A watershed multi-level hydrological forecasting method and system based on dual-mode driving

Abstract

This invention provides a basin-wide multi-level hydrological forecasting method and system based on a dual-mode driven approach. The method includes the following steps: identifying forecast nodes in the river system of the basin based on historical hydrological and meteorological data, and classifying them into Type I and Type II forecast nodes; constructing a corresponding physical mechanism hydrological model for each Type I forecast node; constructing a corresponding machine learning model for each Type II forecast node; and predicting the hydrological forecast results for the corresponding Type I and Type II forecast nodes based on the physical mechanism hydrological model and the machine learning model, respectively, to generate a basin-wide hydrological forecast. This invention deeply mines the correlation characteristics between water level and flow rate, and introduces deep learning algorithms to enhance the accuracy of hydrological forecasts for stations that are not suitable for physical mechanism hydrological models.

Description

Technical Field

[0001] This invention belongs to the field of hydrological forecasting technology, specifically relating to a watershed multi-level hydrological forecasting method and system based on dual-mode driving. Background Technology

[0002] With the intensification of global climate change, extreme weather events, especially river floods triggered by heavy rainfall, are occurring more frequently, posing a significant challenge to hydrological forecasting. Hydrological monitoring, forecasting, and early warning are crucial components of flood control, drought relief, and disaster reduction efforts, playing a key role in minimizing disaster losses and protecting the lives and property of the people. However, existing hydrological forecasting technologies still have many shortcomings when dealing with complex and dynamic climate conditions.

[0003] Traditional hydrological forecasting methods primarily rely on conceptual and distributed hydrological models, which forecast by constructing rainfall-runoff relationships for specific monitoring stations and their corresponding watershed intervals. While these models can simulate hydrological processes to some extent, they often struggle to achieve high accuracy and real-time performance when dealing with complex river systems, large watersheds, and numerous water conservancy projects. Furthermore, most hydrological forecasting systems focus on flow prediction, while water level forecasting is relatively lacking. Especially in automated forecasting systems, the correlation between water level and flow is not fully explored and utilized, which may lead to delayed or inaccurate decisions in actual flood control operations, increasing the risks of disaster prevention and mitigation efforts.

[0004] In recent years, with the rapid development of data-driven technologies such as machine learning and deep learning, an increasing number of studies have attempted to simulate the rainfall-runoff relationship in watersheds based on these methods. Data-driven models can improve forecast accuracy to some extent by mining complex relationships in historical data. However, these models are highly dependent on the correlation between input and output, and their forecasting capabilities are significantly limited, especially when data quality and quantity are insufficient. Furthermore, purely data-driven methods cannot fully reflect the physical mechanisms of hydrological processes, limiting their application in complex watersheds.

[0005] To address these challenges, the integration of data-driven and physical process methods has emerged as a new research direction. Especially for large watersheds with complex river systems and numerous hydraulic engineering projects, hydrological forecasting typically requires dividing the watershed into multiple forecast nodes and conducting forecasts step-by-step from upstream to downstream. This process must balance the impact of reservoir scheduling and forecast correction calculations at upstream and downstream stations. However, as the number of upstream nodes increases, the topological complexity of the watershed significantly increases, and forecast errors may propagate and amplify at each level, affecting the overall accuracy of the forecast. Therefore, reasonable selection is needed when dividing the watershed into nodes to ensure the accuracy and reliability of forecast conclusions for important sections.

[0006] Furthermore, the difficulty of forecasting increases further when the quality of hydrological data between upstream and downstream nodes is poor, or when only water level stations can be deployed due to river characteristics. These areas often lack forecast information and elements, severely restricting the completeness and accuracy of hydrological forecasts. In such cases, how to fully utilize limited data resources and improve the correlation between water level and flow forecasts becomes an urgent technical challenge. Summary of the Invention

[0007] The purpose of this invention is to address the shortcomings of the aforementioned background technology and provide a watershed multi-level hydrological forecasting method and system based on dual-mode driving, ensuring more accurate forecasting results for important cross sections, thereby improving the stability and reliability of the entire watershed forecasting system, ensuring efficient forecasting under complex water system conditions, and enhancing the accuracy of hydrological forecasting at stations that are not suitable for physical mechanism hydrological models by deeply mining the correlation characteristics between water level and flow.

[0008] The technical solution adopted in this invention is: a watershed multi-level hydrological forecasting method based on dual-mode driving, comprising the following steps:

[0009] Based on historical hydrological and meteorological data, forecast nodes in the river system of the basin were identified and classified into Type I and Type II forecast nodes.

[0010] Type I forecast nodes are defined as nodes that are hydrological stations with complete data and comprehensive measurement elements. Type II forecast nodes are defined as nodes that can only observe water level or are not applicable to the physical mechanism hydrological model due to river characteristics or the influence of water conservancy projects.

[0011] For each Type I forecast node, a corresponding physical mechanism hydrological model is constructed.

[0012] For each Type II forecast node, a corresponding machine learning model is constructed.

[0013] Based on the physical mechanism hydrological model and the machine learning model, the hydrological forecast results of the corresponding Type I and Type II forecast nodes are predicted respectively, and the hydrological forecast results of the whole basin are generated.

[0014] In the above technical solution, the construction process of the physical mechanism hydrological model includes: selecting several candidate physical mechanism hydrological models and their model parameters; extracting historical meteorological and hydrological data of the corresponding Type I forecast nodes and organizing them into the form of candidate model input parameters as the corresponding dataset; calibrating the parameters of each candidate model using the corresponding dataset; evaluating the forecasting effect of each candidate model based on the mean absolute error, Nash efficiency coefficient, root mean square error, flood peak error, and peak occurrence time, and selecting the model with the best forecasting effect as the physical mechanism hydrological model of the Type I forecast node.

[0015] In the above technical solution, the machine learning model construction process includes: selecting several candidate machine learning models and their model parameters; extracting historical meteorological and hydrological data of the corresponding Type II forecast nodes and organizing them into the form of candidate model input parameters as the corresponding dataset; training each candidate model using the corresponding dataset; evaluating the forecasting effect of each candidate model based on mean absolute error, Nash efficiency coefficient, root mean square error, flood peak error, and peak occurrence time, and selecting the model with the best forecasting effect as the machine learning model for that Type II forecast node.

[0016] In the above technical solution, the corresponding predefined basic model frameworks are called from the model library as alternative physical mechanism hydrological models or machine learning models; all basic model frameworks adopt standardized model input and output interfaces; the functions, applicable scope, and input and output parameter formats of all basic model frameworks are clearly defined; and all basic model frameworks are registered in the model library.

[0017] The above technical solution also includes: constructing the water system topology of the basin based on Type I forecast nodes and Type II forecast nodes; according to the water system topology, firstly, the hydrological forecast results of Type I forecast nodes are calculated sequentially from upstream to downstream, and then the hydrological forecast results of Type II forecast nodes are calculated sequentially from upstream to downstream, with the flow calculation of each node based on the latest upstream forecast results.

[0018] In the above technical solution, the process of predicting the hydrological forecast result of any Type I forecast node based on the physical mechanism hydrological model includes: obtaining the latest meteorological forecast data corresponding to the Type I forecast node based on external data; organizing the cross-sectional flow calculation results and meteorological forecast data of the Type I forecast node upstream of the Type I forecast node into the model input parameter format defined by the model library, and using it as the model input parameter; calling the physical mechanism hydrological model corresponding to the Type I forecast node from the model library, and using the organized model input parameters to calculate the cross-sectional flow; and storing the calculated cross-sectional flow result corresponding to the Type I forecast node into the forecast result library.

[0019] In the above technical solution, the process of predicting the hydrological forecast result of any Type II forecast node based on the machine learning model includes: obtaining the latest meteorological forecast data corresponding to the Type II forecast node based on external data; organizing the cross-sectional flow calculation results of the Type I forecast node upstream of the Type II forecast node, the hydrological forecast results of the upstream Type II forecast node, and the meteorological forecast data into the model input parameter format defined by the model library, and using them as model input parameters; calling the machine learning model corresponding to the Type II forecast node from the model library, and using the organized model input parameters to perform hydrological forecasting; and storing the calculated hydrological forecast result corresponding to the Type II forecast node into the forecast result library.

[0020] In the above technical solution, the historical hydrological and meteorological data are used in subsequent steps after preprocessing; the preprocessing process includes: collecting hydrological and meteorological data of the river system to be predicted, removing outliers, filling in missing values, unifying the data format, and storing it in the database; the hydrological and meteorological data includes precipitation, water level, flow rate, temperature, humidity, air pressure, wind speed, and evaporation.

[0021] This invention provides a watershed multi-level hydrological forecasting system based on dual-mode driving, comprising:

[0022] The forecast node management module is used to identify forecast nodes in the river system of the basin based on historical hydrological and meteorological data, and classify them into Type I and Type II forecast nodes;

[0023] Type I forecast nodes are defined as nodes that are hydrological stations with complete data and comprehensive measurement elements. Type II forecast nodes are defined as nodes that can only observe water level or are not applicable to the physical mechanism hydrological model due to river characteristics or the influence of water conservancy projects.

[0024] The forecast model library module is used to build the corresponding physical mechanism hydrological model for each Type I forecast node and the corresponding machine learning model for each Type II forecast node.

[0025] The hydrological forecasting module is used to predict the hydrological forecasting results of the corresponding Type I and Type II forecasting nodes based on physical mechanism hydrological models and machine learning models, respectively, and generate hydrological forecasting results for the entire basin.

[0026] The present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the watershed multi-level hydrological forecasting method based on dual-mode driving described in the above technical solution.

[0027] The beneficial effects of this invention are as follows: This invention describes a basin-wide multi-level hydrological forecasting method based on dual-mode driving. By classifying forecast nodes into Type I and Type II, physical mechanism hydrological models and machine learning models are constructed respectively, ultimately generating hydrological forecasting results for the entire basin. By classifying forecast nodes and using the most suitable model (physical model or machine learning model), the hydrological processes of different types of nodes can be simulated more accurately, improving the overall forecast accuracy. Different model methods are used for forecast nodes with different characteristics in complex basins, enhancing the system's adaptability and forecasting effect in complex water systems. Reasonably classifying nodes and using corresponding models avoids resource waste and improves the efficiency of model construction and operation. Combining physical process and data-driven methods, it can accurately simulate the hydrological processes of key nodes and achieve effective forecasting in data-limited areas, ensuring comprehensive forecast coverage of the entire basin.

[0028] Furthermore, the construction process of the physical mechanism hydrological model of the present invention includes model selection, parameter calibration, and model evaluation, ultimately selecting the model with the best forecast performance as the physical model for Type I forecast nodes; through parameter calibration and multi-index evaluation, it is ensured that the selected physical model can best match historical observation data, thereby improving the accuracy of forecasts; the systematic model selection and evaluation process reduces the influence of subjective factors and ensures the scientific validity and reliability of the selected model; based on the historical data and watershed characteristics of specific Type I nodes, the most suitable physical model is customized and selected to enhance the model's adaptability to specific watersheds; by comprehensively considering multiple evaluation indicators, the model performance is fully evaluated to ensure that the forecast results perform excellently under different hydrological scenarios.

[0029] Furthermore, the construction process of the machine learning model of this invention includes model selection, training, and evaluation, ultimately selecting the model with the best forecasting performance as the machine learning model for the Type II forecasting node; by selecting the optimal machine learning model, the accuracy of hydrological forecasts for the Type II forecasting node is improved by utilizing the complex relationships of historical data; the systematic model training and evaluation process ensures that the selected model has good generalization ability and stability under different data conditions; for Type II nodes with limited data quality or observation elements, the machine learning model effectively utilizes limited data to achieve reliable hydrological forecasts; and by applying advanced machine learning methods, the application scenarios of deep learning in the field of hydrological forecasting are expanded, promoting the development and innovation of hydrological forecasting technology.

[0030] Furthermore, this invention calls a predefined basic model framework from the model library, ensuring that all models have standardized input / output interfaces and are registered in the model library. Through standardized model input / output interfaces, compatibility and interchangeability between different models are achieved, simplifying the model calling and integration process. The predefined and registered basic model framework reduces redundant development and configuration time, accelerating model construction and application. The unified model framework and interfaces ensure consistency and coordination within the entire forecasting system, improving overall system performance. Centralized management and registration of models facilitate subsequent model updates, maintenance, and expansion, ensuring long-term stable system operation.

[0031] Furthermore, this invention constructs the watershed's river system topology based on Type I and Type II forecast nodes, and calculates the hydrological forecast results of the forecast nodes sequentially from upstream to downstream according to the river system topology order; forecasting according to the watershed topology order ensures that the prediction of each node is based on the latest upstream forecast results, avoiding data dependency confusion; reasonable calculation order and node division reduce the stepwise propagation and amplification of forecast errors in the watershed, improving the overall forecast accuracy; systematic calculation order and node management improve the logic and efficiency of the forecast process, ensuring the reliability of forecast results; adapting to the forecasting needs of complex water systems, through reasonable division and sequential calculation, it achieves efficient multi-level forecasting, supporting the comprehensive management of water conservancy projects.

[0032] Furthermore, the specific process of predicting hydrological forecast results for Type I forecast nodes based on the physical mechanism hydrological model of this invention includes data acquisition, processing, model invocation, and result storage; by acquiring the latest meteorological forecast data and upstream node flow results, it ensures that the forecasts for Type I nodes are based on the most accurate and up-to-date information; standardizing the model input parameter format ensures data consistency and compatibility, improves model operating efficiency and forecast accuracy; automatically invoking the physical mechanism hydrological model simplifies the forecast process, reduces human intervention, and improves the automation level of the forecast system; and storing the forecast results in the forecast results database ensures data traceability and availability, supporting subsequent analysis and decision-making.

[0033] Furthermore, the present invention describes the specific process of predicting hydrological forecast results for Type II forecast nodes based on machine learning models, including data acquisition, processing, model invocation, and result storage; combining the forecast results of upstream Type I and Type II nodes, and using machine learning models for comprehensive prediction to improve the forecast accuracy and reliability of Type II nodes; by integrating multi-source data (flow results and meteorological data of upstream Type I and Type II nodes), making full use of limited data resources to improve the model's predictive performance; ensuring that the prediction of Type II nodes is based on comprehensive upstream data, achieving coherence and consistency in forecast results, and enhancing the overall forecast capability of the system; adapting to the forecasting needs of Type II nodes in complex water systems, achieving efficient hydrological prediction through advanced machine learning methods, and meeting diverse forecasting application scenarios.

[0034] Furthermore, this invention relates to the preprocessing of historical hydrological and meteorological data, including data collection, outlier removal, missing value imputation, data format standardization, and storage in a database, covering a variety of meteorological and hydrological parameters. By removing outliers and imputing missing values, the integrity and accuracy of historical data are ensured, providing a high-quality data foundation for model training and forecasting. Standardizing data formats and time scales avoids forecasting errors caused by data inconsistencies, improving the stability and reliability of model operation. Centralized storage in a database facilitates rapid data retrieval and management, supporting large-scale data processing and real-time forecasting needs. Covering a variety of key meteorological and hydrological parameters provides comprehensive data support, enhances the model's ability to simulate complex hydrological processes, and improves the overall accuracy of forecasts. Attached Figure Description

[0035] Figure 1 This is a flowchart illustrating the multi-level hydrological forecasting method for river systems based on dual-mode driving provided by the present invention.

[0036] Figure 2 A schematic diagram of the structure of the multi-level hydrological forecasting device for river systems based on dual-mode drive provided by the present invention;

[0037] Figure 3 This is a schematic diagram of the forecast node classification results provided by the present invention;

[0038] Figure 4 A schematic diagram showing the detailed calculation logic of the dual-mode driven multi-level hydrological forecasting system for watersheds provided by this invention. Detailed Implementation

[0039] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments to facilitate a clear understanding of the present invention, but these descriptions do not constitute a limitation on the present invention.

[0040] Example 1

[0041] like Figure 1 As shown in the figure, the multi-level hydrological forecasting method for river systems based on dual-mode driving provided by this invention includes the following steps:

[0042] Step 1: Data Acquisition and Cleaning. Identify the study watershed, collect hydrological and meteorological data, remove outliers, fill in missing values, standardize the data format, and store the data in a database.

[0043] Specifically, hydrological and meteorological data include precipitation, water level, flow rate, temperature, humidity, air pressure, wind speed, evaporation, etc., and may include all of the above types of data or only some of them; the data format may be hour, day, month, year, etc.

[0044] Furthermore, the collected data undergoes preliminary cleaning, outliers are detected and removed, and missing values ​​are imputed to ensure the consistency of time scales for meteorological and hydrological data, while also ensuring that the start and end times of the data are consistent and continuous.

[0045] Optionally, outliers in rainfall and evapotranspiration meteorological data can be identified through meteorological thresholds, statistical outliers, temporal consistency of station data, and spatial consistency with data from neighboring stations.

[0046] Outliers in flow and water level hydrological data can be identified through methods such as statistical outlier detection, time consistency of nearby stations, verification of water level-flow relationship, and machine learning.

[0047] Optionally, for missing values ​​in meteorological data such as rainfall and evapotranspiration, when the missing time period is small, imputation can be performed by using data from previous and subsequent times, data from nearby stations, meteorological satellites, and reanalysis products. When there are many consecutive missing values, the data for that time period should be discarded.

[0048] For missing values ​​in hydrological data such as flow rate and water level, when the missing time period is small, the missing values ​​can be interpolated using observation data from the preceding and following time periods. When there are many consecutive missing values, the data for that time period should be discarded.

[0049] Furthermore, the processed data is stored in a relational database or spatiotemporal database to facilitate subsequent retrieval, querying, and modeling applications. The standardized data format provides a reliable foundation for subsequent data analysis and the use of hydrological forecasting models.

[0050] Step 2: Forecast Node Identification and Classification. Based on the hydrological and meteorological data compiled in Step 1, forecast nodes in the river system of the basin are identified and classified.

[0051] Figure 3 This is the forecast node classification result formed by a watershed containing multiple topological nodes such as reservoirs, hydrological stations, and water level stations. Figure 3 Let's take an example to elaborate on the details.

[0052] Specifically, based on the spatial distribution of monitoring stations such as hydrological stations and water level stations and reservoirs and water conservancy projects within the basin, and combined with the basin's topography, river system characteristics, historical flood event characteristics, water conservancy project scheduling needs and flood and drought disaster prevention needs, important monitoring stations and water conservancy projects in the basin's river system are identified.

[0053] More specifically, these identified stations or reservoirs and water conservancy projects are forecast nodes, which can provide key data support for watershed hydrological forecasting and can serve as the basis for subsequent hydrological forecasting model calculations.

[0054] Furthermore, based on the observation elements of the forecast nodes and the quality of their hydrological data (such as accuracy, continuity, and completeness), forecast nodes are classified into Type I (reservoirs A and E, hydrological stations BD and F) and Type II (hydrological station G). In practical applications, specific indicators and threshold ranges can be selected based on expert experience or actual operating conditions for different application objects to identify Type I and Type II forecast nodes according to the following discrimination criteria.

[0055] Specifically, Type I forecasting nodes refer to nodes that possess complete and reliable hydrological and runoff data. These stations are typically located at key river sections within a watershed and can collect various hydrological data, including flow and water level. This data effectively reflects the hydrological dynamics within the watershed and possesses relatively complete historical data sequences, providing support for the calibration and correction of physical mechanism hydrological models. Therefore, nodes that are hydrological stations with relatively complete data and comprehensive measurement elements are defined as Type I forecasting nodes.

[0056] Specifically, Type II forecast nodes generally refer to stations, such as those downstream of reservoir dams, that can only observe water levels or physical mechanism hydrological models due to river characteristics or the influence of water conservancy projects. The definition of Type II nodes can also be extended to scenarios where other physical hydrological models are not applicable, including but not limited to areas with a large amount of missing flow observation data and limited observation equipment. Because Type II forecast nodes lack sufficient supporting data, it is difficult to use complex physical process models for prediction. However, they may have good correlations with the hydrological elements of Type I forecast nodes upstream and downstream. By leveraging these correlations, the hydrological conditions of Type II nodes can be inferred from the hydrological elements of the associated nodes using a data-driven approach. Through data fusion and inference with neighboring nodes, these nodes can still predict their observed hydrological elements relatively accurately, and are therefore defined as Type II forecast nodes. In the watershed hydrological forecasting system of this invention, their forecasting order follows that of Type I forecast nodes.

[0057] Preferably, the Type I forecast node is defined as follows:

[0058] Station type: Hydrological station (must conform to the definition in the "Hydrological Station Network Planning Guidelines").

[0059] Completeness of information:

[0060] Data elements: at least five consecutive years of reliable flow observation data (with reliability assessment indicators meeting the set requirements), or sufficient flood data to support model calibration.

[0061] Time series completeness: missing rate ≤5% (after completion by interpolation or data from neighboring sites).

[0062] Measurement accuracy: The flow measurement error meets the requirements of the "Hydrological Measurement Specification".

[0063] Physical model applicability: The upstream area of ​​the station is not subject to regulation by large-scale water conservancy projects and is not significantly affected by human activities. The river level-discharge relationship is stable, and the river morphology remains stable. According to the accuracy assessment standards of the "Hydrological Information Forecasting Specification", the accuracy of the physical process model simulation can reach Grade B or above.

[0064] Example:

[0065] Hydrological station A monitors flow rate, water level, rainfall, and sediment content, with a data completeness of 98%. It is not directly regulated by upstream reservoirs, and the riverbed scouring and silting rate is 5%. It meets the above definition requirements and is classified as a type I node.

[0066] Type II forecast nodes are defined as follows:

[0067] Observation limitations: Only water level stations that can acquire water level data.

[0068] Alternatively, the physical model may be unsuitable: Due to one or more of the following factors, such as significant reservoir regulation, strong human activity interference, complex topography, obvious tidal effects, intense sediment movement, variable meteorological conditions, insufficient data support, and significant ecological or vegetation impacts, the physical process model can only achieve a simulation accuracy of grade C or below in this area according to the accuracy assessment standards of the "Hydrological Information Forecasting Specifications", indicating that the model cannot meet the accuracy requirements in practical applications.

[0069] Example:

[0070] Node B, located 10km downstream of a large reservoir, experiences drastic water level fluctuations due to daily reservoir regulation, and the station only measures water level. It meets the above definition requirements and is classified as a Type II node.

[0071] Furthermore, based on the identification and classification of forecast nodes, a forecast topology for the watershed is generated. The watershed forecast topology is a network based on river flow direction, with nodes linked according to their river system connections. This structure clearly represents the type of each forecast node, its upstream and downstream relationships, and its spatial location relative to other nodes. The forecast model can then rationally access data from the corresponding upstream or downstream nodes based on the river system topology.

[0072] Furthermore, after completing node identification, classification, and topology generation, the information of these forecast nodes is stored in a database. Node data includes station location, monitoring data type, forecast data type, classification result (Type I or Type II node), and its correlation in the forecast topology. The database provides unified management of forecast nodes, enabling flexible retrieval of relevant node information in subsequent model building, calibration, validation, and real-time prediction.

[0073] Step 3: Construct a hydrological forecasting model library supporting dual-mode driving. Based on the classification of Type I and Type II forecast nodes, different hydrological model frameworks are built, forming a physical mechanism hydrological model library and a data-driven hydrological model library suitable for the watershed. The construction of the dual-mode driven hydrological forecasting model library includes, but is not limited to, model definition, definition of model input and output interfaces, and model registration.

[0074] Specifically, Type I nodes typically possess relatively complete historical hydrological data (such as flow rate and water level) and relatively clear rainfall-yield and runoff characteristics, making them suitable for physical process-based models (such as rainfall-runoff models, lumped models, and distributed hydrological models). For Type I forecast nodes, a physical process-based hydrological model framework is constructed and integrated and registered in a model library.

[0075] More specifically, the steps for building and integrating Type I forecast nodes into the model library may include:

[0076] Model input definition: Model input includes hydrological and meteorological elements such as rainfall, evaporation, flow rate, and water level (which can be one or more of these elements, or other inputs that the model depends on), as well as the coding, definition, and value range of model parameters of the specific physical mechanism hydrological model. The format, time scale, and data source of these elements are defined.

[0077] ② Output parameter specifications: The model output includes one or more of the following elements: forecast node flow, interval water volume, forecast node water level, etc. Ensure that the output parameters match the forecast requirements and can provide the necessary elements for subsequent hydrological forecasts.

[0078] ③ Model interface preparation: Build the input and output interfaces of the physical mechanism hydrological model to ensure that measured data, rainfall forecast data and watershed characteristic data can be easily connected in the future.

[0079] Furthermore, for Type II forecast nodes, a hydrological model framework based on machine learning is built. The steps for building and integrating this model into the model library may include:

[0080] ① Input Parameter Definition: The model's input data can include historical water level data, precipitation, meteorological data (such as temperature and humidity), and correlated data from nearby forecast nodes. Ensure that the input parameters conform to the architecture requirements of the machine learning model, and that the data format and time scale are consistent. Furthermore, the input data should include relevant model parameters, such as the selection of input features and the importance of those features.

[0081] ② Output parameter specifications: The output data of machine learning models are usually predicted water levels, flow rates, or other key hydrological variables. Ensure that the model output meets forecasting requirements, especially accuracy and timeliness, and matches the hydrological elements required for watershed hydrological management and scheduling decisions.

[0082] ③ Model Architecture Preparation: Build the basic architecture of the machine learning model, including defining the input layer, hidden layers, and output layer, ensuring the architecture can adapt to capturing time-series or spatially correlated features. Alternative machine learning architectures include, but are not limited to, deep learning models such as regression analysis, random forests, Long Short-Term Memory (LSTM) networks, and Transformers. Note that this step does not involve actual model training, but rather building a framework and preparing for subsequent model training, validation, and optimization.

[0083] Furthermore, the physical process model framework and the machine learning model framework are stored in the physical process model library and the data-driven model library, respectively. This ensures that the system can flexibly call the corresponding model for subsequent calibration, validation, and prediction based on the classification of different nodes.

[0084] Step 4: For each Type I forecast node, construct and calibrate a hydrological model based on physical mechanisms using the following steps.

[0085] Specifically, for a given Type I forecast node, based on the water system topology and node division established in step 2, information such as the catchment area of ​​the sub-basin and the upstream inflow node is extracted to support the input requirements of the physical mechanism hydrological model.

[0086] Furthermore, for this Type I forecast node, based on its river system topology information, the required meteorological and hydrological data are extracted from the data cleaned in step 1.

[0087] Specifically, these data typically include meteorological data for the forecast interval (such as rainfall, evaporation, etc.), inflow hydrological data from upstream nodes (such as flow rate, water level, etc.), and the node's own hydrological data. These data can be one or more of the above-mentioned data, and the specific selection depends on the requirements of the chosen physical mechanism hydrological model.

[0088] Furthermore, for this Type I forecast node, an appropriate alternative physical mechanism hydrological model is selected, and the required hydrological and meteorological data formats and model parameters are organized according to the model input parameter format defined in step 3.

[0089] In practical applications, to improve forecast accuracy and adapt to different forecasting needs, the characteristics of the forecasting object should be comprehensively considered and an appropriate forecasting model should be selected.

[0090] Based on their different structures and applicable scopes, forecasting models can be divided into distributed models and lumped models. Distributed hydrological models can include hydrological models based on raster data and hydrological response units, which can accurately simulate the hydrological processes of each unit within the watershed, including but not limited to VIC and SWAT. Lumped models, on the other hand, are usually based on fewer parameters and assume that the hydrological characteristics of the watershed are uniformly distributed, ignoring the spatial distribution characteristics of meteorological and topographical elements, but can complete calculations quickly and provide a certain degree of accuracy, including but not limited to the Xin'anjiang model, HBV model, and HEC-HMS.

[0091] To accurately reflect the evolutionary relationships between river channels, the model design should divide the hydrological process into a section runoff calculation module and a channel confluence calculation module. The channel confluence calculation module simulates the confluence characteristics, velocity variations, and flow interactions of different river segments.

[0092] For reservoirs in the computational topology nodes, the model should include a reservoir regulation module to simulate the reservoir's role in regulating downstream hydrological processes and the reservoir's own safety status.

[0093] Specifically, it is essential to ensure that meteorological data (such as rainfall and evaporation) and hydrological data (such as flow rate and water level) meet the model's input requirements, including time resolution, data format, and unit conversion. Simultaneously, model parameters (such as catchment area, watershed slope, and permeability coefficient) must be coded, defined, and have their value ranges confirmed according to the model's specific needs. This ensures that the model can correctly read and process input data during subsequent calibration phases and can be optimized based on accurate parameter settings.

[0094] Furthermore, for this Type I forecast node, based on the well-organized dataset, the parameters of the physical mechanism hydrological model were calibrated by dividing the test data and calibration data.

[0095] Specifically, one or more of the following indicators are used as model evaluation metrics to comprehensively assess the model's forecasting performance: Mean Absolute Error (MAE), Nash Efficiency Coefficient (NSE), Root Mean Square Error (RMSE), Peak Flood Error, and Peak Flood Time Difference (h). Peak Flood Error is the difference between the model-simulated peak flood value and the measured peak flood value; Peak Flood Time Difference (h) is the difference between the model-simulated peak flood time and the measured peak flood time. The formulas for calculating MAE, NSE, and RMSE are as follows:

[0096] ;

[0097] In the formula: N is the number of data points, Q i for i Measured runoff at time P; i for i Predicted runoff at time Q; avgThis represents the average value of the measured runoff.

[0098] Furthermore, taking into account the above evaluation indicators, models and parameters with better forecast performance are selected and saved to the Type I forecast node model results library, which will be used as the model for subsequent forecasts of the Type I forecast node.

[0099] Step 5: For each Type II forecast node, construct and train a machine learning model using the following steps.

[0100] Specifically, read historical water level, flow rate, precipitation, evapotranspiration and other data (one or more) of a specified Type II forecast node and its upstream and downstream neighboring forecast nodes from the database cleaned in step 1; organize the required hydrological and meteorological data according to the model input parameter format defined in step 3, and prepare the input for the machine learning model of the Type II forecast node.

[0101] Furthermore, the extracted dataset undergoes standardization, typically using normalization methods to scale the data to a uniform range (e.g., [0, 1]). The formula is as follows:

[0102] ;

[0103] In the formula, x represents the original data value, x' represents the normalized value, and min(X) and max(X) are the minimum and maximum values ​​of the feature in the dataset, respectively.

[0104] The alternative machine learning models can employ deep learning methods such as random forests and long short-term memory networks.

[0105] Specifically, when using ensemble learning methods such as random forests for hydrological forecasting, a lag method is employed to generate training samples. This involves setting a certain number of lag steps, combining the current data with data from several past time points to form a feature vector, thus creating new training samples.

[0106] Specifically, when using deep learning methods such as Long Short-Term Memory (LSTM) networks, a sliding window method is employed to generate training samples. This involves setting a fixed-length window, the size of which determines how many time points each training sample will contain. Starting from the beginning of the time series data, the window is moved step-by-step, generating a new training sample with each move, thus capturing the temporal dependencies of the data.

[0107] Furthermore, the dataset can be divided into training and testing sets. For example, 80% of the historical data can be allocated to the training set and 20% to the testing set.

[0108] Furthermore, select appropriate machine learning methods (random forest, long short-term memory network, etc.) and conduct model training, validation, and testing based on the loss function;

[0109] Furthermore, for the trained model, taking into account the above evaluation indicators such as MAE, NSE, RMSE, peak error and peak occurrence time difference, it is saved to the Type II forecast node model result library and used as the model for subsequent forecasts of the Type II forecast node.

[0110] Step 6, Multi-level Hydrological Forecasting of the River Basin. Based on the river basin topology and model results database constructed in Steps 2-5, a multi-level hydrological forecasting system for the river basin is established.

[0111] Figure 4 Taking a forecasting topology that includes various water conservancy objects such as reservoirs, hydrological stations, and water level stations as an example (based on...) Figure 3 The resulting topology, based on the classification of forecast nodes, forms the model's calculation content and logic.

[0112] Specifically, the cross-sectional flow calculation for Type I forecast nodes is carried out first: real-time meteorological forecast data is accessed from external data sources, and the forecast data required for each calculation interval is matched according to the division of the watershed. The model input parameters are organized according to the definition of the model input parameters in step 3, and the model results that have been constructed and stored in step 4 are called. The cross-sectional flow calculation for Type I forecast nodes is completed step by step from the upstream, and the calculation results are stored in the forecast results library.

[0113] The cross-sectional flow calculation for Type I forecast nodes needs to be performed sequentially according to the spatial distribution of the watershed. In step 2, the forecast nodes have been identified and classified, and the watershed's river system topology has been constructed. This topology determines the calculation order of the forecast nodes. Based on the topology, the calculation path from upstream to downstream is determined to ensure that the flow calculation for each node is based on the latest upstream forecast results. An example of the process is as follows:

[0114] Phase 1: Upstream Nodes

[0115] Input data preparation: Access real-time weather forecast data and organize model input parameters.

[0116] Flow calculation: Use the physical process hydrological model built in step 4 to calculate the cross-sectional flow of the upstream node.

[0117] Results storage: The calculation results are stored in the forecast results library as input data for downstream nodes.

[0118] Phase Two: Downstream Nodes

[0119] Dependent on upstream results: Utilize the flow results of upstream nodes in the forecast results database and local real-time weather forecast data as inputs to the downstream node model.

[0120] Flow calculation: Call the corresponding physical process hydrological model to calculate the cross-sectional flow of downstream nodes.

[0121] Results storage: The calculation results are stored in the forecast results repository for use by further downstream nodes.

[0122] Calculations are performed according to the order of watershed distribution, ensuring that all nodes forecast at the same time step, avoiding data misalignment caused by inconsistent time steps. During the forecasting process, the forecast results of upstream nodes can be dynamically adjusted based on real-time data, promptly influencing the predictions of downstream nodes and improving the overall forecast response speed and accuracy. In some complex watersheds, independent sub-watersheds can be identified, allowing parallel computation while maintaining spatial order, thus improving computational efficiency. Calculations are performed according to the order of watershed distribution, which facilitates modular design, making the forecasting system easier to maintain and upgrade.

[0123] Furthermore, after completing the Type I node calculation, the model input parameters are organized according to the definition of the model input parameters in step 3. The model results trained and stored in step 5 are called, and the associated Type I node forecast results are read from the forecast result library to carry out the hydrological prediction results for the Type II forecast nodes. After the forecast is completed, the calculation results are also stored in the database.

[0124] In a watershed, water flows from upstream to downstream. Hydrological processes at upstream nodes (such as precipitation converting into runoff) directly influence the hydrological conditions at downstream nodes. Therefore, forecasts from upstream nodes (whether Type I or Type II) are crucial inputs for downstream node predictions. Predictions from multiple upstream Type II nodes can interact spatially and temporally, accumulating to significantly impact downstream hydrological forecasts. Type II nodes typically have limited data and rely on predictions from multiple upstream nodes to supplement information, thereby improving forecast accuracy. Downstream Type II nodes may be influenced by multiple upstream nodes, including changes in hydrological processes at different temporal and spatial scales. Therefore, integrating predictions from multiple upstream nodes provides a more comprehensive reflection of downstream hydrological dynamics.

[0125] Machine learning models for Type II nodes typically rely on multiple input variables, including local meteorological data, flow data from upstream Type I nodes, and predictions from upstream Type II nodes. This multi-source data provides more information and improves the model's predictive ability. Predictions from upstream Type II nodes can be used as new features input into the downstream Type II node model, helping it capture more complex hydrological processes. In a multi-level hydrological forecasting system, Type II node predictions not only depend on Type I nodes but may also rely on Type II nodes at the same or higher levels. This hierarchical structure allows for a more detailed simulation of complex hydrological processes within the watershed, improving the overall accuracy and reliability of the forecast.

[0126] Based on the watershed's topology, the calculation order is determined to ensure that for each Type II node, the prediction results for all its upstream nodes (including Type I and Type II nodes) are ready when predictions are performed. Calculations are performed according to the topological order to ensure that downstream nodes can obtain all necessary upstream node prediction results, avoiding data loss or inconsistencies. The specific process is as follows:

[0127] Phase 1: Calculation of Type I Nodes

[0128] Access real-time meteorological data (including precipitation, temperature, etc.) and upstream node flow forecasts.

[0129] Flow calculation: Calculate the cross-sectional flow of type I nodes using a physical process model.

[0130] Results storage: Store the flow results in the forecast results database.

[0131] Phase Two: Prediction of Upstream Type II Nodes

[0132] Results dependent on Type I nodes: Use the traffic predictions from Type I nodes as input.

[0133] Traffic calculation: Use machine learning models to predict the traffic of upstream Type II nodes.

[0134] Results storage: Store the prediction results in the forecast results database.

[0135] Phase 3: Prediction of Downstream Type II Nodes

[0136] Results dependent on Type I and upstream Type II nodes: Use the traffic predictions of Type I nodes and all upstream Type II nodes as input.

[0137] Traffic calculation: Use machine learning models to predict the traffic of downstream Type II nodes.

[0138] Results storage: Store the prediction results in the forecast results database.

[0139] The forecasting system should be designed to automatically call the models of Type I and Type II nodes sequentially according to the watershed topology, ensuring the timeliness and accuracy of data transmission. Through the model input / output interfaces defined in step 3, seamless transmission and integration of forecast results from Type I and Type II nodes should be ensured, supporting the efficient operation of data-driven models. The forecast results repository needs to support the storage and retrieval of forecast results from each node in watershed topology order, ensuring that Type II nodes can quickly obtain all necessary upstream data. A real-time data update mechanism should be established to ensure that Type II nodes use the latest upstream forecast results during forecasting, improving the real-time performance and accuracy of the forecast.

[0140] Finally, the calculation results from Class I and Class II stations are integrated to generate multi-level hydrological forecasts for the entire watershed. The combined application of models ensures the comprehensiveness and accuracy of the forecasts. The final watershed-wide forecast results will be stored in a unified database for subsequent decision analysis and system applications.

[0141] Example 2

[0142] Figure 2 This is a schematic diagram of a multi-level hydrological forecasting device for river systems based on dual-mode drive, including:

[0143] The data acquisition and cleaning module is used to acquire historical hydrological data from watershed rain gauges, hydrological stations, and water level stations, as well as real-time rainfall forecast data, remove outliers, fill in missing values ​​using linear interpolation, and standardize the data format.

[0144] The forecast node management module is used to identify forecast nodes in the river system of the basin based on historical hydrological and meteorological data, and classify them into Type I and Type II forecast nodes;

[0145] Type I forecast nodes are defined as nodes that are hydrological stations with complete data and comprehensive measurement elements. Type II forecast nodes are defined as nodes that can only observe water level or are not applicable to the physical mechanism hydrological model due to river characteristics or the influence of water conservancy projects.

[0146] The forecast model library module is used to build the corresponding physical mechanism hydrological model for each Type I forecast node and the corresponding machine learning model for each Type II forecast node.

[0147] The hydrological forecasting module is used to predict the hydrological forecasting results of the corresponding Type I and Type II forecasting nodes based on physical mechanism hydrological models and machine learning models, respectively, and generate hydrological forecasting results for the entire basin.

[0148] Example 3

[0149] This invention also proposes an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions to achieve a corresponding method flow or function. When the processor executes the computer program, it implements the operation of a multi-level hydrological forecasting method for river systems based on dual-mode driving, including the following steps:

[0150] Step 1: Data Acquisition and Cleaning. Identify the study watershed, collect hydrological and meteorological data, remove outliers, fill in missing values, standardize the data format, and store the data in the monitoring database.

[0151] Step 2, Forecast Node Identification and Classification. Based on the hydrological and meteorological data and watershed characteristics from Step 1, forecast nodes are identified and classified into Type I and Type II forecast nodes, forming the watershed topology relationships for each forecast node, and stored in the topology relationship database.

[0152] Step 3: Construct a hydrological forecasting model library based on dual-mode driving. For Type I forecasting nodes, construct a hydrological model based on physical mechanisms; for Type II forecasting nodes, construct a machine learning model (such as random forest or LSTM), and store these models in the dual-mode driven watershed hydrological forecasting model library.

[0153] Step 4: For Type I forecast nodes, construct and calibrate a hydrological model based on physical mechanisms, and store the calibration results in the model results library;

[0154] Step 5: For Type II forecast nodes, construct and train a deep learning-based hydrological model, and store the calibration results in the model results library;

[0155] Step 6, Multi-level Hydrological Forecasting of the River Basin. After accessing external rainfall forecast data, the hydrological model constructed and stored in Step 4 is first used to perform hydrological forecasting for Type I nodes within the basin, and the results are stored in the forecast results repository. Then, the machine learning model constructed and stored in Step 5 is used to forecast for Type II nodes, and the final forecast results are stored in the basin hydrological forecast results repository.

[0156] Example 4

[0157] Finally, the present invention also provides a computer-readable storage medium, which is a memory device in a terminal device for storing programs and data. The computer-readable storage medium here may include the built-in storage medium in the terminal device, or it may include extended storage media supported by the terminal device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, the storage space also stores one or more instructions suitable for loading and execution by a processor, which may be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here may be high-speed RAM or non-volatile memory, such as at least one disk storage device.

[0158] One or more instructions stored in a computer-readable storage medium can be loaded and executed by the processor to implement the corresponding steps of the multi-level hydrological forecasting method for river systems based on dual-mode driving in the above embodiments; one or more instructions in the computer-readable storage medium are loaded and executed by the processor in the following steps:

[0159] Step 1: Data Acquisition and Cleaning. Identify the study watershed, collect hydrological and meteorological data, remove outliers, fill in missing values, standardize the data format, and store the data in the monitoring database.

[0160] Step 2, Forecast Node Identification and Classification. Based on the hydrological and meteorological data and watershed characteristics from Step 1, forecast nodes are identified and classified into Type I and Type II forecast nodes, forming the watershed topology relationships for each forecast node, and stored in the topology relationship database.

[0161] Step 3: Construct a hydrological forecasting model library based on dual-mode driving. For Type I forecasting nodes, construct a hydrological model based on physical mechanisms; for Type II forecasting nodes, construct a machine learning model (such as random forest or LSTM), and store these models in the dual-mode driven watershed hydrological forecasting model library.

[0162] Step 4: For Type I forecast nodes, construct and calibrate a hydrological model based on physical mechanisms, and store the calibration results in the model results library;

[0163] Step 5: For Type II forecast nodes, construct and train a deep learning-based hydrological model, and store the calibration results in the model results library;

[0164] Step 6, Multi-level Hydrological Forecasting of the River Basin. After accessing external rainfall forecast data, the hydrological model constructed and stored in Step 4 is first used to perform hydrological forecasting for Type I nodes within the basin, and the results are stored in the forecast results repository. Subsequently, the machine learning model constructed and stored in Step 5 is used to forecast for Type II nodes, and the final forecast results are stored as the basin hydrological forecast results.

[0165] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0166] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0167] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0168] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0169] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

[0170] The contents not described in detail in this specification are existing technologies known to those skilled in the art.

Claims

1. A watershed multi-level hydrological forecasting method based on dual-mode driving, characterized in that: Includes the following steps: Based on historical hydrological and meteorological data, forecast nodes in the river system of the basin were identified and classified into Type I and Type II forecast nodes. Type I forecast nodes are defined as nodes that are hydrological stations with complete data and comprehensive measurement elements. Type II forecast nodes are defined as nodes that can only observe water level or are not applicable to the physical mechanism hydrological model due to river characteristics or the influence of water conservancy projects. For each Type I forecast node, a corresponding physical mechanism hydrological model is constructed. For each Type II forecast node, a corresponding machine learning model is constructed. Based on the physical mechanism hydrological model and the machine learning model, the hydrological forecast results of the corresponding Type I and Type II forecast nodes are predicted respectively, and the hydrological forecast results of the whole basin are generated. The process of predicting the hydrological forecast result for any Type I forecast node based on a physical mechanism hydrological model includes: obtaining the latest meteorological forecast data corresponding to the Type I forecast node based on external data; organizing the cross-sectional flow calculation results and meteorological forecast data of the Type I forecast node upstream of the Type I forecast node into the model input parameter format defined in the model library, and using them as model input parameters; calling the physical mechanism hydrological model corresponding to the Type I forecast node from the model library, and using the organized model input parameters to calculate the cross-sectional flow; and storing the calculated cross-sectional flow results corresponding to the Type I forecast node into the forecast results library. The process of predicting the hydrological forecast result for any Type II forecast node based on a machine learning model includes: obtaining the latest meteorological forecast data corresponding to the Type II forecast node based on external data; organizing the cross-sectional flow calculation results of the Type I forecast node upstream of the Type II forecast node, the hydrological forecast results of the upstream Type II forecast node, and the meteorological forecast data into the model input parameter format defined in the model library, and using them as model input parameters; calling the machine learning model corresponding to the Type II forecast node from the model library, and using the organized model input parameters to perform hydrological forecasting; and storing the calculated hydrological forecast result corresponding to the Type II forecast node into the forecast result library.

2. The method according to claim 1, characterized in that: The construction process of the physical mechanism hydrological model includes: selecting several candidate physical mechanism hydrological models and their model parameters; extracting historical meteorological and hydrological data of the corresponding Type I forecast nodes and organizing them into the form of candidate model input parameters as the corresponding dataset; calibrating the parameters of each candidate model using the corresponding dataset; evaluating the forecasting effect of each candidate model based on the mean absolute error, Nash efficiency coefficient, root mean square error, flood peak error and peak occurrence time, and selecting the model with the best forecasting effect as the physical mechanism hydrological model of the Type I forecast node.

3. The method according to claim 1, characterized in that: The machine learning model construction process includes: selecting several candidate machine learning models and their model parameters; extracting historical meteorological and hydrological data of the corresponding Type II forecast nodes and organizing them into the form of candidate model input parameters as the corresponding dataset; training each candidate model using the corresponding dataset; evaluating the forecast performance of each candidate model based on mean absolute error, Nash efficiency coefficient, root mean square error, flood peak error, and peak occurrence time, and selecting the model with the best forecast performance as the machine learning model for that Type II forecast node.

4. The method according to claim 2 or 3, characterized in that: The model library calls the corresponding predefined basic model frameworks as alternative physical mechanism hydrological models or machine learning models; all basic model frameworks adopt standardized model input and output interfaces; all basic model frameworks have clearly defined functions, applicable scope, and input and output parameter formats; all basic model frameworks have been registered in the model library.

5. The method according to claim 4, characterized in that: Also includes: The watershed's river system topology is constructed based on Type I and Type II forecast nodes. According to the river system topology, the hydrological forecast results of the Type I forecast nodes are calculated sequentially from upstream to downstream, and then the hydrological forecast results of the Type II forecast nodes are calculated sequentially from upstream to downstream. The flow calculation of each node is based on the latest upstream forecast results.

6. The method according to claim 1, characterized in that: The historical hydrological and meteorological data are used in subsequent steps after undergoing the following preprocessing: collecting hydrological and meteorological data of the river system to be predicted, removing outliers, filling in missing values, standardizing the data format, and storing it in the database; the hydrological and meteorological data include precipitation, water level, flow rate, temperature, humidity, air pressure, wind speed, and evaporation.

7. A watershed multi-level hydrological forecasting system based on dual-mode driving, characterized in that: To implement the method of claims 1-6, the method comprises: The forecast node management module is used to identify forecast nodes in the river system of the basin based on historical hydrological and meteorological data, and classify them into Type I and Type II forecast nodes; Type I forecast nodes are defined as nodes that are hydrological stations with complete data and comprehensive measurement elements. Type II forecast nodes are defined as nodes that can only observe water level or are not applicable to the physical mechanism hydrological model due to river characteristics or the influence of water conservancy projects. The forecast model library module is used to build the corresponding physical mechanism hydrological model for each Type I forecast node and the corresponding machine learning model for each Type II forecast node. The hydrological forecasting module is used to predict the hydrological forecasting results of the corresponding Type I and Type II forecasting nodes based on physical mechanism hydrological models and machine learning models, respectively, and generate hydrological forecasting results for the entire basin.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the watershed multi-level hydrological forecasting method based on dual-mode driving as described in any one of claims 1-6.

Images