Digital twin watershed scenario modeling method based on water cycle dynamic knowledge graph

Abstract

The present invention discloses a method for modeling digital twin river basin scenarios based on a dynamic knowledge graph of the water cycle. The method comprises the following steps: constructing a dynamic knowledge graph of the water cycle based on remote sensing inversion data of the water cycle within the watershed and ground station monitoring data; building a multi-scale nested digital twin framework of the watershed-sub-watershed-aggregate unit-minimum granularity unit, carrying out data spatiotemporal distribution and model configuration on a hierarchical grid, and utilizing the knowledge graph to achieve self-correlation between data and models, thereby realizing digital twin river basin scenario modeling. The present invention uses the dynamic knowledge graph of the water cycle as the knowledge base, constructs a multi-scale nested digital twin framework, couples data with water cycle modules, implements a data organization mechanism that is consistent in time and space, and constructs a high-fidelity mirror digital river basin. This provides technical support for subsequent high-precision hydrological forecasting, flood control scheduling, and simulation prediction based on parallel digital river basins.

Description

Technical Field

[0001] The present invention belongs to the field of digital twin technology for small and medium-sized watersheds, and specifically relates to a digital twin watershed scene modeling method based on a dynamic knowledge graph of water cycle. Background Art

[0002] Small and medium-sized river basins have become a major source of flood-related losses in my country and a weak link in disaster prevention and mitigation efforts. Due to a lack of underlying surface data and inadequate rainfall monitoring, flood forecast accuracy and forecast horizon remain bottlenecks. As water conservancy enters a new stage of development, the Ministry of Water Resources has prioritized the development of smart water conservancy initiatives, and the construction of digital twin river basins has become a core task and goal. By leveraging digital twin technology to create a mirrored digital river basin that mirrors and interacts with the physical river basin, it can effectively enhance flood forecasting and early warning capabilities for small and medium-sized rivers. However, the application of digital twin technology in the water conservancy sector is still in its infancy. While existing business management systems possess a wealth of monitoring data, they still fall short of the comprehensive, full-process management requirements of digital twin river basins. The main problem is that the acquisition accuracy, processing speed, utilization, and update frequency of remote sensing information and ground-based monitoring data from the basin are clearly insufficient. There is no effective theory and technology to organize this information and use it to characterize the elements of the entire basin. At the same time, most hydrological forecasting and hydrodynamic models used in the industry are not universal enough and too inflexible. In other words, the models are not universal, and the models still need to be recompiled when they are transplanted to other problems. In addition, they cannot fully utilize the underlying surface information of the basin, and the applicability and accuracy of the models need to be improved urgently. Therefore, it is necessary to explore and invent a new digital twin watershed construction method. This method can study the temporal and spatial consistency of data organization mechanism for multi-scale water conservancy objects, modularize the water cycle model, and explore the relationship between water conservancy objects, unit characteristics, and water cycle modules. This method can construct a high-fidelity mirror digital watershed, thereby realizing the full-element, full-process, and fine-grained scenario modeling of the digital twin watershed, supporting advanced simulation and rehearsal of the physical watershed, and solving the technical difficulties of applying digital twin technology in the water conservancy field. Summary of the Invention

[0003] Purpose of the invention: The purpose of the present invention is to provide a digital twin watershed scene modeling method based on the dynamic knowledge graph of the water cycle, so as to realize the full-element, full-process and fine-grained scene modeling of the digital twin watershed.

[0004] Technical solution: The present invention provides a digital twin watershed scenario modeling method based on a water cycle dynamic knowledge graph, comprising the following steps:

[0005] (1) Constructing a dynamic knowledge graph of the water cycle: Based on the remote sensing inversion data of the water cycle in the basin and combined with the ground station monitoring data, various structured and unstructured basic data are collected. Through ontology construction, object relationship extraction, and knowledge fusion, the relationship between water conservancy business, water conservancy objects, hydrological models and data is formed, thereby realizing the construction of a dynamic knowledge graph of the water cycle;

[0006] (2) Build a multi-scale nested digital twin framework to carry out data spatiotemporal distribution and model configuration on the grid level by level;

[0007] (3) Utilize the dynamic knowledge graph of the water cycle to achieve self-association and self-matching of data and models, complete the self-organization of multi-scale water cycle data and models, and realize the scenario modeling of all elements and processes of the digital twin basin.

[0008] Furthermore, the step (1) includes the following steps:

[0009] (11) The graph is abstracted into a five-tuple structure, defined as:

[0010] G wa =(C, P, R, A, I)

[0011] Among them, C is the concept set of water conservancy objects, P is the attribute set of related concepts, R is the relationship set, I is the instance set, and A is the stable relationship between concepts. The relationship between axioms and concepts in the water cycle dynamic knowledge graph is:

[0012]

[0013] Among them, A1 is a specific axiom, and the relationship between C1, C2 and C3 constitutes axiom A1, which plays a constraint role in the instantiation process of the three concepts;

[0014] (12) Adding time-varying and spatial characteristics to P and R is done by incorporating the nine-intersection model. For two instances of I, A and B, the boundary (δA, δB), interior (A 0 , B 0 ), and external (A -1 , B -1 ) The intersection of the three is realized by the nine-tuple, which represents the topological relationship between instances A and B, as shown in the following matrix:

[0015]

[0016] (13) Design the model layer and instance layer of the graph. Each node in the graph corresponds to an object directory table and an object basic information table, thus constructing a complete water cycle dynamic knowledge graph.

[0017] Furthermore, the step (2) includes the following steps:

[0018] (21) Digital extraction of river networks based on basic hydrological and geographical data of small and medium-sized river basins;

[0019] (22) With the river network as the framework and the goal of characterizing the evolution of rainfall, runoff and confluence in the water cycle, grid aggregation is carried out within the basin based on the minimum granularity unit grid. Strip aggregation is carried out within the river channel based on the known cross-section. In the non-river channel area, quadrilateral grid aggregation is carried out based on the topography, hydrological and meteorological characteristics and the homogeneity of the underlying surface conditions to form aggregation units.

[0020] (23) Combined with the division of sub-basins, a multi-scale nested digital twin framework of the sub-basin watershed with the smallest granularity unit aggregation unit is finally formed;

[0021] (24) Data sorting and preprocessing are carried out for typical flood data in the basin, and the attribute data of water conservancy objects in the water cycle dynamic knowledge graph are based on the structure of rainfall-evapotranspiration-flow generation-runoff-slope confluence-river confluence that represents the hydrological model. These data are respectively spread on the four-layer grid of the multi-scale nested digital twin framework of basin-sub-basin-aggregation unit-minimum granularity unit.

[0022] Furthermore, the implementation process of using the water cycle dynamic knowledge graph to realize the self-association between data and models in step (3) is as follows:

[0023] Call the model execution agent, based on the water cycle dynamic knowledge graph, retrieve the parameters of the hydrological model in the graph, and use the correlation between the hydrological model parameters and the watershed geomorphological characteristics, soil type and vegetation cover data to call the agent to obtain the parameter values ​​required for the initial calculation of the hydrological model;

[0024] For missing data, the data in the water cycle dynamic knowledge graph is supplemented;

[0025] Data processing is performed on existing data that meets the scale requirements to obtain all parameter values ​​required for hydrological model calculations, and the parameters of the hydrological model are associated with the parameter values ​​in the water cycle dynamic knowledge graph.

[0026] Furthermore, the process of realizing the self-matching of data and model by using the water cycle dynamic knowledge graph in step (3) is as follows:

[0027] Data alignment is performed based on the data association between different grids. At the intersection of grids of different scales, data matching between runoff generation and confluence grids of different scales is achieved based on the water flow exchange rules between grid units and the calculation order between different grids.

[0028] In order to solve the problem of missing data required for the calculation of multi-scale watershed hydrological models, a missing data processing method based on the dynamic knowledge graph of the water cycle and the regression model is used. Multiple data interpolation algorithms such as EMB are used to ensure data consistency and spatiotemporal matching, thereby realizing the preparation of hydrological model data under missing data.

[0029] Furthermore, the process of digital extraction of river network in step (21) is as follows:

[0030] Based on the dynamic knowledge graph of water cycle, the topography and DEM data of small and medium-sized watersheds are obtained. Through filling depressions, calculating water flow direction and water accumulation, extracting river network grids, generating river network vectors, processing vector river networks, establishing spatial topology, classifying watershed systems, and storing layer attributes, the minimum granularity unit grid information is determined and the calculation order of the grid units is obtained. Then, the entire water system is searched according to the source of the water system and natural sub-basins are divided to determine the boundaries of the watershed and extract the geomorphological characteristics of small and medium-sized watersheds.

[0031] Beneficial effects: Compared with the existing technology, the beneficial effects of the present invention are: 1. Focusing on the construction of the dynamic knowledge graph of water cycle, a multi-scale nested digital twin framework of watershed-sub-watershed-aggregation unit-minimum granularity unit is built to carry out spatiotemporal distribution of data and association of data models between scales, and realize a data organization mechanism that is consistent in time and space; 2. Based on the dynamic knowledge graph of water cycle, remote sensing inversion data, ground observation data and hydrological model simulation are coupled and fed back, and multi-agents are introduced to realize knowledge and data dual-driven hydrological model adaptation, liberalization correction and automatic operation, constructing a high-fidelity mirror digital watershed, thereby realizing full-factor, full-process and fine-grained scenario modeling of digital twin watersheds; 3. It can provide technical support for the subsequent high-precision hydrological forecasting, flood control scheduling and simulation prediction based on parallel digital watersheds. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is a water cycle dynamic knowledge graph constructed by the present invention;

[0033] Figure 2 This is a schematic diagram of the principle of constructing a digital twin watershed;

[0034] Figure 3 It is a schematic diagram of the organization mechanism of multi-scale water cycle data and models. DETAILED DESCRIPTION

[0035] The present invention will be described in further detail below with reference to the accompanying drawings.

[0036] The present invention provides a digital twin watershed scene modeling method based on the water cycle dynamic knowledge graph, such as Figure 1 As shown, the following steps are included:

[0037] Step 1: Construct a dynamic knowledge graph of water cycle, such as Figure 1 shown.

[0038] Based on the water cycle remote sensing inversion data such as natural geographical data, hydrological and meteorological data, soil type data, vegetation cover data, water body area data in the basin, and combined with ground station monitoring data such as rainfall data and evapotranspiration data, various structured and unstructured basic data are collected. Through ontology construction, object relationship extraction, knowledge fusion and other technologies, the relationship between water conservancy business, water conservancy objects, hydrological models and data is formed, thereby realizing the construction of a dynamic knowledge graph of the water cycle and supporting real-time updates.

[0039] First, the graph is abstracted into a five-tuple structure, defined as:

[0040] G wa =(C, P, R, A, I)

[0041] Among them, C (Concepts) is the concept set of water conservancy objects, P (Properties) is the attribute set of related concepts, R (Relatiohs) is the relationship set, I (Individuals) is the instance set, and A (Axioms) is the stable relationship between concepts, that is, axioms. Axioms constrain the instances contained in the concepts, and all instances must follow the axioms. The relationship between axioms and concepts can be expressed as follows:

[0042]

[0043] In the above formula, A1 is a specific axiom. The relationship between concepts C1, C2 and C3 constitutes axiom A1, which plays a constraining role in the instantiation process of the three concepts.

[0044] In order to realize the dynamics of the graph, it is necessary to add time-varying and spatial characteristics to P and R, and this operation is completed by incorporating the nine-intersection model. For two instances A and B of I, the boundary (δA, δB), the interior (A 0 , B 0 ), and external (A -1 , B -1 ) The intersection of the three is realized by the nine-tuple to represent the topological relationship between instances A and B, as shown in the following matrix:

[0045]

[0046] Based on water conservancy industry standards such as the "General Principles for the Classification and Coding of Water Conservancy Objects" and the "Water Conservancy Data Directory Service Specification," and incorporating water conservancy object data, the graph's model and instance layers were designed. Each node in the graph corresponds to an object directory table and an object basic information table, thereby constructing a complete dynamic knowledge graph for the water cycle.

[0047] Step 2: Build a multi-scale nested digital twin framework, such as Figure 2 As shown in Figure 2, data spatiotemporal distribution and model configuration are carried out on grids at each level.

[0048] The digital twin of a small and medium-sized river basin is a synchronized digital spatial mapping of the physical basin, including hydrological and meteorological features, and changes in underlying surface conditions. It can represent a multi-scale digital scenario of the basin's water cycle. Based on the construction of a dynamic knowledge graph of the water cycle and a multi-scale digital twin framework of basin-subbasin-aggregation unit-minimum granularity unit, this approach develops spatiotemporal data distribution and data model association across scales, achieving a consistent spatiotemporal data organization mechanism. This coupling of feature data with water cycle modules allows for the construction of a high-fidelity mirror digital basin, forming a full-factor, full-process, and fine-grained scenario model for the digital twin basin. Based on the mirror digital basin, scenario elements are rationally pre-set, enabling advanced simulation rehearsals in subsequent parallel digital basins.

[0049] First, based on the basic hydrological and geographical data of small and medium-sized river basins, digital extraction of river networks and water systems is carried out.

[0050] The digital extraction process for small and medium-sized river networks involves acquiring topographic and geomorphological data and DEMs for these basins based on the dynamic knowledge graph of the water cycle. Through a series of steps, including depression filling, flow direction calculation, flow accumulation calculation, river network raster extraction, river network vector generation, vector river network processing, spatial topology establishment, basin water system classification, and layer attribute storage, the minimum granularity unit grid information is determined and the calculation order of the grid cells is obtained. The entire water system is then searched according to its source and divided into natural sub-basins to determine the boundaries of the basin. Geomorphological features of the small and medium-sized basins are extracted, such as slope and aspect of the minimum granularity unit grid, topographic index, flow path length, basin mean slope, and average channel slope. These geomorphological feature data are then fed back into the dynamic knowledge graph of the water cycle, providing a foundation for the self-organization and self-association of subsequent hydrological model parameters and related data.

[0051] Secondly, with the river network as the skeleton and the goal of characterizing the rainfall-runoff-confluence-evolution process of the water cycle, grid aggregation is carried out within the basin based on the minimum granularity unit grid, strip aggregation is carried out according to the known sections in the river channel, and quadrilateral aggregation is carried out in the non-river channel area based on the topography, hydrological and meteorological characteristics and the homogeneity of the underlying surface conditions, forming multi-scale aggregation units with a grid size of 250m in the runoff generation area, 10m in the confluence area, and 1m in the flooded area.

[0052] Finally, combined with the division of sub-basins, a multi-scale nested digital twin framework of minimum granularity unit-aggregation unit-sub-basin-basin is finally formed.

[0053] Data sorting and preprocessing are carried out for typical flood data in the basin, and the attribute data of water conservancy objects in the water cycle dynamic knowledge graph, including topographic data, soil type data, vegetation cover data, hydrological and meteorological data, rainfall data, evapotranspiration data, etc., are used to characterize the structure of rainfall-evapotranspiration-runoff-runoff-slope confluence-river confluence of the hydrological model. Based on these data, the four-layer grid of the multi-scale nested digital twin framework of basin-sub-basin-aggregation unit-minimum granularity unit is respectively distributed.

[0054] Step 3: Use the dynamic knowledge graph of the water cycle to achieve self-association and self-matching of data and models, complete the self-organization of multi-scale water cycle data and models, and build a high-fidelity mirror digital watershed, thereby realizing the scene modeling of all elements and processes of the digital twin watershed, such as Figure 3 shown.

[0055] Data alignment is performed based on the data association between different grids, and at the intersection of grids of different scales, data matching between runoff-confluence grids of different scales is achieved based on the water flow exchange rules between grid units and the calculation order between different grids; to address the problem of missing data required for the calculation of multi-scale basin hydrological models, a missing data processing method based on the dynamic knowledge graph of the water cycle and the regression model is used, and multiple data interpolation algorithms such as EMB are used to make it conform to data consistency and spatiotemporal matching, so as to realize the preparation of hydrological model data under missing data.

[0056] The process for self-association between hydrological models and data based on the water cycle dynamics knowledge graph is as follows: The model execution agent is invoked to retrieve parameters such as evapotranspiration, runoff, and confluence from the water cycle dynamics knowledge graph. The model parameters are then estimated and calculated using the relationships between them and data such as watershed geomorphological characteristics, soil types, and vegetation cover. For missing data, data substitution and fusion methods, including scaling, are used to complete the data in the water cycle dynamics knowledge graph. Existing data that meets the scaling requirements is processed to obtain all parameter values ​​required for hydrological model calculations. The model parameters are then associated with their values ​​in the water cycle dynamics knowledge graph.

[0057] In this embodiment, the execution process of the flow generation model is as follows: data retrieval: according to the calculation requirements of the model calculation parameters, the data source is retrieved based on the time and space range, and a data set that meets the data processing requirements is output; data processing: the data source is cleaned, filtered, projected, and other operations are performed on the data source, and the data meets the data flow required for the calculation; parameter calculation: calculation is performed based on the data source, consisting of data requirements and calculation process, and the parameter results are output; model calculation: responsible for the specific model / intermediate process calculation, consisting of parameters and calculation process, and outputs the results required by the confluence model; calculation scheduling: for the above calculation process, its execution process is optimized, and the optimized intelligent agent execution strategy is output.

[0058] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A digital twin watershed scenario modeling method based on a water cycle dynamic knowledge graph, characterized in that: The following steps are involved: (1) Constructing a dynamic knowledge graph of the water cycle: Based on the remote sensing inversion data of the water cycle in the basin and combined with the ground station monitoring data, various structured and unstructured basic data are collected. Through ontology construction, object relationship extraction, and knowledge fusion, the relationship between water conservancy business, water conservancy objects, hydrological models and data is formed, thereby realizing the construction of a dynamic knowledge graph of the water cycle; (2) Build a multi-scale nested digital twin framework to carry out data spatiotemporal distribution and model configuration on the grid level by level; (3) Using the dynamic knowledge graph of the water cycle to achieve self-association and self-matching of data and models, complete the self-organization of multi-scale water cycle data and models, and realize the scenario modeling of all elements and processes of the digital twin watershed; The implementation process of using the water cycle dynamic knowledge graph to realize the self-association of data and models in step (3) is as follows: Call the model execution agent, based on the water cycle dynamic knowledge graph, retrieve the parameters of the hydrological model in the graph, and use the correlation between the hydrological model parameters and the watershed geomorphological characteristics, soil type and vegetation cover data to call the agent to obtain the parameter values ​​required for the initial calculation of the hydrological model; For missing data, the data in the water cycle dynamic knowledge graph is supplemented; Process the existing data that meets the scale requirements to obtain all parameter values ​​required for hydrological model calculations, and associate the parameters of the hydrological model with the parameter values ​​in the water cycle dynamic knowledge graph; The process of achieving self-matching between data and models using the water cycle dynamic knowledge graph in step (3) is as follows: Data alignment is performed based on the data association between different grids. At the intersection of grids of different scales, data matching between runoff generation and confluence grids of different scales is achieved based on the water flow exchange rules between grid units and the calculation order between different grids. In order to solve the problem of missing data required for the calculation of multi-scale watershed hydrological models, a missing data processing method based on the dynamic knowledge graph of the water cycle and the regression model is used. Multiple data interpolation algorithms such as EMB are used to ensure data consistency and spatiotemporal matching, thereby realizing the preparation of hydrological model data under missing data.

2. The digital twin watershed scenario modeling method based on the water cycle dynamic knowledge graph according to claim 1 is characterized in that: The step (1) comprises the following steps: (11) The graph is abstracted into a five-tuple structure, defined as: G wa =(C,P,R,A,I) Among them, C is the concept set of water conservancy objects, P is the attribute set of related concepts, R is the relationship set, I is the instance set, and A is the stable relationship between concepts. The relationship between axioms and concepts in the water cycle dynamic knowledge graph is: Among them, A1 is a specific axiom, and the relationship between C1, C2 and C3 constitutes axiom A1, which plays a constraint role in the instantiation process of the three concepts; (12) By incorporating the nine-intersection model into P and R, we can add time-varying and spatial characteristics to the two instances A and B of I. By integrating the boundary (δA, δB) and internal (A 0 ,B 0 ), and external (A -1 ,B -1 ) The intersection of the three is realized by the nine-tuple, which represents the topological relationship between instances A and B, as shown in the following matrix: (13) Design the model layer and instance layer of the graph. Each node in the graph corresponds to an object directory table and an object basic information table, thus constructing a complete water cycle dynamic knowledge graph.

3. The digital twin watershed scenario modeling method based on the water cycle dynamic knowledge graph according to claim 1 is characterized in that: The step (2) comprises the following steps: (21) Digital extraction of river networks based on basic hydrological and geographical data of small and medium-sized river basins; (22) With the river network as the framework and the goal of characterizing the rainfall-runoff-confluence-evolution process of the water cycle, grid aggregation is carried out within the basin based on the minimum granularity unit grid. Strip aggregation is carried out within the river channel based on the known cross-section. In the non-river channel area, quadrilateral grid aggregation is carried out based on the topography, hydrological and meteorological characteristics, and the homogeneity of the underlying surface conditions to form aggregation units. (23) Combined with the division of sub-basins, a multi-scale nested digital twin framework of minimum granularity unit-aggregation unit-sub-basin-basin is finally formed; (24) Data sorting and preprocessing are carried out for typical flood data in the basin, and the attribute data of water conservancy objects in the water cycle dynamic knowledge graph are based on the structure of rainfall-evapotranspiration-flow generation-runoff-slope confluence-river confluence that represents the hydrological model. These data are respectively spread on the four-layer grid of the multi-scale nested digital twin framework of basin-sub-basin-aggregation unit-minimum granularity unit.

4. The digital twin watershed scenario modeling method based on the water cycle dynamic knowledge graph according to claim 3 is characterized in that: The process of digital extraction of river network in step (21) is as follows: Based on the dynamic knowledge graph of water cycle, the topography and DEM data of small and medium-sized watersheds are obtained. Through filling depressions, calculating water flow direction and water accumulation, extracting river network grids, generating river network vectors, processing vector river networks, establishing spatial topology, classifying watershed systems, and storing layer attributes, the minimum granularity unit grid information is determined and the calculation order of the grid units is obtained. Then, the entire water system is searched according to the source of the water system and natural sub-basins are divided to determine the boundaries of the watershed and extract the geomorphological characteristics of small and medium-sized watersheds.

Images