Power multi-agent dynamic cooperative inspection method and system

Abstract

The invention belongs to the technical field of intelligent electric power inspection and multi-agent cooperative control, and provides an electric power multi-agent dynamic cooperative inspection method and system, and the technical scheme is that a path planning scheme is generated by using a reinforcement learning algorithm and a large language model, so that an agent can dynamically adjust an inspection path according to real-time data and task requirements, and the inspection efficiency is improved. Therefore, the minimization of the path and the maximization of the inspection efficiency are realized, and unnecessary travel and resource consumption are reduced. Meanwhile, a dynamic task redistribution mechanism in the system ensures that the intelligent agent can quickly respond and redistribute tasks under the conditions of equipment failure, environment change or task priority adjustment. And real-time communication and data synchronization among the intelligent agents ensure information sharing, so that the system still keeps task continuity and reliability in emergency.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent power inspection and multi-agent collaborative control, and in particular to a power multi-agent dynamic collaborative inspection method and system. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] In the inspection tasks of modern power systems, path planning and emergency dispatch are key links to ensure the safety of equipment operation and the efficiency of maintenance. Existing inspection methods mostly rely on manual or single equipment operation. Although they can meet basic inspection needs, they have obvious deficiencies in task complexity and flexibility. Especially in the face of multi-device and multi-task collaborative inspections, traditional path planning methods are difficult to effectively adapt to the complex and changing power inspection environment. Most existing methods lack the ability to dynamically coordinate and optimize paths between multiple devices. At the same time, they often fail to respond in a timely manner when dealing with emergencies and cannot meet the high reliability requirements of power inspection tasks. Summary of the invention

[0004] In order to solve at least one technical problem existing in the above-mentioned background technology, the present invention provides a method and system for dynamic collaborative inspection of electric power multi-agents, which realizes the path optimization and emergency dispatch of inspection tasks through the collaborative work of inspection equipment such as multiple unmanned aerial vehicles and ground robots. The method dynamically adjusts the path planning of inspection equipment through information sharing and task allocation among multiple agents, and automatically performs task redistribution when equipment failure or task change occurs, thereby ensuring the continuity and efficiency of inspection tasks.

[0005] In order to achieve the above object, the present invention adopts the following technical solution: The first aspect of the present invention provides a power multi-agent dynamic collaborative inspection method, comprising the following steps: Obtain natural language description of power inspection tasks; Parse the natural language description of power inspection tasks and extract the constraints of power inspection goals and path planning; Based on the power inspection objectives and path planning constraints, the corresponding task areas are assigned to each agent according to its characteristics, and the initial inspection path plan for each agent is generated; When new variables or emergencies appear in the environment, the reward function of the state-action pair is combined with the quantitative indicators of path inspection efficiency and path conflict risk to evaluate the adaptability of the current inspection path; When the adaptability evaluation result indicates that the inspection path of each agent needs to be adjusted, the agent gradually selects the path plan that maximizes the reward function until the total expected return is met; During the adjustment process, the working status data of each intelligent body is obtained, and the optimal takeover plan is determined based on the working status data of each intelligent body to obtain the adjusted inspection path plan of each intelligent body.

[0006] Furthermore, the power inspection objectives include minimizing the path length and maximizing the inspection area coverage; the path planning constraints include: first, each inspection point can only be visited once by an agent; second, each agent must complete its assigned inspection task, with the starting point and the end point being the same location.

[0007] Furthermore, the path inspection efficiency score : , Path conflict risk : , in, is the inspection coverage rate, that is, the percentage of inspection points covered. is the ratio of task completion time to the maximum allowed time, is the ratio of the inspection path length to the shortest path length, , and The range is 0 to 1, , , They are the importance weights of coverage, time and path length, and their values ​​can be set according to actual needs; Representing an Agent i and Agent j The path overlap distance, It is an intelligent agent i and j The greater the distance, the smaller the overlap risk. Representing an Agent i and j The probability of inspecting the same area at the same time.

[0008] Furthermore, the reward function for a state-action pair is calculated as: , , in, Indicates the current time t state Take Action The value of is the learning rate, Status Next action It reflects the actual benefit of the agent in the current task assignment step. is a discount factor between 0 and 1, which is used to balance the relationship between current rewards and future rewards. is in the next state Take the next step The value of is in the next state The maximum expected value that can be obtained by choosing the optimal action to achieve this, λ is the weight of path conflict risk.

[0009] Furthermore, the working status data of each intelligent agent is obtained, and the optimal takeover plan is determined based on the working status data of each intelligent agent, including: Obtain the working status data of each intelligent agent; Generate multiple possible emergency plans based on the acquired working status data of each intelligent agent; Score multiple possible emergency plans, select the plan with the highest score as the optimal emergency plan, and send the emergency plan to the corresponding agent to take over the task of the faulty equipment.

[0010] Furthermore, the multiple possible emergency plans are scored and the plan with the highest score is selected as the optimal emergency plan, including: Analyze the urgency of the faulty equipment and obtain an urgency assessment score; Calculate the state weight of each agent based on the working state data of each agent; The comprehensive score of the emergency plan is obtained by combining the evaluation score of the urgency, the state weight of each agent and the weighted sum of proximity.

[0011] Furthermore, after generating the preliminary path plan, LLM assigns tasks to each agent and allocates task areas according to the characteristics of the agent. When allocating task areas, large task areas are divided into multiple sub-areas and independent path plans are generated for each sub-area. These assigned tasks are converted into multiple sub-paths, each of which contains inspection points that the agent needs to visit in sequence.

[0012] A second aspect of the present invention provides a power multi-agent dynamic collaborative inspection system, comprising: Inspection task acquisition module, used to obtain natural language description of power inspection tasks; Inspection task parsing module, used to parse the natural language description of power inspection tasks and extract power inspection targets and path planning constraints; The initial inspection plan generation module is used to generate the initial inspection path plan for each intelligent agent based on the power inspection objectives and path planning constraints, and assign corresponding task areas to each intelligent agent according to its characteristics; The adaptability evaluation module is used to evaluate the adaptability of the current inspection path by using the reward function of the state-action pair combined with the path inspection efficiency and path conflict risk quantitative indicators when new variables or emergencies appear in the environment; Inspection plan adjustment module, which is used for the agents to gradually select the path plan that maximizes the reward function when the adaptability evaluation result indicates that the inspection path of each agent needs to be adjusted until the total expected return is met; During the adjustment process, the working status data of each intelligent body is obtained, and the optimal takeover plan is determined based on the working status data of each intelligent body to obtain the adjusted inspection path plan of each intelligent body.

[0013] A third aspect of the present invention provides a computer-readable storage medium.

[0014] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps in the dynamic collaborative inspection of multiple electric power agents as described above.

[0015] A fourth aspect of the present invention provides a computer device.

[0016] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in the dynamic collaborative inspection of electric power multi-agents as described above are implemented.

[0017] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention innovatively proposes a dynamic collaborative inspection method for multiple power agents, and uses a path planning solution generated by a reinforcement learning algorithm and a large language model to enable the agent to dynamically adjust the inspection path according to real-time data and task requirements, thereby minimizing the path and maximizing the inspection efficiency, reducing unnecessary travel and resource consumption. At the same time, the dynamic task reallocation mechanism in the system ensures that the agent can respond quickly and reallocate tasks in the event of equipment failure, environmental changes, or task priority adjustments. It solves the problem that the existing inspection methods are not responsive in a timely manner and cannot meet high reliability requirements, meets real-time communication and data synchronization between agents, ensures information sharing, and enables the system to maintain task continuity and reliability in emergencies.

[0018] 2. The multi-agent system of the present invention can also sense and adapt to environmental factors such as obstacles and weather changes encountered during the inspection process, and make adaptive adjustments. Each agent adjusts the path according to the data it perceives, and avoids path conflicts and repeated coverage by sharing information, thus achieving optimal inspection coverage.

[0019] 3. The dynamic adjustment and real-time communication driven by reinforcement learning between the intelligent agents of the present invention bring about efficient collaboration and ensure the coordinated completion of task allocation and path planning among multiple intelligent agents. This collaborative working mode enables the system to achieve rapid coverage in a wide range of inspection areas and significantly improves the overall inspection efficiency. Through the task allocation and task takeover mechanism, when an intelligent agent fails or the task requirements change, other intelligent agents can quickly take over the task, thereby ensuring the continuity of the inspection task. The emergency response capability of the system is greatly improved, ensuring the smooth progress of power inspection tasks in complex and dynamic environments.

[0020] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0022] Figure 1 It is a flow chart of the power multi-agent dynamic collaborative inspection method provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0023] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0024] It should be noted that the following detailed descriptions are all illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0025] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0026] The existing methods mentioned in the background technology lack the ability of dynamic coordination and path optimization among multiple devices. At the same time, they often fail to respond in time when dealing with emergencies and cannot meet the high reliability requirements of power inspection tasks.

[0027] The present invention realizes the efficiency, flexibility and reliability of power inspection tasks through dynamic collaboration and path optimization of multiple agents. By using the path planning scheme generated by the reinforcement learning algorithm and the large language model, the agent can dynamically adjust the inspection path according to the real-time data and task requirements, thereby minimizing the path and maximizing the inspection efficiency, reducing unnecessary travel and resource consumption. At the same time, the dynamic task reallocation mechanism ensures that the agent can respond quickly and reallocate tasks in the event of equipment failure, environmental changes or task priority adjustments. Real-time communication and data synchronization between agents ensure information sharing, so that the system can still maintain the consistency and reliability of tasks in emergencies. The multi-agent system can also perceive and adapt to environmental factors such as obstacles and weather changes encountered during the inspection process, and make adaptive adjustments. Each agent adjusts the path according to the data it perceives, and avoids path conflicts and repeated coverage by sharing information, thereby achieving optimal inspection coverage.

[0028] The dynamic adjustment and real-time communication driven by reinforcement learning between agents bring about efficient collaboration and ensure the coordinated completion of task allocation and path planning among multiple agents. This collaborative working mode enables the system to achieve rapid coverage in a wide range of inspection areas and significantly improves the overall inspection efficiency. Through the task allocation and task takeover mechanism, when an agent fails or the task requirements change, other agents can quickly take over the task, thereby ensuring the continuity of the inspection task. The emergency response capability of the system is greatly improved, ensuring the smooth progress of power inspection tasks in complex and dynamic environments. In general, the present invention enables the multi-agent power inspection system to complete tasks efficiently and flexibly, and has a strong ability to cope with dynamic environments and emergencies, thereby greatly improving the automation level and task reliability of power inspections.

[0029] Embodiment 1 like Figure 1 As shown, this embodiment provides a power multi-agent dynamic collaborative inspection method, including the following steps: Step 1: Obtain a natural language description of the power inspection task; In this embodiment, the natural language description includes that the description generally includes the overall goal of the inspection task, the specific area to be covered, the task priority and other possible task requirements, etc. For example, "check all devices in a specific area to ensure maximum coverage and minimize the total path length." Step 2: Convert the natural language description of the power inspection task into a path planning model; In order for multi-agent systems to understand and perform these tasks, these natural language descriptions need to be converted into explicit mathematical form.

[0030] In this embodiment, the natural language description is parsed by a large language model (LLM) to extract the inspection objectives and constraints therein; Among them, the inspection objectives in the task are first identified, including key elements and task requirements, such as which areas need to be inspected, the starting and ending positions, the priority of the inspection task and the capacity limitation of the equipment, etc., and the extracted inspection task is converted into a path planning problem. In this embodiment, the inspection objectives include optimization goals such as minimizing the path length and maximizing the inspection area coverage.

[0031] For each inspection point, in order to avoid path duplication and coverage conflicts, additional constraints are extracted, such as which specific equipment or areas need to be inspected first, and other specific details involving equipment scheduling to ensure the reasonable execution of the task.

[0032] For multi-agent systems, it may also include collaboration between multiple devices, such as using the path planning model of the traveling salesman problem (TSP) or the vehicle routing problem (CVRP) to ensure that each inspection device can complete the task in collaboration.

[0033] In order to improve LLM's understanding of the path planning problem, in addition to generating mathematical formulas for the objective function and constraints, relevant pseudocode and references are also obtained; the pseudocode shows the basic steps of the path planning algorithm, such as how to allocate tasks between inspection devices, how to ensure path connectivity, and how to avoid obstacles. The content of the references usually contains solutions or heuristic strategies for similar tasks, which can provide more guidance and support for LLM's path planning generation process. After all the task descriptions, mathematical formulas, pseudocodes, and literature information are integrated, the system inputs this information into LLM to generate a preliminary path planning solution. The preliminary solution contains the initial configuration of the inspection path and assigns different inspection points to the corresponding agents according to the task requirements.

[0034] This scheme provides the basis for the subsequent self-debugging and self-verification mechanisms. Through these processes, the system can ensure that the initial planning scheme of the inspection task conforms to the task description, and gradually adjust it in the subsequent steps to achieve the optimal execution effect of the task. This series of processing steps enables the natural language description of the power inspection task to be effectively converted into a specific path planning model, and complete complex inspection tasks through multi-agent collaboration.

[0035] Step 3: Generate a preliminary path planning solution for each agent based on the constructed path planning model; This plan includes the order of visits to each inspection point and the initial itinerary. To ensure inspection efficiency, the system assigns different task areas according to the type of agent (such as drone or ground robot). Each agent's path plan will follow the goal of minimizing the itinerary to reduce path length and energy consumption while maximizing coverage of the designated area.

[0036] The path planning model generated by the large language model (LLM) after parsing the task description usually includes the objectives of the inspection points, the minimization of the path length, the capacity constraints of each agent, the time window, and the task priority. For example, assuming that the inspection task involves a series of location points V={v1,v2,...,vn}, the model defines these location points as targets that must be covered in sequence, and sets relevant formulas based on the path minimization objective.

[0037] After generating the mathematical model, LLM applies the path planning algorithm to preliminarily determine the order of visiting each inspection point. LLM regards the agent as a "vehicle" and the inspection point as a "customer", setting the agent to start from the starting point, visit the specified inspection point, and finally return to the starting point. To this end, LLM generates the visit order of each agent based on the distance matrix of the inspection points (such as the distance from each inspection point to other inspection points) and the priority of each point.

[0038] The power inspection objectives include minimizing the path length and maximizing the inspection area coverage; Among them, the minimization of path length can be modeled as a multi-traveling salesman problem (m-TSP), and the following formula is used to minimize the total inspection path length: , in, Representing an Agent i Whether the inspection point has been visited j (A value of 1 indicates access, and a value of 0 indicates no access), For intelligent agents i To the inspection point j The goal is to minimize the path sum of all agents.

[0039] Among them, the calculation formula for maximizing the inspection area coverage is: Assume that there is an area A, which is divided into several inspection points V={v1,v2,…,vn}. The goal of maximizing coverage can be expressed as: , in, Indicates inspection point Whether the state variable is overwritten, if so, ,otherwise The goal is to maximize all to ensure that as much area as possible is covered. LLM uses a path optimization algorithm to generate an initial itinerary that includes the inspection points.

[0040] The algorithm preliminarily determines the direction of the inspection path by evaluating the connectivity and distance of each inspection point. For example, LLM may use a heuristic algorithm (such as the nearest neighbor algorithm) to preferentially visit inspection points that are closer.

[0041] The constraints for path planning include: Each inspection point can only be visited once by an agent: , in, is the set of all inspection points; (2) Each agent must complete its assigned inspection task, with the starting and ending points being the same location: , Among them, M is the set of all agents.

[0042] In this way, LLM can quickly determine the initial order of inspection paths when generating path plans so as to minimize the total path.

[0043] After generating the preliminary path plan, LLM assigns tasks to each agent and allocates appropriate task areas according to the characteristics of the agent; In this embodiment, the agents include various types, such as drones, ground robots, etc.; The corresponding task area can be assigned according to the flight speed of the drone and the load capacity of the ground robot; for example, drones are suitable for quickly completing inspections of large areas due to their high maneuverability and wide coverage, especially in complex terrain and high-altitude scenes, while ground robots are more suitable for detailed inspection tasks, such as equipment inspections in substations. In addition, the system will further optimize the speed, endurance and environmental adaptability of the intelligent agent to ensure that intelligent agents with long endurance complete longer tasks, while intelligent agents with fast speed but short endurance are responsible for short-distance and high-priority tasks. The intelligent agent will also assign special detection tasks, such as infrared imaging or ultrasonic detection, according to task requirements and load capacity to ensure the accuracy of the inspection.

[0044] When assigning task areas, for large task areas, they are divided into multiple sub-areas, and independent path plans are generated for each sub-area. These assigned tasks are converted into multiple "sub-paths", each of which contains inspection points that the agent needs to visit in sequence. Finally, LLM optimizes the generated path plan as a whole to ensure that the tasks of all agents can reach the target with the shortest journey. The system will further correct unreasonable parts of the path plan, such as avoiding repeated coverage or optimizing the connection order of path nodes. After completing the path optimization, LLM will output the initial travel route of each agent to ensure that it can visit each inspection point in sequence in the order generated and finally return to the starting point.

[0045] Step 4: Each agent starts the task according to the initially generated path plan. When new variables or emergencies appear in the environment, the reward function of the state-action pair is combined with the path inspection efficiency and path conflict risk quantitative indicators to evaluate the adaptability of the current path; In reinforcement learning-driven path adjustment, the intelligent agent will use reinforcement learning algorithms to dynamically adjust the path during task execution. This adjustment mechanism relies on real-time environmental feedback (such as obstacles, changes in task priorities, etc.) to ensure the optimality of the inspection path and effective coverage of the task.

[0046] The specific steps include: Step 401: When a new variable or emergency occurs in the environment, the path inspection efficiency and path conflict risk of the current path are calculated; Among them, the path inspection efficiency score : , in, is the inspection coverage rate, that is, the percentage of inspection points covered. is the ratio of task completion time to the maximum allowed time, is the ratio of the inspection path length to the shortest path length, , and The range is 0 to 1, , , They are the importance weights of coverage, time and path length, and their values ​​can be set according to actual needs.

[0047] Path conflict risk : , in, Representing an Agent i and Agent j The path overlap distance, It is an intelligent agenti and j The greater the distance, the smaller the overlap risk. Representing an Agent i and j The probability of inspecting the same area at the same time; Combined with the efficiency of path inspection and path conflict risk Calculate the comprehensive score of the current path plan , , where λ is the weight of the path conflict risk, which is used to balance the weight between efficiency and conflict risk. The higher the final comprehensive score S is, the better the path solution performs in terms of efficiency and conflict, thus selecting a path with higher efficiency and less conflict; Step 402: Based on the reward function of state-action pairs, the quality of the current path is quantitatively evaluated in combination with the path inspection efficiency and path conflict risk of the current path to obtain an evaluation result; When new variables or emergencies appear in the environment (such as new obstacles or high-priority task points), the agent updates the path selection based on the value function in the reinforcement learning algorithm to maximize the inspection efficiency and coverage. The reinforcement learning algorithm evaluates the optimality of the path through the state-action value function, as shown in the following formula: , , in, Indicates the current time t state Take Action The value of is the learning rate, Status Next action It reflects the actual benefits of the agent in the current task assignment step, such as inspection coverage, path efficiency, or task completion. is a discount factor between 0 and 1, which is used to balance the relationship between current rewards and future rewards. is in the next state Take the next step The value of is in the next state The maximum expected value that can be obtained by choosing the optimal action to achieve; This maximization step affects the current moment through the optimal choice in the future The update of makes the agent pay attention not only to the current reward at each step, but also considers the potential maximum reward in the future.

[0048] Step 403: Combine the inspection efficiency and conflict risk of the current path with the reward function of the state-action pair to obtain an adaptability evaluation result, and determine whether the inspection path of the agent needs to be adjusted.

[0049] If the comprehensive score S and the value function If it is greater than the set threshold, the current path is efficient and has low conflict, and the current inspection path is followed; If S or If the value is lower than the set threshold, the agent returns to step 401 to re-evaluate the task allocation based on the current environment state, and the agent gradually selects the task that can maximize the The path plan is then developed until the total expected return is met, ensuring the efficiency and adaptability of the task.

[0050] The total expected return is calculated as: , in, represents the total expected return based on the strategy π, according to the state of the environment and Action To determine the overall task allocation of the agent and ensure that the inspection task is carried out uninterruptedly. It directly affects the decision-making at each step, making the agent tend to choose the allocation plan that has higher benefits for the current task goal.

[0051] Step 5: During the dynamic path adjustment process, the working status data of each intelligent agent is obtained, and the optimal emergency plan is determined based on the working status data of each intelligent agent; When an agent detects an obstacle in the environment, it will pass the relevant information to other agents in real time. Other agents can then adjust their own paths based on the shared data to avoid potential path overlaps and repeated coverage. During execution, the agent updates the strategy based on the inspection feedback. After each adjustment step, the agent will evaluate the effect of the new path, such as path coverage, path adjustment frequency, path length minimization, conflict rate, task completion time, and energy efficiency. If the new path can cover the task area more effectively, the adjustment will be retained; otherwise, the agent will try other path optimizations. This process ensures that the agent can achieve efficient and dynamic path adjustment during inspections, improve inspection coverage, and improve the ability to respond to emergencies.

[0052] The specific steps include: Step 501: When an agent detects a change in the environment or an emergency occurs, the working status data of each agent is obtained; In this embodiment, when the agent detects changes in the environment or an emergency occurs (such as equipment failure), multiple agents share data through wireless networks to achieve information synchronization. Each agent can exchange data (such as current location, remaining power, current task progress, etc.) with each other to ensure the integrity of the path adjustment; Step 502: Generate multiple possible emergency plans based on the acquired working status data of each intelligent agent; For example, in one scenario, the closest agent may take over the task directly, while in another scenario, an agent with sufficient power but farther away is assigned to take over the task.

[0053] Step 503: Score multiple possible emergency plans, select the plan with the highest score as the optimal emergency plan, and send the emergency plan to the corresponding intelligent agent to take over the task of the faulty equipment.

[0054] Specifically include: Step 5031: Analyze the urgency of the faulty equipment to obtain an urgency assessment score; , Among them, for serious failures of core equipment, P is close to the upper limit value, while for minor failures, P is lower.

[0055] Step 50232: Calculate the state weight of each agent based on the working state data of each agent to indicate its suitability for the task. The state weight formula is: , Among them, α is the weight of the power factor, which indicates the impact of power on the priority of task takeover; β is the weight of the load factor, which indicates the importance of load capacity when taking over tasks; γ is the proximity weight, which indicates the priority of the distance between the agent and the fault location; Indicates the current power. Indicates the maximum power. Indicates the current load, Indicates the maximum load, Indicates the distance from the agent to the starting point of the task to be carried out by the faulty agent; Step 50233: Calculate a comprehensive score for each emergency plan , the score is based on the urgency P, the state weight and proximity The weighted sum of: , Among them, λ, μ, ν are the weight coefficients of the optimal solution, which are used to adjust the importance of task urgency, agent state and proximity. is the distance from the current agent to the faulty agent; Step 50234: All generated emergency plans are scored according to their comprehensive scores. After the optimal emergency plan is determined, the system immediately sends instructions to the corresponding agent to instruct it to take over the task of the faulty equipment. At the same time, the communication module notifies other agents to avoid path conflicts.

[0056] This comprehensive weight takes into account the physical distance between the agent and the task, power adequacy, load capacity, and the urgency of the task, ensuring that the takeover plan can find the optimal balance between resource constraints and task requirements, thereby maximizing inspection efficiency and system response speed.

[0057] Once the system determines the emergency plan, the task reallocation process will proceed quickly. The system will send a new task instruction to the agent that takes over the task, instructing it to go to the area where the faulty equipment is located or to handle tasks with higher priority. Other agents will recalculate their own paths and task arrangements according to the nature of the event to ensure the efficiency and consistency of task execution. For example, when a drone fails, the system will quickly select the nearest drone to take over the inspection task in the area, and other devices in the original task will adjust their own paths accordingly to avoid inspection blind spots caused by equipment failure.

[0058] During the entire emergency response process, each intelligent agent always keeps information synchronized to ensure that the inspection equipment can grasp the changes in the task in real time. The system's built-in collaborative management module also supports real-time updating and sharing of inspection task progress information, which means that when the task is redistributed, all devices can immediately obtain the status of the newly assigned task and make corresponding adjustments. The low latency characteristics and data synchronization mechanism of the communication module make these task adjustments almost instantaneous, greatly improving the system's emergency response speed and inspection flexibility.

[0059] Embodiment 2 This embodiment provides a power multi-agent dynamic collaborative inspection system, including: Inspection task acquisition module, used to obtain natural language description of power inspection tasks; Inspection task parsing module, used to parse the natural language description of power inspection tasks and extract power inspection targets and path planning constraints; The initial inspection plan generation module is used to generate the initial inspection path plan for each intelligent agent based on the power inspection objectives and path planning constraints, and assign corresponding task areas to each intelligent agent according to its characteristics; The adaptability evaluation module is used to evaluate the adaptability of the current inspection path by using the reward function of the state-action pair combined with the path inspection efficiency and path conflict risk quantitative indicators when new variables or emergencies appear in the environment; The inspection plan adjustment module is used to dynamically adjust the path of each intelligent agent according to the adaptability evaluation results, re-evaluate the current status and reallocate tasks. During the adjustment process, the working status data of each intelligent agent is obtained, and the optimal takeover plan is determined based on the working status data of each intelligent agent to obtain the adjusted inspection path plan for each intelligent agent.

[0060] Embodiment 3 This embodiment provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps in the power multi-agent dynamic collaborative inspection method as described above are implemented.

[0061] Embodiment 4 This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps in the above-mentioned method for dynamic collaborative inspection of electric power multi-agents are implemented.

[0062] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) containing computer-usable program code.

[0063] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0064] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0065] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0066] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

[0067] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for dynamic collaborative inspection of electric power multi-agents, characterized in that: The steps include: Obtain natural language description of power inspection tasks; Parse the natural language description of power inspection tasks and extract the constraints of power inspection goals and path planning; Based on the power inspection objectives and path planning constraints, the corresponding task areas are assigned to each agent according to its characteristics, and the initial inspection path plan for each agent is generated; When new variables or emergencies appear in the environment, the adaptability of the current inspection path is evaluated by using the reward function of the state-action pair combined with the path inspection efficiency and path conflict risk quantitative indicators; When the adaptability evaluation result indicates that the inspection path of each agent needs to be adjusted, the agent gradually selects the path plan that maximizes the reward function until the total expected return is met; During the adjustment process, the working status data of each intelligent body is obtained, and the optimal takeover plan is determined based on the working status data of each intelligent body to obtain the adjusted inspection path plan of each intelligent body.

2. The power multi-agent dynamic collaborative inspection method according to claim 1, characterized in that: The power inspection objectives include minimizing the path length and maximizing the inspection area coverage; The constraints of path planning include: first, each inspection point can only be visited once by an agent; Second, each agent must complete its assigned inspection task, with the starting and ending points being the same location.

3. The power multi-agent dynamic collaborative inspection method according to claim 1, characterized in that: Path inspection efficiency score : ， Path conflict risk : ， in, is the inspection coverage rate, that is, the percentage of inspection points covered. is the ratio of task completion time to the maximum allowed time, is the ratio of the inspection path length to the shortest path length, , and The range is 0 to 1, , , They are the importance weights of coverage, time and path length, and their values ​​can be set according to actual needs; Representing an Agent i and Agent j The path overlap distance, It is an intelligent agent i and j The greater the distance, the smaller the overlap risk. Representing an Agent i and j The probability of inspecting the same area at the same time.

4. The power multi-agent dynamic collaborative inspection method according to claim 3, characterized in that: The reward function for a state-action pair is calculated as: ， ， in, Indicates the current time t state Take Action The value of is the learning rate, Status Next action It is the instant reward obtained when the agent completes the task, which reflects the actual benefit of the agent in the current task assignment step. is a discount factor between 0 and 1, which is used to balance the relationship between current rewards and future rewards. is in the next state Take the next step The value of is in the next state The maximum expected value that can be obtained by choosing the optimal action to achieve this, λ is the weight of path conflict risk.

5. The power multi-agent dynamic collaborative inspection method according to claim 1, characterized in that: During the adjustment process, the working status data of each intelligent agent is obtained, and the optimal takeover plan is determined based on the working status data of each intelligent agent, including: Obtain the working status data of each intelligent agent; Generate multiple possible emergency plans based on the acquired working status data of each intelligent agent; Score multiple possible emergency plans, select the plan with the highest score as the optimal emergency plan, and send the emergency plan to the corresponding agent to take over the task of the faulty equipment.

6. The electric power multi-agent dynamic collaborative inspection method according to claim 1, characterized in that: The multiple possible emergency plans are scored and the plan with the highest score is selected as the optimal emergency plan, including: Analyze the urgency of the faulty equipment and obtain an urgency assessment score; Calculate the state weight of each agent based on the working state data of each agent; The comprehensive score of the emergency plan is obtained by combining the evaluation score of the urgency, the state weight of each agent and the weighted sum of proximity.

7. The power multi-agent dynamic collaborative inspection method according to claim 1, characterized in that: After generating the preliminary path plan, LLM assigns tasks to each agent and allocates task areas according to the characteristics of the agent. When allocating task areas, large task areas are divided into multiple sub-areas and independent path plans are generated for each sub-area. These assigned tasks are converted into multiple sub-paths, each of which contains inspection points that the agent needs to visit in sequence.

8. The power multi-agent dynamic collaborative inspection system is characterized by: include: Inspection task acquisition module, used to obtain natural language description of power inspection tasks; Inspection task parsing module, used to parse the natural language description of power inspection tasks and extract power inspection targets and path planning constraints; The initial inspection plan generation module is used to generate the initial inspection path plan for each intelligent agent based on the power inspection objectives and path planning constraints, and assign corresponding task areas to each intelligent agent according to its characteristics; The adaptability evaluation module is used to evaluate the adaptability of the current inspection path by using the reward function of the state-action pair combined with the path inspection efficiency and path conflict risk quantitative indicators when new variables or emergencies appear in the environment; Inspection plan adjustment module, which is used for the agents to gradually select the path plan that maximizes the reward function when the adaptability evaluation result indicates that the inspection path of each agent needs to be adjusted until the total expected return is met; During the adjustment process, the working status data of each intelligent body is obtained, and the optimal takeover plan is determined based on the working status data of each intelligent body to obtain the adjusted inspection path plan of each intelligent body.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps in the power multi-agent dynamic collaborative inspection method as described in any one of claims 1 to 7 are implemented.

10. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the power multi-agent dynamic collaborative inspection method as described in any one of claims 1-7 are implemented.

Images