enabled.

[0081] S4, extract the resource competition degree and path transmission efficiency of each task in the risk task group, and generate a task chain transmission impedance factor through a graph convolutional neural network, which includes the following steps:

[0082] For each task record in the risk task group (for example, the risk task group G001 contains tasks P001, P002, P003), the system calls the real-time resource monitoring interface of the AGV scheduling system to obtain the AGV demand value of the task allocation area at the current time (specifically, the total number of AGVs allocated to the area, for example, area Z001 requires 3 AGVs) and the real-time available AGV quantity value of the area (specifically, the number of idle AGVs detected by the area sensor in real time, for example, 2 AGVs). The resource competition degree calculation unit performs resource competition degree value calculation: divide the AGV demand value by the AGV available value to obtain a quotient (3 ÷ 2 = 1.5), which is taken as the resource competition degree value (task P001 resource competition degree 1.5). Each task in the same risk task group independently completes this calculation operation process (tasks P002, P003 perform the same operation process).

[0083] For the same task record as described above (continue to take task P001 as an example), the system queries the path operation history database to retrieve the full path state record of the task within the past 24 hours. The path state record includes a timestamp field and a state marker field (unblocked / blocking). Path unblocked period ratio calculation: count the total time of all unblocked record time (for example, the unblocked record is 540 minutes), divide by the total time of the monitoring period 1440 minutes (24 hours x 60 minutes), and obtain the unblocked period ratio value (540 ÷ 1440 = 0.375). The path conduction efficiency calculation unit takes the reciprocal of the ratio value: 1 divided by the unblocked period ratio value (1 ÷ 0.375 ≈ 2.667), and the calculation result is taken as the path conduction efficiency value (task P001 path conduction efficiency 2.667). Each task in the risk task group independently performs this path conduction efficiency calculation process.

[0084] Feature vector integration engine operation: for each task record that has completed the above two calculations, extract the calculated resource competition degree value and path conduction efficiency value (for example, task P001 resource competition degree 1.5 and path conduction efficiency 2.667), and construct a binary feature vector data composed of two values ([1.5, 2.667]). The system allocates data storage space for each task node of the risk task group in memory, and associates the binary feature vector data to the corresponding task node (task node P001 stores the vector [1.5, 2.667]).

[0085] The system accesses the task resource dependency relationship database to obtain the dependency relationship records between all tasks in the risk task group. Each dependency relationship record contains a dependent task number and a dependent task number field (for example, a record shows that task P001 execution needs to depend on the release of AGV resources by task P002). The task resource dependency topology graph construction performs the following operations: creating a graph structure with each task as an independent node (nodes P001, P002, P003), and creating a directed connection edge between nodes according to the dependency relationship record (for example, an edge from node P002 to node P001). Each node carries the aforementioned stored binary feature vector data. The graph structure data is converted into an adjacency matrix form for storage.

[0086] The topology processing unit loads the complete task resource dependency topology graph data into the graph convolutional neural network computing environment. The graph convolutional neural network is configured as a two-layer structure: the first layer performs a neighborhood feature aggregation operation: for each task node, the binary feature vector data of its first-order adjacent nodes is collected, a weighted sum operation is performed (the weighting coefficient is determined by the dependency strength coefficient of the connection edge, and the default value is 1.0), and the result is generated through an activation function to generate an intermediate feature vector; the second layer repeats this feature aggregation and transformation process, and the final output layer performs a linear regression operation to generate a task chain conduction impedance factor value for each node (for example, task P001 impedance factor 0.82). The system combines the task number of each task with the corresponding task chain conduction impedance factor value into an output data record and writes it to the shared memory output queue.

[0087] Implementation requirements: The path state record database requires a throughput of 1000 write operations per second; the graph neural network computing environment needs to be configured with a graphics processing unit accelerator card with a video memory capacity of not less than 16 GB; the memory system needs to support 64 GB of physical memory allocation. Abnormal data processing mechanism: when the AGV available quantity value is 0, the resource competition degree value is taken as the square value of the AGV demand quantity value; when the path state record is less than 12 hours, the unobstructed period proportion is calculated based on the corresponding data of the latest complete working day; when a circular dependency conflict is detected (for example, A depends on B, and B depends on A), the latest established dependency relationship is automatically removed. The resource competition degree value is a dimensionless ratio; the path conduction efficiency value is a dimensionless reciprocal; the task chain conduction impedance factor value is a dimensionless regression value.

[0088] S5, input the task chain conduction impedance factor into the gated recurrent unit model, output the AGV cooperative allocation weight of the risk task group, and the specific implementation includes the following steps:

[0089] The system reads the task chain conduction impedance factor value sequence of all tasks in the risk task group from the shared data buffer (for example, the risk task group G001 contains three tasks: the task chain conduction impedance factor value of task number P001 is 0.82, the task chain conduction impedance factor value of task number P002 is 1.15, and the task chain conduction impedance factor value of task number P003 is 0.73). The system accesses the task resource dependency database to query the dependency timing information between tasks in the risk task group (for example, the record shows that task P003 must be executed before task P002, and task P002 must be executed before task P001). The timing reconstruction engine reorders the task chain conduction impedance factor value sequence in the order of dependency: the task chain conduction impedance factor value of the task without pre-dependence is placed at the beginning (such as the task chain conduction impedance factor value of P003 is 0.73), then the task chain conduction impedance factor value of the task directly dependent on the former (such as the task chain conduction impedance factor value of P002 is 1.15), and finally the task chain conduction impedance factor value of the final dependent task (such as the task chain conduction impedance factor value of P001 is 0.82), forming an impedance factor timing input vector [0.73, 1.15, 0.82].

[0090] The sorted impedance factor timing input vector is input to the pre-trained gated recurrent unit model inference engine. Model initialization operation: set the hidden state vector to a 128-dimensional all-zero array, and set the model time step value equal to the input vector length value (in this example, the time step is 3). The gated recurrent unit model performs timing correlation fusion calculation: the first step processes the first element 0.73 of the input vector, and determines the proportion of historical information to be retained through reset gate calculation (reset gate weight parameters are loaded from model file), and determines the proportion of new current input through update gate calculation (update gate weight parameters are loaded); the second step processes the second element 1.15, and the hidden state vector output by the previous step is used as the basis data, and the reset gate and update gate calculation processes are repeated; the third step processes the last element 0.82 to complete the timing fusion operation. The final time step hidden state vector is input to the fully connected output layer, and the Sigmoid activation function is processed to generate the task weight parameters corresponding to each task (for example, the task weight parameter of task P001 is 0.36, the task weight parameter of task P002 is 0.45, and the task weight parameter of task P003 is 0.19).

[0091] The task weight parameters of all tasks in the risk task group are normalized: first, the sum of all task weight parameters is calculated (task P001 task weight parameter 0.36 plus task P002 task weight parameter 0.45 plus task P003 task weight parameter 0.19 equals 1.00), and then the task weight parameter of each task is divided by the sum value to obtain the normalized AGV collaborative allocation weight value (for example, task P001 AGV collaborative allocation weight value 0.36 ÷ 1.00 = 0.36). The system creates structured output data: the task number field stores the task identification (such as "P001"), the weight value field stores the AGV collaborative allocation weight value (such as 0.36), the normalization status flag is set to true, and a timestamp is generated to record the completion time. The final data is written to the AGV collaborative allocation weight result database table of the risk task group.

[0092] Gated recurrent unit model pre-training implementation: train the model using flexible production line historical scheduling data, and the training data set contains 50,000 groups of task chain conduction impedance factor time series vectors and corresponding optimal AGV collaborative allocation weight value labels (optimal weight value labels are determined by production line scheduling experts based on task urgency and resource occupation). The training process uses the Adam optimization algorithm, and the loss function is defined as the mean square error value between the predicted weight value and the optimal weight value label. Model structure configuration parameters: single-layer gated recurrent unit network structure, 128 hidden units; the number of nodes in the fully connected output layer is 1. Training termination condition: stop training when the validation set loss value decreases by less than 0.001 for 10 consecutive iterations.

[0093] Implementation environment hardware requirements: model inference requires a graphics processor accelerator card (single-precision floating-point computing power is not less than 50 trillion operations per second). Exception handling mechanism: when the task dependency time series information is missing, arrange the task chain conduction impedance factor value sequence in ascending order of task number character encoding; when the model output task weight parameter is negative or zero, take the absolute value and add a 0.01 offset; when the normalization denominator (the sum of task weight parameters) is zero or negative, assign equal weight values to each task (AGV collaborative allocation weight value of each task = 1 ÷ task total number). All values generated during the calculation process are dimensionless parameters (resource competition degree, path conduction efficiency, impedance factor, weight parameter, etc. are all dimensionless values).

[0094] S6, according to the AGV collaborative allocation weight, reassign the execution order of all tasks in the risk task group and generate AGV scheduling instructions, which includes the following steps:

[0095] The system reads the AGV collaborative allocation weight value data records corresponding to all tasks in the risk task group from the AGV collaborative allocation weight database page by page (for example, the risk task group G001 contains three tasks: the AGV collaborative allocation weight value of task number P001 is 0.36, the AGV collaborative allocation weight value of task number P002 is 0.45, and the AGV collaborative allocation weight value of task number P003 is 0.19). The scheduling engine initializes the priority queue data structure and arranges all tasks in the risk task group according to the descending order of AGV collaborative allocation weight value: first, select the task with the highest AGV collaborative allocation weight value (such as P002 weight 0.45 ranked first), then arrange the task with the second highest weight value (such as P001 weight 0.36 ranked second), and finally arrange the task with the lowest weight value (such as P003 weight 0.19 ranked third), forming the execution priority order list [P002, P001, P003]. During the sorting process, tasks with the same weight value are arranged in ascending order of task number letters.

[0096] Task resource allocation operation: for each task in the execution priority order list, the system accesses the task resource demand database to obtain the task basic resource demand quantity of the task (for example, the task P002 basic resource demand quantity is 2 AGVs). The integer coefficient dynamic generator determines the current coefficient value according to the total quantity of the production line AGV cluster: obtain the real-time AGV total number (for example, 80) through the AGV position sensor network, and select the integer coefficient 10 when the total number is in the interval of 50 to 100. The task allocation AGV cluster resource quantity calculation engine executes: first multiply the AGV collaborative allocation weight value by the integer coefficient (0.45×10=4.5), apply the ceiling function operation to the product to obtain the integer value (4.5→5), then multiply the task basic resource demand quantity by the integer value (2 AGVs×5=10 AGVs), and the result is the AGV cluster resource allocation quantity of the task. Each task in the risk task group is calculated according to the same rule.

[0097] The timing queue generator loads the execution priority order list and the AGV cluster resource allocation quantity result, and creates a task scheduling timing queue data structure: each element of the queue contains five fields: the order position field stores the index value of the task in the execution priority order list (such as P002 index value 1), the task number field stores the task identification (such as "P002"), the AGV cluster resource allocation quantity field stores the allocation result (such as 10), the planned start time field stores an ISO8601 format timestamp (such as "2024-07-05T10:00:00Z"), and the planned end time field stores a timestamp (such as "2024-07-05T10:30:00Z"). The planned start time of adjacent tasks is set: the planned start time of the first task is the current system time; the planned start time of the subsequent task is equal to the planned end time of the previous task plus the buffer time interval value (the buffer time interval is set to 4 minutes according to historical data statistics).

[0098] The scheduling instruction conversion converts the task scheduling timing queue into a set of executable scheduling instructions recognizable by the AGV control system: the instruction set is encapsulated in a JSON format array, and each instruction object contains eight fields: the command_id field stores the unique identifier of the instruction, the task_id field stores the task number, the priority field stores the execution priority order value (such as 1), the agv_count field stores the AGV cluster resource allocation quantity, the start_time field stores the planned start time, the end_time field stores the planned end time, the command_type field is fixed as "task scheduling instruction", and the status field is initialized as "to be executed". Instruction objects are arranged in the array in the order of the task scheduling timing queue (index 0 is the highest priority instruction).

[0099] The integer coefficient dynamic adjustment mechanism: the system monitors the total quantity of AGV clusters in real time and updates the coefficient value every hour: when the total quantity of AGV is less than 50, the integer coefficient is set to 5, when it is 50-100, it is set to 10, and when it is greater than 100, it is set to 15. Buffer time interval determination rule: statistical analysis of the task switching time records of the production line in the last 30 days, calculation of the 90th percentile value of the task switching time (such as 180 seconds), plus a safety redundancy time of 60 seconds to get the buffer time interval value (180+60=240 seconds). Task planning time length calculation: obtain the task standard processing time from the process database, multiply the task standard processing time by the inverse ratio coefficient of the AGV cluster resource allocation quantity and the task basic resource requirement quantity (such as task standard time 60 minutes, allocation quantity 10 ÷ requirement quantity 2 = 5 times, calculate actual time 60 ÷ 5 = 12 minutes).

[0100] Implementation environment requirements: the scheduling server needs to be equipped with a 32-core processor and 64 GB of memory, and the AGV position sensor network needs to support a 5 Hz update frequency. Abnormal processing mechanism: when the number of AGV cluster resource allocation exceeds the upper limit of the regional AGV capacity, the system reduces the number of task allocation according to the descending order of execution priority (decreases by 1 each time) until the capacity limit is met; when the time window conflict occurs in the task scheduling time queue (such as the planned start time being earlier than the end time of the previous task), the start time of the low-priority task is automatically delayed and the buffer time interval is maintained; when the integer coefficient calculation result causes the allocation number to be zero, 1 AGV is forced to be allocated and an alarm log is triggered.

[0101] The task coupling correlation strength evaluation mechanism (S2) of the embodiment is different from the limitation of the traditional dynamic priority scheduling which only considers single task attributes, and reveals the resource conflict conduction effect between high-priority tasks in the time and space dimensions. The risk task group is constructed based on the topological connectivity relationship (S3), which breaks through the simple threshold filtering rule and accurately captures the implicit conflict chain caused by indirect coupling. The double-factor graph convolution fusion of resource competition degree and path conduction efficiency (S4) quantifies the resource blocking diffusion characteristics of the task chain, solves the modeling deviation problem of a single resource competition index. The gated recurrent unit time sequence modeling of the task chain conduction impedance factor (S5) effectively analyzes the dynamic evolution law of the priority conflict, realizes the causal correlation decision of the weight allocation. The integer coefficient resource allocation rule of AGV cooperative allocation weight (S6) avoids the resource fragmentation defect of the conventional weight allocation. The above-mentioned links form a technical closed loop of "conflict detection-dynamic quantization-collaborative decision".

[0102] Embodiment 2: Figure 2 A structure diagram of a flexible production line scheduling system based on multi-AGV cooperation is given, and the flexible production line scheduling system based on multi-AGV cooperation comprises the following modules:

[0103] A data acquisition module is configured to acquire the task priority of all to-be-executed tasks in the flexible production line.

[0104] A conflict detection module is configured to evaluate the task coupling correlation strength between any two high-priority tasks and output a task coupling evaluation result.

[0105] A risk identification module is configured to identify a risk task group based on the task coupling evaluation result, and the risk task group is composed of tasks whose task coupling correlation strength exceeds a preset coupling threshold.

[0106] A resource modeling module is configured to extract the resource competition degree and path conduction efficiency of each task in the risk task group, and generate a task chain conduction impedance factor through a graph convolution neural network.

[0107] The cooperative decision module is configured to input the task chain transmission impedance factor into a gated recurrent unit model, and output an AGV cooperative allocation weight of the risk task group.

[0108] The instruction generation module is configured to reassign execution sequences of all tasks in the risk task group according to the AGV cooperative allocation weight, and generate an AGV scheduling instruction.

[0109] In the embodiments, all calculations are dimensionless numerical calculations, and preset parameters and threshold values in the calculations are set by a person skilled in the art according to actual conditions.

[0110] The above embodiments can be realized wholly or partially by software, hardware, firmware, or any combination thereof. When realized by software, the above embodiments can be realized wholly or partially in the form of a computer program product.

[0111] Those skilled in the art can appreciate that the modules and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized by hardware or software depends on the specific application and the constraints of the technical solution. A person 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.

[0112] In addition, each functional module in each embodiment of the present application can be integrated in one processing module, or each module can exist physically independently, or two or more modules can be integrated in one module.

[0113] In the several embodiments provided by the present application, it should be understood that the disclosed system, device and method can be implemented in other ways. For example, the device embodiments described above are only schematic. The division of the modules is only a logical function division. There can be another division manner in actual implementation, for example, a plurality of modules or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the modules shown or discussed can be indirect coupling or communication connection through some interface, device or module, and can be electrical, mechanical or other forms.

[0114] The above describes only the specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0115] Finally: the above only for the preferred embodiments of the present application, and not for limiting the present application, any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application, should be included in the scope of protection of the present application.

Claims

1. A flexible production line scheduling method based on multi-AGV cooperation, characterized in that, The method comprises the following steps: S1, obtaining the task priority of all tasks to be executed in the flexible production line; S2, evaluating the task coupling correlation strength between any two high-priority tasks and outputting the task coupling evaluation result, including: obtaining the execution path coordinate set of each high-priority task in the current layout diagram of the flexible production line; calculating the path overlap rate between the execution path coordinate sets of the two high-priority tasks; obtaining the task execution time overlap rate of the two high-priority tasks within the scheduling time window; outputting the product of the path overlap rate and the task execution time overlap rate as the task coupling correlation strength value of the two high-priority tasks and outputting it as the task coupling evaluation result; S3, identifying a risk task group based on the task coupling evaluation result, the risk task group being composed of tasks whose task coupling correlation strength exceeds a preset coupling threshold; S4, extracting the resource competition degree and path transmission efficiency of each task in the risk task group, and generating a task chain transmission impedance factor through a graph convolutional neural network, including: for each task in the risk task group, obtaining the real-time AGV demand value and available value of the region where the task is located, and calculating the resource competition degree as the ratio of the real-time AGV demand value to the available value; for the same task, obtaining the historical path running log data of the task, and calculating the path transmission efficiency as the inverse of the path smoothness period proportion value; combining the resource competition degree value and the path transmission efficiency value into a binary feature vector, and associating it to the task node corresponding to the risk task group; based on the resource dependency relationship data between the tasks in the risk task group, constructing a task resource dependency topology graph with nodes representing tasks and edges representing dependency relationships; inputting the task resource dependency topology graph into the graph convolutional neural network to perform neighborhood feature aggregation operation, and outputting the task chain transmission impedance factor value of each task node; S5, inputting the task chain transmission impedance factor into a gated recurrent unit model to output the AGV collaborative allocation weight of the risk task group, including: obtaining the sequence of task chain transmission impedance factor values of all tasks in the risk task group; arranging the sequence of task chain transmission impedance factor values into an impedance factor time sequence input vector according to the task dependency time sequence; inputting the impedance factor time sequence input vector into the pre-trained gated recurrent unit model to perform time sequence correlation fusion operation; generating the task weight parameters of the tasks in the risk task group through the output layer of the gated recurrent unit model; normalizing the task weight parameters of all tasks in the risk task group to obtain the AGV collaborative allocation weight value of the risk task group; S6, redistributing the execution order of all tasks in the risk task group according to the AGV collaborative allocation weight and generating AGV scheduling instructions. 2.The method of claim 1, wherein, obtaining the task priority of all tasks to be executed in the flexible production line, including: collecting the task priority data of each task to be executed from the flexible production line control system in real time; wherein the task priority data is dynamically generated by the flexible production line control system according to the urgency of the task triggering event; the task triggering event includes equipment fault repair instructions and single insertion production instructions.

3. The method according to claim 1, characterized in that, The path overlap rate is equal to the proportion of the sum of the Euclidean distances of the path overlap segments to the total length of the two task paths; and the task execution time overlap rate is equal to the proportion of the length of the overlap of the two task execution periods to the total length of the two tasks.

4. The method according to claim 1, characterized in that, The risk task group is identified based on the task coupling evaluation result, and the risk task group is composed of tasks with a task coupling correlation strength exceeding a preset coupling threshold, and includes: Obtaining the task coupling correlation strength values of all high-priority task pairs; Comparing the task coupling correlation strength values with the preset coupling threshold; Screening high-priority task pairs with a task coupling correlation strength value exceeding the preset coupling threshold; Based on the topological connection relationship of the screened high-priority task pairs, a risk task group is constructed; Outputting all high-priority task sets contained in the risk task group as the identification result.

5. The method according to claim 1, wherein, The execution order of all tasks in the risk task group is redistributed according to the AGV collaborative allocation weight, and AGV scheduling instructions are generated, including: Obtaining the AGV collaborative allocation weight values corresponding to all tasks in the risk task group; Arranging the execution priority order of all tasks in the risk task group in descending order according to the AGV collaborative allocation weight values; Allocating AGV cluster resource quantities to each task according to the execution priority order; Based on the execution priority order and the allocated AGV cluster resource quantities, a task scheduling time sequence queue is generated; Converting the task scheduling time sequence queue into an executable scheduling instruction set of the AGV control system.

6. The method according to claim 5, wherein, The task allocation AGV cluster resource quantity is equal to the integer coefficient of the product of the task basic resource requirement quantity and the AGV collaborative allocation weight value.

7. A flexible production line scheduling system based on multi-AGV cooperation, used to implement the flexible production line scheduling method based on multi-AGV cooperation according to any one of claims 1-6, characterized in that, The system includes the following modules: A data acquisition module for obtaining the task priority of all tasks to be executed in the flexible production line; A conflict detection module for evaluating the task coupling correlation strength between any two high-priority tasks and outputting a task coupling evaluation result; A risk identification module for identifying a risk task group based on the task coupling evaluation result, the risk task group being composed of tasks with a task coupling correlation strength exceeding a preset coupling threshold; A resource modeling module for extracting the resource competition degree and path transmission efficiency of each task in the risk task group, and generating a task chain transmission impedance factor through a graph convolutional neural network; A collaborative decision-making module for inputting the task chain transmission impedance factor into a gated recurrent unit model and outputting an AGV collaborative allocation weight of the risk task group; An instruction generation module for redistributing the execution order of all tasks in the risk task group according to the AGV collaborative allocation weight and generating AGV scheduling instructions.

Images