Large-model-driven emergency disposal process generation method and system

Abstract

The invention discloses a method and a system for generating an emergency disposal process driven by a large model. The method comprises the following steps of: performing data preprocessing on an emergency disposal case; using a father-child partitioning method to perform document partitioning on the pre-processed emergency disposal case, and constructing a knowledge base and a vector base; carrying out keyword and vector retrieval based on a BM25 sparse retrieval method and a BGE dense retrieval method, and carrying out weighted fusion on retrieval results to obtain a mixed retrieval result; building an emergency disposal process to generate cue words, and inputting the emergency description text, the mixed retrieval result and the cue words into a large language model to generate a disposal process conforming to a preset specification; verifying the disposal process, and storing the verified disposal process as a new emergency disposal case in the knowledge base and the vector base; according to the invention, a clear and operable standardized action guide can be provided for emergency handling, and the response accuracy and efficiency are effectively improved.

Description

Technical Field

[0001] This invention belongs to the field of information processing technology, and in particular relates to a method and system for generating emergency response procedures driven by a large model. Background Technology

[0002] In the field of emergency management, emergency response is a core issue of concern to the government and society. Currently, the industry generally relies on a contingency plan development model based on historical experience and expert knowledge. This model involves extracting response procedures through manual analysis and expert review, and using process modeling languages ​​(such as BPMN) to standardize the descriptions of tasks, participants, and task sequences in historical cases.

[0003] However, existing technologies have significant shortcomings: most contingency plans only provide general principles, organizational structures, and operational mechanisms, lacking detailed descriptions of the participants, specific tasks, and inter-task relationships in the handling of specific events, thus failing to provide accurate and actionable guidance for handling emergencies. Summary of the Invention

[0004] Purpose of the invention: The purpose of this invention is to provide a large-model-driven method for generating emergency response procedures to achieve precise guidance for emergency response; on the other hand, it provides a large-model-driven system for generating emergency response procedures.

[0005] Technical solution: The emergency response process generation method of the present invention includes: Data preprocessing is performed on emergency response cases, including standardizing document naming and normalizing document content. The parent-child block method is used to divide the pre-processed emergency response cases into document blocks, build a knowledge base, and use the BGE embedding model to build the vector library corresponding to the knowledge base. Keyword retrieval is performed from the knowledge base based on the BM25 sparse retrieval method, and vector retrieval is performed from the vector library based on the BGE dense retrieval method. The results of the keyword retrieval and vector retrieval are weighted and fused to obtain a hybrid retrieval result. Construct emergency response procedures and generate prompt words. Input the emergency description text, the mixed search results and prompt words into the big language model to generate a response procedure that conforms to the preset specifications. The handling process is verified, and the verified handling process is stored as a new emergency handling case in the knowledge base and vector library.

[0006] This invention significantly improves the structuring level and information consistency of historical case handling by uniformly naming and normalizing case data, laying a solid foundation for subsequent refined retrieval. It constructs a knowledge base using a parent-child block method and generates a vector library using a BGE model, maintaining the logical integrity of document content and achieving deep semantic representation of handling tasks, executing entities, and temporal relationships between tasks. By integrating keyword and semantic search results, it accurately captures key participants and specific task nodes in the handling process and comprehensively associates handling logic in similar scenarios, ensuring precise matching of search results to the current event. The handling process generated based on the large language model can deeply integrate the retrieved refined case information, outputting an actionable solution that clearly defines the responsible entity, specific task content, and the flow of tasks. Finally, through a verification mechanism and dynamic expansion of the case library, the system can continuously optimize its ability to guide the handling of different types of emergencies, ultimately providing emergency command personnel with precise, actionable process guidance down to the execution level.

[0007] Preferably, the data preprocessing includes: Documents for emergency response cases should be named in a standardized manner according to the format of "Event Name-Event Type-Event Level"; Extract the event name, occurrence time, event type, event level, location, event details, and handling process from emergency response cases, and store them in a structured format, with the handling process represented in BPMN specification.

[0008] By adopting a unified naming convention of "event name-event type-event level," the standardization and retrieval efficiency of case management have been significantly improved, enabling emergency personnel to quickly locate relevant cases. By extracting key event elements and storing them in a structured format, the systematic organization and efficient utilization of case information have been achieved. In particular, the handling process is represented using the BPMN standard, ensuring the machine readability and clarity of task nodes, execution roles, temporal logic, and branch conditions in the process. This provides a standardized foundation for the automatic parsing, accurate matching, and generation of subsequent processes, thereby comprehensively improving the collaborative efficiency and operational feasibility of emergency response.

[0009] Preferably, the parent-child block partitioning method includes: Use the entire document of emergency response case studies as the parent section; Each stage of the key-value pair information and processing flow in the case document is treated as a sub-block; Embed parent block identifier information in sub-blocks to establish association relationships.

[0010] By employing a parent-child block approach, the complete case document is used as the parent block to preserve the global context, while key-value pairs and each stage of the handling process are used as child blocks for refined processing. The parent block identifier is embedded in the child blocks to establish clear relationships. This method maintains the overall semantic integrity of the case document while enabling precise location and independent retrieval of specific handling details. It significantly improves the structuring level of the knowledge base and the efficiency of information retrieval, allowing the system to simultaneously meet the query needs for both the macro framework of the case and the micro operational details. This provides multi-granular knowledge support for generating accurate and operable emergency response procedures.

[0011] Preferably, the keyword retrieval from the knowledge base based on the BM25 sparse retrieval method includes: Extract keywords from the description text of the emergency, wherein the keywords include one or more of the following: event name, event type, and event level; Based on the keywords, a matching search is performed in the knowledge base, and a document relevance score is calculated; The search results are returned based on the document relevance score, which returns the top-K relevant emergency response cases with the highest scores, where K is a preset positive integer.

[0012] By accurately extracting core keywords such as event name, type, and level from the description text of emergencies, and performing BM25 matching retrieval and relevance scoring in the knowledge base based on these keywords, the system can quickly locate historical handling cases that highly match the characteristics of the current event. Based on the relevance score, it returns the top-K most relevant results. This method effectively ensures that the search results and the current emergency are accurately matched in key attributes, providing highly relevant and reliable case references for subsequent process generation, and significantly improving the accuracy and pertinence of emergency response decisions.

[0013] Preferably, the vector retrieval from the vector database based on the BGE dense retrieval method includes: The description text of the emergency event is encoded into a query vector; Calculate the cosine similarity between the query vector and each sub-block vector in the vector library to obtain the similarity score of each sub-block; Based on the relationship between sub-blocks and parent blocks, calculate the average similarity score of all related sub-blocks under the same parent block document, and use this average score as the overall score of the parent block document; Based on the overall score of the parent block documents, the top-N parent block documents with the highest scores are returned as search results, where N is a preset positive integer.

[0014] By encoding the description text of an emergency into a query vector and calculating its semantic similarity to sub-blocks in the vector library, BGE dense retrieval can deeply capture the semantic association between the current event and historical cases. By aggregating sub-block scores into an overall score by parent document, this method takes into account both the semantic matching accuracy of details of the handling process and the document-level completeness of the returned results. Finally, it returns the top-N most relevant complete cases, so that the retrieval results can provide a detailed reference for specific task links and maintain the contextual coherence of the complete handling process, which significantly improves the comprehensiveness and practicality of semantic retrieval and provides rich semantic knowledge support for generating a rigorous and highly operable emergency handling process.

[0015] Preferably, the formula for calculating the cosine similarity is:

[0016] Where a represents a vector of the text of the emergency, and b represents any vector in the vector library.

[0017] By using cosine similarity to calculate the cosine of the angle between the query vector and the case vector, this method can effectively measure the strength of semantic association between texts. Its vector direction consistency measurement method can ignore the absolute numerical difference and focus on the relative distribution relationship in the semantic space. This calculation method not only ensures the accuracy of semantic similarity judgment, but also improves the adaptability to texts of different lengths and expressions, providing a reliable mathematical basis for accurately retrieving semantically related handling cases, thereby significantly enhancing the system's semantic understanding ability and case matching accuracy for diverse emergency event descriptions.

[0018] Preferably, the weighted fusion of the results of keyword retrieval and vector retrieval includes: The final score is calculated using the following weighted formula:

[0019] in, The score of emergency case documents is calculated in a pre-defined database using BM25 sparse retrieval. The score of emergency event case documents is calculated from a pre-defined vector library through BGE dense retrieval. The weights are those of the weighted average. This is the final score.

[0020] By employing a weighted fusion formula to organically combine the precise keyword matching score of BM25 sparse retrieval with the deep semantic similarity score of BGE dense retrieval, this method fully leverages the complementary advantages of the two retrieval technologies: it retains the precise ability of keyword matching to capture core event elements (such as type and level) while incorporating the deep understanding of contextual association and task logic by semantic retrieval; through adjustable weight parameters, the system can flexibly balance literal matching accuracy and semantic association strength, thereby significantly improving the comprehensive relevance, coverage, and practicality of the retrieval results, providing more reliable multi-dimensional knowledge support for generating comprehensive and accurate emergency response procedures.

[0021] Secondly, the emergency response process generation system of the present invention includes: The data preprocessing module is used to preprocess data from emergency response cases. The data preprocessing includes standardizing document naming and normalizing the document content. The knowledge base construction module is used to divide the pre-processed emergency response cases into document blocks using the parent-child block method, build a knowledge base, and use the BGE embedding model to build the vector library corresponding to the knowledge base. The hybrid retrieval module is used to perform keyword retrieval from the knowledge base based on the BM25 sparse retrieval method and vector retrieval from the vector library based on the BGE dense retrieval method, and to perform weighted fusion of the results of the keyword retrieval and vector retrieval to obtain hybrid retrieval results; The process generation module is used to generate prompts for emergency response processes. It inputs the emergency description text, the mixed search results and prompts into the large language model to generate a response process that conforms to preset specifications. The verification and update module is used to verify the handling process and store the verified handling process as a new emergency handling case in the knowledge base and vector library.

[0022] Thirdly, the present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and executed as described in the large model-driven method for generating emergency response procedures.

[0023] Fourthly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the large model-driven emergency response process generation method.

[0024] Beneficial Effects: Compared with existing technologies, this invention has the following significant advantages: 1. By retrieving and integrating the handling processes of similar historical cases (represented by the BPMN specification), it can generate a refined process that includes specific participating departments, clear task nodes, and logical flow relationships between tasks, providing a clear and operable standardized action guide for emergency response, effectively improving the accuracy and efficiency of the response; 2. Combining the advantages of BM25 keyword retrieval and BGE vector retrieval, it can accurately match key meta-information of cases (such as type and level) through keywords, and capture deep contextual information in descriptive text through semantic understanding. The weighted fusion of the two ensures accurate retrieval results. 1. The results are comprehensive and relevant, providing a high-quality reference for process generation; 2. By requiring the large language model to generate handling processes based on preset prompts and BPMN specifications, the output results are no longer free text, but a standardized process description that is clearly structured, machine-readable, and conforms to industry standards, facilitating subsequent system integration, automated execution, and unified management; 3. The system has closed-loop learning capabilities. By using newly generated handling processes after verification as cases to feed back into the knowledge base and vector library, the knowledge base can be continuously expanded and updated. Thus, when facing new or evolving emergencies, the retrieval and generation capabilities are continuously enhanced, reflecting the system's adaptability and intelligence. Attached Figure Description

[0025] Figure 1 This is a simplified flowchart of the method of the present invention; Figure 2 This is a detailed flowchart illustrating the method of the present invention; Figure 3 This is a flowchart of the emergency response process based on the BPMN specification of the present invention. Detailed Implementation

[0026] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0027] This invention provides a large-model-driven method for generating emergency response procedures. Based on existing historical emergency response procedures, it addresses the problem of rapidly generating standardized emergency response procedures for new emergencies by constructing a knowledge base using historical emergency response procedures. Figure 1 , Figure 2 and Figure 3 As shown, the method includes the following steps: Step 1: Perform data preprocessing on emergency response cases and name the historical emergency response case documents uniformly according to the format of "event name-event type-event level".

[0028] The case documents of emergency response are normalized and stored in JSON format. The event name, event time, event type, event level, event location, event details, handling status, and handling process information are extracted from the case documents. The handling process information is represented in the standard form of the BPMN specification.

[0029] Step 2: After data preprocessing, the case study documents for handling emergencies are divided into blocks using a parent-child block method. In this embodiment, a complete case study document is considered a parent block, and each key-value pair within the document is considered a child block. Similarly, for the handling process data within the case study document, each stage of the process is considered a child block. Parent and child blocks are linked by adding the parent block's identifier to the child blocks. The child blocks are used for dense retrieval in the vector database, and the parent block associated with each child block document is returned as a document, providing complete semantic context information. The case study document block data is stored in an Elasticsearch database as a knowledge base, facilitating subsequent sparse retrieval.

[0030] Using the BGE embedding model, all sub-chunk documents are vectorized, and the sub-chunk document vectors and their corresponding parent chunk document IDs are simultaneously stored in the vector database. In this embodiment, the BGE-base-zh-v1.5 embedding model is selected to generate 768-dimensional semantic vectors, and the vector database chosen is ChromaDB, which supports high-concurrency approximate nearest neighbor retrieval.

[0031] Step 3: Use the BM25 sparse search method to perform keyword retrieval on the knowledge base. In this embodiment, ElasticSearch is used to implement BM25 sparse search, extracting professional keywords such as event name, event type, and event level from the description text of the emergency. Keyword matching is performed in ElasticSearch to obtain the top-10 relevant emergency handling cases with the highest document relevance scores.

[0032] Vector retrieval is performed in the vector library using the BGE dense retrieval method. In this embodiment, the BGE-base-zh-v1.5 embedding model is used to vectorize the text describing the emergency, generating a 768-dimensional semantic vector. The cosine similarity of this vector with all vectors in the vector library is calculated, and the cosine similarity is used as the score for each sub-block document. Based on the parent block document ID associated with each sub-block document, all sub-block documents associated with each parent block document are obtained. The average cosine similarity of all sub-block documents associated with each parent block document is calculated as the score of the parent block document, and the top-10 parent block documents with the highest scores are obtained, which are the emergency response cases.

[0033] The formula for calculating the cosine similarity between two vectors is shown below, where a represents the vector of the sudden event text, and b represents any vector in the vector library:

[0034] Furthermore, a weighted average is calculated for the BM25 sparse search and the BGE dense search to obtain the final score. The weighted average formula is shown below, where... The score of incident case documents is calculated in the ElasticSearch database using BM25 sparse retrieval. The score of emergency case documents is calculated from the ChromaDB vector library through BGE dense retrieval. This is the weight of the weighted average, which is set to 0.5 in this embodiment. This is the final score.

[0035]

[0036] In this embodiment, the top-5 emergency response plans with the highest final scores are used as the final mixed search results.

[0037] Step 4: Construct an emergency response process and generate prompts. In this embodiment, a sample prompt is provided: ### Role As an expert in designing emergency response procedures, you need to integrate historical emergency response cases (referencing national / local emergency plans) with the characteristics of new types of events to generate standardized and executable BPMN procedures.

[0038] ### Task Objectives Output the flowchart of the BPMN specification, including the flowchart of MBPN specification elements such as the process definition root element, process start event, process end event, activities, sequence flow, and gateway.

[0039] ### Input Data { "Description of the emergency": { Event Name: A 5.2-magnitude earthquake struck Kunming, Yunnan Province on July 5. "Date of Occurrence": "July 5, 2025", Location of the incident: Guandu District, Kunming City, Yunnan Province Event Type: Earthquake Event Level: "Normal" Event Details: On July 5, 2025, a 5.2-magnitude earthquake struck Guandu District, Kunming City, Yunnan Province. The epicenter was located in Puzi Community, Yiliu Subdistrict, with a focal depth of 10 kilometers and an intensity of 5.2. }, "Mixed search results": [ { Event Name: A 5.2-magnitude earthquake struck Kunming, Yunnan Province on July 5. "Date of Occurrence": "July 5, 2025", Location of the incident: Guandu District, Kunming City, Yunnan Province Event Type: Earthquake Event Level: "Normal" Event Details: On July 5, 2025, a 5.2-magnitude earthquake struck Guandu District, Kunming City, Yunnan Province. The epicenter was located in Puzi Community, Yiliu Subdistrict, with a focal depth of 10 kilometers and an intensity of 5.2. "Handling Process":" <process id="earthquake_emergency_handle" name="地震事件处置流程样例"> <!--1. Start event to receive earthquake emergency alerts--> <startevent id="startEvent" name="地震告警"> <messageeventdefinition messageRef="EarthquakeAlertMsg" / > < / startevent> <!--2. Event assessment task--> <sequenceflow id="flow1" sourceref="startEvent" targetref="assessTask"> <servicetask id="assessTask" name="事件评估"> <documentation> Analyze the types of events and their scope of impact.< / documentation> < / servicetask> < / sequenceflow> <!--3. Resource scheduling task--> <sequenceflow id="flow2" sourceref="assessTask" targetref="dispatchTask"> <servicetask id="dispatchTask" name="资源调度"> <documentation> Deploying emergency personnel and supplies< / documentation> < / servicetask> < / sequenceflow> <!--4. Disposal execution task--> <sequenceflow id="flow3" sourceref="dispatchTask" targetref="executeTask"> <servicetask id="executeTask" name="处置执行"> <documentation> Implement emergency response measures< / documentation> < / servicetask> < / sequenceflow> <!--5. End event--> <sequenceflow id="flow4" sourceRef="executeTask" targetRef="endEvent" / > <endevent id="endEvent" name="处置完成"> <messgae id="EarthquakeAlertMsg" name="地震突发事件告警消息" / > < / endevent> < / process> " } ... / / Four search results omitted ] } ### Output Data Output XML format text conforming to the BPMN specification

[0040] The emergency description text, the historical emergency handling cases obtained by the mixed retrieval based on the emergency description text in step 3, and the big model prompt words are passed to the big model. In this embodiment, the DeepSeek-R1 big model is selected, and the big model generates process information represented in accordance with the BPMN specification.

[0041] Step 5: Combine the description text of the emergency with the process information generated by the large model in Step 4, make manual fine-tuning, and store it as a new emergency response case in the knowledge base and vector library to complete the dynamic update of the data.

[0042] Based on a similar inventive concept, this invention also provides an emergency response process generation system corresponding to the aforementioned emergency response process generation method, comprising: The data preprocessing module is used to preprocess data from emergency response cases. The data preprocessing includes standardizing document naming and normalizing the document content. The knowledge base construction module is used to divide the pre-processed emergency response cases into document blocks using the parent-child block method, build a knowledge base, and use the BGE embedding model to build the vector library corresponding to the knowledge base. The hybrid retrieval module is used to perform keyword retrieval from the knowledge base based on the BM25 sparse retrieval method and vector retrieval from the vector library based on the BGE dense retrieval method, and to perform weighted fusion of the results of the keyword retrieval and vector retrieval to obtain hybrid retrieval results; The process generation module is used to generate prompts for emergency response processes. It inputs the emergency description text, the mixed search results and prompts into the large language model to generate a response process that conforms to preset specifications. The verification and update module is used to verify the handling process and store the verified handling process as a new emergency handling case in the knowledge base and vector library.

[0043] The present invention also discloses an electronic device.

[0044] Specifically, the electronic device can be a desktop computer, laptop computer, handheld computer, or cloud server, etc. This computer device may include, but is not limited to, a processor and memory. The processor and memory can be connected via a bus or other means. The processor can be a Central Processing Unit (CPU). The processor can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) or other programmable logic devices, graphics processing units (GPUs), embedded neural network processing units (NPUs) or other dedicated deep learning coprocessors, discrete gate or transistor logic devices, discrete hardware components, or combinations of the above types of chips.

[0045] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer-executable programs, and modules. The processor executes various functional applications and data processing by running non-transitory software programs, instructions, and modules stored in memory. Memory may include a program storage area and a data storage area. The program storage area may store the control unit and the application program required for at least one function; the data storage area may store data created by the processor, etc. Furthermore, memory may include high-speed random access memory and non-transitory memory. In some embodiments, memory may optionally include memory remotely located relative to the processor, which can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0046] The present invention also discloses a computer-readable storage medium.

[0047] Specifically, the computer-readable storage medium is used to store a computer program, which, when executed by a processor, implements the methods described in the above method implementation.

[0048] Those skilled in the art will understand that all or part of the processes in the methods described above can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk drive (HDD), or solid-state drive (SSD), etc.; the storage medium can also include combinations of the above types of memory.

Claims

1. A method for generating emergency response procedures driven by a large model, characterized in that, include: Data preprocessing is performed on emergency response cases, including standardizing document naming and normalizing document content. The parent-child block method is used to divide the pre-processed emergency response cases into document blocks, build a knowledge base, and use the BGE embedding model to build the vector library corresponding to the knowledge base. Keyword retrieval is performed from the knowledge base based on the BM25 sparse retrieval method, and vector retrieval is performed from the vector library based on the BGE dense retrieval method. The results of the keyword retrieval and vector retrieval are weighted and fused to obtain a hybrid retrieval result. Construct emergency response procedures and generate prompt words. Input the emergency description text, the mixed search results and prompt words into the big language model to generate a response procedure that conforms to the preset specifications. The handling process is verified, and the verified handling process is stored as a new emergency handling case in the knowledge base and vector library.

2. The method for generating emergency response procedures according to claim 1, characterized in that, The data preprocessing includes: Documents for emergency response cases should be named in a standardized manner according to the format of "Event Name-Event Type-Event Level"; Extract the event name, occurrence time, event type, event level, location, event details, and handling process from emergency response cases, and store them in a structured format, with the handling process represented in BPMN specification.

3. The method for generating emergency response procedures according to claim 1, characterized in that, The parent-child block partitioning method includes: Use the entire document of emergency response case studies as the parent section; Each stage of the key-value pair information and processing flow in the case document is treated as a sub-block; Embed parent block identifier information in sub-blocks to establish association relationships.

4. The method for generating emergency response procedures according to claim 1, characterized in that, The BM25-based sparse retrieval method for keyword retrieval from the knowledge base includes: Extract keywords from the description text of the emergency, wherein the keywords include one or more of the following: event name, event type, and event level; Based on the keywords, a matching search is performed in the knowledge base, and a document relevance score is calculated; The search results are returned based on the document relevance score, which returns the top-K relevant emergency response cases with the highest scores, where K is a preset positive integer.

5. The method for generating emergency response procedures according to claim 1, characterized in that, The BGE-based dense retrieval method for vector retrieval from the vector database includes: The description text of the emergency event is encoded into a query vector; Calculate the cosine similarity between the query vector and each sub-block vector in the vector library to obtain the similarity score of each sub-block; Based on the relationship between sub-blocks and parent blocks, calculate the average similarity score of all related sub-blocks under the same parent block document, and use this average score as the overall score of the parent block document; Based on the overall score of the parent block documents, the top-N parent block documents with the highest scores are returned as search results, where N is a preset positive integer.

6. The method for generating emergency response procedures according to claim 5, characterized in that, The formula for calculating the cosine similarity is: Where a represents a vector of the text of the emergency, and b represents any vector in the vector library.

7. The method for generating emergency response procedures according to claim 1, characterized in that, The weighted fusion of the results of keyword retrieval and vector retrieval includes: The final score is calculated using the following weighted formula: ;in, The score of emergency case documents is calculated in a pre-defined database using BM25 sparse retrieval. The score of emergency event case documents is calculated from a pre-defined vector library through BGE dense retrieval. The weights are those of the weighted average. This is the final score.

8. A large-model-driven emergency response process generation system, characterized in that, include: The data preprocessing module is used to preprocess data from emergency response cases. The data preprocessing includes standardizing document naming and normalizing the document content. The knowledge base construction module is used to divide the pre-processed emergency response cases into document blocks using the parent-child block method, build a knowledge base, and use the BGE embedding model to build the vector library corresponding to the knowledge base. The hybrid retrieval module is used to perform keyword retrieval from the knowledge base based on the BM25 sparse retrieval method and vector retrieval from the vector library based on the BGE dense retrieval method, and to perform weighted fusion of the results of the keyword retrieval and vector retrieval to obtain hybrid retrieval results; The process generation module is used to generate prompts for emergency response processes. It inputs the emergency description text, the mixed search results and prompts into the large language model to generate a response process that conforms to preset specifications. The verification and update module is used to verify the handling process and store the verified handling process as a new emergency handling case in the knowledge base and vector library.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the large model-driven emergency response process generation method according to any one of claims 1 to 7.

10. An electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method for generating a large-model-driven emergency response process according to any one of claims 1 to 7.

Images