A large-scale intelligent agent application system for the water conservancy industry

Abstract

This invention provides a large-scale intelligent agent application system for the water conservancy industry, belonging to the field of intelligent water conservancy. It includes: a large language model base unit for providing a standard interface to interact with open-source or non-open-source LLMs and respond to user requests; a business knowledge base unit whose training dataset is used to accumulate knowledge for training the LLM, and the knowledge base dataset is used to support customized requirements, expressing customized business as knowledge for intelligent agents to call; a large-scale model-specific water conservancy component library unit for providing a management platform for all large-scale model-specific components; a water conservancy large-scale model service decision support system unit for assembling business components into an application system and providing operation tools to adjust the assembly process to form an applicable system; and a knowledge accumulation unit for extracting system knowledge generated by the water conservancy large-scale model service decision support system unit. The water conservancy application system architecture provided by this invention is fully functional, flexible and adaptable to changing business needs, and has a short and long usage cycle.

Description

Technical Field

[0001] This invention belongs to the field of intelligent water conservancy technology, specifically relating to a large-scale intelligent agent application system for the water conservancy industry. Background Technology

[0002] In recent years, generative AI technology, multimodal data processing and preprocessing, pre-training of large models, and reinforcement learning have become mainstream trends in the field of artificial intelligence applications. Today, LLM (Large Language Model) has had a significant impact on the AI ​​community, and the emergence of ChatGPT and GPT-4 has prompted a rethinking of the possibilities of AGI (Artificial General Intelligence). The adoption of these technologies has enabled AI applications to possess diverse interaction methods, the ability to process multiple tasks in parallel, and intelligent information perception, making them a global development direction for AI technology applications.

[0003] Currently, the entire industry is attempting to leverage LLM to empower water conservancy operations and create intelligent software. However, to date, no truly intelligent application has emerged that meets actual business needs. While some simple applications exist, they lack deep integration into industry operations, resulting in subpar performance. A mature system framework has not yet been established to deeply penetrate the industry and integrate with new technologies to generate unified applications.

[0004] Under the traditional interactive model of information system design, existing water conservancy industry application systems have two major drawbacks: First, due to their architectural limitations, existing business application systems are fixed after development and delivery, resulting in complex operation and a lack of business flexibility, making it difficult to adapt to the changing needs of the industry; Second, currently, systems combining LLM with the water conservancy industry are still mainly limited to intelligent question answering and text interaction, and no in-depth industry business applications have yet emerged. Summary of the Invention

[0005] To address the issues of complex operation, insufficient business flexibility, and low industry integration in existing water conservancy industry application systems, this invention provides a large-scale intelligent agent application system for the water conservancy industry.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A large-scale intelligent agent application system for the water conservancy industry includes:

[0008] The Large Language Model Base Unit provides a standard interface for interacting with open-source or non-open-source Large Language Models (LLMs) and responding to user requests.

[0009] The business knowledge base unit includes a training dataset for training the LLM model and a knowledge base dataset for LLM add-ons. The training dataset is used to accumulate knowledge to train the LLM, and the knowledge base dataset is used to support customized needs, expressing customized business as knowledge for intelligent agents to call.

[0010] The large-scale model-specific hydraulic component library unit is used to provide a management platform for all large-scale model-specific components, provide component service functions, and respond to system requirements.

[0011] The water conservancy big model service decision support system unit is used to assemble business components into an application system and provide operation tools to adjust the assembly process to form an applicable system. According to the dynamic changes in demand, the system can be modified by adjusting the components to adapt to the changes in demand.

[0012] The knowledge accumulation unit is used to extract system knowledge generated by the water conservancy big model service decision support system unit, and then organize the knowledge and incorporate it into the dataset.

[0013] Preferably, the large language model base unit responds to user requests, specifically including:

[0014] When the large language model base unit receives a user request, it performs intent recognition of the request. First, the request is vectorized to obtain a request vector, and then the request vector is classified to obtain a classification result.

[0015] The classification results are returned to the LLM. It is then determined whether the business knowledge base unit needs to participate. If so, the business knowledge base unit is invoked, and the results are reconstructed based on the output of the business knowledge base unit. The LLM then processes and outputs the results. If the business knowledge base unit does not need to participate, the LLM processes the results directly and outputs them directly.

[0016] Preferably, the standard interface is an interface that takes a string as input and returns a string.

[0017] Preferably, the operation of responding to user requests specifically includes:

[0018] The business knowledge base is divided into topic domains. User needs are matched through intent recognition, and then LLM provides recommended topic domains. Under each topic domain, all business descriptions are vectorized to obtain business description vectors, and then a matching process is performed. After matching the corresponding business, the business process description standard and the corresponding business input are used as prompt words in LLM. Then, LLM generates business process content, forming a business expression chain.

[0019] Preferably, the large-scale model-specific hydraulic component library unit is specifically used for:

[0020] Provide a unified access address for component registration;

[0021] The service strategy for embedded components, and the component service governance strategy;

[0022] Access to the component library unit is granted through authentication and authorization;

[0023] All connected components are managed in a unified group, and blacklist / whitelist groups are created.

[0024] After the component is invoked, a log of the request access is generated;

[0025] It has a built-in authentication system for authorization, auditing, and access control when accessing the component library through other means.

[0026] Preferably, the large-scale model dedicated water conservancy component library unit includes the following for component access: component purpose and scope of application, component input parameters, component output parameters, and component protocol authentication.

[0027] Preferably, the large-scale model dedicated water conservancy component library unit responds to the LLM's component retrieval request, decomposes the generated business description chain, feeds it back to the LLM, gradually finds the corresponding component to match it, feeds back the upstream and downstream relationships of the component to the LLM, and generates the corresponding component retrieval relationship; integrates the business description chain to generate business component relationship, and passes the business component relationship to the water conservancy large-scale model service decision support system unit to generate the business application system.

[0028] Preferably, the water conservancy large model service decision support system unit includes:

[0029] The BS architecture is a browser client architecture that runs within a browser.

[0030] A knowledge graph generation strategy is used to generate a logical topology knowledge graph based on the business expression chain, and to provide an interface that can be modified on each generated topology graph node;

[0031] A knowledge graph manipulation toolset for modifying nodes in a knowledge graph, including functions for adding, deleting, and modifying nodes;

[0032] Generate standard knowledge strategies so that business systems generated based on knowledge graphs can be output as text data that conforms to business process description standards.

[0033] Log recording is used to generate system process logs.

[0034] Preferably, the operation of the water conservancy large model service decision support system unit for component access is as follows:

[0035] The generated business description chain is input into this unit, which parses the relationships and generates a business process topology framework diagram. This diagram is then sent as a draft to the front-end system interface to determine the initial version of the business process topology framework diagram. The system is then run to generate the business application system. The system is then tested, and this unit supports system rollback corrections until it meets business requirements. At this point, the correction process is terminated, and the business application system is obtained.

[0036] Preferably, the knowledge accumulation unit performs the following operations for knowledge accumulation: extracting knowledge from the generated business application system, cleaning and data mining of the text knowledge, and entering the business knowledge pool, where the knowledge is managed; configuring the knowledge in the knowledge pool into a knowledge base specific to a particular region to provide add-on services to LLM, and using it as a training dataset for LLM to guide the emergence of industry-specific business optimization methods.

[0037] The beneficial effects of the large-scale intelligent agent application system for the water conservancy industry proposed in this invention are as follows:

[0038] This invention responds to industry needs by using an LLM (Limited Learning Model) combined with a series of tools to generate business practices that meet those needs. Business components are then transferred from a component library, and the components are assembled by the Water Resources Big Data Model Service Decision Support System to generate a business system. Finally, the business system transfers knowledge back to the LLM, forming a closed-loop, universal solution paradigm for the water resources industry. Its advantages are:

[0039] Firstly, it can deeply integrate water resources knowledge with LLM, enabling LLM to leverage its reasoning and emergent abilities when facing water resources professional business. Combined with water resources professional knowledge, it can generate business practices across the entire industry. Theoretically, with sufficient training data, LLM can even provide business understanding practices that surpass those of industry experts, thanks to its extensive learning of a large amount of basic knowledge that is difficult for humans to achieve.

[0040] Next, the business logic is described in a standardized way through business knowledge base units, which is beneficial to the long-term development of the industry. A unified description method will enable the large model to better understand the intentions of personnel needs.

[0041] By using a dedicated hydraulic component library for large models to modularize all functional modules, all systems can be decoupled, making the resulting business application system more flexible and able to respond to changes in requirements in a timely manner. On the other hand, modularized modules also eliminate the need for repetitive development work, and each component can be reused.

[0042] The decision support system unit for water conservancy big model services can flexibly assemble all components. At this stage, industry professionals participate in the system generation, making the generated system more in line with the needs of actual applications. Because it is based on a component-based assembly system, the system development cycle is short and the timeliness is strong. All generated business application systems can quickly respond to changes in demand, making the application system more flexible and easy to use.

[0043] By leveraging the knowledge accumulation unit, the industry expertise implicit in the generated business application system can be fed back to the LLM for learning. Its dataset can be used to train the LLM, forming a scientific closed-loop application model in which ideas guide practice, and practice in turn provides feedback to guide ideas. Attached Figure Description

[0044] To more clearly illustrate the embodiments and design schemes of the present invention, the accompanying drawings required for this embodiment will be briefly described below. The drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 This is a block diagram of the water conservancy industry large-scale intelligent agent application system according to Embodiment 1 of the present invention; Detailed Implementation

[0046] To enable those skilled in the art to better understand and implement the technical solutions of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and should not be construed as limiting the scope of protection of the present invention.

[0047] To address the shortcomings of existing water conservancy application systems, such as their simplistic architecture, inflexibility in adapting to changing business needs, and short lifecycles, a more efficient, professional, accurate, and secure intelligent agent application system based on a large language model for the water conservancy industry, along with its development framework, is required to meet the comprehensive needs of the water conservancy sector. This new technology needs to overcome the deficiencies and problems of existing technologies, requires guidance from interdisciplinary industry expertise, and necessitates a dynamic response mechanism and closed-loop learning logic to address the new demands and challenges of the water conservancy industry.

[0048] In conclusion, with the continuous development of artificial intelligence and water conservancy technologies, smart water conservancy has become the industry's development direction. The application of intelligent agents based on large language models in the water conservancy industry will challenge traditional applications, defining a new paradigm for water conservancy business software applications and improving the intelligence level, application efficiency, information-based decision-making level, and knowledge accumulation of business operations. In the future, after accumulating sufficient professional and useful knowledge, this technology will generate emergent capabilities—the ability to generate solutions to industry problems. These capabilities can be combined with existing intelligent equipment in the industry to create more valuable applications, realizing comprehensive, efficient, and highly secure smart applications for the water conservancy industry.

[0049] Example 1

[0050] Based on this, the present invention provides a large-scale intelligent agent application system for the water conservancy industry. Specifically, it is a system capable of grasping the core of water conservancy business analysis and decision-making, supporting full-process information development, and possessing dynamic adaptability. Specifically, as follows... Figure 1 As shown, this system identifies and processes water conservancy business needs through a large language model base unit, generates business application steps based on these needs using a business knowledge base unit, outputs the necessary components for these business steps using a large model-specific water conservancy component library unit, assembles these components using a water conservancy large model service decision support system unit, fine-tunes them by industry application personnel to generate the application system, and finally accumulates and reuses business knowledge using a knowledge accumulation unit. The entire process is closed-loop; each application is a feedback learning process, making the intelligent agent more aligned with the user's needs, forming a virtuous cycle. This achieves a new application paradigm definition for the water conservancy industry, ensuring security, efficiency, intelligence, and knowledge accumulation. It provides the water conservancy industry with a solution for the development and application of large model intelligent agent software that is usable, easy to use, manageable, essential, frequently used, and enjoyable.

[0051] Specifically, this invention provides a water conservancy industry large-scale intelligent agent application system based on a large language model, including a large language model base unit, a business knowledge base unit, a large-scale model-specific water conservancy component library unit, a water conservancy large-scale model service decision support system unit, and a knowledge accumulation unit. The functions of each unit are further explained below.

[0052] Large Language Model Base Unit:

[0053] The Large Language Model (LLM) foundation unit is a collection for supporting LLMs (All Common Large Language Models). Within this unit, there are method classes for interacting with most mainstream LLMs. Its functionality is universal across all LLM types, providing a standard interface for interacting with both open-source and closed-source LLMs. This interface takes strings as input and returns strings, conforming to the REST API standard. This allows for flexible selection of the LLM foundation based on specific application scenarios.

[0054] Open-source LLMs, due to their open-source nature, can be deployed locally or via cloud resources, depending on the specific needs. Their advantage lies in the complete control over the base model's capabilities, allowing for flexible training of all parameters or fine-tuning of specific parameters. Examples include GLM, Llama2, and Gemini bases. Closed-source (i.e., not open-source) LLMs are often models with a large number of parameters, exhibiting superior performance in their fundamental capabilities. Currently, these LLMs offer services through standard interfaces and charge a fee; their capabilities depend on the depth of the owner's model development. Examples include GPT-4, Qwen, and ERNIE Bot. The large language model base unit contains method classes supporting both types of LLMs.

[0055] Furthermore, the large language model base unit responds to user requests, and the specific operation process of the response is as follows:

[0056] When the large language model base unit receives a user request, it performs intent recognition of the request. First, it embeds the request (vectorization) to obtain a request vector. This vector is then classified to obtain the classification result. The result is returned to the LLM to determine whether the business knowledge base unit needs to participate. If so, the business knowledge base unit is invoked, and the result is reconstructed based on its output. The LLM then processes and outputs the result. If the business knowledge base unit does not need to participate, the LLM processes the result directly and outputs it. For example: No knowledge base unit participation required: (Q: "Hello", A: "Hello, I am XX"). Knowledge base unit participation required: (Q: "Rainfall situation in XX area, how does the flood generated by this rainfall affect the XX section?" A: "① Obtain rainfall situation in XX area, call the rainfall query component to obtain the rainfall sequence; ② Obtain the location relationship between the XX section and XX area, call the location relationship component to obtain the geographical location topology; ③ Obtain the production-confluence model construction relationship, call the production-confluence model component to obtain the model calculation result; ④ Obtain the result display method, call the software application display component to obtain the software application interface").

[0057] Furthermore, when the large language model base unit needs to retrieve the business knowledge base unit, it will input the requirements and the business knowledge base into the LLM. The business knowledge base unit will be further introduced below.

[0058] Business knowledge base unit:

[0059] The business knowledge base unit is divided into a training dataset for training the LLM model and a knowledge base dataset for LLM add-ons. Their purposes differ: the former is used to accumulate knowledge for training the LLM, making it more suitable for industry applications; the latter supports customized needs, primarily used to express customized business requirements as knowledge for intelligent agents to use.

[0060] Since the intelligent agent is driven by a large language model, the storage type of the business knowledge base data is text; the storage format of business knowledge conforms to the business process description standard. Under this invention, the business process description standard is defined as follows: the business needs to be described in general, including but not limited to: (business theme, business description set, business type, business scope, etc.). The business knowledge is described step by step, and each step needs to have three elements (① definition: how to do things in this step ② function: what type of component this step needs to call ③ location: the relationship between this step and the steps above and below). The components involved in the steps need to have three elements (① definition: the purpose and scope of application of this component ② input parameters: what data this component needs ③ output parameters: what this component outputs).

[0061] Furthermore, the business knowledge base unit responds to user requests, and the specific operation process for responding is as follows:

[0062] First, the business knowledge base is divided into subject domains, such as (flood and drought disaster domain, irrigation area allocation domain, ecological compensation domain, forecasting and prediction domain, water rights trading domain, allocation and regulation domain, etc.). Each subject domain has descriptive text, keywords, etc., for that domain. The description is embedded to obtain a subject domain vector. Then, user needs are matched through intent recognition, and LLM provides recommended subject domains. Under that subject domain, all business descriptions are embedded to obtain business description vectors. Another round of matching is performed. After matching the corresponding business, the business process description standard is input into the LLM with the corresponding business to perform prompt (prompt word engineering). Then, the LLM generates the business process content, i.e., the business description chain.

[0063] Furthermore, the business knowledge base unit and the large language model base unit work together to generate a business expression chain, and extract corresponding components from the large model component library unit based on the business expression chain.

[0064] Large-scale model dedicated hydraulic component library unit:

[0065] The large-scale model-specific hydraulic component library unit provides a management platform for all large-scale model-specific components. It integrates all component services and deploys them as a separate layer to address service governance issues. This unit is used for adding, deleting, modifying, and querying components; it functions as a component library, allowing new components to be integrated into this unit. It mainly comprises six aspects:

[0066] 1. Provide a unified access address for component registration.

[0067] This unit supports the access of all external components with standard interface protocols. Due to the different development methods of components, some provide HTTP services, some provide RPC services, and some provide Web Service services. Within this unit, all components will be dynamically routed together, and different services will be converted into REST API interface specifications using protocols.

[0068] 2. Possesses component service governance strategies

[0069] Service strategies for components can be embedded within this unit, such as service circuit breaking, traffic offloading, and degradation.

[0070] 3. Authentication and Authorization

[0071] When integrating component interfaces that require authentication and authorization, such as OAuth protocol authentication, Cookie authentication, and Token authentication, this unit provides a unified entry point for processing. Configuration within this unit is all that's needed to integrate the component library unit.

[0072] 4. Blacklists and whitelists

[0073] This unit can manage all connected components in a unified group, such as blacklists and whitelists based on IP addresses.

[0074] 5. Log Recording

[0075] All components in this unit will have access request logs after they are invoked, which facilitates the maintenance of this unit.

[0076] 6. Unified Authentication Exposure

[0077] The unit has a built-in authentication system for authorization, auditing, and access control when accessing the component library in other ways.

[0078] Furthermore, the specific operation process for connecting components in the large-scale model-specific hydraulic component library unit is as follows:

[0079] Clarify the purpose and scope of application of the component. This item is a required option. This step is a text-based description. This description will be embedded within the unit to obtain a vector of component introduction, so that LLM can better understand the function of the component.

[0080] Clearly define what data this component requires. This option is mandatory. This step mainly clarifies the component's input parameters. Similarly, the vector of component input parameters is obtained through embedding based on the text description. For example, {time series data, integer}.

[0081] Clearly define what the component outputs. This option is mandatory. This step mainly clarifies the component's output parameters. Similarly, the output parameter vector is obtained by embedding the text description. For example, {GIS data, grid data}.

[0082] Specify how the component will be connected to the component library unit. This is a mandatory option. Choose the type of component to be connected from the three main service types, such as connecting via RPC service.

[0083] Specify the required protocol authentication for this component. This option is optional. If the component does not require authentication, it can be directly connected. If authentication is required, select the corresponding access method and configure the rules to connect to the unit.

[0084] Furthermore, the large-scale model-specific hydraulic component library unit responds to the LLM's component retrieval requests. The specific operation process of the response is as follows:

[0085] The business description chain generated in the previous step is broken down and fed back to the LLM decomposition method with an example. The decomposed steps are then used to find the corresponding components to match them. The upstream and downstream relationships of the components need to be fed back to the LLM with an example, and then the LLM generates the corresponding component retrieval relationship.

[0086] Furthermore, the large-scale water conservancy component library integrates business expression chains to generate business component relationships, and based on these relationships, transmits them to the water conservancy large-scale model service decision support system to generate business application systems.

[0087] Water Conservancy Large Model Service Decision Support System Unit:

[0088] The Water Conservancy Large Model Service Decision Support System Unit is an integrated application system designed and developed specifically for large model services. Its function is to assemble business components into an application system and provide operation tools to adjust the assembly process. After adjustment, an applicable system is directly formed. It can be dynamically changed according to needs. Only the components need to be adjusted to quickly modify the system to adapt to changes in needs. Finally, the knowledge of the assembled system is extracted and further transferred to the knowledge accumulation unit.

[0089] Furthermore, the main components of the water conservancy large model service decision support system unit are as follows:

[0090] 1. B / S architecture

[0091] This unit belongs to the browser client architecture and runs within the browser.

[0092] 2. Knowledge Graph Generation Strategies

[0093] This unit can generate a logical topology knowledge graph based on the business description chain, and provides a modifiable interface on each generated topology graph node.

[0094] 3. Knowledge Graph Manipulation Toolset

[0095] The toolset allows modification of knowledge graph nodes, providing functions for adding, deleting, and changing nodes.

[0096] 4. Strategies for Generating Standard Knowledge

[0097] The business system generated by this unit based on the knowledge graph can be output as text data that conforms to the business process description standard.

[0098] 5. Log Recording

[0099] All production system processes in this unit will be logged to facilitate the maintenance of this unit.

[0100] Furthermore, the specific operation process for component access in the water conservancy big data model service decision support system unit is as follows:

[0101] Input the business component relationship (business expression chain integrated by the large model dedicated water conservancy component library unit) generated in the previous step into this unit. The system of this unit will parse the relationship and then generate a business process topology framework diagram. The business process topology framework diagram is then sent to the front-end system interface as a sketch.

[0102] Furthermore, water conservancy industry experts participated in revising the business process topology framework diagram, including revisions to the overall topology structure, node attributes, and node data.

[0103] Next, water conservancy industry experts determine the initial version of the business process topology framework diagram, run it to generate the business application system, and then test the system. This unit supports system rollback and correction until it meets business requirements, at which point the correction is terminated, and the business application system is obtained.

[0104] Furthermore, when requirements change, water conservancy industry experts can quickly roll back the system, modify the business logic in the business process topology knowledge graph, and regenerate the business application system.

[0105] Furthermore, the water conservancy big model service decision support system unit can translate the generated business application system into text knowledge that conforms to the business process description standard, and then transfer it to the knowledge accumulation unit.

[0106] Knowledge Accumulation Unit:

[0107] The knowledge accumulation unit extracts system knowledge generated by the water conservancy big model service decision support system unit, and then organizes and incorporates the knowledge into the dataset.

[0108] Furthermore, the specific operational process of knowledge accumulation within the knowledge accumulation unit is as follows:

[0109] First, knowledge is extracted from the business application system generated in the previous step, and then cleaned through text knowledge data mining before entering the business knowledge pool. The business knowledge pool allows for knowledge management, such as knowledge sharing, knowledge reuse, and knowledge privatization. Furthermore, the knowledge in the knowledge pool can be configured as a knowledge base specific to a particular region to provide add-on services to LLM. On the other hand, it can also serve as a training dataset for LLM to guide the emergence of industry-specific business optimization methods.

[0110] The modules in the aforementioned large-scale intelligent agent application system for the water conservancy industry can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0111] The advantages of the large-scale intelligent agent application system for the water conservancy industry proposed in this invention are as follows:

[0112] Firstly, it can deeply integrate water resources knowledge with LLM, enabling LLM to leverage its reasoning and emergent abilities when facing water resources professional business. Combined with water resources professional knowledge, it can generate business practices across the entire industry. Theoretically, with sufficient training data, LLM can even provide business understanding practices that surpass those of industry experts, thanks to its extensive learning of a large amount of basic knowledge that is difficult for humans to achieve.

[0113] Next, the business logic is described in a standardized way through business knowledge base units, which is beneficial to the long-term development of the industry. A unified description method will enable the large model to better understand the intentions of personnel needs.

[0114] By using a dedicated hydraulic component library for large models to modularize all functional modules, all systems can be decoupled, making the resulting business application system more flexible and able to respond to changes in requirements in a timely manner. On the other hand, modularized modules also eliminate the need for repetitive development work, and each component can be reused.

[0115] The decision support system unit for water conservancy big model services can flexibly assemble all components. At this stage, industry professionals participate in the system generation, making the generated system more in line with the needs of actual applications. Because it is based on a component-based assembly system, the system development cycle is short and the timeliness is strong. All generated business application systems can quickly respond to changes in demand, making the application system more flexible and easy to use.

[0116] By leveraging the knowledge accumulation unit, the industry expertise implicit in the generated business application system can be fed back to the LLM for learning. Its dataset can be used to train the LLM, forming a scientific closed-loop application model in which ideas guide practice, and practice in turn provides feedback to guide ideas.

[0117] Those skilled in the art will understand that embodiments of the present invention can provide methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied 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.

[0118] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0119] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0120] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0121] It should be noted that the specific embodiments described above enable those skilled in the art to more fully understand the present invention, but do not limit the present invention in any way. Therefore, although the present invention has been described in detail in this specification and embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the present invention; and all technical solutions and improvements that do not depart from the spirit and scope of the present invention are covered within the protection scope of the present invention patent. No reference numerals in the claims should be construed as limiting the scope of the claims. Any simple variations or equivalent substitutions of technical solutions that can be readily obtained by those skilled in the art within the scope of the technology disclosed in the present invention are within the protection scope of the present invention.

Claims

1. A large-scale intelligent agent application system for the water conservancy industry, characterized in that, include: The Large Language Model Base Unit provides a standard interface for interacting with open-source or non-open-source Large Language Models (LLMs) and responding to user requests. The business knowledge base unit includes a training dataset for training the LLM model and a knowledge base dataset for LLM add-ons. The training dataset is used to accumulate knowledge to train the LLM, and the knowledge base dataset is used to support customized needs, expressing customized business as knowledge for intelligent agents to call. The large-scale model-specific hydraulic component library unit is used to provide a management platform for all large-scale model-specific components, provide component service functions, and respond to system requirements. The water conservancy big model service decision support system unit is used to assemble business components into an application system and provide operation tools to adjust the assembly process to form an applicable system. According to the dynamic changes in demand, the system can be modified by adjusting the components to adapt to the changes in demand. The knowledge accumulation unit is used to extract system knowledge generated by the water conservancy big model service decision support system unit, and then organize the knowledge and put it into the dataset. The large language model base unit responds to user requests, specifically including: When the large language model base unit receives a user request, it performs intent recognition of the request. First, the request is vectorized to obtain a request vector, and then the request vector is classified to obtain a classification result. The classification results are returned to the LLM to determine whether the business knowledge base unit needs to participate. If so, the business knowledge base unit is called, and the results are reconstructed based on the output of the business knowledge base unit. The LLM then processes and outputs the results. If the business knowledge base unit does not need to participate, the LLM processes the results and outputs them directly. The specific operations for responding to user requests include: The business knowledge base is divided into topic domains. User needs are matched through intent recognition, and then LLM provides recommended topic domains. Under each topic domain, all business descriptions are vectorized to obtain business description vectors, and then a matching process is performed. After matching the corresponding business, the business process description standard and the corresponding business input are used as prompt words in LLM. Then, LLM generates business process content to form a business expression chain. The water conservancy large model service decision support system unit includes: A knowledge graph generation strategy is used to generate a logical topology knowledge graph based on the business expression chain, and to provide an interface that can be modified on each generated topology graph node; Generate standard knowledge strategies so that business systems generated based on knowledge graphs can be output as text data that conforms to business process description standards. The generated business description chain is input into this unit, which parses the relationships and generates a business process topology framework diagram. This business process topology framework diagram is then sent as a draft to the front-end system interface to determine the initial version of the business process topology framework diagram. This diagram is then run to generate the business application system.

2. The water conservancy industry large-scale intelligent agent application system according to claim 1, characterized in that, The standard interface is one that takes a string as input and returns a string.

3. The water conservancy industry large-scale intelligent agent application system according to claim 1, characterized in that, The large-scale model-specific hydraulic component library unit is specifically used for: Provide a unified access address for component registration; The service strategy for embedded components, and the component service governance strategy; Access to the component library unit is granted through authentication and authorization; All connected components are managed in a unified group, and blacklist / whitelist groups are created. After the component is invoked, a log of the request access is generated; It has a built-in authentication system for authorization, auditing, and access control when accessing the component library through other means.

4. The water conservancy industry large-scale model intelligent agent application system according to claim 1, characterized in that, The large-scale model dedicated water conservancy component library unit performs unified group management of the connected components. The management content includes: the purpose and scope of application of the components, the input parameters of the components, the output parameters of the components, and the protocol authentication of the components.

5. The water conservancy industry large-scale intelligent agent application system according to claim 3, characterized in that, The large-scale model dedicated water conservancy component library unit responds to the component retrieval requests of the LLM, decomposes the generated business description chain, feeds it back to the LLM, gradually finds the corresponding components to match, feeds back the upstream and downstream relationships of the components to the LLM, and generates the corresponding component retrieval relationship; integrates the business description chain to generate business component relationship, and based on the business component relationship, passes it to the water conservancy large-scale model service decision support system unit to generate the business application system.

6. The water conservancy industry large-scale intelligent agent application system according to claim 1, characterized in that, The water conservancy large model service decision support system unit also includes: The BS architecture is a browser client architecture that runs within a browser. A knowledge graph manipulation toolset for modifying nodes in a knowledge graph, including functions for adding, deleting, and modifying nodes; Log recording is used to generate system process logs.

7. The water conservancy industry large-scale model intelligent agent application system according to claim 6, characterized in that, The operation of the water conservancy large model service decision support system unit also includes: The system is tested, and this unit supports system rollback corrections until the system meets business requirements, at which point the correction process is terminated, and the business application system is obtained.

8. The water conservancy industry large-scale model intelligent agent application system according to claim 7, characterized in that, The knowledge accumulation unit performs the following operations for knowledge accumulation: extracting knowledge from the generated business application system, and then cleaning and mining the text knowledge before entering the business knowledge pool, where the knowledge is managed. The knowledge in the knowledge pool can be configured into a knowledge base specific to a particular region to provide add-on services to LLM, and can be used as a training dataset for LLM to guide the emergence of industry-specific business optimization methods.

Images