Intelligent agent task cooperative scheduling method based on HC-MOGA in industrial internet

Abstract

The invention provides an HC-MOGA-based agent task collaborative scheduling method in an industrial internet, and aims to solve the key problems of low scheduling efficiency, poor system robustness and the like caused by heterogeneous types of equipment (subsequent modeling is an agent), complex task dependence and non-uniform resource distribution in an industrial site. According to the method, a heterogeneous agent scheduling model with task identification, resource matching and dependent modeling capabilities is constructed for a multi-stage operation task cooperatively completed by a data acquisition agent and an operation type agent in an industrial internet environment. An HC-MOGA (heterogeneous constraint multi-objective genetic algorithm) is provided, multi-dimensional chromosome coding, a layered mapping mechanism, a dual-stage variation strategy and a dynamic constraint repair mechanism are introduced, and a dual-objective optimization model fusing work completion income and comprehensive resource consumption and potential loss is established. The method is suitable for industrial internet typical application scenes such as intelligent manufacturing, automatic production lines and multi-agent task collaborative management.

Description

Technical Field

[0001] The present invention relates to the field of industrial Internet, and in particular to an HC-MOGA-based intelligent agent task collaborative scheduling method in the industrial Internet. Background Art

[0002] Driven by the rapid evolution of the Industrial Internet, modern manufacturing systems are developing towards a highly intelligent, networked, and flexible approach. To meet diverse and customized market demands, enterprises have gradually introduced a large number of diverse production and operation units, including fixed and mobile data collection agents and agents with autonomous operation capabilities. These agents exhibit significant differences in functionality, performance, operational processes, and communication capabilities, creating a highly heterogeneous production environment. Simultaneously, the complexity of on-site operations is increasing, often characterized by multiple stages, multiple objectives, and multiple constraints, with strict execution sequences and resource competition between tasks. This staged, parallel, and highly interdependent nature presents significant challenges for traditional scheduling methods in practical applications.

[0003] First, traditional scheduling algorithms are typically based on assumptions of homogeneous agents or static resources. They lack precise modeling and dynamic response mechanisms for heterogeneous resource capabilities, making it difficult to effectively handle the uncertainty in task execution caused by differences in agents. Second, during task execution, due to the uneven distribution of resources (such as time, energy, and operators), traditional scheduling mechanisms cannot achieve optimal resource allocation, resulting in some agents being idle while others are overloaded, reducing overall operational efficiency. Furthermore, current industrial systems are increasingly demanding high robustness and adaptive scheduling. In particular, the ability of a scheduling system to quickly respond and maintain operational stability in the face of emergencies, task changes, or agent anomalies has become a crucial indicator of the system's intelligence level.

[0004] Therefore, developing a scheduling method for the industrial Internet environment that can simultaneously handle complex issues such as agent heterogeneity, task dependencies, and resource matching optimization has become a core technical challenge that needs to be solved urgently. This method should have the ability to intelligently identify task characteristics, reasonably allocate resources, dynamically adjust task order and paths, and optimize the overall system efficiency while taking into account both task completion quality and resource consumption costs. By adopting a multi-objective genetic algorithm with global search capabilities and strong adaptability, and introducing a multi-dimensional encoding strategy, a complex constraint processing mechanism, and a robustness-enhancing design, it will provide an efficient and scalable agent task collaborative scheduling solution for industrial systems, thereby playing a significant role in key areas such as intelligent manufacturing, automated production, and multi-agent linkage management, and promoting the intelligent development of the industrial Internet. Summary of the Invention

[0005] Technical issues: The heterogeneity of intelligent agents leads to uncertainty in task execution: Traditional scheduling algorithms are based on the assumption of homogeneous intelligent agents and lack the ability to differentiate modeling of heterogeneous intelligent agents such as data collection agents and operational agents in the industrial Internet. It is difficult to accurately match task requirements with intelligent agent capabilities, resulting in unreasonable task allocation and low execution efficiency.

[0006] Dynamic resource allocation and dependency conflicts: Industrial tasks have multi-stage dependencies (e.g., monitoring tasks must be executed before maintenance tasks). Traditional methods cannot effectively handle the dynamic competition between logical dependencies and resources (time, energy consumption, communication bandwidth) between tasks, which can easily lead to local resource overload or idleness, reducing the overall system utilization.

[0007] Multi-objective optimization and constraint conflicts: Existing scheduling mechanisms find it difficult to simultaneously optimize competing objectives such as task completion rate, system energy consumption, and task deadlines. Furthermore, they lack robust processing capabilities for complex constraints (such as the maximum energy consumption of intelligent agents and the upper limit of task processing capabilities), resulting in poor feasibility of scheduling solutions.

[0008] Insufficient dynamic response capabilities: In the face of emergencies (such as agent failures and task changes), traditional methods are unable to quickly adjust scheduling strategies, and the system's robustness and adaptability are insufficient, affecting production continuity.

[0009] Technical Solution: This invention proposes an agent task collaborative scheduling technology for the complex scheduling environment of the Industrial Internet, which integrates multi-objective optimization strategies and heterogeneous agent modeling methods. Specifically, it introduces a heterogeneous constrained multi-objective genetic algorithm (HCMOGA) to intelligently solve the agent scheduling scheme. This method focuses on solving the problems of low scheduling efficiency, high system overhead, and strong job uncertainty caused by heterogeneous agent types, stage-by-stage task dependencies, uneven distribution of resource capabilities, and scheduling goal conflicts in industrial scenarios. The technical solution mainly includes the following core steps and modules:

[0010] First, the mobile data collection devices and mobile operational equipment in the Industrial Internet are modeled as agents with different capabilities: data collection agents and operational agents. Specifically, during the scheduling model construction phase, the system heterogeneously models the mobile agents in the industrial field, dividing all agents into two categories: data collection agents, which can only perform monitoring tasks; and operational agents, which can perform both monitoring and operational tasks. Each agent's capabilities are described by attribute parameters such as its energy budget, response speed, maximum task load, and mobility cost. An agent-task adaptation table is then established for scheduling matching.

[0011] Secondly, to address the complexity of industrial tasks, the present invention divides system tasks into two atomic tasks: monitoring tasks and operation tasks. The execution order between tasks is clearly defined based on their stage-by-stage dependencies. Furthermore, the task attribute set also incorporates information such as task coordinate location, risk level, business value, and resource consumption intensity to support processes such as task prioritization, risk assessment, and agent matching. In terms of task scheduling coding design, to meet the stage-by-stage dependencies of task execution and achieve optimal resource allocation, the present invention designs a multidimensional hierarchical chromosome encoding structure. This employs a multidimensional chromosome encoding approach to hierarchically map tasks, ensuring that monitoring tasks are completed before operation tasks. Tasks and resources are then matched based on agent capabilities and task attributes. Specifically, the agent set and task set are systematically mapped to the chromosome gene space. A task type-based allocation mechanism is employed to ensure that monitoring tasks are assigned to data acquisition agents or operation agents, while operation tasks are assigned only to operation agents, meeting actual capability constraints. This chromosome structure supports distributed parsing, facilitating the rapid identification of task attribution and execution order in genetic operations.

[0012] At the algorithm solving level, in order to achieve the dual-objective optimization of maximizing the task completion rate and minimizing the system energy consumption, the present invention proposes a heterogeneous constrained multi-objective genetic algorithm (HCMOGA), whose main features include a two-stage gene mutation strategy and a task distribution repair mechanism. Under the multiple constraints of communication bandwidth, agent energy consumption and task deadline, the migration path and execution order of tasks are dynamically optimized to improve the overall scheduling efficiency and the balance of system resource utilization. Specifically: in the mutation operation, the algorithm first identifies the execution unit based on the chromosome task block structure, adopts differentiated target agent selection methods for different task types, and implements cross-agent task transfer and random re-arrangement of task sequences. And in the crossover operation, the "exchange block priority" deduplication mechanism and the missing task reconstruction mechanism are introduced to ensure the uniqueness and integrity of all tasks in the offspring chromosomes. In the actual industrial Internet scheduling scenario, the overall system benefits brought by completing all tasks must be comprehensively considered. At the same time, it's also necessary to consider the various operational costs incurred by the agent during task execution, including energy consumption, time delays, and the impact of high-risk operating environments on the agent's lifespan or performance stability. Therefore, to meet the multi-objective optimization requirements of industrial environments, this paper constructs a dual-objective optimization model that integrates task benefit and system loss. The task benefit objective function measures the total business value of task completion, reflecting the task's comprehensive contribution to system efficiency, agent stability, and product quality assurance. The cost objective function evaluates scheduling overhead from three dimensions: energy consumption, time delay, and task risk. This objective measures the comprehensive operational costs incurred by the agent during task execution, primarily including energy consumption, response latency, and the impact of operating in high-risk task areas on the agent's health. The algorithm aims to maximize task completion benefits while minimizing task scheduling costs, achieving an overall optimal balance between system benefit and agent load. To achieve a dynamic balance between these optimization objectives, the algorithm utilizes a Pareto non-dominated sorting mechanism to generate candidate solutions. This, combined with an elite retention and constraint violation penalty mechanism, effectively improves the convergence efficiency and feasibility of the optimal solution. In terms of constraint handling, the algorithm considers multiple scheduling rules that must be followed in real-world industrial scenarios, including energy consumption limits, agent capacity limitations, task dependencies, and task assignment uniqueness. By introducing a constraint repair strategy during the genetic iteration process, scheduling schemes that do not meet dependency order or resource requirements are dynamically adjusted, significantly enhancing the algorithm's stability and robustness under complex constraints.

[0013] Beneficial effects:

[0014] (1) Improving scheduling efficiency and task completion rate By adopting the heterogeneous constrained multi-objective genetic algorithm (HCMOGA), the task allocation and execution order among different types of agents are effectively coordinated, which significantly improves the overall scheduling efficiency and task completion rate in the multi-stage operation process.

[0015] (2) Supporting the modeling of heterogeneous agent capability differences. Taking into full account the heterogeneity of agents in industrial sites in terms of function, performance, energy consumption, and movement speed, a capability matching-driven scheduling model is established to achieve precise adaptation between tasks and agents, thus avoiding resource mismatch and scheduling imbalance.

[0016] (3) Enhance the robustness and adaptability of the scheduling system. By introducing a two-stage mutation mechanism and task allocation repair strategy, we can effectively deal with uncertainties such as sudden task changes, agent failures, or resource bottlenecks, and improve the stability and emergency response capabilities of the scheduling system. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a variation operation flow chart showing the intelligent agent task collaborative scheduling method of heterogeneous constrained multi-objective genetic algorithm in the industrial Internet.

[0018] Figure 2 It is a cross-operation flow chart showing the intelligent agent task collaborative scheduling method of heterogeneous constrained multi-objective genetic algorithm in the industrial Internet.

[0019] Figure 3 It is the main principle diagram of the method of the present invention. DETAILED DESCRIPTION

[0020] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention.

[0021] like Figure 3 As shown, the present invention provides an intelligent agent task collaborative scheduling method based on HC-MOGA in the industrial Internet.

[0022] (1) The intelligent agents of the Industrial Internet implement overall task scheduling based on the attributes of themselves and other intelligent agents and the cost of executing tasks, taking into account the dual optimization objectives of task completion benefits and risk losses, so as to improve the overall scheduling efficiency of the system and increase the task completion rate. In the initial stage, the intelligent agent model, task model, task benefit objective function model, consumption objective function model, and constraint model are defined. The model is described as follows:

[0023] Agent model definition: In the Industrial Internet agent task scheduling scenario, the system includes two heterogeneous agents: data collection agents and operational agents. Let m represent the total number of agents, including both types. The attribute set of each agent is defined as follows: Among them, (x i ,y i ) represents the initial deployment position of the agent, which is used to calculate the distance between the agent and the task point; V iRepresents the movement response speed of the agent, which affects its task switching and execution efficiency; E i Represents the energy consumption budget of the agent, which is used to limit the continuous working time; F i Represents the overhead of the agent's movement path; C i Represents the maximum job-carrying capacity of the agent. The carrying capacity of the data collection agent is defined as 0, and it can only perform monitoring tasks. The job-type agent has the composite ability to perform monitoring and job tasks; Represents the task assignment set, which defines the tasks that the agent needs to complete during the scheduling period and the specific order of task execution.

[0024] ● Task model definition: In the industrial Internet task scheduling scenario, let n represent the total number of tasks. The system tasks consist of two types: monitoring tasks and job tasks. Usually, the number of tasks is greater than the number of agents (m < n). The attribute set of each task is defined as follows: Among them, (x<00000​​​​​​​​​​​​​​​​​​​​​​​​​​​In the Industrial Internet task scheduling model, the maximum task carrying capacity of an agent can be compared to its available resource budget, while the resource intensity required by the task represents the specific amount of resources that must be consumed to complete the task. i Indicates whether the i-th task is successfully completed. If the task is successfully completed, there is TC i =1, otherwise TC i = 0. If and only if the agent's current remaining resource carrying capacity is greater than or equal to the resource intensity required for task i, the task is considered to be successfully completed and the complete business benefits corresponding to the task can be obtained. remain If the task requirements cannot be fully met and only partially supported, the task will be considered partially completed. In this case, the actual benefits of the task will be calculated based on the remaining resources UC of the agent. remain The amount of resources required for the task is converted into a proportion and multiplied by an unfinished penalty coefficient (set to 0.7) to reflect the reduction in benefits due to insufficient resources.

[0028] Consumption objective function model: In the industrial Internet task scheduling model, in order to accurately evaluate the comprehensive resource consumption and potential loss of the intelligent agent during the execution of the task, the model defines the following intelligent agent cluster consumption objective function (CostFunction), which mainly includes three aspects: energy consumption cost, time delay cost and operation risk cost. Energy consumption is mainly calculated based on the energy consumption characteristics of the intelligent agent in different operation stages. The intelligent agent usually goes through three stages during the task execution: standby stage, mobile stage, and operation execution stage. The energy consumption of each stage is different. The energy consumption per unit time of the intelligent agent in the standby stage is P wait , the energy consumption in the mobile stage is the agent's mobile path cost F i , the energy consumption per unit time during the job execution phase is P work , if the total standby time is T wait , the total operation time is T work , then the total energy consumption of the intelligent agent cluster is considered to be

[0029]

[0030] The time delay cost reflects the total time required for all agents to complete their tasks, especially the waiting and coordination costs when considering task dependencies. Consider the geometric distance d between two task points. ij , which can be obtained according to the Euclidean geometric distance formula

[0031]

[0032] The actual response distance of the agent can be obtained based on the task coordinates and the task assignment set of the UAV, that is,

[0033]

[0034] There is a front-to-back dependency between the monitoring task and the operation task. If the operation task to be performed by the agent depends on the completion of the previous monitoring, the agent needs to determine whether to wait. If the previous monitoring task is not completed, the agent needs to wait in place, increasing the waiting time. If the monitoring is completed, it can move and operate directly. The execution time of each task is different. Let the execution time of the monitoring task (by the data collection agent) be T detect The execution time of the monitoring task (temporarily executed by the task-based agent) is T detect-work , the formal task execution time is T work-task , the maximum time taken by all agents to complete all tasks is the overall time delay cost:

[0035] F time =max{T 作业完成}

[0036] Finally, consider the risk cost of the agent when performing tasks. When performing tasks in high-risk task areas, it may bring potential losses to the agent. The loss risk cost of the agent performing tasks is related to the following factors: the task risk level coefficient A i ,Task influence radius R i , Agent movement response speed V i , in order to facilitate calculation, the loss risk of the agent when performing the task is uniformly considered as The total risk function is designed as follows:

[0037]

[0038] In order to achieve unified optimization of multiple factors, a weighted combination of three costs is introduced to construct the overall loss objective function:

[0039] F c =α1F energy +α2F time +α3F risk

[0040] Constraint model: In an industrial context, agent task scheduling typically involves allocating and sequencing multiple tasks, subject to the following key constraints. First, the energy consumption limits of the agents must be met. The energy consumption of each agent is determined by the distance it moves, and each agent's energy consumption must not exceed its energy budget:

[0041]

[0042] Each industrial task must be executed and completed by one and only one agent to prevent redundant scheduling or task omissions:

[0043]

[0044] In addition, there are task dependency restrictions, that is, for a task objective, the corresponding operational task can only be executed after the monitoring task is fully completed.

[0045] (2) Chromosome encoding stage: The system contains m1 data acquisition agents and m2 operational agents, forming an agent cluster with a total size of m = m1 + m2. It needs to complete a two-stage task process (monitoring task → operational task) with n task points, involving a total of 2n atomic tasks. The encoding space is systematically divided into three functional discrete intervals: the first m1 gene numbers represent the data acquisition agent number set Γ s ={1,2,...,m1}, then m2 gene numbers correspond to the set of job-type agent numbers Γ a ={m1+1,...,m}. Task coding adopts a hierarchical mapping mechanism to define the monitoring task set T s ={m+1,...,m+n} and the task set T a ={m+n+1,...,m+2n}, where task pair (t i ,t {n+i} )∈T s ×T a The complete processing flow corresponding to the i-th target point. The chromosome parsing algorithm adopts a distributed task deconstruction strategy, firstly splitting the gene sequence: s ∪Γ a As separator, the chromosome is divided into k subsequences S1, S2, ..., S k , then the gene sequence should satisfy the task allocation constraints: (The data acquisition agent only performs monitoring tasks), (Work-type intelligent agents can perform monitoring tasks and work tasks).

[0046] (3) Mutation stage: Figure 1As shown in the figure, a two-stage mutation mechanism is used to optimize industrial task scheduling. This method first performs dynamic task block parsing and constructs an agent-task mapping table by linearly scanning chromosomes, accurately identifying the task execution sequence structure of each mobile agent. Subsequently, an adaptive gene mutation operation is implemented to randomly select a source agent and its task units. Differentiated agent selection strategies are implemented based on the task type: if the task is a monitoring task, a target agent is randomly selected from the set of data acquisition agents and task-based agents; if the task is a work task, only legitimate target agents are selected from the set of task-based agents. After determining the target agent, the system performs a cross-agent gene transfer operation, migrating the selected task to the target agent's task sequence. To further optimize the task scheduling structure, the algorithm applies the Fisher-Yates shuffle algorithm to the target agent's task sequence to generate a new permutation that satisfies the uniform distribution property, while keeping the task sequences of the remaining agents unchanged, thus completing the overall mutation process.

[0047] (4) Crossover stage: Figure 2 As shown in the figure, an innovative crossover deduplication and repair mechanism is designed. This mechanism first randomly selects a mobile agent and uses the task sequences corresponding to the agent in the two parent chromosomes as the exchange unit. Subsequently, the complete task sequence of the agent in parent 1 is exchanged with the corresponding sequence in parent 2. The exchanged chromosomes often produce duplicate task elements or missing tasks, so systematic deduplication and repair processing is required. This algorithm adopts a "swap block first" duplication elimination strategy: that is, in the daughter chromosome, the task sequences of other agents except the swap block are scanned and all instances with duplicate task IDs in the swap block are deleted. Subsequently, the system detects the missing task items in the current daughter chromosome based on the entire task set and randomly assigns them to data acquisition or operation-type agents with corresponding execution capabilities according to the task type (monitoring task or operation task), thus completing the crossover operation.

[0048] (5) Pareto sorting and elimination phase: To ensure that the solution generated by the algorithm meets the actual constraints of industrial task scheduling, a special constraint processing mechanism is introduced. During the iterative generation of chromosomes, if individuals violate the scheduling constraints, it mainly manifests as two types of problems: one is resource constraint violation, such as individual scheduling schemes exceeding the physical limitations of the mobile agent's maximum energy consumption, task processing capacity, etc.; the other is logical dependency violation, such as some tasks are scheduled in advance without completing the preceding monitoring tasks, which may cause system scheduling blockage or logical deadlock. In multi-objective optimization task scheduling, we often face the problem of optimizing multiple performance indicators at the same time. There is usually a significant competition and constraint relationship between these objectives, that is, optimizing a certain objective often leads to the performance degradation of other objectives. To deal with this conflict, a hierarchical non-dominated sorting method based on the Pareto dominance relationship is adopted to construct a dual-objective optimization model with the maximization of task benefits and the minimization of system operation losses as the core. For two different individuals (each individual has an allocation plan), the dominance relationship is only satisfied when the following relationship is satisfied:

[0049] F e (x i )≥F e (x j )and F c (x i )≤F c (x j )

[0050] A strict non-dominated sorting process is used to generate a multi-level candidate solution set. To maintain population quality and solution feasibility, a constraint violation evaluation mechanism is introduced in the selection phase, setting elimination priorities based on the degree of deviation of individuals from the constraints.

Claims

1. A method for collaborative scheduling of intelligent tasks based on HC-MOGA in the industrial Internet, characterized by the following steps: As follows: First, the mobile data acquisition devices and mobile operational devices in the industrial Internet are modeled as intelligent agents with different capabilities, namely data acquisition agents and operational agents. The data acquisition agents can only perform monitoring tasks, while the operational agents can perform both monitoring and operational tasks. Secondly, in order to meet the stage-by-stage dependency of task execution and achieve optimal resource allocation, a multi-dimensional chromosome encoding method is used to map tasks in layers to ensure that monitoring tasks are completed before operational tasks, and tasks and resources are matched according to the agent capabilities and task attributes. Finally, in order to achieve the dual-objective optimization of maximizing task completion rate and minimizing energy consumption, a two-stage gene mutation strategy and task distribution repair mechanism are introduced. Under the multiple constraints of communication bandwidth, agent energy consumption and task deadline, the migration path and execution order of tasks are dynamically optimized to improve the overall scheduling efficiency and the balance of resource utilization.

2. The method for collaborative task scheduling of intelligent agents based on HC-MOGA in the industrial Internet according to claim 1 is characterized by: In the data collection agent and the task-oriented agent, let m represent the total number of agents, including two types of agents; the attribute set of each agent is defined as follows: Among them, (x i ,y i ) represents the initial deployment position of the agent, which is used to calculate the distance between the agent and the task point; V i Indicates the mobile response speed of the agent, which affects its task switching and execution efficiency; E i Represents the energy consumption budget of the intelligent agent, which is used to limit the continuous working time; F i represents the agent's moving path cost; C i Indicates the maximum operation carrying capacity of the agent. The carrying capacity of the data acquisition agent is defined as 0, and it can only perform monitoring tasks. The operation-type agent has the combined ability to perform monitoring and operation tasks; It represents a set of task assignments, defining the tasks that the agent needs to complete within the scheduling period and the specific order in which the tasks are executed.

3. The method for collaborative task scheduling of intelligent agents based on HC-MOGA in the industrial Internet according to claim 2 is characterized by: In the monitoring tasks and operation tasks, let \(n\) represent the total number of tasks, and the number of tasks is greater than the number of agents, \(m < n\); the attribute set of each task is defined as follows: Among them, \((x Ti , y Ti ) represents the physical coordinate position of the task, which is used to calculate the scheduling distance from the agent; \(A i represents the risk level coefficient of the task, which is used to measure the urgency of task execution and the possible scheduling uncertainty caused; \(R i represents the task influence radius, which reflects the degree of restriction of the task on the agent scheduling ability in space; \(P i represents the business value of the task, which is used to establish the priority sorting strategy in scheduling; \(CR i represents the resource consumption intensity required for the task, which is used to control the complexity of task execution and the demand for agent capabilities, and is used as the key basis for agent selection and task matching; The decision variable is a binary variable, indicating whether a certain task is assigned to an industrial agent; specifically, when , it means that the \(j\)th task is successfully assigned to the \(i\)th agent; while when it means that the task is not assigned to this agent; Corresponding decision variables are set for the monitoring tasks and operation tasks and respectively to represent the agent assignment status of the monitoring tasks and operation tasks.

4. The method for collaborative task scheduling of intelligent agents based on HC-MOGA in the industrial Internet according to claim 3 is characterized by: Two optimization objective functions are defined to maximize the task completion benefit while minimizing the task scheduling cost, including the task benefit objective function and the consumption objective function. The task benefit objective function is used to measure the total business value brought by completing various tasks within the scheduling cycle. The consumption objective function is used to measure the comprehensive operating cost of the agent during the task execution, mainly including energy consumption, response delay cost, and the impact of working in high-risk task areas on the health status of the agent. It aims to maximize the task completion benefit while minimizing the task scheduling cost, so as to achieve an overall optimization balance between benefit and agent load.

5. The method for collaborative task scheduling of intelligent agents based on HC-MOGA in the industrial Internet according to claim 4 is characterized by: In the task benefit objective function, when calculating task benefits, only benefits are taken into account when the task is successfully completed. The benefits of monitoring tasks are not taken into account. The overall benefits are determined based on the completed tasks: Among them, TC i Indicates whether the i-th task is successfully completed. If the task is successfully completed, there is TC i =1, otherwise TC i = 0; if and only if the agent's current remaining resource carrying capacity is greater than or equal to the resource intensity required for task i, the task is considered to be successfully completed and the complete business benefits corresponding to the task can be obtained; if the agent's remaining resources UC when executing the task are ... remain If the task requirements cannot be fully met and only partially support the task execution, the task is considered partially completed; in this case, the actual benefits of the task will be calculated based on the remaining resources UC of the agent. remain The amount is converted into the proportion of the resources required for the task and multiplied by an unfinished penalty coefficient of 0.7 to reflect the reduction in benefits due to insufficient resources.

6. The method for collaborative task scheduling of intelligent agents based on HC-MOGA in the industrial Internet according to claim 5, characterized in that: In the consumption objective function, energy consumption is comprehensively calculated based on the energy consumption characteristics of the agent in different operation stages. The agent goes through three stages during task execution: standby stage, mobile stage, and operation execution stage. The energy consumption of each stage is different. The energy consumption per unit time of the agent in the standby stage is P wait , the energy consumption in the mobile stage is the agent's mobile path cost F i , the energy consumption per unit time during the job execution phase is P work , if the total standby time is T wait , the total operation time is T work , then the total energy consumption of the intelligent agent cluster is considered to be The time delay cost reflects the total time required for all agents to complete their tasks, especially the waiting and coordination costs under the condition of task dependencies; considering the geometric distance d between two task points ij , according to the Euclidean geometric distance formula, we get The actual response distance of the agent is obtained according to the task coordinates and the task assignment set of the UAV, that is, There is a front-to-back dependency between the monitoring task and the operation task. If the operation task to be performed by the agent depends on the completion of the previous monitoring, the agent needs to determine whether to wait; if the previous monitoring task is not completed, the agent needs to wait in place, increasing the waiting time. If the monitoring task is completed, it can move and operate directly; the execution time of each task is different. Let the execution time of the monitoring task by the data acquisition agent be T detect The monitoring task is temporarily executed by the task agent for a time period of T detect-work , the formal task execution time is T work-task , the maximum time taken by all agents to complete all tasks is the overall time delay cost: F time =max{T 作业完成 } Finally, consider the risk cost of the agent when performing tasks. When performing tasks in high-risk task areas, it may bring potential losses to the agent. The loss risk cost of the agent performing tasks is related to the following factors: the task risk level coefficient A i ,Task influence radius R i , Agent movement response speed V i , in order to facilitate calculation, the loss risk of the agent when performing the task is uniformly considered as The total risk function is designed as follows: In order to achieve unified optimization of multiple factors, a weighted combination of three costs is introduced to construct the overall loss objective function: F c =α1F energy +α2F time +α3F risk 。 7. The method for collaborative task scheduling of intelligent agents based on HC-MOGA in the industrial Internet according to claim 6, characterized in that: Agent scheduling involves the allocation and sequencing of multiple tasks, subject to the following constraints. First, the energy consumption limits of the agents must be met. The energy consumption of each agent is determined by the distance it moves, and each agent's energy consumption must not exceed its energy budget: Each industrial task must be executed and completed by one and only one agent to prevent redundant scheduling or task omissions: In addition, there are task dependency restrictions, that is, for a task objective, the corresponding operational task can only be executed after the monitoring task is fully completed.

8. The method for collaborative task scheduling of intelligent agents based on HC-MOGA in the industrial Internet according to claim 7, characterized in that: In the heterogeneous constrained multi-objective genetic algorithm, chromosomes are encoded to represent the scheduling scheme of agents and tasks, a fitness function is designed to evaluate the quality of the scheduling solution, and an elite retention strategy is introduced to preserve the non-dominated solution set. The algorithm generates a new population through crossover and mutation operations, and combines a special constraint processing mechanism to effectively deal with the limitations brought by the ability differences of heterogeneous agents, balance the conflicts and trade-offs between multiple objectives, and proposes a new multi-dimensional chromosome encoding structure and its parsing method. The agents include m1 data acquisition agents and m2 operation-type agents, forming an agent cluster with a total size of m = m1 + m2, which needs to complete a two-stage task process of n task points, i.e., monitoring task → operation task, involving a total of 2n atomic tasks.

9. The method for collaborative task scheduling of intelligent agents based on HC-MOGA in the industrial Internet according to claim 8, characterized in that: The encoding space is systematically divided into three functional discrete intervals: the first m1 gene numbers represent the set of data collection agent numbers Γ s ={1,2,...,m1}, then m2 gene numbers correspond to the set of job-type agent numbers Γ a ={m1+1,...,m}; Task coding adopts a hierarchical mapping mechanism to define the monitoring task set T s ={m+1,...,m+n} and the task set T a ={m+n+1,...,m+2n}, where task pair (t i ,t {n+i} )∈T s ×T a The complete processing flow corresponding to the i-th target point; the chromosome parsing algorithm adopts a distributed task deconstruction strategy, first splitting the gene sequence: with agent ID∈Γ s ∪Γ a As separator, the chromosome is divided into k subsequences S1, S2, ..., S k , then the gene sequence should satisfy the task allocation constraints: The data collection agent only performs monitoring tasks; Operational intelligent agents can perform monitoring tasks and operational tasks.

10. The method for collaborative task scheduling of intelligent agents based on HC-MOGA in the industrial Internet according to claim 9, characterized in that: The mutation operation is specifically as follows: first, dynamic task block parsing is performed, and an agent-task mapping table is constructed by linearly scanning chromosomes to accurately identify the task execution sequence structure of each mobile agent; Subsequently, an adaptive gene mutation operation is implemented to randomly select source agents and their task units, and a differentiated agent selection strategy is implemented according to the task type: if the task is a monitoring task, the target agent is randomly selected from the set of data collection agents and task-oriented agents; if the task is a task, only the legal target agent is selected from the set of task-oriented agents; After determining the target agent, perform the gene transfer operation across agents to transfer the selected tasks to the task sequence of the target agent; To further optimize the task scheduling structure, the Fisher-Yates shuffling algorithm is applied to the task sequence of the target agent to generate a new arrangement that satisfies the uniform distribution property, while keeping the task sequences of the remaining agents unchanged, thus completing the overall mutation process; The crossover operation is as follows: First, an agent is randomly selected and its corresponding task sequences in the two parent chromosomes are used as the exchange unit. Then, the complete task sequence of the agent in parent 1 is exchanged with the corresponding sequence in parent 2. After the exchange, a "swap block first" duplication elimination strategy is adopted: that is, in the daughter chromosome, the task sequences of other agents except the swap block are scanned and all instances with duplicate task IDs in the swap block are deleted. Then, based on the entire task set, the missing task items in the current daughter chromosome are detected and randomly assigned to data acquisition or operation-type agents with corresponding execution capabilities according to the task type, thus completing the crossover operation. The constraint handling mechanism during the crossover and mutation process is as follows: During the iterative generation of chromosomes, if individuals violate scheduling constraints, this can manifest as two types of problems: first, resource constraint violations, including individual scheduling schemes that exceed the maximum energy consumption of the agent or the physical limitations of task processing capabilities; second, logical dependency violations, including certain monitoring tasks being scheduled in advance without completing their predecessor tasks, which may lead to system scheduling blockage or logical deadlock. In multi-objective optimization task scheduling, a hierarchical non-dominated sorting method based on the Pareto dominance relationship is adopted to construct a dual-objective optimization model centered on maximizing task benefits and minimizing system operation losses. For two different individuals, the dominance relationship is only satisfied when the following relationship is met: F e (x i )≥F e (x j )and Fc (x i )≤F c (x j ) A multi-level candidate solution set is generated through a strict non-dominated sorting process. In order to maintain the population quality and the feasibility of the solution, a constraint violation degree evaluation mechanism is introduced in the selection stage, and the elimination priority is set according to the degree of deviation of individuals from the constraints.

Images