A method and device for generating results of dredging a large model, an electronic device, and a medium

Abstract

Embodiments of the present application disclose a dredging large model result generation method and device, electronic equipment and medium, which are applied to a total control scheduling intelligent agent of a dredging large model. The method comprises the following steps: receiving user context information output by a base model of the dredging large model for a current user problem, and calling a domain recognition intelligent agent to output a system prompt word corresponding to a current domain according to the user context information, so that the domain recognition intelligent agent outputs a domain judgment result according to the system prompt word; when the domain judgment result is the dredging domain, calling a question classification intelligent agent to determine a question classification result corresponding to the current user problem according to an analysis and reasoning requirement corresponding to the current domain and the user context information; calling a tool matching intelligent agent to determine a tool calling strategy according to the question classification result, the user context information and a dredging domain tool library, calling a target tool according to the tool calling strategy, and obtaining and displaying a target generation result corresponding to the current user problem.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of dredging technology, and in particular to a method, apparatus, electronic device and medium for generating results of a large-scale dredging model. Background Technology

[0002] With the development of intelligent technologies, the demand for efficient decision support, accurate computational analysis, and knowledge reuse in the dredging engineering field is becoming increasingly urgent. Currently, using the natural language processing and reasoning capabilities of general large models to process dredging engineering knowledge involves many professional formulas and historical data. Therefore, it can only rely on simple analysis of the user's context to determine the user's purpose and call upon known general knowledge bases to generate results that can solve the user's problem.

[0003] However, for professionals in the dredging field, the problems they use to deal with by large models generally contain industry-specific terminology and are prone to involve a large number of mechanism calculations. The problem results generated by general large models relying solely on general knowledge bases are difficult to meet the accuracy requirements of engineering-level answers. Summary of the Invention

[0004] This invention provides a method, apparatus, electronic device, and medium for generating results of a large-scale dredging model. It can accurately select appropriate tools according to the problem type corresponding to different problems in the dredging field to accurately solve the current user problem and obtain the target generated result, which can achieve and meet the accuracy requirements of engineering-level answers.

[0005] In a first aspect, embodiments of the present invention provide a method for generating results for a large-scale dredging model, applied to a central control and scheduling agent for a large-scale dredging model, the method comprising: The system receives user context information output by the base model of the dredging big model for the current user problem, and calls the domain recognition agent to output the system prompt words corresponding to the current domain based on the user context information, so that the domain recognition agent can output the domain judgment result based on the system prompt words.

[0006] When the domain determination result is dredging, the problem classification agent is invoked to determine the problem classification result corresponding to the current user's problem based on the analysis and reasoning requirements of the current domain and the user's context information.

[0007] The tool matching agent determines the tool invocation strategy based on the problem classification results, user context information, and dredging domain tool library. It then invokes the target tool according to the tool invocation strategy to obtain and display the target generation result corresponding to the current user problem.

[0008] The dredging large-scale model result generation method provided in this invention first utilizes the base model to output user context information, enabling rapid analysis of the user context information corresponding to the current user's question using the base model's natural language processing capabilities. A neighborhood recognition agent dynamically generates system prompts based on the technical field to which the current user's question belongs, providing dynamic guidance, constraint, and enhancement for the domain recognition agent when determining the domain judgment result. When the domain judgment result is determined to be dredging, a problem classification agent is invoked to analyze the user context information based on the analysis and reasoning requirements corresponding to the current domain, obtaining the problem classification result corresponding to the current user's question. This allows for rapid differentiation of the problem type of the current user's question using the problem classification agent. By utilizing a tool-matching agent, the system determines the tool invocation strategy from the dredging tool library based on the current user's question classification results and user context information. Following this strategy, the system invokes the target tool to generate the desired result. This addresses the current limitation of using only general-purpose large models to handle dredging issues, which cannot meet the accuracy requirements for engineering-level responses. The system enables precise and rapid selection of appropriate tools based on the question type corresponding to different dredging problems to accurately solve the current user's question and obtain the target result, thus achieving and meeting the accuracy requirements for engineering-level responses.

[0009] Secondly, embodiments of the present invention also provide a result generation device for a large-scale dredging model, which is applied to the overall control and scheduling agent of the large-scale dredging model. The device includes: The receiving module is used to receive the user context information output by the base model of the dredging large model in response to the current user problem, and call the domain recognition agent to output the system prompt words corresponding to the current domain based on the user context information, so that the domain recognition agent can output the domain judgment result based on the system prompt words.

[0010] The classification module is used to, when the domain judgment result is dredging, call the problem classification agent to determine the problem classification result corresponding to the current user's problem based on the analysis and reasoning requirements of the current domain and the user's context information.

[0011] The output module is used to call the tool matching agent to determine the tool calling strategy based on the problem classification results, user context information and dredging domain tool library, call the target tool according to the tool calling strategy, and obtain and display the target generation result corresponding to the current user problem.

[0012] Thirdly, embodiments of the present invention also provide an electronic device, the electronic device comprising: At least one processor; and A memory that is communicatively connected to at least one processor; wherein, The memory stores a computer program that can be executed by at least one processor, such that the at least one processor is able to perform the method for generating results of a large dredging model according to any embodiment of the present invention.

[0013] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing computer instructions that, when executed by a processor, implement the method for generating results of a large-scale dredging model according to any embodiment of the present invention.

[0014] Fifthly, embodiments of the present invention also provide a computer program product, including a computer program that, when executed by a processor, implements the method for generating results of a large-scale dredging model according to any embodiment of the present invention.

[0015] It should be noted that the aforementioned computer instructions may be stored, in whole or in part, on a computer-readable storage medium. This computer-readable storage medium may be packaged together with the processor of the dredging model result generation device, or it may be packaged separately from the processor of the dredging model result generation device; this application does not impose any limitations on this.

[0016] The descriptions of the second, third, fourth, and fifth aspects in this application can be referred to the detailed description of the first aspect; and the beneficial effects of the descriptions of the second, third, fourth, and fifth aspects can be referred to the analysis of the beneficial effects of the first aspect, which will not be repeated here.

[0017] In this application, the name of the device for generating the results of the large-scale dredging model described above does not limit the equipment or functional module itself. In actual implementation, these devices or functional modules may appear under other names. As long as the function of each device or functional module is similar to that of this application, it falls within the scope of the claims of this application and its equivalents.

[0018] These or other aspects of this application will become more readily apparent in the following description. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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.

[0020] Figure 1 A flowchart illustrating a method for generating results of a large-scale dredging model according to an embodiment of the present invention; Figure 2A flowchart illustrating another method for generating results of a large-scale dredging model provided in an embodiment of the present invention; Figure 3 A schematic diagram of the structure of a dredging large model result generation device provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0021] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present invention, and not all of the structures.

[0022] In this article, the term "and / or" is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone.

[0023] The terms “initial” and “target” in the specification and drawings of this application are used to distinguish different objects or to distinguish different treatments of the same object, rather than to describe a specific order of objects.

[0024] Furthermore, the terms "comprising" and "having," and any variations thereof, used in the description of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the steps or units listed, but may optionally include other steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus.

[0025] Before discussing the exemplary embodiments in more detail, it should be noted that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe the operations (or steps) as sequential processes, many of these operations can be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the operations can be rearranged. The process can be terminated when its operation is completed, but may also have additional steps not included in the figures. The process can correspond to a method, function, procedure, subroutine, subroutine, etc. Moreover, without conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0026] It should be noted that in the embodiments of this application, the words "exemplary" or "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design scheme described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of the words "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0027] In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0028] Figure 1 This is a flowchart illustrating a method for generating results from a large-scale dredging model according to an embodiment of the present invention. This embodiment is applicable to generating results after analyzing user problems using a proprietary large-scale dredging model. The method can be executed by a large-scale dredging model result generation device, which can be implemented in hardware and / or software and can be configured within the overall control and scheduling intelligent agent of the large-scale dredging model. In this embodiment, the overall control and scheduling intelligent agent of the large-scale dredging model can run as a software entity or functional module on an electronic device. (Continue to refer to...) Figure 1 This embodiment specifically includes: S101. Receive the user context information output by the base model of the dredging large model for the current user problem, and call the domain recognition agent to output the domain judgment result based on the system prompt words corresponding to the current domain according to the user context information.

[0029] The dredging master model is specifically designed for the dredging field and is obtained through optimization training based on the general master model. The base model serves as the foundation for the dredging master model. The current user question is the question input by the user in the current query round. The user context information is the information obtained by the base model after parsing the context of the current user question. The central control scheduling agent is responsible for task distribution, state management, and exception handling throughout the entire process, receiving user input (the current user question) and calling the corresponding agents according to the process sequence. The domain identification agent analyzes whether the user's current question belongs to the dredging field and generates a system prompt word corresponding to the current user question after determining that it belongs to the dredging field. The domain judgment result is used to determine the professional domain to which the current user question belongs.

[0030] Specifically, after the user inputs their current question through the large model interface, the central control and scheduling agent performs tasks such as intent recognition, entity extraction, and contextual understanding and parsing on the question. Based on prompt word engineering, retrieval enhancement, and long text truncation, it constructs a structured prompt word for the large model's base model. The base model transforms the structured prompt word into user context information containing the user question, context, and a temporary user profile corresponding to the current question. Then, the central control and scheduling agent receives the user context information output from the base model and invokes the domain recognition agent to sequentially perform feature extraction, determine domain routing labels and rule prompt word packages, generate dynamic domain system prompt words based on an embedded multi-domain prompt word routing framework, and finally output the domain judgment result using the generated system prompt words.

[0031] In this embodiment, the user context information is first output using the base model, enabling rapid analysis of the user context information corresponding to the current user's question using the base model's natural language processing capabilities. Then, the neighborhood recognition agent outputs dynamic system prompts corresponding to the current domain based on the user context information. This allows for the dynamic generation of system prompts based on the technical field of the current user's question, providing dynamic guidance, constraint, and enhancement for the domain recognition agent when determining the domain's judgment result. It also provides a foundation for subsequently identifying the current domain as dredging and calling upon an agent more suitable for that domain.

[0032] S102. When the domain judgment result is dredging domain, the problem classification agent is invoked to determine the problem classification result corresponding to the current user problem based on the analysis and reasoning requirements of the current domain and the user context information.

[0033] The dredging field primarily refers to engineering operations involving the excavation, cleaning, transportation, and disposal of sediments such as silt, rocks, and waste underwater (riverbeds, lakebeds, ports, waterways, etc.) using manpower, machinery, or hydraulic methods. The problem classification agent is used to classify the current user's problem within its current domain, determining whether it belongs to a knowledge-based reasoning problem or a problem requiring formula calculation and analysis. Analysis and reasoning requirements characterize the analytical or reasoning needs required for different problem types within the current domain. The problem classification result characterizes the problem type corresponding to the current user's problem.

[0034] Specifically, in this embodiment, the domain determination result only needs to be determined as either a dredging domain or not a dredging domain. Therefore, when the domain determination result is a dredging domain, the central control scheduling agent can directly call the problem classification agent to analyze which problem type's analysis and reasoning requirements are met by the user's context information, using the analysis and reasoning requirements corresponding to the current domain as the standard, and thus determine the problem classification result corresponding to the current user's problem.

[0035] Optionally, when the domain determination result is a non-dredging domain, since this embodiment is mainly based on the dredging domain, for the current user problem in the non-dredging domain, the base model can be directly called to output the generation result for the current user problem.

[0036] In this embodiment, when the domain judgment result is determined to be the dredging domain, the problem classification agent is invoked to judge and analyze the user's context information based on the analysis and reasoning requirements corresponding to the current domain, so as to obtain the problem classification result corresponding to the current user's problem. This enables the problem classification agent to quickly distinguish the problem type of the current user's problem, providing a basis for dynamically selecting and invoking appropriate tools to solve the problem according to different types of current user problems.

[0037] S103. The tool matching agent determines the tool invocation strategy based on the problem classification results, user context information, and dredging domain tool library. It then invokes the target tool according to the tool invocation strategy to obtain and display the target generation result corresponding to the current user problem.

[0038] The tool matching agent maintains a tool index (tool name, applicable scenarios, input parameter requirements, functional description, etc.), performs matching searches between problem intent and tools, and determines whether a matching tool exists. The dredging domain tool library is a user-pre-configured knowledge tool library in the dredging domain, which may include a retrieval-enhanced generation plugin knowledge base, a knowledge graph-enhanced generation plugin knowledge base, and a dredging domain context protocol toolchain. The tool invocation strategy indicates the tools that need to be used or invoked for the current user problem. The target tool is the tool that needs to be invoked for the current user problem. The target generated result is the response that needs to be returned to the user for the current user problem.

[0039] Specifically, after obtaining the problem classification results output by the problem classification agent, the tool matching agent can be invoked to select appropriate tools from the dredging tool library based on the problem classification results and user context information, while simultaneously determining the tool invocation strategy. Then, the central control scheduling agent can invoke the target tool according to the tool invocation strategy and process the current user's problem based on the user context information, ultimately obtaining the processed result, i.e., the target generation result, which is then displayed to the user.

[0040] In this embodiment, a tool matching agent determines the tool invocation strategy from the dredging domain tool library based on the current user's question classification result and user context information. The target tool is then invoked according to the tool invocation strategy to generate the target result. This solves the problem that current methods using only general large models to process dredging issues are insufficient to meet the accuracy requirements of engineering-level answers. By using a domain identification agent to determine the domain of the current user's question, a question classification agent to determine the question type, and a tool matching agent to select appropriate tools from the dredging domain tool library for different question types, this embodiment achieves precise and rapid selection of appropriate tools based on the question type corresponding to different problems in the dredging domain to accurately solve the current user's problem and obtain the target result, thus meeting the accuracy requirements of engineering-level answers.

[0041] The dredging large-scale model result generation method provided in this invention first utilizes the base model to output user context information, enabling rapid analysis of the user context information corresponding to the current user's question using the base model's natural language processing capabilities. A neighborhood recognition agent dynamically generates system prompts based on the technical field to which the current user's question belongs, providing dynamic guidance, constraint, and enhancement for the domain recognition agent when determining the domain judgment result. When the domain judgment result is determined to be dredging, a problem classification agent is invoked to analyze the user context information based on the analysis and reasoning requirements corresponding to the current domain, obtaining the problem classification result corresponding to the current user's question. This allows for rapid differentiation of the problem type of the current user's question using the problem classification agent. By utilizing a tool-matching agent, the system determines the tool invocation strategy from the dredging tool library based on the current user's question classification results and user context information. Following this strategy, the system invokes the target tool to generate the desired result. This addresses the current limitation of using only general-purpose large models to handle dredging issues, which cannot meet the accuracy requirements for engineering-level responses. The system enables precise and rapid selection of appropriate tools based on the question type corresponding to different dredging problems to accurately solve the current user's question and obtain the target result, thus achieving and meeting the accuracy requirements for engineering-level responses.

[0042] Figure 2 This is a flowchart illustrating another method for generating results for a large-scale dredging model, provided by an embodiment of the present invention. This embodiment is based on the above embodiment, specifying the execution steps for calling each intelligent agent and other optional steps. In this embodiment, the method may specifically include: S201. Receive user context information output by the base model of the dredging large model for the current user problem.

[0043] Specifically, the central control and scheduling agent receives the current user's question through the front-end interface - access layer (API gateway) - service layer (agent service), and passes the received question to the base model for contextual semantic parsing of the current user's question. After obtaining the user's context information, the base model feeds it back to the central control and scheduling agent.

[0044] For example, the base model of the dredging large-scale model in this embodiment can be based on an existing general-purpose large-scale model. A model base adapted to the dredging field can be selected, requiring data interaction interfaces, tool interfaces, and interactive functional interfaces to support the input and output of external data and enable the large-scale model to access and call various tool programs. For example, a general-purpose large-scale model base with dialogue interaction and data interface interaction can be selected. Dialogue-based interactive data is also connected to the data interface. A Hypertext Transfer Protocol-Transmission Control Protocol interface is established as the input / output interface. The selected base model needs to have a standardized tool call interface, supporting: function call protocol, multimodal input / output, and streaming interaction interface. For the base model, parameters strongly related to the dredging profession need to be adjusted. Only adaptation parameters should be adjusted, without modifying the model's original weights / structure. All parameters should be optimized around "dredging data interaction and terminology understanding." Optionally, interface interaction parameters can be adjusted. Since dredging calculations require high-frequency transmission of numerical / graphical data, too few connections will lead to delayed feedback of calculation results and excessive server load. Therefore, a persistent connection number of the parallel computing transmission control protocol adapted to single / multi-vessel construction parameters is selected. Determine an appropriate streaming frame rate to adapt to the incremental output of long dredging texts such as construction plans and calculation reports, as well as visual charts such as mud pump QH curves and pipe resistance velocity curves, ensuring smoothness. Optionally, adjust the word / embedding parameters to determine the embedding vector dimension of newly added dredging words, ensuring consistency between the vector space of newly added dredging terms and the native vocabulary, generally 1024 dimensions, to avoid misunderstanding of terms; when training alone, set the learning rate to 1 / 5 of the base value to prevent damage to the native semantic capabilities. Optionally, adjust the multimodal parameters to adapt to high-definition parsing of dredging engineering charts, ensuring the accuracy of visualization of calculation results, generally with a chart parsing resolution (SVG format) of no less than 1920×1080.

[0045] S202. Call the domain recognition agent so that the domain recognition agent can extract features based on the user's purpose, context information and user question profile in the user context information to obtain the domain feature vector. Based on the domain feature vector, extract the domain system prompt words from the domain prompt word library, and output the domain judgment result based on the domain system prompt words and user context information.

[0046] Among these, "User Purpose" represents the main purpose of the user's current question. "Contextual Information" represents the relevance of the information provided in the current question, its connection to historical dialogues, and its relevance to the user's purpose. "User Question Profile" is a predicted user profile based on the current question, aiming to predict the user's role. "Domain Feature Vector" is a feature vector reflecting the domain, obtained after feature extraction and standardization by the input layer of the neighborhood recognition agent. "Domain Prompt Word Library" is a multi-domain routing repository within the multi-domain routing layer of the domain recognition agent, responsible for routing to the corresponding domain processor based on the domain feature vector, thereby determining the domain judgment result. "Domain System Prompt Words" are dynamic system prompt words generated by the prompt word generation layer of the domain recognition agent.

[0047] Specifically, the domain recognition agent in this embodiment is divided into an input layer, a multi-domain routing layer, a prompt word generation layer, and an output layer. After receiving the user's context information, the domain recognition agent parses the user's context information at the input layer to obtain the user's purpose, context information, and user question profile, and extracts and standardizes their features to obtain a domain feature vector. Then, the domain recognition agent passes the domain feature vector from the input layer to the multi-domain routing layer. The domain routing engine in the multi-domain routing layer outputs domain routing labels and recognition rule prompt word packages based on the domain feature vector, locating the prompt word generation layer. The prompt word generation layer receives the domain routing labels, recognition rule prompt word packages, and domain feature vector, matches the corresponding domain from the domain prompt word library, and dynamically generates domain recognition system prompt words specific to the current scene. Finally, the output layer performs domain recognition reasoning based on the domain system prompt words, and the final reasoning result serves as the domain judgment result.

[0048] In this embodiment, by initially calling the domain recognition agent to identify the domain to which the current user's question belongs, and obtaining the domain judgment result, it is possible to quickly and accurately assign the user's question to the processor of the corresponding domain and generate dynamic domain system prompt words. This not only provides accurate guidance for subsequent use of the dredging big model to process dredging domain questions, avoiding the problem that using the dredging big model to process non-dredging domain questions may lead to inaccurate output results, but also provides a foundation for providing accurate system prompt words for subsequent use of the dredging big model to process the current user's question.

[0049] S203. When the domain judgment result is dredging domain, the problem classification agent is invoked so that the problem classification agent can determine the problem classification feature vector based on the user context information, determine the classification system prompt words based on the problem classification feature vector and the domain routing label in the domain judgment result, and determine the problem classification result based on the classification system prompt words and the problem classification criteria.

[0050] The question classification feature vector is the feature vector obtained by the question classification agent after parsing the user's context information. The classification system prompts are dynamic system prompts generated by the prompt generation layer of the question classification agent to determine the question type.

[0051] Specifically, if the domain determination result is determined to be the dredging domain, the problem classification agent can be directly invoked to analyze the user's purpose, context information, user problem profile, and optionally added domain routing tags in the user's context information. In this embodiment, after obtaining the user's context information, the problem classification agent analyzes and parses it, for example, by completing the context information to restore the user's complete core needs and obtain a problem classification feature vector; then, based on the domain routing tags, it filters the dredging domain classification standards and positive and negative example libraries from the full scenario library of the dredging domain, and compares the obtained dredging domain classification standards and positive and negative example libraries with the problem classification feature vector to determine the classification system prompt words; then, based on the classification system prompt words and the problem classification standards, it analyzes the problem classification feature vector, such as determining the core of the problem and the processing methods that may need to be used, and finally determines the problem classification result.

[0052] For example, in this embodiment, the problem classification criteria can be as follows: If the core of the problem involves information retrieval, concept explanation, standard interpretation, case analysis, qualitative scheme evaluation, and knowledge-based reasoning; and it can be answered solely through dredging knowledge without requiring any numerical calculations, engineering formulas, or statistical analysis functions / tools, then it falls under the category of knowledge reasoning. If the core of the problem involves quantitative calculations, numerical analysis, parameter verification, data statistics, and formula solving; and it requires the use of dredging-specific calculation formulas, numerical models, statistical functions, and calculation tools to output accurate results, and the problem contains explicit numerical parameters and operating conditions, requiring the output of quantitative results, verification conclusions, and fitted data, then it falls under the category of calculation and analysis.

[0053] In this embodiment, the problem classification agent is invoked to classify the current user's problem. This enables the initial selection of tool types that are more closely related to the problem type, and then the selection of appropriate tools from the tool types corresponding to the problem type is carried out later. This achieves the rapid selection of appropriate tool categories based on problem type, reducing the complexity and time of selecting suitable tools one by one from a large number of tools.

[0054] Optionally, in this embodiment, after the domain recognition agent outputs the system prompt word corresponding to the current domain based on the user context information, the method further includes: When the domain determination result is not in the dredging domain, the domain determination result and user context information are input into the base model so that the base model can determine the natural language understanding result based on the domain determination result and user context information, and display the natural language understanding result as the target generation result.

[0055] Specifically, when the domain judgment result is determined to be a non-dredging domain, the domain judgment result and user context information can be directly passed to the base model. The base model can then generate a natural language understanding result for the current user question by using its own natural language understanding ability or by directly calling the knowledge base of the corresponding domain based on the domain routing identifier in the domain judgment result, based on the domain judgment result and the user context information. The natural language understanding result is then displayed as the target generated result.

[0056] In this embodiment, if the domain determination result is not a dredging domain, the target generation result corresponding to the current user problem can be directly given based on the base model, which enables a rapid response even when facing problems in non-dredging domains.

[0057] S204. If the problem classification result is a computational analysis class, the tool matching agent determines the tool invocation strategy containing the target invocation function based on the model context protocol toolchain, user context information, and dredging domain tool library corresponding to the computational analysis class.

[0058] The computational analysis category requires calling calculation formulas based on the known conditions given in the current user question to perform mathematical calculations and generate a response to the user question. In this embodiment, the Model Context Protocol (MCP) toolchain is a dredging-specific MCP toolchain, employing a three-layer architecture. It serves the intelligent agent in calling external function programs, enabling functions such as database retrieval, empirical data statistics, and mechanism calculation. The target calling function can be a pre-packaged dredging scenario-based intelligent agent from a dredging tool library; in this embodiment, the dredging scenario-based intelligent agents include: cutter suction dredging construction intelligent agents, trailing suction dredging construction intelligent agents, pipe resistance calculation intelligent agents, and ecological dredging intelligent agents, etc.

[0059] Specifically, if the problem classification result is a computational analysis type, the tool matching agent can be directly invoked to select the dredging scenario-specific agent that best matches the known conditions given by the user in the user context information based on the MCP toolchain corresponding to the computational analysis type, the known conditions given by the user in the user context information (such as numerical parameters, working conditions, required output quantitative results, etc.), and the dredging domain tool library. Then, the MCP toolchain is used to form a call chain, and a tool call strategy with a determined target call function is formed.

[0060] S205. If the problem classification result is knowledge reasoning, the tool matching agent is invoked to determine the information organization structure classification corresponding to the current user problem based on the user context information and the dredging domain tool library, and to determine the tool invocation strategy that includes the knowledge enhancement tools corresponding to the information organization structure classification.

[0061] Among them, the information organization structure classification is used to clarify the information organization architecture of the current user's question. The knowledge enhancement tool is used to inject real-time, private, and professional knowledge into the model through retrieval, searching, and querying, thereby improving the accuracy and credibility of the answer.

[0062] Specifically, if the problem classification result is knowledge reasoning, the tool matching agent can be invoked to first determine the information organization structure classification corresponding to the current user's problem based on the user's context information. After determining the information organization structure classification, a more suitable knowledge enhancement tool can be selected from the dredging domain tool library to determine the tool invocation strategy.

[0063] For example, the tool matching agent determines the information organization structure corresponding to the current user problem based on user context information and the dredging domain tool library. This classification includes fragmented and structured information organization structures. If the information organization structure is classified as fragmented, the corresponding knowledge enhancement tool is a retrieval enhancement generation tool; if the information organization structure is classified as structured, the corresponding knowledge enhancement tool is a knowledge graph generated through knowledge enhancement.

[0064] Among these, the retrieval enhancement generation tool injects relevant information into prompt words by retrieving external knowledge bases, enabling large models to generate accurate and traceable answers based on factual knowledge. The knowledge graph generated through knowledge enhancement stores knowledge in an entity-relationship-triple format, supporting multi-hop queries and logical reasoning, providing large models with reasonable and interpretable enhanced knowledge. Fragmented information organization can be understood as the current user question consisting of simple language or fragmented known conditions. Structured information organization can be understood as the current user question presented in a structured form, such as a diagram or a clear organizational structure.

[0065] In this embodiment, by determining whether the question type belongs to the calculation and analysis category or the knowledge reasoning category, the question type of the current user's question is clarified, and the type of tool to be used to solve the question is then determined. By clarifying the question type of the current user's question, the difference between its processing logic and the tool used is reflected, which provides a more accurate and faster basis for selecting the appropriate tool later.

[0066] Specifically, if the information organization structure is determined to be a fragmented information organization structure, then the knowledge enhancement tool can be identified as a retrieval enhancement generation tool. If the information organization structure is determined to be a structured information organization structure, then the knowledge enhancement tool can be identified as a knowledge graph generated through knowledge enhancement.

[0067] Optionally, the process for establishing the retrieval enhancement generation tool in this embodiment can be as follows: Dredging knowledge fragmentation: Terminology, cases, and standards in the dredging professional knowledge base are fragmented into single knowledge points / single questions, with the fragment length controlled between 50-200 characters to suit retrieval precision; Vectorization: A Transformer-based bidirectional encoder representation embedding model is used to convert the fragmented knowledge into vectors. The embedding model is fine-tuned in advance using a dredging terminology dataset to improve the accuracy of terminology vectorization; Construction of a dedicated dredging vector database: Milvus / Pinecone is used to categorize databases by knowledge type (terminology database, case database, standard database, equipment database) to suit the classification system of dredging knowledge; Retrieval link construction: User question → Vector conversion → Cosine similarity retrieval → Extraction of the first N knowledge fragments → Merging with the question and inputting into the large dredging model → Generating an answer.

[0068] Furthermore, the knowledge base of the retrieval enhancement generation tool needs parameter adjustments adapted to the dredging field. This primarily targets vector dimension, the number of the first N searches, the cosine similarity retrieval threshold, and the knowledge fragment segment length. Optionally, the vector dimension is typically 768, adapting to the low-dimensional characteristics of fragmented dredging knowledge and balancing retrieval speed and accuracy. The number of the first N searches is set to 3 to 5; dredging knowledge has strong uniqueness, and too many searches will introduce irrelevant information, ensuring both accuracy and richness of the answers. The cosine similarity retrieval threshold is ≥0.8; values ​​below 0.8 are considered irrelevant, preventing incorrect answers due to irrelevant content. The knowledge fragment segment length is 50 to 200 characters; dredging professional knowledge points are mostly short texts, such as equipment parameters and process precautions. Excessively long segments will include irrelevant information, while segments that are too short will result in incomplete knowledge.

[0069] Optionally, the knowledge graph creation process in this embodiment can be as follows: Dataset detection: Quality detection and scoring are performed on the acquired high-quality dredging dataset. Knowledge extraction: "Entity-relationship-attribute" triples are extracted from the high-quality dredging dataset using a hybrid strategy to construct the basic units of the knowledge graph. Entity extraction uses an embedding model to identify dredging-specific entities, converting them into feature vectors and comparing their matching degree with knowledge fragments to convert them into a probability distribution for entity type ranking. Relationship extraction: Implicit associations between entities are mined based on the embedding model. Attribute extraction: Entity attributes are extracted using the embedding model. Knowledge fusion: Entity alignment and conflict resolution address entity duplication and attribute conflicts in multi-source data, ensuring the consistency of the knowledge graph. Knowledge storage and querying: A graph database is used for storage. Based on the semantic understanding capabilities of the large model foundation, multiple query mechanisms are set up to achieve rapid utilization of the knowledge graph.

[0070] Furthermore, the knowledge graph generated by knowledge augmentation needs parameter adjustments adapted to the dredging field. This mainly targets dataset detection, knowledge extraction, knowledge fusion, and knowledge query. Optionally, adjust the dataset detection parameters: set scoring weights (completeness α1 / accuracy α2 / consistency α3). Generally, in dredging projects, knowledge accuracy (such as formulas and parameters) is the core and should be given the highest weight; consistency (such as terminology) is second, and completeness is last. Set a qualified dataset threshold, generally requiring ≥0.8, to ensure that the dredging data entering the graph has no parameter errors or logical conflicts and matches engineering-level accuracy requirements. Adjust the knowledge extraction parameters: set an entity extraction confidence threshold, generally requiring ≥0.95. Dredging entities (such as dredgers and other dredging equipment or dredging-specific entities) are highly specialized, and the extraction accuracy must be ensured to avoid identifying non-entities as entities; set a relationship extraction scoring threshold, according to the translation embedding model formula, the smaller the score, the more reasonable the relationship, generally set ≤0.1 to ensure relationships between dredging entities. Adjusting knowledge fusion parameters: Set an entity alignment similarity threshold of ≥0.9 to accurately merge different representations of the same entity in the dredging field, avoiding duplicate nodes in the graph; set conflict resolution weighted voting weights (authoritative source / engineering measurement), prioritizing national standards for dredging parameters, and using engineering measurement data as corrections to match engineering compliance requirements. Generally, authoritative source (national standard) = 0.6, engineering measurement = 0.4. Adjusting knowledge query parameters: Set the maximum number of steps for multi-hop reasoning. Dredging process reasoning typically involves 5 to 8 steps, such as soil quality → applicable technology → supporting equipment → core parameters → construction requirements. Too many steps can lead to deviations in reasoning results.

[0071] In this embodiment, after determining that the problem classification result is knowledge reasoning, it is also necessary to determine the information organization structure classification corresponding to the current user problem, and then determine different knowledge enhancement tools based on the information organization structure classification. This achieves the goal of selecting a more suitable knowledge enhancement tool corresponding to the current user problem based on the specific information organization structure given by the current user problem, thereby providing a more accurate, easier and faster processing effect for generating the target generation result using the knowledge enhancement tool later.

[0072] Optionally, in this embodiment, after the tool matching agent determines the tool invocation strategy based on the problem classification result, the user context information, and the dredging tool library, the method further includes: If the tool invocation strategy is "no target invocation tool is found", then the unmatched target invocation tool will be identified as the target generated result and displayed.

[0073] Specifically, in practical applications, regardless of whether the problem classification result is in the computational analysis category or the knowledge reasoning category, there will be cases where no target calling tool is matched. For example, in the computational analysis category, if the required target calling function cannot be completely found or matched in the dredging scenario-based intelligent agent; or in the knowledge reasoning category, if the reasoning for the current user's problem cannot be completed by searching the augmented generation tool or the knowledge graph generated by knowledge augmentation, it can be regarded as a case where no target calling tool is matched. In this case, the target generation result is a case where no target calling tool is matched. At the same time, the target generation result can be displayed to the user, and the target generation result can be directly fed back to the maintenance staff of the dredging large model. Then, the maintenance staff can manually add and update the calling tool according to the current user's problem to further improve the dredging large model. Optionally, there is another implementation method, that is, when no target calling tool is matched, the base model can directly perform simple processing to obtain the obtainable result, and display the obtained result together with the result of no target calling tool matched to the user.

[0074] In this embodiment, if it is determined that no target calling tool is matched, the target generation result of the unmatched target calling tool can be displayed to the user and maintenance staff. This not only enables the user to receive an answer even when no target calling tool is matched, but also encourages maintenance staff to expand and improve the tool library of the dredging model.

[0075] S206. Invoke the target tool according to the tool invocation strategy, obtain and display the target generation result corresponding to the current user's problem.

[0076] The target tool is the "target calling function", "retrieval enhancement generation tool" or "knowledge graph generated by knowledge enhancement" as defined above.

[0077] Specifically, after determining the tool invocation strategy, the corresponding target tool can be invoked according to the tool invocation strategy to answer the user's context information (i.e., the current user's question), obtain the target generation result, and finally display the target generation result to the user.

[0078] Optionally, the dredging toolkit in this embodiment mainly includes professional knowledge and computational knowledge in the dredging field. Professional knowledge includes dredging terminology and common terminology; parameters of currently operational dredging vessels, specifically trailing suction hopper dredgers, cutter suction dredgers, grab bucket dredgers, and bucket wheel dredgers; popular science content related to major dredging companies, specifically the four major international dredging companies; typical engineering cases in dredging in recent years, specifically various dredging projects; dredging equipment and devices, specifically mainstream domestic and international mud pumps, pipelines, cutterheads, and rake heads; and timely updates on cutting-edge technologies in the dredging field. Computational knowledge in the dredging field includes: cutter suction dredger-related knowledge, trailing suction dredger-related knowledge, joint analysis frameworks, and other computational frameworks. Furthermore, the calculations related to the reamer include those for the cutter head, mud pump, pipelines, steel piles, and lateral anchors. The cutter head calculation module uses the Midemar cutter tooth cutting theory to calculate the cutting resistance of different soil types such as dry / saturated sand, clay, and rock, based on factors like tooth angle, soil properties, cutting thickness, and cutting speed. This calculation is used for reamer performance evaluation and excavability analysis. The mud pump calculation uses the Midemar formula to calculate and analyze the QH curve of the mud pump characteristics. Based on the clear water mud pump curve, and considering the full-power state of the mud pump, it is clear that there is a significant difference from actual construction, being significantly higher than the actual discharge pressure. Therefore, the efficiency percentage of the mud pump power under actual operating conditions is introduced for correction. To more accurately describe the law of discharge pressure change with rotational speed, a quadratic polynomial relationship between discharge pressure and the square of rotational speed can be established. Through the above-mentioned mud pump performance slurry QH curve calculation and analysis method, the operating characteristics of mud pumps at different wear stages of underwater and deck pumps can be analyzed. Based on the QH performance curve variation characteristics of mud pump slurry at different wear stages and the field test results of the minimum practical flow rate, construction suggestions and low-power delivery operation points for mud pump pipeline systems at different wear stages under single-pump and multi-pump construction conditions can be provided. Based on the variation characteristics of mud pump efficiency, mud pump purchase cost, and fuel consumption per 10,000 cubic meters with different wear stages, an optimal economic strategy for replacing cutter suction dredger mud pumps is achieved. The pipeline calculation module uses flow velocity, K coefficient, roughness, relative viscosity, concentration, density, particle settling velocity, and friction coefficient to perform calculations primarily on the transport resistance of dredged pipelines, using relevant models and formulas. In the pipe resistance calculation, the settling velocity of sand particles has a significant impact on the transport resistance calculation. Gravity theory states that the energy consumption of two-phase flow is greater than that of transporting clear water due to the increase in gravity; therefore, the Wushui formula is used to calculate the settling velocity of silt particles.A joint analysis and calculation framework is used to improve construction capacity across multiple dimensions throughout the entire process of cutter suction dredger excavation, pipeline transportation, and hydraulic reclamation. It constrains soil properties and engineering requirements, focusing on key parameters for overall dredger optimization. With overall dredger optimization as the goal, it comprehensively considers soil balance and dynamically allocates load based on individual cutter head and pump calculation parameters, providing construction process recommendations for different discharge distances in common soil types for specified equipment. It also includes process optimization and performance analysis under extreme working conditions for pre-construction cutter suction dredger construction process parameters, and the development of relay pump configuration schemes. For trailing suction hopper dredgers, the framework optimizes the entire process of dredging, navigation, and dumping based on haul distance, speed, vessel capacity, maximum load capacity, key soil parameters, energy consumption data for dredging, dumping, transportation, hydraulic reclamation, and unloading, and historical construction data such as historical output. Based on differences in haul distance and excavated soil types in actual projects, it balances and optimizes output and energy consumption, proposing overall dredger process recommendations for loading time / shipment, shipment / day, and earthwork volume / month. Other calculation frameworks include theoretical calculation methods related to other dredging vessels such as grab bucket dredgers, shovel bucket dredgers, and chain bucket dredgers; and theoretical calculation methods related to combined scraper and winch construction or combined construction of multiple types of dredging vessels.

[0079] Optionally, for the professional knowledge base in the dredging field, the core adjustment parameters for knowledge processing and incremental updates are adjusted to ensure the standardization and timeliness of knowledge updates. For example, adjusting general knowledge processing parameters: setting an optical character recognition accuracy threshold. Dredging project reports / drawings contain a large number of numerical parameters (such as equipment power and construction concentration). A low optical character recognition accuracy will lead to knowledge errors; an optical character recognition accuracy ≥ 0.95 matches the engineering precision. Adjusting the terminology standardization matching threshold. Dredging terminology has strong uniqueness; it is necessary to ensure that the matching degree between standardized terms and the domain dictionary is ≥ 0.98 to avoid terminology confusion leading to model comprehension bias. Adjusting specialized knowledge processing parameters: setting the granularity of process step decomposition, splitting by single operation steps, adapting to the model's accurate understanding and reasoning of the process, and avoiding the loss of process logic due to step merging. Adjust incremental update parameters: Set a similarity threshold for new knowledge fusion. When the similarity between newly added dredging knowledge and existing knowledge is ≥0.8, fusion and update are performed to avoid duplicate storage; if it is below 0.8, a new knowledge node is created. Set an update cycle for engineering cases. Classic dredging engineering cases are updated slowly, so quarterly updates are sufficient to match the accumulation speed of engineering cases. Set an update cycle for equipment / processes. The cutting-edge technologies of dredging equipment and processes are updated quickly, so monthly updates are performed to ensure the timeliness of model knowledge.

[0080] Optionally, parameter tuning for large dredging models can be divided into tuning of low-rank adaptation (LoRA) core fine-tuning parameters, training hyperparameters, and professional evaluation parameters. Specifically, for the core fine-tuning parameters of LoRA, the LoRA rank (R) is adjusted. Given the moderate amount of knowledge in the dredging domain and the dataset size being much smaller than general datasets, a R that is too small cannot fully inject professional knowledge, while a R that is too large easily leads to overfitting; 16-32 is the optimal range. The LoRA alpha (LoRA gradient scaling factor) is also adjusted. α=2×R achieves reasonable gradient scaling, ensuring efficient injection of dredging knowledge without compromising the model's original capabilities. The LoRA regularization is adjusted. Regularization is the probability of randomly deactivating some neurons during training, preventing the model from over-relying on noise in the training data. Since the dredging fine-tuning dataset has a small sample size, setting the regularization to 0.05-0.1 effectively prevents overfitting while retaining sufficient knowledge information. Finally, the fine-tuning modules are adjusted. The core of the attention layer is the Q (query vector), K (key vector), and V (value vector) matrices. Query projection is the projection layer that generates the Q matrix, and value projection is the projection layer that generates the V matrix. Only these two modules are trained, without changing the fully connected layers. This ensures the injection of dredging knowledge while minimizing training costs, matching the requirement for low training costs. Specifically, for training hyperparameters, adjust the learning rate. Dredging fine-tuning is domain-adaptive fine-tuning, so the learning rate needs to be much smaller than that of general model pre-training; 1e-4 to 3e-4 can avoid covering the native capabilities of general models. Adjust the batch size. The batch size is the number of samples fed into the model for each parameter update. Dredging fine-tuning datasets are relatively small, and the core rules between samples have high overlap. Too large a batch size will lead to training non-convergence; 2-4 is suitable for small sample training. Adjust the number of rounds. The number of rounds is the number of times the model completely traverses the entire training dataset. 3-5 rounds of training can achieve sufficient fitting of dredged knowledge. Too many rounds (such as 10+) will lead to severe overfitting, and the model will lose its general reasoning ability. Adjust the maximum sequence length. The maximum sequence length is the upper limit of the number of tokens that the model can process at one time. If it exceeds this, the text will be truncated, resulting in the loss of key information. Dredging computational question-answering pairs contain a large number of parameters and formulas, so the sequence length needs to be adapted to long texts. 1024-2048 can cover the complete calculation process and results, which is much longer than general dialogue text. Adjust the optimizer, using 8-bit quantized Adam. W-optimizer quantization significantly reduces memory usage, adapting to the hardware deployment conditions of dredging projects while ensuring training accuracy. A learning rate adjustment strategy using cosine annealing is employed, with a high learning rate initially for rapid learning of core rules, followed by a lower learning rate later for fine-tuning parameters and smooth convergence. This avoids parameter oscillations caused by excessively high learning rates in the later stages of training, ensuring stable injection of dredging knowledge. A dedicated evaluation framework is established for professional evaluation parameters, enabling multi-dimensional evaluation of the training results. Specifically, the evaluation considers knowledge accuracy, computational reliability, terminology standardization, inference depth, and robustness. For example, as shown in Table 1, after passing the evaluation, the fine-tuned model can be returned to the larger model base for use in professional question answering within the dredging field. Optionally, in this embodiment, the large-scale dredging model can continuously realize multiple rounds of questioning by the same user. The only difference between this and a single question is the "user context information". That is, in this embodiment, if the user asks questions in multiple rounds, the user context information in each round will include the historical context.

[0081] Figure 3 This is a schematic diagram of the structure of a dredging large-scale model result generation device provided in an embodiment of the present invention, as shown below. Figure 3 As shown, this device is applied to the overall control and scheduling intelligent agent of a large-scale dredging model; the device includes: The receiving module 301 is used to receive the user context information output by the base model of the dredging large model in response to the current user problem, and call the domain recognition agent to output the system prompt words corresponding to the current domain based on the user context information, so that the domain recognition agent can output the domain judgment result based on the system prompt words.

[0082] The classification module 302 is used to call the problem classification agent to determine the problem classification result corresponding to the current user problem based on the analysis and reasoning requirements of the current domain and the user context information when the domain judgment result is the dredging domain.

[0083] The output module 303 is used to call the tool matching agent to determine the tool calling strategy based on the problem classification results, user context information and dredging domain tool library, call the target tool according to the tool calling strategy, and obtain and display the target generation result corresponding to the current user problem.

[0084] Optionally, the domain recognition agent is invoked to output system prompts corresponding to the current domain based on the user's context information, so that the domain recognition agent outputs the domain judgment result based on the system prompts. The receiving module 301 is specifically used for: The domain recognition agent is invoked to extract features based on the user's purpose, context information, and question profile in the user's context information, thereby obtaining a domain feature vector. Based on the domain feature vector, the domain system prompt words are extracted from the domain prompt word library, and the domain judgment result is output based on the domain system prompt words and the user's context information.

[0085] Optionally, the classification module 302 is specifically used for: The problem classification agent is invoked to determine the problem classification feature vector based on the user context information, determine the classification system prompt words based on the problem classification feature vector and the domain routing label in the domain judgment result, and determine the problem classification result based on the classification system prompt words and the problem classification criteria.

[0086] Optionally, the tool invocation matching agent determines the tool invocation strategy based on the problem classification results, user context information, and the dredging domain tool library. The output module 303 is specifically used for: If the problem classification result is computational analysis, the tool matching agent determines the tool invocation strategy containing the target invocation function based on the model context protocol toolchain, user context information, and dredging domain tool library corresponding to the computational analysis category. If the problem classification result is knowledge reasoning, the tool matching agent determines the information organization structure classification corresponding to the current user problem based on the user context information and dredging domain tool library, and determines the tool invocation strategy containing the knowledge enhancement tool corresponding to the information organization structure classification.

[0087] Optionally, after the tool invocation matching agent determines the tool invocation strategy based on the problem classification results, user context information, and the dredging domain tool library, the output module 303 is further used for: If the tool invocation strategy is "no target invocation tool is found", then the unmatched target invocation tool will be identified as the target generated result and displayed.

[0088] Optionally, after the domain recognition agent outputs the system prompt word corresponding to the current domain based on the user context information, the receiving module 301 is further configured to: When the domain determination result is not in the dredging domain, the domain determination result and user context information are input into the base model so that the base model can determine the natural language understanding result based on the domain determination result and user context information, and display the natural language understanding result as the target generation result.

[0089] The dredging large model result generation device provided in the embodiments of the present invention can execute the dredging large model result generation method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0090] It is worth noting that in the embodiments of the above-mentioned dredging large model result generation device, the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy differentiation and are not used to limit the scope of protection of the present invention.

[0091] Figure 4This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. In this embodiment, the electronic device is the actual carrier of the overall control and scheduling intelligent agent of the dredging large-scale model (that is, the overall control and scheduling intelligent agent of the dredging large-scale model can run in a software entity or functional module on the electronic device). The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workbenches, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (such as helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the invention described and / or claimed herein.

[0092] like Figure 4 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0093] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0094] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as the results generation method for dredging large models.

[0095] In some embodiments, the method for generating the results of a large-scale dredging model can be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the method for generating the results of a large-scale dredging model described above can be performed. Alternatively, in other embodiments, processor 11 can be configured to perform the method for generating the results of a large-scale dredging model by any other suitable means (e.g., by means of firmware).

[0096] This invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements, for example, the method for generating results of a large-scale dredging model provided in this invention.

[0097] The computer storage medium of this invention can be any combination of one or more computer-readable media. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0098] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of sending, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0099] Program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0100] This invention also provides a computer program product, including a computer program that, when executed by a processor, implements the method for generating results of a large dredging model as provided in any embodiment of this invention.

[0101] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages—such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0102] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computing device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0103] Furthermore, the acquisition, storage, use, and processing of data in the technical solution of this invention all comply with the relevant provisions of national laws and regulations.

[0104] Note that the above description is merely a preferred embodiment of the present invention and the technical principles employed. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions can be made without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in detail through the above embodiments, the present invention is not limited to the above embodiments, and may include many other equivalent embodiments without departing from the concept of the present invention, the scope of which is determined by the scope of the appended claims.

Claims

1. A method for generating results of a large-scale dredging model, characterized in that, A master control and scheduling agent applied to a large-scale dredging model; the method includes: The system receives user context information output by the base model of the dredging model in response to the current user's problem, and calls the domain recognition agent to output system prompt words corresponding to the current domain based on the user context information, so that the domain recognition agent outputs the domain judgment result based on the system prompt words. When the domain determination result is dredging, the problem classification agent is invoked to determine the problem classification result corresponding to the current user's problem based on the analysis and reasoning requirements corresponding to the current domain and the user context information. The tool matching agent determines the tool invocation strategy based on the problem classification results, the user context information, and the dredging tool library, and invokes the target tool according to the tool invocation strategy to obtain and display the target generation result corresponding to the current user problem.

2. The method according to claim 1, characterized in that, The domain identification agent outputs a system prompt word corresponding to the current domain based on the user context information, so that the domain identification agent outputs a domain judgment result based on the system prompt word, including: The domain identification agent is invoked so that it can extract features based on the user's purpose, context information, and user question profile in the user context information to obtain a domain feature vector. Based on the domain feature vector, the agent extracts domain system prompt words from the domain prompt word library and outputs the domain judgment result based on the domain system prompt words and the user context information.

3. The method according to claim 1, characterized in that, The invoking question classification agent determines the question classification result corresponding to the current user's question based on the analysis and reasoning requirements of the current domain and the user context information, including: The problem classification agent is invoked so that it determines a problem classification feature vector based on the user context information, determines a classification system prompt word based on the problem classification feature vector and the domain routing label in the domain judgment result, and determines the problem classification result based on the classification system prompt word and the problem classification criteria.

4. The method according to claim 1, characterized in that, The tool invocation matching agent determines the tool invocation strategy based on the problem classification results, the user context information, and the dredging tool library, including: If the problem classification result is a computational analysis class, then the tool matching agent is invoked to determine the tool invocation strategy containing the target invocation function based on the model context protocol toolchain corresponding to the computational analysis class, the user context information, and the dredging domain tool library. If the problem classification result is a knowledge reasoning type, then the tool matching agent is invoked to determine the information organization structure classification corresponding to the current user problem based on the user context information and the dredging domain tool library, and to determine the tool invocation strategy that includes the knowledge enhancement tools corresponding to the information organization structure classification.

5. The method according to claim 4, characterized in that, The tool matching agent classifies the information organization structure corresponding to the current user problem as determined by the user context information and the dredging domain tool library, including fragmented information organization structure and structured information organization structure. If the information organization structure is classified as a fragmented information organization structure, then the knowledge enhancement tool corresponding to the information organization structure classification is a retrieval enhancement generation tool. If the information organization structure is classified as a structured information organization structure, then the knowledge enhancement tool corresponding to the information organization structure classification is a knowledge graph generated by knowledge enhancement.

6. The method according to claim 1, characterized in that, After the tool invocation matching agent determines the tool invocation strategy based on the problem classification results, the user context information, and the dredging domain tool library, the process further includes: If the tool invocation strategy is that no target invocation tool is matched, then the unmatched target invocation tool is determined as the target generation result, and the target generation result is displayed.

7. The method according to claim 6, characterized in that, After the domain recognition agent outputs the system prompt word corresponding to the current domain based on the user context information, the process also includes: When the domain determination result is not a dredging domain, the domain determination result and the user context information are input into the base model, so that the base model determines the natural language understanding result based on the domain determination result and the user context information, and displays the natural language understanding result as the target generation result.

8. A device for generating results of a large-scale dredging model, characterized in that, A master control and scheduling intelligent agent applied to a large-scale dredging model; the device includes: The receiving module is used to receive user context information output by the base model of the dredging large model in response to the current user problem, and call the domain recognition agent to output the system prompt word corresponding to the current domain based on the user context information, so that the domain recognition agent outputs the domain judgment result based on the system prompt word; The classification module is used to, when the domain judgment result is the dredging domain, call the problem classification agent to determine the problem classification result corresponding to the current user's problem based on the analysis and reasoning requirements corresponding to the current domain and the user context information; The output module is used to call the tool matching agent to determine the tool calling strategy based on the problem classification result, the user context information and the dredging domain tool library, call the target tool according to the tool calling strategy, and obtain and display the target generation result corresponding to the current user problem.

9. An electronic device, characterized in that, include: One or more processors; Memory, used to store one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the method for generating results of a large dredging model as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method for generating results of the large dredging model as described in any one of claims 1 to 7.

Images