A Knowledge Graph-Driven Centralized Flood Forecasting Method

Abstract

This invention discloses a knowledge graph-driven ensemble flood forecasting method. First, pre-acquired watershed data is cleaned and integrated; a knowledge graph related to flood forecasting operations in the target watershed is constructed. Second, based on the forecasting process, the knowledge graph is used to acquire the data, knowledge, and models required for each forecasting step, and a flood forecasting scheme for the target watershed is formulated and published. Finally, based on the watershed topology, areal rainfall calculation scheme, lead time, and time period provided by the forecasting scheme, combined with measured rainfall-evaporation data and flood data provided by the knowledge graph, operational forecasting for the target watershed is conducted, and the forecast results are finally published. This invention, based on the strong correlation and data management capabilities of knowledge graphs, can effectively solve the problem of scarce and scattered data resources in flood forecasting, achieving dynamic adaptation of model structure, parameters, and state variables under changing watershed underlying surfaces, thus improving the accuracy of flood forecasting and the efficiency of operational processing.

Description

Technical Field

[0001] This invention belongs to the field of knowledge graph technology and flood forecasting, specifically relating to a knowledge graph-driven ensemble flood forecasting method. Background Technology

[0002] Flood forecasting is a crucial basis for flood control command, dispatch, and decision-making. Improving forecast accuracy and reducing flood losses are the core themes of "four early warnings" (flood forecasting, early warning, and early reporting). The entire flood forecasting process requires querying and storing a large amount of data and knowledge, involving complex and diverse watershed objects and interactions. Currently, this data is disorganized and scattered, lacking effective organization and unified management methods. This makes it difficult to support the rapid operation of flood forecasting processes, directly affecting the rapid and accurate decision-making of flood control departments and threatening the production and lives of the people.

[0003] Knowledge graphs are an efficient way to organize, represent, and manage information. They describe concepts, entities, and relationships in the objective world in a form close to human cognition, enabling more efficient and accurate retrieval of the knowledge and data needed in business processes. Currently, there is limited research on knowledge graphs specifically for flood forecasting. There is an urgent need to deepen and refine the business logic of flood forecasting, utilizing knowledge graph technology to organize and process data and knowledge in the field of flood forecasting. This will enhance the processing and scheduling capabilities of knowledge and data in business processes, continuously enrich and accumulate flood forecasting business knowledge and experience, and assist in improving the decision-making capabilities of flood control departments. Furthermore, flood forecasting driven by knowledge graphs is also one of the key technologies for improving the level of digital twin watersheds. Summary of the Invention

[0004] Purpose of the invention: The purpose of this invention is to provide a knowledge graph-driven lumped flood forecasting method that enables dynamic adaptation of model structure, parameters, and state variables under changing watershed underlying surface conditions, thereby improving the accuracy of flood forecasting and the efficiency of operational processing.

[0005] Technical solution: This invention provides a knowledge graph-driven ensemble flood forecasting method, comprising the following steps:

[0006] (1) Cleaning and integrating the pre-acquired watershed data;

[0007] (2) Construct a knowledge graph related to flood forecasting operations in the target watershed, including a watershed data graph, a watershed object relationship graph, a forecasting model graph, and a forecasting flowchart;

[0008] (3) Based on the compilation process, knowledge graphs are used to obtain the data, knowledge and models required for each compilation step, and flood forecasting schemes for the target watershed are formulated and released.

[0009] (4) Based on the watershed topology, areal rainfall calculation scheme, forecast period and time period provided by the forecast scheme, combined with the measured rainfall evaporation data and flood data provided by the knowledge graph, the target watershed operation forecast is carried out, and the forecast results are finally released.

[0010] Further, step (1) includes the following steps:

[0011] (11) Extract digital elevation data, underlying surface data, hydrological and meteorological data, daily flood model data and flood event data of the watershed, carry out data cleaning, remove invalid and redundant values, and ensure data quality;

[0012] (12) Divide the watershed data into basic data, monitoring data, geospatial data, and business management data, etc.;

[0013] (13) Standardize the classification of relevant entities and their attributes, use the RDF framework to organize and integrate the aforementioned watershed data into structured information and store it; integrate the key attributes and relationships of each water conservancy object; integrate the database structure and statistical characteristics of water and rainfall monitoring elements; compile the element names, types, etc. of geospatial data; integrate the inputs and outputs, parameter names and type constraints of hydrological forecasting models or modules;

[0014] (14) Establish a relationship between watershed data and watershed objects, and realize the traceability and updability of watershed data through metadata registration.

[0015] Furthermore, the implementation process of step (2) is as follows:

[0016] The data map of the watershed is aggregated with objects related to the measured data, and the monitored objects are linked with the compiled monitoring data, geospatial data and remote sensing data;

[0017] The watershed object relationship map categorizes core watershed objects into rivers and lakes, water conservancy projects, monitoring stations, and other management objects. The watershed object relationship map includes basic information on water conservancy objects within the target watershed, as well as the names and upstream-downstream relationships of each calculation section. Calculation sections are divided into sub-watershed intervals, headwater sections, and connecting sections. The watershed object relationship map stores underlying surface features and meteorological flood features generated during the forecasting process.

[0018] Construction of forecast model maps; extraction of the structure of the hydrological forecast model, including evapotranspiration module, runoff generation module, slope runoff module, and river runoff module; sorting out model parameters and associating them with the watershed objects required for parameter calibration; defining calibration methods and real-time correction methods for state variables in the maps;

[0019] The forecast flowchart abstracts each step of the flood forecasting scheme preparation process and operational forecasting process into events, forming an event chain that drives the entire flood forecasting scheme preparation process. First, the forecast flowchart is abstracted into a triplet structure, as follows:

[0020] G event =(E,R,L)

[0021] E represents the event set, R represents the set of relationships between events, and L represents the evolution rules between events; the set of relationships between events mainly includes the following:

[0022] R={before,after,simultaneous,contains}

[0023] The rules for representing events can be expressed as follows:

[0024]

[0025] Among them, E1, E2 and E3 are three specific events. These three events, based on different combination relationships, together constitute the evolution rule L between events.

[0026] Furthermore, the implementation process of step (3) is as follows:

[0027] (31) Digital watershed feature extraction and water conservancy engineering analysis: Quickly obtain watershed objects and associated digital elevation data from watershed data maps and process them intelligently to extract elements such as watershed boundaries, sub-watershed division, and watershed geomorphological features; obtain the distribution of water conservancy engineering objects and monitoring stations from watershed object relationship maps, generate watershed topology by combining watershed features, and transfer the data to the watershed object relationship map.

[0028] (32) Meteorological and flood characteristics analysis and watershed underlying surface characteristics analysis: call the monitoring stations and associated rainfall, flood data and geospatial data in the watershed data map to calculate the watershed climate characteristics, rainstorm characteristics, historical wet and dry years, historical major floods, etc.; obtain the vegetation cover, soil type products, etc. retrieved from the watershed data map, extract the watershed vegetation cover characteristic index, soil type spatial distribution, etc., and store them;

[0029] (33) Forecasting scheme preparation: Select the corresponding daily model data and flood data from the watershed data map to determine the lead time and duration of this forecasting scheme; formulate the areal rainfall calculation scheme based on the distribution of rain gauges in the watershed topology;

[0030] (34) Forecast scheme modeling; Based on the watershed characteristics generated in steps (31) to (33), select model modules from the forecast model map and assemble them, and obtain rainfall, evaporation and flood data from the watershed data map according to the correlation between parameters and watershed objects, and complete parameter calibration, sensitivity or correlation analysis, and model simulation effect analysis and verification; continuously repeat steps (31) to (34) to make the simulation effect of the forecast model reach the optimal level;

[0031] (35) Forecast scheme release: The generated watershed topology, areal rainfall calculation scheme, forecast model structure and parameters, and scheme calculation conditions are integrated into a forecast scheme and released.

[0032] Furthermore, the implementation process of step (4) is as follows:

[0033] (41) Operational Forecast Calculation - Measured Section: First, select the forecast section and determine the section type through the watershed object relationship map. Based on the watershed topology given in the forecast scheme, trace the calculation nodes through the calculation sequence. For the runoff generation process of the sub-watershed interval, based on the areal rainfall calculation scheme in the forecast scheme, retrieve the real-time rainfall data of the corresponding rain gauge station from the watershed data map and integrate it with the rainfall products retrieved from remote sensing. Combine the evapotranspiration data and remote sensing evaporation data to obtain the evapotranspiration. Then, complete the runoff generation calculation and lag confluence calculation. For the runoff generation process of the leading node, it is necessary to... Based on the watershed topology, we continue to find the measured outflow from associated reservoirs and process it according to the time period, then direct the flow to downstream nodes for river channel confluence. For the initiating nodes, we need to continue to trace their associated sub-watershed intervals or leading nodes to complete the runoff calculation. We overlay the runoff calculation results and flow from the calculation nodes, and continuously recursively traverse to obtain the calculated flow of the forecast object. We obtain the model state quantity correction method from the forecast model map, retrieve the model state observations such as soil moisture content and small water body area from the watershed data map, as well as the measured flow from the forecast nodes, to complete the real-time correction of the state quantities.

[0034] (42) Operation forecast calculation - forecast segment: Based on the state variables of the hydrological model after the actual measurement segment is corrected, the planned outflow is adopted for the head node, and the forecast products such as numerical forecast and radar extrapolation are adopted for the sub-basin interval. After the forecast results are intervened by manual or intelligent means, the forecast results are finally released.

[0035] Beneficial effects: Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention can quickly integrate and reconstruct the basic data, monitoring data, geospatial data, and business management data of the watershed. Based on knowledge graph technology, it condenses and organizes the objects and relationships of the watershed, the structure and parameters of the hydrological model, and the business logic of flood forecasting. It breaks down the barriers that support the rapid flow of data, knowledge and business in flood forecasting, improves the efficiency of the entire forecasting process, and enhances the rapid and accurate decision-making ability of flood control departments. Attached Figure Description

[0036] Figure 1 This is a flowchart of the invention;

[0037] Figure 2 This is a schematic diagram of the forecasting scheme development process based on knowledge graphs;

[0038] Figure 3 It is a schematic diagram of the relationships between various objects in the watershed topology and watershed object relationship map;

[0039] Figure 4 This is a flowchart illustrating the process of predicting and calculating work progress and the actual measurement.

[0040] Figure 5 This is a flowchart illustrating the forecast calculation and forecast segment process for this operation. Detailed Implementation

[0041] The present invention will now be described in further detail with reference to the accompanying drawings.

[0042] This invention proposes a knowledge graph-driven ensemble flood forecasting method, such as... Figure 1 As shown, it includes the following steps:

[0043] Step 1: Clean and integrate the pre-acquired watershed data.

[0044] Extract digital elevation data, underlying surface data, hydrological and meteorological data, daily flood model data, and flood event data of the watershed, and carry out data cleaning to remove invalid and redundant values ​​to ensure data quality; based on the outline for the construction of digital twin watersheds, the watershed data is divided into basic data, monitoring data, geospatial data, and business management data.

[0045] Based on the industry standard "General Rules for Classification and Coding of Water Conservancy Objects" (SL / T213-2020), the relevant entities and their attributes are classified in a standardized manner. The RDF framework is used to organize and integrate the aforementioned watershed data into structured information and store it. The key attributes and relationships of each water conservancy object are integrated. The database structure and statistical characteristics of water and rainfall monitoring elements are integrated. The element names and types of geospatial data are compiled. The inputs and outputs, parameter names and type constraints of hydrological forecasting models / modules are integrated.

[0046] Step 2: Construct a knowledge graph related to flood forecasting services for the target watershed.

[0047] Integrate various structured and unstructured watershed data related to flood forecasting operations, such as hydrological and rainfall data, geospatial data, and water conservancy project data, within the target watershed. Based on existing standards and specifications and under the guidance of domain experts, utilize technologies such as ontology construction and object relationship extraction to construct watershed data maps, watershed object relationship maps, forecasting model maps, and forecasting flowcharts that can support forecasting operations.

[0048] Construct a watershed data map; this map focuses on the monitoring attributes of water conservancy objects, such as the monitoring frequency of rain gauges and real-time rainfall values, and aggregates objects related to measured data (monitoring objects such as hydrological stations, rain gauges, evaporation stations, and reservoirs), and links them with the compiled monitoring data, geospatial data, and remote sensing data.

[0049] A watershed object relationship map is constructed. Based on the operational needs of flood forecasting, and considering the natural entities, water conservancy facilities, and management concepts involved in water management activities, the core watershed objects are categorized into rivers and lakes, water conservancy projects, monitoring stations, and other management objects. The watershed object relationship map needs to include basic information and relationships of water conservancy objects within the target watershed. Basic information includes the codes, names, and locations of monitoring stations, while relationships include the scheduling relationships of reservoirs, ponds, dams, and other water conservancy projects, as well as the names and upstream-downstream relationships of each calculation section. In this invention, calculation sections are divided into sub-watershed intervals, headwater sections (without upstream sections), and connecting sections. This map can store underlying surface features and meteorological flood features generated during the forecasting scheme preparation process.

[0050] The forecast model map is constructed; the structure of the hydrological forecast model is extracted, including the evapotranspiration module, runoff generation module, slope runoff module, and river runoff module; the model parameters are sorted out and associated with the watershed objects required for parameter calibration; the calibration method and the real-time correction method of state variables are defined in the map.

[0051] Forecast flowchart construction: Based on the standardized flood forecasting scheme preparation process and operational forecasting process, including meteorological and flood analysis steps, watershed underlying surface analysis steps, forecasting model selection, etc., the object attributes and data attributes of each step are extracted and abstracted into various events. Based on business management relationships and business dependencies, the sequence relationships between each step are organized in the flowchart, such as before, after, simultaneous, and contains relationships, forming an event chain, which ultimately drives the entire flood forecasting scheme preparation process.

[0052] First, the forecast flowchart is abstracted into a triplet structure, as follows:

[0053] G event =(E,R,L)

[0054] E represents the event set, R represents the set of relationships between events, and L represents the evolution rules between events; the set of relationships between events mainly includes the following:

[0055] R={before,after,simultaneous,contains}

[0056] The rules for representing events can be expressed as follows:

[0057]

[0058] Among them, E1, E2 and E3 are three specific events. These three events, based on different combination relationships, together constitute the evolution rule L between events.

[0059] Step 3: Based on the compilation process, utilize knowledge graphs to acquire the data, knowledge, and models required for each compilation step, and formulate and publish flood forecasting plans for the target watershed. Specifically, as follows... Figure 2 As shown:

[0060] S31: Digital Watershed Feature Extraction and Hydraulic Engineering Analysis: This involves rapidly acquiring watershed objects and their associated digital elevation data from watershed data maps, processing them intelligently, and extracting elements such as watershed boundaries, sub-watershed divisions, and watershed geomorphological features. It also involves obtaining the distribution of watershed hydraulic engineering objects and monitoring stations from watershed object relationship maps, generating watershed topology based on watershed features, and transferring this data to the watershed object relationship map. (See attached...) Figure 3 As shown, in the watershed object relationship graph, the generated watershed topology is associated with related objects.

[0061] S32: Meteorological and flood characteristic analysis and watershed underlying surface characteristic analysis: Call the monitoring stations and associated rainfall, flood data and geospatial data in the watershed data map to calculate the watershed climate characteristics, rainstorm characteristics, historical wet and dry years, historical major floods, etc.; obtain the vegetation cover, soil type products, etc. from the watershed data map through remote sensing inversion, extract the watershed vegetation cover characteristic index, soil type spatial distribution, etc., and store them.

[0062] S33: Forecasting scheme preparation: Select relevant daily model data and flood data from the watershed data map to determine the lead time and duration of this forecasting scheme; formulate an areal rainfall calculation scheme based on the distribution of rain gauges in the watershed topology.

[0063] S34: Forecast Scheme Modeling: Based on the watershed characteristics generated in S21, S22, and S23, select and assemble model modules from the forecast model map, and obtain rainfall, evaporation, and flood data from the watershed data map according to the correlation between parameters and watershed objects. Complete tasks such as parameter calibration, sensitivity / correlation analysis, and model simulation effect analysis and verification. Continuously repeat the above steps to optimize the simulation effect of the forecast model.

[0064] S35: Forecast scheme release; integrate the watershed topology, areal rainfall calculation scheme, forecast model structure and parameters, and scheme calculation conditions (forecast period and duration) generated in the above steps into a forecast scheme and release it.

[0065] Step 4: Based on the watershed topology, areal rainfall calculation scheme, forecast period and time period provided by the forecast scheme, and combined with the measured rainfall and evaporation data and flood data provided by the knowledge graph, conduct operational forecasts for the target watershed, and finally release the forecast results.

[0066] Work forecast calculation - measured section, such as Figure 4 As shown: First, the forecast section is selected, and the type of section is determined through the watershed object relationship map. Based on the watershed topology given in the forecast scheme, the calculation nodes are traced back according to the calculation sequence. For the runoff generation process of the sub-watershed interval, based on the areal rainfall calculation scheme in the forecast scheme, real-time rainfall data of the corresponding rain gauge station is retrieved from the watershed data map and fused with the rainfall products retrieved from remote sensing. Combined with evapotranspiration data and remote sensing evaporation data, the evapotranspiration is obtained. Then, the runoff generation calculation and lag confluence calculation are completed. For the runoff generation process of the leading node, it is necessary to consider the watershed topology. The process continues to identify and process the measured outflow from associated reservoirs based on time periods, and then directs the flow to downstream nodes. For initiating nodes, it is necessary to trace their associated sub-basin intervals or leading nodes to complete the runoff calculation. The runoff calculation results and flow rates of the calculation nodes are superimposed and recursively traversed to obtain the calculated flow rate of the forecast object. The model state quantity correction method is obtained from the forecast model map, and the model state observations such as soil moisture content and small water body area in the watershed data map, as well as the measured flow rates of the forecast nodes, are retrieved to complete the real-time correction of the state quantities.

[0067] Operation forecast calculation - forecast segment, such as Figure 5 As shown, based on the state variables of the hydrological model after the actual measurement section is corrected, the planned outflow is used at the head node, and the forecast products such as numerical forecasting and radar extrapolation are used for the sub-basin intervals. After the forecast results are intervened by manual or intelligent means, the final forecast results are released.

[0068] The embodiments described above are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A knowledge graph-driven ensemble flood forecasting method, characterized in that, Includes the following steps: (1) Cleaning and integrating the pre-acquired watershed data; (2) Construct a knowledge graph related to flood forecasting operations in the target watershed, including a watershed data graph, a watershed object relationship graph, a forecasting model graph, and a forecasting flowchart; (3) Based on the compilation process, knowledge graphs are used to obtain the data, knowledge and models required for each compilation step, and flood forecasting schemes for the target watershed are formulated and released. (4) Based on the watershed topology, areal rainfall calculation scheme, forecast period and time period provided by the forecast scheme, combined with the measured rainfall and evaporation data and flood data provided by the knowledge graph, carry out the target watershed operation forecast, and finally release the forecast results; The implementation process of step (3) is as follows: (31) Digital watershed feature extraction and water conservancy engineering analysis: Quickly obtain watershed objects and associated digital elevation data from watershed data maps and process them intelligently to extract watershed boundaries, sub-watershed divisions, and watershed geomorphological features; obtain the distribution of water conservancy engineering objects and monitoring stations from watershed object relationship maps, generate watershed topology by combining watershed features, and transfer the data to the watershed object relationship map. (32) Meteorological and flood characteristics analysis and watershed underlying surface characteristics analysis: call the monitoring stations and associated rainfall, flood data and geospatial data in the watershed data map to calculate the watershed climate characteristics, rainstorm characteristics, historical abundant and dry years and historical major floods; obtain the vegetation cover and soil type products retrieved from the watershed data map by remote sensing, extract the watershed vegetation cover characteristic index and soil type spatial distribution, and store them; (33) Forecasting scheme preparation: Select the corresponding daily model data and flood data from the watershed data map to determine the lead time and duration of this forecasting scheme; formulate the areal rainfall calculation scheme based on the distribution of rain gauges in the watershed topology; (34) Forecast scheme modeling; Based on the watershed characteristics generated in steps (31) to (33), model modules are selected from the forecast model map and assembled. Rainfall, evaporation and flood data are obtained from the watershed data map according to the correlation between parameters and watershed objects. Parameter calibration, sensitivity or correlation analysis and model simulation effect analysis and verification are completed. Steps (31) to (34) are continuously repeated to make the simulation effect of the forecast model reach the optimal level. (35) Forecast scheme release: The generated watershed topology, areal rainfall calculation scheme, forecast model structure and parameters, and scheme calculation conditions are integrated into a forecast scheme and released.

2. The knowledge graph-driven ensemble flood forecasting method according to claim 1, characterized in that, Step (1) includes the following steps: (11) Extract digital elevation data, underlying surface data, hydrological and meteorological data, daily flood model data and flood event data of the watershed, carry out data cleaning, remove invalid and redundant values, and ensure data quality; (12) Divide the watershed data into basic data, monitoring data, geospatial data, and operational management data; (13) Standardize the classification of relevant entities and their attributes, use the RDF framework to organize and integrate the aforementioned watershed data into structured information and store it; integrate the key attributes and relationships of each water conservancy object; integrate the database structure and statistical characteristics of water and rainfall monitoring elements; compile the element names and types of geospatial data; integrate the inputs, outputs, parameter names and type constraints of hydrological forecasting models or modules; (14) Establish a relationship between watershed data and watershed objects, and realize the traceability and updability of watershed data through metadata registration.

3. The knowledge graph-driven ensemble flood forecasting method according to claim 1, characterized in that, The implementation process of step (2) is as follows: The data map of the watershed is aggregated with objects related to the measured data, and the monitored objects are linked with the compiled monitoring data, geospatial data and remote sensing data; The watershed object relationship map categorizes core watershed objects into rivers and lakes, water conservancy projects, monitoring stations, and other management objects. The watershed object relationship map includes basic information on water conservancy objects within the target watershed, as well as the names and upstream-downstream relationships of each calculation section. Calculation sections are divided into sub-watershed intervals, headwater sections, and connecting sections. The watershed object relationship map stores underlying surface features and meteorological flood features generated during the forecasting process. Construction of forecast model maps; extraction of the structure of the hydrological forecast model, including evapotranspiration module, runoff generation module, slope runoff module, and river runoff module; sorting out model parameters and associating them with the watershed objects required for parameter calibration; defining calibration methods and real-time correction methods for state variables in the maps; The forecast flowchart abstracts each step of the flood forecasting scheme preparation process and operational forecasting process into events, forming an event chain that drives the entire flood forecasting scheme preparation process. First, the forecast flowchart is abstracted into a triplet structure, as follows: G event =(E,R,L) E represents the event set, R represents the set of relationships between events, and L represents the evolution rules between events; the set of relationships between events mainly includes the following: R={before,after,simultaneous,contains} The rules for representing events are expressed as follows: Among them, E1, E2 and E3 are three specific events. These three events, based on different combination relationships, together constitute the evolution rule L between events.

4. The knowledge graph-driven ensemble flood forecasting method according to claim 1, characterized in that, The implementation process of step (4) is as follows: (41) Operational forecast calculation - measured section: First, select the forecast section and determine the type of section through the watershed object relationship map. According to the watershed topology given by the forecast scheme, trace the calculation nodes through the calculation order. For the runoff generation process of the sub-watershed interval, based on the areal rainfall calculation scheme in the forecast scheme, retrieve the real-time rainfall data of the corresponding rain gauge station from the watershed data map and integrate it with the rainfall product retrieved by remote sensing. Combine the evapotranspiration data and remote sensing evaporation data to obtain the evapotranspiration. Then complete the runoff generation calculation and the lag confluence calculation. For the runoff generation process of the head node, it is necessary to continue to find the measured outflow of the associated reservoir according to the watershed topology and process it according to the time period, and carry out the river confluence to the downstream node. For the connecting node, it is necessary to continue to trace its associated sub-watershed interval or head node to complete the runoff generation and confluence calculation. The flow calculation results and flow rates of the superimposed computing nodes are continuously recursively traversed to obtain the calculated flow rate of the forecast object; the model state quantity correction method is obtained from the forecast model map, and the state observations of the soil moisture content model and small water body area model in the watershed data map, as well as the measured flow rate of the forecast node, are retrieved to complete the real-time correction of the state quantity. (42) Operation forecast calculation - forecast segment: Based on the state variables of the hydrological model after the actual measurement segment is corrected, the planned outflow is adopted for the head node, and the forecast products including numerical forecast and radar extrapolation are adopted for the sub-basin interval. After the forecast results are intervened by manual or intelligent means, the forecast results are finally released.

Images