Scheduling method and system for AGV cluster

Abstract

The invention relates to the technical field of Internet of Things industrial control. The invention relates to an AGV cluster scheduling method and system, and more specifically relates to a scheduling method and system for an AGV cluster. The method comprises the steps of calculating a production rhythm disturbance coefficient; carrying out weighted calculation on the production rhythm disturbance coefficient and the comprehensive execution adaptability of the AGV to obtain a distribution utility value; constructing a task allocation model with the goal of maximizing the sum of all possible allocation utility values, and solving the model to obtain an optimal AGV-task allocation pair; for the determined AGV-task allocation pairs, according to the sequence of the production rhythm disturbance coefficients of the tasks from high to low, driving paths are planned for the AGVs in sequence; wherein an improved A * algorithm is adopted, and when the passing cost of the path segment is calculated, a time-space conflict probability cost is superposed; and issuing the driving path to a corresponding AGV for execution. According to the invention, the success rate of path planning and the passing efficiency of the AGV are improved, and the operation of the whole logistics system is smoother and more efficient.

Description

Technical Field

[0001] The present invention relates to the field of Internet of Things industrial control technology. More specifically, the present invention relates to a scheduling method and system for AGV clusters. Background Art

[0002] In the fields of smart manufacturing and intelligent logistics, automated guided vehicle (AGV) swarm systems have been deployed and applied as a core technology for automated and flexible material handling. The performance of the AGV swarm scheduling system directly determines the efficiency and stability of the entire material handling system. Traditional AGV swarm scheduling methods typically focus on two key steps: task allocation and path planning. During the task allocation phase, existing technologies mostly employ relatively simple strategies, such as a first-come, first-served principle, which assigns tasks in the order they are generated, or a proximity principle, which assigns tasks to the idle AGV closest to the task's starting point. Some systems employ optimization models that aim to minimize the total travel distance or total waiting time of all AGVs. However, these approaches often treat all tasks as equally important and fail to fully consider the actual cycle time and dynamic demands of the production system. When faced with multiple concurrent tasks, they are unable to effectively identify and prioritize those most critical tasks that have the greatest impact on production continuity at downstream workstations. This can easily lead to downtime at critical workstations due to material shortages, disrupting the overall production rhythm.

[0003] In terms of path planning, the classic A* algorithm and its variants are the most widely used single-unit path planning methods. They are effective in finding the shortest path in static environments. However, in the dynamic environment of multiple AGVs working together, these algorithms' limitations become apparent. When planning a path for a single AGV, they typically fail to consider the real-time positions and future trajectories of other AGVs. This can easily lead to spatial and temporal conflicts in the planned paths, causing traffic congestion, deadlocks, and even collision risks. To address this issue, some improved methods have introduced time-window-based reservation mechanisms or dynamic local obstacle avoidance strategies. However, the former often sacrifices path optimality and is computationally complex, while the latter can cause AGVs to frequently slow down, wait, or take detours, reducing traffic efficiency. Furthermore, existing technologies have limited consideration of AGV status when deciding between tasks and AGVs. They typically focus only on basic information such as battery level and location, while ignoring the AGV's historical performance and potential congestion encountered during specific tasks. This makes allocation decisions lack foresight and comprehensiveness, making it difficult to maximize the overall efficiency of the AGV cluster. Summary of the Invention

[0004] In order to solve the above technical problems, the present invention provides the following solutions.

[0005] A scheduling method for an AGV cluster includes the following steps: Obtaining real-time status data of each AGV in the AGV cluster and a set of tasks to be scheduled, wherein the real-time status data includes the position, power and load status of the AGV; For each task in the task set, a production rhythm disturbance coefficient is calculated to represent the urgency of the task based on its timeliness requirements, the material cache status of the associated downstream workstation, and the future material consumption rate predicted based on historical data; For each combination of a pending task and an idle AGV, a weighted calculation is performed on the task's production cycle disturbance coefficient and the AGV's comprehensive execution suitability to obtain an allocation utility value. The comprehensive execution suitability is based on the AGV's current battery level, its historical task execution score, and the estimated congestion probability of the AGV's path for executing the task, calculated using a preset spatiotemporal gridded occupancy prediction model. Construct a task allocation model with the goal of maximizing the sum of all possible allocation utility values, and solve the model to obtain the optimal AGV-task allocation pair; For each AGV-task pair, the AGVs are routed according to the production cycle disturbance coefficients of their tasks, from high to low. This approach employs an improved A* algorithm, whose cost function incorporates a spatiotemporal conflict probability cost calculated based on the spatiotemporal trajectories of other AGVs that have already completed path planning, when calculating the cost of each path segment. The driving path is sent to the corresponding AGV for execution.

[0006] By introducing a production rhythm disturbance coefficient that integrates timeliness, downstream workstation material status, and future consumption forecasts, the risk of critical workstation downtime due to untimely material delivery is avoided. In task allocation decisions, an allocation utility value that integrates task urgency and the AGV's comprehensive execution suitability is constructed to ensure optimal AGV-task matching and significantly improve the cluster's overall operational efficiency. In path planning, an improved A* algorithm that prioritizes planning by task urgency and incorporates a probability cost for spatiotemporal conflicts is used to proactively avoid potential spatiotemporal conflicts between AGVs during the planning phase.

[0007] Furthermore, the step of calculating the production rhythm disturbance coefficient representing the urgency of the task includes: The task timeliness, downstream workstation material cache status, and future material consumption rate are normalized and weighted to obtain the production rhythm disturbance coefficient P, which is calculated as follows: P=w1×[(current time - task creation time) / (task completion deadline - task creation time)]+w2×[1-(current cache size / maximum cache size of downstream workstations)]+w3×(material consumption rate within a preset future time window predicted based on historical production data / designed peak consumption rate of the production line); where w1, w2, and w3 are preset weight coefficients, and w1+w2+w3=1.

[0008] The production rhythm disturbance coefficient P is quantified by weighting, and the task timeliness, downstream workstation material cache status and future material consumption rate are reasonably integrated.

[0009] Furthermore, the step of generating the comprehensive execution suitability includes: The current power of the AGV, the historical task execution score, and the estimated path congestion probability are normalized and weighted to obtain the comprehensive execution suitability F, which is calculated as follows: F = c1 × (AGV current power / AGV full power) + c2 × (AGV historical task average execution score / full score) + c3 × (1-estimated path congestion probability); Among them, c1, c2, and c3 are preset weight coefficients; and when the current power of the AGV is lower than the sum of the estimated power consumption for executing the task and a preset safety power, the comprehensive execution suitability F of the AGV for the task is set to 0.

[0010] The comprehensive execution suitability is calculated in a weighted manner, taking into full consideration the AGV's current power level, historical task execution score, and estimated path congestion probability.

[0011] Furthermore, the construction of a task allocation model with the goal of maximizing the sum of allocation utility values ​​of all possible allocations includes: Construct a utility matrix with the tasks to be assigned as rows and the idle AGVs as columns. The elements in the matrix are the allocation utility values ​​of the corresponding AGV-task combinations. The Hungarian algorithm or KM algorithm is used to solve the utility matrix to obtain the optimal matching solution that maximizes the total utility value. This solution is the optimal AGV-task assignment pair.

[0012] Furthermore, the cost function f(n) used by the improved A* algorithm is: f(n)=g(n)+C(n,t)+h(n); in: g(n) is the actual physical path length cost from the path starting point to the current node n; C(n,t) is the probability cost of spatiotemporal conflict, and its value is positively correlated with the situation where the preset spatiotemporal neighborhood around the AGV is occupied by the planned paths of other AGVs when the AGV passes through the node n at the estimated time t; h(n) is the estimated cost from the current node n to the target node, which is obtained by calculating the Manhattan distance or Euclidean distance between node n and the target node.

[0013] The key to the above scheme is to increase the probability cost of spatiotemporal conflict, which reflects the situation that when an AGV passes through node n at the estimated time t, the preset spatiotemporal neighborhood around it is occupied by the planned paths of other AGVs.

[0014] Furthermore, the space-time conflict probability cost C(n,t) is calculated as follows: Divide the map environment into grids of preset size and divide time into time slices of preset length; Count the number of grid-time slices N occupied by other AGVs' planned paths within the preset neighborhood centered on node n within the time slice at time t when the AGV is expected to arrive at node n; C(n,t)=α×N, where α is the preset conflict cost coefficient.

[0015] By dividing the map environment into grids, conditions are created for the calculation of the spatiotemporal conflict probability cost C(n,t), reducing the computational difficulty.

[0016] Furthermore, the material consumption rate is output through a time series prediction model trained on historical production data based on a long short-term memory network (LSTM).

[0017] Furthermore, the generation of comprehensive execution suitability also includes a safety constraint: if the current power of the AGV is less than the estimated power consumption of the task + the safety redundancy power, then F=0.

[0018] Furthermore, the grid is 1 meter × 1 meter, and the time slice is 1 second.

[0019] A scheduling system for an AGV cluster includes a processor and a memory, wherein the memory stores a computer program and the processor executes the computer program to implement the above-mentioned scheduling method for the AGV cluster.

[0020] Compared with existing technologies, the present invention offers the following advantages: by introducing a production cycle disturbance coefficient that integrates timeliness, downstream workstation material status, and future consumption forecasts, the scheduling system can accurately identify and prioritize critical tasks that have the greatest impact on production line continuity, effectively ensuring stable production cycles and avoiding the risk of downtime at critical workstations due to untimely material delivery. In task allocation decisions, a utility value is constructed that integrates task urgency with the AGV's comprehensive execution suitability (including power consumption, historical efficiency, and estimated congestion probability), and is solved with the goal of maximizing global utility. This achieves a more scientific and forward-looking allocation of AGV resources, ensures optimal AGV-task matching, and significantly improves the overall operational efficiency of the cluster. In path planning, an improved A* algorithm, which prioritizes planning based on task urgency and incorporates spatiotemporal conflict probability costs, proactively avoids potential spatiotemporal conflicts between AGVs during the planning phase, reducing congestion and deadlock during operation, improving the success rate of path planning and the efficiency of AGV traffic, and ensuring smoother and more efficient operation of the entire logistics system. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 is a flowchart schematically illustrating a scheduling method for an AGV cluster according to an embodiment of the present invention; Figure 2 is a schematic diagram schematically illustrating a rasterized map environment according to an embodiment of the present invention; Figure 3 FIG. 1 is a schematic diagram schematically illustrating a scheduling system for an AGV cluster according to an embodiment of the present invention. DETAILED DESCRIPTION

[0022] like Figure 1 As shown, a scheduling method for an AGV cluster includes the following steps: S1, obtaining the real-time status data of each AGV in the AGV cluster and the task set to be scheduled, wherein the real-time status data includes the position, power and load status of the AGV.

[0023] Specifically, through a wireless communication network, such as a 5G network, the scheduling server periodically obtains the real-time position coordinates of the AGV on-board controller, the remaining power percentage reported by the battery management system BMS, and the empty or full load status detected by the load sensor, and obtains a list of pending material handling tasks, which includes the task number, starting station, target station, and material demand time.

[0024] S2, for each task in the task set, based on its timeliness requirements, the material cache status of the associated downstream workstations, and the future material consumption rate predicted based on historical data, calculate the production rhythm disturbance coefficient that represents the urgency of the task.

[0025] Specifically, the production cycle disturbance coefficient is calculated by the weighted sum of the task timeliness factor, the workstation buffer emergency factor, and the future consumption rate factor. The task timeliness factor is calculated based on the difference between the task's required completion time and the current time; the workstation buffer emergency factor is calculated based on the difference between the real-time material quantity at the downstream workstation and the safety stock threshold; and the future material consumption rate is output using a time series prediction model trained on historical production data using a long short-term memory (LSTM) network.

[0026] In one embodiment, the task timeliness, the downstream workstation material buffer status, and the future material consumption rate are normalized and weighted to obtain the production rhythm disturbance coefficient P, which is calculated as follows: P=w1×[(current time - task creation time) / (task completion deadline - task creation time)]+w2×[1-(current cache size / maximum cache size of downstream workstations)]+w3×(material consumption rate within a preset future time window predicted based on historical production data / designed peak consumption rate of the production line); where w1, w2, and w3 are preset weight coefficients, and w1+w2+w3=1.

[0027] Specifically, the production cycle disturbance coefficient (P) integrates three core dimensions that influence production continuity. The first component reflects the time urgency of the task through time ratios. For example, if a handling task is created at 9:00 AM and required to be completed by 9:30 AM, and the current time is 9:20 AM, the calculated value is (20 minutes / 30 minutes), which is approximately 0.67. As time passes, the closer the value approaches 1, indicating a more urgent task. The second component reflects the urgency of material demand at downstream workstations. Suppose a downstream workstation has a maximum buffer capacity of 100 pieces, and only 20 pieces are currently available. The calculated value is 1 minus (20 / 100), or 0.8. This high value indicates that the workstation is about to shut down due to material shortages and has a very urgent demand. The third component proactively considers future consumption. If historical data predicts that the workstation will consume 30 pieces of material within the next 15 minutes, and the production line is designed for a peak consumption rate of 40 pieces per 15 minutes, the value is (30 / 40), or 0.75, indicating significant future consumption pressure. By setting weights such as w1 = 0.4, w2 = 0.4, and w3 = 0.2, and summing these three factors, we can obtain a comprehensive production rhythm disturbance coefficient. A higher value indicates that the task is more important to maintaining smooth production line operation and should be given a higher execution priority, thus enabling accurate quantification and proactive response to production rhythm disturbances.

[0028] S3, for each combination of a task to be assigned and each idle AGV, the production rhythm disturbance coefficient of the task and the comprehensive execution suitability of the AGV are weightedly calculated to obtain an allocation utility value; wherein, the comprehensive execution suitability is generated based on the current power of the AGV, the historical task execution score, and the estimated path congestion probability of the AGV performing this task calculated using a preset spatiotemporal rasterization occupancy prediction model.

[0029] Specifically, the assigned utility value is a linear weighted sum of the task's production cycle disturbance coefficient and the AGV's overall execution suitability. The overall execution suitability itself is composed of three normalized weighted scores: the current AGV battery level, which drops sharply when the battery level falls below a preset threshold; the historical task execution score, calculated using an exponentially weighted moving average; and the estimated path congestion probability. This probability is calculated by using a pre-trained Convolutional Long Short-Term Memory (ConvLSTM) model to grid the factory map in time and space, predicting the probability of each grid being occupied over a period of time. This probability is then accumulated along the estimated path from the AGV to the task's starting point.

[0030] In one embodiment, the current power of the AGV, the historical task execution score, and the estimated path congestion probability are normalized and weighted to obtain the comprehensive execution suitability F, which is calculated as follows: F=c1×(AGV current power / AGV full power)+c2×(the AGV's average historical task execution score / full score)+c3×(1-estimated path congestion probability); where c1, c2, and c3 are preset weight coefficients; and when the AGV's current power is lower than the sum of the estimated power consumption for executing the task and a preset safety power, the AGV's comprehensive execution suitability F for the task is set to 0.

[0031] Specifically, the comprehensive execution suitability F is designed to comprehensively assess whether an AGV is the best choice for a specific task. This calculation integrates the AGV's own status, historical performance, and external environmental influences. For example, if an AGV has a full battery capacity of 100 units and is currently at 80 units, its battery score is 0.8, reflecting its endurance. Furthermore, if the AGV's average score for past tasks is 95 out of 100, its historical performance score is 0.95, representing its reliability and efficiency.

[0032] Furthermore, based on real-time traffic data, the system predicts a 10% probability of congestion on the AGV's path for the current mission, or 0.1. Therefore, the path smoothness score is 1 minus 0.1, or 0.9. Assuming the weight coefficients c1, c2, and c3 are set to 0.5, 0.3, and 0.2, respectively, the AGV's overall suitability for execution, F, is 0.5 times 0.8 plus 0.3 times 0.95 plus 0.2 times 0.9, for a total score of 0.865. A higher value indicates a more suitable AGV for the mission.

[0033] In one embodiment, this step may also include a mandatory safety constraint. For example, a task is estimated to consume 30 units of electricity, and the system sets a safety redundancy of 15 units, then the minimum power requirement for executing the task is 45 units. If at this time another AGV has excellent indicators, but its current power is only 40 units, which is lower than the threshold of 45 units, then the system will directly set its suitability F for this task to 0, thereby excluding it from the candidate list to prevent the AGV from shutting down due to power exhaustion during the task, thereby ensuring the stable operation of the entire logistics system. The above safety constraints can be expressed as follows: If the current power of the AGV ≥ the estimated power consumption of the task + the safety redundancy power, then F is calculated according to the above weighted method; if the current power of the AGV < the estimated power consumption of the task + the safety redundancy power, then F = 0.

[0034] S4, construct a task allocation model with the goal of maximizing the sum of the allocation utility values ​​of all possible allocations, and solve the model to obtain the optimal AGV-task allocation pair.

[0035] Specifically, the task allocation model is constructed as an assignment problem, where all idle AGVs form one side and all pending tasks form the other. The weights of the edges connecting AGVs and tasks are the allocation utilities calculated in the previous steps. The goal of the model is to find an optimal match that maximizes the sum of the utilities of the selected AGVs and tasks. This optimization problem is solved using, for example, the Kuhn-Monquets algorithm (KM algorithm), resulting in a globally optimal allocation solution.

[0036] In one embodiment, a utility matrix can be constructed with the tasks to be assigned as rows and the idle AGVs as columns. The elements in the matrix are the allocation utility values ​​of the corresponding AGV-task combinations. The Hungarian algorithm or the KM algorithm is used to solve the utility matrix to obtain the optimal matching solution that maximizes the total utility value. This solution is the optimal AGV-task assignment pair.

[0037] Specifically, the allocation utility value (hereinafter referred to as utility) of each idle AGV for each pending task is calculated. This value is typically obtained by multiplying the task's production cycle disturbance coefficient by the AGV's overall execution suitability. For example, given Tasks 1 and 2, and idle AGVs A and B, the calculated utility values ​​are as follows: AGV A's utility for Task 1 is 0.9, and its utility for Task 2 is 0.4; AGV B's utility for Task 1 is 0.6, and its utility for Task 2 is 0.8.

[0038] Based on this data, the system constructs a two-dimensional utility matrix, where rows represent tasks and columns represent idle AGVs. An example is shown in the following table:

[0039] In this example, the first row of the matrix is ​​0.9 and 0.4, and the second row is 0.6 and 0.8. This matrix clearly shows all possible allocation choices and their corresponding utilities. The system then calls an efficient matching algorithm, such as the Hungarian algorithm, to process this matrix. The algorithm's goal is to find a one-to-one matching method that maximizes the sum of the utility values ​​of all selected matches. In this example, there are two matching schemes. Scheme 1 is to match AGV A to Task 1 and AGV B to Task 2, with a total utility of 0.9 plus 0.8, which equals 1.7; Scheme 2 is to match AGV A to Task 2 and AGV B to Task 1, with a total utility of 0.4 plus 0.6, which equals 1.0. The algorithm quickly determines that Scheme 1 is the optimal solution and generates the final scheduling instructions, which dispatch AGV A to perform Task 1 and AGV B to perform Task 2.

[0040] S5, for the determined AGV-task assignment pairs, plan the driving paths for the AGVs in descending order of the production rhythm disturbance coefficients of their tasks; an improved A* algorithm is used, in which the cost function superimposes a spatiotemporal conflict probability cost calculated based on the spatiotemporal trajectories of other AGVs that have completed path planning when calculating the passage cost of the path segment.

[0041] Specifically, the assigned tasks are first sorted from high to low according to the production rhythm disturbance coefficient value. Then, the scheduling system calculates the path for the AGV corresponding to each task in turn. In the improved A* algorithm, the spatiotemporal conflict probability cost is introduced. The spatiotemporal conflict probability cost is determined by querying a dynamically updated path reservation table, which records the map grids that all AGVs with planned paths will occupy at each future time step. When the next path segment explored by the A* algorithm overlaps with an existing reservation in the table within the corresponding time window, a high penalty cost will be included in g(n).

[0042] In an optional embodiment, the cost function f(n) used by the improved A* algorithm is: f(n)=g(n)+C(n,t)+h(n); where: g(n) is the actual physical path length cost from the path starting point to the current node n; C(n,t) is the probability cost of spatiotemporal conflict, and its value is positively correlated with the situation where the preset spatiotemporal neighborhood around the AGV is occupied by the planned paths of other AGVs when the AGV passes through the node n at the estimated time t; h(n) is the estimated cost from the current node n to the target node, which is obtained by calculating the Manhattan distance or Euclidean distance between node n and the target node.

[0043] Specifically, this cost function is an optimization of the traditional A* algorithm, aiming to plan an efficient path that is not only short in distance but also has few conflicts.

[0044] Here, g(n) is the actual cost incurred. For example, if the AGV moves from starting point A to the current location node n, and the map shows that it has traveled 15 meters, then the value of g(n) is 15. This is a certain backtracking cost that has already occurred.

[0045] h(n) is an estimate of future costs, or a heuristic. If the coordinates of the current node n are (20, 30) and the coordinates of the target destination B are (50, 70), using the Manhattan distance, the value of h(n) is (50 minus 20) plus (70 minus 30), which equals 70 meters. This estimate guides the algorithm to prioritize exploring nodes toward the destination. The key improvement lies in the introduction of C(n, t), the space-time conflict cost. This cost term provides path planning with predictive and avoidance capabilities.

[0046] For example, when the algorithm evaluates whether to take node n as the next step on a path, it not only considers distance but also checks the time t at which the AGV is expected to arrive at node n, and whether the surrounding area has already been reserved by other AGVs at a similar time. Assuming that at the estimated arrival time, an intersection near node n is reserved by another AGV, the system assigns a positive value, such as 5, to C(n, t) based on the severity of the conflict. For node m on another alternative path, although the physical distance may be slightly farther, there are no other AGV activities nearby during the corresponding time period, and its C(m, t) value is 0. Therefore, when selecting the next node, the improved algorithm may abandon the physically closer node n due to the high conflict cost and instead choose the safer node m, thereby proactively avoiding future traffic congestion and potential collisions.

[0047] In an optional embodiment, the temporal-spatial conflict probability cost C(n,t) is calculated as follows: Divide the map environment into grids of preset size and divide time into time slices of preset length; Count the number of grid-time slices N occupied by other AGVs' planned paths within the preset neighborhood centered on node n within the time slice at time t where the AGV is expected to arrive at node n; C(n,t)=α×N, where α is the preset conflict cost coefficient.

[0048] Specifically, the entire factory map is discretized into a 1-meter by 1-meter square grid, for example, and time is divided into time slices, for example, one second each. This allows the precise location of any AGV at any given moment to be described by a grid-time slice combination. When planning a path for an AGV, the system must assess the collision risk associated with passing through a grid cell n, such as the grid with coordinates (35, 42), at a future time t, such as the 58th second.

[0049] Check a neighborhood centered on grid n, covering a total of 9 grid cells (3 by 3), and focus on the same time slice as time t, i.e., the 58th second. The system queries a global path reservation table, which is updated in real time and records the map grids occupied by all AGVs with planned paths at time t. After each AGV's path planning is completed, the map grid corresponding to time t is also determined. Count how many grid-time slice cells in this 9-square-meter area are already occupied by other AGVs' established paths at the 58th second. If two cells, such as grid (35, 43) and grid (36, 42), are found to be occupied, then the number of occupied cells, N, is equal to 2.

[0050] like Figure 2 In the grid map environment shown, it is observed that the grid n (35, 42) is the center and covers a neighborhood of 9 grids (3 by 3). Grids (35, 43) and (36, 42) are occupied, and the other grids are not occupied.

[0051] The number N is multiplied by a preset conflict cost coefficient α. The coefficient α can be set based on experience, for example, to 5. Then, the spatiotemporal conflict cost C(n,t) passing through node n at this moment is equal to 5 multiplied by 2, that is, 10. This cost of 10 will be added to the cost function of the A* algorithm, significantly increasing the cost of selecting this path point, thereby guiding the path planner to find other paths with smaller or zero N values, that is, paths with fewer conflicts. In the above embodiment, for ease of calculation, the dimension of C(n,t) can be the same as g(n) and h(n), both in meters.

[0052] S6: Send the driving path to the corresponding AGV for execution.

[0053] Specifically, the dispatch server breaks down the final planned driving path into a series of path instruction sequences, including coordinate points, desired speed, and estimated arrival time. This instruction sequence is transmitted via a wireless local area network to the onboard control unit of the target AGV. The AGV's control unit receives and interprets the path instruction sequence, and controls the drive and steering systems accordingly, ensuring that the AGV precisely follows the predetermined trajectory to complete the material handling task.

[0054] like Figure 3 As shown, the present invention also relates to a scheduling system, including a processor and a memory, the memory storing a computer program, the processor can interact with the memory, can call the computer program (for example, through a bus), and then the processor executes the computer program. When the computer program is executed by the processor, the scheduling method of the above embodiment is implemented.

[0055] In the present invention, the aforementioned memory may be any tangible medium that contains or stores a program, such as a magnetic storage medium or a magneto-optical storage medium.

[0056] While several embodiments of the present invention have been shown and described herein, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Numerous modifications, variations, and alternatives will occur to those skilled in the art without departing from the concept and spirit of the present invention. It should be understood that various alternatives to the embodiments of the present invention described herein may be employed in practicing the present invention.

Claims

1. A scheduling method for an AGV cluster, characterized in that: The following steps are involved: Obtaining real-time status data of each AGV in the AGV cluster and a set of tasks to be scheduled, wherein the real-time status data includes the position, power and load status of the AGV; For each task in the task set, a production rhythm disturbance coefficient is calculated to represent the urgency of the task based on its timeliness requirements, the material cache status of the associated downstream workstation, and the future material consumption rate predicted based on historical data; For each combination of a pending task and an idle AGV, a weighted calculation is performed on the task's production cycle disturbance coefficient and the AGV's comprehensive execution suitability to obtain an allocation utility value. The comprehensive execution suitability is based on the AGV's current battery level, its historical task execution score, and the estimated congestion probability of the AGV's path for executing the task, calculated using a preset spatiotemporal gridded occupancy prediction model. Construct a task allocation model with the goal of maximizing the sum of all possible allocation utility values, and solve the model to obtain the optimal AGV-task allocation pair; For each AGV-task pair, the AGVs are routed according to the production cycle disturbance coefficients of their tasks, from high to low. This approach employs an improved A* algorithm, whose cost function incorporates a spatiotemporal conflict probability cost calculated based on the spatiotemporal trajectories of other AGVs that have already completed path planning, when calculating the cost of each path segment. The driving path is sent to the corresponding AGV for execution.

2. The method according to claim 1, characterized in that The step of calculating the production rhythm disturbance coefficient representing the urgency of the task includes: The task timeliness, downstream workstation material cache status, and future material consumption rate are normalized and weighted to obtain the production rhythm disturbance coefficient P, which is calculated as follows: P = w1 × [(current time - task creation time) / (task completion deadline - task creation time)] + w2 × [1 - (current buffer size / maximum buffer size of downstream stations)] + w3 × (material consumption rate within a preset future time window predicted based on historical production data / production line design peak consumption rate); Among them, w1, w2, and w3 are preset weight coefficients, and w1+w2+w3=1.

3. The method according to claim 1, characterized in that The step of generating the comprehensive execution suitability includes: The current power of the AGV, the historical task execution score, and the estimated path congestion probability are normalized and weighted to obtain the comprehensive execution suitability F, which is calculated as follows: F = c1 × (AGV current power / AGV full power) + c2 × (AGV historical task average execution score / full score) + c3 × (1-estimated path congestion probability); Among them, c1, c2, and c3 are preset weight coefficients; Furthermore, when the current power level of the AGV is lower than the sum of the estimated power consumption for executing the task and a preset safety power level, the comprehensive execution suitability F of the AGV for the task is set to 0.

4. The method according to claim 1, wherein The construction of a task allocation model with the goal of maximizing the sum of the allocation utility values ​​of all possible allocations includes: Construct a utility matrix with the tasks to be assigned as rows and the idle AGVs as columns. The elements in the matrix are the allocation utility values ​​of the corresponding AGV-task combinations. The Hungarian algorithm or KM algorithm is used to solve the utility matrix to obtain the optimal matching solution that maximizes the total utility value. This solution is the optimal AGV-task assignment pair.

5. The method according to claim 1, wherein The cost function f(n) used by the improved A* algorithm is: f(n)=g(n)+C(n,t)+h(n); in: g(n) is the actual physical path length cost from the path starting point to the current node n; C(n,t) is the spatiotemporal conflict probability cost, and its value is positively correlated with the situation where the preset spatiotemporal neighborhood around the AGV is occupied by the planned paths of other AGVs when the AGV passes through the node n at the estimated time t; h(n) is the estimated cost from the current node n to the target node, which is obtained by calculating the Manhattan distance or Euclidean distance between node n and the target node.

6. The method according to claim 5, characterized in that The calculation method of the spatiotemporal conflict probability cost C(n,t) is: Divide the map environment into grids of preset size and divide time into time slices of preset length; Count the number of grid-time slices N occupied by other AGVs' planned paths within the preset neighborhood centered on node n within the time slice at time t where the AGV is expected to arrive at node n; C(n,t)=α×N, where α is the preset conflict cost coefficient.

7. The method according to claim 2, characterized in that The material consumption rate is output through a time series prediction model trained on historical production data based on a long short-term memory network (LSTM).

8. The method according to claim 3, characterized in that The generation of comprehensive execution suitability also includes a safety constraint: if the current power of the AGV is less than the estimated power consumption of the task + the safety redundancy power, then F=0.

9. The method according to claim 6, characterized in that The grid is 1 meter × 1 meter, and the time slice is 1 second.

10. A scheduling system for an AGV cluster, characterized in that: The system comprises a processor and a memory, wherein the memory stores a computer program and the processor executes the computer program to implement the scheduling method for an AGV cluster according to any one of claims 1 to 9.

Images