A hydrological model coupling method based on a flood control large model

Abstract

The application discloses a hydrological model coupling method based on a flood control large model, belongs to the field of hydrological models and flood control technologies, and comprises the following steps: collecting and sorting massive multi-source water conservancy industry data, and constructing a flood control knowledge base; based on a general large model, carrying out supervision fine tuning based on the water conservancy knowledge base, and constructing a flood control large model for a water conservancy flood control and disaster reduction scene; proposing a hydrological model decomposition method based on a physical mechanism, utilizing a knowledge graph to construct a coupling relationship between each calculation unit and hydrological knowledge, realizing automatic decomposition of a model calculation process and automatic association of business data through a flood control large model agent, and calling each calculation unit of the hydrological model and the corresponding knowledge base to realize coupling calculation, and realizing automatic calibration of hydrological model parameters. The application aims to propose a method for realizing effective connection and coupling of flood control knowledge and the hydrological model based on the flood control large model, achieve the purpose of intelligent optimization of hydrological model parameters, continuously improve the accuracy of hydrological prediction, and scientifically support the four-prediction application of flood control.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of hydrological models and flood control, and particularly relates to a hydrological model coupling method based on a flood control large model. BACKGROUND

[0002] Intelligent means for flood control are increasingly becoming an important part of flood control and disaster reduction work. Traditional flood forecasting methods are often based on a single hydrological model, and mainly use artificial verification to determine parameters. The parameters are closely related to factors such as region, river basin, soil, vegetation, and rain intensity, and the accuracy of the model also varies from region to region. In order to reduce the manual participation in the model calibration process and cross-region transplantation and reuse, as well as continuously improve the accuracy of the model, a coupling method is needed that can integrate multiple hydrological models and use flood control and disaster reduction knowledge to intelligently drive model calculation, realize intelligent parameter optimization of hydrological models, improve the accuracy of hydrological forecasting, and scientifically support the flood control "four predictions" scenario application. SUMMARY

[0003] The purpose of the present application is to provide a hydrological model coupling method based on a flood control large model, which realizes intelligent parameter optimization of hydrological models and improves the accuracy of hydrological forecasting.

[0004] The technical solution adopted by the present application to solve its technical problems is:

[0005] A hydrological model coupling method based on a flood control large model, comprising the following steps:

[0006] S1, collect and organize massive multi-source water conservancy industry data, use a full-factor data extraction and fusion method to build a flood control knowledge base, and establish a disaster scenario library and a whole-process defense knowledge base for different flood types;

[0007] S2, based on a general large language model, build a special supervised fine-tuning data set and training strategy, carry out professional pre-training and fine-tuning with the water conservancy knowledge base, and build a flood control large model for water conservancy flood control and disaster reduction scenarios;

[0008] S3, decompose the important links of the conceptual hydrological model evapotranspiration and runoff simulation of hydrological cycle, comprehensively consider the calculation factors, calculation methods and calculation range, and obtain the smallest independently calculable unit;

[0009] S4, based on the hydrological model domain knowledge graph ontology and concept relationship, use the knowledge graph to build the coupling relationship between each calculation unit of the hydrological model and the hydrological knowledge, and establish the corresponding knowledge representation for each calculation unit;

[0010] S5, realize automatic decomposition of the model calculation process and automatic association of business data through the flood control large model agent, and perform coupling calculation based on the flood control large model calling each calculation unit of the hydrological model and its corresponding knowledge base.

[0011] Further techniques of the present application:

[0012] Preferably, S3 is specifically:

[0013] 1) The evapotranspiration unit divides the hydrological response unit according to the climate of the watershed, land use and soil type, inputs the measured water evaporation and soil moisture through the evapotranspiration calculation model, and calculates the evapotranspiration of the watershed;

[0014] 2) The runoff unit considers the precipitation, soil moisture and vegetation coverage in the watershed, divides the grid or sub-watershed, calculates the water quantity converted from precipitation into runoff on the ground or in the soil according to the input measured rainfall and the calculated evapotranspiration result, and respectively substitutes into the runoff source equation to calculate the ground runoff, soil runoff and underground runoff according to the path of runoff formation;

[0015] 3) The confluence unit considers the topography, landform and flow velocity in the watershed, divides the sub-watershed or grid, respectively calculates the runoff flow process of different water sources, obtains the outlet flow process of each unit area, and then respectively passes the outlet flow process of each unit through the river network confluence unit to the watershed outlet section through river flood routing to obtain the flow process of the watershed outlet.

[0016] Preferably, in S4, the coupling relationship between the hydrological knowledge and the calculation unit is established through the knowledge graph triple, the calculation unit provides the model and the parameter, and the hydrological knowledge provides the coupling relationship and the data required for calculation;

[0017] The coupling relationship between the calculation unit and the hydrological knowledge is represented in the form of RDF triple.

[0018] Preferably, the specific steps of S5 are:

[0019] S51, a large model reasoning constraint is provided through a knowledge graph flood control relationship database, a hydrological model algorithm scheduling knowledge chain is constructed, a large model driven model calling calculation after reasoning logic strengthening is carried out;

[0020] S52, a multi-path calculation optimization comparison is realized through the perception, planning, action and execution memory ability of the intelligent agent, and the model algorithm calling path is planned;

[0021] S53, according to the planned hydrological model calling path, the service request, function method and data corresponding to the calculation unit are called in turn, the hydrological prediction results are output through result calculation, and the knowledge driven model calculation is realized;

[0022] S54, the calculation results of the hydrological model are integrated, and the flood control knowledge base is updated.

[0023] Advantages of the present application:

[0024] The method of the present application carries out large model reasoning logic reinforcement through the water conservancy knowledge graph, carries out coupling calculation on the basis of the intelligent calling of each calculation unit of the flood control large model and the corresponding knowledge base of the hydrological model, solves the problems of high artificial dependence of hydrological model parameter calibration and difficulty in continuously improving the prediction accuracy, realizes knowledge intelligent driving model for business decision-making, and provides a scientific basis for flood control and disaster reduction. BRIEF DESCRIPTION OF DRAWINGS

[0025] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0026] Figure 1 The flowchart of the hydrological model coupling method based on the flood control large model.

[0027] Figure 2 The flood control knowledge graph construction process.

[0028] Figure 3 The hydrological model calculation unit and hydrological knowledge coupling linkage technology line diagram. DETAILED DESCRIPTION

[0029] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be further described in detail below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0030] Please refer to Figure 1 The flowchart of the hydrological model coupling method based on the flood control large model includes the following steps:

[0031] S1, collect and organize massive multi-source water conservancy industry data, adopt a full-factor data extraction and fusion method to construct a flood control knowledge base, and establish a disaster scene library and a whole-process defense knowledge base for different flood types.

[0032] S2, based on a general large language model, construct a special supervised fine-tuning data set and training strategy, carry out professional pre-training and fine-tuning with the water conservancy knowledge base, and construct a flood control large model for water conservancy flood control and disaster reduction scenarios.

[0033] S3, decompose important links of the conceptual hydrological model evapotranspiration, runoff generation and other simulation of hydrological cycle, comprehensively consider calculation factors, calculation methods and calculation range, and obtain the smallest independently calculable unit.

[0034] S4, based on the hydrological model field knowledge graph ontology and concept relationship, the knowledge graph is used to construct the coupling relationship between each calculation unit of the hydrological model and the hydrological knowledge, and the corresponding knowledge representation is established for each calculation unit.

[0035] S5, the model calculation process is automatically decomposed and the business data is automatically associated through the flood control big model agent, and the coupling calculation of each calculation unit of the hydrological model and the corresponding knowledge base is realized based on the flood control big model.

[0036] The specific steps of S1 are as follows, and the implementation process of the flood control knowledge graph is as shown in Figure 2

[0037] S11, the water conservancy industry data includes five categories of basic data, monitoring data, geographic spatial data, business data and cross-industry shared data.

[0038] S12, the water conservancy data source also includes unstructured or semi-structured text, image, table and other multi-modal data published externally from water conservancy encyclopedias, policy documents, electronic journals, standards and specifications.

[0039] S13, the collected and sorted water conservancy industry data is preprocessed, including text preprocessing, data cleaning and multi-source data alignment. The non-text content of the file is processed, and the redundant spaces and line breaks are removed. The file text is converted into standard format text data. Align the same entities in different data sources to ensure data consistency, avoid repeated extraction and information redundancy, improve the efficiency and accuracy of knowledge extraction, and further improve the construction quality of the knowledge graph.

[0040] S14, structured data directly uses D2R (Database to RDF) technology to convert database information into RDF triple information. RDF triple is generally represented as: G={E,R,F}, wherein G represents a knowledge graph, E represents an entity set {e1, e2,…, e E}, the entity e is the most basic component element in the knowledge graph, which is represented by a node in the graph, R represents a relationship set {r1, r2,…, r R}, the relationship r is the edge in the knowledge graph, which represents the connection between different entities. F represents a fact set {f1, f2,…, f F}, each fact f is defined as a triple (h, r, t) ∈ f. Wherein, h represents the head entity, r represents the relationship, and t represents the tail entity.

[0041] ​S15, unstructured and structured data, using large model end-to-end atlas automatic construction technology, text classification, automatic labeling, knowledge extraction and other natural language processing work, in-depth mining of implicit relationships in massive data. A multi-modal pre-training model is used to learn the semantic consistency between different modalities to ensure that text, images and table data can be represented in the same feature space, maintaining information exchange and semantic consistency between different modalities. Finally, the fused multi-modal features are represented as knowledge and stored in the vector knowledge base.

[0042] S16, the ontology model of the knowledge graph includes forecast schemes, historical scenarios, expert experience, business rule base and preplans. The knowledge graph formed by extraction and fusion includes a flood control ontology base, a semantic relationship base and an application scenario base. The flood control ontology base includes flood control object ontology and flood control measure ontology. The object ontology is divided into natural objects, engineering objects and management objects. The measure ontology is divided into risk, perception, analysis, simulation, early warning and influence. The semantic relationship base includes spatial relationship, time relationship, logical relationship, organizational relationship and physical relationship of flood control objects. The ontology objects and their attributes are linked through semantic relationships. The entity set after knowledge fusion is mapped to nodes, and the relationships in the relationship triplets are mapped to edges and stored in a Neo4j database. Visualization is performed through a graph database.

[0043] S17, the knowledge graph is logically divided into two layers of ontology layer and instance layer, and the ontology layer is located above the instance layer. The ontology layer includes class hierarchy, relationship, entity type and attribute definition. The ontology model is used to constrain the types of entities and the relationships between entities. The instance layer is composed of a series of data. Entity, relationship extraction triplets are performed through the flood control large model to complete the construction of the instance layer.

[0044] The S2 step includes:

[0045] S21, a supervised fine-tuning (SFT) training data set suitable for water conservancy knowledge is constructed, data types that are most significant for improving the effect of flood control tasks are mined, and a diversified data sampling strategy is adopted to realize efficient data construction.

[0046] S22, the training stability and convergence speed of the model are improved by optimizing the reinforcement learning algorithm through discount factor adjustment and sampling method improvement. The corresponding training strategy is constructed in combination with the flood control and disaster reduction task scene to ensure that the model can stably learn and align the result feedback.

[0047] S23, the prompt (prompt word engineering) technology of the large model is used to design effective prompt words to guide the large model to better understand water conservancy professional terms and special knowledge. The chapter recall technology is used to improve the knowledge recall accuracy.

[0048] The S3 step comprises:

[0049] S31, the evapotranspiration unit divides hydrological response units according to hydrological characteristics such as basin climate, land use and soil type, inputs measured water surface evaporation and soil moisture into the evapotranspiration calculation model, and calculates the evapotranspiration of the basin.

[0050] S32, the runoff unit divides grids or sub-basins considering factors such as precipitation, soil moisture and vegetation coverage in the basin. According to the input measured rainfall and the calculated evapotranspiration result, the water quantity of the precipitation converted into runoff on the ground or in the soil is calculated, and according to the path of the runoff formation, the ground runoff, interflow and groundwater runoff are calculated by substituting into the water source equation respectively.

[0051] S33, the confluence unit divides sub-basins or grids considering factors such as topography, landform and flow velocity in the basin, respectively calculates the runoff flow process of different water sources, and obtains the outlet flow process of each unit area. Then, the river network confluence unit respectively carries out the river channel flood routing to the outlet section of the basin according to the outlet flow process of each unit, and obtains the flow process of the outlet of the basin.

[0052] Figure 3 It is a hydrological model calculation unit and hydrological knowledge coupling linkage technical line diagram, involving steps S4 and S5.

[0053] The S4 step comprises:

[0054] S41, the coupling relationship between hydrological knowledge and the calculation unit is established through the triplets of the knowledge graph, the calculation unit provides the model and the parameters, and the hydrological knowledge provides the coupling relationship and the data required for calculation.

[0055] S42, the coupling relationship between the calculation unit and the hydrological knowledge is represented in the form of RDF triplets (entity-relation-entity). The triplet construction paradigm of the hydrological forecast knowledge graph is represented as: [

[0057] ("hydrology", "forecast", "distributed hydrological model type");

[0058] ("distributed hydrological model", "start", "runoff model");

[0059] ("runoff model", "dispatch", "S({method:fun(v)})");

[0060] ("runoff model", "end", "confluence model");

[0061] ("confluence model", "dispatch", "F(x1,x2,x3,…,k)");

[0062] ("distributed hydrological model", "end", "F(x)"); ]

[0064] Convention S(*) is a service request, F(*) is a function method.

[0065] S43, establish the interface of the coupling linkage of the model parameters of each calculation unit.

[0066] S44, extract the entities and relationships required by the calculation unit through the interface, and then match the data in the knowledge base through the entities and relationships.

[0067] The S5 step includes:

[0068] S51, provide large model reasoning constraints through the flood control relationship database of the knowledge graph, construct a hydrological model algorithm scheduling knowledge chain, and carry out large model driven model calling calculation after reasoning logic reinforcement.

[0069] S52, realize multi-path calculation optimization comparison through the perception, planning, action and execution memory ability of the agent, and plan the model algorithm calling path.

[0070] S53, according to the planned hydrological model calling path, the service request, function method and data corresponding to the calculation unit are called in turn, the hydrological forecast results are output through the result calculation, and the knowledge driven model calculation is realized.

[0071] S54, integrate all the calculation results of the hydrological model, and update the flood control knowledge base.

[0072] The embodiment is based on the clustering of similar environmental characteristics of a basin, constructs a deep learning algorithm of a flood control large model driven hydrological model automatic rate parameter, realizes intelligent parameter optimization of the hydrological model, and improves the accuracy of hydrological prediction.

[0073] The specific steps are:

[0074] Collect and arrange meteorological and hydrological data of different regions under similar environmental conditions.

[0075] Construct a multi-target parameter cluster (key hydrological parameter) under different rainfall conditions, and the parameter cluster is defined as:

[0076] W=(K,B,C,UM,IM,WM,LM,…,CI)

[0077] W is a parameter cluster, K is an evapotranspiration conversion coefficient, B is a water storage capacity distribution curve index, C is a deep evaporation coefficient, UM is an average upper water storage capacity, IM is an impervious area ratio, WM is an average water storage capacity, LM is an average middle water storage capacity, and CI is a soil infiltration coefficient.

[0078] A deep learning algorithm is established for rating influence factors of hydrological process under complex underlying surface conditions by coupling achievements of a flood control large model and a hydrological prediction model, so as to realize automatic rating of hydrological model parameters. By constraining total flood volume, flood peak value and flood peak time qualified rate in calculation achievements of the hydrological model, initial values of the above parameter groups are assigned on the basis of known characteristic attributes. By using the deep learning algorithm, a solution algorithm for parameter group weights under the constraint condition of the hydrological process is constructed, and convergence of each parameter weight value is realized through intelligent trial iteration.

[0079] The construction condition of the deep learning algorithm is represented as: Q = f (w1x2, w2x2, w3x3,..., w n x n )

[0080] Q = {q1, q2,..., q n} is hydrological process data under the constraint condition, w1-w n is parameter group characteristic index data, x1-x n is a solution parameter weight, and f (*) is an algorithm solution objective function.

[0081] The above is only a preferred embodiment of the present application, and does not limit the present application in any form; any person skilled in the art can make many possible changes and modifications to the technical solution of the present application, or modify equivalent embodiments, without departing from the scope of the technical solution of the present application, by using the disclosed methods and technical contents. Therefore, any simple modification, equivalent replacement, equivalent change and modification of the above embodiments according to the technical essence of the present application, which does not depart from the technical solution of the present application, still belongs to the protection scope of the technical solution of the present application.

Claims

1. A hydrological model coupling method based on a large flood control model, characterized in that, Includes the following steps: S1. Collect and organize massive amounts of multi-source water conservancy industry data, and use the full-element data extraction and fusion method to build a flood control knowledge base. Establish a disaster scenario database and a full-process defense knowledge base for different flood types. S2. Based on a general large language model, construct a dedicated supervised fine-tuning dataset and training strategy, and conduct professional pre-training and fine-tuning using a water conservancy knowledge base to build a large flood control model for water conservancy flood control and disaster reduction scenarios. S3. Decompose the key links of the hydrological cycle simulation of evapotranspiration and runoff generation in the conceptual hydrological model, and comprehensively consider the calculation factors, calculation methods and calculation range to obtain the smallest independently calculable unit. S4. Based on the ontology and concept relationships of the knowledge graph in the field of hydrological models, the coupling relationship between each computational unit of the hydrological model and hydrological knowledge is constructed using the knowledge graph, and a corresponding knowledge representation is established for each computational unit. S5. The model calculation process is automatically decomposed and business data is automatically associated through the flood control big model intelligent agent. Based on the flood control big model, the calculation units of the hydrological model and their corresponding knowledge bases are called for coupled calculation. The specific steps of S5 are as follows: S51. Provide large-scale model reasoning constraints through the knowledge graph flood control relationship database, construct a knowledge chain for hydrological model algorithm scheduling, and carry out large-scale model-driven model call calculation after strengthening reasoning logic; S52. Through the agent's perception, planning, action, and execution memory capabilities, multi-path computation optimization comparison is achieved, and the algorithm call paths of each model are planned. S53. According to the planned hydrological model call path, call the service requests, function methods and data corresponding to the calculation unit in sequence, and output the hydrological forecast results through result aggregation to realize knowledge-driven model calculation. S54. Integrate all calculation results from the hydrological model and update the flood control knowledge base.

2. The hydrological model coupling method based on a large flood control model as described in claim 1, characterized in that, S3 specifically refers to: 1) The evapotranspiration unit is divided into hydrological response units based on the watershed climate, land use and soil type. The watershed evapotranspiration is calculated by inputting the measured water surface evaporation and soil moisture through the evapotranspiration calculation model. 2) The runoff generation unit considers precipitation, soil moisture, and vegetation cover within the watershed, divides it into grids or sub-watersheds, and calculates the amount of precipitation that is converted into runoff on the surface or in the soil based on the input measured rainfall and calculated evapotranspiration results. According to the runoff formation path, the surface runoff, interflow and groundwater runoff are calculated by substituting them into the water source equation respectively. 3) The confluence unit takes into account the topography, geomorphology, and water flow velocity within the watershed, and divides it into sub-watersheds or grids. The runoff flow process of different water sources is calculated separately to obtain the outflow process of each unit area. Then, by using the river network confluence unit, the outflow process of each unit is calculated to the outlet section of the basin through river flooding, and the outflow process of the basin outlet is obtained.

3. The hydrological model coupling method based on a large flood control model as described in claim 1, characterized in that, In S4, the coupling relationship between hydrological knowledge and computing units is established through knowledge graph triples. The computing units provide models and parameters, while hydrological knowledge provides coupling relationships and data required for computation. The coupling relationship between computational units and hydrological knowledge is represented using RDF triples.

Images