Multi-agent cooperative electric power inspection method, system and device and storage medium

Abstract

The invention discloses a multi-agent cooperative electric power inspection method, system and device and a storage medium, and relates to the technical field of electric power inspection, and the method comprises the steps: generating a prompt according to a user service instruction through employing a pre-established prompt word agent; analyzing a user service instruction from the prompt by using a pre-established language understanding agent; and according to the analyzed user service instruction, intelligent routing and decision making are carried out among a plurality of service agents by using a pre-established routing agent, and a specific task process is completed through the pre-established service agent. According to the method, a more intelligent and automatic task decomposition and execution mode is adopted to process complex tasks, the capabilities of a large model in language understanding and task arrangement aspects are fully utilized, automatic processing of the complex tasks in a power system is achieved, manual participation in the processing process is reduced, the labor cost is greatly reduced, the task processing efficiency is improved, and the task processing efficiency is improved. And the working threshold is reduced, and the overall stability is improved by using a plurality of service agents to work together.

Description

Technical Field

[0001] The present invention relates to the technical field of electric power inspection, and in particular to a multi-agent collaborative electric power inspection method, system, equipment and storage medium. Background Art

[0002] In existing power scenarios, when dealing with complex tasks, tasks are usually broken down manually and the execution of each subtask is controlled manually. Although this method has its flexibility, it also has some significant disadvantages, including:

[0003] (1) High time cost: Manual task decomposition requires a lot of time and effort to carefully consider the details of each subtask and ensure the connection and coordination between them. In addition, human control of the execution of each subtask also requires continuous attention and adjustment, which also consumes a lot of time.

[0004] (2) Error-prone: When tasks are manually decomposed, they may be unreasonable or have omissions due to incomplete understanding of the tasks or lack of consideration. In the process of manually controlling the execution of subtasks, execution errors or deviations may also occur due to human factors.

[0005] (3) Low efficiency: It is usually difficult to achieve parallel processing when manually controlling the execution of each subtask, which leads to low efficiency of task execution. At the same time, it is also difficult for human control to fully utilize modern technologies such as computers to improve the automation and intelligence level of task execution.

[0006] (4) Strong dependence on personnel: The method of manually decomposing and controlling tasks requires high professional ability and experience of personnel. If the personnel are not capable or experienced enough, the effectiveness and efficiency of task execution may be affected. Summary of the invention

[0007] The purpose of the present invention is to address the problems in the above-mentioned prior art and provide a multi-agent collaborative power inspection method, system, equipment and storage medium, which adopts a more intelligent and automated task decomposition and execution method to handle complex tasks, fully utilize the capabilities of large models in language understanding and task scheduling, realize the automatic processing of complex tasks in the power system, reduce manual participation in the processing process, greatly reduce labor costs, improve task processing efficiency, and lower the work threshold.

[0008] In order to achieve the above object, the present invention has the following technical solutions:

[0009] In a first aspect, a multi-agent collaborative power inspection method is provided, comprising:

[0010] Generate prompts based on user business instructions using pre-established prompt word agents;

[0011] Utilize pre-built language understanding agents to parse user business instructions from prompts;

[0012] According to the parsed user business instructions, the pre-established routing agent is used to perform intelligent routing and decision-making among multiple business agents, and the specific task process is completed through the pre-established business agent.

[0013] As a preferred solution, in the step of using a pre-established prompt word agent to generate prompts according to user business instructions, an automatic prompt generation AutoPrompt method is used to establish a prompt word agent, and the automatic prompt generation AutoPrompt method generates a target prompt word sequence that can guide the language model to output a specific task answer by iteratively selecting and adjusting prompt words; the step of generating prompts according to user business instructions includes:

[0014] Receive and analyze user business instructions to identify the user's actual needs;

[0015] The pre-trained masked language model is used to encode the context of the user's business instructions, and an initial prompt word sequence is generated based on the encoding results according to the user's actual needs. By evaluating the impact of the generated prompt words on task performance, the initial prompt word sequence is iteratively selected and adjusted to determine the target prompt word sequence.

[0016] As a preferred solution, the step of parsing the user's business instructions from the prompt using a pre-established language understanding agent includes:

[0017] Set priorities for each task based on prompts generated by user business instructions;

[0018] Analyze the dependencies between tasks and classify them; execute tasks at the same level in parallel, and wait for the tasks at the next level to complete before starting to execute; set the level of tasks that do not depend on other tasks to zero, and set the level of tasks that depend on other tasks to the maximum level of the dependent tasks plus 1:

[0019] level i =max(level j )+1,j=d i,0 ,...d i,n-1

[0020] In the formula, level i is the level of the ith task, d i,0 is the number of the 0th task that the ith task depends on, and n is the number of tasks that the ith task depends on;

[0021] Allocate resources based on the difficulty of the task and the resources required;

[0022] Generate task schedule by integrating task priority, task classification and task resource allocation;

[0023] Confirm and optimize the generated task scheduling to make it meet actual needs.

[0024] As a preferred solution, it also includes establishing a business agent description table, and recording the specific information of a business agent in each record of the business agent description table. The specific information includes the business agent application programming interface API, the business agent function description, the device where the business agent is located, and the number of resources on the device where the business agent is located.

[0025] As a preferred solution, the step of using a pre-established routing agent to perform intelligent routing and decision-making among multiple business agents includes:

[0026] Match each task in the task arrangement with the functional description of the business intelligence entity to obtain a matching list of all tasks;

[0027] In the matching list of all tasks, traverse each task. For tasks with only one business agent matching, select the corresponding business agent to bind to the corresponding task; for tasks with multiple business agents matching, select the business agent with the lowest bid to bind to the corresponding task through the auction-based task allocation model;

[0028] For business agents that have been bound to corresponding tasks, the execution of the business agents is controlled according to the task orchestration process. Tasks that can be executed in parallel are executed in parallel, and tasks with dependencies are executed sequentially. In the process of controlling the execution of the business agents, the routing agent obtains the input and output of the business agents, completes the data format conversion and input data list organization according to the next business agent application programming interface API, and calls the next business agent application programming interface API.

[0029] As a preferred solution, when selecting the business agent with the lowest bid to bind to the corresponding task through the auction-based task allocation model, each business agent bids based on the resources required to complete the task, the task completion speed and the number of currently unfinished tasks.

[0030] As a preferred solution, the business agent is composed of multiple agents, each of which completes one or more specific tasks; each business agent is connected to a routing agent, and the routing agent transmits information between different business agents.

[0031] In a second aspect, a multi-agent collaborative power inspection system is provided, comprising:

[0032] A prompt generation module, for generating prompts according to user business instructions using pre-established prompt word agents;

[0033] An instruction parsing module, which is used to parse user business instructions from prompts using a pre-established language understanding agent;

[0034] The task execution module is used to perform intelligent routing and decision-making among multiple business agents based on the parsed user business instructions using pre-established routing agents, and complete the specific task process through pre-established business agents.

[0035] As a preferred solution, the prompt generation module uses an automatic prompt generation AutoPrompt method to establish a prompt word agent, and the automatic prompt generation AutoPrompt method generates a target prompt word sequence that can guide the language model to output a specific task answer by iteratively selecting and adjusting prompt words;

[0036] The steps of generating prompts according to user business instructions include:

[0037] Receive and analyze user business instructions to identify the user's actual needs;

[0038] The pre-trained masked language model is used to encode the context of the user's business instructions, and an initial prompt word sequence is generated based on the encoding results according to the user's actual needs. By evaluating the impact of the generated prompt words on task performance, the initial prompt word sequence is iteratively selected and adjusted to determine the target prompt word sequence.

[0039] As a preferred solution, the instruction parsing module sets a priority for each task based on the prompt generated by the user's business instruction;

[0040] Analyze the dependencies between tasks and classify them; execute tasks at the same level in parallel, and wait for the tasks at the next level to complete before starting to execute; set the level of tasks that do not depend on other tasks to zero, and set the level of tasks that depend on other tasks to the maximum level of the dependent tasks plus 1:

[0041] level i =max(level j )+1,j=d i,0 ,...d i,n-1

[0042] In the formula, level i is the level of the ith task, d i,0is the number of the 0th task that the ith task depends on, and n is the number of tasks that the ith task depends on;

[0043] Allocate resources based on the difficulty of the task and the resources required;

[0044] Generate task schedule by integrating task priority, task classification and task resource allocation;

[0045] Confirm and optimize the generated task scheduling to make it meet actual needs.

[0046] As a preferred solution, when the task execution module uses the pre-established routing agent to perform intelligent routing and decision-making among multiple business agents, a business agent description table is established, and specific information of a business agent is recorded in each record of the business agent description table, the specific information includes the business agent application programming interface API, the business agent function description, the device where the business agent is located, and the number of resources of the device where the business agent is located; the steps of performing intelligent routing and decision-making among multiple business agents include:

[0047] Match each task in the task arrangement with the functional description of the business intelligence entity to obtain a matching list of all tasks;

[0048] In the matching list of all tasks, traverse each task. For tasks with only one business agent matching, select the corresponding business agent to bind to the corresponding task; for tasks with multiple business agents matching, select the business agent with the lowest bid to bind to the corresponding task through the auction-based task allocation model;

[0049] For business agents that have been bound to corresponding tasks, the execution of the business agents is controlled according to the task orchestration process. Tasks that can be executed in parallel are executed in parallel, and tasks with dependencies are executed sequentially. In the process of controlling the execution of the business agents, the routing agent obtains the input and output of the business agents, completes the data format conversion and input data list organization according to the next business agent application programming interface API, and calls the next business agent application programming interface API.

[0050] As a preferred solution, when the task execution module selects the business agent with the lowest bid to bind to the corresponding task through the auction-based task allocation model, each business agent bids according to the resources required to complete the task, the task completion speed and the number of currently unfinished tasks;

[0051] The business agent is composed of multiple agents, each of which completes one or more specific tasks; each business agent is connected to a routing agent, and the routing agent transmits information between different business agents.

[0052] According to a third aspect, an electronic device is provided, comprising a processor and a memory, wherein the processor is configured to execute a computer program stored in the memory to implement the multi-agent collaborative power inspection method as described in the first aspect.

[0053] In a fourth aspect, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores at least one instruction, and when the at least one instruction is executed by a processor, the multi-agent collaborative power inspection method as described in the first aspect is implemented.

[0054] Compared with the prior art, the first aspect of the present invention has at least the following beneficial effects:

[0055] The prompt word agent, language understanding agent, routing agent and business agent are established, and the power inspection task of multi-agent collaboration of perception, cognition and decision-making is realized based on the big model. The ability of the big model in language understanding and task scheduling is fully utilized to realize the automatic processing of complex tasks in the power system, reduce the manual participation in the processing process, greatly reduce the labor cost, improve the efficiency of task processing, and lower the work threshold. The prompt word agent generates high-quality and robust prompts according to the user's business instructions, the language understanding agent is used to understand and parse the user's business instructions, the routing agent is responsible for intelligent routing and decision-making between multiple business agents, and the business agent is responsible for completing the specific task process. The present invention uses multi-agent collaboration technology to realize end-to-end processing of power tasks. At the same time, distributed computing is adopted, and different agents can be distributed on different computing nodes, so that large-scale parallel computing can be realized, and the efficiency and robustness of business processing can be improved. Each agent of the present invention is responsible for solving specific problems or subtasks, and the modular design makes the big model more flexible, and can be quickly adjusted and optimized for different scenarios and environments. The present invention uses multiple business agents to work together. Even if a business agent has a problem and goes offline, other business agents can still continue to work, thereby improving the stability of the overall method during actual use. The present invention uses a large model as the core of the language understanding agent. The large model has the advantages of powerful semantic understanding ability, context dependency, potential knowledge discovery ability, general task solving ability, autonomous learning ability and interactive ability. When processing complex tasks, it is more efficient and reliable to adopt a more intelligent and automated task decomposition and execution method.

[0056] It can be understood that the beneficial effects of the second to fourth aspects mentioned above can be found in the relevant description of the first aspect mentioned above, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0058] Figure 1 A schematic diagram of the principle architecture of a multi-agent collaborative power inspection method according to an embodiment of the present invention;

[0059] Figure 2 Schematic diagram of a routing agent performing intelligent routing and decision-making among multiple business agents in an embodiment of the present invention;

[0060] Figure 3 Flow chart of a multi-agent collaborative power inspection method according to an embodiment of the present invention;

[0061] Figure 4 Structural block diagram of a multi-agent collaborative power inspection system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0062] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0063] In existing power scenarios, when dealing with complex tasks, tasks are generally decomposed manually and the execution of each subtask is controlled manually. However, when dealing with complex tasks, it is more efficient and reliable to adopt a more intelligent and automated task decomposition and execution method. With the powerful capabilities of large models in language understanding and task scheduling, they can understand and generate natural language and handle a variety of natural language tasks, such as text generation, knowledge question and answer, reasoning calculation and reading comprehension. They improve the accuracy of natural language processing by training on large-scale corpus and learning rich knowledge in various fields. In terms of task scheduling, large models can carry out systematic planning and adjustment according to specific needs and situations. By managing the relationships and resources between subsystems, changes and challenges can be flexibly responded to.

[0064] See also Figure 3 The embodiment of the present invention proposes a multi-agent collaborative power inspection method, comprising the following steps:

[0065] S1, using the pre-established prompt word agent to generate prompts according to the user's business instructions;

[0066] S2, using a pre-established language understanding agent to parse the user's business instructions from the prompt;

[0067] S3. Based on the parsed user business instructions, use the pre-established routing agent to perform intelligent routing and decision-making among multiple business agents, and complete the specific task process through the pre-established business agent.

[0068] like Figure 1 As shown, the multi-agent collaborative power inspection method of an embodiment of the present invention is based on multi-agent collaboration in perception, cognition, and decision-making, and simplifies the processing flow of power business and improves processing efficiency through collaboration and distributed computing among agents.

[0069] The multi-agent of the embodiment of the present invention mainly includes a prompt word agent, a language understanding agent, a routing agent and a business agent. Among them, the prompt word agent is used to generate high-quality and robust prompts according to user business instructions. The language understanding agent is used to understand and parse user business instructions. The routing agent is responsible for intelligent routing and decision-making between multiple business agents. The business agent is responsible for completing the specific task process.

[0070] In one possible implementation, the prompt word agent of the embodiment of the present invention can obtain prompt words based on user language instructions. By using prompt words, the model can more accurately understand the user's intentions, needs, and contextual information, thereby avoiding the manpower cost and guesswork involved in manually generating prompt words.

[0071] The embodiment of the present invention uses an automated method to generate prompt words, thereby improving the efficiency and wide applicability of the analysis. In step S1, an automatic prompt generation AutoPrompt method is used to establish a prompt word intelligent agent. The automatic prompt generation AutoPrompt method automatically generates a target prompt word sequence that can guide the language model to output a specific task answer by iteratively selecting and adjusting the prompt words. This process is similar to reinforcement learning, but focuses on optimizing the selection of prompt words. By evaluating the impact of the generated prompt words on the task performance, and constantly adjusting to find the optimal prompt word combination. Therefore, the automatic prompt generation AutoPrompt method effectively alleviates the problem of excessive sensitivity that may occur when prompts face minor changes, and also reduces the ambiguity of the prompts themselves, making the generated prompts clearer and more stable. For example, the automatic prompt generation AutoPrompt method can quickly respond and generate accurate prompts such as "Please read the following article and write a concise summary", thereby driving the language model to accurately complete the summary generation work. The steps of generating prompts according to user business instructions in the embodiment of the present invention include:

[0072] Step 1: Receive and parse user business instructions to identify the user's actual needs. This step ensures that the user's intentions and expectations can be accurately understood.

[0073] Step 2: Use a pre-trained masked language model (such as BERT, a bidirectional encoding representation model based on Transformer) to encode the context of the user's business instructions, and generate an initial prompt word sequence based on the encoding results according to the user's actual needs.

[0074] In a possible implementation, the language understanding agent of the embodiment of the present invention can analyze the user's true intention and generate corresponding operation commands according to the user's language instructions and the prompt words generated by the prompt word agent. The embodiment of the present invention uses a large model as the core of the language understanding agent, which has the advantages of powerful semantic understanding ability, context dependency, potential knowledge discovery ability, general task solving ability, autonomous learning ability and interactive ability.

[0075] In the embodiment of the present invention, step S2 of parsing the user service instruction from the prompt specifically includes:

[0076] Step 1: After receiving the user's instructions and prompts, the language understanding agent will perform task analysis, understand the user's intentions, and identify the specific tasks that need to be completed and the relationships between tasks.

[0077] Step 2: For complex tasks, the language understanding agent will break them down into multiple subtasks. These subtasks are easier to manage and distribute, and each subtask can be performed independently.

[0078] Step 3: According to the prompts generated by the user's business instructions (such as "urgent" or "important") or the built-in algorithm of the model, the language understanding intelligent body will set the priority for each task. Tasks with high priority will be processed first to ensure that important and urgent tasks can be completed on time. The urgency of the embodiment of the present invention is set to 4 levels. The urgency of ordinary tasks is set to 0, the urgency of dependent tasks is set to 1, and the urgency of tasks with user prompts (such as "urgent" or "important") is set to 2. Urgency 3 is generally reserved unless there are particularly urgent and system-related tasks.

[0079] Step 4: Analyze the dependencies between tasks to determine which tasks need to be completed first and which tasks can be performed in parallel. This helps avoid conflicts between tasks and delays caused by dependencies. When analyzing task dependencies, the embodiment of the present invention classifies tasks. Tasks at the same level can be performed in parallel, and tasks at the next level need to wait for the completion of tasks at the previous level before they can be executed. For tasks that do not depend on other tasks, the level is set to zero. For tasks that depend on other tasks, the level is set to the maximum level of the dependent tasks plus 1. The specific setting expression is as follows:

[0080] level i =max(level j )+1,j=d i,0 ,...d i,n-1

[0081] In the formula, level i is the level of the ith task, d i,0 is the number of the 0th task that the ith task depends on, and n is the number of tasks that the ith task depends on.

[0082] Step 5: Provide users with suggestions for resource allocation based on the difficulty of the task and the resources required. This helps users better manage resources and ensure the smooth progress of the task.

[0083] Step 6. After completing the above steps, the language understanding agent will generate a detailed task schedule, which will include the specific steps of each task, the dependencies between tasks, and the priority.

[0084] Step 7: Users need to confirm and optimize the task arrangement generated by the large model. Users can adjust and modify the arrangement based on their own experience and needs to ensure that it meets actual needs.

[0085] In a possible implementation, the routing agent of the embodiment of the present invention can route the business agent according to the task arrangement generated by the large model. Furthermore, the embodiment of the present invention establishes a business agent description table, and each record in the business agent description table records the specific information of a business agent, including the business agent application programming interface API, the business agent function description, the device where the business agent is located, and the number of resources on the device where the business agent is located. Figure 2 When the routing agent receives a task schedule, it forwards each task in the task schedule to the target business agent and controls the target business agent to execute in sequence. When a task corresponds to multiple business agents, the embodiment of the present invention adopts an auction-based task allocation model. Each business agent bids for the task according to its own capabilities and resources, and the task will be assigned to the business agent with the lowest bid. This model can effectively utilize the capabilities and resources of business agents and improve the efficiency of task execution. When a business agent fails, the routing agent can redirect the request to other available business agents to ensure service continuity. The working process of the routing agent is as follows:

[0086] Step 1: Match each task in the task orchestration with the functional description of the business agent to obtain a matching list of all tasks;

[0087] Step 2: In the matching list of all tasks, traverse each task. For tasks that have only one business agent matching, directly select the corresponding business agent to bind to the corresponding task; for tasks that have multiple business agents matching, select the business agent with the lowest bid and bind it to the corresponding task through the auction-based task allocation model; each business agent bids based on the resources required to complete the task, the task completion speed, and the number of currently unfinished tasks.

[0088] Step 3: Control the execution of the business agent according to the task orchestration process, execute tasks that can be executed in parallel in parallel, and execute tasks with dependencies sequentially; in the process of controlling the execution of the business agent, the routing agent obtains the input and output of the business agent, completes the data format conversion and input data list organization according to the next business agent application programming interface API, and calls the next business agent application programming interface API.

[0089] In a possible implementation, the business agent in the embodiment of the present invention is composed of multiple agents, each of which completes one or more specific tasks. Figure 2As shown in FIG. 1 , the business agents of the embodiment of the present invention can be divided into an image binarization agent, an object segmentation agent, an image classification agent, and a report generation agent. For business functions that are used frequently, multiple agents can be set, such as Figure 2 Image classification agent in .

[0090] Each business agent is connected to the routing agent, and the routing agent serves as a bridge to complete the information transmission between different business agents. The business agent of the embodiment of the present invention adopts a distributed computing method to disperse the business agent computing tasks to multiple computing nodes, so as to realize the efficient operation of the power business scenario agent. The distributed computing system has good fault tolerance and robustness. Even if some nodes or agents fail, the entire system can continue to operate to ensure the continuity and stability of decision-making. At the same time, the computing resources and agent configuration can be dynamically adjusted according to environmental changes and task requirements. It has strong adaptability and flexibility, and can effectively improve the operating efficiency and stability of the multi-agent system in the power business scenario.

[0091] The multi-agent collaborative power inspection method of the embodiment of the present invention realizes end-to-end processing of power tasks based on the multi-agent collaboration of perception, cognition and decision-making. At the same time, by adopting distributed computing, different agents can be distributed on different computing nodes, so that large-scale parallel computing can be realized and the business processing efficiency and robustness can be improved. Each agent in the embodiment of the present invention is responsible for solving specific problems or subtasks. The modular design makes the large model more flexible and can be quickly adjusted and optimized for different scenarios and environments. The embodiment of the present invention uses multiple business agents to work together. Even if a business agent has a problem and goes offline, other business agents can still continue to work, thereby improving the stability of the entire power inspection method in the actual use process. The embodiment of the present invention uses a large model as the core of the language understanding agent. The large model has the advantages of powerful semantic understanding ability, context dependency, potential knowledge discovery ability, general task solving ability, autonomous learning ability and interaction ability. When processing complex tasks, it is more efficient and reliable to adopt a more intelligent and automated task decomposition and execution method.

[0092] See also Figure 4 Another embodiment of the present invention provides a multi-agent collaborative power inspection system, comprising:

[0093] A prompt generation module, for generating prompts according to user business instructions using pre-established prompt word agents;

[0094] An instruction parsing module, which is used to parse user business instructions from prompts using a pre-established language understanding agent;

[0095] The task execution module is used to perform intelligent routing and decision-making among multiple business agents based on the parsed user business instructions using pre-established routing agents, and complete the specific task process through pre-established business agents.

[0096] In a possible implementation, the prompt generation module uses an automatic prompt generation AutoPrompt method to establish a prompt word agent, and the automatic prompt generation AutoPrompt method generates a target prompt word sequence that can guide the language model to output a specific task answer by iteratively selecting and adjusting prompt words;

[0097] The steps of generating prompts according to user business instructions include:

[0098] Receive and analyze user business instructions to identify the user's actual needs;

[0099] The pre-trained masked language model is used to encode the context of the user's business instructions, and an initial prompt word sequence is generated based on the encoding results according to the user's actual needs. By evaluating the impact of the generated prompt words on task performance, the initial prompt word sequence is iteratively selected and adjusted to determine the target prompt word sequence.

[0100] In a possible implementation, the instruction parsing module sets a priority for each task based on a prompt generated by a user business instruction;

[0101] Analyze the dependencies between tasks and classify them; execute tasks at the same level in parallel, and wait for the tasks at the next level to complete before starting to execute; set the level of tasks that do not depend on other tasks to zero, and set the level of tasks that depend on other tasks to the maximum level of the dependent tasks plus 1:

[0102] level i =max(level j )+1,j=d i,0 ,...d i,n-1

[0103] In the formula, level i is the level of the ith task, d i,0 is the number of the 0th task that the ith task depends on, and n is the number of tasks that the ith task depends on;

[0104] Allocate resources based on the difficulty of the task and the resources required;

[0105] Generate task schedule by integrating task priority, task classification and task resource allocation;

[0106] Confirm and optimize the generated task scheduling to make it meet actual needs.

[0107] In a possible implementation, when the task execution module uses a pre-established routing agent to perform intelligent routing and decision-making among multiple business agents, a business agent description table is established, and specific information of a business agent is recorded in each record of the business agent description table, the specific information includes the business agent application programming interface API, the business agent function description, the device where the business agent is located, and the number of resources of the device where the business agent is located; the steps of performing intelligent routing and decision-making among multiple business agents include:

[0108] Match each task in the task arrangement with the functional description of the business intelligence entity to obtain a matching list of all tasks;

[0109] In the matching list of all tasks, traverse each task. For tasks with only one business agent matching, select the corresponding business agent to bind to the corresponding task; for tasks with multiple business agents matching, select the business agent with the lowest bid to bind to the corresponding task through the auction-based task allocation model;

[0110] For business agents that have been bound to corresponding tasks, the execution of the business agents is controlled according to the task orchestration process. Tasks that can be executed in parallel are executed in parallel, and tasks with dependencies are executed sequentially. In the process of controlling the execution of the business agents, the routing agent obtains the input and output of the business agents, completes the data format conversion and input data list organization according to the next business agent application programming interface API, and calls the next business agent application programming interface API.

[0111] In one possible implementation, when the task execution module selects the business agent with the lowest bid to bind to the corresponding task through an auction-based task allocation model, each business agent bids based on the resources required to complete the task, the task completion speed, and the number of currently unfinished tasks.

[0112] The business agent is composed of multiple agents, each of which completes one or more specific tasks; each business agent is connected to a routing agent, and the routing agent transmits information between different business agents.

[0113] Another embodiment of the present invention further proposes an electronic device, including a processor and a memory, wherein the processor is used to execute a computer program stored in the memory to implement the multi-agent collaborative power inspection method.

[0114] Another embodiment of the present invention further proposes a computer-readable storage medium, wherein the computer-readable storage medium stores at least one instruction, and when the at least one instruction is executed by a processor, the multi-agent collaborative power inspection method is implemented.

[0115] The computer program includes computer program code, which may be in source code form, object code form, executable file or some intermediate form. The computer-readable storage medium may include: any entity or device, medium, USB flash drive, mobile hard disk, disk, optical disk, computer memory, read-only memory, random access memory, electric carrier signal, telecommunication signal and software distribution medium, etc. that can carry the computer program code. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals. For ease of explanation, the above content only shows the part related to the embodiment of the present invention. For specific technical details not disclosed, please refer to the method part of the embodiment of the present invention. The computer-readable storage medium is non-temporary and can be stored in a storage device formed by various electronic devices, which can realize the execution process recorded in the method of the embodiment of the present invention.

[0116] It will be appreciated by those skilled in the art 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 a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, 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, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0117] 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 process in the flowchart. 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.

[0118] 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.

[0119] 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.

[0120] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A multi-agent collaborative power inspection method, characterized in that: include: Generate prompts based on user business instructions using pre-established prompt word agents; Utilize pre-built language understanding agents to parse user business instructions from prompts; According to the parsed user business instructions, the pre-established routing agent is used to perform intelligent routing and decision-making among multiple business agents, and the specific task process is completed through the pre-established business agent.

2. The multi-agent collaborative power inspection method according to claim 1, characterized in that: In the step of using the pre-established prompt word agent to generate prompts according to the user's business instructions, an automatic prompt generation AutoPrompt method is used to establish the prompt word agent, and the automatic prompt generation AutoPrompt method generates a target prompt word sequence that can guide the language model to output a specific task answer by iteratively selecting and adjusting prompt words; The steps of generating prompts according to user business instructions include: Receive and analyze user business instructions to identify the user's actual needs; The pre-trained masked language model is used to encode the context of the user's business instructions, and an initial prompt word sequence is generated based on the encoding results according to the user's actual needs. By evaluating the impact of the generated prompt words on task performance, the initial prompt word sequence is iteratively selected and adjusted to determine the target prompt word sequence.

3. The multi-agent collaborative power inspection method according to claim 1, characterized in that: The step of parsing the user's business instructions from the prompt using the pre-established language understanding agent comprises: Set priorities for each task based on prompts generated by user business instructions; Analyze the dependencies between tasks and classify them; execute tasks at the same level in parallel, and wait for the tasks at the next level to complete before starting to execute; set the level of tasks that do not depend on other tasks to zero, and set the level of tasks that depend on other tasks to the maximum level of the dependent tasks plus 1: level i =max(level j )+1,j=d i,0 ,...d i,n-1 In the formula, level i is the level of the ith task, d i,0 is the number of the 0th task that the ith task depends on, and n is the number of tasks that the ith task depends on; Allocate resources based on the difficulty of the task and the resources required; Generate task schedule by integrating task priority, task classification and task resource allocation; Confirm and optimize the generated task scheduling to make it meet actual needs.

4. The multi-agent collaborative power inspection method according to claim 3 is characterized in that: It also includes establishing a business agent description table, recording specific information of a business agent in each record of the business agent description table, the specific information includes the business agent application programming interface API, the business agent function description, the device where the business agent is located, and the number of resources on the device where the business agent is located.

5. The multi-agent collaborative power inspection method according to claim 4 is characterized in that: The step of using a pre-established routing agent to perform intelligent routing and decision-making among multiple business agents includes: Match each task in the task arrangement with the functional description of the business intelligence entity to obtain a matching list of all tasks; In the matching list of all tasks, traverse each task. For tasks with only one business agent matching, select the corresponding business agent to bind to the corresponding task; for tasks with multiple business agents matching, select the business agent with the lowest bid to bind to the corresponding task through the auction-based task allocation model; For business agents that have been bound to corresponding tasks, the execution of the business agents is controlled according to the task orchestration process. Tasks that can be executed in parallel are executed in parallel, and tasks with dependencies are executed sequentially. In the process of controlling the execution of the business agents, the routing agent obtains the input and output of the business agents, completes the data format conversion and input data list organization according to the next business agent application programming interface API, and calls the next business agent application programming interface API.

6. The multi-agent collaborative power inspection method according to claim 5, characterized in that: When the business agent with the lowest bid is selected to bind to the corresponding task through the auction-based task allocation model, each business agent bids according to the resources required to complete the task, the task completion speed and the number of currently unfinished tasks.

7. The multi-agent collaborative power inspection method according to claim 1, characterized in that: The business agent is composed of multiple agents, each of which completes one or more specific tasks; each business agent is connected to a routing agent, and the routing agent transmits information between different business agents.

8. A multi-agent collaborative power inspection system, characterized in that: include: A prompt generation module, for generating prompts according to user business instructions using pre-established prompt word agents; An instruction parsing module, which is used to parse user business instructions from prompts using a pre-established language understanding agent; The task execution module is used to perform intelligent routing and decision-making among multiple business agents based on the parsed user business instructions using pre-established routing agents, and complete the specific task process through pre-established business agents.

9. The multi-agent collaborative power inspection system according to claim 8, characterized in that: The prompt generation module uses an automatic prompt generation AutoPrompt method to establish a prompt word agent, and the automatic prompt generation AutoPrompt method generates a target prompt word sequence that can guide the language model to output a specific task answer by iteratively selecting and adjusting prompt words; The steps of generating prompts according to user business instructions include: Receive and analyze user business instructions to identify the user's actual needs; The pre-trained masked language model is used to encode the context of the user's business instructions, and an initial prompt word sequence is generated based on the encoding results according to the user's actual needs. By evaluating the impact of the generated prompt words on task performance, the initial prompt word sequence is iteratively selected and adjusted to determine the target prompt word sequence.

10. The multi-agent collaborative power inspection system according to claim 8, characterized in that: The instruction parsing module sets a priority for each task based on the prompt generated by the user's business instruction; Analyze the dependencies between tasks and classify them; execute tasks at the same level in parallel, and wait for the tasks at the next level to complete before starting to execute; set the level of tasks that do not depend on other tasks to zero, and set the level of tasks that depend on other tasks to the maximum level of the dependent tasks plus 1: level i =max(level j )+1,j=d i,0 ,...d i,n-1 In the formula, level i is the level of the ith task, d i,0 is the number of the 0th task that the ith task depends on, and n is the number of tasks that the ith task depends on; Allocate resources based on the difficulty of the task and the resources required; Generate task schedule by integrating task priority, task classification and task resource allocation; Confirm and optimize the generated task scheduling to make it meet actual needs.

11. The multi-agent collaborative power inspection system according to claim 8, characterized in that: When the task execution module uses the pre-established routing agent to perform intelligent routing and decision-making among multiple business agents, a business agent description table is established, and specific information of a business agent is recorded in each record of the business agent description table, the specific information includes the business agent application programming interface API, the business agent function description, the device where the business agent is located, and the number of resources of the device where the business agent is located; the steps of performing intelligent routing and decision-making among multiple business agents include: Match each task in the task arrangement with the functional description of the business intelligence entity to obtain a matching list of all tasks; In the matching list of all tasks, traverse each task. For tasks with only one business agent matching, select the corresponding business agent to bind to the corresponding task; for tasks with multiple business agents matching, select the business agent with the lowest bid to bind to the corresponding task through the auction-based task allocation model; For business agents that have been bound to corresponding tasks, the execution of the business agents is controlled according to the task orchestration process. Tasks that can be executed in parallel are executed in parallel, and tasks with dependencies are executed sequentially. In the process of controlling the execution of the business agents, the routing agent obtains the input and output of the business agents, completes the data format conversion and input data list organization according to the next business agent application programming interface API, and calls the next business agent application programming interface API.

12. The multi-agent collaborative power inspection system according to claim 11, characterized in that: When the task execution module selects the business agent with the lowest bid to bind to the corresponding task through the auction-based task allocation model, each business agent bids according to the resources required to complete the task, the task completion speed and the number of currently unfinished tasks; The business agent is composed of multiple agents, each of which completes one or more specific tasks; each business agent is connected to a routing agent, and the routing agent transmits information between different business agents.

13. An electronic device, characterized in that: It comprises a processor and a memory, wherein the processor is used to execute a computer program stored in the memory to implement a multi-agent collaborative power inspection method as described in any one of claims 1 to 7.

14. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores at least one instruction, and when the at least one instruction is executed by the processor, it implements the multi-agent collaborative power inspection method as described in any one of claims 1 to 7.

Images