Intelligent generation method, system, electronic equipment and storage medium for flood disaster prevention plans based on inductive feedback.

Abstract

This invention discloses an intelligent generation method, system, electronic device, and storage medium for flood disaster prevention plans based on inductive feedback. The intelligent generation method for flood disaster prevention plans includes: constructing a knowledge base for flood disaster prevention plans; generating an initial prevention plan; generating multiple candidate plan schemes; inputting the multiple candidate plan schemes into a water conservancy professional model to perform flood evolution induction, obtaining the induction results corresponding to each candidate plan scheme; evaluating the consistency between each induction result and the preset expected result on multiple preset dimensions, selecting the optimal scheme based on the evaluation results, and revising the initial prevention plan based on the optimal scheme to obtain an optimized prevention plan. This invention significantly improves the intelligence level, scientific decision-making ability, and adaptive evolution ability of flood disaster prevention plans, providing practical and feasible technical support for the plan function in the "four preparations" of digital twin water conservancy.

Description

Technical Field

[0001] This invention relates to the field of flood prevention and early warning technology, and specifically to a method, system, electronic device, and storage medium for intelligent generation of flood disaster prevention plans based on inductive feedback. Background Technology

[0002] In related technologies, to effectively respond to floods, water conservancy departments need to develop detailed defense plans for different flood protection zones, flash flood basins, and reservoir projects. Traditional defense plan development mainly relies on manual experience and is stored in the form of static documents, which has many prominent problems in actual work:

[0003] (1) The efficiency of emergency plan preparation is low and the updates are lagging. Traditional emergency plans are usually written manually. It takes a long time for an emergency plan to go from data collection, analysis and demonstration to completion. Once it is completed, it often remains unchanged for many years. It is difficult to dynamically generate or quickly adjust according to the real-time flood disaster scenario forecast, resulting in the emergency plan being out of touch with the actual situation.

[0004] (2) Most existing plans are static texts based on experience and lack coupling verification with water conservancy professional models such as hydrological models and hydrodynamic models. Whether the key measures set in the plans, such as evacuation routes and scheduling rules, can be effectively implemented in the actual flood evolution process lacks quantitative simulation and scientific evaluation.

[0005] (3) A large amount of unstructured knowledge, such as emergency plans, dispatch schemes, and expert experience, has been accumulated in historical flood disasters, but the existing system lacks an effective mechanism for extracting, organizing, and reusing this knowledge. The emergency plans and simulation results generated during a flood control process cannot be fed back into the knowledge base to optimize the subsequent emergency plan generation model, and there is a lack of "mutual driving and feedback" between knowledge and decision-making.

[0006] (4) When faced with the same disaster scenario, decision-makers often need to consider multiple candidate solutions. However, existing methods are difficult to generate multiple candidate solutions quickly and automatically, and there is a lack of a mechanism to conduct consistency evaluation and select the best candidate solution based on the results of water conservancy professional model simulation.

[0007] In recent years, with the advancement of digital twin water conservancy construction, some research has begun to explore the use of deep learning models (such as Transformer) to generate contingency plan texts, or to assist flood control decision-making through knowledge graphs and large language models. However, existing technologies have not yet established a complete chain from "knowledge base—contingency plan generation—water conservancy model deduction—evaluation and optimization—feedback optimization," and in particular, lack a "deduction-feedback" type intelligent generation method that can feed the deduction results back to the contingency plan generation model for iterative optimization. Therefore, it is urgent to propose a new technical solution to address the above problems. Summary of the Invention

[0008] This invention addresses the shortcomings of existing flood disaster prevention plans, such as reliance on manual labor in their preparation, inability to dynamically adapt to disaster scenarios, lack of verification and multi-scheme selection mechanisms based on water conservancy professional models, and the inability to form a feedback optimization loop between knowledge and models. It provides an intelligent generation method for flood disaster prevention plans based on simulation feedback, which enables automated generation of prevention plans, consistency evaluation and selection of multiple schemes based on flood evolution simulation, and iterative optimization of the plan generation model using simulation results, thereby improving the scientificity, timeliness, and adaptability of the plans.

[0009] To achieve the above objectives, the present invention employs the following technical solution:

[0010] A method for intelligently generating flood disaster prevention plans based on inductive feedback includes the following steps:

[0011] Construct a knowledge base for flood disaster prevention plans, which includes at least disaster scenario attributes, historical plan texts, risk levels, evacuation routes, early warning levels, and dispatch rules.

[0012] Based on real-time or forecasted flood disaster scenario information, an initial defense plan is generated by calling a deep learning-based initial plan generation model.

[0013] For the same flood disaster scenario, multiple candidate contingency plans are generated based on the knowledge base and the initial defense plan.

[0014] The multiple candidate contingency plans are input into the water conservancy professional model to perform flood evolution simulation, and the simulation results corresponding to each candidate contingency plan are obtained. The water conservancy professional model includes at least a hydrological model and a hydrodynamic model.

[0015] Based on the multi-agent self-consistency evaluation method, the consistency of each deduction result with the preset expected result is evaluated on multiple preset dimensions. Based on the evaluation results, the optimal solution is selected from multiple candidate contingency planning schemes. The initial defense plan is then modified according to the optimal solution to obtain the optimized defense plan.

[0016] The optimized defense plan is stored in the knowledge base as a historical case for the generation of subsequent plans.

[0017] Preferably, the steps for constructing the knowledge base include:

[0018] Acquire multi-source data related to flood disasters, use large models for knowledge extraction, and extract entities, attributes, and relationships from the contingency plans.

[0019] The content of each type of flood control plan is decomposed in a structured manner, a plan template is constructed and stored in the database, and the plan template includes dynamically populated attribute fields and fixed text paragraphs.

[0020] Extract the activation conditions, protected objects, relocation and resettlement methods, early warning levels, and scheduling methods of the contingency plan as key attributes, and associate and store these key attributes with the corresponding contingency plan templates.

[0021] Preferably, the initial contingency plan generation model is a deep learning model based on a self-attention mechanism, used to capture the temporal features and contextual relationships of disaster scenarios and generate a sequence of contingency plan texts.

[0022] Preferably, the initial contingency plan generation model is based on the Transformer architecture, and its training process includes: extracting historical disaster scenario attributes and risk levels from the knowledge base as model input, extracting evacuation routes, warning levels, and scheduling methods from the corresponding historical contingency plans as labels, and training it using a supervised learning method; when generating the initial defense plan, using an autoregressive method to predict the output sequence word by word.

[0023] Preferably, the method of generating multiple candidate contingency plans includes: subjecting the key parameters in the initial defense plan to multiple perturbations, or retrieving multiple historical contingency plans with similar scenarios from the knowledge base for adaptation and adjustment.

[0024] Preferably, the multi-agent self-consistency evaluation method includes: scoring the deduction results of each candidate plan scheme on each preset dimension, and then using a majority voting or weighted voting strategy to integrate the scoring results of each preset dimension, and taking the candidate plan scheme with the highest comprehensive score as the optimal scheme.

[0025] Preferably, after revising the initial defense plan, a corresponding preset template is extracted from the plan template database based on template document generation technology. Then, entity recognition and attribute auto-fill technology are used to fill the key decision information in the optimal plan into the preset template to generate a structured plan document.

[0026] Preferably, the intelligent generation method for flood disaster prevention plans further includes generating emergency response plans based on a retrieval-enhanced generation model to supplement the optimized prevention plan. The steps for generating the emergency response plans include:

[0027] Indexing phase: Flood disaster response cases, expert experience documents, and scheduling rule documents are segmented into text blocks, converted into vectors through an embedding model, and stored in a vector database.

[0028] Retrieval phase: Real-time emergency information is converted into query vectors using the same embedding model, and the most relevant text blocks are retrieved based on vector similarity and keyword matching algorithms.

[0029] Generation phase: Real-time emergency information and retrieved text blocks are input into a large language model to generate an emergency response plan.

[0030] Preferably, after the emergency response plan is generated, a multi-objective optimization algorithm is used to rank and recommend the multiple candidate response plans, with the objectives of minimizing response time, resource consumption, and flooding impact.

[0031] This invention also provides an intelligent flood disaster prevention plan generation system based on inductive feedback, used to execute the above-mentioned intelligent flood disaster prevention plan generation method, the intelligent flood disaster prevention plan generation system comprising:

[0032] The contingency plan knowledge construction module is used to acquire multi-source data related to flood disasters, extract knowledge using a large model, and form a structured knowledge base;

[0033] The initial contingency plan generation module has a built-in initial contingency plan generation model, which is used to generate initial defense plans based on flood disaster scenario information;

[0034] A multi-candidate solution generation module is used to generate multiple candidate solution planning schemes based on the knowledge base and the initial defense plan;

[0035] The feedback simulation module is used to input each candidate contingency plan into the water conservancy professional model to perform flood evolution simulation and obtain the simulation results corresponding to each candidate contingency plan.

[0036] The self-consistency evaluation module is used to evaluate the consistency between each simulation result and the preset expected result across multiple preset dimensions. Based on the evaluation results, it selects the optimal solution from multiple candidate contingency plans and modifies the initial defense plan according to the optimal solution to obtain an optimized defense plan.

[0037] The knowledge update module is used to store the optimized defense plan into the knowledge base, which serves as a historical case for the generation of subsequent plans.

[0038] Preferably, the water conservancy professional model includes a hydrological model and a hydrodynamic model.

[0039] Preferably, the self-consistency assessment module includes a majority voting assessment unit, which is used to conduct consistency assessments on the simulation results corresponding to each candidate contingency plan in terms of risk level, warning level, flooding range, and affected area, and select the optimal plan.

[0040] Preferably, the intelligent generation system for flood disaster prevention plans further includes an emergency response module, which includes a vector database, a retrieval unit, and a large model generator. The vector database is used to store the embedding vectors of case text blocks, the retrieval unit is used to perform vector similarity retrieval and keyword matching retrieval, and the large model generator is used to fuse the retrieval results and generate an emergency response plan.

[0041] The present invention also provides an electronic device, including a processor and a memory, wherein the memory stores machine-executable instructions that can be executed by the processor, and the processor can execute the machine-executable instructions to implement the above-mentioned intelligent generation method for flood disaster prevention plans.

[0042] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-described method for intelligently generating flood disaster prevention plans.

[0043] Compared with existing technologies, this invention, through the technical concept of inference and feedback, opens up the entire chain from plan generation to model verification, from scheme selection to feedback optimization, significantly improving the intelligence level, scientific decision-making ability and adaptive evolution ability of flood disaster prevention plans, and providing practical and feasible technical support for the plan function of digital twin water conservancy "four plans" (forecast, early warning, rehearsal and plan).

[0044] Other features and advantages of the present invention will be described in detail in the following specific embodiments. Attached Figure Description

[0045] Figure 1 This is a flowchart of an intelligent generation method for flood disaster prevention plans based on inference and feedback provided in an embodiment of the present invention;

[0046] Figure 2 This is a schematic diagram of a model training and risk scenario contingency plan generation process provided by an embodiment of the present invention;

[0047] Figure 3 This is a schematic diagram of a Transformer deep learning structure provided in an embodiment of the present invention;

[0048] Figure 4 This is a schematic diagram of a typical paradigm of a general large model and a retrieval enhancement generation model provided by an embodiment of the present invention. Detailed Implementation

[0049] To make the technical means, creative features, objectives, and effects of this invention readily understandable, the invention is further illustrated below with reference to specific embodiments. It should be understood that the specific embodiments described herein are for illustrative and explanatory purposes only and are not intended to limit the invention.

[0050] As mentioned above, combined with Figure 1 As shown, this embodiment of the invention provides a method for intelligently generating flood disaster prevention plans based on inductive feedback, including the following steps:

[0051] S1: Construct a knowledge base for flood disaster prevention plans. The knowledge base shall include at least disaster scenario attributes, historical plan texts, risk levels, evacuation routes, early warning levels, and dispatch rules.

[0052] In this invention, the disaster scenario attributes refer to key characteristics describing a flood disaster, such as drainage area, rainfall intensity, soil saturation, and initial river level; the risk level is the degree of danger assessed based on historical data and real-time information, such as extremely serious, serious, relatively serious, and general; the evacuation route is the path planning for the evacuation of people from the danger zone to safe resettlement points; the warning level can be classified into four levels, such as red, orange, yellow, and blue; and the scheduling rules are the operating procedures for water conservancy projects such as reservoirs and sluice gates at different water levels.

[0053] In this invention, a knowledge base for flood disaster prevention plans is constructed to provide a data foundation and knowledge support for subsequent plan generation, scheme retrieval, and model training. In traditional methods, this data is scattered across various documents and systems, making it difficult to utilize effectively. This invention, by constructing a unified structured knowledge base, transforms multi-source heterogeneous data, including text, tables, and images, into machine-understandable knowledge, enabling the initial plan generation model to directly learn from historical experience.

[0054] S2: Based on real-time or forecasted flood disaster scenario information, call the deep learning-based initial plan generation model to generate an initial defense plan.

[0055] By calling a deep learning-based initial plan generation model to generate initial defense plans, the defense plans no longer rely solely on manual writing. Instead, they can automatically access real-time monitoring or forecast data and quickly output initial plans before a disaster strikes. The response time is reduced from several hours or even days in the traditional manual process to minutes, buying valuable time for flood control decisions.

[0056] S3: For the same flood disaster scenario, based on the knowledge base and the initial defense plan, generate multiple candidate plan planning schemes; for example, generate multiple variant schemes by changing the water level threshold for initiating evacuation, replacing the evacuation route, adjusting the timing of early warning release, etc.

[0057] Since a single contingency plan is insufficient to cover the uncertainties in the evolution of a disaster, generating multiple candidate contingency plans provides a sufficient pool of alternatives for subsequent simulation, evaluation, and selection.

[0058] S4: Input the multiple candidate contingency plans into the water conservancy professional model to perform flood evolution simulation and obtain the simulation results corresponding to each candidate contingency plan. The water conservancy professional model includes at least a hydrological model and a hydrodynamic model.

[0059] In this invention, the hydrological model is a simulation tool that generalizes the physical or logical processes of natural hydrological systems. Its core value lies in revealing and predicting hydrological patterns by establishing mathematical structures with specific physical meaning, thereby solving practical problems such as flood forecasting, water resource assessment, and water environment protection. Representative hydrological models include distributed physical models (DPMAs) or conceptual hydrological models such as the Xin'anjiang model. Specific examples of DPMAs include the SWAT model, which focuses on simulating watershed water balance and pollutant migration, and the Parflow model, which can accurately simulate the water and energy balance processes from groundwater to vegetation. The hydrodynamic model is a mathematical model built based on fluid mechanics principles. It describes the laws of water flow through a system of differential equations such as mass conservation and momentum conservation, and is widely used in water conservancy engineering, marine simulation, and urban flood control. This model uses numerical methods to solve boundary value problems, supports spatial simulation from one-dimensional to three-dimensional, and typical software tools include MIKE, HEC-RAS, and Delft3D. Dynamic simulation and risk assessment can be achieved through digital twin technology.

[0060] This invention overcomes the shortcomings of traditional plans that cannot be quantitatively evaluated by inputting multiple candidate contingency plans into a water conservancy professional model to simulate flood evolution. The hydrological and hydrodynamic models can simulate the inundation range and water level process of floods under different plans, thereby identifying potential risks in the plans in advance, such as flooding of evacuation routes and failure of dispatching measures, and thus providing objective and quantitative simulation results for subsequent evaluation.

[0061] S5: Based on the multi-agent self-consistency evaluation method, the consistency evaluation of each deduction result and the preset expected result is carried out on multiple preset dimensions. Based on the evaluation results, the optimal solution is selected from multiple candidate contingency planning schemes. The initial defense plan is modified according to the optimal solution to obtain the optimized defense plan.

[0062] In this invention, multi-agent self-consistent evaluation refers to setting up multiple independent agents, each of which can be a rule-based scorer, a trained neural network evaluation model, or an agent encapsulating expert decision-making logic, allowing them to independently evaluate the deduction results of the same candidate plan. If the conclusions reached by multiple agents after judging the deduction results of the same plan from different perspectives are highly consistent, for example, concluding that the candidate plan can meet the expected results, then the candidate plan is considered reliable and self-consistent.

[0063] The preset expected results can be pre-set safety thresholds, such as the flooding range not exceeding a certain area, the maximum water depth being lower than the warning value, and the time for the evacuation of the masses being completed being earlier than the time the flood arrives.

[0064] In this invention, by having multiple intelligent agents independently evaluate the simulation results of different candidate contingency plans under the same disaster scenario, and then making a comprehensive judgment through voting or weighting strategies, the bias of a single evaluation perspective can be effectively avoided, and the best comprehensive performance in multiple dimensions such as risk level, warning level, flood range, and affected area can be selected, making the final adopted contingency plan more robust and reliable.

[0065] S6: Store the optimized defense plan in the knowledge base as a historical case for subsequent plan generation.

[0066] In this invention, optimized defense plans are stored in the knowledge base as historical cases for subsequent plan generation. This ensures that each successful or optimized plan becomes part of the knowledge base, allowing for reference in generating new plans through similar scenario retrieval, avoiding "reinventing the wheel" and achieving knowledge reuse and inheritance. In a preferred embodiment, the optimized defense plan and the deduction results of the optimal solution can serve as new training data to iteratively optimize the initial plan generation model. This allows successful plans and corresponding deduction verification data from each flood control operation to flow back to the model training stage, continuously improving the prediction accuracy and scenario adaptability of the initial plan generation model. After multiple iterations, the initial plan generation model can automatically learn "which plans are better in the deduction," thus directly generating higher-quality initial plans when encountering similar disaster scenarios again. This positive feedback mechanism combining knowledge-driven and model-driven approaches fundamentally solves the problem of the lack of mutual feedback between knowledge and decision-making in traditional solutions.

[0067] According to the intelligent generation method for flood disaster prevention plans provided by this invention, the entire chain from plan generation to model verification, from scheme selection to feedback optimization is opened up through the technical architecture of "deduction and feedback". This significantly improves the intelligence level, scientific decision-making ability and adaptive evolution ability of flood disaster prevention plans, and provides practical technical support for the plan function in the "four early warnings" (forecast, early warning, early warning and plan) of digital twin water conservancy.

[0068] In some embodiments of the present invention, the knowledge base construction steps specifically include:

[0069] S11: Obtain multi-source data related to flood disasters, use a large model to extract knowledge, and extract entities, attributes and relationships from the contingency plan; wherein, the multi-source data includes, for example, historical flood records, meteorological reports, water level and flow tables, remote sensing images, etc.; the entities include, for example, names of water conservancy projects, names of villages, etc.; the attributes include, for example, reservoir capacity, resettlement capacity, etc.; and the relationships include, for example, upstream and downstream relationships, relocation correspondence, etc.

[0070] S12: The content of each type of flood control plan is decomposed in a structured manner, a plan template is constructed and stored in the database. The plan template includes dynamically populated attribute fields and fixed text paragraphs, so that when generating a plan later, only key data needs to be filled in to obtain a complete document.

[0071] S13: Extract the activation conditions, protected objects, relocation and resettlement methods, early warning levels, and scheduling methods of the contingency plan as key attributes, and associate and store the key attributes with the corresponding contingency plan template.

[0072] The knowledge base on flood disaster prevention plans constructed in the above manner not only stores historical plans, but also forms a reusable template library and attribute library, providing a high-quality data foundation for subsequent intelligent generation and retrieval.

[0073] In this embodiment of the invention, the initial contingency plan generation model is a deep learning model based on a self-attention mechanism, which is used to capture the temporal features and contextual relationships of disaster scenarios and generate a sequence of contingency plan texts.

[0074] Furthermore, in some embodiments of the present invention, the initial contingency plan generation model is based on the Transformer architecture. Its training process includes: extracting historical disaster scenario attributes and risk levels from the knowledge base as model input; extracting evacuation routes, warning levels, and scheduling methods from corresponding historical contingency plans as annotation labels; and training using supervised learning methods, for example, using historical flood scenario information and text fragments of successful contingency plans as a pair of training samples. When generating the initial defense plan, an autoregressive method is used to predict the output sequence word by word, that is, predicting the next most likely word based on the already generated words and the input scenario information, until a complete contingency plan text is generated. Through the above training, the initial contingency plan generation model can learn the mapping relationship from disaster scenarios to contingency plan measures, and can generate grammatically correct and logically sound contingency plan text based on real-time scenarios, with a generation speed significantly faster than manual compilation.

[0075] In some embodiments of the present invention, the method of generating multiple candidate contingency planning schemes includes: subjecting multiple sets of perturbations to the key parameters in the initial defense plan, for example, setting the "critical water level for initiating transfer" to different thresholds, such as once in 10 years or once in 20 years; or selecting different road combinations for the "transfer route", thereby generating multiple candidate contingency planning schemes.

[0076] In other embodiments of the present invention, the method of generating multiple candidate contingency planning schemes includes: retrieving multiple historical contingency plans with similar scenarios from the knowledge base and adapting them; specifically, retrieving multiple historical contingency plans most similar to the current scenario from the knowledge base and adapting these historical contingency plans, for example, replacing the location name and time parameter in the historical contingency plan with the current scenario value, thereby obtaining multiple candidate contingency planning schemes.

[0077] It should be noted that the methods for generating multiple candidate planning schemes provided in the above embodiments can be used individually or in combination.

[0078] In this embodiment of the invention, the multi-agent self-consistency evaluation method includes: scoring the deduction results of each candidate planning scheme on each preset dimension; then using a majority voting or weighted voting strategy to synthesize the scoring results of each preset dimension; and selecting the candidate planning scheme with the highest comprehensive score as the optimal scheme. This method can integrate the opinions of multiple evaluation subjects, avoid the bias of a single evaluation model, and improve the reliability and robustness of the optimal scheme selection.

[0079] In this embodiment of the invention, after the initial defense plan is modified, a corresponding preset template is extracted from the plan template database based on template document generation technology. Entity recognition and attribute auto-fill technology are used to fill the key decision information in the optimal solution into the preset template to generate a structured plan document.

[0080] Specifically, after revising the initial defense plan to obtain an optimized defense plan, the following steps are used to generate a standardized, directly deployable plan document:

[0081] (1) Template extraction: Based on template document generation technology, such as OpenOffice, POI, etc., extract the preset template corresponding to the current disaster type from the emergency plan template database.

[0082] (2) Information filling: Entity recognition technology is used to identify the content that needs to be filled from the key decision information of the optimized defense plan, such as the name of the transfer route, the address of the resettlement point, the warning level, the start time, etc. Then, attribute automatic filling technology is used to fill these contents into the corresponding fields of the preset template.

[0083] (3) Format generation: The final structured contingency plan document is generated, supporting common formats such as Word and PDF, ensuring that the structure, style and content of the contingency plan document meet the relevant requirements for contingency plan preparation.

[0084] This method automatically converts the optimized defense plan into a document that can be directly executed and archived, reducing the workload of manual sorting and formatting while ensuring the standardization and consistency of the plan document.

[0085] In a specific embodiment of the present invention, considering the temporal nature of flood disaster processes, a contingency plan recommendation model is constructed based on Transformer (a deep learning model with a self-attention mechanism) to provide algorithmic support for the generation of flood disaster contingency plans. The overall structure is as follows: Figure 2 As shown in the figure, this diagram illustrates the model training and contingency plan generation process.

[0086] The Transformer model employs a self-attention mechanism to replace the sequence information extraction method of traditional recurrent neural networks. This mechanism assigns different weights to different parts of the input data based on their importance, significantly improving the model's ability to capture contextual information in long-term sequences. Based on this, a dynamic disaster preparedness model can intelligently analyze the disaster evolution process and influencing factors using spatiotemporal data under different disaster scenarios, formulate response strategies, and generate scenario-based contingency plans. The model architecture is as follows: Figure 3 As shown.

[0087] Considering the positional relationships involved in the transfer process, and the fact that the Transformer model lacks sequence information, positional encoding is added to provide this information. This is based on absolute positional encoding using sine and cosine:

[0088] ; ;

[0089] Where pos is the position number, i is the dimension index, and d is the embedding dimension.

[0090] The knowledge base stores various flood disaster emergency plans and monitoring information, providing favorable conditions for model training. The large model extracts activation conditions, protected objects, etc. from various plans as model inputs, and plans such as relocation and resettlement, early warning levels, and scheduling methods as data labels. The system extracts these data and uses supervised learning methods to train the model.

[0091] For data extracted from the knowledge base, preprocessing is required. Taking flash floods as an example, for each flash flood scenario, the following key attributes are extracted: scenario attribute A, including location, terrain, and water system structure; risk level R, assessed based on historical flood data; and contingency plan evacuation route P, including warning level, path points, and resettlement points. After preprocessing, the data structure can be represented as follows: Where A is the basic attribute of the i-th scenario, R is its corresponding risk level, P is the transfer route in the plan, and N is the number of training set samples.

[0092] The input text data includes basic attributes of flash flood scenarios and historical evacuation routes. To adapt to the model's input requirements, the text is first tokenized using a word-level segmentation strategy, dividing the sentence into several tokens, where W... i Represents the i-th token: Through the embedding matrix E, each token is mapped to a low-dimensional vector representation, thereby transforming the input scene text into a vector sequence. : In addition, due to the relative scarcity of historical disaster data in some regions, the model may lack generalization ability in specific scenarios. Data augmentation techniques can be used to generate more diverse training data, thereby increasing the diversity and representativeness of the training set.

[0093] In terms of contingency plan quality verification, multiple planning options are implemented by calling intelligent agents during the training and evaluation of contingency plan texts. This is coupled with a water conservancy professional model to obtain two candidate planning schemes: the flood inundation model calculation results and the contingency plan recommendation results under similar scenarios. The self-consistency evaluation method of the large model intelligent agent is used to analyze whether similar results can be produced. A simple majority voting strategy is used to evaluate several dimensions such as risk level, warning level, inundation range, and affected area. At the same time, the calculation results of the professional model are considered. Through multiple rounds of question and answer, the plan with the most votes is regarded as the optimal choice. Adjustments are made according to the content of the flood contingency plan in the final confirmed plan.

[0094] In terms of generating contingency plan texts, based on OpenOffice technology, the corresponding preset templates are extracted from the contingency plan template database according to the data returned by the model. Entity recognition and attribute auto-fill technology are used to update the content of the contingency plan model and automatically generate Word, PDF and other documents. At the same time, the structure, style and content of the generated contingency plan documents meet the relevant requirements for contingency plan preparation.

[0095] In this embodiment of the invention, the intelligent generation method for flood disaster prevention plans further includes generating emergency response plans based on a retrieval-enhanced generation model to supplement the optimized prevention plan. This method is particularly suitable for scenarios where a disaster has already occurred and specific response measures need to be generated quickly. The generation steps of the emergency response plan include an indexing stage, a retrieval stage, and a generation stage. In the indexing stage, flood disaster response cases, expert experience documents, and scheduling rule documents are segmented into text blocks, converted into vectors through an embedding model, and stored in a vector database. It can be understood that an inverted index can also be constructed to support keyword retrieval. In the retrieval stage, real-time disaster information is converted into query vectors through the same embedding model, and the most relevant text blocks are retrieved based on vector similarity and keyword matching algorithms. Specifically, in the retrieval stage, on the one hand, the similarity between the query vector and the text block vectors in the vector database is calculated to find the most semantically relevant text blocks; on the other hand, a keyword matching algorithm, such as the BM25 algorithm, is used to find document blocks containing relevant keywords. The two retrieval results are then fused to obtain the most relevant text blocks. During the generation phase, real-time hazard information and retrieved text blocks are input into a large language model to generate an emergency response plan. Specifically, the model inputs real-time hazard information and retrieved text blocks, designs appropriate prompts, and requires the model to generate an emergency response plan based on the retrieved content. For example, after retrieving historical cases of similar hazard responses, the model combines the current hazard description with specific response steps, required materials, and personnel configuration suggestions.

[0096] It is understandable that developing flood emergency response plans requires strong domain-specific knowledge. Generating emergency response plans solely through general-purpose models is limited and inaccurate. The retrieval-enhanced generative model provided in this invention accesses a private dataset through a general-purpose model, improving its reasoning ability in terms of domain-specific knowledge. This breaks away from the limitations of traditional generative models that rely solely on their own parameters to predict output. It introduces a retrieval strategy to obtain relevant information from external professional knowledge bases, and then uses this information to guide the model in generating more accurate and reasonable content, significantly improving the model's performance on various tasks such as question answering, dialogue systems, and text summarization. Typical paradigms of general-purpose models and retrieval-enhanced generative models are as follows: Figure 4 As shown.

[0097] This invention, based on the Research Retrieval Enhanced Generative Model (RAG), incorporates a professional flood disaster response case knowledge base into the emergency response plan generation process. Enhanced prompts are introduced during query processing to achieve accurate, diverse, and context-sensitive flood disaster response plans generated based on this large-scale flood disaster response case knowledge base. To address data security concerns, a private case knowledge base is constructed and deployed on a dedicated water resources network to prevent its dissemination on the internet, thereby enhancing the security of water resources data.

[0098] In this embodiment of the invention, the indexing stage segments documents such as disaster relief cases, expert experience, and scheduling rules into manageable blocks, converts them into vectors through an embedding model, and stores them in a vector database. The pre-retrieval processing stage improves the original query to enhance retrieval performance, addressing issues such as unclear query wording, language complexity, and ambiguity. The case retrieval stage establishes semantic similarity between the question and the document. Appropriate retrieval tools, including sparse, dense, and hybrid retrieval tools, are selected based on task requirements. Operations include constructing an inverted index and using ranking algorithms such as BM25 to sort the retrieval results. The retrieved information can be text fragments, keywords, or structured data. In the fusion stage, the retrieved information needs to be fused with the input text to provide rich context for the generative model. The information fusion method can be adjusted according to the specific task to maximize the role of the retrieved information. The case generation stage rapidly constructs and generates cases using large-scale natural language generation technology, generating corresponding response plans based on the input information about the flood disaster emergency.

[0099] In some embodiments of the present invention, after the emergency response plan is generated, a multi-objective optimization algorithm is used to rank and recommend the multiple candidate response plans with the objectives of minimizing response time, resource consumption, and flooding impact.

[0100] In this embodiment of the invention, by introducing retrieval-enhanced generation, the illusion problem of professional knowledge in general large models is effectively alleviated, the speed of knowledge updates is improved, the traceability of generated content is enhanced, and the hybrid retrieval strategy of vector similarity and keyword matching can simultaneously recall semantically similar and precisely matched cases, thereby improving the accuracy and relevance of generated content.

[0101] This invention also provides an intelligent flood disaster prevention plan generation system based on inference and feedback, used to execute the above-mentioned intelligent flood disaster prevention plan generation method. The intelligent flood disaster prevention plan generation system includes a plan knowledge construction module, an initial plan generation module, a multi-candidate scheme generation module, a feedback inference module, a self-consistency evaluation module, and a knowledge update module. The contingency plan knowledge construction module is used to acquire multi-source data related to flood disasters, extract knowledge using a large model, and form a structured knowledge base. The initial contingency plan generation module has a built-in initial contingency plan generation model, used to generate an initial defense plan based on flood disaster scenario information. The multi-candidate scheme generation module is used to generate multiple candidate contingency plan planning schemes based on the knowledge base and the initial defense plan. The mutual feedback inference module is used to input each candidate contingency plan planning scheme into a water conservancy professional model to perform flood evolution inference and obtain the inference results corresponding to each candidate contingency plan planning scheme. The self-consistency evaluation module is used to evaluate the consistency between each inference result and the preset expected result on multiple preset dimensions, select the optimal scheme from multiple candidate contingency plan planning schemes based on the evaluation results, and revise the initial defense plan based on the optimal scheme to obtain an optimized defense plan. The knowledge update module is used to store the optimized defense plan in the knowledge base as a historical case for subsequent contingency plan generation. In a preferred embodiment, the knowledge update module is also used to use the optimized defense plan and the inference results of the optimal scheme as training data to iteratively optimize the initial contingency plan generation model.

[0102] In the intelligent flood disaster prevention plan generation system provided by this invention, the plan knowledge construction module outputs a knowledge base; the initial plan generation module reads the knowledge base based on scenario information and generates an initial prevention plan; the multi-candidate scheme generation module generates multiple candidate plan planning schemes based on the initial prevention plan and the knowledge base; the mutual feedback simulation module calls a water conservancy professional model and returns simulation results; the self-consistency evaluation module selects the optimal scheme and corrects it based on the simulation results; and the knowledge update module feeds the results back to the knowledge base and the initial plan generation model. The modules communicate with each other through API interfaces and can be deployed using a microservice architecture.

[0103] In some embodiments, the hydraulic engineering model includes a hydrological model and a hydrodynamic model; it is understood that other professional models, such as dam-break models and sediment transport models, may also be added as needed.

[0104] In some embodiments, the self-consistency evaluation module includes a majority voting evaluation unit, which is used to evaluate the consistency of the deduction results corresponding to each candidate contingency plan in terms of risk level, warning level, flood range, and affected area, and select the optimal plan. Specifically, multiple agents independently evaluate the performance of each candidate contingency plan in each dimension, and the majority voting evaluation unit counts the votes in each dimension and determines the plan with the best overall performance as the final selection.

[0105] In this embodiment of the invention, the intelligent flood disaster prevention plan generation system further includes an emergency response module. The emergency response module comprises a vector database, a retrieval unit, and a large model generator. The vector database stores the embedding vectors of case text blocks, supporting efficient vector similarity retrieval. The retrieval unit performs vector similarity retrieval and keyword matching retrieval, returning the most relevant text blocks based on the query terms. The large model generator integrates the retrieval results and generates an emergency response plan, specifically by integrating the retrieved text blocks and real-time hazard information, and calling a large language model to generate the emergency response plan.

[0106] The emergency response module provided by this invention serves as an enhancement and supplement to the aforementioned intelligent generation system for flood disaster prevention plans. It can quickly provide response measures when a dangerous situation occurs, forming a complementary solution.

[0107] This invention also provides an electronic device, including a processor and a memory, wherein the memory stores machine-executable instructions that can be executed by the processor, and the processor can execute the machine-executable instructions to implement the above-mentioned intelligent generation method for flood disaster prevention plans.

[0108] In this invention, the processor contains a kernel, which retrieves the corresponding program units from memory. One or more kernels can be configured.

[0109] The memory may include non-permanent memory in a computer-readable storage medium, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0110] This invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-described method for intelligently generating flood disaster prevention plans.

[0111] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0112] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0113] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0114] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0115] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0116] Memory may include non-persistent memory in computer-readable storage media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable storage medium.

[0117] Computer-readable storage media include both permanent and non-permanent, removable and non-removable media that can store information by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable storage media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0118] It should also be noted that the term "comprising," or any other variation thereof, is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0119] The foregoing has shown and described the basic principles, main features, and characteristics of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A method for intelligently generating flood disaster prevention plans based on inductive feedback, characterized in that, Includes the following steps: Construct a knowledge base for flood disaster prevention plans, which includes at least disaster scenario attributes, historical plan texts, risk levels, evacuation routes, early warning levels, and dispatch rules; Based on real-time or forecasted flood disaster scenario information, an initial defense plan is generated by calling a deep learning-based initial plan generation model. For the same flood disaster scenario, multiple candidate contingency planning schemes are generated based on the knowledge base and the initial defense plan; The multiple candidate contingency plans are input into the water conservancy professional model to perform flood evolution simulation, and the simulation results corresponding to each candidate contingency plan are obtained. The water conservancy professional model includes at least a hydrological model and a hydrodynamic model. Based on the multi-agent self-consistency evaluation method, the consistency evaluation of each inference result and the preset expected result is carried out on multiple preset dimensions. Based on the evaluation results, the optimal solution is selected from multiple candidate contingency planning schemes. The initial defense plan is then modified according to the optimal solution to obtain the optimized defense plan. The optimized defense plan is stored in the knowledge base as a historical case for the generation of subsequent plans.

2. The intelligent generation method for flood disaster prevention plans according to claim 1, characterized in that, The steps for constructing the knowledge base include: Acquire multi-source data related to flood disasters, use large models for knowledge extraction, and extract entities, attributes, and relationships from the contingency plans; The content of each type of flood control plan is decomposed into a structure, a plan template is constructed and stored in the database, and the plan template includes dynamically populated attribute fields and fixed text paragraphs; Extract the activation conditions, protected objects, relocation and resettlement methods, early warning levels, and scheduling methods of the contingency plan as key attributes, and associate and store these key attributes with the corresponding contingency plan templates.

3. The intelligent generation method for flood disaster prevention plans according to claim 1, characterized in that, The initial contingency plan generation model is a deep learning model based on a self-attention mechanism, used to capture the temporal features and contextual relationships of disaster scenarios and generate a sequence of contingency plan texts; The initial contingency plan generation model is based on the Transformer architecture. Its training process includes: extracting historical disaster scenario attributes and risk levels from the knowledge base as model inputs, extracting evacuation routes, warning levels, and scheduling methods from the corresponding historical contingency plans as labels, and training it using a supervised learning approach; when generating the initial defense plan, an autoregressive approach is used to predict the output sequence word by word.

4. The intelligent generation method for flood disaster prevention plans according to claim 1, characterized in that, The methods for generating multiple candidate contingency plans include: perturbing the key parameters in the initial defense plan with multiple sets of perturbations, or retrieving multiple historical contingency plans with similar scenarios from the knowledge base for adaptation and adjustment; The multi-agent self-consistency evaluation method includes: scoring the deduction results of each candidate plan scheme on each preset dimension, and then using a majority voting or weighted voting strategy to integrate the scoring results of each preset dimension, and taking the candidate plan scheme with the highest comprehensive score as the optimal scheme.

5. The intelligent generation method for flood disaster prevention plans according to claim 1, characterized in that, After revising the initial defense plan, a corresponding preset template is extracted from the plan template database based on template document generation technology. Entity recognition and attribute auto-fill technology are then used to fill the preset template with the key decision information from the optimal solution, generating a structured plan document.

6. The intelligent generation method for flood disaster prevention plans according to claim 1, characterized in that, It also includes generating emergency response plans based on a retrieval-enhanced generative model to supplement the optimized defense plan. The steps for generating the emergency response plan include: Indexing phase: Flood disaster response cases, expert experience documents, and scheduling rule documents are segmented into text blocks, converted into vectors through an embedding model, and stored in a vector database; Retrieval phase: Real-time emergency information is converted into query vectors using the same embedding model, and the most relevant text blocks are retrieved based on vector similarity and keyword matching algorithms; Generation phase: Real-time emergency information and retrieved text blocks are input into a large language model to generate an emergency response plan; After the emergency response plan is generated, a multi-objective optimization algorithm is used to rank and recommend the multiple candidate response plans, with the objectives of minimizing response time, resource consumption, and flooding impact.

7. A flood disaster prevention plan intelligent generation system based on inductive feedback, used to execute the flood disaster prevention plan intelligent generation method according to any one of claims 1-6, characterized in that, The intelligent generation system for flood disaster prevention plans includes: The contingency plan knowledge construction module is used to acquire multi-source data related to flood disasters, extract knowledge using a large model, and form a structured knowledge base; The initial contingency plan generation module has a built-in initial contingency plan generation model, which is used to generate initial defense plans based on flood disaster scenario information; A multi-candidate solution generation module is used to generate multiple candidate solution planning schemes based on the knowledge base and the initial defense plan; The feedback simulation module is used to input each candidate contingency plan into the water conservancy professional model to perform flood evolution simulation and obtain the simulation results corresponding to each candidate contingency plan. The self-consistency evaluation module is used to evaluate the consistency between each simulation result and the preset expected result across multiple preset dimensions. Based on the evaluation results, it selects the optimal solution from multiple candidate contingency plans and modifies the initial defense plan according to the optimal solution to obtain an optimized defense plan. The knowledge update module is used to store the optimized defense plan into the knowledge base, which serves as a historical case for the generation of subsequent plans.

8. The intelligent generation system for flood disaster prevention plans according to claim 7, characterized in that, The water conservancy professional model includes a hydrological model and a hydrodynamic model; The self-consistency assessment module includes a majority voting assessment unit, which is used to conduct consistency assessments on the deduction results corresponding to each candidate contingency plan in terms of risk level, warning level, flooding range, and affected area, and select the optimal plan. The intelligent generation system for flood disaster prevention plans also includes an emergency response module, which includes a vector database, a retrieval system, and a large model generator. The vector database is used to store the embedding vectors of case text blocks, the retrieval system is used to perform vector similarity retrieval and keyword matching retrieval, and the large model generator is used to fuse the retrieval results and generate an emergency response plan.

9. An electronic device, characterized in that, The system includes a processor and a memory, wherein the memory stores machine-executable instructions that can be executed by the processor, and the processor can execute the machine-executable instructions to implement the intelligent generation method for flood disaster prevention plans as described in any one of claims 1-6.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the intelligent generation method for flood disaster prevention plans as described in any one of claims 1-6.

Images