Drainage basin ecological hydrological process simulation and prediction system based on mechanism hydrological model and AI coupling

Abstract

The invention discloses a drainage basin ecological hydrological process simulation and prediction system based on a mechanism hydrological model and AI coupling, and relates to the technical field of drainage basin ecological hydrological simulation and prediction. Multi-source data integration and preprocessing are carried out through a data input layer, and a standardized data set is output and provided for a model coupling layer; the model coupling layer directly receives the standardized data set of the data input layer and outputs an optimized simulation result and a predicted value to the simulation and prediction layer; the simulation prediction layer receives a simulation result and a prediction value of the model coupling layer, performs multi-dimensional ecological hydrological element collaborative simulation and short-term-medium and long-term scene adaptive prediction, and outputs a simulation prediction result to the result output layer; and the result output layer receives the simulation prediction result of the simulation prediction layer and provides visual interaction and decision suggestion generation functions. The invention provides an innovative scheme of deep fusion of a mechanism model and an AI technology, and comprehensive requirements of physical rationality, data fitting precision, multi-element coverage and efficient response are covered.

Description

Technical Field

[0001] This invention relates to the field of watershed eco-hydrological simulation and prediction technology, and more specifically to a watershed eco-hydrological process simulation and prediction system based on the coupling of mechanistic hydrological models and AI. Background Technology

[0002] Watershed eco-hydrological processes are complex systems of climate-land surface interaction, and their simulation and prediction are the core technological foundation for water resource management and ecological protection. Currently, mainstream technical solutions in the industry are mainly divided into three categories, but all of them have insurmountable shortcomings and cannot meet the needs of complex watersheds and diverse scenarios: (I) Pure Mechanistic Hydrological Model Represented by SWAT, HEC-HMS, and VIC, their core principle is to construct mathematical equations based on physical laws to describe the intrinsic patterns of hydrological processes. Limitations: Strong parameter dependence: The model requires a large number of physical parameters as input. Among them, key parameters such as the hydraulic properties of deep soil are difficult to obtain through field observation and need to be estimated based on experience or assigned regional averages, which leads to significant simulation errors in complex watersheds and large runoff simulation errors. Low computational efficiency: Mechanism models require solving physical equations grid by grid and time period by time. In large watershed or medium- to long-term prediction scenarios, the computation time can reach several hours or even days, which cannot meet the needs of real-time decision-making. Poor adaptability: The response to abrupt changes in the underlying surface of the watershed or extreme hydrological events is delayed. Because the physical equations cannot cover the nonlinear changes in "unconventional" scenarios, the prediction results deviate from reality.

[0003] (ii) Pure AI Model

[0004] Represented by LSTM, Random Forest, and BP Neural Network, the core principle is to learn the statistical regularities of hydrological processes through historical observation data, thereby achieving data-driven simulation and prediction. Limitations: The model exhibits significant "black box" characteristics: it focuses only on the mapping relationship between input and output, cannot explain the physical meaning of the predicted results, and is difficult to support causal analysis in scientific decision-making. Weak generalization ability: Low reliability in predicting scenarios that have not been experienced, and samples outside the historical data distribution are prone to "prediction drift"; High data dependence: It requires continuous and complete long-sequence observation data, which cannot be effectively deployed in watersheds with scarce data, and is sensitive to data noise. A small number of outliers can seriously affect the accuracy of the model.

[0005] (III) Simplified Coupled Model

[0006] Currently, coupling within the industry is mostly a one-way call pattern, such as: using an AI model to predict a key parameter of a mechanistic model, then substituting it into the mechanistic model for calculation; or using an AI model to correct the output of a mechanistic model. Drawbacks: Shallow coupling level: AI and mechanism model are independent of each other, lack dynamic feedback mechanism, and cannot work together to deal with the complex nonlinear correlation of hydrological processes. There are still problems of "lack of physical logic" or "data fitting deviation". Single function: Most of them only focus on a single hydrological element and do not integrate multi-dimensional eco-hydrological processes such as precipitation, evapotranspiration, and pollutant migration, and cannot provide comprehensive data support for the coordinated management of watershed "water resources-ecology-environment"; Poor scenario adaptability: The inability to dynamically adjust the model strategy according to the hydrological scenario, and the problems of insufficient short-term prediction accuracy or poor medium- and long-term prediction stability have not yet been solved.

[0007] In summary, existing technologies cannot simultaneously meet the comprehensive requirements of "physical rationality, data fitting accuracy, multi-factor coverage, and efficient response." There is an urgent need for an innovative solution that deeply integrates mechanistic models and AI technologies to break through the current technological bottlenecks. Summary of the Invention

[0008] In view of this, the present invention provides a watershed eco-hydrological process simulation and prediction system based on the coupling of mechanistic hydrological model and AI, in order to solve the problems existing in the background technology.

[0009] To achieve the above objectives, the present invention adopts the following technical solution: A watershed eco-hydrological process simulation and prediction system based on mechanistic hydrological modeling coupled with AI includes: a data input layer, a model coupling layer, a simulation prediction layer, and a result output layer. The data input layer integrates and preprocesses multi-source data, outputs a standardized dataset, and provides it to the model coupling layer. The model coupling layer directly receives the standardized dataset from the data input layer and outputs optimized simulation results and predicted values ​​to the simulation prediction layer. The simulation prediction layer receives the simulation results and predicted values ​​from the model coupling layer, performs multi-dimensional eco-hydrological element collaborative simulation and short-term-medium-term scenario adaptive prediction, and outputs simulation prediction results to the result output layer. The result output layer receives the simulation prediction results from the simulation prediction layer and provides visualization interaction and decision suggestion generation functions.

[0010] Preferably, the data integrated by the data input layer includes basic geographic data and multi-source observation data.

[0011] Preferably, the data input layer further includes data cleaning based on the Python Pandas library, geographic data format conversion based on the GDAL library, and meteorological data from sparse stations in multi-source observation data completed using Kriging interpolation, outputting a standardized dataset.

[0012] Preferably, the model coupling layer includes a mechanism constraint module, an AI parameter inversion module, and a dynamic feedback module. The mechanism constraint module uses a SWAT model for its mechanism hydrological model, inputting standardized data, setting initial difficult-to-observe parameters to default values, and outputting multi-dimensional eco-hydrological intermediate variables. The initial difficult-to-observe parameters include soil saturated hydraulic conductivity, runoff curve number, and Manning coefficient. The AI ​​model of the AI ​​parameter inversion module adopts an LSTM model combined with an attention mechanism, inputting the standardized dataset from the data input layer and the intermediate variables of the mechanism hydrological model, and outputting inverted values ​​of difficult-to-observe parameters and predicted values ​​of short-term hydrological elements. The dynamic feedback module directly inputs the multi-dimensional eco-hydrological intermediate variables of the mechanism model and the inverted values ​​of difficult-to-observe parameters and predicted values ​​of short-term hydrological elements of the AI ​​model, calculates the coupling bias, and when the coupling bias is >5%, re-collects multi-source data to update the parameters of the mechanism constraint module and the AI ​​parameter inversion module.

[0013] Preferably, the simulation prediction layer includes: The multi-factor collaborative simulation module directly inputs the optimized coupling results of the model coupling layer and the standardized dataset of the data input layer, and directly outputs multi-dimensional eco-hydrological simulation data, including runoff spatial distribution data; actual evapotranspiration data; pollutant migration concentration data; and daily water consumption data for different vegetation types. The scene adaptation module, based on real-time meteorological data, outputs prediction results according to the following logic: When the 24-hour precipitation is ≥50mm, the AI ​​model of the AI ​​parameter inversion module of the model coupling layer is used to directly output the short-term prediction results with an hourly calculation step and a 30-day training window. When the 24-hour precipitation is less than 50 mm, the SWAT model of the AI ​​parameter inversion module of the model coupling layer is applied, and the daily calculation step and 1-year training window are used to directly output the medium- and long-term prediction results.

[0014] The short-term forecast specifically refers to the forecast for the next 1-3 days (hourly continuous output), while the medium- and long-term forecast specifically refers to the forecast for the next 1 month to 1 year (daily output).

[0015] Preferably, the result output layer includes a visualization interaction module, which directly inputs the short-term or medium-to-long-term prediction results from the simulation prediction layer, multi-dimensional eco-hydrological simulation data, and measured data from the data input layer, and directly outputs an interactive interface that supports access from PC or mobile devices, with functions for spatiotemporal distribution display, data query, and result comparison; and a decision suggestion module, which directly inputs the short-term or medium-to-long-term prediction results from the simulation prediction layer and directly outputs a decision report, including risk warnings and optimization schemes.

[0016] As can be seen from the above technical solution, compared with the prior art, the present invention discloses a watershed eco-hydrological process simulation and prediction system based on the coupling of mechanistic hydrological model and AI, with the following beneficial effects: Significantly Improved Accuracy: In complex and diverse watershed environments, this invention significantly improves the accuracy of hydrological simulations. Whether it's routine runoff forecasting, peak flow prediction for extreme rainfall events, or simulation of water pollutant concentrations, the accuracy of the simulations is significantly enhanced. Significantly enhanced adaptability: When faced with situations where the underlying surface conditions of the watershed change drastically, this invention demonstrates excellent predictive robustness, can quickly adapt and provide reliable prediction results, outperforming single artificial intelligence models. Significantly improved efficiency: This invention greatly reduces the time required for model operation when handling short-term and medium-to-long-term prediction tasks, meeting the needs of real-time decision-making; Reduced management costs: This invention significantly reduces reliance on on-site parameter monitoring, effectively reducing the investment in field sampling; at the same time, its accurate flood disaster prediction capabilities enable significant cost savings in advance response and emergency handling. Supporting scientific decision-making: It provides multi-dimensional data support for watershed water resource allocation and ecological restoration. For example, in agricultural watersheds, it can accurately predict the water consumption demand of vegetation, thereby optimizing irrigation strategies, achieving not only significant water-saving effects, but also effectively reducing water pollution caused by fertilizer runoff. Enhancing disaster prevention capabilities: Under extreme rainstorm events, this invention can predict runoff peaks in advance and accurately, buying valuable time for the scientific management of reservoirs and the timely evacuation of people; Industry promotion value: The system supports model replacement (such as replacing SWAT with HEC-HMS) and data source expansion (such as remote sensing data replacing measured data), and can be adapted to watersheds of different sizes and climate zones. It has broad prospects for promotion and application and is expected to become the industry standard solution for watershed eco-hydrological simulation. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the structure provided by the present invention. Detailed Implementation

[0019] 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, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] This invention discloses a watershed eco-hydrological process simulation and prediction system based on the coupling of mechanistic hydrological models and AI, such as... Figure 1 As shown, it includes: a data input layer, a model coupling layer, a simulation prediction layer, and a result output layer. The data input layer integrates and preprocesses multi-source data, outputs a standardized dataset, and provides it to the model coupling layer. The model coupling layer directly receives the standardized dataset from the data input layer and outputs optimized simulation results and predicted values ​​to the simulation prediction layer. The simulation prediction layer receives the simulation results and predicted values ​​from the model coupling layer, performs multi-dimensional eco-hydrological element collaborative simulation and short-term-medium-term scenario adaptive prediction, and outputs simulation prediction results to the result output layer. The result output layer receives the simulation prediction results from the simulation prediction layer and provides visualization interaction and decision suggestion generation functions.

[0021] Preferably, the data integrated by the data input layer includes basic geographic data and multi-source observation data.

[0022] Basic geographic data: Watershed digital elevation model (DEM, 30m resolution), land use type map (30m resolution), soil type distribution map (500m-1km resolution), river network vector map; Multi-source observation data: meteorological station (precipitation, temperature, wind speed), hydrological station (runoff), water quality station (TN, TP), soil moisture station (soil moisture content, soil nutrients), vegetation observation station (NDVI vegetation coverage) data, supporting real-time access via wireless sensor network (LoRa, NB-IoT) and offline import from third-party databases (such as the National Meteorological Data Network); Preferably, the data input layer further includes data cleaning based on the Python Pandas library, geographic data format conversion based on the GDAL library, and meteorological data of sparse stations in multi-source observation data completed based on Kriging interpolation, outputting a standardized dataset (format: CSV / GeoTIFF, time step: hour / day).

[0023] Preferably, the model coupling layer includes a mechanism constraint module, an AI parameter inversion module, and a dynamic feedback module. The mechanism constraint module uses a SWAT model for its mechanism hydrological model, inputting standardized data, setting initial difficult-to-observe parameters to default values, and outputting multi-dimensional eco-hydrological intermediate variables. The initial difficult-to-observe parameters include soil saturated hydraulic conductivity, runoff curve number, and Manning coefficient. The AI ​​model of the AI ​​parameter inversion module adopts an LSTM model combined with an attention mechanism, inputting the standardized dataset from the data input layer and the intermediate variables of the mechanism hydrological model (precipitation, surface runoff, soil moisture content, total watershed outflow, nitrogen content, phosphorus content, organic carbon content, sediment transport, etc.), and outputting inverted values ​​of difficult-to-observe parameters and predicted values ​​of short-term hydrological elements. The dynamic feedback module directly inputs the multi-dimensional eco-hydrological intermediate variables of the mechanism model and the inverted values ​​of difficult-to-observe parameters and predicted values ​​of short-term hydrological elements of the AI ​​model, calculates the coupling bias, and when the coupling bias is >5%, re-collects multi-source data to update the parameters of the mechanism constraint module and the AI ​​parameter inversion module.

[0024] The mechanism constraint module converts the core SWAT physics equations into the loss function regularization term of the AI ​​model, ensuring that the AI ​​output conforms to physical laws and avoiding "black box drift". The AI ​​model learns the mapping relationship between "parameters-simulation results-measured results" to invert parameters such as Ksat and CN.

[0025] Preferably, the simulation prediction layer includes: The multi-factor collaborative simulation module directly inputs the optimized coupling results of the model coupling layer and the standardized dataset of the data input layer, and directly outputs multi-dimensional eco-hydrological simulation data, including: Runoff simulation: Based on a coupled model, output the flow rate and spatial distribution of runoff depth at the watershed outlet section; Evapotranspiration calculation: Combine the vegetation cover retrieved by AI, correct the Penman-Monteith formula results of the SWAT model, and obtain the actual evapotranspiration. Pollutant migration simulation: Input the initial concentration of pollutants (such as the intensity of fertilizer application in farmland), and simulate the concentration changes of surface runoff / groundwater through the SWAT pollutant migration module. The AI ​​model corrects the diffusion coefficient. Vegetation water consumption analysis: Based on NDVI data and AI-derived vegetation interception coefficients, the daily water consumption of different vegetation types (forest land, cultivated land) is calculated. The scene adaptation module, based on real-time meteorological data, outputs prediction results according to the following logic: When the 24-hour precipitation is ≥50mm, the AI ​​model of the AI ​​parameter inversion module of the model coupling layer is used to directly output the short-term prediction results with an hourly calculation step and a 30-day training window. When the 24-hour precipitation is less than 50 mm, the SWAT model of the AI ​​parameter inversion module of the model coupling layer is applied, and the daily calculation step and 1-year training window are used to directly output the medium- and long-term prediction results.

[0026] Preferably, the results output layer includes a visualization and interactive module, developed based on WebGIS (OpenLayers) and ECharts, supporting access from PCs and mobile devices. Functions include: ① Spatiotemporal distribution display (e.g., runoff depth map, pollutant concentration time series map); ② Data query (filtering element values ​​by time / space); ③ Comparative analysis (intuitive comparison of the system's results with measured values ​​and pure model results). The decision-making recommendation module automatically generates reports (format: PDF / Excel) based on simulation and prediction results, including: ① risk warning; ② solution optimization.

[0027] Technological Innovation Points

[0028] The "two-way dynamic feedback" deep coupling architecture breaks through the existing simple coupling one-way call mode. It achieves deep integration of AI and mechanism model through mechanism constraint injection (AI conforms to physical logic), AI parameter inversion (mechanism model parameter optimization), and dynamic deviation correction (the two work together), rather than simple splicing. Attention-based parameter inversion algorithm: For parameters that are difficult to observe in the mechanism model, the attention mechanism is used to enhance the influence of key inputs (such as precipitation during rainstorms and vegetation cover) on parameter inversion, thereby reducing the dependence on field observations; Multi-dimensional eco-hydrological element integration module: For the first time, precipitation, runoff, evapotranspiration, pollutant migration, and vegetation water consumption are incorporated into the same coupled system to achieve multi-element collaborative simulation and solve the limitations of existing single-element simulation systems; Adaptive prediction switching logic: Automatically adjust the AI ​​training window and the calculation step size of the mechanism model according to the hydrological scenario to balance short-term real-time performance and medium- to long-term stability; Physically interpretable AI output: By embedding physical equations into the AI ​​loss function through the mechanism constraint module, the AI ​​prediction results can be explained by physical laws, breaking the "black box" defect of pure AI.

[0029] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. The methods disclosed in the embodiments are described simply because they correspond to the methods disclosed in the embodiments; relevant parts can be found in the method section.

[0030] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A watershed eco-hydrological process simulation and prediction system based on the coupling of mechanistic hydrological models and AI, characterized in that, include: The system comprises a data input layer, a model coupling layer, a simulation prediction layer, and a result output layer. The data input layer integrates and preprocesses multi-source data, outputs a standardized dataset, and provides it to the model coupling layer. The model coupling layer directly receives the standardized dataset from the data input layer and outputs optimized simulation results and predicted values ​​to the simulation prediction layer. The simulation prediction layer receives the simulation results and predicted values ​​from the model coupling layer, performs multi-dimensional eco-hydrological element collaborative simulation and short- to medium-term scenario adaptive prediction, and outputs simulation prediction results to the result output layer. The result output layer receives the simulation prediction results from the simulation prediction layer and provides visualization interaction and decision suggestion generation functions.

2. The watershed eco-hydrological process simulation and prediction system based on the coupling of mechanistic hydrological model and AI according to claim 1, characterized in that, The data input layer integrates basic geographic data and multi-source observation data.

3. The watershed eco-hydrological process simulation and prediction system based on the coupling of mechanistic hydrological model and AI according to claim 2, characterized in that, The data input layer also includes data cleaning based on the Python Pandas library, geographic data format conversion based on the GDAL library, and meteorological data from sparse stations in multi-source observation data completed using Kriging interpolation, outputting a standardized dataset.

4. The watershed eco-hydrological process simulation and prediction system based on the coupling of mechanistic hydrological model and AI according to claim 1, characterized in that, The model coupling layer includes a mechanism constraint module, an AI parameter inversion module, and a dynamic feedback module. The mechanism constraint module uses a SWAT model for its mechanistic hydrological model, inputting standardized data and setting initial difficult-to-observe parameters to default values, and outputting multi-dimensional eco-hydrological intermediate variables. The initial difficult-to-observe parameters include soil saturated hydraulic conductivity, runoff curve number, and Manning coefficient. The AI ​​model of the AI ​​parameter inversion module adopts an LSTM model combined with an attention mechanism, inputting the standardized dataset from the data input layer and the intermediate variables of the mechanism hydrological model, and outputting inverted values ​​of difficult-to-observe parameters and predicted values ​​of short-term hydrological elements. The dynamic feedback module directly inputs the multi-dimensional eco-hydrological intermediate variables of the mechanism model and the inverted values ​​of difficult-to-observe parameters and predicted values ​​of short-term hydrological elements from the AI ​​model, calculates the coupling bias, and when the coupling bias is >5%, multi-source data is collected again to update the parameters of the mechanism constraint module and the AI ​​parameter inversion module.

5. The watershed eco-hydrological process simulation and prediction system based on the coupling of mechanistic hydrological model and AI according to claim 4, characterized in that, The simulation prediction layer includes: The multi-factor collaborative simulation module directly inputs the optimized coupling results of the model coupling layer and the standardized dataset of the data input layer, and directly outputs multi-dimensional eco-hydrological simulation data, including runoff spatial distribution data; actual evapotranspiration data; pollutant migration concentration data; and daily water consumption data for different vegetation types. The scene adaptation module, based on real-time meteorological data, outputs prediction results according to the following logic: When the 24-hour precipitation is ≥50mm, the AI ​​model of the AI ​​parameter inversion module of the model coupling layer is used to directly output the short-term prediction results with an hourly calculation step and a 30-day training window. When the 24-hour precipitation is less than 50 mm, the SWAT model of the AI ​​parameter inversion module of the model coupling layer is applied, and the daily calculation step and 1-year training window are used to directly output the medium- and long-term prediction results.

6. The watershed eco-hydrological process simulation and prediction system based on the coupling of mechanistic hydrological model and AI according to claim 1, characterized in that, The results output layer includes a visualization and interaction module, which directly inputs short-term or medium-to-long-term prediction results from the simulation prediction layer, multi-dimensional eco-hydrological simulation data, and measured data from the data input layer. It directly outputs an interactive interface that supports access from PC or mobile devices, and features spatiotemporal distribution display, data query, and result comparison functions. The decision suggestion module directly inputs short-term or medium-to-long-term prediction results from the simulation prediction layer and directly outputs a decision report, which includes risk warnings and optimization schemes.

Images