Intelligent early warning method and device for urban extreme weather disasters based on digital-physical fusion

Abstract

The application provides a kind of city extreme weather disaster intelligent early warning method and device based on data and object fusion, it is related to meteorological disaster simulation and artificial intelligence decision-making technical field.The method comprises: simulating the wind disaster and flood of target city to obtain numerical simulation feature vector;The historical plan text is processed to obtain the plan text vector;A vector database containing the extreme disaster of the target city is constructed;Design retrieval enhancement to generate agent, disaster assessment agent, report abstract agent and meteorological disaster dialogue agent, and build multi-agent damage decision-making large language model according to the agent scheduling framework;Through the combination of vector database and multi-agent damage decision-making large language model, a customized damage decision-making scheme is generated for a specific disaster scenario.The application solves the problem of meteorological forecast information to accurate decision-making, realizes the change from single meteorological parameter forecast to scenario-based damage decision support, and significantly improves the intelligent level of city disaster warning.

Description

Technical Field

[0001] This invention relates to the field of meteorological disaster simulation and artificial intelligence decision-making technology, and in particular to an intelligent early warning method and device for urban extreme meteorological disasters based on data-physical fusion. Background Technology

[0002] With the increasing frequency of extreme weather events due to global climate change, complex disasters such as windstorms and floods in urban areas pose serious threats to infrastructure, public safety, and economic operations. Existing meteorological early warning and disaster management systems generally suffer from problems such as coarse warning resolution, low accuracy, and weak intelligence, making it difficult to meet the needs of refined urban governance and emergency decision-making. Traditional disaster simulation and early warning methods are mostly based on meteorological numerical simulations of single physical mechanisms, which are insufficient to comprehensively reflect the coupling relationships between multi-source data. Simulation results obtained through physical simulations are often too specialized, making it difficult for non-professional emergency personnel to directly obtain emergency rescue information from numerical distribution maps. While traditional large-scale models have strong semantic understanding, they lack awareness of specific geographic spatial physical constraints, making them prone to misinterpretation. Therefore, constructing a city-level intelligent disaster early warning technology system that integrates risk quantification assessment, numerical simulation, and artificial intelligence has significant scientific and application value. However, currently, there is a lack of early warning and decision-making schemes that combine physical simulation and artificial intelligence generation. Therefore, developing a city-level intelligent disaster early warning system that integrates risk quantification assessment, numerical simulation feedback, and large-scale model logical reasoning to achieve a closed-loop method from physical parameter derivation to cognitive decision generation has significant scientific value and practical application significance. Summary of the Invention

[0003] To address the current technical problem of lacking a collaborative approach between numerical simulation and large-scale models for generating disaster mitigation strategies for specific locations, this invention provides a method and device for intelligent early warning of urban extreme weather disasters based on data-physical fusion. The technical solution is as follows: On the one hand, a smart early warning method for urban extreme weather disasters based on data-physical fusion is provided. This method is implemented by a smart early warning device for urban extreme weather disasters based on data-physical fusion, and includes: S1. Simulate and extract features of wind disasters and floods in the target city to obtain numerical simulation feature vectors; segment, clean, slice and transform the historical emergency plan text of the target city to obtain emergency plan text vectors; construct a vector database containing extreme disasters in the target city based on the numerical simulation feature vectors and emergency plan text vectors.

[0004] S2. Design a retrieval enhancement generation agent, a disaster assessment agent, a report summary agent, and a meteorological disaster dialogue agent, and construct a multi-agent loss reduction decision-making big language model based on the agent scheduling framework.

[0005] Among them, the retrieval enhancement generative agent is used to generate the input context of the large language model based on user instructions and vector database; the disaster assessment agent is designed to obtain the disaster risk score based on the input context of the large language model; the report summary agent is used to extract key information from the disaster risk assessment report; and the meteorological disaster dialogue agent is used to reason and respond to user instructions.

[0006] S3. By combining a vector database with a multi-agent loss reduction decision-making language model, customized loss reduction decision schemes are generated for specific disaster scenarios.

[0007] Optionally, in S1, the windstorms and floods in the target city are simulated and features are extracted to obtain a numerical simulation feature vector, including: The wind disaster in the target city was simulated using a semi-implicit method based on computational fluid dynamics, pressure coupling equations, and a turbulence model, and the numerical simulation results of the wind disaster were obtained.

[0008] The flooding of the target city was simulated based on the Mohid-Land model and the finite volume method, and the numerical simulation results of the flooding were obtained.

[0009] Feature extraction was performed on the numerical simulation results of wind disaster and flood to obtain numerical simulation feature vectors.

[0010] Optionally, in S1, the historical contingency plan text of the target city is segmented, cleaned, sliced, and transformed to obtain a contingency plan text vector, including: The historical contingency plan texts are obtained, segmented, and cleaned to obtain cleaned data. The historical contingency plan texts include historical meteorological data, disaster news, and emergency plans. The cleaned data is sliced ​​to obtain multiple text blocks.

[0011] Each text block is transformed by a pre-trained large language model, and the vector embedding corresponding to each text block is used as the pre-planned text vector.

[0012] Optionally, the input context for the large language model is generated based on user instructions and a vector database, including: The enhanced generative agent transforms user commands into query vectors by retrieving data.

[0013] Calculate the cosine similarity score between the query vector and the numerical simulation feature vector, and obtain the semantically matched numerical simulation feature vector based on the cosine similarity score.

[0014] Calculate the cosine similarity score between the query vector and the proposed text vector, and obtain the semantically matched proposed text vector based on the cosine similarity score.

[0015] The numerical simulation feature vectors of semantic matching and the pre-planned text vectors are fused to obtain a fused feature vector, which serves as the input context for the large language model.

[0016] Optionally, a disaster risk score is obtained based on the input context of the large language model, including: The disaster assessment agent receives the input context from the large language model, constructs prompt words through the thinking chain technology, inputs the input context and prompt words of the large language model into the pre-trained large language model, obtains the disaster risk score, and generates a customized loss reduction decision scheme.

[0017] Optionally, inference responses to user commands include: The meteorological disaster dialogue agent parses the user's instructions to determine whether to invoke the retrieval enhancement generation agent. If so, the retrieval enhancement generation agent generates the input context of the large language model, and the agent performs inference and response based on the input context of the large language model. If not, the agent performs inference and response based on the user's instructions.

[0018] Optionally, S3 includes: The input context and prompt words of the large language model are fed into the pre-trained large language model to obtain a disaster risk score and generate a customized loss reduction decision scheme.

[0019] The loss reduction decision-making schemes are divided into three categories: departments, enterprises, and the public, and a disaster risk assessment report is output, which includes real-time early warning, evacuation advice, and resource allocation.

[0020] The disaster risk score is dynamically bound to the geographic information system coordinates in the numerical simulation results of wind disaster and flood to obtain the geographic information system coordinates of the affected area. The geographic information system coordinates of the affected area are then spatially overlaid with the pre-stored urban infrastructure database to obtain the analysis results.

[0021] On the other hand, a smart early warning device for urban extreme weather disasters based on data-physical fusion is provided. This device is applied to a smart early warning method for urban extreme weather disasters based on data-physical fusion. The device includes: The vector database construction module is used to simulate and extract features of wind disasters and floods in the target city to obtain numerical simulation feature vectors; to process the historical emergency plan text of the target city through segmentation, text cleaning, slicing and transformation to obtain emergency plan text vectors; and to construct a vector database containing extreme disasters in the target city based on the numerical simulation feature vectors and the emergency plan text vectors.

[0022] The module for building a large multi-agent loss reduction decision model is used to design a retrieval enhancement generation agent, a disaster assessment agent, a report summary agent, and a meteorological disaster dialogue agent, and to build a large multi-agent loss reduction decision language model based on the agent scheduling framework.

[0023] Among them, the retrieval enhancement generative agent is used to generate the input context of the large language model based on user instructions and vector database; the disaster assessment agent is designed to obtain the disaster risk score based on the input context of the large language model; the report summary agent is used to extract key information from the disaster risk assessment report; and the meteorological disaster dialogue agent is used to reason and respond to user instructions.

[0024] The data-physical fusion solution generation module is used to generate customized loss reduction decision-making solutions for specific disaster scenarios by combining a vector database with a multi-agent loss reduction decision-making big language model.

[0025] Optionally, the vector database building module is further used for: The wind disaster in the target city was simulated using a semi-implicit method based on computational fluid dynamics, pressure coupling equations, and a turbulence model, and the numerical simulation results of the wind disaster were obtained.

[0026] The flooding of the target city was simulated based on the Mohid-Land model and the finite volume method, and the numerical simulation results of the flooding were obtained.

[0027] Feature extraction was performed on the numerical simulation results of wind disaster and flood to obtain numerical simulation feature vectors.

[0028] Optionally, the vector database building module is further used for: The historical contingency plan texts are obtained, segmented, and cleaned to obtain cleaned data. The historical contingency plan texts include historical meteorological data, disaster news, and emergency plans. The cleaned data is sliced ​​to obtain multiple text blocks.

[0029] Each text block is transformed by a pre-trained large language model, and the vector embedding corresponding to each text block is used as the pre-planned text vector.

[0030] Optionally, the input context for the large language model is generated based on user instructions and a vector database, including: The enhanced generative agent transforms user commands into query vectors by retrieving data.

[0031] Calculate the cosine similarity score between the query vector and the numerical simulation feature vector, and obtain the semantically matched numerical simulation feature vector based on the cosine similarity score.

[0032] Calculate the cosine similarity score between the query vector and the proposed text vector, and obtain the semantically matched proposed text vector based on the cosine similarity score.

[0033] The numerical simulation feature vectors of semantic matching and the pre-planned text vectors are fused to obtain a fused feature vector, which serves as the input context for the large language model.

[0034] Optionally, the multi-agent loss reduction decision-making large model construction module can be further used for: The disaster assessment agent receives the input context from the large language model, constructs prompt words through the thinking chain technology, inputs the input context and prompt words of the large language model into the pre-trained large language model, obtains the disaster risk score, and generates a customized loss reduction decision scheme.

[0035] Optionally, the multi-agent loss reduction decision-making large model construction module can be further used for: The meteorological disaster dialogue agent parses the user's instructions to determine whether to invoke the retrieval enhancement generation agent. If so, the retrieval enhancement generation agent generates the input context of the large language model, and the agent performs inference and response based on the input context of the large language model. If not, the agent performs inference and response based on the user's instructions.

[0036] Optionally, the data-to-thing fusion solution generation module is further used for: The input context and prompt words of the large language model are fed into the pre-trained large language model to obtain a disaster risk score and generate a customized loss reduction decision scheme.

[0037] The loss reduction decision-making schemes are divided into three categories: departments, enterprises, and the public, and a disaster risk assessment report is output, which includes real-time early warning, evacuation advice, and resource allocation.

[0038] The disaster risk score is dynamically bound to the geographic information system coordinates in the numerical simulation results of wind disaster and flood to obtain the geographic information system coordinates of the affected area. The geographic information system coordinates of the affected area are then spatially overlaid with the pre-stored urban infrastructure database to obtain the analysis results.

[0039] On the other hand, a smart early warning device for urban extreme weather disasters based on data-physical fusion is provided. The smart early warning device for urban extreme weather disasters based on data-physical fusion includes: a processor; a memory, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, any one of the methods in the above-described smart early warning method for urban extreme weather disasters based on data-physical fusion is implemented.

[0040] On the other hand, a computer-readable storage medium is provided, wherein at least one instruction is stored in the storage medium, the at least one instruction being loaded and executed by a processor to implement any of the above-described intelligent early warning methods for urban extreme weather disasters based on data-physical fusion.

[0041] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: This invention provides an intelligent early warning method and device for urban extreme weather disasters based on data-physical fusion, in order to solve the problem of the lack of numerical simulation and large model collaborative generation of disaster reduction strategies for specific locations in the current technology, and realize the generation of customized disaster reduction schemes for designated areas. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 This is a flowchart of an intelligent early warning method for urban extreme weather disasters based on data-physical fusion, provided by an embodiment of the present invention. Figure 2 This is a schematic diagram of the execution flow of the intelligent early warning method for urban extreme weather disasters based on data-physical fusion provided in an embodiment of the present invention; Figure 3 This is a block diagram of an intelligent early warning device for urban extreme weather disasters based on data-physical fusion, provided by an embodiment of the present invention. Figure 4 This is a schematic diagram of the structure of an intelligent early warning device for urban extreme weather disasters based on data-physical fusion, provided in an embodiment of the present invention. Detailed Implementation

[0044] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0045] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0046] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0047] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0048] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0049] This invention provides a method for intelligent early warning of urban extreme weather disasters based on data-physical fusion. This method can be implemented by an intelligent early warning device for urban extreme weather disasters based on data-physical fusion, which can be a terminal or a server. Figure 1 The flowchart shown is for an intelligent early warning method for urban extreme weather disasters based on data-physical fusion. The processing flow of this method may include the following steps: S1. Simulate and extract features of wind disasters and floods in the target city to obtain numerical simulation feature vectors; segment, clean, slice and transform the historical emergency plan text of the target city to obtain emergency plan text vectors; construct a vector database containing extreme disasters in the target city based on the numerical simulation feature vectors and emergency plan text vectors.

[0050] In one feasible implementation, the present invention constructs a vector database containing extreme disaster data for a target city. This vector database integrates structured data such as numerical simulation results of windstorms and floods in the target city, as well as unstructured text content such as contingency plans, loss mitigation guidelines, and historical cases. This enables the vectorized storage and representation of multi-source heterogeneous disaster data, specifically including the following steps 1.1-1.6: Step 1.1: Based on the semi-implicit method of computational fluid dynamics, pressure coupling equations and turbulence model, the wind disaster of the target city is simulated to obtain the numerical simulation results of the wind disaster.

[0051] In one feasible implementation, based on the CFD (Computational Fluid Dynamics) simulation method, an open-source CFD engine is used in conjunction with the SIMPLE (Semi-Implicit Method for Pressure Linked Equations) algorithm and the k-ε Realizable turbulence model to simulate wind disasters in the target city and obtain maps of building damage risk, pedestrian comfort risk, and tree fall risk.

[0052] For example, the vector database FAISS (Facebook AI Similarity Search) can be used to build a vector database containing extreme disasters in a specific region, integrating structured data such as numerical simulation results of wind and flood disasters with unstructured text content such as clauses and plans.

[0053] Specifically, based on CFD simulation methods, a wind disaster in a specific area is simulated by using an open-source CFD engine combined with the SIMPLE algorithm and the RANS turbulence model to obtain maps of building damage risk, pedestrian comfort risk, and tree fall risk.

[0054] In this embodiment, the RANS (Reynolds Average Navier-Stokes) method is used for steady-state simulation, and the k-εRealizable turbulence model is selected to improve the prediction accuracy of complex flows.

[0055] The pressure-velocity coupling algorithm employs the SIMPLE algorithm, which is suitable for steady-state calculations, and sets appropriate relaxation factors (pressure 0.3, momentum 0.7) to ensure convergence stability.

[0056] The total number of calculation steps was 2000, and the residual convergence criterion was set to 10⁻⁴ (continuity, momentum, and turbulence equations). At the same time, the velocity and pressure changes at key locations were monitored to verify the convergence.

[0057] To improve computational efficiency, a second-order discretization scheme (such as QUICK or second-order upwind) is adopted to balance accuracy and computational cost, and multigrid is used to accelerate convergence.

[0058] Step 1.2: Simulate the flooding of the target city based on the Mohid-Land model and the finite volume method to obtain the numerical simulation results of the flooding.

[0059] In one feasible implementation, an urban flood model is obtained based on the Mohid-Land model and the finite volume method to obtain urban waterlogging conditions under different rainfall amounts.

[0060] For example, based on the Mohid-Land model and the finite volume method, a flood model for a specific region can be obtained to obtain the urban waterlogging situation under different rainfall amounts.

[0061] The target area covered 22 streets and 7 towns, spanning 43,100 hectares. The topographic data used was a 30m × 30m grid DEM (Digital Elevation Model).

[0062] A two-dimensional surface runoff model is constructed by spatially discretizing DEM topographic data based on the finite volume method. This model is then coupled with a constructed one-dimensional pipe network model in the vertical direction between the surface and underground throughout the entire time period to simulate the process of water exchange between the underground drainage network and the surface.

[0063] Step 1.3: Extract features from the numerical simulation results of wind disaster and flood to obtain numerical simulation feature vectors.

[0064] In one feasible implementation, the numerical simulation results of wind field and flow field generated by physical simulation are subjected to feature extraction and transformed into high-dimensional structured vectors.

[0065] Step 1.4: Obtain historical contingency plan texts, segment and clean the historical contingency plan texts to obtain cleaned data; among them, historical contingency plan texts include historical meteorological data, disaster news and emergency plans.

[0066] In one feasible implementation, 20,000 historical meteorological data, disaster news, and emergency response plan texts are processed in segments and cleaned. The 20,000 texts are illustrative and not intended to limit the scope of this invention.

[0067] Specifically, the traversal process is first started based on the preset root directory path (ROOT_DIR).

[0068] Furthermore, the NFKC (Normalization Form KC) standard is used to transcode the text, unifying the encoding of full-width / half-width characters and special symbols.

[0069] Step 1.5: Slice the cleaned data to obtain multiple text blocks; transform each text block using a pre-trained large language model to obtain the vector embedding corresponding to each text block as the pre-planned text vector.

[0070] In one feasible implementation, a segmentation tool (load_and_split_text) is invoked to slice the cleaned long text according to a preset chunk size and overlap threshold. The text is sliced ​​into chunks of 1500 characters each with an overlap threshold of 50 characters. A remote large model embedding service is invoked through an API interface. In this embodiment, Qwen3 is preferred for testing. The model returns a corresponding high-dimensional vector that captures the deep semantic features of the text.

[0071] To achieve efficient retrieval and association of text information, the pre-trained large language model Qwen3 is used as a text vectorization tool to convert each text block into a vector embedding form.

[0072] Step 1.6: Further import the vector embedding results into a vector database to establish an efficient vector database.

[0073] S2. Design a retrieval enhancement generation agent, a disaster assessment agent, a report summary agent, and a meteorological disaster dialogue agent, and construct a multi-agent loss reduction decision-making big language model based on the agent scheduling framework.

[0074] Among them, the retrieval enhancement generative agent is used to generate the input context of the large language model based on user instructions and vector database; the disaster assessment agent is designed to obtain the disaster risk score based on the input context of the large language model; the report summary agent is used to extract key information from the disaster risk assessment report; and the meteorological disaster dialogue agent is used to reason and respond to user instructions.

[0075] In one feasible implementation, an agent scheduling framework is utilized, and a large language model is employed to construct a multi-agent loss mitigation decision-making model. This model integrates RAG (Retrieval Augmented Generation) and CoT (Chain of Thought) technologies to achieve chained invocation and logical reasoning among retrieval augmented generation agents, disaster assessment agents, report summarization agents, and dialogue agents. Specifically, this includes the following steps 2.1-2.5: Step 2.1: Design a retrieval-enhanced generative agent to convert user commands into query vectors; calculate the cosine similarity score between the query vector and the numerical simulation feature vector, and obtain the semantically matched numerical simulation feature vector based on the cosine similarity score; calculate the cosine similarity score between the query vector and the proposed text vector, and obtain the semantically matched proposed text vector based on the cosine similarity score; fuse the semantically matched numerical simulation feature vector and the proposed text vector to obtain the fused feature vector as the input context of the large language model.

[0076] In one feasible implementation, a retrieval-enhanced generative agent is designed to convert user commands into query vectors, and then semantically matches them with the vector database from step 1.6 to calculate a cosine similarity score. .

[0077] Parallel extraction of the numerical simulation feature vector (physical features) from step 1.3 and the contingency plan text vector from step 1.5 is performed to achieve context injection of data-physical fusion, enabling high-concurrency, millisecond-level semantic retrieval for large-scale urban disaster contingency plans. The formula is as follows: (1) in, Indicates the first Document blocks and query vectors Similarity score; For document blocks Embedded vector, Indicates the first document blocks The original text contained therein.

[0078] By using weighted concatenation, aligned projection, or multimodal attention mechanisms, the structured feature vectors extracted from physical simulation are fused with the retrieved text embedding vectors to form a fused feature vector, which serves as the input context for the large model.

[0079] Step 2.2: Design a disaster assessment agent. The disaster assessment agent receives the input context from the large language model, constructs prompt words through the thinking chain technology, inputs the input context and prompt words of the large language model into the pre-trained large language model, obtains the disaster risk score, and generates a customized loss reduction decision scheme.

[0080] In one feasible implementation, a disaster assessment agent is designed. The disaster assessment agent receives injected multimodal context, constructs prompt words through CoT technology, and guides a large language model to perform logical deduction.

[0081] After identifying high-risk areas and damage levels in the numerical simulation results, corresponding measures in the emergency response plan are linked, and consistent damage reduction recommendations are generated in real time.

[0082] Based on the concept of data-physical fusion, the physical feature vector extracted in step 1.3 is combined with the historical case data obtained through retrieval. Real-time environmental parameter set and the spatial infrastructure characteristics of the target location Alignment, function Disaster risk scoring is achieved through the reasoning process of a large language model. The formula is: (2) To adapt to multi-terminal display and quick front-end access, the intelligent agent output results adopt a dual format combining JSON and Markdown. The JSON format is used to encapsulate machine-readable fields such as risk quantification indicators, while the Markdown format is used to generate readable structured emergency plans.

[0083] Step 2.3: Design a report summarization agent to extract key information from lengthy disaster risk assessment reports, enabling users to quickly grasp the core content. This agent uses prompts to guide a large language model to automatically evaluate key information in the text and organizes the summary information according to a logical structure.

[0084] Guided by prompts, the model determines the importance of each piece of information in the overall disaster assessment and then aggregates them to form the final summary.

[0085] Step 2.4: Design a meteorological disaster dialogue agent. The meteorological disaster dialogue agent performs intent parsing on user commands. Based on the intent parsing result, it determines whether to call the retrieval enhancement generation agent. If so, the retrieval enhancement generation agent generates the input context of the large language model, and the inference response is performed based on the input context of the large language model. If not, the inference response is performed based on the user command.

[0086] In one feasible implementation, a meteorological disaster dialogue agent is designed to dynamically determine whether a retrieval-enhanced generation agent needs to be invoked during the dialogue process, and to directly perform reasoning and response based on the current context.

[0087] When a user asks a question, the agent first analyzes the user's intent and decides whether to perform knowledge retrieval based on the complexity of the intent.

[0088] If the question involves a specific disaster history or complex response plans, enhanced search is triggered.

[0089] Otherwise, the model completes the reasoning within the existing context.

[0090] The implementation of multi-turn dialogue logic is based on the state machine thinking model, which systematically manages the context information in the human-computer interaction process by defining a clear set of states and state transition rules.

[0091] In each round of dialogue interaction, the large language model captures and stores the user's historical input sequence and corresponding system responses in real time, forming structured dialogue state data. This state tracking mechanism ensures the integrity and temporal continuity of contextual information in multi-round interactions, avoiding dialogue breaks caused by information loss.

[0092] Step 2.5: Implement task scheduling based on the LangChain intelligent agent scheduling framework, decompose the disaster response process into specialized sub-tasks, and score tasks based on task priority (combined with relevance scores). Computational complexity and user context weight To achieve efficient chained calls and decision-making loops, the formula is: (3) in, , , These are the weighting coefficients for the three parameters, used to adjust the influence of each factor.

[0093] S3. By combining a vector database with a multi-agent loss reduction decision-making language model, customized loss reduction decision schemes are generated for specific disaster scenarios.

[0094] In one feasible implementation, the present invention combines physical simulation with LLM (Large Language Model) semantic understanding to extract physical boundary conditions and align logic for specific disaster scenarios, thereby generating customized loss reduction decision schemes.

[0095] Furthermore, based on the characteristics of different social entities, the generated loss reduction decision-making schemes are divided into three categories: departments, enterprises, and the public, and a hierarchical disaster prevention decision-making report is output, which includes real-time early warning, evacuation advice, and resource allocation.

[0096] Optionally, step S3 above may include: The input context and prompt words of the large language model are fed into the pre-trained large language model to obtain a disaster risk score and generate a customized loss reduction decision scheme.

[0097] The loss reduction decision-making schemes are divided into three categories: departments, enterprises, and the public, and a disaster risk assessment report is output, which includes real-time early warning, evacuation advice, and resource allocation.

[0098] The disaster risk score is dynamically bound to the geographic information system coordinates in the numerical simulation results of wind disaster and flood to obtain the geographic information system coordinates of the affected area. The geographic information system coordinates of the affected area are then spatially overlaid with the pre-stored urban infrastructure database to obtain the analysis results.

[0099] In one feasible implementation, the risk score output in step 2.2 is... Dynamically bind the coordinates with the geographic information system coordinates from steps 1.1 and 1.2; Furthermore, when calling the large language model for logical deduction, the inference parameters Temperature are set to 0.2 and Top_p to 0.95. The generated fused feature vector and the system prompt are input together with the model's 128k tokens long context window to guide the model to output a structured risk assessment report.

[0100] Furthermore, for different object-oriented subjects, a pre-trained large language model is used to parse high-dimensional physical simulation data into structured action instructions: By analyzing the physical field numerical results obtained in step 1.3, the GIS (Geographic Information System) coordinates of the affected area are extracted in real time and spatially overlaid with the pre-stored urban infrastructure database.

[0101] By using RAG technology to extract defense standards and historical cases that match the current physical field parameters from the vector database, and combining them with CoT prompt words, the large language model is driven to perform logical reasoning to make disposal suggestions.

[0102] For departments: Issue macro-level management instructions.

[0103] For businesses: Output professional production instructions that include a list of potential shutdown risks and reinforcement points.

[0104] For the public: Provide survival guidelines that include real-time warnings of falling objects from heights.

[0105] like Figure 2 As shown, this invention constructs a vector database containing extreme disaster data for target cities, integrating structured data such as numerical simulation results of wind and flood disasters with unstructured text content such as clauses and plans. It utilizes an agent scheduling framework to construct a large-scale multi-agent loss reduction decision-making model, integrating RAG and CoT technologies to achieve chained invocation and logical reasoning of retrieval-enhanced generation agents, disaster assessment agents, report summarization agents, and dialogue agents. The method combines physical simulation with LLM semantic understanding to generate customized loss reduction decision-making schemes for specific disaster scenarios. This invention solves the problem of transforming meteorological forecast information into accurate decision-making, realizing the transformation from single meteorological parameter forecasting to scenario-based loss reduction decision support, and significantly improving the intelligence level of urban disaster early warning.

[0106] In this embodiment of the invention, a method and device for intelligent early warning of urban extreme weather disasters based on data-physical fusion are provided to solve the problem of the lack of numerical simulation and large model collaborative generation of disaster reduction strategies for specific locations in the current technology, and to realize the generation of customized disaster reduction schemes for designated areas.

[0107] Figure 3This is a block diagram illustrating an intelligent early warning device for urban extreme weather disasters based on data-physical fusion, according to an exemplary embodiment. The device is used in an intelligent early warning method for urban extreme weather disasters based on data-physical fusion. (Refer to...) Figure 3 The device includes a vector database construction module 310, a multi-agent loss reduction decision-making large model construction module 320, and a data-object fusion scheme generation module 330. Among them: The vector database construction module 310 is used to simulate and extract features of wind disasters and floods in the target city to obtain numerical simulation feature vectors; to perform segmentation, text cleaning, slicing and transformation on the historical contingency plan text of the target city to obtain contingency plan text vectors; and to construct a vector database containing extreme disasters in the target city based on the numerical simulation feature vectors and the contingency plan text vectors.

[0108] The multi-agent loss reduction decision-making large model construction module 320 is used to design retrieval enhancement generation agents, disaster assessment agents, report summarization agents, and meteorological disaster dialogue agents, and to construct a multi-agent loss reduction decision-making large language model based on the agent scheduling framework.

[0109] Among them, the retrieval enhancement generative agent is used to generate the input context of the large language model based on user instructions and vector database; the disaster assessment agent is designed to obtain the disaster risk score based on the input context of the large language model; the report summary agent is used to extract key information from the disaster risk assessment report; and the meteorological disaster dialogue agent is used to reason and respond to user instructions.

[0110] The data-physical fusion solution generation module 330 is used to generate customized loss reduction decision schemes for specific disaster scenarios by combining a vector database with a multi-agent loss reduction decision big language model.

[0111] In this embodiment of the invention, a method and device for intelligent early warning of urban extreme weather disasters based on data-physical fusion are provided to solve the problem of the lack of numerical simulation and large model collaborative generation of disaster reduction strategies for specific locations in the current technology, and to realize the generation of customized disaster reduction schemes for designated areas.

[0112] Figure 4 This is a schematic diagram of the structure of an intelligent early warning device for urban extreme weather disasters based on data-physical fusion, provided in an embodiment of the present invention. Figure 4 As shown, the intelligent early warning device for urban extreme weather disasters based on data-physical fusion can include the above-mentioned... Figure 3 The illustrated intelligent early warning device for urban extreme weather disasters based on data-physical fusion. Optionally, the intelligent early warning device 410 for urban extreme weather disasters based on data-physical fusion may include a first processor 2001.

[0113] Optionally, the intelligent early warning device 410 for urban extreme weather disasters based on data-physical fusion may also include a memory 2002 and a transceiver 2003.

[0114] The first processor 2001, memory 2002, and transceiver 2003 can be connected via a communication bus.

[0115] The following is combined with Figure 4 A detailed introduction to each component of the 410 intelligent early warning device for urban extreme weather disasters based on data-physical fusion is provided below: The first processor 2001 is the control center of the intelligent early warning device 410 for urban extreme weather disasters based on the fusion of data and physical systems. It can be a single processor or a collective term for multiple processing elements. For example, the first processor 2001 can be one or more central processing units (CPUs), application-specific integrated circuits (ASICs), or one or more integrated circuits configured to implement the embodiments of the present invention, such as one or more digital signal processors (DSPs), or one or more field-programmable gate arrays (FPGAs).

[0116] Optionally, the first processor 2001 can execute various functions of the smart early warning device 410 for urban extreme weather disasters based on data-physical fusion by running or executing software programs stored in the memory 2002 and calling data stored in the memory 2002.

[0117] In a specific implementation, as one example, the first processor 2001 may include one or more CPUs, for example... Figure 4 CPU0 and CPU1 are shown in the diagram.

[0118] In a specific implementation, as one example, the intelligent early warning device 410 for urban extreme weather disasters based on data-physical fusion may also include multiple processors, for example... Figure 4 The first processor 2001 and the second processor 2004 are shown in the diagram. Each of these processors can be a single-core processor or a multi-core processor. Here, a processor can refer to one or more devices, circuits, and / or processing cores used to process data (such as computer program instructions).

[0119] The memory 2002 is used to store the software program that executes the present invention, and is controlled by the first processor 2001 to execute it. The specific implementation method can be referred to the above method embodiment, and will not be repeated here.

[0120] Optionally, the memory 2002 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. The memory 2002 may be integrated with the first processor 2001 or exist independently, and may be connected via the interface circuit of the data-physical fusion-based urban extreme weather disaster intelligent early warning device 410. Figure 4 (Not shown in the image) is coupled to the first processor 2001, and this embodiment of the invention does not specifically limit this.

[0121] The transceiver 2003 is used to communicate with network devices or with terminal devices.

[0122] Alternatively, transceiver 2003 may include a receiver and a transmitter. Figure 4 (Not shown separately). The receiver is used to implement the receiving function, and the transmitter is used to implement the transmitting function.

[0123] Optionally, the transceiver 2003 can be integrated with the first processor 2001, or it can exist independently, and it can be connected to the interface circuit of the smart early warning device for urban extreme weather disasters based on data-physical fusion 410. Figure 4 (Not shown in the image) is coupled to the first processor 2001, and this embodiment of the invention does not specifically limit this.

[0124] It should be noted that, Figure 4 The structure of the smart early warning device 410 for urban extreme weather disasters based on data-physical fusion shown in the figure does not constitute a limitation on the router. The actual knowledge structure recognition device may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0125] Furthermore, the technical effects of the intelligent early warning device 410 for urban extreme weather disasters based on data-physical fusion can be referred to the technical effects of the intelligent early warning method for urban extreme weather disasters based on data-physical fusion described in the above method embodiments, and will not be repeated here.

[0126] It should be understood that the first processor 2001 in the embodiments of the present invention may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0127] It should also be understood that the memory in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDR SDRAM), enhanced synchronous DRAM (ESDRAM), synchronous linked DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0128] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0129] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0130] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0131] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0132] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0133] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0134] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0135] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0136] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0137] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0138] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A smart early warning method for urban extreme weather disasters based on data-physical fusion, characterized in that, The method includes: S1. Simulate and extract features of wind disasters and floods in the target city to obtain numerical simulation feature vectors; segment, clean, slice and transform the historical contingency plan text of the target city to obtain contingency plan text vectors; construct a vector database containing extreme disasters in the target city based on the numerical simulation feature vectors and contingency plan text vectors. S2. Design a retrieval enhancement generation agent, a disaster assessment agent, a report summary agent, and a meteorological disaster dialogue agent, and construct a multi-agent loss reduction decision-making big language model based on the agent scheduling framework; Among them, the retrieval enhancement generative agent is used to generate the input context of the large language model based on user instructions and vector database; the disaster assessment agent is designed to obtain the disaster risk score based on the input context of the large language model; the report summary agent is used to extract key information from the disaster risk assessment report; and the meteorological disaster dialogue agent is used to reason and respond to user instructions. S3. By combining a vector database with a multi-agent loss reduction decision-making big language model, customized loss reduction decision schemes are generated for specific disaster scenarios. The process of generating the input context for the large language model based on user instructions and a vector database includes: The enhanced generative agent transforms user commands into query vectors through retrieval enhancement. Calculate the cosine similarity score between the query vector and the numerical simulation feature vector, and obtain the semantically matched numerical simulation feature vector based on the cosine similarity score; Calculate the cosine similarity score between the query vector and the proposed text vector, and obtain the semantically matched proposed text vector based on the cosine similarity score; The numerical simulation feature vectors of semantic matching and the pre-planned text vectors are fused to obtain a fused feature vector, which serves as the input context for the large language model.

2. The intelligent early warning method for urban extreme weather disasters based on data-physical fusion as described in claim 1, characterized in that, The simulation and feature extraction of windstorms and floods in the target city in S1 yields a numerical simulation feature vector, including: The wind disaster in the target city was simulated using a semi-implicit method based on computational fluid dynamics, pressure coupling equations, and a turbulence model, and the numerical simulation results of the wind disaster were obtained. The flooding of the target city was simulated based on the Mohid-Land model and the finite volume method, and the numerical simulation results of the flooding were obtained. Feature extraction was performed on the numerical simulation results of wind disaster and flood to obtain numerical simulation feature vectors.

3. The intelligent early warning method for urban extreme weather disasters based on data-physical fusion according to claim 1, characterized in that, The process in S1 involves segmenting, cleaning, slicing, and transforming the historical contingency plan text of the target city to obtain a contingency plan text vector, including: The historical contingency plan texts are obtained, segmented, and cleaned to obtain cleaned data. The historical contingency plan texts include historical meteorological data, disaster news, and emergency plans. The cleaned data is sliced ​​to obtain multiple text blocks; Each text block is transformed by a pre-trained large language model, and the vector embedding corresponding to each text block is used as the pre-planned text vector.

4. The intelligent early warning method for urban extreme weather disasters based on data-physical fusion according to claim 1, characterized in that, The process of obtaining a disaster risk score based on the input context of a large language model includes: The disaster assessment agent receives the input context from the large language model, constructs prompt words through the thinking chain technology, inputs the input context and prompt words of the large language model into the pre-trained large language model, obtains the disaster risk score, and generates a customized loss reduction decision scheme.

5. The intelligent early warning method for urban extreme weather disasters based on data-physical fusion according to claim 1, characterized in that, The inference response to the user command includes: The meteorological disaster dialogue agent parses the user's instructions to determine whether to invoke the retrieval enhancement generation agent. If so, the retrieval enhancement generation agent generates the input context of the large language model, and the agent performs inference and response based on the input context of the large language model. If not, the agent performs inference and response based on the user's instructions.

6. The intelligent early warning method for urban extreme weather disasters based on data-physical fusion according to claim 1, characterized in that, The S3 includes: The input context and prompt words of the large language model are fed into the pre-trained large language model to obtain a disaster risk score and generate a customized loss reduction decision scheme; The loss reduction decision-making schemes are divided into three categories: departments, enterprises, and the public, and a disaster risk assessment report is output, which includes real-time early warning, evacuation advice, and resource allocation. The disaster risk score is dynamically bound to the geographic information system coordinates in the numerical simulation results of wind disaster and flood to obtain the geographic information system coordinates of the affected area. The geographic information system coordinates of the affected area are then spatially overlaid with the pre-stored urban infrastructure database to obtain the analysis results.

7. A smart early warning device for urban extreme weather disasters based on data-physical fusion, wherein the smart early warning device for urban extreme weather disasters based on data-physical fusion is used to implement the smart early warning method for urban extreme weather disasters based on data-physical fusion as described in any one of claims 1-6, characterized in that, The device includes: The vector database construction module is used to simulate and extract features of wind disasters and floods in the target city to obtain numerical simulation feature vectors; to perform segmentation, text cleaning, slicing and transformation on the historical emergency plan text of the target city to obtain emergency plan text vectors; and to construct a vector database containing extreme disasters in the target city based on the numerical simulation feature vectors and the emergency plan text vectors. The module for building a large multi-agent loss reduction decision model is used to design a retrieval enhancement generation agent, a disaster assessment agent, a report summarization agent, and a meteorological disaster dialogue agent, and to build a large multi-agent loss reduction decision language model based on the agent scheduling framework. Among them, the retrieval enhancement generative agent is used to generate the input context of the large language model based on user instructions and vector database; the disaster assessment agent is designed to obtain the disaster risk score based on the input context of the large language model; the report summary agent is used to extract key information from the disaster risk assessment report; and the meteorological disaster dialogue agent is used to reason and respond to user instructions. The data-physical fusion solution generation module is used to generate customized loss reduction decision-making solutions for specific disaster scenarios by combining a vector database with a multi-agent loss reduction decision-making big language model.

8. A smart early warning device for urban extreme weather disasters based on data-physical fusion, characterized in that, The intelligent early warning device for urban extreme weather disasters based on data-physical fusion includes: processor; A memory storing computer-readable instructions that, when executed by the processor, implement the method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium contains program code that can be invoked by a processor to execute the method as described in any one of claims 1 to 6.

Images