Multi-robot collaborative job scheduling optimization method based on graph neural network

Abstract

The invention discloses a multi-robot collaborative job scheduling optimization method based on a graph neural network. The method comprises the following steps: S1, constructing a collaborative scheduling graph comprising robot nodes, task nodes and meeting point nodes; s2, initializing a feature vector of each node; s3, node feature aggregation and scheduling representation vector generation are executed through the graph neural network; s4, an improved Co-DPSIPP algorithm is adopted to calculate a handover score of a meeting point node, and optimization is carried out based on task completion time, energy consumption change and communication load; s5, selecting an optimal handover mode according to the handover score, and updating the scheduling graph; s6, executing task redistribution and path planning based on the updated scheduling graph; and S7, repeatedly executing reasoning and scheduling control until all tasks are completed, and outputting a scheduling scheme and a path sequence. The scheduling efficiency and path planning of the multi-robot system are remarkably improved, and the method is widely applied to automatic production and robot cooperation tasks.

Description

Technical Field

[0001] The present invention relates to the field of robotics technology, and in particular to a multi-robot collaborative operation scheduling optimization method based on graph neural network. Background Art

[0002] With the widespread application of multi-robot systems, particularly in industrial production, logistics distribution, and warehouse management, task scheduling and path planning for collaborative robot operations have become crucial for improving system efficiency. However, existing multi-robot collaborative operation scheduling technologies typically rely on heuristic algorithms or rule-based optimization methods, which often struggle to cope with the complex environments of large-scale, multi-robot collaborative operations.

[0003] At present, traditional scheduling methods often regard task allocation and path planning as independent problems, lacking the ability to dynamically adapt to robot status and task requirements. Most methods ignore the interdependence between robots and the real-time changes during task execution, resulting in the inability to make fast and accurate scheduling decisions in complex environments. For example, dynamic changing factors such as the robot's power, load status, and task priority when performing tasks are often not effectively considered, resulting in low task scheduling efficiency, unbalanced resource utilization, and even task delays or waste of robot resources.

[0004] In addition, traditional path planning algorithms are usually calculated based on static assumptions, ignoring the synergistic effects of multiple robots when performing tasks. Existing path planning methods mostly use local information to make decisions, which are prone to path conflicts and resource occupation problems between robots. Especially in complex environments, the risk of path conflicts is greatly increased, affecting overall operational efficiency. For multi-robot collaborative scheduling in complex scenarios, existing methods lack sufficient global optimization when processing path planning, resulting in low computational efficiency and an inability to efficiently resolve path conflicts.

[0005] Existing technologies for optimizing the scheduling of multi-robot collaborative tasks still face the problem of low computational efficiency. As the number of tasks and robots increases, traditional scheduling methods often become computationally complex, making them ineffective in real-time environments. Because these methods fail to fully consider the collaborative interactions between robots and the complexity of tasks, they often require significant computing resources, leading to scheduling delays and making them difficult to meet practical application requirements.

[0006] Therefore, although the existing multi-robot collaborative operation scheduling optimization methods can provide basic scheduling functions in some simple tasks and small-scale environments, they often exhibit shortcomings such as poor flexibility, low efficiency, and inaccurate path planning when dealing with complex environments, large-scale tasks, and multi-robot collaborative operations, and are unable to effectively cope with changing environments and task requirements.

[0007] Therefore, how to provide a multi-robot collaborative work scheduling optimization method based on graph neural network is an urgent problem that technicians in this field need to solve. Summary of the Invention

[0008] One purpose of the present invention is to propose a multi-robot collaborative task scheduling optimization method based on graph neural networks. The present invention makes full use of graph neural network technology, improved Co-DPSIPP algorithm and dynamic scheduling control strategy, and describes in detail how to realize robot task allocation and path planning optimization through graph neural networks, with the advantages of high efficiency, flexibility and strong adaptability. By constructing a collaborative scheduling graph and combining the attention mechanism and deep learning ability of graph neural networks, the present invention can accurately handle complex multi-robot collaborative tasks, automatically adjust the scheduling scheme in a dynamic environment, and achieve global optimization of task allocation and path planning. Compared with traditional methods, the present invention significantly reduces task conflicts and path blockages in multi-robot collaborative scheduling, improves computational efficiency, and can adaptively adjust in complex and dynamic working environments to ensure efficient and stable task completion.

[0009] A multi-robot collaborative operation scheduling optimization method based on a graph neural network according to an embodiment of the present invention includes the following steps:

[0010] S1. Construct a collaborative scheduling graph including robot nodes, task nodes and rendezvous nodes;

[0011] S2. Initialize the feature vectors of various nodes in the collaborative scheduling graph;

[0012] S3. Input the collaborative scheduling graph into the graph neural network model and perform node feature aggregation and scheduling representation vector generation through the graph attention mechanism;

[0013] S4. Using the improved Co-DPSIPP algorithm, the handover score of each rendezvous node is calculated based on the scheduling representation vector. The scoring is based on the task completion time increment, system energy consumption change and communication load level.

[0014] S5. Based on the handover scoring results, the scheduling controller selects the optimal handover mode, including a single rendezvous point mode and a multi-rendezvous point mode, and updates the collaborative scheduling graph structure.

[0015] S6. Based on the updated collaborative scheduling graph, perform robot task reallocation and path planning;

[0016] S7. Repeat the graph neural network reasoning, handover score calculation and scheduling control operations until all task nodes are completed, and output the multi-robot collaborative operation scheduling plan and path execution sequence.

[0017] Optionally, the connection edges of the collaborative scheduling graph represent the allocation relationship between robots and tasks, the handover relationship between robots and meeting points, and the interaction relationship between tasks and meeting points.

[0018] Optionally, the feature vectors of each type of node include:

[0019] Robot node characteristics include current location, remaining power, task status, and load capacity;

[0020] Task node characteristics include task type, deadline, and priority;

[0021] The characteristics of the rendezvous point node include reachability index and communication cost.

[0022] Optionally, the improved Co-DPSIPP algorithm includes:

[0023] By constructing a collaborative scheduling graph consisting of robots, tasks, and rendezvous nodes, and introducing a graph neural network to generate a scheduling representation vector for each node, the rendezvous handover score is calculated based on the scheduling representation vector, replacing the original heuristic handover judgment mechanism.

[0024] Set a handover score vector based on the output of the graph neural network to drive the handover controller to perform adaptive switching of handover modes, including single rendezvous point mode, multiple rendezvous point mode, and direct task execution mode;

[0025] A joint scheduling objective function is embedded in the handover score calculation process. The objective function considers task completion time, energy consumption cost and communication load cost simultaneously, and is used to optimize the scheduling graph structure and path allocation strategy.

[0026] Optionally, the S3 specifically includes:

[0027] S31. For each node in the collaborative scheduling graph, calculate the attention coefficient with the adjacent nodes:

[0028] e ij =LeakyReLU(a T ·[Wh i ||Wh j ]);

[0029] Among them, e ij is the attention weight of node i to node j, h i is the feature vector of node i, h j is the feature vector of node j, W is the trainable weight matrix used to linearly transform node features, a is the learnable parameter vector of the attention mechanism, || represents the vector concatenation operation, and LeakyReLU is the leaky rectified linear unit activation function;

[0030] S32. Normalize the attention coefficient to obtain the standardized attention weight:

[0031]

[0032] Among them, α ij is the normalized attention weight of node i to node j, represents the set of adjacent nodes of node i, and exp is an exponential function;

[0033] S33. Perform node feature aggregation to generate updated node features:

[0034]

[0035] Where h′ i is the updated feature vector of node i, σ is the ReLU activation function;

[0036] S34. Globally pool the updated node features to generate a scheduling representation vector:

[0037]

[0038] Where z is the global scheduling representation vector, N is the total number of nodes in the collaborative scheduling graph, and ∑ represents the summation operation.

[0039] Optionally, the S4 specifically includes:

[0040] S41. Based on the scheduling representation vector z, construct a state evaluation tuple M for each rendezvous node k. k :

[0041] M k =(z k ,h r ,h t );

[0042] Among them, z k represents the scheduling representation vector of the rendezvous point k, h r represents the associated robot node feature vector, h t Represents the associated task node feature vector;

[0043] S42. Calculate the handover score of each rendezvous node using the improved Co-DPSIPP algorithm:

[0044]

[0045] Among them, Score k represents the handover score of the rendezvous point k, ΔT k represents the task completion time increment when selecting the rendezvous point k, T maxIndicates the maximum time increment allowed by the system, ΔE k Indicates the change in system energy consumption, E max Indicates the maximum energy consumption increment allowed by the system, L k Indicates the communication load level, L max represents the maximum communication load allowed by the system, α, β, and γ represent weight coefficients and satisfy α+β+γ=1;

[0046] S43, verify the reachability of each rendezvous point node, and select the robot whose distance to the rendezvous point is no more than D max And the distance between the meeting point and the mission point does not exceed D max The meeting point, where D max Indicates the maximum reachable distance preset by the system;

[0047] S44. Select a connection point with a score not less than S from the reachable rendezvous point. min The meeting points of S constitute the valid meeting point set V, where S min Indicates the system preset score acceptance threshold.

[0048] Optionally, the S5 specifically includes:

[0049] S51, receiving the handover scores of each rendezvous point node and the connection relationship in the collaborative scheduling graph, and constructing a handover mode candidate set, wherein the candidate set includes a single rendezvous point mode, a multiple rendezvous point mode, and a direct task execution mode;

[0050] S52. Set the handover mode selection function M to determine the optimal handover mode based on the handover score and the current task status:

[0051]

[0052] Where M represents the selected handover mode, Represents the set of all candidate handover modes, T m represents the task completion time under handover mode m, E m Indicates energy consumption, C m represents the communication load, ω1, ω2, and ω3 are weighted coefficients, and satisfy ω1+ω2+ω3=1;

[0053] S53, adjusting the structure of the collaborative scheduling graph according to the selected handover mode, and updating the connection edge between the robot node and the meeting point node;

[0054] S54. Synchronously update the allocation status of the task nodes, mark the task nodes and robot nodes involved in the decided handover mode, and enter the path planning process.

[0055] Optionally, the S6 specifically includes:

[0056] S61. According to the updated collaborative scheduling graph, the path requirement pair between each robot node and the corresponding task node is extracted, and the path requirement set is set as where r i represents the robot node, t j Represents the task node assigned to the robot;

[0057] S62: Use the path planning algorithm to calculate a feasible path for each pair of nodes in the path requirement set, and set the path as π ij ={v0,v1,...,v n}, where π ij Indicates robot r i Leading to the mission node t j The path, v k represents the kth graph node in the path;

[0058] S63. Introduce a conflict detection mechanism in the path planning process to detect whether there is a resource occupation conflict between paths and construct a conflict graph G. c =(V c ,E c ), where V c Represents the conflict unit in the path, E c Indicates path pairs with time or space conflicts;

[0059] S64. Use conflict avoidance strategies to modify the conflict graph, adjust some path sequences, and generate a final conflict-free path set Π. * ;

[0060] S65, set the path set Π * Each path in the task is sent to the corresponding robot node, bound to the task node identifier and registered the execution status.

[0061] Optionally, the S7 specifically includes:

[0062] S71. In each scheduling cycle, re-execute the node feature propagation process and update the scheduling representation vectors of all nodes in the collaborative scheduling graph. in represents the scheduling representation vector of the i-th node in the scheduling period t;

[0063] S72. Based on the updated scheduling representation vector, repeatedly calculate the handover scores of all rendezvous nodes And reselect handover mode M (t) , update the connection relationship of the collaborative scheduling graph;

[0064] S73: After the graph structure is updated, the path planning and task allocation process is re-executed to generate a path set Π within the current scheduling period. tand task execution allocation table A t , where π t Represents the robot path set, A t Represents the task allocation mapping relationship;

[0065] S74: Determine whether all task nodes in the collaborative scheduling graph have completed scheduling. If there are unfinished nodes, enter the next scheduling cycle t+1 and continue to execute S71 to S73;

[0066] S75. When all task nodes have completed scheduling, the iteration process is terminated and the final scheduling result is output, including the scheduling graph structure G * , path set Π * With the task completion record R, where G * is the final collaborative scheduling graph, Π * is the final path set, and R is the completion status identifier corresponding to each task node.

[0067] The beneficial effects of the present invention are:

[0068] This invention overcomes many of the shortcomings of existing multi-robot collaborative scheduling techniques by introducing a multi-robot collaborative task scheduling optimization method based on a graph neural network. Traditional methods often rely on heuristic algorithms or fixed rules and are unable to effectively handle task allocation, path planning, and robot state changes in complex environments, resulting in wasted resources, path conflicts, and inefficient scheduling. This invention, on the other hand, constructs a collaborative scheduling graph that includes robots, tasks, and meeting points. It utilizes the attention mechanism of a graph neural network to aggregate and optimize node features, enabling robots to dynamically adjust task allocation and path planning based on the current environment, thereby significantly improving the scheduling efficiency and flexibility of multi-robot systems.

[0069] By introducing the improved Co-DPSIPP algorithm, the present invention can accurately score the handover of rendezvous nodes, replacing the traditional heuristic handover judgment mechanism and further improving the accuracy of task allocation. During the task scheduling process, the present invention uses a graph neural network to model the interdependencies between robots, ensuring a more balanced task allocation and avoiding the problems of uneven load and resource waste in traditional methods. In addition, the dynamic scheduling control strategy of the present invention can adjust the scheduling plan in real time during the task execution process, allowing the robot to still perform tasks efficiently and stably when facing complex dynamic environments.

[0070] The advantages of the present invention are also reflected in the optimization of path planning. By updating the scheduling graph structure based on graph neural networks, robots can optimize paths in real time when performing tasks, avoiding path conflicts and resource occupation, and ensuring the efficiency and stability of multi-robot systems in collaborative operations. Compared with traditional path planning methods, the present invention can not only better adapt to complex tasks in multi-robot environments, but also provide significant improvements in real-time performance and computing efficiency. Ultimately, the present invention enables multi-robot collaborative operations to complete tasks quickly and efficiently in more complex and dynamic environments, improving the work efficiency and resource utilization of the overall system. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0072] Figure 1 This is a flowchart of a multi-robot collaborative operation scheduling optimization method based on graph neural network proposed by the present invention;

[0073] Figure 2 This is a schematic diagram of the collaborative scheduling graph construction process of a multi-robot collaborative operation scheduling optimization method based on graph neural network proposed in the present invention;

[0074] Figure 3 This is a schematic diagram of the path planning and task allocation optimization process of a multi-robot collaborative work scheduling optimization method based on graph neural network proposed in this invention. DETAILED DESCRIPTION

[0075] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0076] refer to Figure 1-3 , a multi-robot collaborative operation scheduling optimization method based on graph neural network, including the following steps:

[0077] S1. Construct a collaborative scheduling graph including robot nodes, task nodes and rendezvous nodes;

[0078] S2. Initialize the feature vectors of various nodes in the collaborative scheduling graph;

[0079] S3. Input the collaborative scheduling graph into the graph neural network model and perform node feature aggregation and scheduling representation vector generation through the graph attention mechanism;

[0080] S4. Using the improved Co-DPSIPP algorithm, the handover score of each rendezvous node is calculated based on the scheduling representation vector. The scoring is based on the task completion time increment, system energy consumption change and communication load level.

[0081] S5. Based on the handover scoring results, the scheduling controller selects the optimal handover mode, including a single rendezvous point mode and a multi-rendezvous point mode, and updates the collaborative scheduling graph structure.

[0082] S6. Based on the updated collaborative scheduling graph, perform robot task reallocation and path planning;

[0083] S7. Repeat the graph neural network reasoning, handover score calculation and scheduling control operations until all task nodes are completed, and output the multi-robot collaborative operation scheduling plan and path execution sequence.

[0084] This method addresses the inability of traditional scheduling methods to effectively account for the complex relationships between multiple robots by constructing a collaborative scheduling graph comprising robot nodes, task nodes, and rendezvous point nodes. Through model optimization using a graph neural network, the dynamic relationships between robots, tasks, and rendezvous points can be accurately expressed, improving task scheduling efficiency, avoiding resource waste and path conflicts, and significantly enhancing the overall performance of multi-robot collaborative operation systems.

[0085] In this embodiment, the connection edges of the collaborative scheduling graph represent the allocation relationship between robots and tasks, the handover relationship between robots and meeting points, and the interaction relationship between tasks and meeting points.

[0086] This method initializes the feature vectors of various nodes in the collaborative scheduling graph, enabling each node to fully represent the robot's state, task requirements, and rendezvous point information. This feature initialization process facilitates subsequent task allocation and path planning, ensuring that the characteristics of each node in collaborative scheduling are accurately captured and effectively reflecting the impact of various factors on scheduling results, further improving scheduling optimization.

[0087] In this embodiment, the characteristic vectors of each type of node include:

[0088] Robot node characteristics include current location, remaining power, task status, and load capacity;

[0089] Task node characteristics include task type, deadline, and priority;

[0090] The characteristics of the rendezvous point node include reachability index and communication cost.

[0091] This paper uses the graph attention mechanism of a graph neural network to aggregate node features and generate scheduling representation vectors, effectively combining local and global information about each node. This feature aggregation approach makes scheduling more accurate, intelligently adjusting task allocation and path planning based on multiple factors such as task priority and robot load, and reducing task scheduling imbalances and path conflicts.

[0092] In this embodiment, the improved Co-DPSIPP algorithm includes:

[0093] By constructing a collaborative scheduling graph consisting of robots, tasks, and rendezvous nodes, and introducing a graph neural network to generate a scheduling representation vector for each node, the rendezvous handover score is calculated based on the scheduling representation vector, replacing the original heuristic handover judgment mechanism.

[0094] Set a handover score vector based on the output of the graph neural network to drive the handover controller to perform adaptive switching of handover modes, including single rendezvous point mode, multiple rendezvous point mode, and direct task execution mode;

[0095] A joint scheduling objective function is embedded in the handover score calculation process. The objective function considers task completion time, energy consumption cost and communication load cost simultaneously, and is used to optimize the scheduling graph structure and path allocation strategy.

[0096] This paper uses an improved Co-DPSIPP algorithm to calculate the handover score for each rendezvous node based on the scheduling representation vector, taking into account multi-dimensional indicators such as task completion time, system energy consumption changes, and communication load level. This algorithm effectively replaces the traditional heuristic handover judgment mechanism, not only improving the accuracy of handover control, but also enabling the scheduling system to flexibly respond to various complex situations and optimize the efficiency of multi-robot collaboration.

[0097] In this embodiment, S3 specifically includes:

[0098] S31. For each node in the collaborative scheduling graph, calculate the attention coefficient with the adjacent nodes:

[0099] e ij =LeakyReLU(a T ·[Wh i ||Wh j ]);

[0100] Among them, e ij is the attention weight of node i to node j, h i is the feature vector of node i, h jis the feature vector of node j, W is the trainable weight matrix used to linearly transform node features, a is the learnable parameter vector of the attention mechanism, || represents the vector concatenation operation, and LeakyReLU is the leaky rectified linear unit activation function;

[0101] S32. Normalize the attention coefficient to obtain the standardized attention weight:

[0102]

[0103] Among them, α ij is the normalized attention weight of node i to node j, represents the set of adjacent nodes of node i, and exp is an exponential function;

[0104] S33. Perform node feature aggregation to generate updated node features:

[0105]

[0106] Where h′ i is the updated feature vector of node i, σ is the ReLU activation function;

[0107] S34. Globally pool the updated node features to generate a scheduling representation vector:

[0108]

[0109] Where z is the global scheduling representation vector, N is the total number of nodes in the collaborative scheduling graph, and ∑ represents the summation operation.

[0110] This method uses handover scores and the connections in the collaborative scheduling graph to select the optimal handover mode and update the collaborative scheduling graph structure, enabling adaptive switching between single and multiple rendezvous point modes. This process ensures that robots can flexibly adjust their operating methods in different task execution scenarios, optimizing overall task execution efficiency. Furthermore, by optimizing handover scores, resource conflicts and delays during task execution are further reduced.

[0111] In this embodiment, the S4 specifically includes:

[0112] S41. Based on the scheduling representation vector z, construct a state evaluation tuple M for each rendezvous node k. k :

[0113] M k =(z k ,h r ,h t );

[0114] Among them, z krepresents the scheduling representation vector of the rendezvous point k, h r represents the associated robot node feature vector, h t Represents the associated task node feature vector;

[0115] S42. Calculate the handover score of each rendezvous node using the improved Co-DPSIPP algorithm:

[0116]

[0117] Among them, Score k represents the handover score of the rendezvous point k, ΔT k represents the task completion time increment when selecting the rendezvous point k, T max Indicates the maximum time increment allowed by the system, ΔE k Indicates the change in system energy consumption, E max Indicates the maximum energy consumption increment allowed by the system, L k Indicates the communication load level, L max represents the maximum communication load allowed by the system, α, β, and γ represent weight coefficients and satisfy α+β+γ=1;

[0118] S43, verify the reachability of each rendezvous point node, and select the robot whose distance to the rendezvous point is no more than D max And the distance between the meeting point and the mission point does not exceed D max The meeting point, where D max Indicates the maximum reachable distance preset by the system;

[0119] S44. Select a connection point with a score not less than S from the reachable rendezvous point. min The meeting points of S constitute the valid meeting point set V, where S min Indicates the system preset score acceptance threshold.

[0120] This invention effectively resolves resource competition and path conflicts between tasks by performing robot task reallocation and path planning based on an updated collaborative scheduling graph. The updated graph structure reflects robot state changes and task progress in real time, ensuring that each robot is always in optimal working condition. This reduces task delays and path conflicts, significantly improving the collaborative efficiency of multi-robot systems.

[0121] In this embodiment, the S5 specifically includes:

[0122] S51, receiving the handover scores of each rendezvous point node and the connection relationship in the collaborative scheduling graph, and constructing a handover mode candidate set, wherein the candidate set includes a single rendezvous point mode, a multiple rendezvous point mode, and a direct task execution mode;

[0123] S52. Set the handover mode selection function M to determine the optimal handover mode based on the handover score and the current task status:

[0124]

[0125] Where M represents the selected handover mode, Represents the set of all candidate handover modes, T m represents the task completion time under handover mode m, E m Indicates energy consumption, C m represents the communication load, ω1, ω2, and ω3 are weighted coefficients, and satisfy ω1+ω2+ω3=1;

[0126] S53, adjusting the structure of the collaborative scheduling graph according to the selected handover mode, and updating the connection edge between the robot node and the meeting point node;

[0127] S54. Synchronously update the allocation status of the task nodes, mark the task nodes and robot nodes involved in the decided handover mode, and enter the path planning process.

[0128] The present invention ensures that the scheduling representation vectors of all nodes in the collaborative scheduling graph are updated in a timely manner by re-executing node feature propagation in each scheduling cycle. This dynamic update mechanism enables the scheduling process to adapt to environmental changes and task adjustments in real time, greatly improving the response speed and flexibility of the scheduling system, and providing strong support for multi-robot collaborative operations in complex environments.

[0129] In this embodiment, S6 specifically includes:

[0130] S61. According to the updated collaborative scheduling graph, the path requirement pair between each robot node and the corresponding task node is extracted, and the path requirement set is set as where r i represents the robot node, t j Represents the task node assigned to the robot;

[0131] S62: Use the path planning algorithm to calculate a feasible path for each pair of nodes in the path requirement set, and set the path as π ij ={v0,v1,...,v n}, where π ij Indicates robot r i Leading to the mission node t j The path, v k represents the kth graph node in the path;

[0132] S63. Introduce a conflict detection mechanism in the path planning process to detect whether there is a resource occupation conflict between paths and construct a conflict graph G. c =(Vc ,E c ), where V c Represents the conflict unit in the path, E c Indicates path pairs with time or space conflicts;

[0133] S64. Use conflict avoidance strategies to modify the conflict graph, adjust some path sequences, and generate a final conflict-free path set Π. * ;

[0134] S65, set the path set Π * Each path in the task is sent to the corresponding robot node, bound to the task node identifier and registered the execution status.

[0135] This invention further optimizes the efficiency of the scheduling system by calculating the handover score for each rendezvous node and reselecting the handover mode. Intelligent switching of handover modes selects the most appropriate task execution mode based on the actual needs of the task, reducing unnecessary waiting and resource waste during the scheduling process, significantly improving the operating efficiency of the robotic system.

[0136] In this embodiment, the S7 specifically includes:

[0137] S71. In each scheduling cycle, re-execute the node feature propagation process and update the scheduling representation vectors of all nodes in the collaborative scheduling graph. in represents the scheduling representation vector of the i-th node in the scheduling period t;

[0138] S72. Based on the updated scheduling representation vector, repeatedly calculate the handover scores of all rendezvous nodes And reselect handover mode M (t) , update the connection relationship of the collaborative scheduling graph;

[0139] S73: After the graph structure is updated, the path planning and task allocation process is re-executed to generate a path set Π within the current scheduling period. t and task execution allocation table A t , where π t Represents the robot path set, A t Represents the task allocation mapping relationship;

[0140] S74: Determine whether all task nodes in the collaborative scheduling graph have completed scheduling. If there are unfinished nodes, enter the next scheduling cycle t+1 and continue to execute S71 to S73;

[0141] S75. When all task nodes have completed scheduling, the iteration process is terminated and the final scheduling result is output, including the scheduling graph structure G * , path set Π* With the task completion record R, where G * is the final collaborative scheduling graph, Π * is the final path set, and R is the completion status identifier corresponding to each task node.

[0142] The present invention re-executes the path planning and task allocation process after the graph structure is updated, generates a path set and task execution allocation table within the current scheduling cycle, and ensures that each task can be allocated and executed in a timely and effective manner. Finally, through the scheduling completion and path execution records of each task node, it ensures that all tasks are completed smoothly according to the predetermined plan, optimizes the scheduling cycle and execution results of the entire system, and improves the overall operating efficiency of the system.

[0143] Example 1:

[0144] In order to verify the feasibility of the present invention in implementation, the present invention is applied to the multi-robot collaborative work scheduling task in a large production workshop. The workshop uses five automated robots to perform production tasks, including handling, assembly and quality inspection. The tasks in the production workshop have different priorities and time limit requirements. Each robot needs to consider its own load, remaining power, task progress and other information when performing tasks. Because traditional scheduling methods fail to fully consider these factors, robots have problems such as uneven load and path conflicts during task execution, which in turn affects production efficiency. In this context, the multi-robot collaborative work scheduling optimization method based on graph neural network proposed in the present invention shows its important practical significance.

[0145] During implementation, a task node was first constructed for each task, containing information such as the task's priority and deadline. Each robot node contained information such as the robot's power level, load, and the current task's completion progress. Meeting point nodes represented the intersections between robots. Using the graph neural network's reasoning mechanism, the collaborative scheduling graph aggregated features for each node, calculating the optimal match between each task and robot, and based on this, performed path planning. Based on the graph neural network's reasoning results, the system dynamically adjusted task allocation strategies and path planning to ensure balanced allocation of robot resources and avoid path conflicts and task delays.

[0146] Specifically, when five robots work together in a production workshop, traditional scheduling methods are unable to address the problems of uneven loads and path conflicts between robots. However, a scheduling optimization method based on graph neural networks can fully consider each robot's power level, load, and task priority when assigning tasks, allowing for precise task allocation and path planning, thereby improving overall operational efficiency. If a robot's power level is low, the system automatically adjusts task allocation to allow it to perform tasks that are closer and have lower priority, avoiding resource waste and path conflicts.

[0147] To verify the effectiveness of this method, we compared our graph neural network-based scheduling optimization approach with a traditional heuristic algorithm. In the same production scenario, the traditional heuristic algorithm took 14.2 hours to complete the task, while the graph neural network-based scheduling optimization approach only took 10.8 hours, saving approximately 24%. Regarding the number of path conflicts, the traditional approach experienced 9, while the graph neural network approach only experienced 3, significantly reducing the number of conflicts. Regarding task allocation time, the graph neural network-based approach reduced it from 130 minutes to 55 minutes, saving over 50%.

[0148] These comparative data demonstrate that the scheduling optimization method based on graph neural networks demonstrates significant advantages in terms of task completion time, number of path conflicts, and task allocation time. This method not only improves task completion efficiency but also optimizes the use of robot resources, reducing path conflicts and energy consumption. The application of this method in real-world production environments provides a more efficient and intelligent solution for multi-robot collaborative operations.

[0149] Ultimately, the scheduling optimization method of this invention significantly improved the multi-robot collaborative operation system in a production workshop in terms of task allocation, path planning, and energy consumption. Compared with traditional heuristic scheduling methods, this method not only improves task completion efficiency but also optimizes the use of robot resources, reducing path conflicts and energy consumption. Through these optimizations, the overall efficiency of the system has been greatly improved, providing an efficient and intelligent solution for multi-robot collaborative operations in real-world production environments.

[0150] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A multi-robot collaborative operation scheduling optimization method based on graph neural network, characterized in that: The steps include: S1. Construct a collaborative scheduling graph including robot nodes, task nodes and rendezvous nodes; S2. Initialize the feature vectors of various nodes in the collaborative scheduling graph; S3. Input the collaborative scheduling graph into the graph neural network model and perform node feature aggregation and scheduling representation vector generation through the graph attention mechanism; S4. Using the improved Co-DPSIPP algorithm, the handover score of each rendezvous node is calculated based on the scheduling representation vector. The scoring is based on the task completion time increment, system energy consumption change and communication load level. S5. Based on the handover scoring results, the scheduling controller selects the optimal handover mode, including a single rendezvous point mode and a multi-rendezvous point mode, and updates the collaborative scheduling graph structure. S6. Based on the updated collaborative scheduling graph, perform robot task reallocation and path planning; S7. Repeat the graph neural network reasoning, handover score calculation and scheduling control operations until all task nodes are completed, and output the multi-robot collaborative operation scheduling plan and path execution sequence.

2. The multi-robot collaborative operation scheduling optimization method based on graph neural network according to claim 1 is characterized in that: The connection edges of the collaborative scheduling graph represent the allocation relationship between robots and tasks, the handover relationship between robots and meeting points, and the interaction relationship between tasks and meeting points.

3. The multi-robot collaborative operation scheduling optimization method based on graph neural network according to claim 1 is characterized in that: The characteristic vectors of each type of node include: Robot node characteristics include current location, remaining power, task status, and load capacity; Task node characteristics include task type, deadline, and priority; The characteristics of the rendezvous point node include reachability index and communication cost.

4. The multi-robot collaborative operation scheduling optimization method based on graph neural network according to claim 1 is characterized in that: The improved Co-DPSIPP algorithm includes: By constructing a collaborative scheduling graph consisting of robots, tasks, and rendezvous nodes, and introducing a graph neural network to generate a scheduling representation vector for each node, the rendezvous handover score is calculated based on the scheduling representation vector, replacing the original heuristic handover judgment mechanism. Set a handover score vector based on the output of the graph neural network to drive the handover controller to perform adaptive switching of handover modes, including single rendezvous point mode, multiple rendezvous point mode, and direct task execution mode; A joint scheduling objective function is embedded in the handover score calculation process. The objective function considers task completion time, energy consumption cost and communication load cost simultaneously, and is used to optimize the scheduling graph structure and path allocation strategy.

5. The multi-robot collaborative operation scheduling optimization method based on graph neural network according to claim 1 is characterized in that: The S3 specifically includes: S31. For each node in the collaborative scheduling graph, calculate the attention coefficient with the adjacent nodes: e ij =LeakyReLU(a T ·[Wh i ‖Wh j ]); Among them, e ij is the attention weight of node i to node j, h i is the feature vector of node i, h j is the feature vector of node j, W is the trainable weight matrix used to linearly transform node features, a is the learnable parameter vector of the attention mechanism, ‖ represents the vector concatenation operation, and LeakyReLU is the leaky rectified linear unit activation function; S32. Normalize the attention coefficient to obtain the standardized attention weight: Among them, α ij is the normalized attention weight of node i to node j, represents the set of adjacent nodes of node i, and exp is an exponential function; S33. Perform node feature aggregation to generate updated node features: Where h′ i is the updated feature vector of node i, σ is the ReLU activation function; S34. Globally pool the updated node features to generate a scheduling representation vector: Where z is the global scheduling representation vector, N is the total number of nodes in the collaborative scheduling graph, and ∑ represents the summation operation.

6. The multi-robot collaborative operation scheduling optimization method based on graph neural network according to claim 1 is characterized in that: The S4 specifically includes: S41. Based on the scheduling representation vector z, construct a state evaluation tuple M for each rendezvous node k. k : M k =(z k ,h r ,h t ); Among them, z k represents the scheduling representation vector of the rendezvous point k, h r represents the associated robot node feature vector, h t Represents the associated task node feature vector; S42. Calculate the handover score of each rendezvous node using the improved Co-DPSIPP algorithm: Among them, Score k represents the handover score of the rendezvous point k, ΔT k represents the task completion time increment when selecting the rendezvous point k, T max Indicates the maximum time increment allowed by the system, ΔE k Indicates the change in system energy consumption, E max Indicates the maximum energy consumption increment allowed by the system, L k Indicates the communication load level, L max represents the maximum communication load allowed by the system, α, β, and γ represent weight coefficients and satisfy α+β+γ=1; S43, verify the reachability of each rendezvous point node, and select the robot whose distance to the rendezvous point is no more than D max And the distance between the meeting point and the mission point does not exceed D max The meeting point, where D max Indicates the maximum reachable distance preset by the system; S44. Select a connection point with a score not less than S from the reachable rendezvous point. min The meeting points of S constitute the valid meeting point set V, where S min Indicates the system preset score acceptance threshold.

7. The multi-robot collaborative operation scheduling optimization method based on graph neural network according to claim 1 is characterized in that: The S5 specifically includes: S51, receiving the handover score S of each rendezvous node k and constructing a handover mode candidate set based on the connection relationship in the collaborative scheduling graph, wherein the candidate set includes a single rendezvous point mode, a multiple rendezvous point mode, and a direct task execution mode; S52. Set the handover mode selection function M to determine the optimal handover mode based on the handover score and the current task status: Where M represents the selected handover mode, Represents the set of all candidate handover modes, T m represents the task completion time under handover mode m, E m Indicates energy consumption, C m represents the communication load, ω1, ω2, and ω3 are weighted coefficients, and satisfy ω1+ω2+ω3=1; S53, adjusting the structure of the collaborative scheduling graph according to the selected handover mode, and updating the connection edge between the robot node and the meeting point node; S54. Synchronously update the allocation status of the task nodes, mark the task nodes and robot nodes involved in the decided handover mode, and enter the path planning process.

8. The multi-robot collaborative operation scheduling optimization method based on graph neural network according to claim 1 is characterized in that: The S6 specifically includes: S61. According to the updated collaborative scheduling graph, the path requirement pair between each robot node and the corresponding task node is extracted, and the path requirement set is set as where r i represents the robot node, t j Represents the task node assigned to the robot; S62: Use the path planning algorithm to calculate a feasible path for each pair of nodes in the path requirement set, and set the path as π ij ={v0,v1,...,v n }, where π ij Indicates robot r i Leading to the mission node t j The path, v k represents the kth graph node in the path; S63. Introduce a conflict detection mechanism in the path planning process to detect whether there is a resource occupation conflict between paths and construct a conflict graph G. c =(V c ,E c ), where V c Represents the conflict unit in the path, E c Indicates path pairs with time or space conflicts; S64. Use conflict avoidance strategies to modify the conflict graph, adjust some path sequences, and generate a final conflict-free path set Π. * ; S65, set the path set Π * Each path in the task is sent to the corresponding robot node, bound to the task node identifier and registered the execution status.

9. The multi-robot collaborative operation scheduling optimization method based on graph neural network according to claim 1 is characterized in that: The S7 specifically includes: S71. In each scheduling cycle, re-execute the node feature propagation process and update the scheduling representation vectors of all nodes in the collaborative scheduling graph. in represents the scheduling representation vector of the i-th node in the scheduling period t; S72. Based on the updated scheduling representation vector, repeatedly calculate the handover scores of all rendezvous nodes And reselect handover mode M (t) , update the connection relationship of the collaborative scheduling graph; S73: After the graph structure is updated, the path planning and task allocation process is re-executed to generate a path set Π within the current scheduling period. t and task execution allocation table A t , where π t Represents the robot path set, A t Represents the task allocation mapping relationship; S74: Determine whether all task nodes in the collaborative scheduling graph have completed scheduling. If there are unfinished nodes, enter the next scheduling cycle t+1 and continue to execute S71 to S73; S75. When all task nodes have completed scheduling, the iteration process is terminated and the final scheduling result is output, including the scheduling graph structure G * , path set Π * With the task completion record R, where G * is the final collaborative scheduling graph, Π * is the final path set, and R is the completion status identifier corresponding to each task node.

Images