An AI agent-based defense briefing assisted generation method and system

Abstract

The application relates to the field of intelligent water affairs and artificial intelligence technology, and provides a defense briefing auxiliary generation method and system based on an AI intelligent agent, which comprises the following steps: performing semantic analysis and intention recognition, and extracting basic parameters; performing integrity checking on the basic parameters; converting selected defense briefing template texts into high-dimensional vector representations; performing similarity retrieval, obtaining candidate knowledge document segments, scoring the correlation degrees, reordering the candidate knowledge segments, and outputting the top N most relevant knowledge segments; identifying data requirements indicated by each mask information in the template, and generating a structured query statement; querying required actual data from each relevant data table; and based on the actual data, replacing mask identifiers in the template item by item to generate a defense briefing. The application realizes full-process automation of defense briefing generation, can shorten the briefing generation time from several hours to several minutes, and significantly reduces the labor cost and time cost.

Description

Technical Field

[0001] This invention relates to the fields of smart water management and artificial intelligence technology. Specifically, it relates to a method and system for assisting in the generation of defense briefings based on AI intelligent agents. In particular, it relates to the use of large language models, retrieval-enhanced generation technology, template mask intelligent filling, and workflow orchestration technology to achieve automated generation and multi-terminal push of defense briefings in urban flood control management scenarios. Background Technology

[0002] Currently, there are several existing technical solutions related to this invention in the field of flood control and emergency management. Regarding the generation of flood control briefings and reports, existing technologies have proposed template-based methods for generating flood control briefings. These methods involve manually collecting data from various business systems and filling it into a pre-set template to generate a flood control briefing. Other existing technologies involve flood control decision support systems based on data warehouses, which generate flood control situation analysis reports by integrating multi-source data. Furthermore, existing technologies have disclosed methods for rapidly generating flood control plans based on knowledge graphs, utilizing knowledge graph technology to assist in generating flood control plan documents.

[0003] In existing technologies, generating flood control and defense briefings requires manual data retrieval from multiple operational systems, including meteorological, hydrological monitoring, engineering condition monitoring, and urban flooding management systems, followed by manual aggregation and input into a pre-defined briefing template. Each briefing generation consumes significant manpower and time, especially when generating frequent briefings (such as daily, weekly, and emergency reports), where the repetitive workload is extremely heavy. Current technologies lack an automated mechanism capable of automatically extracting data from multi-source heterogeneous data systems and intelligently populating the briefing template. Therefore, existing technologies lack an automated mechanism capable of automatically extracting data from multi-source heterogeneous data systems and intelligently populating the briefing template. Summary of the Invention

[0004] To automate the entire process of defense briefing generation, this invention provides a defense briefing auxiliary generation method and system based on AI intelligent agents, the specific technical solution of which is as follows: A method for assisting in the generation of defense briefings based on AI agents includes the following steps: S1 receives the defense briefing generation instruction input by the user, performs semantic parsing and intent recognition, and extracts the basic parameters required for briefing generation; S2, after recognizing the user's intent to generate the defense briefing, performs a completeness check on the basic parameters. If any required parameters are missing, the missing parameter items are identified and a request to the user to supplement them is sent until the completeness check of the basic parameters passes. S3, select the corresponding defense briefing template from the preset template library and convert the selected defense briefing template text into a high-dimensional vector representation to obtain the template vector. The defense briefing template contains preset mask information, which is used to identify the positions in the template that need to be replaced by actual data. S4. Based on the template vector, perform similarity retrieval in the vector database of the local water conservancy knowledge base to obtain candidate knowledge document fragments related to the content of the defense briefing template. Score the degree of correlation between the candidate knowledge fragments and the template content, and re-rank the candidate knowledge fragments according to the score. Output the top N most relevant knowledge fragments. S5 uses the top N most relevant knowledge fragments and defense briefing templates as input prompts to identify the data requirements indicated by each mask information in the template and generate structured query statements. S6 submits the structured query statement to the relational database of the smart water management platform for execution, and retrieves the required actual data from the relevant data tables; S7, based on actual data, replaces the mask identifiers in the template item by item according to the correspondence with the mask information in the template, and generates a defense briefing.

[0005] The AI-based defense briefing generation method described above constructs a complete technical chain encompassing instruction understanding, intelligent parameter acquisition, knowledge retrieval, multi-source data querying, and template mask filling. Users only need to input simple commands such as "Please generate today's defense briefing" in natural language, and the system can automatically complete the entire process from parameter acquisition and data querying to briefing generation. Compared to existing technologies that require manual data querying from multiple business systems and manual template filling, this invention automates the entire defense briefing generation process, reducing briefing generation time from hours to minutes, significantly lowering labor and time costs.

[0006] Preferably, step S1 specifically includes: S11, obtains the defense briefing generation instructions input by the user in natural language text through the large language model interaction module; S12: Identify whether the defense briefing generation instruction belongs to the defense briefing generation intention. If so, extract the basic parameters required for briefing generation from the defense briefing generation instruction and jump to S2. If not, proceed to the general question and answer process or prompt the user to re-enter.

[0007] Preferably, step S3 specifically includes: S31. After the parameter integrity verification is passed, the large language model calls the defense briefing function and selects the corresponding defense briefing template from the preset template library. The template library has multiple formats of defense briefing templates. Different defense briefing templates are suitable for different defense scenarios. The defense briefing template contains preset mask information, which is used to identify the positions in the template that need to be replaced by actual data. S32, using the Embedding model, the selected defense briefing template text is transformed into a high-dimensional vector representation, resulting in the template vector.

[0008] Preferably, step S5 specifically includes: S51, the top N most relevant knowledge fragments and defense briefing templates are used as input prompts for the large language model, and the large language model identifies the data requirements indicated by each mask information in the template; S52, after identifying the data requirements of each mask information, generates a structured query statement from the large language model.

[0009] Preferably, step S6 specifically includes: S61, submit the structured query statement to the relational database of the smart water platform for execution, query the required actual data from the relevant data tables, and return the query results in the form of structured data; S62. If a certain type of data query fails or the query result is empty, the system records the abnormal situation and fills the corresponding mask position with a preset identifier when the subsequent briefing is generated. At the same time, a data missing log is generated for management personnel to refer to and supplement the data later.

[0010] Preferably, step S7 specifically includes: S71, perform data mapping, and establish a one-to-one correspondence between the fields in the query results and the mask identifiers in the defense briefing template; S72 performs data formatting and aggregation, formatting the raw data according to the format requirements of each data item in the defense briefing template; S73, perform template filling, fill the formatted data item by item into the corresponding mask position of the defense briefing template to generate a complete defense briefing.

[0011] Preferably, the method for assisting in the generation of defense briefings based on AI agents further includes: S8. The defense briefing function of the smart water management platform is called through the briefing generation module to process the defense briefing generated in step S7 and generate a formal defense briefing.

[0012] Preferably, step S12 specifically includes: S121, a pre-set business keyword library associated with the defense briefing is configured, and word segmentation is performed on the defense briefing generation instructions; S122, if the defense briefing generation instruction after word segmentation hits any keyword, it is determined to be a defense briefing generation intention; otherwise, the defense briefing generation instruction is converted into a vector using the Embedding model and cosine similarity is calculated with the preset defense briefing generation intention standard template vector. S123, if the similarity exceeds the preset threshold, it is determined to be an intention to generate a defense briefing; otherwise, it proceeds to the general Q&A process or prompts the user to re-enter the information.

[0013] Preferably, step S4 specifically includes: S41, the candidate knowledge fragments and the defense briefing template are encoded independently, the semantic vectors of the fragments and the semantic vectors of the template are obtained, and the cosine similarity between the two is calculated to obtain the semantic similarity score. S42, extract business keywords from candidate knowledge fragments and defense briefing templates respectively, obtain fragment keyword sets and template keyword sets, calculate the Jaccard coefficient of the two, and obtain the business keyword overlap. S43, extract water conservancy entities from candidate knowledge fragments and defense briefing templates respectively, obtain fragment water conservancy entity set and template water conservancy entity set, and obtain water conservancy entity alignment degree based on fragment water conservancy entity set and template water conservancy entity set; S44. The semantic similarity score, the overlap of business keywords, and the alignment of water conservancy entities are weighted and fused to obtain the relevance score.

[0014] A defense briefing assistance generation system based on AI agents, used to implement the aforementioned defense briefing assistance generation method based on AI agents, includes: The large language model interaction module is used to receive user input instructions for generating defense briefings, perform semantic parsing and intent recognition, and extract the basic parameters required for briefing generation. The parameter missing identification and proactive inquiry module is used to perform integrity verification on the basic parameters after identifying the user's intent to generate the defense briefing. If any required parameters are missing, the module identifies the missing parameter items and sends a supplementary request to the user until the integrity verification of the basic parameters passes. The knowledge base retrieval module is used to select the corresponding defense briefing template from the preset template library and convert the selected defense briefing template text into a high-dimensional vector representation to obtain the template vector. The defense briefing template contains preset mask information, which is used to identify the positions in the template that need to be replaced by actual data. The Reranker module is used to perform similarity retrieval in the vector database of the local water conservancy knowledge base based on template vectors, obtain candidate knowledge document fragments related to the content of the defense briefing template, score the degree of correlation between the candidate knowledge fragments and the template content, re-rank the candidate knowledge fragments according to the score, and output the top N most relevant knowledge fragments. The SQ auto-generation module is used to identify the data requirements indicated by each mask information in the template by taking the top N most relevant knowledge fragments and defense briefing templates as input prompts and generating structured query statements. The defense briefing template filling module is used to submit structured query statements to the relational database of the smart water management platform for execution. It retrieves the required actual data from various relevant data tables, and based on the actual data, replaces the mask identifiers in the template item by item according to the correspondence with the mask information in the template, thereby generating a defense briefing. Attached Figure Description

[0015] The invention will be further understood from the following description taken in conjunction with the accompanying drawings. The components in the drawings are not necessarily drawn to scale, but rather the emphasis is on illustrating the principles of the embodiments. In different views, the same reference numerals designate corresponding parts.

[0016] Figure 1 This is a schematic diagram of the overall process of a defense briefing auxiliary generation method based on AI intelligent agents in one embodiment of the present invention; Figure 2 This is a flowchart illustrating the specific method of step S1 in one embodiment of the present invention; Figure 3 This is a flowchart illustrating the specific method of step S12 in one embodiment of the present invention; Figure 4 This is a flowchart illustrating the specific method of step S3 in one embodiment of the present invention; Figure 5 This is a flowchart illustrating the specific method of step S4 in one embodiment of the present invention; Figure 6 This is a flowchart illustrating the specific method of step S5 in one embodiment of the present invention; Figure 7 This is a flowchart illustrating the specific method of step S6 in one embodiment of the present invention; Figure 8 This is a flowchart illustrating the specific method of step S7 in one embodiment of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to its embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not limit the scope of protection of the invention.

[0018] like Figure 1 As shown, an embodiment of the present invention provides a defense briefing assistance generation method based on AI intelligent agents, comprising the following steps: S1 receives the defense briefing generation instructions input by the user, performs semantic parsing and intent recognition, and extracts the basic parameters required for briefing generation.

[0019] As a preferred technical solution, such as Figure 2 As shown, step S1 specifically includes: S11 obtains the defense briefing generation instructions input by the user in natural language text through the large language model interaction module.

[0020] Users can input defense briefing generation commands in natural language text format through the large language model interaction module. For example, users can input commands such as "Please generate today's defense briefing" or "Generate a defense briefing including weather conditions for the next three days".

[0021] S12: Identify whether the defense briefing generation instruction belongs to the defense briefing generation intention. If so, extract the basic parameters required for briefing generation from the defense briefing generation instruction and jump to S2. If not, proceed to the general question and answer process or prompt the user to re-enter.

[0022] The large language model performs semantic parsing and intent recognition on user-input defense briefing generation instructions, specifically including: (1) Intent Classification: Identify whether the user's instruction belongs to the "defense briefing generation" intent. If it does, proceed to the briefing generation process; if it does not, proceed to the general Q&A process or prompt the user to re-enter the information.

[0023] (2) Extraction of basic parameters: Extract the basic parameters required for the generation of the briefing from the user instructions, including but not limited to the generation date (such as "today" or "June 29, 2026"), forecast period (such as "the next 3 days" or "the next 7 days"), weather description (such as "sunny", "rainy", or "typhoon" given by the user) and briefing type (such as "daily defense briefing" or "emergency defense briefing").

[0024] As a preferred technical solution, such as Figure 3 As shown, step S12 specifically includes: S121, a pre-set business keyword library associated with the defense briefing is configured, and word segmentation is performed on the defense briefing generation instructions.

[0025] S122, if the defense briefing generation instruction after word segmentation hits any keyword, it is determined to be a defense briefing generation intention; otherwise, the defense briefing generation instruction is converted into a vector using the Embedding model and its cosine similarity is calculated with the preset defense briefing generation intention standard template vector.

[0026] S123, if the similarity exceeds the preset threshold, it is determined to be an intention to generate a defense briefing; otherwise, it proceeds to the general Q&A process or prompts the user to re-enter the information.

[0027] Specifically, in this invention, the identification of the intent to generate a defense briefing can employ a two-tiered strategy: keyword matching as a priority and semantic similarity as a fallback. The system pre-configures a database of business keywords strongly associated with defense briefings, including but not limited to: defense briefing, flood control briefing, briefing generation, push briefing, defense information, briefing push, defense status, and flood control dynamics. After segmenting the user-input defense briefing generation command, if any business keyword is matched, it is directly determined as the intent to generate a defense briefing. If no keyword is matched, the system uses an embedding model to convert the user's command into a vector and performs cosine similarity calculation with a pre-defined standard template vector for defense briefing generation intent. If the similarity exceeds a pre-defined threshold, such as 0.75, it is also classified as the intent to generate a defense briefing. If both levels of matching fail, the system proceeds to a general question-and-answer process or returns a prompt. After successful matching, core parameters such as the generation date and future weather conditions are extracted; if parameters are missing, follow-up questions are proactively asked to supplement them.

[0028] This invention automatically verifies the completeness of parameters after the user inputs a command. If key parameters are detected to be missing (such as release date, weather conditions, etc.), the system proactively asks the user clarifying questions, guiding them to supplement the necessary information in a friendly and natural manner. Compared to existing technologies that exit with an error or generate an incomplete report upon missing parameters, this invention's proactive questioning mechanism ensures that complete parameter input is obtained for each report generation, improving the process completion rate and user experience.

[0029] S2, after recognizing the user's intent to generate the defense briefing, performs a completeness check on the basic parameters. If any required parameters are missing, it identifies the missing parameter items and sends a request to the user to supplement them, until the completeness check of the basic parameters passes.

[0030] The integrity of the basic parameters extracted in step S1 can be verified by constructing a parameter missing identification and proactive query module. Required parameters for generating the defense briefing include, but are not limited to: Release Date / Generation Date: The effective date of the briefing; Future weather conditions: A description of the weather conditions during the forecast period.

[0031] If all required parameters have been extracted completely, proceed to step S3. If any required parameters are missing, the system automatically identifies the missing parameters and initiates clarifying follow-up questions to the user through the large language model interaction module, specifically including the following points: (1) Missing parameter identification: For example, if a user enters "generate defense briefing" but does not specify the date and weather, the system will identify that the two parameters "generation date" and "future weather conditions" are missing.

[0032] (2) Proactive follow-up question generation: The system calculates the specific dates for the next 3 days (e.g., "June 30, July 1, July 2") based on the current time (e.g., "2026-06-29 10:59"), and asks the user in a clear and friendly manner: "Based on your needs, the following parameters are required to generate a defense briefing: 1. Release date: The next 3 days are June 30, July 1, and July 2. Which day's defense briefing would you like to generate? 2. Weather conditions: Do you have a specific description of the weather conditions (e.g., sunny, rainy, typhoon, etc.)?"

[0033] (3) User supplementation: After receiving follow-up questions, the user provides supplementary parameter information. After the system obtains the supplementary information, it merges it into the parameter set and continues to execute the subsequent process.

[0034] S3. Select the corresponding defense briefing template from the preset template library and convert the selected defense briefing template text into a high-dimensional vector representation to obtain the template vector. The defense briefing template contains preset mask information, which is used to identify the positions in the template that need to be replaced by actual data.

[0035] As a preferred technical solution, such as Figure 4 As shown, step S3 specifically includes: S31. After the parameter integrity verification is passed, the large language model calls the defense briefing function and selects the corresponding defense briefing template from the preset template library. The template library has multiple formats of defense briefing templates. Different defense briefing templates are suitable for different defense scenarios. The defense briefing template contains preset mask information, which is used to identify the positions in the template that need to be replaced by actual data.

[0036] S32, using the Embedding model, the selected defense briefing template text is transformed into a high-dimensional vector representation, resulting in the template vector.

[0037] Specifically, different templates are suitable for different defense scenarios, such as daily defense briefings, emergency response briefings, and flood season daily reports. The mask information is used to identify the locations in the template that need to be replaced with actual data, for example: {{Weather Conditions}}: This field requires the weather conditions for the forecast period to be entered; {{Warning Status}}: This field requires the currently active warning level and warning information to be entered; {{Emergency Response Status}}: This field requires the current emergency response level and status to be entered. {{Reservoir Water Information}}: This field requires information such as water level and capacity of each reservoir. {{River Water Conditions}}: This field requires information such as water level and flow rate of the main river. {{Flooding Situation}}: This indicates the location where the water accumulation information for each flooded point needs to be filled in; {{Manhole Water Level}}: This indicates that manhole water level monitoring information needs to be entered at this location; {{Rainfall Information}}: This field requires information such as rainfall amount and intensity. {{Facility Operating Conditions}}: This indicates the operating conditions of drainage facilities, pumping stations, etc., that need to be filled in at this location; {{Deployment Status}}: This indicates that information on the deployment of flood control materials and personnel needs to be entered at this location; {{Flooding Incident}}: This field requires information about a flooding incident that has already occurred. {{Emergency Incident}}: This field requires information about an emergency incident that has already occurred. {{Event Handling Status}}: This field should be filled with the progress of handling various events.

[0038] The Embedding model can be used to convert the selected defense briefing template text into a high-dimensional vector representation, thus obtaining the template vector.

[0039] S4. Based on the template vector, perform similarity retrieval in the vector database of the local water conservancy knowledge base to obtain candidate knowledge document fragments related to the content of the defense briefing template. Score the degree of correlation between the candidate knowledge fragments and the template content, and re-rank the candidate knowledge fragments according to the score. Output the top N most relevant knowledge fragments.

[0040] Specifically, the template vector generated in step S3 is used as the query vector to perform a similarity search in the vector database of the local water resources knowledge base, retrieving knowledge document fragments related to the content of the defense briefing template. The search process uses vector similarity measurement methods such as cosine similarity or Euclidean distance to return the top K candidate knowledge fragments (K is a preset value, such as K=20) in terms of similarity.

[0041] Input the K candidate knowledge fragments obtained from the preliminary retrieval into the Reranker re-ranking module. The Reranker model adopts the Cross-Encoder architecture to perform fine-grained scoring on the correlation degree between each candidate knowledge fragment and the template content, re-rank the candidate knowledge fragments according to the scores, and output the top N (N is a preset value, N<K, for example, N=5) most relevant knowledge fragments.

[0042] The function of re-ranking lies in that the preliminary vector retrieval may recall some fragments with low correlation. Through the fine re-ranking of the Reranker model, the accuracy of knowledge retrieval can be effectively improved, ensuring that the subsequent large language model can obtain the most relevant flood control professional knowledge to assist in understanding various mask information in the template as well as their business meanings and data sources.

[0043] The Reranker model adopting the Cross-Encoder architecture splices the template text and each candidate knowledge fragment into the format of <[BOS_never_used_51bce0c785ca2f68081bfa7d91973934]> template content [SEP] candidate fragment [SEP], and outputs a matching score through the fully connected layer after Transformer encoding.

[0044] Exemplarily, the scoring formula is: Score=α×semantic_sim+β×key_overlap+γ×entity_alignment, where semantic_sim is the cosine similarity between <[BOS_never_used_51bce0c785ca2f68081bfa7d91973934]> vectors, key_overlap is the Jaccard coefficient of business keywords (such as waterlogging, water level, early warning, etc.), entity_alignment is the hit rate of water conservancy entities (such as section names, waterlogging points, indicators, etc.), α=0.6, β=0.2, γ=0.2. Finally, the Top-N most relevant knowledge fragments are selected in descending order of the weighted total score.

[0045] The present invention adopts a two-stage retrieval strategy of vector retrieval and Reranker re-ranking, which can ensure that the large language model can obtain high-quality professional knowledge support when understanding template mask information and generating SQL statements, and improves the accuracy of mask semantic understanding and the standardization of data query.

[0046] As a preferred technical solution, such as Figure 5 shown, step S4 specifically includes: S41, independently encode the candidate knowledge fragments and the defense brief template respectively, obtain the fragment semantic vectors and the template semantic vectors, and calculate the cosine similarity between the two to obtain the semantic similarity score.

[0047] Candidate knowledge fragments and defense briefing templates can be independently encoded using Transformer to extract their fragment semantic vectors and template semantic vectors. Cosine similarity is between 0 and 1, with higher values ​​indicating stronger overall semantic association. This semantic similarity score is primarily used to capture deeper semantic correspondences beyond literal expressions, such as the semantic matching between the mask for "{{reservoir water conditions}}" in the template and the candidate fragment "reservoir real-time water level and storage capacity monitoring specifications".

[0048] S42, extract business keywords from candidate knowledge fragments and defense briefing templates respectively, obtain fragment keyword sets and template keyword sets, and calculate the Jaccard coefficient of the two to obtain the business keyword overlap.

[0049] A dedicated keyword library for flood control and defense briefings can be pre-built, covering five major categories: meteorology, hydrology, engineering conditions, emergency response, and handling. It includes professional terms such as "waterlogging," "warning water level," "pump station," "reservoir capacity," "hazard deployment," and "early warning response." A domain-specific word segmentation tool is used to extract business keywords from candidate knowledge fragments and the briefing template, obtaining both a set of fragment keywords and a set of template keywords. The Jaccard coefficient between the two is calculated, and this coefficient serves as the business keyword overlap rate. The overlap rate ranges from 0 to 1, with higher values ​​indicating more overlapping keywords. This overlap rate is primarily used to forcibly filter knowledge fragments closely related to the flood control business theme, excluding similar semantic texts from general domains such as general emergency management and other industry defense texts, ensuring the domain relevance of the recalled content.

[0050] S43, extract water conservancy entities from candidate knowledge fragments and defense briefing templates respectively, obtain fragment water conservancy entity sets and template water conservancy entity sets, and obtain water conservancy entity alignment based on fragment water conservancy entity sets and template water conservancy entity sets.

[0051] This dimension is designed to meet the structured data requirements of defense briefings, measuring the degree of match between the water conservancy entities contained in the candidate segments and the template requirements. A localized water conservancy entity database can be built first, covering hydrological facility entities, indicator entities, event entities, and response entities. Hydrological facility entities include the names of reservoirs, river sections, flood-prone areas, pumping stations, and sluice gates; indicator entities include water levels, flow rates, rainfall, reservoir capacity, and water depth; event entities include rainstorm warnings, emergencies, and flooding events; and response entities include emergency response teams, flood control materials, and deployment locations. At the same time, key entities such as local main rivers and key reservoirs are given higher weights.

[0052] Named Entity Recognition (NER) is used to extract water conservancy entities from candidate knowledge fragments and defense briefing templates, and the alignment ratio is calculated according to the entity weight. The alignment degree of the water conservancy entity is between 0 and 1, which is mainly used to ensure that the recalled knowledge fragments contain the specific business objects and indicators required by the template. For example, when the template contains {{waterlogging situation}}, knowledge fragments containing specific waterlogging points and water depth indicators are prioritized.

[0053] For example, the alignment degree of the hydraulic entity = E1 represents the set of all water conservancy professional entities extracted by the Named Entity Recognition (NER) model in the water conservancy field in the defense briefing template, i.e., the template water conservancy entity set; if the template contains the masks {{reservoir water conditions}} and {{early warning status}}, then E1 = {reservoir, water level, water storage, rainstorm warning, warning level}.

[0054] E2 represents the set of all water conservancy professional entities extracted from the same water conservancy NER model within a single candidate knowledge fragment, i.e., the fragment water conservancy entity set; if the content of a candidate fragment is "Reservoir water level monitoring needs to be reported daily, and rainstorm warning information should be updated synchronously", then E2 = reservoir, water level, rainstorm warning. E1∩E2 represents the intersection between the fragment water conservancy entity set and the template water conservancy entity set.

[0055] "e" generally refers to a single water conservancy professional entity and is a basic component of the set; This is the business weight coefficient corresponding to a single water conservancy entity e, used to reflect the importance of the entity in the defense briefing. The higher the importance, the greater the weight value.

[0056] For core entities, which directly correspond to the core data items in the briefing, such as the name of key reservoirs, warning water levels, locations of waterlogged areas, emergency response levels, and hazard levels, these entities are the core information carriers of the defense briefing. When a match is successful, they contribute a higher score, and the business weight coefficient is set to 2.

[0057] Ordinary entities refer to general water conservancy terms, auxiliary facilities or indicators, such as pumping stations, gates, rainfall, flood control materials, etc., which are routine business descriptions. The matching contribution is the benchmark value, and the business weight coefficient is set to 1.

[0058] S44. The semantic similarity score, the overlap of business keywords, and the alignment of water conservancy entities are weighted and fused to obtain the relevance score.

[0059] For example, the scoring formula is: Relevance Score = α × Semantic Similarity Score + β × Business Keyword Overlap + γ × Water Conservancy Entity Alignment.

[0060] S5 uses the top N most relevant knowledge fragments and defense briefing templates as input prompts to identify the data requirements indicated by each mask information in the template and generate structured query statements.

[0061] As a preferred technical solution, such as Figure 6 As shown, step S5 specifically includes: S51, the top N most relevant knowledge fragments and defense briefing templates are used as input prompts for the large language model, which identifies the data requirements indicated by each mask information in the template.

[0062] S52, after identifying the data requirements of each mask information, generates a structured query statement from the large language model.

[0063] Specifically, the large language model uses the N most relevant knowledge fragments after the reordering in step S4, along with the flood control briefing template, as input prompts. The large language model understands the data requirements indicated by each mask information in the template based on two aspects: First, semantic analysis of the mask information: analyzing the literal meaning of mask identifiers (such as {{reservoir water level}}) and their business context in the flood control briefing template; second, auxiliary reference of knowledge fragments: referring to retrieved flood control professional knowledge fragments to understand the standard expression methods, data sources, and collection specifications of various water level, rainfall, and engineering data in the flood control briefing.

[0064] After fully understanding the data requirements of each mask information, the large language model automatically generates structured query language (SQL) statements for querying the relational database of the smart water management platform. The generated SQL statements include, but are not limited to, queries for the following data types: Weather forecast: Check weather conditions, temperature, wind speed, and precipitation probability for the forecast period; Warning status inquiry: Inquire about currently effective warning information such as rainstorm warnings, typhoon warnings, and flood warnings; Emergency Response Status Inquiry: Inquire about the current emergency response level, response activation time, response status, etc. Reservoir water information query: query the real-time water level, water storage, reservoir capacity ratio, inflow and outflow of each major reservoir; River water level inquiry: Inquire about the water level, flow rate, and warning water level difference of major rivers; Flooding Information Inquiry: Inquire about the water depth, area, and affected range of each flooded location; Manhole water level inquiry: Inquire about manhole water level, overflow status, etc.; Rainfall information query: query rainfall amount, rainfall intensity, rainfall duration, etc.; Facility status inquiry: Inquire about the operating status, opening and closing frequency, etc. of facilities such as drainage pumping stations and gates; Deployment status inquiry: Inquire about flood control material reserves, emergency rescue team deployment locations, personnel configuration, etc.; Flooding Event Inquiry: Search for recorded flooding events, including their name, occurrence time, and handling status; Emergency Incident Inquiry: Inquire about recorded emergency incident types, locations, urgency levels, etc. Incident handling status inquiry: Inquire about the handling measures, progress, and responsible persons for each incident.

[0065] The SQL statement generation takes into full consideration the database table structure, field names, data types, and relationships between tables, ensuring that the generated SQL statements can be correctly executed in the relational database of the smart water management platform.

[0066] This invention leverages the semantic understanding capabilities of a large language model, enabling the model to autonomously analyze the business meaning and data requirements of each mask information (such as {{reservoir water level}}, {{urban flooding event}}, etc.) in the template and automatically generate corresponding SQL query statements. Compared to existing technologies that require manual pre-configuration of data source tables and field names for each mask, this invention eliminates the need for manual maintenance of complex mapping relationships. When the database table structure changes or the template is adjusted, the large language model can automatically adapt, significantly reducing system maintenance costs.

[0067] Generating defense briefings requires querying more than ten categories of data, including meteorological, early warning, emergency response, reservoir, river, urban flooding, manhole, rainfall, engineering status, deployment, incident, and handling data. This data is scattered across different subsystems of the smart water management platform. This invention automatically generates cross-table SQL queries using a large language model, retrieving all necessary data at once without requiring manual table-by-table queries.

[0068] S6 submits the structured query statement to the relational database of the smart water management platform for execution, and retrieves the required actual data from the relevant data tables.

[0069] As a preferred technical solution, such as Figure 7 As shown, step S6 specifically includes: S61 submits the structured query statement to the relational database of the smart water management platform for execution, retrieves the required actual data from the relevant data tables, and returns the query results in the form of structured data.

[0070] S62. If a certain type of data query fails or the query result is empty, the system records the abnormal situation and fills the corresponding mask position with a preset identifier when the subsequent briefing is generated. At the same time, a data missing log is generated for management personnel to refer to and supplement the data later.

[0071] The SQL statement generated in step S5 is submitted to the relational database of the smart water management platform for execution, and the required actual data is retrieved from the relevant data tables. The query results are returned in the form of structured data, including the aforementioned data on water conditions, rainfall, engineering conditions, events, etc.

[0072] If a certain type of data query fails or the query result is empty (e.g., a reservoir has not yet been connected to the real-time monitoring system), the system records the abnormal situation and fills the corresponding mask position with a specific identifier (e.g., "data collection abnormal" or "-") when generating subsequent reports. At the same time, a data missing log is generated for management personnel to refer to and supplement the data later.

[0073] In this invention, the relational database of the smart water management platform can adopt a custom table structure, designed in modules according to the business requirements of defense briefings. The core tables include: a meteorological forecast data table (including fields such as region ID, date, weather conditions, temperature, and wind force), an early warning information table (including early warning ID, early warning type, level, release time, and status), an emergency response table (including response ID, level, activation time, and status), a water monitoring table (including station ID, timestamp, water level, and flow rate, categorized by different types such as reservoirs, rivers, urban flooding, and manholes), a rainfall data table (including rain gauge ID, timestamp, and rainfall), an operational data table (including facility ID, operating status, and start / stop records), a deployment status table (including deployment ID, materials, personnel, and location), and an event management table (including event ID, type, occurrence time, handling progress, and responsible person). Each table is linked by primary and foreign keys such as region ID and station ID, supporting flexible retrieval based on time range, region, and other conditions. Field types and lengths are customized according to actual business needs, and extended fields are reserved to adapt to future business adjustments.

[0074] S7, based on actual data, replaces the mask identifiers in the template item by item according to the correspondence with the mask information in the template, and generates a defense briefing.

[0075] The large language model will replace the mask identifiers in the template item by item with the structured data retrieved from the database in step S6, according to the correspondence of the mask information in the template.

[0076] As a preferred technical solution, such as Figure 8 As shown, step S7 specifically includes: S71, perform data mapping, and establish a one-to-one correspondence between the fields in the query results and the mask identifiers in the defense briefing template.

[0077] For example, the value of the "weather description" field in the query results can be mapped to the {{weather conditions}} mask position in the template.

[0078] S72 performs data formatting and aggregation, formatting the raw data according to the format requirements of each data item in the defense briefing template.

[0079] For example: unified conversion of date and time formats, such as converting "2026-06-29 10:59:00" to "2026 June 29 10:59"; control of numerical precision, such as keeping water level data to two decimal places; and aggregation of multi-source data, such as merging hydrological data from multiple reservoirs into a descriptive text according to the template requirements.

[0080] S73, perform template filling, fill the formatted data item by item into the corresponding mask position of the defense briefing template to generate a complete defense briefing.

[0081] The defense briefing function of the smart water management platform is called through the briefing generation module to process the defense briefing generated in step S7 and generate a formal defense briefing.

[0082] The generated briefing supports the following two output modes: Mode 1: Interactive Interface Output. The defense briefing is presented to the user in the form of a question-and-answer session within the large language model's interactive module. The briefing content is displayed in a structured text format or table format, and users can directly read, copy, or ask further questions about the details in the briefing within the interactive interface.

[0083] Mode 2: Generation and Push via Smart Water Management Platform. The defense briefing is simultaneously generated on the smart water management platform and displayed in a standard format (such as HTML, PDF, etc.). The system automatically pushes the generated briefing to the corresponding module on the smart water management platform for administrators to view. The push status will be provided to the user.

[0084] Both output modes can be enabled simultaneously, or one can be selected based on user instructions or system configuration.

[0085] This invention automatically pushes the defense briefing to the smart water management platform for display after it is generated, and provides feedback on the push status to the user. Compared with the existing technology that requires manual uploading or pushing of briefings to the business platform, the automatic push mechanism of this invention ensures that the briefing can reach the business system and relevant personnel in a timely manner, improving the efficiency of information flow.

[0086] This invention supports the simultaneous output of defense briefings in a question-and-answer format within a large language model interactive interface and in a formally generated and displayed format on a smart water management business platform. The interactive interface output facilitates user interaction and follow-up inquiries, while the business platform output enables standardized archiving and cross-platform sharing of the briefings. This complementary approach to output caters to the needs of different usage scenarios, enhancing the system's practicality and user-friendliness.

[0087] In summary, the AI-based defense briefing generation method constructs a complete technical chain encompassing instruction understanding, intelligent parameter acquisition, knowledge retrieval, multi-source data querying, template mask filling, and multi-terminal output. Users only need to input simple commands such as "Please generate today's defense briefing" in natural language, and the system can automatically complete the entire process from parameter acquisition and data querying to briefing generation. Compared to existing technologies that require manual data querying from multiple business systems and manual template filling, this invention automates the entire defense briefing generation process, reducing briefing generation time from hours to minutes, significantly lowering labor and time costs.

[0088] An embodiment of the present invention also provides a defense briefing auxiliary generation system based on AI intelligent agents, used to implement the defense briefing auxiliary generation method based on AI intelligent agents, including a large language model interaction module, a knowledge base retrieval module, a Reranker reordering module, a parameter missing identification and active query module, an SQL automatic generation module, a defense briefing template filling module, a briefing generation module, and a multi-terminal output module.

[0089] Users input defense briefing generation commands in text form through an interactive interface. The large language model understands the user's intent and extracts the generation parameters. If key parameters are missing (such as weather conditions or generation date), the system proactively asks the user for clarification to obtain the complete parameters. The system then invokes the defense briefing function, selecting a preset briefing template containing masking information. The template text is vectorized using an embedding model, and a vector retrieval is performed in the local water resources knowledge base. The retrieval results are then reordered using a Reranker model. The large language model then performs a final analysis based on the reordered retrieval results. The system automatically interprets the data requirements indicated by the mask information in the template based on the results and template content, and generates SQL query statements to query the multi-source data tables of the smart water management platform. The queried data includes weather conditions, early warning conditions, emergency response conditions, reservoir water conditions, river water conditions, urban flooding conditions, manhole water conditions, rainfall conditions, facility conditions, deployment conditions, urban flooding events, dangerous events, and event handling conditions. The query results are then filled into the template mask positions to generate a complete defense briefing. The briefing can be output in the form of a question and answer in the interactive interface, or it can be officially generated in the smart water management platform and automatically pushed for display.

[0090] Specifically, the large language model interaction module receives defense briefing generation instructions input by the user in natural language text form, performs semantic parsing and intent recognition, extracts the basic parameters required for briefing generation, and engages in multi-turn dialogue interaction with the user. This module embeds a large language model, possessing natural language understanding and generation capabilities.

[0091] The parameter missing identification and proactive inquiry module is used to verify the completeness of basic parameters after recognizing the user's intent to generate a defense briefing. If any required parameters are missing, the module identifies the missing parameters and sends a request to the user to supplement them until the basic parameter completeness verification passes. For example, after the user enters a command, the module automatically identifies whether the key parameters required to generate the defense briefing are complete. If any are missing, the module proactively asks the user for supplementary information.

[0092] The knowledge base retrieval module comprises a local water resources knowledge base and a vector database. The local water resources knowledge base stores unstructured knowledge such as professional documents on flood control, specifications for writing flood control briefings, historical flood control briefing cases, and flood control emergency plans. The vector database stores the vector representations of the above knowledge after being transformed by the Embedding model. The knowledge base retrieval module is used to select the corresponding flood control briefing template from the preset template library and convert the selected flood control briefing template text into a high-dimensional vector representation to obtain the template vector. The flood control briefing template contains preset mask information, which is used to identify the positions in the template that need to be replaced by actual data.

[0093] The Reranker module is used to perform similarity retrieval in the vector database of the local water conservancy knowledge base based on template vectors. It obtains candidate knowledge document fragments related to the content of the defense briefing template, scores the degree of correlation between the candidate knowledge fragments and the template content, and re-ranks the candidate knowledge fragments according to the scores. It outputs the top N most relevant knowledge fragments to improve the accuracy of the relevance between the search results and the briefing template content.

[0094] The SQ auto-generation module receives the understanding results of the large language model on the template mask information and automatically generates structured query statements (SQL) for the relational database of the smart water management platform. Specifically, it uses the top N most relevant knowledge fragments and defense briefing templates as input prompts to identify the data requirements indicated by each mask information in the template and generate structured query statements.

[0095] The defense briefing template filling module is used to submit structured query statements to the relational database of the smart water management platform for execution. It retrieves the required actual data from various relevant data tables, and based on the actual data, replaces the mask identifiers in the template item by item according to the correspondence with the mask information in the template, thereby generating a defense briefing.

[0096] The briefing generation module is used to call the defense briefing function of the smart water management platform and generate a formal defense briefing document based on the filled template; the multi-terminal output module is used to support the output of defense briefings in the form of question and answer in the big language model interactive interface, as well as to officially generate and automatically push the display in the smart water management platform; the workflow orchestration engine is used to orchestrate and schedule the collaborative workflow between the above modules to ensure that each link is executed in sequence and data is transmitted; the MCP protocol adaptation module is used to realize standardized interaction between the modules and between the system and external tools and data sources.

[0097] This invention adopts a modular architecture design, with modules interacting through standardized interfaces. When it is necessary to support new presentation template formats, add new data sources, or upgrade the large language model, only the corresponding modules need to be replaced or extended, without refactoring the entire system, thus reducing the system's maintenance and upgrade costs.

[0098] This invention automates the entire process of defense briefing generation, from instruction understanding, intelligent parameter acquisition, knowledge retrieval, multi-source data query, template mask filling to multi-terminal output. It significantly improves the efficiency and accuracy of defense briefing generation, reduces manual operation costs, and enhances the response speed of flood control emergency management.

[0099] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A method for assisting in the generation of defense briefings based on AI intelligent agents, characterized in that, Includes the following steps: S1 receives the defense briefing generation instruction input by the user, performs semantic parsing and intent recognition, and extracts the basic parameters required for briefing generation; S2, after recognizing the user's intent to generate the defense briefing, performs a completeness check on the basic parameters. If any required parameters are missing, the missing parameter items are identified and a request to the user to supplement them is sent until the completeness check of the basic parameters passes. S3, select the corresponding defense briefing template from the preset template library and convert the selected defense briefing template text into a high-dimensional vector representation to obtain the template vector. The defense briefing template contains preset mask information, which is used to identify the positions in the template that need to be replaced by actual data. S4. Based on the template vector, perform similarity retrieval in the vector database of the local water conservancy knowledge base to obtain candidate knowledge document fragments related to the content of the defense briefing template. Score the degree of correlation between the candidate knowledge fragments and the template content, and re-rank the candidate knowledge fragments according to the score. Output the top N most relevant knowledge fragments. S5 uses the top N most relevant knowledge fragments and defense briefing templates as input prompts to identify the data requirements indicated by each mask information in the template and generate structured query statements. S6 submits the structured query statement to the relational database of the smart water management platform for execution, and retrieves the required actual data from the relevant data tables; S7, based on actual data, replaces the mask identifiers in the template item by item according to the correspondence with the mask information in the template, and generates a defense briefing.

2. The defense briefing generation method based on AI intelligent agents as described in claim 1, characterized in that, Step S1 specifically includes: S11, obtains the defense briefing generation instructions input by the user in natural language text through the large language model interaction module; S12: Identify whether the defense briefing generation instruction belongs to the defense briefing generation intention. If so, extract the basic parameters required for briefing generation from the defense briefing generation instruction and jump to S2. If not, proceed to the general question and answer process or prompt the user to re-enter.

3. The defense briefing generation method based on AI intelligent agents as described in claim 2, characterized in that, Step S3 specifically includes: S31. After the parameter integrity verification is passed, the large language model calls the defense briefing function and selects the corresponding defense briefing template from the preset template library. The template library has multiple formats of defense briefing templates. Different defense briefing templates are suitable for different defense scenarios. The defense briefing template contains preset mask information, which is used to identify the positions in the template that need to be replaced by actual data. S32, using the Embedding model, the selected defense briefing template text is transformed into a high-dimensional vector representation, resulting in the template vector.

4. The defense briefing generation method based on AI intelligent agents as described in claim 3, characterized in that, Step S5 specifically includes: S51, the top N most relevant knowledge fragments and defense briefing templates are used as input prompts for the large language model, and the large language model identifies the data requirements indicated by each mask information in the template; S52, after identifying the data requirements of each mask information, generates a structured query statement from the large language model.

5. The defense briefing generation method based on AI intelligent agents as described in claim 4, characterized in that, Step S6 specifically includes: S61, submit the structured query statement to the relational database of the smart water platform for execution, query the required actual data from the relevant data tables, and return the query results in the form of structured data; S62. If a certain type of data query fails or the query result is empty, the system records the abnormal situation and fills the corresponding mask position with a preset identifier when the subsequent briefing is generated. At the same time, a data missing log is generated for management personnel to refer to and supplement the data later.

6. The defense briefing generation method based on AI intelligent agents as described in claim 5, characterized in that, Step S7 specifically includes: S71, perform data mapping, and establish a one-to-one correspondence between the fields in the query results and the mask identifiers in the defense briefing template; S72 performs data formatting and aggregation, formatting the raw data according to the format requirements of each data item in the defense briefing template; S73, perform template filling, fill the formatted data item by item into the corresponding mask position of the defense briefing template to generate a complete defense briefing.

7. The defense briefing generation method based on AI intelligent agents as described in claim 6, characterized in that, Also includes: S8. The defense briefing function of the smart water management platform is called through the briefing generation module to process the defense briefing generated in step S7 and generate a formal defense briefing.

8. The defense briefing generation method based on AI intelligent agents as described in claim 7, characterized in that, Step S12 specifically includes: S121, a pre-set business keyword library associated with the defense briefing is configured, and word segmentation is performed on the defense briefing generation instructions; S122, if the defense briefing generation instruction after word segmentation hits any keyword, it is determined to be a defense briefing generation intention; otherwise, the defense briefing generation instruction is converted into a vector using the Embedding model and cosine similarity is calculated with the preset defense briefing generation intention standard template vector. S123, if the similarity exceeds the preset threshold, it is determined to be an intention to generate a defense briefing; otherwise, it proceeds to the general Q&A process or prompts the user to re-enter the information.

9. The defense briefing generation method based on AI intelligent agents as described in claim 8, characterized in that, Step S4 specifically includes: S41, the candidate knowledge fragments and the defense briefing template are encoded independently, the semantic vectors of the fragments and the semantic vectors of the template are obtained, and the cosine similarity between the two is calculated to obtain the semantic similarity score. S42, extract business keywords from candidate knowledge fragments and defense briefing templates respectively, obtain fragment keyword sets and template keyword sets, calculate the Jaccard coefficient of the two, and obtain the business keyword overlap. S43, extract water conservancy entities from candidate knowledge fragments and defense briefing templates respectively, obtain fragment water conservancy entity set and template water conservancy entity set, and obtain water conservancy entity alignment degree based on fragment water conservancy entity set and template water conservancy entity set; S44. The semantic similarity score, the overlap of business keywords, and the alignment of water conservancy entities are weighted and fused to obtain the relevance score.

10. A defense briefing assistance generation system based on AI agents, used to implement the defense briefing assistance generation method based on AI agents as described in any one of claims 1-9, characterized in that, include: The large language model interaction module is used to receive user input instructions for generating defense briefings, perform semantic parsing and intent recognition, and extract the basic parameters required for briefing generation. The parameter missing identification and proactive inquiry module is used to perform integrity verification on the basic parameters after identifying the user's intent to generate the defense briefing. If any required parameters are missing, the module identifies the missing parameter items and sends a supplementary request to the user until the integrity verification of the basic parameters passes. The knowledge base retrieval module is used to select the corresponding defense briefing template from the preset template library and convert the selected defense briefing template text into a high-dimensional vector representation to obtain the template vector. The defense briefing template contains preset mask information, which is used to identify the positions in the template that need to be replaced by actual data. The Reranker module is used to perform similarity retrieval in the vector database of the local water conservancy knowledge base based on template vectors, obtain candidate knowledge document fragments related to the content of the defense briefing template, score the degree of correlation between the candidate knowledge fragments and the template content, re-rank the candidate knowledge fragments according to the score, and output the top N most relevant knowledge fragments. The SQ auto-generation module is used to identify the data requirements indicated by each mask information in the template by taking the top N most relevant knowledge fragments and defense briefing templates as input prompts and generating structured query statements. The defense briefing template filling module is used to submit structured query statements to the relational database of the smart water management platform for execution. It retrieves the required actual data from various relevant data tables, and based on the actual data, replaces the mask identifiers in the template item by item according to the correspondence with the mask information in the template, thereby generating a defense briefing.

Images