An improved flow simulation method and system of Xin'anjiang model coupled with neural network

Abstract

This invention discloses an improved Xin'anjiang model coupled with a neural network for flow simulation and a corresponding method and system. The method includes collecting sub-basin data of the target watershed, including DEM data from a semi-distributed model; constructing sub-basin topological relationships, establishing sub-basin boundaries and their mapping relationships with forecast sections, and clarifying the flow mapping rules and data transmission methods between upstream and downstream sub-basins, and between sub-basins and the river channel; configuring the sub-basin dataset with the forecast section as the core, constructing a simulation scheme, and generating basic boundary conditions; parsing the boundary conditions and allocating them to nodes according to the topology for iterative operation, outputting the outflow data of the improved Xin'anjiang model's outlet section; tensorizing the outflow, precipitation, and evapotranspiration data and inputting them into the neural network for training to obtain the simulated flow sequence. This invention combines the advantages of neural network time-series fitting with the physical mechanism of the improved Xin'anjiang model to achieve efficient simulation of watershed outlet section flow, improving the accuracy, timeliness, and refinement of runoff prediction.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of computer science, hydrology, and water resources, and specifically relates to an improved method and system for flow simulation using a coupled neural network in the Xin'anjiang model. Background Technology

[0002] With the increasing demand for refined management and precise control of hydrological infrastructure, traditional stand-alone hydrological models are increasingly revealing significant problems when simulating and analyzing large-scale, complex hydrological systems. These problems include low computational efficiency, difficulties in data sharing, and insufficient simulation accuracy, making them inadequate to meet the practical application needs of current watershed flood prevention and water resource optimization. For example, when simulating complex hydrological scenarios such as flood evolution simulation and water resource allocation optimization in large-scale watersheds, traditional stand-alone hydrological models, limited by hardware computing power and algorithm design, often cannot complete the computation and processing of large-scale hydrological data within the predetermined time and struggle to effectively integrate multi-source data for comprehensive analysis.

[0003] Currently, mainstream physics-based hydrological models still have many limitations when dealing with large-scale watersheds, especially trans-regional watersheds with significant spatiotemporal heterogeneity. Specifically: First, traditional conceptual hydrological models rely heavily on monitoring data and require manual parameter calibration. In watersheds with scarce or no data, insufficient data often renders the model unusable. Furthermore, these models are typically only applicable to specific geographical and climatic zones, exhibiting poor adaptability to different regions and insufficient generalization ability, making them unsuitable for hydrological simulation needs across regions and multiple climate types. Second, while lumped or semi-distributed hydrological models, such as the Xin'anjiang model, can initially characterize the spatial heterogeneity of the watershed's underlying surface through water storage capacity curves, their structural design limits their ability to effectively simulate the non-uniform characteristics of hydrological sequences under climate change, making it difficult to accurately capture the evolution of river runoff. This results in simulation accuracy failing to meet practical application requirements, especially in simulations of extreme hydrological events where errors are significant. Third, distributed hydrological models such as the SWAT (Soil and Water Assessment Tool) model use HRU (Hydrological Response Unit) for watershed aggregation calculations. This aggregation method easily overlooks anomalous hydraulic connections caused by drastic changes in the underlying surface (such as urbanization, abrupt changes in land use types, etc.), resulting in insufficient detail in the model's characterization of hydrological processes within the watershed, thus affecting the accuracy of the simulation results. Fourth, while hydraulic models such as the HEC-HMS (Hydrologic Engineering Center's Hydrologic Modeling System) model can capture the hydrodynamic evolution of watershed channels relatively well, the linear reservoir approximation method used in these models fails to effectively couple the complex river system topology network characteristics revealed by high-precision DEM (Digital Elevation Model) analysis. This makes it difficult to accurately quantify the flow acceleration effect during the sub-watershed merging stage, ultimately causing the model to fail to capture the flow at the watershed outlet section. Summary of the Invention

[0004] The purpose of this invention is to address the problems in the prior art by providing an improved flow simulation method and system using a coupled neural network in the Xin'anjiang model. By utilizing the high-precision fitting capability of neural networks for time-series hydrological data, this invention enables efficient simulation of the flow at the outlet section of the entire basin, thereby improving the accuracy, timeliness, and precision of basin runoff prediction.

[0005] To achieve the above objectives, the present invention provides the following technical solution: Firstly, an improved method for simulating flow in the Xin'anjiang River model coupled with a neural network is provided, including: Collect sub-basin data of the target watershed, including digital elevation model (DEM) data of a semi-distributed model; Construct sub-basin topological relationships, and establish sub-basin boundaries and mapping relationships between sub-basins and target basin forecast sections based on these relationships. Between upstream and downstream sub-basins, the flow calculation results of the upstream sub-basin are used as input boundary conditions for the downstream sub-basin. For the connection between a sub-basin and the river channel, the sub-basin outlet flow is mapped to the corresponding basin outlet section. Mapping rules are formulated to clarify the conversion and transmission methods of different data types between sub-basins. Using the target watershed forecast section as the core, configure the dataset required for each sub-watershed and construct a simulation scheme; Using the constructed simulation scheme, the basic boundary conditions of the target watershed are generated; The basic boundary conditions of the target watershed are analyzed, and various boundary data are distributed to various nodes according to the sub-watershed and water system topology for iterative operation, outputting the outflow data of the improved Xin'anjiang model at the outlet section of the target watershed; The outflow data, precipitation and evapotranspiration data of the target watershed outlet section are tensor-quantized and then input into the neural network for training to obtain the simulated flow sequence of the target watershed outlet section.

[0006] As a preferred approach, in the step of configuring the dataset required for each sub-basin based on the target watershed forecast section, and constructing the simulation scheme, the dataset required for each sub-basin is configured according to flood control needs or water resource conditions. The construction of the simulation scheme includes: Create a calculation scheme and name it; Select the start and end sub-basins, and match the computational region of the simulation scheme according to the topological relationship; Within the calculation area, the exit sections are listed, and the calculation logic for the exit sections is determined; Match various types of data mapped to the corresponding exit section; After determining the various calculation data for the outlet section, set the initial parameters of the model; Identify the reservoirs that have a significant impact on the watershed and set their impact coefficients; Complete the initial calculation configuration for each outlet section according to the topological relationship and save the simulation scheme.

[0007] As a preferred embodiment, the steps of selecting the start and end sub-basins and matching the computational region of the simulation scheme according to the topological relationship include: Select the target watershed simulation section; Analyze topological relationships; Based on the analysis results of the target watershed simulation cross section and topological relationship, the computational region corresponding to the simulation scheme is determined; The steps for analyzing topological relationships include: Retrieve pre-constructed watershed channel topology data and store the upstream and downstream connections of the water system in the form of a directed graph. Use the breadth-first search algorithm in graph theory to search along the topology path from the starting sub-watershed until the end outlet section is reached, and include all water system segments and sub-watersheds involved in the path into the calculation area.

[0008] As a preferred embodiment, the steps of establishing the sub-basin boundary and the mapping relationship between the sub-basin and the target basin forecast sections based on the sub-basin topological relationship include: preliminary selection of forecast sections based on forecast requirements, wherein the forecast sections include flood control forecast sections, water resource management sections, and ecological flow guarantee sections; diagnosis of the fit of the outlet sections; optimization and verification of the outlet sections based on model simulation; determination of the basin outlet sections; and basin outlet section information including at least the section location and basic parameters.

[0009] As a preferred embodiment, the step of collecting sub-basin data of the target watershed obtains the computational data of the sub-basins from different database sources and uses data cleaning methods to remove outliers from the data; The step of collecting sub-basin data of the target watershed also includes sub-basin characteristic analysis, as follows: Calculate the geometric properties of each sub-basin; Analyze the resolution and coverage of sub-basins to determine the accuracy of the sub-basins in describing the spatial characteristics of the watershed; The catchment area and boundary morphology of sub-basins are determined by using a watershed drainage system extraction algorithm. By using spatial overlay analysis, we can compare the spatial overlap of sub-basins under different calculation and analysis tools and identify potential data conflict areas.

[0010] As a preferred embodiment, the steps of constructing sub-basin topological relationships include constructing a river system topology, building upstream and downstream topological relationships of river channels based on the river system of the basin, assigning a unique identifier to each sub-basin of the river system, and establishing a topology table to record the confluence relationships between sub-basins; simultaneously determining the boundaries of sub-basins and establishing a spatial logical topological network between sub-basins; in a semi-distributed model, constructing the local topological structure of sub-basins based on the water flow direction of adjacent sub-basins, and clarifying the transmission path of water flow between sub-basins.

[0011] As a preferred approach, establishing the mapping relationship between the forecast sections of the sub-basin and the target basin includes spatial overlay analysis, distance overlay analysis, and topological relationship analysis.

[0012] As a preferred embodiment, the steps of analyzing the basic boundary conditions of the target watershed, allocating various boundary data to various nodes according to sub-watersheds and river system topology for iterative operation, and outputting the outflow data of the improved Xin'anjiang model at the outlet section of the target watershed include: Start the simulated work process; The task of parsing nodes is executed to decompose the basic boundary conditions of the target watershed and, in combination with the constructed sub-watershed topology, clarify the allocation rules for various types of boundary data. Data reading and computation preparation; The calculation logic is planned, and the iterative calculation method is determined by combining the sub-basin topology and outlet section requirements; Determine the boundary conditions and, in conjunction with the actual hydrological characteristics of the target watershed, determine the iterative convergence criteria; Perform iterative calculations according to the planned computational logic and output the traffic data corresponding to each node; Monitor and update the status, and after the iterative run is completed, output the outflow data of the target watershed outlet section.

[0013] As a preferred embodiment, the step of tensor-quantizing the outflow data, precipitation, and evapotranspiration data of the target watershed outlet section and then inputting them into a neural network for training to obtain the simulated flow sequence of the target watershed outlet section includes: Outflow data of the target basin outlet section were derived from the improved Xin'anjiang model; The collected precipitation and evapotranspiration data were weighted by watershed area. The outflow data, weighted precipitation data, and evapotranspiration data at the outlet section are merged to construct an integrated feature vector of outflow, precipitation, and evapotranspiration. Tensor quantization is performed on the integrated feature vectors of outflow, precipitation, and evapotranspiration to unify the data dimensions; Configure and initialize the core parameters of the Long Short-Term Memory (LSTM) neural network; The tensor-quantized feature vectors are input into the initialized long short-term memory neural network, and through multiple rounds of iterative calculations, the simulated flow sequence of the target watershed outlet section is obtained.

[0014] Secondly, an improved flow simulation system using a coupled neural network based on the Xin'anjiang model is provided, comprising: The sub-basin data collection module is used to collect sub-basin data of the target basin, including digital elevation model (DEM) data of the semi-distributed model. The mapping relationship establishment module is used to construct sub-basin topological relationships, establish sub-basin boundaries and mapping relationships between sub-basins and target basin forecast sections based on the sub-basin topological relationships; between upstream and downstream sub-basins, the flow calculation results of the upstream sub-basin are used as the input boundary conditions of the downstream sub-basin; for the connection relationship between the sub-basin and the river, the outlet flow of the sub-basin is mapped to the corresponding basin outlet section; and mapping rules are formulated to clarify the conversion and transmission methods of different types of data between sub-basins. The simulation scheme construction module is used to construct simulation schemes by configuring the datasets required for each sub-basin, with the target watershed forecast section as the core. The basic boundary condition generation module is used to generate basic boundary conditions for the target watershed using the constructed simulation scheme. The outlet section outflow data acquisition module is used to analyze the basic boundary conditions of the target watershed, distribute various boundary data to various nodes according to sub-watersheds and river system topology for iterative operation, and output the outlet section outflow data of the improved Xin'anjiang model in the target watershed. The neural network simulation module is used to process the outflow data, precipitation and evapotranspiration data of the target watershed outlet section into tensor quantization and then input them into the neural network for training, so as to obtain the simulated flow sequence of the target watershed outlet section.

[0015] Compared with the prior art, the present invention has at least the following beneficial effects: By collecting sub-basin data and constructing topological relationships, the problems of insufficient spatial characterization and chaotic data transmission in traditional models are solved. The sub-basin data of the target watershed include digital elevation model (DEM) data from a semi-distributed model. By collecting DEM data and constructing sub-basin topological relationships, the boundaries of sub-basins can be accurately delineated and the upstream and downstream connections between sub-basins can be clarified. This solves the problem that traditional lumped hydrological models cannot reflect the spatial differences of watersheds (such as topography and confluence paths), making the division of sub-basins more consistent with the actual hydrological characteristics of the watershed. At the same time, each sub-basin corresponds to a computing node, realizing the fine-grained decomposition of watershed hydrological simulation. By establishing a mapping relationship between sub-basins and the forecast sections of the target basin, this invention clarifies the rules for "upstream flow as downstream input boundary condition" between upstream and downstream sub-basins and "sub-basin outlet flow mapped to the basin outlet section" between sub-basins and the river channel. Simultaneously, it formulates conversion and transmission methods for different types of data, solving the problems of chaotic sub-basin data transmission and unclear boundary conditions in traditional models. This ensures that the computational data of each sub-basin can be transmitted and summarized in an orderly manner according to topological relationships, improving the coherence and reliability of the simulation process. This invention uses the target basin forecast section as the core, configures the required datasets for each sub-basin, and constructs a simulation scheme. It can specifically configure the datasets of each sub-basin based on the actual forecasting needs of the basin (such as flood control and water resource management), avoiding the problems of "generalized configuration and insufficient specificity" in traditional models. Furthermore, based on the basic boundary conditions generated by the simulation scheme, it can accurately match the iterative operation requirements of the subsequently improved Xin'anjiang model, ensuring that the boundary conditions are compatible with the sub-basin topology and computational node configuration, reducing simulation errors caused by unreasonable boundary conditions, and improving the model's adaptability to the target basin. This invention distributes boundary data to corresponding computational nodes according to sub-basins and river system topology. Each node independently runs the improved Xin'anjiang model, enabling accurate calculation of runoff generation and confluence processes in each sub-basin. The data is then aggregated to the outlet section through topological relationships, solving the problem of traditional models that "calculate the whole and ignore local runoff generation and confluence differences." This improves the calculation accuracy of outflow data at the outlet section, providing high-quality foundational data for subsequent neural network training. By distributing data according to topological relationships and iteratively running the model, the calculation results of each sub-basin can be connected in real time, ensuring that the flow data of upstream sub-basins is promptly transmitted to downstream, avoiding computational gaps. Simultaneously, the iterative running mode can gradually correct calculation deviations, ensuring that the output outflow data truly reflects the actual hydrological processes of the basin, laying a solid foundation for subsequent coupled neural network optimization. While the improved Xin'anjiang model possesses a clear physical mechanism, its ability to fit hydrological time-series data (such as the dynamic correlation between precipitation, evapotranspiration, and flow) is insufficient. This invention addresses this by tensorizing outflow, precipitation, and evapotranspiration data, unifying data dimensions to suit the input requirements of neural networks, and leveraging the neural network's strength in capturing time-series data correlations to optimize the outflow data output by the improved Xin'anjiang model. This effectively corrects the simulation bias of the physical model and improves the accuracy of outflow simulation at the outlet section. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is a flowchart of the improved Xin'anjiang model coupled with a neural network in Embodiment 1 of the present invention for traffic simulation. Figure 2 This is a schematic diagram illustrating the working principle of the flow simulation system based on the improved Xin'anjiang model coupled with a neural network, as described in Embodiment 2 of the present invention. Figure 3 This is a schematic diagram of the flow simulation system structure of the improved Xin'anjiang model coupled with a neural network according to Embodiment 2 of the present invention; Figure 4 This is a schematic diagram of an electronic device for implementing the improved Xin'anjiang model coupled with a neural network for flow simulation according to the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, those skilled in the art can obtain other embodiments without creative effort.

[0019] This invention proposes an improved flow simulation method using a coupled neural network based on the Xin'anjiang model. By achieving efficient integration and collaborative operation between the hydrological model and the neural network, the scientific nature and accuracy of flood disaster prevention and water resource allocation are improved.

[0020] Example 1 Please see Figure 1 The present invention improves the flow simulation method of the Xin'anjiang model coupled with a neural network, including: Step 1000: Obtain the digital elevation model, land use data, soil type data, precipitation data, and evapotranspiration data for the target watershed. In order to support the mixed construction of the model in different sub-watersheds within the watershed, a unified basic outlet section needs to be constructed.

[0021] The construction of basic exit sections is divided into four stages: preliminary data collection and analysis, initial selection of exit sections based on forecast requirements, verification of exit section suitability, and section optimization and final determination. The specific details are as follows: Step 1010, Watershed Basic Data Collection and Analysis, includes the following sub-steps: Step 1011 involves collecting topographic and geomorphological data within the watershed, including a high-precision digital elevation model (DEM). Software such as ArcSWAT is used to delineate sub-watershed boundaries and extract drainage systems, clarifying basic information such as the area and river network distribution of each sub-watershed. Simultaneously, land use data and soil type data, as well as other underlying surface data, are acquired, as these data will influence precipitation runoff processes.

[0022] Historical hydrological and meteorological data were collected, covering meteorological elements such as precipitation and evapotranspiration over many years, as well as flow data from hydrological stations. Data analysis tools such as Python and R were used to perform statistical analysis on the historical data, plotting precipitation-runoff relationship curves to understand the basic hydrological change patterns of the watershed.

[0023] Step 1012: Review of water conservancy projects and monitoring networks; Statistics should be compiled on water conservancy projects such as reservoirs that have a major impact on the basin. For reservoirs, basic parameters such as their capacity should be obtained.

[0024] Determine the distribution of existing hydrological monitoring stations, including the location, monitoring period, and data quality of rain gauges and flow meters. Analyze the coverage and representativeness of the monitoring stations to determine if any monitoring blind spots exist.

[0025] Step 1020: Preliminary selection of outlet sections based on forecast requirements. Outlet sections include flood control forecast sections, water resource management sections, and ecological flow guarantee sections. The specific selection methods are as follows: 1. Selection of flood forecast sections In key flood control areas, such as cities, densely populated areas, and areas surrounding important infrastructure, cross-sections that effectively reflect the flood's evolution should be selected. These cross-sections are typically chosen from river sections that are straight, have regular cross-sections, and lack significant backwater or diversion. Forecasting cross-sections are usually selected at a suitable distance upstream of the flood protection zone to provide early warning. For example, an outlet cross-section might be selected in a stable river section 10-20 kilometers upstream of a city, allowing sufficient response time for urban flood control based on historical flood propagation times.

[0026] Meanwhile, considering the connection with water conservancy facilities such as reservoirs, an outlet section is selected at an appropriate location downstream of the reservoir to monitor the impact of reservoir discharge on the downstream river channel.

[0027] 2. Selection of water resource management sections To address water resource management needs, outlet sections are selected at key water intakes, areas with concentrated water users, and transboundary rivers. Sections upstream of water intakes are selected to forecast inflow and ensure water supply security; sections along transboundary rivers are used to monitor water volume changes and provide data support for water resource allocation. For example, outlet sections are selected at the confluence of inter-provincial rivers to monitor water volume in real time and prevent water resource disputes.

[0028] 3. Selection of ecological flow guarantee sections In ecologically sensitive areas, such as wetlands, fish spawning grounds, and habitats of rare aquatic organisms, outlet sections that reflect ecological water demand should be selected. Combining ecological protection goals and ecological flow indicators, outlet sections should be selected in key river sections to ensure the ecological base flow of the river. For example, an outlet section can be selected upstream of a wetland reserve, and water resources can be rationally allocated through flow forecasting to maintain the stability of the wetland ecosystem.

[0029] Step 1030, Export Section Fit Verification, includes the following sub-steps: Using collected historical hydrological data, mathematical statistical methods, such as correlation analysis and frequency analysis, are employed to evaluate the correlation between the data from the initially selected outlet section hydrological station and the discharge data from surrounding hydrological stations. The correlation coefficient between the discharges of adjacent hydrological stations is calculated; a higher value indicates better hydrological representativeness of the outlet section. If the correlation coefficient is below 0.7, it is necessary to reselect the outlet section or adopt measures such as data interpolation to improve representativeness.

[0030] Step 1040, Final selection of the watershed outlet section: Based on the selected watershed outlet section, it is necessary to use software such as ArcSWAT to spatially overlay the section with the river, and configure the topological relationship of the watershed outlet section according to the hierarchical topological relationship of the river channels within the watershed.

[0031] Step 1100: Based on the Digital Elevation Model (DEM), the target watershed is divided into sub-watersheds, and the sub-watershed topology is constructed. This includes the following sub-steps: Step 1110, Sub-basin data collection and preprocessing Obtain the digital elevation model (DEM) of the target watershed from the official source, standardize it into a standard format supported by Geographic Information System (GIS) (such as Shapefile, GeoJSON), and standardize the attribute field names to ensure data compatibility.

[0032] Step 1120, Sub-basin characteristic analysis, includes the following sub-steps: Step 1121: Use ArcSWAT software to export the divided sub-basins and water system data tables. Based on the area, use data analysis tools such as Python and R to calculate other geometric attributes of each sub-basin, such as centroid coordinates. Step 1122: Analyze the resolution and coverage of the sub-basin data to determine its accuracy in describing the spatial characteristics of the watershed; Step 1123: For each sub-basin, analyze whether its water system data is consistent with the actual catchment area and boundary morphology of the sub-basin; Step 1124: Through spatial overlay analysis, compare the spatial overlap of each sub-basin to identify potential data conflict areas.

[0033] Step 1130, Calculate the sub-basin topology construction: Based on the watershed system, the ArcSWAT River Network Tracing Algorithm (D8 algorithm) is used to construct the upstream and downstream topological relationships of sub-watersheds. A unique identifier is assigned to each sub-watershed, and a topology table is established to record the connections between sub-watersheds, for example, identifying upstream and downstream relationships by starting and ending sub-watershed numbers. For watersheds with bifurcations, a directed acyclic graph (DAG) structure from graph theory is used to describe complex flow paths. In the semi-distributed hydrological model, the local topological structure of the sub-watersheds is constructed, clarifying the transmission paths of water flow between sub-watersheds.

[0034] Step 1140: Mapping the computational unit boundaries based on topological relationships; establishing direct mapping relationships between sub-basins based on the constructed topological network. Between upstream and downstream sub-basins, the flow calculation results of the upstream sub-basin are directly used as the input boundary conditions for the downstream sub-basin, until the watershed outlet section is reached. Efficient data transmission between sub-basins is achieved through the connection information in the topology table.

[0035] Step 1150: Develop detailed boundary mapping rules, clarify the migration methods of different types of data between sub-basins, and compile the mapping rules into a configuration file for easy operation and maintenance later.

[0036] Step 1200: Using sub-basins as the basic calculation unit, precipitation is divided into extremely fast current, medium fast current, and base current based on evapotranspiration. This includes the following sub-steps: Step 1210: Based on the evapotranspiration of the daytime watershed and the underlying surface properties, classify the net rainfall generated by precipitation. Soil moisture content is determined by evaporation; potential evapotranspiration is constructed. .

[0037] Step 1220: Based on the actual evapotranspiration of the sub-basin constrained by tension water Among them, tension water first deducts evaporation. This information is used for subsequent allocation of soil moisture, infiltration, and fast and slow flows within the sub-basin.

[0038] Step 1230: Divide the sub-basin into infiltration and direct runoff generation; define wettability. Assign infiltration coefficient: ; Cut off the infiltration coefficient: Therefore, rainfall is divided into infiltration recharge and direct runoff: , This process is mainly regulated through soil type and land use data of each sub-basin.

[0039] Step 1240: Differentiate between the over-osmosis of the tension water and its leakage into free water; first, replace the tension water. Exceeding WM Partially became superosmotic And the tension water is cut back to its maximum capacity: Then, tension water seeps into free water. Therefore, the tension water at the end of the period is Therefore, the total input of free water is This replaces the original Xin'anjiang model SM-EX water storage capacity curve.

[0040] Step 1250: Divide the net rainfall into three parts: extremely fast current, medium fast current, and base current; Current total fast flow ratio Cut it into The fast stream is further split into smaller parts, with the ultra-fast stream having a weight of [value missing]. Cut it into Therefore, the weight of this fast stream is Therefore, it always flows quickly. The total base current is The total amount of ultrafast flow is The total amount of the second fast flow is The above process replaces the flow physics decomposition system of surface runoff, interflow, and groundwater runoff in the original Xin'anjiang model with the time-scaled decomposition system of extremely fast flow, sub-fast flow, and baseflow.

[0041] Step 1260: Linear calculation of ultrafast flow, subfast flow, and basestream library; Calculate the outflow from each of the ultrafast flow, subfast flow, and base flow pools: , , Update the status of the three libraries: , , Therefore, the runoff generated in the sub-basin is deep. Here, two fast-flow reservoirs replace the linear decay logic of free water in the soil runoff and groundwater runoff in the original Xin'anjiang model; the runoff unit is converted from mm / d to m. 3 / s, introducing a constant α=0.011574, therefore each sub-basin has ,in, Area j Units are km 2 .

[0042] Step 1270: Regulate the water levels of the sub-basin modules based on the major regulating reservoirs within the basin; Ruozi River Basin j Including reservoirs, And smoothed according to first-order linear library: For sub-basins without reservoirs: This processing is a visual extension of the original Xin'anjiang model.

[0043] Step 1280, Muskingum River Road; First, adjust the propagation time according to the sub-basin slope: , , ,make t =1d, then we have: , , , If any of the three coefficients are negative, then the following restrictions apply. Otherwise, normalize to: , , Then, calculate hourly: ,in Then, the outflow is accumulated along the downstream topology to the downstream sub-basin to calculate the outflow at the outlet section.

[0044] Step 1290: Perform scale correction on the outflow at the outlet section; Let the set of outlet sections of the target watershed be n. Then, for section i ∈ n, we have: , , where g is the outlet section of the watershed.

[0045] Step 1300: Based on the underlying surface properties, reservoir regulation, and slope correction of the Mustingen River route, obtain the outflow of each sub-basin, including the following sub-steps: Step 1310: Start the simulation job process; The system receives simulation calculation instructions through the user's click to run. When the instruction arrives, the system triggers the simulation job process. First, the system performs a validity check on the instruction (such as calculating the time range). If the instruction has missing key information, an error message is returned. If the check passes, the instruction is stored in the task queue and a unique process ID is generated for subsequent job tracking and management.

[0046] Step 1320, parse the process task; 1. Geospatial data parsing; By utilizing the zoning function of a Geographic Information System (GIS), geospatial data such as land use data and soil type data are matched with each sub-basin based on the topological relationship between sub-basins, thereby obtaining the corresponding land use data and soil type data for each sub-basin.

[0047] 2. Meteorological data instruction parsing; For meteorological data instructions, use data processing tools such as Python, R, or MatLab to extract daily precipitation and evapotranspiration data (usually in CSV format) for each sub-basin. Read the precipitation and evapotranspiration data using file reading functions (such as the `read_csv` function in the pandas library of Python), perform format conversion and quality checks, remove outliers (such as data with negative precipitation), and then distribute the processed data to the corresponding sub-basins.

[0048] 3. Analysis of the outlet section; The outlet section command reads the target watershed outlet section information table (usually in CSV format) extracted from the national hydrological station dataset to obtain basic information such as the name and latitude and longitude of the outlet section. Then, through the XY data loading function of the Geographic Information System (GIS), the target watershed outlet section is matched to the corresponding sub-watershed and marked in the sub-watershed topology table for subsequent calculation operations on the outlet section.

[0049] Step 1330, Information Reading and Calculation Preparation; 1. Obtaining regional information; Based on the analyzed watershed spatial extent, GIS spatial analysis tools (such as ArcGIS's SpatialAnalyst module) are used to calculate basic attributes of each sub-watershed, such as slope, using the target watershed digital elevation model (DEM). Topological relationship data of rivers and sub-watersheds within the watershed are retrieved from the watershed topology database, including upstream and downstream river connections and sub-watershed confluence paths, providing spatial data support for subsequent calculations.

[0050] 2. Meteorological data acquisition; Based on the meteorological data instruction parsing results, daily-scale precipitation and evapotranspiration data for the corresponding time range are retrieved from the meteorological database. If the data is daily-scale precipitation, it is cropped to the target watershed area; if the data is evapotranspiration, it is synchronized to the target watershed area. Data cleaning algorithms (such as outlier removal and missing value imputation) are used to preprocess the precipitation and evapotranspiration data to ensure data integrity and accuracy.

[0051] 3. Obtaining exit section information; Basic information about the outlet sections of the target watershed, such as section name and latitude and longitude, is obtained through a nationwide cross-sectional monitoring network. This information is then cross-referenced with sub-watershed matching information to ensure the accuracy of the outlet section's geographical location information.

[0052] Step 1340, planning and calculation logic; 1. The relationship between the outlet section and the sub-basin; Based on the geographical location of the outlet section, the sub-basins associated with the outlet section are determined using GIS spatial overlay analysis. Since the improved Xin'anjiang model is a semi-distributed hydrological model, the mapping relationship between the section and its sub-basin can be established by determining whether the section coordinates lie within a certain raster cell. A data association table is created to record the correspondence between the outlet section and the sub-basin, as well as the data transmission direction and format requirements (generally, geospatial data is in Shapefile or GeoJSON format, and time series data is in CSV format).

[0053] 2. Handling relationships and logical planning; For the target watershed outlet section to be calculated in this invention, the execution order and data transfer relationships are determined according to pre-configured processing rules. A directed acyclic graph (DAG) data structure is used to describe the computational logic, where nodes represent sub-watersheds and edges represent data flow directions and process dependencies. For example, the upstream node output obtained from the improved Xin'anjiang model is used as the input to the downstream node. The execution order of the computational tasks is generated using a breadth-first search (BFS) algorithm within the DAG, ensuring the rationality of the model computation.

[0054] Step 1350: Determine boundary information; 1. Determination of hydrological calculation boundary conditions; Upstream boundary conditions typically utilize the output results from the upstream sub-basin and are transmitted via a topology link (topology database). Downstream boundary conditions can combine the topology link (topology database) and actual river conditions, using boundary information fed back from the downstream sub-basin.

[0055] 2. Determination of reservoir disturbance boundary; The input boundary of the reservoir's disturbance to the target watershed includes the reservoir capacity (obtained from the national official reservoir database). The regulation capacity of different reservoirs is converted into a regulation coefficient based on the reservoir capacity and used as the input boundary for the downstream sub-watershed and the entire target watershed.

[0056] Step 1360: Start and execute the computing unit; Employing a multi-threaded or distributed computing architecture, the system invokes one or more Central Processing Units (CPUs) and simultaneously utilizes Python's CUDA acceleration module to invoke one or more Graphics Processing Units (GPUs) to initiate computations for each sub-basin. Before each sub-basin is started, a status query function checks the completeness of its input data; if the data is complete, the computation task begins. During computation, a logging function records the computation progress and intermediate results in real time. After computation is complete, the output results are stored in a specified data storage location (such as the cloud, database, or file system), and the process status of the target basin's exit section is updated for subsequent querying and retrieval at the exit section.

[0057] Step 1370: Monitor and update the status; The system utilizes Time technology to acquire real-time operational status information of the flow simulation device coupled with the improved Xin'anjiang model's neural network. For the model iteration calculation module, it monitors its computation progress, resource usage (CPU utilization, memory usage), and the accuracy of calculation results. For the outlet section, it monitors changes in key elements such as the simulated flow rate. When the status of the model iteration calculation module or the outlet section changes, the system automatically updates the boundary state and outlet section status databases and calibrates the sub-basin calculations in real time, ensuring the consistency and accuracy of the entire calculation and scheduling process.

[0058] Step 1400: Based on the total outflow of the target basin outlet section, obtain the flow process line of the target basin outlet section through river confluence calculation.

[0059] Initialization settings: Create a session task, set the simulation start and end times, and complete the time dimension configuration.

[0060] Sub-basin outflow loading: Retrieve daily outflow data of each sub-basin from the data center, and extrapolate the outflow of each sub-basin to the sub-basin where the outlet section is located through the Muskingum channel confluence algorithm according to the topological relationship between the sub-basin and the outlet section (topology database).

[0061] Flow scheme synthesis: Based on the characteristics of the watershed and the topological relationship between the sub-watershed and the outlet section, the flow scaling factor of the outlet section is called to adjust the magnitude of the outflow and then synthesize the flow process line of the outlet section according to the time span.

[0062] Step 1500 involves merging and tensing the flow, precipitation, and evapotranspiration data calculated by the improved Xin'anjiang model into a feature vector, which is then input into a Long Short-Term Memory (LSTM) neural network for training to obtain the final flow process curve at the target outlet section. This step includes the following sub-steps: Step 1510: Construct the input window for the Long Short-Term Memory (LSTM) neural network; let the input time series... ,in , which are the feature vector elements of each improved Xin'anjiang model outflow, precipitation, and evapotranspiration required to input the neural network.

[0063] Step 1520: Perform Z-Score normalization on each element of the input feature vector. ,in These are the elements of the normalized feature vector. x t Original sequence X middle t Elements of time μ x Time series X The mean, σ x Time series X The standard deviation.

[0064] Step 1530: Perform recursive calculations using a large number of hidden layers in a Long Short-Term Memory (LSTM) neural network; , , , , , ,in i t , f t , o t These are the input gate, the forget gate, and the output gate. x t , h t , c t These represent the input features, hidden states, and cell states, respectively. W i , W f , W o These are the weight coefficients for the input gate, forget gate, and output gate, respectively. U i , U f , U o These are the weight coefficients for the hidden states of each gate. b i , b f , b oHere, σ represents the bias of each gate, tan(·) represents the hyperbolic tangent function, and the flow rate is iteratively derived through the above calculations and the state update of the neural network.

[0065] Step 1540: Output the flow residual through recursive calculations using the massive hidden layers of a Long Short-Term Memory (LSTM) neural network. ; continuously through the loss function Reducing the gap between simulated and actual residuals completes the flow residual correction term after training the Long Short-Term Memory (LSTM) neural network. .

[0066] Step 1550: Superimpose the flow residual correction term with the outflow at the outlet section of the improved Xin'anjiang model to obtain the final simulated flow rate; superimpose the flow residual correction term, optimized by a complete Long Short-Term Memory (LSTM) neural network, onto the outflow at the section of the improved Xin'anjiang model, then we have: Let denoted as the final simulated flow of the improved Xin'anjiang model coupled with a neural network for flow simulation.

[0067] Step 1560: Save the target watershed outlet section flow obtained from the simulation in this embodiment of the invention into a database (cloud, database, file system, etc.) for subsequent system operation and maintenance.

[0068] Example 2 Please see Figure 3 Another embodiment of the present invention also proposes an improved flow simulation system for the Xin'anjiang model coupled with a neural network, comprising: The sub-basin data collection module 201 is used to collect sub-basin data of the target basin, wherein the sub-basin data of the target basin includes digital elevation model (DEM) data of the semi-distributed model; The mapping relationship establishment module 202 is used to construct sub-basin topological relationships, establish sub-basin boundaries and mapping relationships between sub-basins and target basin forecast sections based on the sub-basin topological relationships; between upstream and downstream sub-basins, the flow calculation results of the upstream sub-basin are used as the input boundary conditions of the downstream sub-basin; for the connection relationship between the sub-basin and the river, the outlet flow of the sub-basin is mapped to the corresponding basin outlet section; and mapping rules are formulated to clarify the conversion and transmission methods of different types of data between sub-basins. The simulation scheme construction module 203 is used to construct a simulation scheme by configuring the dataset required for each sub-basin, with the target watershed forecast section as the core. The basic boundary condition generation module 204 is used to generate basic boundary conditions for the target watershed using the constructed simulation scheme. The outlet section outflow data acquisition module 205 is used to analyze the basic boundary conditions of the target watershed, distribute various boundary data to various nodes according to the sub-watershed and water system topology for iterative operation, and output the outlet section outflow data of the improved Xin'anjiang model in the target watershed. The neural network simulation module 206 is used to process the outflow data, precipitation and evapotranspiration data of the target watershed outlet section into tensor quantization and then input them into the neural network for training to obtain the simulated flow sequence of the target watershed outlet section.

[0069] Taking the Jialing River Basin as an example, in this embodiment, the sub-basin data collection module 201 acquires geospatial data and meteorological data of the target basin; the geospatial data includes at least a digital elevation model (DEM), land use data, and soil type data; the meteorological data includes at least precipitation data and evapotranspiration data.

[0070] The sub-basin data collection module 201 uses the sub-basin division tool in ArcSWAT software based on the digital elevation model (DEM) to divide the target watershed into sub-basins. This tool uses the flow direction tracking algorithm (D8) to extract the watershed water system and establish the water network topology relationship between sub-basins according to the slope. At the same time, it obtains the basic information of each sub-basin, such as watershed area and coordinates, in order to extract the sub-basin and water system data table (usually in CSV format).

[0071] Net rainfall is classified into runoff based on the evapotranspiration of sub-basins. According to the differences in evapotranspiration and underlying soil moisture content, net rainfall is divided into three components—very fast runoff, medium fast runoff, and base runoff—in a certain proportion for runoff generation.

[0072] The outflow data acquisition module 205 at the outlet section adjusts the propagation time based on the slope of each sub-basin to facilitate the confluence of the Muskingum channel. First, it adjusts the propagation time coefficient K of the runoff in each sub-basin according to the slope. j The process involves making corrections by calculating the three major coefficients based on the Muskingum confluence equation. If any coefficient has a negative value, the default configuration (0, 1, 0) is used; otherwise, the confluence coefficients are normalized. Then, the Muskingum channel confluence equation is used to perform hourly topological channel confluence calculations on the runoff generation, inflow, and outflow of each sub-basin. The topological accumulation is performed from upstream to downstream along the sub-basin topological relationship table until the outflow is extrapolated to the outlet section, thus obtaining the outflow of the improved Xin'anjiang model outlet section.

[0073] The neural network simulation module 206 is used to improve the outflow at the outlet section of the Xin'anjiang model. Tensorization training of precipitation and evapotranspiration data is used to obtain the final outflow at the outlet section of the target basin. The neural network simulation module 206 combines the residual term of the improved Xin'anjiang model's outflow at the outlet section with precipitation and evapotranspiration to form a feature vector. This feature vector is then tensorized and input into a Long Short-Term Memory (LSTM) neural network for training. The loss function is used to continuously correct the flow residual term, gradually approximating the measured flow value. The optimal solution that minimizes the loss function is recursively calculated and used as the final simulated flow at the outlet section of the Jialing River basin for the improved Xin'anjiang model coupled with a neural network.

[0074] This application's embodiments improve the coupling design of the Xin'anjiang model and neural network, combining sub-basin topology construction, data mapping, and distributed computing logic. Addressing the technical pain points of traditional hydrological models, such as low computational efficiency, insufficient simulation accuracy, and poor adaptation to spatial heterogeneity, this invention achieves efficient and accurate simulation of basin flow. By incorporating GPU acceleration technology, this invention significantly improves the computational efficiency of neural network training and the overall simulation process, solving the problem of slow computation speed and inability to meet real-time forecasting requirements in large-scale basin simulations using traditional models. Furthermore, the entire technical solution revolves around "accurate simulation of outlet section flow," aligning with the actual operational needs of basin hydrological forecasting and water resource management, and possesses strong practicality.

[0075] Example 3 Please see Figure 4 The present invention proposes an electronic device for implementing the improved Xin'anjiang model coupled neural network for flow simulation. The electronic device includes a memory 310 and a processor 300. The memory 310 is used to store computer programs, and the processor 300 runs the computer programs to enable the electronic device to perform the above-described method.

[0076] Furthermore, combined Figure 4 The electronic device shown also includes a bus 320 and a communication interface 330, with the processor 300, communication interface 330 and memory 310 connected via the bus 320.

[0077] The memory 310 may include high-speed random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device. Communication between this system network element and at least one other network element is achieved through at least one communication interface 330 (which can be wired or wireless), such as the Internet, wide area network, local area network, metropolitan area network, etc. The bus 320 may be an ISA bus, PCI bus, or EISA bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, in... Figure 4 The symbol is represented by a single double-headed arrow, but this does not mean that there is only one bus or one type of bus.

[0078] The processor 300 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry of the processor 300's hardware or by instructions in software form. The processor 300 may be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it may also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components, capable of implementing or executing the methods, steps, and logic block diagrams disclosed in the embodiments of this invention. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this invention can be directly manifested as execution by a hardware decoding processor, or execution by a combination of hardware and software modules in the decoding processor. The software modules may reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The storage medium is located in the memory 310. The processor 300 reads the information in the memory 310 and, in conjunction with its hardware, completes the steps of the method described in the foregoing embodiment.

[0079] The present invention provides a readable storage medium storing computer program instructions, which are read and executed by a processor to perform the improved Xin'anjiang model coupled neural network flow simulation method.

[0080] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system and apparatus described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0081] Furthermore, in the description of the embodiments of the present invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "joining" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the connection within two components. Those skilled in the art will understand the specific meaning of the above terms in the present invention according to the specific circumstances.

[0082] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a server, or a network device such as cloud computing, blockchain, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0083] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0084] Finally, it should be noted that the above embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical details; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An improved method for simulating flow in the Xin'anjiang River model coupled with a neural network, characterized in that, include: Collect sub-basin data of the target watershed, including digital elevation model (DEM) data from a semi-distributed model; Construct sub-basin topological relationships, and establish sub-basin boundaries and mapping relationships between sub-basins and target basin forecast sections based on these relationships. Between upstream and downstream sub-basins, the flow calculation results of the upstream sub-basin are used as input boundary conditions for the downstream sub-basin. For the connection between a sub-basin and the river channel, the sub-basin outlet flow is mapped to the corresponding basin outlet section. Mapping rules are formulated to clarify the conversion and transmission methods of different data types between sub-basins. Using the target watershed forecast section as the core, configure the dataset required for each sub-watershed and construct a simulation scheme; Using the constructed simulation scheme, the basic boundary conditions of the target watershed are generated; The basic boundary conditions of the target watershed are analyzed, and various boundary data are distributed to various nodes according to the sub-watershed and water system topology for iterative operation, outputting the outflow data of the improved Xin'anjiang model at the outlet section of the target watershed; The outflow data, precipitation and evapotranspiration data of the target watershed outlet section are tensor-quantized and then input into the neural network for training to obtain the simulated flow sequence of the target watershed outlet section.

2. The improved Xin'anjiang model coupled with a neural network for flow simulation method according to claim 1, characterized in that, In the step of constructing a simulation scheme by configuring the dataset required for each sub-basin based on the target watershed forecast section, the dataset required for each sub-basin is configured according to flood control needs or water resource conditions. The construction of the simulation scheme includes: Create a calculation scheme and name it; Select the start and end sub-basins, and match the computational region of the simulation scheme according to the topological relationship; Within the calculation area, the exit sections are listed, and the calculation logic for the exit sections is determined; Match various types of data mapped to the corresponding exit section; After determining the various calculation data for the outlet section, set the initial parameters of the model; Identify the reservoirs that have a significant impact on the watershed and set their impact coefficients; Complete the initial calculation configuration for each outlet section according to the topological relationship and save the simulation scheme.

3. The improved Xin'anjiang model coupled with a neural network for flow simulation method according to claim 2, characterized in that, The steps of selecting the start and end sub-basins and matching the computational region of the simulation scheme according to the topological relationship include: Select the target watershed simulation section; Analyze topological relationships; Based on the analysis results of the target watershed simulation cross section and topological relationship, the computational region corresponding to the simulation scheme is determined; The steps for analyzing topological relationships include: Retrieve pre-constructed watershed channel topology data and store the upstream and downstream connections of the water system in the form of a directed graph. Use the breadth-first search algorithm in graph theory to search along the topology path from the starting sub-watershed until the end outlet section is reached, and include all water system segments and sub-watersheds involved in the path into the calculation area.

4. The improved Xin'anjiang model coupled with a neural network for flow simulation method according to claim 1, characterized in that, The steps for establishing the sub-basin boundary and the mapping relationship between the sub-basin and the target basin forecast sections based on the sub-basin topological relationship include: preliminary selection of forecast sections based on forecast requirements, wherein the forecast sections include flood control forecast sections, water resource management sections, and ecological flow guarantee sections; diagnosis of the fit of the outlet sections; optimization and demonstration of the outlet sections based on model simulation; determination of the basin outlet sections; and basin outlet section information including at least the section location and basic parameters.

5. The improved Xin'anjiang model coupled with a neural network for flow simulation method according to claim 1, characterized in that, The step of collecting sub-basin data of the target watershed obtains the computational data of the sub-basins from different database sources and uses data cleaning methods to remove outliers from the data. The step of collecting sub-basin data of the target watershed also includes sub-basin characteristic analysis, as follows: Calculate the geometric properties of each sub-basin; Analyze the resolution and coverage of sub-basins to determine the accuracy of the sub-basins in describing the spatial characteristics of the watershed; The catchment area and boundary morphology of sub-basins are determined by using a watershed drainage system extraction algorithm. By using spatial overlay analysis, we can compare the spatial overlap of sub-basins under different calculation and analysis tools and identify potential data conflict areas.

6. The improved Xin'anjiang model coupled with a neural network for flow simulation method according to claim 1, characterized in that, The steps for constructing sub-basin topological relationships include constructing a river system topology, building upstream and downstream river topological relationships based on the river system, assigning a unique identifier to each sub-basin, and establishing a topology table to record the confluence relationships between sub-basins; simultaneously determining the boundaries of sub-basins and establishing a spatial logical topological network between sub-basins; in the semi-distributed model, constructing the local topological structure of sub-basins based on the water flow direction of adjacent sub-basins, and clarifying the transmission path of water flow between sub-basins.

7. The improved Xin'anjiang model coupled with a neural network for flow simulation method according to claim 1, characterized in that, Establishing the mapping relationship between the forecast sections of the sub-basin and the target basin includes spatial overlay analysis, distance overlay analysis, and topological relationship analysis.

8. The improved Xin'anjiang model coupled with a neural network for flow simulation method according to claim 1, characterized in that, The steps of analyzing the basic boundary conditions of the target watershed, allocating various boundary data to various nodes according to sub-watersheds and river system topology for iterative operation, and outputting the outflow data of the improved Xin'anjiang model at the outlet section of the target watershed include: Start the simulated work process; The task of parsing nodes is executed to decompose the basic boundary conditions of the target watershed and, in combination with the constructed sub-watershed topology, clarify the allocation rules for various types of boundary data. Data reading and computation preparation; The calculation logic is planned, and the iterative calculation method is determined by combining the sub-basin topology and outlet section requirements; Determine the boundary conditions and, in conjunction with the actual hydrological characteristics of the target watershed, determine the iterative convergence criteria; Perform iterative calculations according to the planned computational logic and output the traffic data corresponding to each node; Monitor and update the status, and after the iterative run is completed, output the outflow data of the target watershed outlet section.

9. The improved Xin'anjiang model coupled with a neural network for flow simulation method according to claim 1, characterized in that, The step of tensor quantization processing of the outflow data, precipitation, and evapotranspiration data of the target watershed outlet section, followed by inputting them into a neural network for training to obtain the simulated flow sequence of the target watershed outlet section includes: Outflow data of the target basin outlet section were derived from the improved Xin'anjiang model; The collected precipitation and evapotranspiration data were weighted by watershed area. The outflow data, weighted precipitation data, and evapotranspiration data at the outlet section are merged to construct an integrated feature vector of outflow, precipitation, and evapotranspiration. Tensor quantization is performed on the integrated feature vectors of outflow, precipitation, and evapotranspiration to unify the data dimensions; Configure and initialize the core parameters of the Long Short-Term Memory (LSTM) neural network; The tensor-quantized feature vectors are input into the initialized long short-term memory neural network, and through multiple rounds of iterative calculations, the simulated flow sequence of the target watershed outlet section is obtained.

10. A flow simulation system based on an improved Xin'anjiang model coupled with a neural network, characterized in that, include: The sub-basin data collection module is used to collect sub-basin data of the target basin, including digital elevation model (DEM) data of the semi-distributed model. The mapping relationship establishment module is used to construct sub-basin topological relationships, establish sub-basin boundaries and mapping relationships between sub-basins and target basin forecast sections based on the sub-basin topological relationships; between upstream and downstream sub-basins, the flow calculation results of the upstream sub-basin are used as the input boundary conditions of the downstream sub-basin; for the connection relationship between the sub-basin and the river, the outlet flow of the sub-basin is mapped to the corresponding basin outlet section; and mapping rules are formulated to clarify the conversion and transmission methods of different types of data between sub-basins. The simulation scheme construction module is used to construct simulation schemes by configuring the datasets required for each sub-basin, with the target watershed forecast section as the core. The basic boundary condition generation module is used to generate basic boundary conditions for the target watershed using the constructed simulation scheme. The outlet section outflow data acquisition module is used to analyze the basic boundary conditions of the target watershed, distribute various boundary data to various nodes according to sub-watersheds and river system topology for iterative operation, and output the outlet section outflow data of the improved Xin'anjiang model in the target watershed. The neural network simulation module is used to process the outflow data, precipitation and evapotranspiration data of the target watershed outlet section into tensor quantization and then input them into the neural network for training to obtain the simulated flow sequence of the target watershed outlet section.

Images