Monitoring service scheme generation method and device, electronic equipment, storage medium and computer product

Abstract

The invention relates to the technical field of Internet, and provides a monitoring service scheme generation method and device, electronic equipment, a storage medium and a computer product, and the method comprises the steps: receiving monitoring service demand information, inputting the monitoring service demand information to a first agent, and obtaining an initial monitoring service scheme outputted by the first agent; the first agent is generated based on a large language model and is used for generating a business scheme based on input information; inputting the initial monitoring service scheme and the monitoring service demand information into a second intelligent agent to obtain a scheme verification result output by the second intelligent agent; the second agent is generated based on a large language model and is used for verifying a business scheme in the input information based on demand information in the input information; if the scheme verification result is that the scheme passes, feeding back the user or suggesting modification; and based on the scheme verification result, determining a target monitoring service scheme. According to the invention, the accuracy of the generated monitoring service scheme can be improved.

Description

Technical Field

[0001] This application relates to the field of Internet technologies, and in particular, to a method, apparatus, electronic device, storage medium, and computer product for generating a monitoring service solution. Background Art

[0002] In the current technology of performing intent parsing and generating a monitoring service solution through a large language model, a method of using prompts for process orchestration for foreseeable scenarios is usually adopted. Through a fixed process template, parsing is carried out layer by layer and step by step, and finally a complete solution is formed. Among them, a prompt is an input instruction used to guide an artificial intelligence model (especially a large language model) to complete a specific task. It is usually a text-based prompt word, question, or task description, aiming to help the model understand the user's needs and generate an output that meets the requirements. However, the currently pre-orchestrated process is too fixed. To meet the continuously emerging user needs, it is necessary to pre-orchestrate for each possibility. Once the user asks a question that has not been pre-orchestrated, the system will not be able to execute according to the expected flow chain, thereby resulting in an inability to accurately generate a monitoring service solution. Summary of the Invention

[0003] This application aims to solve at least one of the technical problems existing in the related art. For this purpose, this application proposes a method, apparatus, electronic device, storage medium, and computer product for generating a monitoring service solution, so as to solve the problem that the current technology of performing intent parsing and generating a monitoring service solution through a large language model cannot accurately generate a monitoring service solution.

[0004] According to an embodiment of the first aspect of this application, the method for generating a monitoring service solution includes: Receiving monitoring service requirement information, and inputting the monitoring service requirement information into a first intelligent agent to obtain an initial monitoring service solution output by the first intelligent agent; the first intelligent agent is generated based on a large language model and is used for generating a service solution based on the input information; Inputting the initial monitoring service solution and the monitoring service requirement information into a second intelligent agent to obtain a solution verification result output by the second intelligent agent; the second intelligent agent is generated based on a large language model and is used for verifying the service solution in the input information based on the requirement information in the input information; the solution verification result is that the solution passes, feedback to the user, or suggestions for modification; Determining a target monitoring service solution based on the solution verification result.

[0005] According to an embodiment of this application, the first intelligent agent is used for: Generate a dependency thinking chain for the required capabilities of the monitored service requirement information according to the principle of coordination between reasoning and action; wherein, the required capabilities include at least one or more of computing power capabilities, network capabilities, link capabilities, artificial intelligence capabilities, and storage cycle capabilities; Invoke a tool to parse each single-point capability in the dependency thinking chain and generate an initial monitored service plan, store the plan generation process in the private memory area of the first intelligent agent, and store the monitored service requirement information and the initial monitored service plan in the public memory area.

[0006] According to an embodiment of the present application, the computing power capability includes the selection of a resource pool; the resource pool selection strategy includes a performance priority strategy, a cost priority strategy, and a preset resource pool selection strategy.

[0007] According to an embodiment of the present application, the link capability includes the selection of a link; the link selection strategy includes a latency priority strategy, a smoothness priority strategy, and a preset link selection strategy.

[0008] According to an embodiment of the present application, the second intelligent agent is used to adopt a reinforcement learning strategy from human feedback, generate a plan verification result according to the initial monitored service plan and the monitored service requirement information, store the verification process in the private memory area of the second intelligent agent, and store the monitored service requirement information and the feedback to the user or the suggested modification in the plan verification result in the public memory area.

[0009] According to an embodiment of the present application, determining the target monitored service plan based on the plan verification result includes: If the plan verification result is that the plan passes, determine the initial monitored service plan as the target monitored service plan; If the plan verification result is to feedback to the user, output the feedback information in the plan verification result, and input the supplementary information returned based on the feedback information, the monitored service requirement information, and the initial monitored service plan into the first intelligent agent to obtain the target monitored service plan output by the first intelligent agent; If the plan verification result is a suggested modification, input the modification suggestion in the plan verification result, the monitored service requirement information, and the initial monitored service plan into the first intelligent agent to obtain the target monitored service plan output by the first intelligent agent.

[0010] The monitored service plan generation device according to the second aspect embodiment of the present application includes: A generation module, configured to receive monitoring service requirement information, input the monitoring service requirement information into a first intelligent agent, and obtain an initial monitoring service plan output by the first intelligent agent; the first intelligent agent is generated based on a large language model and is used to generate a service plan based on input information; A verification module, configured to input the initial monitoring service plan and the monitoring service requirement information into a second intelligent agent, and obtain a plan verification result output by the second intelligent agent; the second intelligent agent is generated based on a large language model and is used to verify the service plan in the input information based on the requirement information in the input information; the plan verification result is that the plan passes, feedback to the user, or suggest modification; A determination module, configured to determine a target monitoring service plan based on the plan verification result.

[0011] An electronic device according to an embodiment of the third aspect of the present application includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the method for generating a monitoring service plan as described in any one of the above is implemented.

[0012] A storage medium according to an embodiment of the fourth aspect of the present application is a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the method for generating a monitoring service plan as described in any one of the above is implemented.

[0013] A computer program product according to an embodiment of the fifth aspect of the present application includes a computer program. When the computer program is executed by a processor, the method for generating a monitoring service plan as described in any one of the above is implemented.

[0014] One or more of the above technical solutions in the embodiments of the present application at least have the following technical effects: After receiving the monitoring service requirement information, first generate a service plan based on the monitoring service requirement information through the first intelligent agent to obtain an initial monitoring service plan; then verify the initial monitoring service plan by combining the monitoring service requirement information through the second intelligent agent. Further, a target monitoring service plan can be generated according to the plan verification result. Thus, by changing the service acceptance mode arranged in a fixed process to a task-driven dual-intelligent-agent collaboration mode, the second intelligent agent can verify the initial monitoring service plan generated by the first intelligent agent. Furthermore, when the plan verification result is to feedback to the user, the monitoring service plan can be regenerated according to the content supplemented by the user, and when the plan verification result is to suggest modification, the monitoring service plan can be regenerated according to the modification suggestion. Therefore, the accuracy of the finally generated monitoring service plan can be improved.

[0015] Additional aspects and advantages of the present application will be given in part in the following description, become apparent in part from the following description, or be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0017] Figure 1 is one of the flow diagrams of the method for generating a monitoring service solution provided by an embodiment of the present application.

[0018] Figure 2 is a schematic diagram of a dual-Agent system scenario of the method for generating a monitoring service solution provided by an embodiment of the present application.

[0019] Figure 3 is the second of the flow diagrams of the method for generating a monitoring service solution provided by an embodiment of the present application.

[0020] Figure 4 is a schematic diagram of the structure of an electronic device provided by the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0021] The following will further describe in detail the embodiments of the present application in conjunction with the drawings and embodiments. The following embodiments are used to illustrate the present application, but cannot be used to limit the scope of the present application.

[0022] In the description of the embodiments of the present application, it should be noted that the orientation or positional relationship indicated by the terms "center", "longitudinal", "transverse", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the embodiments of the present application and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation to the embodiments of the present application. In addition, the terms "first", "second", and "third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0023] In the description of the embodiments of the present application, it should be noted that unless otherwise clearly specified or limited, the terms "connected" and "coupled" should be understood in a broad sense. For example, it may be a fixed connection, a detachable connection, or an integral connection; it may be a mechanical connection or an electrical connection; it may be directly connected or indirectly connected through an intermediate medium. For those of ordinary skill in the art, the specific meanings of the above terms in the embodiments of the present application can be understood according to specific circumstances.

[0024] In the embodiments of the present application, unless otherwise clearly specified or limited, the first feature being "on" or "under" the second feature may be that the first and second features are in direct contact, or the first and second features are indirectly in contact through an intermediate medium. Moreover, the first feature being "above", "over" and "on top of" the second feature may be that the first feature is directly above or obliquely above the second feature, or merely indicates that the first feature has a higher horizontal height than the second feature. The first feature being "under", "below" and "beneath" the second feature may be that the first feature is directly below or obliquely below the second feature, or merely indicates that the first feature has a lower horizontal height than the second feature.

[0025] In the description of this specification, the descriptions with reference to terms such as "one embodiment", "some embodiments", "example", "specific example", or "some examples" mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the embodiments of the present application. In this specification, the schematic descriptions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or more embodiments or examples. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0026] With the rapid development of the enterprise intelligent monitoring and security field, the global intelligent video fusion monitoring market scale has been increasing year by year. In order to meet the high-quality and multi-service integration requirements of industry users for intelligent video monitoring systems, there is an urgent need to add a new task-driven dynamic scheme generation mode to efficiently and accurately meet the needs of industry users to cope with increasingly complex security threats and monitoring requirements.

[0027] In the current solutions for intent parsing through large language models, a method of using prompts for process orchestration for foreseeable scenarios is usually adopted. Through a fixed process template, parsing is carried out layer by layer and step by step to finally form a complete solution.

[0028] For example, when the user's input is "I hope to enable the camera in the parking lot to recognize license plates, with the requirements of being able to view the playback in high definition, storing the video in the cloud for 7 days, selecting resources with lower costs and links with lower latency. What solutions are recommended?", the system will respectively extract the calls of multi-level process large language models (LLMs) such as "action-level determination", "scenario-level determination", and "ability list determination" according to the corresponding prompt orchestration template. After each layer is called, rules are used for determination to make the program jump to the next level. Finally, after all the ability lists are determined, the corresponding prompt word templates are found according to each ability list to guide the LLM to give answers and integrate them to achieve the business acceptance work. Among them, the large language model is a deep learning model trained with a large amount of text data, which can generate natural language text or understand the meaning of language text. The large language model can handle various natural language tasks, such as text classification, question answering, dialogue, etc., and is an important way to artificial intelligence.

[0029] However, in the current process of generating solutions through prompt process orchestration, due to the problem that the pre-orchestrated process is too fixed, the following two pain points will be faced: Harsh requirements for single-point generation effects: Due to the prompt process orchestration method, the input of each step depends on the output of the previous step's model parsing result, so the entire system has relatively harsh requirements for the effects of each single point. Once an ideal output cannot be obtained in a certain link, it will directly affect the subsequent execution actions of all subsequent links. Moreover, the model itself is a model with strong randomness and uncontrollability. At the same time, due to its large parameter volume, the deployment hardware cost is relatively high currently. Most organizations may choose a smaller-scale version when using it, and the model understanding and reasoning ability of the smaller-scale version is limited, further increasing the uncertainty in the link; Low intelligence and difficulty in adding new scenarios: For each possible user demand scenario, it is necessary to conduct deductions in advance and orchestrate the prompt transfer chain, resulting in too single supported scenarios of the system. Once the user asks a question that has not been pre-orchestrated, the system will not be able to execute according to the expected transfer chain, resulting in a poor user experience and insufficient intelligence level. At the same time, to meet the continuously emerging user needs, it is necessary to pre-orchestrate for each possibility, resulting in a huge amount of manual work, and the development speed can never meet the new requirements.

[0030] Based on this, the present application proposes a method, device, electronic device, storage medium, and computer product for generating a monitoring service solution.

[0031] Figure 1 It is one of the flow diagrams of the method for generating a monitoring service solution provided by the embodiments of the present application, as Figure 1As shown, the method for generating a monitoring service solution includes: Step 110: Receive monitoring service requirement information, input the monitoring service requirement information into the first intelligent agent, and obtain an initial monitoring service solution output by the first intelligent agent; the first intelligent agent is generated based on a large language model and is used to generate a service solution based on the input information.

[0032] Step 120: Input the initial monitoring service solution and the monitoring service requirement information into the second intelligent agent, and obtain a solution verification result output by the second intelligent agent; the second intelligent agent is generated based on a large language model and is used to verify the service solution in the input information based on the requirement information in the input information; the solution verification result is that the solution passes, feedback to the user, or suggestions for modification.

[0033] Step 130: Determine the target monitoring service solution based on the solution verification result.

[0034] It should be noted that the execution subject of the method for generating a monitoring service solution provided in the embodiments of the present application may be a computer device, such as a mobile phone, a tablet computer, a notebook computer, a handheld computer, a vehicle-mounted electronic device, a wearable device, an Ultra-mobile Personal Computer (UMPC), a netbook, or a Personal Digital Assistant (PDA), etc. A page generation platform or page generation software, etc., may be set in the computer device. It should be noted that all data that needs to be obtained in the present application is obtained through formal channels after being authorized by relevant users.

[0035] A monitoring service solution generation device may be set or connected in the computer device of the present application, so as to control the monitoring service solution generation device to execute the method for generating a monitoring service solution of the present application.

[0036] It should be noted that the method for generating a monitoring service solution provided in the embodiments of the present application may be specifically applied to a service acceptance system. Among them, the service acceptance system may be deployed on a computer device. Figure 2 is a schematic diagram of a dual-Agent system scenario of the method for generating a monitoring service solution provided in the embodiments of the present application, Figure 3 is a second process schematic diagram of the method for generating a monitoring service solution provided in the embodiments of the present application, as Figure 2 and Figure 3 As shown, a first intelligent agent (which may be referred to as a product solution recommendation Agent in the present application) and a second intelligent agent (which may be referred to as an optimization and reflection Agent in the present application) may be set in the service acceptance system. Moreover, both the first intelligent agent and the second intelligent agent are generated based on a large language model.

[0037] In this application, whether it is the first agent or the second agent, they are both "moving" in the "environment" (Enviroment). The environment describes the state space where the agent is located, and movement is an abstract concept that can be understood as the sum of all the agent's behaviors.

[0038] The "movement" of the agent can be roughly classified into three parts: perception, internal processing, and action, corresponding to the three modules of Perception, Brain, and Action respectively. The Brain module undertakes internal tasks such as memory, thinking, and decision-making; the Perception module is responsible for perceiving and processing multimodal information from the external environment; the Action module is responsible for using tools to execute tasks and affect the surrounding environment.

[0039] The internal information path of the agent is Perception -> Brain -> Action. This is similar to the transmission of human nerve signals and determines some issues such as the range of information that the agent can represent, information loss, and computational efficiency. Macroscopically, the agent should complete a task through one or more rounds of "input -> processing -> output", and whether to end the task should be determined by external or internal feedback information. Generally speaking, an agent based on an LLM is an entity that takes the LLM as the core, acts in a certain external environment, and can perform some form of input and output.

[0040] In this application, each agent takes the LLM as the base and the principle of Reasoning and Acting (ReACT) as the core idea, customizes and constructs the required tool set, and is guided by a specific prompt, so as to enable the agent to automatically disassemble task modules, build task chains, observe task status, and formulate task strategies.

[0041] For example, the prompt design of the product solution recommendation agent can be as follows: Currently, you are a business acceptance staff member of the intelligent computing network project team. Your job is to communicate with users, understand and identify the users' intentions, and complete relevant requirement operations. Usually, users will describe their requirements in the integrated monitoring service. You need to provide the five capabilities of "network", "storage period", "AI", "link", and "computing power" that can meet the users' requirements.

[0042] You need to answer the following questions to the best of your ability.

[0043] It should be clear that no one can help you, and you must try to solve all problems independently.

[0044] If you have clearly defined questions, you can answer directly.

[0045] If you are unsure or unable to give an accurate answer to some questions, you can visit the following tool list to try to find available tools: {tool_descs}; At the same time, a solution verifier will verify the solution you finally generated. You need to fully listen to their suggestions, optimize and reflect based on the relevant suggestions, and adjust the solution.

[0046] If there are no suggestions, you can come up with your own ideas.

[0047] If the suggestion is "reasonable", the output will be based on the current solution.

[0048] All historical information available for reference (including the user's previous requirements, the solutions you previously provided, and the modification suggestions given by the verifier or the verifier's feedback to the user will be stored in public_history, which you need to refer to as a summary of the previous situation).

[0049] Please be sure to use the following format: Question: An input question that you must answer; Think: You should always think about what to do; Action: Answer directly or select a tool (the action to be taken should be one of [{tool_names}]); Action input: input of the action; Observation: the result of an action; …(this thought / operation / action input / observation can be repeated zero or more times); I think: "I now know the final answer, the answer is: ***" or "I can't provide a solution this time, I need to give feedback to the user, the feedback content is: ***"; Final answer: The final answer to the original input question; start; Historical interaction information available for reference: {public_history}; Question: {query}.

[0050] The business acceptance system in this application uses a dual-agent nested collaboration model. The product solution recommendation agent takes the lead in generating business solutions. The generated solutions are then passed to the optimization and reflection agent, which verifies the artificial intelligence (AI) capabilities, selected resource pools, and selected link information in the product solution. Ultimately, the optimization and reflection agent interprets the user's intent into a structured, deterministic resource solution, completing the business acceptance.

[0051] Understandably, in the intelligent computing and networking system, users have a need to order intelligent monitoring products for their monitoring devices. Therefore, in this application, users can input their requirement information into the service acceptance system as monitoring service requirement information. For example, users can input monitoring service requirement information such as "I hope to enable the cameras in the parking lot to have the ability to recognize license plates, require high-definition viewing of the playback, store the video in the cloud for 7 days, select resources with lower costs and links with lower latency, and what solutions are recommended".

[0052] Furthermore, the monitoring service requirement information can be input into the first intelligent agent, and the first intelligent agent generates a solution as the initial monitoring service solution based on the input monitoring service requirement information.

[0053] Furthermore, this application can input the initial monitoring service solution and the monitoring service requirement information into the second intelligent agent. The second intelligent agent verifies each piece of information in the initial monitoring service solution in combination with the monitoring service requirement information, and the second intelligent agent outputs the solution verification result of passing the solution, feedback to the user, or suggesting modification.

[0054] The core design point of this application is different from the conventional multi-Agent peer-to-peer collaboration mode in the current industry. Instead, it adopts a dual-Agent nested collaboration mode, where one Agent completes the optimization and reflection work of another Agent. The key point to realize this design lies in the Memory module design in this application. This application adopts a form of combining a public memory area and a private memory area, enabling each Agent to independently complete its respective job on the one hand, and on the other hand, to obtain the public information needed to cooperate.

[0055] Furthermore, this application can determine the target monitoring service solution according to the corresponding processes after different solution verification results of passing the solution, feedback to the user, or suggesting modification.

[0056] According to the monitoring service solution generation method of the embodiments of this application, after receiving the monitoring service requirement information, first, the first intelligent agent generates a service solution based on the monitoring service requirement information to obtain the initial monitoring service solution; then, the second intelligent agent verifies the initial monitoring service solution in combination with the monitoring service requirement information. Further, according to the solution verification result, the target monitoring service solution can be generated. Thus, by changing the service acceptance mode arranged in a fixed process to a task-driven dual-intelligent-agent collaboration mode, the second intelligent agent can verify the initial monitoring service solution generated by the first intelligent agent. Furthermore, when the solution verification result is to feedback to the user, the monitoring service solution can be regenerated according to the content supplemented by the user, and when the solution verification result is to suggest modification, the monitoring service solution can be regenerated according to the modification suggestion. Therefore, the accuracy of the finally generated monitoring service solution can be improved.

[0057] Based on the above embodiments, the first intelligent agent is used for: Generating a dependency thinking chain for the required capabilities of the monitoring service requirement information according to the principle of coordinated reasoning and action; wherein, the required capabilities include at least one or more of computing power capabilities, network capabilities, link capabilities, artificial intelligence capabilities, and storage cycle capabilities; Invoking tools to parse each single-point capability in the dependency thinking chain and generating an initial monitoring service plan, storing the plan generation process in the private memory area of the first intelligent agent, and storing the monitoring service requirement information and the initial monitoring service plan in the public memory area.

[0058] Specifically, in this application, the business system can transmit the monitoring service requirement information to the product solution recommendation Agent through the interaction entrance. The Agent generates a dependency thinking chain for the five major capabilities of "computing power", "network", "link", "AI", and "storage cycle" required for the monitoring service requirement information (that is, finally completing business acceptance) according to the ReACT principle, and respectively invokes tools to parse each single-point capability to generate the current plan as the initial monitoring service plan. Among them, ReACT is that the large language model can construct a complete series of actions (Act) according to logical reasoning (Reason) to achieve the expected goal. The key to the ReAct method is to coordinate the large language model and external information acquisition and interact with other functions: the large language model is the brain, and through the ReAct framework, the brain can control the hands and feet.

[0059] The essence of the Chain of Thought (COT) is to decompose complex tasks into multiple simple subtasks. It refers to a sequence of continuous logical reasoning steps or associations in a thinking process, which is a series of interconnected ideas, viewpoints, or concepts in the thinking process. The Chain of Thought is usually used to solve problems, make decisions, or perform reasoning. It can connect and organize thinking in a logical order, decompose complex problems into simpler steps or concepts, so as to better understand and solve problems.

[0060] The generation of the specific plan can be achieved through the following process: For AI capabilities: The Retrieval-Augmented Generation (RAG) method is adopted to obtain necessary industry knowledge and historical experience information for enhancing the parsing accuracy from the industry knowledge base of integrated monitoring for reference. The specific AI capability name is obtained through intention understanding and analysis by the LLM based on the results retrieved by RAG and combined with the user description content.

[0061] It should be noted that before the emergence of large models, RAG mainly focused on the direct optimization of end-to-end models. Common practices in retrieval include dense retrieval, such as using vector-based dense passage retrieval, and training smaller models in generation. Due to the relatively small overall parameter scale, the retriever and generator usually undergo synchronous end-to-end training or fine-tuning. Since the emergence of LLMs, generative language models have become mainstream and demonstrated impressive performance in various language tasks. However, LLMs still face issues such as hallucinations, knowledge updates, and data-related problems. This affects the reliability of LLMs, and in some serious task scenarios, especially knowledge-intensive tasks that require access to a large amount of knowledge, such as open-domain question answering and common sense reasoning, LLMs perform poorly. The implicit knowledge within the parameters may be incomplete and insufficient. Incorporating RAG into the in-context learning of large language models in this application can alleviate the above problems, with significant and easily achievable effects. During the inference process, RAG dynamically retrieves information from external knowledge sources and uses the retrieved data as a reference to organize answers. This significantly improves the accuracy and relevance of responses, effectively solving the hallucination problem existing in LLMs. This technology has quickly gained attention since the advent of LLMs and has become one of the most popular techniques for improving chatbots and making LLMs more practical. By separating real-time knowledge from the training parameters of LLMs, RAG cleverly combines the powerful capabilities of generative models and the flexibility of retrieval modules, providing an effective solution to the problem of incomplete and insufficient knowledge inherent in pure parametric models.

[0062] For network capabilities and storage cycle capabilities: Use preset formulas for calculation to obtain the bandwidth resources required for playing back videos at a specified clarity and the storage resources required for a specified storage cycle.

[0063] For the selection of resource pools: Call the interface of the computing and network awareness system, and the system will feedback a list of resource addresses with this AI capability. The service acceptance system selects a suitable resource pool according to the "performance priority" strategy or "cost priority" specified by the user. When the user does not specify an activation strategy, the default resource pool provided by the computing and network awareness system is selected.

[0064] Performance priority strategy formula: Let the set of resource pools be , and each resource pool 's performance score is determined by the following four parameters: CPU performance , memory performance , GPU performance , and hard disk read / write speed . Then the comprehensive performance score of the resource pool can be expressed as: ; Among them, , , , are the weights of CPU, memory, GPU, and hard disk read / write speed respectively, and satisfy .

[0065] Under the performance - priority strategy, the selected resource pool is: ; Cost - priority strategy formula: Suppose the set of resource pools is still , and the price of each resource pool is , the minimum CPU performance required by the user is , the minimum memory performance is , the minimum GPU performance is , and the minimum hard disk read / write speed is . Then under the cost - priority strategy, the selected resource pool satisfies the following conditions: ; ; ; ; Under these conditions, select the resource pool with the lowest price: ; Among them, needs to meet all the above performance requirements.

[0066] For the selection of the link: After determining the location of the resource pool, call the computing and network perception system interface. The system allocates a list of optional AZ link sets from the user side to the resource side according to the user's address and the location of the selected resource pool. The service acceptance system makes a suitable link selection according to the "delay - priority" strategy or "smooth - priority" strategy specified by the user. When the user does not specify an opening strategy, the default link provided by the computing and network perception system is selected.

[0067] Delay - priority strategy formula: Suppose the set of candidate links is , and each link is composed of a series of nodes , and the delay between every two nodes is , where .

[0068] Then the total delay of the link can be expressed as: .

[0069] Under the delay priority strategy, the selected link for: .

[0070] Smooth priority strategy formula: Assume that the link set is still , each link It consists of a series of nodes. The bandwidth usage between each two nodes is ,in .

[0071] Then the link Maximum bandwidth usage It can be expressed as: .

[0072] Under the smooth priority policy, the selected link for: .

[0073] in, The highest occupancy rate among all links The lowest link represents the most unobstructed path in the network.

[0074] Thus, the product solution recommendation agent can follow the ReACT model to gradually observe, reflect, decide, and act, generating a final answer as the initial monitoring solution for this recommendation. The steps of observation, reflection, decision-making, and action during the solution generation process are stored in the first agent's independent private memory area, the private history, which serves as the basis for ReACT's iterative thinking. The final answer, the initial monitoring solution, is stored in the public history area and simultaneously transmitted to the optimization and reflection agent.

[0075] In this application, the product solution recommendation agent is responsible for generating solutions based on the user's needs description, which can quickly generate the initial monitoring business solution and improve the efficiency of generating monitoring business solutions.

[0076] Based on the above embodiment, the second intelligent agent is used to adopt a reinforcement learning strategy from human feedback, generate a solution verification result based on the initial monitoring business plan and the monitoring business demand information, store the verification process in the private memory interval of the second intelligent agent, and store the feedback user or suggested modifications in the monitoring business demand information and the solution verification result in the public memory interval.

[0077] Specifically, in this application, the optimized reflection Agent can adopt the Reinforcement Learning from Human Feedback (RLHF) strategy to align the human intentions of the LLM. Thus, it can verify the initial monitoring service plan in combination with the monitored business requirement information and output the verification results of passing the plan, feeding back to the user, or suggesting modifications.

[0078] In this application, the reflection optimization Agent can verify the AI capabilities, resource pools, and link information in the initial monitoring service plan according to a pre-set orchestration process: among which, the verification of AI capabilities is completed by the LLM's own intention understanding ability combined with the RAG knowledge base retrieval ability, and the rationality verification of resource pools and link information is inferred through the model aligned by RLHF described above. Among them, the verification results of each step are stored in the private memory interval private history of the second intelligent agent, and finally, the verification results of all modules are extracted and analyzed by the LLM. If abnormal feedback to the user is required or the plan needs to be modified, the modification suggestions or feedback information and the monitored business requirement information are stored in the public memory interval public history together.

[0079] Specifically, the verification of the reflection optimization Agent can be achieved through the following process: Let D be the historical data set, which contains the live metric data of the requirements confirmed and accepted by the user during actual operation. The training process of the optimized reflection Agent can be expressed as: : Among them, R represents the set of metric parameters of the resource pool, and L represents the set of metric parameters of the link. The specific metric parameters include: Resource pool metrics: ; Link metrics: ; Among them, is the queries per second rate, is the fastest processing speed, is the slowest processing speed, is the packet loss rate, is the network jitter.

[0080] The analysis conclusion of the optimized reflection Agent can be expressed as: ; Among them, is the analysis conclusion, is the analysis function, Q is the user requirement input, is the current plan, that is, the initial monitoring service plan.

[0081] This application designs a solution for continuously iterating the optimized reflection Agent using the RLHF mode. In the product recommendation Agent, performance data is used for selective evaluation of the link and resource pool. When the order is placed for the solution, operation metric data is collected, and the operation metric data and user requirement description are presented to the expert for evaluation. The evaluation result forms key-value (KV) pairs to perform continuous RLHF reinforcement learning on the optimized reflection Agent. The theoretical performance parameters are judged by the real metric results to optimize the generated solution.

[0082] This application validates the initial monitoring service solution generated by the first intelligent agent through the second intelligent agent. Thus, when the solution validation result is to feedback to the user, the monitoring service solution can be regenerated according to the content supplemented by the user, and when the solution validation result is to suggest modification, the monitoring service solution can be regenerated according to the modification suggestion. Therefore, the accuracy of the finally generated monitoring service solution can be improved.

[0083] Based on the above embodiments, based on the solution validation result, determining the target monitoring service solution includes: If the solution validation result is that the solution passes, determine the initial monitoring service solution as the target monitoring service solution; If the solution validation result is to feedback to the user, output the feedback information in the solution validation result, input the supplementary information returned based on the feedback information, the monitoring service requirement information, and the initial monitoring service solution into the first intelligent agent, and obtain the target monitoring service solution output by the first intelligent agent; If the solution validation result is to suggest modification, input the modification suggestion in the solution validation result, the monitoring service requirement information, and the initial monitoring service solution into the first intelligent agent, and obtain the target monitoring service solution output by the first intelligent agent.

[0084] Specifically, after the solution is validated by the reflection and optimization Agent in this application, the solution validation results are divided into three types: the solution passes (P), feedback to the user (F), and suggest modification (M).

[0085] Specifically, when the initial monitoring service solution is determined as the target monitoring service solution and used as the final solution and presented to the user for confirmation: ; If the user has not subscribed, the status is pending; if the user confirms the subscription, execute the subscription process: ; Among them, represents the confirmation status of the user for the final solution After the user confirms, execute the solution subscription process, and send the user requirement Q and the final solution Write to the historical experience library E: , while regularly collecting real-time indicator data and , combined with the solution information for manual review to form new labeled data , to continuously train the model: .

[0086] In this application, on the one hand, the historical experience library supports the manual upload of offline user requirements and the corresponding recommended solution KV pair sample data. On the other hand, it will save the online user questions and the KV pair data of the final answers given by the business acceptance Agent in real time. The online data supports manual evaluation and labeling. The finally determined reasonable solutions and the local data uploaded offline will be used as reference samples to match with the user requirements in the form of RAG, for the product solution recommendation Agent to refer to when deconstructing user requirements.

[0087] Furthermore, according to different strategies of "modification suggestions" or "feedback content", this application executes different subsequent processes. If it is necessary to modify the existing solution, the user input, recommended solution, and modification suggestions are extracted from the public memory area public history and then passed into the solution recommendation Agent again to generate the solution for the second time according to the modification suggestions. If it is necessary to give feedback to the user, after the feedback information is given, wait for the user's re-input, and pass the original user input information, recommended solution, and user's new input information to the solution recommendation Agent for complete solution completion.

[0088] When , it means that the situation where the user input information is missing and a complete solution cannot be generated is encountered. At this time, the LLM module in the optimization and reflection Agent generates feedback information for the missing content , guiding the user to conduct multi-round conversations to supplement the required information: ; Furthermore, the supplementary information returned based on the feedback information, the monitored business requirement information, and the initial monitored business solution are input into the first intelligent agent to obtain the target monitored business solution output by the first intelligent agent.

[0089] When , that is, the analysis conclusion is to recommend modification. At this time, the optimization and reflection Agent will generate modification suggestions , and feedback it to the product solution recommendation Agent to regenerate the solution: ; Among them, h represents the ability to generate modification suggestions designed in the optimization and reflection Agent, and s is the original solution. The product solution recommendation Agent regenerates the solution according to to improve the original solution.

[0090] That is, the modification suggestions, monitoring service requirement information in the solution verification result and the initial monitoring service solution are input into the first intelligent agent to obtain the target monitoring service solution output by the first intelligent agent.

[0091] In this application, the product solution recommendation Agent is responsible for generating an intention deconstruction solution according to the user requirement description. The optimization and reflection Agent verifies the AI capabilities, selected resource pools, and selected links of the generated solution. At the same time, the interaction between agents is completed in the form of a combination of a private memory area + a public memory area set in the Memory module, isolating the iteration within the intelligent body and the interaction between intelligent bodies, so as to achieve the purpose of better parsing user requirements and continuous optimization, and can improve the accuracy of the finally generated monitoring service solution.

[0092] The purpose of this application is to provide a multi-dimensional fusion solution acceptance process for computing power, network, link, AI, and storage cycle for users through the intelligent task splitting, task decision-making, task execution, task observation, and solution optimization modes of intelligent agents. The user interacts through voice on the application side and describes the requirements in an unstructured and uncertain expression way. The service acceptance system adopts a dual-Agent nested cooperation mode, with the product solution recommendation Agent as the leader to generate a service solution. The generated solution result is transmitted to the optimization and reflection Agent through the public historical record. The optimization and reflection Agent verifies the AI capabilities, selected resource pools, and selected link information in the product solution. Finally, the user's intention is parsed into a structured and deterministic resource solution to complete the service acceptance.

[0093] Next, the monitoring service solution generation device provided by this application will be described. The monitoring service solution generation device described below can be correspondingly referred to the monitoring service solution generation method described above.

[0094] Furthermore, this application also provides a monitoring service solution generation device.

[0095] The monitoring service solution generation device includes: A generation module, configured to receive monitoring service requirement information, input the monitoring service requirement information into a first intelligent agent, and obtain an initial monitoring service solution output by the first intelligent agent; the first intelligent agent is generated based on a large language model and is used for generating a service solution based on the input information; A verification module, configured to input the initial monitoring service solution and the monitoring service requirement information into a second intelligent agent, and obtain a solution verification result output by the second intelligent agent; the second intelligent agent is generated based on a large language model and is used for verifying the service solution in the input information based on the requirement information in the input information; the solution verification result is that the solution passes, feedback to the user, or suggest modification; A determination module, configured to determine a target monitoring service solution based on the solution verification result.

[0096] After receiving the monitoring service requirement information, the monitoring service solution generation device of the present application first generates an initial monitoring service solution through a first intelligent agent based on the monitoring service requirement information, and then verifies the initial monitoring service solution by combining the monitoring service requirement information through a second intelligent agent. Further, a target monitoring service solution can be generated according to the solution verification result. Thus, by changing the service acceptance mode arranged in a fixed process to a task-driven dual-intelligent-agent collaboration mode, the second intelligent agent can verify the initial monitoring service solution generated by the first intelligent agent. Furthermore, when the solution verification result is to feedback to the user, the monitoring service solution can be regenerated according to the content supplemented by the user, and when the solution verification result is a suggestion for modification, the monitoring service solution can be regenerated according to the modification suggestion. Therefore, the accuracy of the finally generated monitoring service solution can be improved.

[0097] In one embodiment, the determination module is specifically configured to: If the solution verification result is that the solution passes, determine the initial monitoring service solution as the target monitoring service solution; If the solution verification result is to feedback to the user, output the feedback information in the solution verification result, input the supplementary information returned based on the feedback information, the monitoring service requirement information, and the initial monitoring service solution into the first intelligent agent, and obtain the target monitoring service solution output by the first intelligent agent; If the solution verification result is a suggestion for modification, input the modification suggestion in the solution verification result, the monitoring service requirement information, and the initial monitoring service solution into the first intelligent agent, and obtain the target monitoring service solution output by the first intelligent agent.

[0098] Figure 4 An example of the physical structure diagram of an electronic device is shown as Figure 4 As shown, the electronic device may include: a processor 410, a communication interface 420, a memory 430, and a communication bus 440. Among them, the processor 410, the communication interface 420, and the memory 430 communicate with each other through the communication bus 440. The processor 410 can call the logical instructions in the memory 430 to execute the following method: receive the monitoring service requirement information, input the monitoring service requirement information into the first intelligent agent, and obtain the initial monitoring service solution output by the first intelligent agent; the first intelligent agent is generated based on a large language model and is used to generate a service solution based on the input information. Input the initial monitoring service plan and the monitoring service requirement information into the second intelligent agent to obtain the plan verification result output by the second intelligent agent; the second intelligent agent is generated based on a large language model and is used to verify the service plan in the input information based on the requirement information in the input information; the plan verification result is that the plan passes, feedback to the user, or suggest modification; Determine the target monitoring service plan based on the plan verification result.

[0099] In addition, when the logical instructions in the above-mentioned memory 430 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the related technology, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of this application. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0100] In another aspect, an embodiment of this application also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it is configured to execute the methods provided in the above-mentioned various embodiments, for example, including: receiving monitoring service requirement information, inputting the monitoring service requirement information into the first intelligent agent to obtain the initial monitoring service plan output by the first intelligent agent; the first intelligent agent is generated based on a large language model and is used to generate a service plan based on the input information; Input the initial monitoring service plan and the monitoring service requirement information into the second intelligent agent to obtain the plan verification result output by the second intelligent agent; the second intelligent agent is generated based on a large language model and is used to verify the service plan in the input information based on the requirement information in the input information; the plan verification result is that the plan passes, feedback to the user, or suggest modification; Determine the target monitoring service plan based on the plan verification result.

[0101] In another aspect, an embodiment of the present application further provides a computer program product, on which a computer program is stored. When the computer program is executed by a processor, it is configured to execute the methods provided in the above embodiments. For example, it includes: receiving monitoring service requirement information, inputting the monitoring service requirement information into a first intelligent agent, and obtaining an initial monitoring service plan output by the first intelligent agent; the first intelligent agent is generated based on a large language model and is used to generate a service plan based on input information. Inputting the initial monitoring service plan and the monitoring service requirement information into a second intelligent agent, and obtaining a plan verification result output by the second intelligent agent; the second intelligent agent is generated based on a large language model and is used to verify the service plan in the input information based on the requirement information in the input information; the plan verification result is that the plan passes, feedback to the user, or suggest modification. Based on the plan verification result, determine the target monitoring service plan.

[0102] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative work.

[0103] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the essence of the above technical solution, or the part that contributes to the related technology, can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0104] Finally, it should be noted that the above embodiments are only used to illustrate the present application, rather than to limit the present application. Although the present application has been described in detail with reference to the embodiments, those of ordinary skill in the art should understand that various combinations, modifications, or equivalent replacements of the technical solutions of the present application do not depart from the spirit and scope of the technical solutions of the present application.

Claims

1. A method for generating a monitoring service solution, characterized in that, Including: Receiving monitoring service requirement information, inputting the monitoring service requirement information into a first intelligent agent, and obtaining an initial monitoring service plan output by the first intelligent agent; The first intelligent agent is generated based on a large language model and is used to generate a service plan based on input information; Inputting the initial monitoring service plan and the monitoring service requirement information into a second intelligent agent, and obtaining a plan verification result output by the second intelligent agent; The second intelligent agent is generated based on a large language model and is used to verify the service plan in the input information based on the requirement information in the input information; The plan verification result is that the plan passes, feedback to the user, or suggestion for modification; Based on the plan verification result, determining a target monitoring service plan.

2. The method for generating a monitoring service plan according to claim 1, wherein The first intelligent agent is used for: Generating a dependency thinking chain for the required capabilities of the monitoring service requirement information according to the principle of coordination of reasoning and action; wherein, the required capabilities include at least one or more of computing power capabilities, network capabilities, link capabilities, artificial intelligence capabilities, and storage cycle capabilities; Invoking tools to parse each single-point capability in the dependency thinking chain and generating an initial monitoring service plan, storing the plan generation process in the private memory area of the first intelligent agent, and storing the monitoring service requirement information and the initial monitoring service plan in the public memory area.

3. The method for generating a monitoring service solution according to claim 2, wherein The computing power capabilities include the selection of resource pools; the resource pool selection strategies include a performance priority strategy, a cost priority strategy, and a preset resource pool selection strategy.

4. The method for generating a monitoring service solution according to claim 2, wherein The link capabilities include the selection of links; the link selection strategies include a latency priority strategy, a smoothness priority strategy, and a preset link selection strategy.

5. The method for generating a monitoring service solution according to claim 1, wherein The second intelligent agent is used to adopt a reinforcement learning strategy from human feedback, generate a plan verification result according to the initial monitoring service plan and the monitoring service requirement information, store the verification process in the private memory area of the second intelligent agent, and store the monitoring service requirement information and the feedback to the user or the suggestion for modification in the plan verification result in the public memory area.

6. The method for generating a monitoring service solution according to claim 1, wherein, The determining the target monitoring service plan based on the plan verification result includes: If the plan verification result is that the plan passes, determining the initial monitoring service plan as the target monitoring service plan; If the plan verification result is feedback to the user, outputting the feedback information in the plan verification result, inputting the supplementary information returned based on the feedback information, the monitoring service requirement information, and the initial monitoring service plan into the first intelligent agent, and obtaining the target monitoring service plan output by the first intelligent agent; If the plan verification result is a suggestion for modification, inputting the modification suggestion in the plan verification result, the monitoring service requirement information, and the initial monitoring service plan into the first intelligent agent, and obtaining the target monitoring service plan output by the first intelligent agent.

7. A monitoring service solution generation device, characterized in that, Including: A generation module, configured to receive monitoring service requirement information, input the monitoring service requirement information into a first intelligent agent, and obtain an initial monitoring service plan output by the first intelligent agent; The first intelligent agent is generated based on a large language model and is used to generate a service plan based on input information; A verification module, configured to input the initial monitoring service plan and the monitoring service requirement information into a second intelligent agent, and obtain a plan verification result output by the second intelligent agent; The second intelligent agent is generated based on a large language model and is configured to verify a service plan in input information based on the requirement information in the input information; The plan verification result is that the plan passes, feedback to the user, or suggest modification; A determination module, configured to determine a target monitoring service plan based on the plan verification result.

8. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the monitoring service plan generation method according to any one of claims 1-6.

9. A storage medium, the storage medium being a non-transitory computer-readable storage medium, on which a computer program is stored, characterized in that, When the computer program is executed by a processor, it implements the monitoring service plan generation method according to any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the monitoring service plan generation method according to any one of claims 1-6.

Images