Multi-agent distributed task collaboration method and system

Abstract

The invention discloses a multi-agent distributed task collaboration method and system, and relates to the technical field of computers, and the method comprises the steps: broadcasting a to-be-allocated task set to each agent, enabling the agents to estimate the task expected completion duration of each to-be-allocated task, and generating a corresponding to-be-allocated task expected duration matrix; receiving a to-be-allocated task expected duration matrix and a historical allocation task estimated duration from each agent; constructing a constraint optimization model according to the to-be-allocated task expected duration matrix corresponding to each agent and the historical allocation task estimated duration; and solving the constraint optimization model through a global constraint optimization algorithm, and allocating each task to be allocated to a corresponding agent according to an optimal solution. Therefore, the flexibility of task allocation is improved, and resource waste and unreasonable allocation in task allocation are reduced.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a multi-agent distributed task collaboration method and system. Background Art

[0002] As a key link in the logistics industry, warehouse management no longer relies solely on manual operations and single automated equipment. With the demand for e-commerce, rapid turnover and large-scale supply chains, warehouse management systems are gradually developing towards more efficient, flexible and intelligent ones.

[0003] Traditional warehousing systems typically rely on centralized scheduling systems for task allocation and resource scheduling. At its core, a centralized control unit manages and schedules all warehousing tasks and resources. This system worked well in early, small-scale warehousing environments. However, as the number of tasks and resource types increased, the computational and management pressures placed on the centralized scheduling system increased significantly, slowing the system's response speed and creating a bottleneck in processing capacity. Furthermore, warehousing systems need to rapidly respond to dynamic changes in tasks and emergencies, but traditional centralized approaches often lack flexibility and are unable to quickly adjust to environmental changes in real time.

[0004] A multi-agent system (MAS) consists of multiple agents with autonomous perception, decision-making, and execution capabilities. Multi-agent task collaboration refers to the coordination and cooperation between agents within the system to jointly complete specific tasks. Task allocation, as a key step in task collaboration, directly impacts the operational efficiency and success rate of the entire multi-agent system.

[0005] Currently, multi-agent task allocation methods mainly fall into two categories: centralized control and traditional distributed control. Centralized control relies on one or a few central nodes to be responsible for task allocation, information synchronization, and agent coordination. It performs well in small-scale scenarios with relatively simple tasks. Although it can unify coordination and optimize global goals, as the number of agents increases and the environment becomes more dynamic, the computational load and communication delay of the central node increase dramatically, which can easily lead to decision-making bottlenecks and single points of failure. Traditional distributed task allocation methods rely on static or predefined rules. Faced with increasing task complexity and environmental uncertainty, they cannot effectively adapt to the rapid changes in the task environment, resulting in low coordination efficiency and even task failure.

[0006] To address the above issues, the industry has not yet proposed a better technical solution. Summary of the Invention

[0007] The present application provides a multi-agent distributed task collaboration method, system, storage medium, computer program product and electronic device to at least solve the problems of insufficient centralized scheduling flexibility, insufficient real-time adaptability and low collaboration efficiency of multi-agent systems in current related technologies.

[0008] In a first aspect, an embodiment of the present application provides a multi-agent distributed task collaboration method, comprising: broadcasting a set of tasks to be assigned to each agent, so that the agent estimates the expected completion time of each task to be assigned and generates a corresponding expected time matrix of tasks to be assigned; receiving the expected time matrix of tasks to be assigned and the estimated time of historical assigned tasks from each of the agents; wherein, if the agent state is idle, the estimated time of queued tasks is 0; if the agent state is running, the estimated time of historical assigned tasks is the sum of the expected completion time of tasks corresponding to at least one assigned task to be executed in the agent; constructing a constraint optimization model based on the expected time matrix of tasks to be assigned and the estimated time of historical assigned tasks corresponding to each of the agents; the constraint optimization model includes task execution constraints, agent queued task quantity constraints, agent maximum load constraints and objective functions; solving the constraint optimization model by a global constraint optimization algorithm, and assigning each of the tasks to be assigned to the corresponding agent according to the optimal solution;

[0009] The objective function is expressed as follows:

[0010]

[0011] Where M represents the total number of tasks in the task set to be assigned, N represents the total number of agents; x ij is a decision variable, indicating whether task i is assigned to agent j. If so, then x ij =1, otherwise x ij =0;t ij represents the time required for agent j to perform task i, which is determined according to the expected task duration distribution; is the total estimated time of the assigned tasks of agent j;

[0012] The task execution constraints restrict each task to be assigned to one agent:

[0013]

[0014] The agent queue task quantity constraint is used to limit the upper limit of the number of tasks assigned to the agent to avoid agent task backlog:

[0015]

[0016] Where Q jis the threshold number of assigned tasks for agent j;

[0017] The agent maximum load constraint is used to ensure that the agent will not be overloaded while executing queued tasks and currently assigned tasks:

[0018]

[0019] Where C j represents the maximum workload that agent j can bear.

[0020] In a second aspect, an embodiment of the present application provides a multi-agent distributed task collaboration system, comprising: a task set broadcasting unit, for broadcasting a set of tasks to be assigned to each agent, so that the agent estimates the expected completion time of each task to be assigned and generates a corresponding expected time matrix of the task to be assigned; an estimated time receiving unit, for receiving the expected time matrix of the task to be assigned and the estimated time of historical assigned tasks from each of the agents; wherein, if the agent state is idle, the estimated time of the queued task is 0; if the agent state is running, the estimated time of the historical assigned tasks is the sum of the expected completion time of the tasks corresponding to at least one assigned task to be executed in the agent; a constraint optimization modeling unit, for constructing a constraint optimization model according to the expected time matrix of the task to be assigned and the estimated time of historical assigned tasks corresponding to each of the agents; the constraint optimization model includes task execution constraints, agent queued task quantity constraints, agent maximum load constraints and objective functions; a global constraint solving unit, for solving the constraint optimization model through a global constraint optimization algorithm, and assigning each of the tasks to be assigned to the corresponding agent according to the optimal solution;

[0021] The objective function is expressed as follows:

[0022]

[0023] Where M represents the total number of tasks in the task set to be assigned, N represents the total number of agents; x ij is a decision variable, indicating whether task i is assigned to agent j. If so, then x ij =1, otherwise x ij =0;t ij represents the time required for agent j to perform task i, which is determined according to the expected task duration distribution; is the total estimated time of the assigned tasks of agent j;

[0024] The task execution constraints restrict each task to be assigned to one agent:

[0025]

[0026] The agent queue task quantity constraint is used to limit the upper limit of the number of tasks assigned to the agent to avoid agent task backlog:

[0027]

[0028] Where Q j is the threshold number of assigned tasks for agent j;

[0029] The agent maximum load constraint is used to ensure that the agent will not be overloaded while executing queued tasks and currently assigned tasks:

[0030]

[0031] Where C j represents the maximum workload that agent j can bear.

[0032] In a third aspect, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the steps of the multi-agent distributed task collaboration method of any embodiment of the present application.

[0033] In a fourth aspect, an embodiment of the present application provides a storage medium on which a computer program is stored, characterized in that when the program is executed by a processor, the steps of the multi-agent distributed task collaboration method of any embodiment of the present application are implemented.

[0034] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the multi-agent distributed task collaboration method of any embodiment of the present application.

[0035] The multi-agent distributed task collaboration method and system provided by this application can produce at least the following technical effects:

[0036] (1) By broadcasting the set of tasks to be assigned and requiring each agent to estimate the expected completion time of each task, task assignment can take into account the load of each agent and the task execution time. By generating a task expected duration matrix and combining it with historical task data, a state-based dynamic adjustment method can more accurately evaluate the timing of each task assignment and the most suitable agent. This makes task assignment not only dependent on static rules or simple centralized control, but also based on real-time calculation and data feedback. This improves the flexibility of task assignment, reduces resource waste and unreasonable allocation in task assignment, and thus improves the response speed and efficiency of the entire warehousing system.

[0037] (2) Compared with the traditional centralized control method, each intelligent agent independently evaluates the task duration of each task to be assigned and shares load information. Through distributed task collaboration, the excessive computational burden of the central node is avoided, while also reducing communication delays, enhancing the response speed and flexibility of the system, and maintaining a high task allocation efficiency when facing a large number of intelligent agents and a dynamically changing environment.

[0038] (3) By constructing a constrained optimization model and using a global constrained optimization algorithm to solve it, we ensure that each task is assigned to the appropriate agent in the best way. While meeting multiple constraints such as task execution time, agent load limit, and the number of task queues, we maximize resource utilization and ensure that the agent can complete the task within a reasonable load range, thereby avoiding task execution delays or failures caused by overload, and also avoiding idle resources, thus achieving global optimal scheduling.

[0039] Through this technical solution, by combining the global constraint optimization algorithm and the distributed collaboration mechanism of the multi-agent system, the efficiency, stability and reliability of the warehouse management system in dealing with complex task allocation and environmental changes are significantly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0041] Figure 1 A flowchart of an example of a multi-agent distributed task collaboration method according to an embodiment of the present application is shown;

[0042] Figure 2 A structural block diagram of an example of a multi-agent distributed task collaboration system according to an embodiment of the present application is shown;

[0043] Figure 3 This is a schematic structural diagram of an embodiment of an electronic device of the present application. DETAILED DESCRIPTION

[0044] To make the purpose, technical solutions, and advantages of the embodiments of this application more clear, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0045] In the technical solutions of this application, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved shall comply with the provisions of relevant laws and regulations and shall not violate public order and good morals.

[0046] Figure 1 A flowchart of an example of a multi-agent distributed task collaboration method according to an embodiment of the present application is shown.

[0047] The execution entity of the method of the embodiment of the present application can be any controller or processor with computing or processing capabilities. Specifically, it can be implemented by a multi-agent task allocation management platform. Through a distributed architecture and constraint optimization model, the response speed, computing efficiency, and global optimization capabilities of the warehouse management system are significantly improved, the flexibility and robustness of the system in complex dynamic environments are enhanced, and operation and maintenance costs are reduced, achieving efficient and reliable task allocation and resource scheduling.

[0048] In some examples, it can be integrated into an electronic device or terminal through software, hardware, or a combination of software and hardware, and the type of terminal or electronic device can be diverse, such as a mobile phone, tablet computer, or desktop computer, etc.

[0049] like Figure 1 As shown, in step S110, the set of tasks to be assigned is broadcast to each agent, so that the agent estimates the expected completion time of each task to be assigned and generates a corresponding expected completion time matrix of the tasks to be assigned.

[0050] In some embodiments, the platform first identifies all tasks to be assigned and transmits all task information to each intelligent agent through network broadcast or multicast. These tasks can be warehouse operation tasks or other types of tasks, and can contain corresponding task attribute information, such as task type, task resource requirements, task priority and other related parameters.

[0051] Exemplarily, the agents are warehouse agents, such as automated guided vehicles (AGVs), intelligent forklifts, and sorting robots. Tasks are warehouse operation tasks, and the corresponding task types include any of the following: receiving tasks, picking tasks, and inventory tasks. Thus, each task type has different standard execution times and resource requirements. Task resource requirements refer to the physical resources required to complete the task, such as storage space, transportation equipment, and robotic equipment. This allows each agent to independently estimate the completion time of each task based on the task attributes and the agent's own state or performance, and generate an expected completion time matrix for each task to be assigned. This matrix contains the agent's expected completion time for each task to be assigned. Exemplarily, each agent estimates the expected completion time for each task through its own computational model, taking into account factors such as task attributes and dynamic changes in the environment. The agent then returns the expected completion time matrix for each task to the management platform.

[0052] Therefore, the set of tasks to be assigned is broadcast to all intelligent agents through the central management platform to ensure transparent information sharing, so that each intelligent agent can timely understand the task status within the system. Each intelligent agent can estimate the completion time of the task in real time based on historical experience and its own execution capabilities, thereby improving the system's sensitivity and response speed to task execution.

[0053] In step S120, the expected duration matrix of the tasks to be assigned and the estimated duration of the historical assigned tasks are received from each agent.

[0054] Here, if the agent state is idle, the estimated duration of the queued task is If it is 0, it means that the agent currently has no tasks waiting to be executed. If the agent is in the running state, the estimated duration of the historical assigned tasks is the sum of the expected completion times of at least one assigned task to be executed by the agent, representing its current load status.

[0055] In this way, the management platform receives the expected duration matrix of the tasks to be assigned from each agent and synchronizes the historical task assignment data of each agent, realizing a real-time feedback mechanism for the task status, helping the management platform understand the current workload of each agent, and thus flexibly adjusting the task assignment strategy to ensure that task assignment can reasonably balance the load and avoid overload or idleness of the agent.

[0056] In step S130, a constraint optimization model is constructed based on the expected duration matrix of the tasks to be assigned corresponding to each agent and the estimated duration of the historical assigned tasks. The constraint optimization model includes task execution constraints, agent queue task quantity constraints, agent maximum load constraints and objective functions.

[0057] Here, a comprehensive constrained optimization model is constructed based on the expected duration distribution of each agent and the estimated duration of historical tasks. The goal is to optimize task allocation so that the task execution time of the entire system is minimized while satisfying the load and resource constraints of the agents.

[0058] Specifically, task execution constraints ensure that each task is appropriately assigned to an agent. Constraints on the number of queued tasks limit the number of tasks each agent can execute, preventing an agent from being overloaded with too many tasks. Maximum load constraints ensure that each agent's workload (including the estimated duration of the current task and previously assigned queued tasks) does not exceed its maximum capacity, preventing agent overload.

[0059] More specifically, the objective function is expressed as follows:

[0060]

[0061] Where M represents the total number of tasks in the task set to be assigned, N represents the total number of agents; x ij is a decision variable, indicating whether task i is assigned to agent j. If so, then x ij =1, otherwise x ij =0;t ij represents the time required for agent j to perform task i, which is determined according to the expected task duration distribution; is the total estimated time of the assigned tasks for agent j.

[0062] In the objective function of formula (1), the core purpose is to minimize the total execution time of all tasks, thereby optimizing the efficiency of the entire system. The execution time of each task i includes not only the estimated duration t of the task itself, but also the total execution time t of all tasks. ij , we also need to consider the current load of the agent, that is, the estimated time of the queued tasks This allows task allocation to not only focus on the execution of the current task, but also consider the actual workload of the agent, effectively balancing the task allocation of each agent and avoiding task delays due to overload of an agent.

[0063] The task execution constraints restrict each task to be assigned to one agent:

[0064]

[0065] In the task execution constraints, each task i must be assigned to one and only one agent j, ensuring the uniqueness of task assignment, thereby avoiding the situation where the task is executed repeatedly or by multiple agents.

[0066] The agent queue task quantity constraint is used to limit the upper limit of the number of tasks assigned to the agent to avoid agent task backlog:

[0067]

[0068] Where Q j is the threshold number of assigned tasks for agent j.

[0069] In the agent queue task quantity constraint, the number of queue tasks for each agent j is limited, that is, at most Q tasks are assigned at the same time. j Therefore, by limiting the number of queued tasks, the system can reasonably schedule the tasks of each agent, avoid task backlogs, and reduce task delays caused by agent overload.

[0070] The maximum agent load constraint is used to ensure that the agent does not become overloaded while executing queued tasks and currently assigned tasks:

[0071]

[0072] Where C j represents the maximum workload that agent j can bear.

[0073] According to formula (4), the maximum load constraint of the agent ensures that the workload of agent j will not exceed its maximum capacity C during the execution of the task. j , ensuring that the agent does not exceed its resource and capacity limits when performing tasks. By introducing the agent maximum load constraint, the system can limit the workload of each agent, so that when assigning a new task to an agent, it is necessary to consider the total duration of the tasks it has already executed and the tasks in the queue, to ensure that the maximum workload is not exceeded. This effectively avoids the problem of overloading the agent and reduces task delays and execution failures caused by overload. In addition, reasonable load control for each agent improves the stability of the entire warehouse management system and the on-time completion rate of tasks, ensuring the quality of the agent's task execution.

[0074] Therefore, by introducing multiple constraints, the management platform can comprehensively consider factors such as task execution time, agent load and task queue, not only to meet the needs of each task, but also to ensure that the load of each agent is within an acceptable range, optimize task allocation on a global scale, and reasonably schedule the tasks of the agents to avoid excessive load or task backlogs on the agents, thereby improving the task execution efficiency and the overall stability of the multi-agent system.

[0075] In step S140, the constraint optimization model is solved by a global constraint optimization algorithm, and each task to be assigned is assigned to a corresponding intelligent agent according to the optimal solution.

[0076] Specifically, when solving the constrained optimization model using a global constrained optimization algorithm, the task allocation scheme is adjusted through multiple iterations, gradually approaching the global optimal solution. This global optimization algorithm ensures that the task allocation scheme is optimal under all constraints, thereby improving the task execution efficiency of the entire multi-agent system.

[0077] It should be noted that the types of global constraint optimization algorithms can be diverse. For example, in the task allocation problem, since the decision variable x ij Since the problem is binary (each task is assigned to an agent or not), it can be solved using a MILP (Mixed-Integer Linear Programming) algorithm with integer decision variables and linear constraints. Furthermore, due to the large number and types of tasks and agents in a warehouse environment, and the complex constraints, various heuristic algorithms (such as genetic algorithms and particle swarm optimization) can also be used. These algorithms can achieve low computational time consumption despite high computational complexity, and are not subject to any restrictions here.

[0078] Through the embodiments of the present application, through the distributed task collaboration method combined with global constraint optimization, it is possible to respond to environmental changes in real time, effectively avoid the risk of single point failure, optimize resource utilization and load balancing, and achieve flexibility in task allocation strategies.

[0079] Regarding the details of estimating the expected completion time of the task in step S110, in some examples of the embodiments of the present application, it can be achieved through a deep learning model based on a multi-layer perceptron deployed in the intelligent agent.

[0080] In some embodiments, the intelligent agent is used to perform the following operations to estimate the expected completion time of the task to be assigned: parse the task type and task resource requirements corresponding to the task to be assigned, and obtain the device performance indicators of the intelligent agent to construct a multivariate input feature; and then input the multivariate input feature into a deep learning model based on a multilayer perceptron (MLP) to determine the expected completion time of the corresponding task.

[0081] Regarding task type parsing, in some implementations, each assigned task includes task type information (e.g., handling, sorting, storage, etc.), and the agent must parse the task's execution characteristics based on this task type. Different task types correspond to varying operational complexity, required resources, and execution time. To enable the agent to adapt to different tasks, task types can be converted using one-hot encoding.

[0082] Regarding the details of task resource requirement analysis, the resource requirements of a task include the equipment required for the task (such as robots, sensors, etc.), storage space, power requirements, etc., such as the number of robots required, the sensor accuracy required for each task, and the required storage space or processing power.

[0083] Regarding the description of device performance indicators, each agent will regularly update its device performance indicators, such as robot movement speed, handling efficiency, sensor accuracy, battery capacity and processor performance, etc., which reflect the processing ability of the agent when performing tasks and directly affect the execution time of the task.

[0084] By analyzing the type and resource requirements of the task, the intelligent agent can accurately identify the basic requirements of each task. By obtaining the intelligent agent's equipment performance indicators and matching them with the task resource requirements, the intelligent agent can dynamically adjust the task duration estimate based on its own capabilities to ensure the efficiency and accuracy of task allocation.

[0085] A multi-dimensional input feature vector is constructed using the above information and input into a deep learning model based on MLP. The MLP model is used to perform regression prediction on the task duration through a multi-layer neural network.

[0086] It should be noted that the MLP model can include multiple hidden layers, each containing multiple neurons. The number of neurons in each layer and the number of layers can be tuned through cross-validation to ensure the training effect of the model. Nonlinear activation functions (for example, ReLU activation function) are used between each layer to improve the nonlinear modeling capabilities of the network. In particular, through the multi-layer perceptron (MLP) model, the platform can accurately predict the execution time of each task based on the type of task (receiving task, picking task, inventory counting task) and the device performance of the intelligent agent.

[0087] Through the embodiments of the present application, a multivariate prediction model based on MLP is deployed in the intelligent agent, and the intelligent agent can accurately predict the execution time of the task according to the type of task, resource requirements and its own equipment performance. Since the MLP model has a strong nonlinear fitting ability, it can capture the relationship between complex features, thereby making the estimation of task duration more accurate. In addition, with the dynamic changes of the task environment (such as the type of task, changes in resource requirements, etc.), the MLP model can continuously optimize the prediction of task duration by training and learning new data, thereby adapting to tasks of different types and complexities. With the accumulation and feedback of data, the model can automatically adjust and optimize the duration prediction by continuously learning the feedback of the intelligent agent performing the task, and continuously iterate to improve the accuracy of the task duration prediction.

[0088] In some examples of the embodiments of the present application, the global constrained optimization algorithm adopts a collaborative particle swarm optimization algorithm. The solution process based on the global constrained optimization algorithm mainly includes the following operations:

[0089] Particle swarm initialization:

[0090] Each particle is defined as a task allocation scheme with a dimension of M×N, where the dimension corresponds to the allocation relationship between each task and each agent; the particle position is defined using decision variables. It indicates whether task i is assigned to agent j in particle k.

[0091] Here, in the particle swarm optimization algorithm, the task allocation scheme is defined by the particle position matrix of each particle. The position of each particle is Corresponding to the assignment relationship between task i and agent j, in the initial state, the particle position matrix and the corresponding task assignment scheme are random, which ensures that all possible task assignments are explored without any prior knowledge by covering a wider range of possibilities in the search space.

[0092] Under the premise of satisfying the task execution constraints, the number of agent queued tasks and the maximum load constraints of the agent, the initial position of the particle is randomly generated; the particle speed is used to generate the particle position. Represents the change in the assignment of task i to agent j in particle k. The initial value of the particle's velocity is zero. The particle's velocity will control the change in the task assignment scheme. The particle optimizes task scheduling based on velocity and position updates in each generation.

[0093] Specifically, the task execution constraint ensures that each task is assigned to an agent, and the agent queue task quantity constraint ensures that the number of tasks assigned to each agent does not exceed the preset upper limit Q. jThe maximum load constraint of the agent ensures that the total load of each agent (the execution time of the task to be assigned plus the estimated time of the queued task) does not exceed the maximum load capacity C of the agent. j .

[0094] Particle swarm update based on neighborhood collaborative search:

[0095] Particles not only rely on their own optimal solutions, but also use the information sharing mechanism of neighboring particles to accelerate the convergence of the global optimal solution through collaborative search.

[0096] The particle velocity update formula is:

[0097]

[0098] Where, represents the particle velocity of particle k at the tth iteration, which represents the change when task i is assigned to agent j; w(t) is the dynamic inertia weight, which determines the balance between global search and local development of the particle and gradually decreases with the number of iterations; c1(t) and c2(t) represent the acceleration constants of the personal optimal solution and the global optimal solution, respectively; r1 and r2 are random numbers in the range [0,1] used to increase the randomness of the search; is the personal optimal solution of particle k in the tth iteration when task i is assigned to agent j; is the task allocation scheme of particle k in the global optimal solution, which represents the optimal task allocation scheme among all particles; N(k) is the set of neighboring particles of particle k, and particles exchange information through collaborative search; l represents the neighboring particles of particle k; represents the particle position of particle k at the tth iteration, indicating whether task i is assigned to agent j; α is the influence coefficient of collaborative search, which controls the intensity of collaboration between particles.

[0099] In the particle velocity update formula shown in Equation (5), the particle velocity is used to represent the change in the task allocation scheme, controlling the direction and step size of the task allocation change. Specifically, the guiding role of particles between the current optimal solution and the global optimal solution is considered. At the same time, collaborative search is performed through information transmission between neighboring particles, allowing the particle swarm to consider neighborhood information during local search, thereby accelerating convergence.

[0100] Specifically, based on It provides an information sharing mechanism between neighboring particles. Particles adjust their speed and position according to the performance of neighboring particles, which enhances the cooperation between particles, avoids falling into local optimality, ensures that the optimal solution is found in a larger search space, and accelerates the convergence of the global optimal solution.

[0101] In this way, global search and local development are effectively balanced, and neighborhood collaborative search is integrated, so that particles can find the global optimal solution more quickly during the search process and avoid falling into the local optimum.

[0102] The particle position update formula is:

[0103]

[0104] Where σ(·) represents the Sigmoid activation function, which is used to convert particle velocity into probability to determine whether task i is assigned to agent j; r ij is a random number in the range [0,1], used to control whether task i is assigned to agent j.

[0105] In Equation (6), particle updates are based on the Sigmoid activation function, which converts velocity into a probability value to determine whether a task is assigned to agent j. This allows particles to adjust their task assignments in each iteration, ensuring the diversity of particle searches. This ensures that task assignment decisions are based on the particle's current velocity while enabling particles to conduct appropriate exploration during the search process, improving the randomness of task assignments and enhancing the algorithm's global search capabilities.

[0106] After the particle position is updated, check whether the task allocation scheme corresponding to the updated particle satisfies the task execution constraint, the number of agent queued tasks constraint, and the maximum load constraint of the agent at the same time; if so, perform the iteration stop condition judgment; if not, reinitialize the particle velocity to zero.

[0107] Here, after each particle position update, the task allocation solution represented by the particle undergoes a constraint verification. The particle will only continue to participate in subsequent iterations if the task execution constraints, the number of tasks queued by the agent, and the maximum load constraints are met. This constraint verification mechanism ensures that the particle's update after each iteration meets all constraint requirements, thus avoiding invalid solutions and ensuring the rationality and feasibility of the final solution.

[0108] The update formula of dynamic inertia weight is:

[0109]

[0110] Where w max and w min is the maximum and minimum value of the inertia weight, T max is the maximum number of iterations.

[0111] Referring to Equation (7), the inertia weight w(t) controls the balance between the global search and local search of the particles. The inertia weight gradually decreases with the increase in the number of iterations, so that in the early stages of the search, the particles have a larger inertia weight and a stronger global search capability. In the later stages, the inertia weight gradually decreases, making the particles more focused on local search. This helps the particle swarm avoid premature convergence in the early stages of the search and improves the accuracy of the search in the later stages.

[0112] The updating formula of the adaptive acceleration constant is:

[0113]

[0114] Where c min,1 and c max,1 They represent the minimum acceleration constant and the maximum acceleration constant of the individual optimal solution respectively; c min,2 and c max,2 They represent the minimum acceleration constant and the maximum acceleration constant of the global optimal solution respectively.

[0115] It should be noted that c1(t) is the optimal solution for controlling particles to individuals. The acceleration constant of the degree of dependence. Since the particle is more dependent on the global optimal solution in the early stage of the search space Conduct extensive exploration, and later rely more on personal optimal solutions To perform fine local optimization, the acceleration constant c1(t) should decrease with the increase of iteration number to guide particles to gradually turn to local search and enhance local optimization ability.

[0116] In addition, c2(t) controls the particle to the global optimal solution In the early stages of the search, particles should rely more on the global optimal solution for extensive exploration, while in the later stages, particles gradually turn to the individual optimal solution. Therefore, c2(t) should be large in the early stage to enhance the global search capability, and gradually decrease in the later stage to enhance the local search capability.

[0117] Therefore, in the early stages, particles rely on the global optimal solution through a large c2(t), allowing them to explore a wider range of solution spaces and avoid being trapped in local optimal solutions. As the number of iterations increases, particles gradually reduce their dependence on the global optimal solution and rely more on their own optimal solution for fine-grained local search, thereby accelerating the convergence process and improving the accuracy of the solution.

[0118] In the iterative stop condition judgment, when it is detected that the number of iterations reaches the maximum number of iterations T maxWhen the preset fitness convergence condition is met, the iteration is terminated and the particle with the largest fitness in the current particle swarm is returned as the optimal solution. Then, each task to be assigned is assigned to the corresponding agent according to the optimal solution.

[0119] Specifically, the algorithm terminates when the maximum number of iterations is reached or the fitness changes of all particles meet the convergence conditions, ensuring that the algorithm can stop at the appropriate time to avoid excessive iterations and waste of computing resources, and returns the particle with the highest fitness as the optimal solution, and finally assigns the task to the agent.

[0120] Through the embodiments of the present application, a collaborative particle swarm optimization algorithm is adopted. By dynamically adjusting the inertia weight and adaptive acceleration constant, particles can effectively balance global exploration and local optimization during the search process, and can flexibly respond to different search stages, thereby improving efficiency and accuracy. In addition, the collaborative search between particles improves the global convergence rate and avoids local optimality, enhancing the global search capability of the algorithm. Ultimately, the task allocation scheme is determined by the particle position of the optimal solution, ensuring that tasks can be efficiently allocated to appropriate agents under constraints.

[0121] In some examples of the embodiments of the present application, the fitness function of the collaborative particle swarm optimization algorithm is expressed by the following formula:

[0122] F=F1+λF2, Equation (10)

[0123]

[0124] Where F represents the fitness function, F1 represents the optimization term for the total execution time of the task, F2 represents the penalty term for task timeout, and λ represents the fitness balance factor; i,limit represents the maximum deadline of task i, which is determined according to a preset mapping relationship between task type and / or task priority and deadline.

[0125] Here, the fitness function F, used to evaluate the quality of each particle's solution, consists of two components: the total task execution time optimization term F1 and the task timeout penalty term F2. It aims to simultaneously minimize both the total task execution time and the timeout penalty. The balancing factor λ controls the relative importance of these two objectives, adjusting the relative weights of execution time and penalty terms during the optimization process. This allows the algorithm to comprehensively consider both task execution time and deadline requirements during task scheduling, enabling a flexible balance between execution efficiency and deadline constraints to meet the dynamic adjustment needs of different scenarios.

[0126] In Equation (11), the total task execution time optimization term F1 refers to the objective function of the constrained optimization model and calculates the total time required for task execution, including the execution time of the task itself and the time the agent queues tasks. The total execution time of all tasks on all agents is calculated, thereby minimizing the total task allocation time and improving the overall efficiency of the multi-agent system. In particular, by considering the impact of agent queued tasks, it is possible to effectively avoid uneven agent load or overload caused by task allocation, ensuring the efficiency of task scheduling.

[0127] In formula (12), the task timeout penalty term F2 is used to evaluate the penalty for processing tasks that exceed the maximum time limit by calculating the difference between the task execution time and the task maximum time limit. Specifically, it calculates whether all tasks have timed out and punishes the timed out tasks. The penalty amount for the timed out tasks is max(0,t ij -T i,limit ), which is the difference between the actual execution time of a task and its maximum deadline. The larger the difference, the greater the penalty, thus minimizing task timeouts. Therefore, the penalty term effectively prevents task timeouts and, by penalizing timed-out tasks, encourages the algorithm to find a task allocation solution that meets deadlines.

[0128] Through the embodiments of the present application, by introducing the particle swarm optimization algorithm and combining the dual optimization goals of the total execution time of multi-agent tasks and the time limit penalty items of each task, the time and time limit requirements of task execution can be efficiently balanced, so that in the task scheduling process, the task timeout problem can be effectively avoided while maximizing the efficiency of task execution.

[0129] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity of description, they are all expressed as a series of combined actions, but those skilled in the art should be aware that this application is not limited by the order of the actions described, because according to this application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by this application. In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0130] Figure 2 A structural block diagram of an example of a multi-agent distributed task collaboration system according to an embodiment of the present application is shown.

[0131] like Figure 2As shown, the multi-agent distributed task collaboration system 200 includes a task set broadcasting unit 210, an estimated duration receiving unit 220, a constraint optimization modeling unit 230 and a global constraint solving unit 240.

[0132] The task set broadcasting unit 210 is used to broadcast the to-be-assigned task set to each agent, so that the agent can respectively estimate the expected completion time of each to-be-assigned task and generate a corresponding to-be-assigned task expected completion time matrix.

[0133] The estimated duration receiving unit 220 is used to receive the expected duration matrix of the tasks to be assigned and the estimated duration of the historical assigned tasks from each of the intelligent agents; wherein, if the intelligent agent state is idle, the estimated duration of the queued tasks is 0; if the intelligent agent state is running, the estimated duration of the historical assigned tasks is the sum of the expected completion durations of the tasks corresponding to at least one assigned task to be executed in the intelligent agent.

[0134] The constraint optimization modeling unit 230 is used to construct a constraint optimization model based on the expected duration matrix of the tasks to be assigned corresponding to each of the intelligent agents and the estimated duration of the historical assigned tasks; the constraint optimization model includes task execution constraints, the number of tasks queued by the intelligent agent, the maximum load constraint of the intelligent agent and the objective function.

[0135] The global constraint solving unit 240 is used to solve the constraint optimization model through a global constraint optimization algorithm, and assign each of the tasks to be assigned to a corresponding intelligent agent according to the optimal solution.

[0136] The objective function is expressed as follows:

[0137]

[0138] Where M represents the total number of tasks in the task set to be assigned, N represents the total number of agents; x ij is a decision variable, indicating whether task i is assigned to agent j. If so, then x ij =1, otherwise x ij =0;t ij represents the time required for agent j to perform task i, which is determined according to the expected task duration distribution; is the total estimated time of the assigned tasks of agent j;

[0139] The task execution constraints restrict each task to be assigned to one agent:

[0140]

[0141] The agent queue task quantity constraint is used to limit the upper limit of the number of tasks assigned to the agent to avoid agent task backlog:

[0142]

[0143] Where Q j is the threshold number of assigned tasks for agent j;

[0144] The agent maximum load constraint is used to ensure that the agent will not be overloaded while executing queued tasks and currently assigned tasks:

[0145]

[0146] Where C j represents the maximum workload that agent j can bear.

[0147] In some embodiments, an embodiment of the present application provides a non-volatile computer-readable storage medium, which stores one or more programs including execution instructions, and the execution instructions can be read and executed by an electronic device (including but not limited to a computer, server, or network device, etc.) to execute the steps of any of the above-mentioned multi-agent distributed task collaboration methods of the present application.

[0148] In some embodiments, the embodiments of the present application also provide a computer program product, which includes a computer program stored on a non-volatile computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer performs any step of the above-mentioned multi-agent distributed task collaboration method.

[0149] In some embodiments, an embodiment of the present application also provides an electronic device comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the steps of the multi-agent distributed task collaboration method.

[0150] Figure 3 This is a hardware structure diagram of an electronic device for executing a multi-agent distributed task collaboration method provided by another embodiment of the present application. Figure 3 As shown, the device includes:

[0151] One or more processors 310 and memory 320, Figure 3 A processor 310 is taken as an example.

[0152] The device for executing the multi-agent distributed task collaboration method may further include: an input device 330 and an output device 340.

[0153] The processor 310, the memory 320, the input device 330 and the output device 340 may be connected via a bus or other means. Figure 3 The bus connection is taken as an example.

[0154] Memory 320, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer executable programs, and modules, such as the program instructions / modules corresponding to the multi-agent distributed task collaboration method in the embodiments of the present application. Processor 310 executes the non-volatile software programs, instructions, and modules stored in memory 320 to execute various server functional applications and data processing, thereby implementing the multi-agent distributed task collaboration method in the above-mentioned method embodiment.

[0155] The memory 320 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and applications required for at least one function; the data storage area may store data created based on the use of the electronic device, etc. In addition, the memory 320 may include a high-speed random access memory and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 320 may optionally include a memory remotely located relative to the processor 310, and these remote memories may be connected to the electronic device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0156] The input device 330 can receive input digital or character information and generate signals related to user settings and function control of the electronic device. The output device 340 can include a display device such as a display screen.

[0157] The one or more modules are stored in the memory 320 and, when executed by the one or more processors 310, perform the multi-agent distributed task collaboration method in any of the above method embodiments.

[0158] The above-mentioned product can execute the method provided in the embodiment of this application, and has the functional modules and beneficial effects corresponding to the execution method. For technical details not fully described in this embodiment, please refer to the method provided in the embodiment of this application.

[0159] The electronic devices of the embodiments of the present application exist in various forms, including but not limited to:

[0160] (1) Mobile communication devices: These devices are characterized by their mobile communication capabilities and their primary purpose is to provide voice and data communications. These terminals include smartphones, multimedia phones, feature phones, and low-end phones.

[0161] (2) Ultra-mobile personal computer devices: These devices fall under the category of personal computers and have computing and processing capabilities, and generally also have mobile Internet access. These terminals include PDAs, MIDs, and UMPCs.

[0162] (3) Portable entertainment devices: These devices can display and play multimedia content. They include audio and video players, handheld game consoles, e-books, smart toys, and portable car navigation devices.

[0163] (4) Other onboard electronic devices with data interaction functions, such as onboard computer devices installed in vehicles.

[0164] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of this embodiment.

[0165] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a general hardware platform, or of course, by hardware. Based on this understanding, the above technical solution, in essence, or the part that contributes to the relevant technology, can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or certain parts of the embodiment.

[0166] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A multi-agent distributed task collaboration method, the method comprising: Broadcast the set of tasks to be assigned to each agent, so that the agent can estimate the expected completion time of each task to be assigned and generate the corresponding expected completion time matrix of the tasks to be assigned; Receive from each of the agents a matrix of expected durations of tasks to be assigned and estimated durations of historically assigned tasks; wherein, if the agent is in an idle state, the estimated duration of the queued tasks is 0; if the agent is in a running state, the estimated duration of historically assigned tasks is the sum of the expected completion times of at least one assigned task to be executed in the agent; Constructing a constrained optimization model based on the expected duration matrix of the tasks to be assigned to each of the agents and the estimated duration of the historical assigned tasks; the constrained optimization model includes task execution constraints, agent queue task quantity constraints, agent maximum load constraints and an objective function; Solving the constraint optimization model by a global constraint optimization algorithm, and assigning each of the tasks to be assigned to a corresponding agent according to the optimal solution; The objective function is expressed as follows: Where M represents the total number of tasks in the task set to be assigned, N represents the total number of agents; x ij is a decision variable, indicating whether task i is assigned to agent j. If so, then x ij =1, otherwise x ij =0;t ij represents the time required for agent j to perform task i, which is determined according to the expected task duration distribution; is the total estimated time of the assigned tasks of agent j; The task execution constraints restrict each task to be assigned to one agent: The agent queue task quantity constraint is used to limit the upper limit of the number of tasks assigned to the agent to avoid agent task backlog: Where Q j is the threshold number of assigned tasks for agent j; The agent maximum load constraint is used to ensure that the agent will not be overloaded while executing queued tasks and currently assigned tasks: Where C j represents the maximum workload that agent j can bear.

2. The method according to claim 1, wherein The agent is equipped with a deep learning model based on a multi-layer perceptron. The agent performs the following operations to estimate the expected completion time of the assigned task: Analyze the task type and task resource requirements corresponding to the task to be assigned, and obtain the device performance indicators of the intelligent agent to construct multivariate input features; The multivariate input features are input into the deep learning model based on the multilayer perceptron to determine the expected completion time of the corresponding task.

3. The method according to claim 1 or 2, wherein: The agent is a warehouse agent, and the task is a warehouse operation task, and the corresponding task type includes any one of the following: receiving task, picking task and inventory counting task.

4. The method according to claim 1, wherein The global constraint optimization algorithm adopts the collaborative particle swarm optimization algorithm. Solving the constraint optimization model by a global constraint optimization algorithm and assigning each of the tasks to be assigned to a corresponding agent according to the optimal solution includes: Particle swarm initialization: Each particle is defined as a task allocation scheme with a dimension of M×N, where the dimension corresponds to the allocation relationship between each task and each agent; the particle position is defined using decision variables. It indicates whether task i is assigned to agent j in particle k; Under the premise of satisfying the task execution constraints, the number of agent queued tasks and the maximum load constraints of the agent, the initial position of the particle is randomly generated; the particle speed is used to generate the particle position. represents the change in the assignment of task i to agent j in particle k. The initial value of the particle's velocity is zero. The particle's velocity controls the change in the task assignment scheme. The particle optimizes task scheduling based on its velocity and position during each iteration. Particle swarm update based on neighborhood collaborative search: Particles not only rely on their own optimal solutions, but also use the information sharing mechanism of neighboring particles to accelerate the convergence of the global optimal solution through collaborative search; The particle velocity update formula is: Where, represents the particle velocity of particle k at the tth iteration, and represents the change when task i is assigned to agent j; w(t) is the dynamic inertia weight, which determines the balance between global search and local development of the particle and gradually decreases with the number of iterations; c1(t) and c2(t) represent the personal optimal solution acceleration constant and the global optimal solution acceleration constant, respectively; r1 and r2 are random numbers in the range [0, 1] used to increase the randomness of the search; is the personal optimal solution of particle k when task i is assigned to agent j in iteration t; is the task allocation scheme of particle k in the global optimal solution, which represents the optimal task allocation scheme among all particles; N(k) is the set of neighboring particles of particle k, and particles exchange information through collaborative search; l represents the neighboring particles of particle k; represents the particle position of particle k at the tth iteration, indicating whether task i is assigned to agent j; α is the influence coefficient of collaborative search, which controls the intensity of collaboration between particles; The particle position update formula is: Where σ(·) represents the Sigmoid activation function, which is used to convert particle velocity into probability to determine whether task i is assigned to agent j; r ij is a random number in the range [0, 1], used to control whether task i is assigned to agent j; After the particle position is updated, check whether the task allocation scheme corresponding to the updated particle satisfies the task execution constraint, the number of agent queued tasks constraint, and the maximum load constraint of the agent at the same time; if so, perform the iteration stop condition judgment; if not, reinitialize the particle velocity to zero; Among them, the update formula of dynamic inertia weight is: Where w max and w min is the maximum and minimum value of the inertia weight, T max is the maximum number of iterations; Among them, the update formula of the adaptive acceleration constant is: Where c min，1 and c max，1 They represent the minimum acceleration constant and the maximum acceleration constant of the individual optimal solution respectively; c min，2 and c max，2 They represent the minimum acceleration constant and the maximum acceleration constant of the global optimal solution respectively; In the iterative stop condition judgment, when it is detected that the number of iterations reaches the maximum number of iterations T max When the fitness convergence condition is met, the iteration is terminated, and the particle with the largest fitness in the current particle swarm is returned as the optimal solution, and each of the tasks to be assigned is assigned to the corresponding agent according to the optimal solution.

5. The method according to claim 4, wherein The fitness function of the collaborative particle swarm optimization algorithm is expressed as follows: F=F1+λF2, Where F represents the fitness function, F1 represents the optimization term for the total execution time of the task, F2 represents the penalty term for task timeout, and λ represents the fitness balance factor. t i，limit represents the maximum deadline of task i, which is determined according to a preset mapping relationship between task type and / or task priority and deadline.

6. A multi-agent distributed task collaboration system, comprising: The task set broadcast unit is used to broadcast the task set to be assigned to each agent, so that the agent can estimate the expected completion time of each task to be assigned and generate a corresponding expected completion time matrix of the task to be assigned; An estimated duration receiving unit is used to receive from each of the agents an expected duration matrix of tasks to be assigned and an estimated duration of tasks previously assigned; wherein, if the agent is in an idle state, the estimated duration of the queued task is 0; if the agent is in a running state, the estimated duration of tasks previously assigned is the sum of the expected completion durations of at least one task to be executed by the agent; A constraint optimization modeling unit is used to construct a constraint optimization model based on the expected duration matrix of the tasks to be assigned to each of the agents and the estimated duration of the historical assigned tasks; the constraint optimization model includes task execution constraints, agent queue task quantity constraints, agent maximum load constraints and objective function; A global constraint solving unit, configured to solve the constraint optimization model using a global constraint optimization algorithm, and assign each of the tasks to be assigned to a corresponding agent according to the optimal solution; The objective function is expressed as follows: Where M represents the total number of tasks in the task set to be assigned, N represents the total number of agents; x ij is a decision variable, indicating whether task i is assigned to agent j. If so, then x ij =1, otherwise x ij =0;t ij represents the time required for agent j to perform task i, which is determined according to the expected task duration distribution; is the total estimated time of the assigned tasks of agent j; The task execution constraints restrict each task to be assigned to one agent: The agent queue task quantity constraint is used to limit the upper limit of the number of tasks assigned to the agent to avoid agent task backlog: Where Q j is the threshold number of assigned tasks for agent j; The agent maximum load constraint is used to ensure that the agent will not be overloaded while executing queued tasks and currently assigned tasks: Where C j represents the maximum workload that agent j can bear.

Images