A method, system, and medium for watershed flood reduction

Abstract

The application relates to a basin flood reduction method, system and medium, the method comprising the following steps: acquiring a historical hydrological sequence and digital terrain data of a calculation area, extracting a flow deviation mode caused by regulation and storage to reduce natural flow, calculating a dynamic water level increment constrained by a physical law and superimposing the dynamic water level increment on a measured datum to obtain a reduced natural water level, using the reduced natural element as a time-varying boundary condition to drive a two-dimensional water dynamic model to perform calculation, and acquiring a daily flood evolution process of the whole basin. Through the method of fusing a water level deduction mechanism constrained by a physical boundary and two-dimensional water dynamic coupling, the application overcomes the problem that a traditional pure data-driven model is flattened in extreme flood extrapolation, causes water level inversion, and solves the defect that single-point prediction lacks a spatial physical mechanism, and significantly improves the physical logical consistency of extreme flood process reduction and the space-time simulation accuracy of the whole basin.

Description

Technical Field

[0001] This application relates to the interdisciplinary fields of computational hydrology, artificial intelligence and disaster prevention and mitigation, and in particular to a method, system and medium for watershed flood reconstruction. Background Technology

[0002] Analyzing the evolution patterns of river basin floods and planning flood control operations are major challenges facing the national water resources and water security strategy. In river basins severely affected by large-scale water conservancy projects (such as reservoirs and dams), accurately reconstructing the natural flood processes (flow and water level) obscured by regulation and storage is crucial for verifying flood control benchmarks and designing water-related engineering projects. Traditional reconstruction methods typically employ pure hydrodynamic physical models (such as algorithms based on one-dimensional / two-dimensional Saint-Venant equations). However, such methods have extremely high requirements for boundary conditions such as river channel topography and roughness during periods without historical data, and parameter calibration suffers from strong "heterogeneous but identical spectrum" uncertainties; moreover, long-sequence physical inversion calculations are extremely time-consuming, making it difficult to achieve rapid and flexible reconstruction of historical extreme scenarios.

[0003] Another option is to introduce purely data-driven machine learning techniques (such as random forests and support vector machines) to establish a nonlinear mapping relationship between historical water levels and flow rates. While this approach significantly reduces computational costs, existing machine learning methods are often limited by the mathematical flaws of their underlying algorithms when dealing with extreme floods (i.e., extreme samples that exceed the distribution of the historical training set). For example, traditional tree-based algorithms essentially perform zero-order extrapolation at leaf nodes, which cannot exceed the envelope of historical extreme values, leading to a severe numerical truncation effect (i.e., "peak clipping") when predicting extreme flood peaks. More critically, because they are decoupled from the underlying constraints of physical mechanisms, during periods of strong flood control and peak shifting at reservoirs, data-driven models often output the erroneous result that "the restored natural water level is lower than the measured water level." This not only violates the fundamental physical law of energy and mass conservation in confined river channels but also directly renders the inversion results unusable in engineering flood control demonstrations.

[0004] Furthermore, current mainstream data-driven reconstruction methods are often limited to numerical sequence predictions of single hydrological stations, which are discrete zero-dimensional or one-dimensional spatial extrapolations. This single-point prediction severs the spatial continuity and hydrodynamic relationship of flood propagation in complex river networks and floodplains, and cannot provide information on the spatiotemporal evolution of floods, inundation extent, and flow field vector distribution. It is difficult to meet the urgent need of modern digital twin watersheds for high-precision, multi-dimensional spatial flood extrapolation across the entire watershed. Summary of the Invention

[0005] The purpose of this application is to provide a watershed flood reconstruction method, system and medium. By introducing a water level increment extrapolation mechanism with physical boundary constraints into the data-driven model and using it as a time-varying boundary condition to drive the physical hydrodynamic model, the physical logic consistency of extreme flood process reconstruction and the spatiotemporal simulation accuracy of the entire watershed are significantly improved while overcoming the defects of pure data-driven extreme extrapolation.

[0006] To achieve the above objectives, this application provides the following technical solution:

[0007] In a first aspect, embodiments of this application provide a method for reconstructing watershed floods, comprising the following steps:

[0008] Step 1: Obtain the historical daily flow and water level sequences of typical control stations in the target watershed, extract the time features of the corresponding time, and divide the dataset into a baseline dataset without reservoir influence and an impact dataset with reservoir influence, with the reservoir construction and water storage node as the boundary.

[0009] Step 2: Construct a unified flow pattern neural network with integrated status identifiers; introduce a database identifier variable, assign a first set value to the database identifier variable of all samples in the baseline period dataset, and assign a second set value to the database identifier variable of all samples in the influence period dataset; concatenate the two datasets to construct a training set with time features and database identifier variables as joint inputs and flow as output, and train the unified flow pattern neural network using a neural network.

[0010] Step 3: Decouple and extract the flow deviation pattern and restore the natural flow using the unified flow pattern neural network; combine the time characteristics of the extreme flood year to be restored with the first set value and the second set value respectively, and input them into the trained unified flow pattern neural network twice in parallel; calculate the predicted flow difference output by the network under different flag activations, and accurately decouple to obtain the daily average flow deviation pattern affected by reservoir regulation in that year; then, use the measured controlled flow to remove the daily average flow deviation pattern to obtain the restored natural flow.

[0011] Step 4: Extract the flow rate and time features from the baseline dataset as input and the water level as output, train a water level prediction neural network with nonlinear extrapolation capability, and configure it as a nonlinear hydrodynamic response operator.

[0012] Step 5: Construct a water level extrapolation mechanism that integrates physical boundary constraints to eliminate the mathematical truncation error of the pure data-driven model when extrapolating extreme floods: Set the measured water level as the lower physical boundary of the extrapolation, and input the restored natural flow and the measured flow into the nonlinear hydrodynamic response operator respectively; By extracting the difference response of the model in the feature space, calculate the dynamic water level increment constrained by the monotonicity of physical laws.

[0013] Step 6: Strictly couple the dynamic water level increment to the measured water level, which serves as the physical reference boundary, to achieve the transfer mapping from the constrained state to the natural state and obtain the restored natural water level of the target station.

[0014] Step 7: Construct a regional two-dimensional hydrodynamic model of the target watershed;

[0015] Step 8: Use the restored natural flow rate obtained in Step 3 and the restored natural water level obtained in Step 6 as the time-varying boundary conditions of the two-dimensional hydrodynamic model to drive the two-dimensional hydrodynamic model to perform calculations and obtain the daily natural evolution process of floods in the entire basin.

[0016] The time feature in step 1 is specifically the day sequence feature DOY.

[0017] In step 3, the first setting value is 0, the second setting value is 1, and the calculation formula for the daily average flow deviation mode and the restored natural flow is as follows:

[0018]

[0019]

[0020] in, To unify the neural network for traffic patterns; To restore natural traffic, To measure the actual flow rate, This refers to the daily average flow deviation pattern; and for A minimum physical threshold is set as a safety net, which is a set percentage of the historical minimum flow in the baseline period, to prevent abnormally low values ​​of the inverted flow that violate hydrological common sense.

[0021] The neural networks in steps 2 and 4 are both backpropagation (BP) feedforward neural networks, which utilize the nonlinear activation function of the hidden layer to provide continuous mathematical extrapolation capability when dealing with extreme flood peaks that exceed historical extreme values.

[0022] In step 5, the water level estimation mechanism that integrates physical boundary constraints specifically calculates the dynamic water level increment using the following difference mapping formula:

[0023]

[0024] in, For dynamic water level increment, For the nonlinear hydrodynamic response operator, The time characteristic is used; this mechanism forces the model output water level increment response to maintain a physical positive correlation with the positive perturbation of the input flow rate.

[0025] In step 6, the restored natural water level at the target site is calculated using the following confined state transition formula:

[0026]

[0027] in, To restore the natural water level, The measured water level is used as the reference; by strictly anchoring the actual measured value, the physical logic of restoring the natural water level to be always greater than the measured water level during the flood control peak reduction period is ensured.

[0028] In step 7, constructing the overall two-dimensional hydrodynamic model of the region requires acquiring and inputting underwater elevation data, digital elevation model (DEM) data, river cross-section topographic data, and surface roughness parameters of the target watershed in order to construct and solve the two-dimensional shallow water dynamic equations.

[0029] In step 8, the time-varying boundary conditions specifically include: using the restored natural flow rate of typical upstream control stations as the upstream inflow boundary of the two-dimensional hydrodynamic model, and using the restored natural water level of typical downstream control stations as the downstream water level control boundary; and outputting the spatial distribution of water depth, velocity vector field and flood inundation range of the grid nodes of the entire basin through boundary driving.

[0030] Secondly, embodiments of this application provide a watershed flood reconstruction system, comprising:

[0031] The data processing module is used to acquire historical hydrological data of the target watershed, extract time features, and divide the dataset into baseline period and impact period datasets.

[0032] The flow restoration module is used to introduce database construction status identifier variables and train a unified conditional control neural network. By extracting the differential output of the same network under different identifier activation states, the climatological flow deviation pattern is accurately decoupled, and the natural flow is restored from the measured regulated flow accordingly.

[0033] The constrained water level extrapolation module has a built-in predictive neural network configured as a nonlinear hydrodynamic response operator. Under the physical lower bound constraint of the measured water level, it calculates the dynamic water level increment by inputting the characteristic difference between the natural flow and the measured flow, and maps it to the measured water level to output the restored natural water level.

[0034] The whole basin hydrodynamic calculation module has a built-in two-dimensional shallow water dynamic equation solving engine. It is used to receive the natural elements output by the flow restoration module and the restricted water level extrapolation module as time-varying physical boundaries, and calculate and output the spatiotemporal evolution process of natural floods in the whole basin.

[0035] Thirdly, embodiments of this application provide a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the watershed flood restoration method as described above.

[0036] Compared with the prior art, the beneficial effects of this application are:

[0037] 1. Overcoming the "Extreme Value Shaving" Deficiency and Recreating the True Extreme Flood Peak: Traditional machine learning algorithms (such as random forests) often fail to predict water levels exceeding historical training records when faced with historically rare extreme floods, leading to the flood peak being erroneously "shaving off." This application utilizes the excellent trend extrapolation capability of BP neural networks, abandoning rigid absolute value prediction and instead predicting the "relative increment" brought about by feature changes. This approach successfully breaks through the data limitations of historical extreme values, enabling a more accurate and realistic recreation of the magnitude of extreme floods.

[0038] 2. Achieving multi-domain feature joint learning to improve model generalization and decoupling accuracy: This application overcomes the data fragmentation defects caused by traditional separate modeling (building two models before and after reservoir construction). It innovatively introduces a "reservoir construction status identifier variable," integrating data from different periods into a unified conditional control neural network. Through joint learning, while sharing the underlying seasonal hydrological runoff characteristics, the network can explicitly and precisely separate and decouple "dam regulation disturbance" from natural evolution with high precision, significantly enhancing the model's generalization ability and anti-overfitting capability.

[0039] 3. Ensuring Correct Physics and Completely Eliminating "Water Level Inversion": During reservoir flood control and peak reduction, the natural flood peak water level will inevitably be higher than the measured water level controlled by the reservoir. However, traditional algorithms often calculate erroneous results due to prediction errors, such as "the natural restored water level is actually lower than the measured water level." This application cleverly introduces a "physical baseline fallback" mechanism, no longer allowing the model to directly guess the final high water level, but instead using the "measured water level" as the minimum baseline for calculation. The model only needs to predict the "water level rise" caused solely by the increase in flow. The sum of the two fundamentally ensures that the calculation results conform to hydrological physics, eliminating the possibility of negative calculated flow rates.

[0040] 4. Achieving a Spatial Leap from "Single Station" to "Entire Basin": Traditional machine learning reconstruction methods typically only output the numerical process line of a specific hydrological station (single point), lacking spatial dimension information. This application uses the high-precision single-station reconstruction results calculated by neural networks as the "source boundary" input into a two-dimensional physical hydrodynamic model that includes actual terrain. This combination of "AI data + physical model" can not only calculate the water level at a specific point, but also intuitively simulate how the flood evolves throughout the entire basin and where the inundation is deepest, achieving a leap from single numerical prediction to multi-dimensional spatial panoramic flood simulation. Attached Figure Description

[0041] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 This is a flowchart of the calculation process of the method of the present invention. Detailed Implementation

[0043] The technical solutions of the embodiments of this application will now be described with reference to the accompanying drawings. It should be noted that similar reference numerals and letters in the following drawings indicate similar items; therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0044] The terms “comprising,” “including,” or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase “comprising one…” does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0045] The terms “first,” “second,” etc., are used only to distinguish one entity or operation from another, and should not be construed as indicating or implying relative importance, nor as requiring or implying any such actual relationship or order between these entities or operations.

[0046] Reference Figure 1 This application provides a watershed flood reconstruction method based on the coupling of neural networks and two-dimensional hydrodynamics, including the following steps:

[0047] Step 1: Obtain the historical daily flow and water level sequences of typical control stations in the target watershed, extract the time features of the corresponding time, and divide the dataset into a baseline dataset without reservoir influence and an impact dataset with reservoir influence, with the reservoir construction and water storage node as the boundary.

[0048] Step 2: Construct a unified flow pattern neural network with integrated status identifiers; introduce a database identifier variable, assign a first set value to the database identifier variable of all samples in the baseline period dataset, and assign a second set value to the database identifier variable of all samples in the influence period dataset; concatenate the two datasets to construct a training set with time features and database identifier variables as joint inputs and flow as output, and train the unified flow pattern neural network using a neural network.

[0049] Step 3: Decouple and extract the flow deviation pattern and restore the natural flow using the unified flow pattern neural network; combine the time characteristics of the extreme flood year to be restored with the first set value and the second set value respectively, and input them into the trained unified flow pattern neural network twice in parallel; calculate the predicted flow difference output by the network under different flag activations, and accurately decouple to obtain the daily average flow deviation pattern affected by reservoir regulation in that year; then, use the measured controlled flow to remove the daily average flow deviation pattern to obtain the restored natural flow.

[0050] Step 4: Extract the flow rate and time features from the baseline dataset as input and the water level as output, train a water level prediction neural network with nonlinear extrapolation capability, and configure it as a nonlinear hydrodynamic response operator.

[0051] Step 5: Construct a water level extrapolation mechanism that integrates physical boundary constraints to eliminate the mathematical truncation error of the pure data-driven model when extrapolating extreme floods: Set the measured water level as the lower physical boundary of the extrapolation, and input the restored natural flow and the measured flow into the nonlinear hydrodynamic response operator respectively; By extracting the difference response of the model in the feature space, calculate the dynamic water level increment constrained by the monotonicity of physical laws.

[0052] Step 6: Strictly couple the dynamic water level increment to the measured water level, which serves as the physical reference boundary, to achieve the transfer mapping from the constrained state to the natural state and obtain the restored natural water level of the target station.

[0053] Step 7: Construct a regional two-dimensional hydrodynamic model of the target watershed;

[0054] Step 8: Use the restored natural flow rate obtained in Step 3 and the restored natural water level obtained in Step 6 as the time-varying boundary conditions of the two-dimensional hydrodynamic model to drive the two-dimensional hydrodynamic model to perform calculations and obtain the daily natural evolution process of floods in the entire basin.

[0055] This embodiment takes the natural reconstruction of the extreme flood event in 2020 as an example to illustrate in detail the specific implementation steps of the basin flood reconstruction method based on the coupling of neural network and two-dimensional hydrodynamics proposed in this invention.

[0056] Step 1: Obtain historical hydrological data and dataset division for the calculation area.

[0057] Specifically, daily flow sequences of typical control stations in the target watershed were collected over the years. ) and water level sequence ( Extract the time features corresponding to each data point. In this embodiment, the preferred feature is the day of year (DOY, which is the day of the year and ranges from 1 to 366).

[0058] Using the year the target reservoir was built and began impounding water (2003) as the physical time node, the dataset was strictly divided into:

[0059] Baseline datasets without reservoir impact (e.g., 1983–2003);

[0060] Data sets showing the impact of intensive reservoir regulation during certain periods (e.g., 2004–2021).

[0061] The target flood year data to be restored (e.g., 2020, the year of the extreme flood).

[0062] Before building the model, the baseline and impact period data were cleaned, missing values ​​(NaN) were removed, and a structured training sample matrix was constructed.

[0063] Step 2: Train the neural network for the flow patterns during the baseline and impact periods.

[0064] This embodiment constructs a unified Conditioned BP Neural Network to achieve joint learning and parameter sharing of multi-basin state data. Specifically, a one-dimensional database construction identifier variable (denoted as Regime Indicator) is introduced. ). All samples from the baseline dataset Assigning a value of 0 will affect all samples in the dataset during the affected period. The value is assigned to 1. The data from the two periods are vertically combined and concatenated to construct a structure containing feature variables. The two-dimensional joint input matrix, with measured flow rate As output labels. In the deep learning framework included in MATLAB, hidden layer neurons are configured (e.g., 10 neurons), and the unified flow pattern neural network is trained using a mixed dataset. Through this joint learning mechanism, the network's hidden layers can learn the shared seasonal evolution patterns of river runoff using full-sample data, while explicitly internalizing the flow disturbance mechanism caused by reservoir regulation. Nonlinear weight mapping for variable activation.

[0065] Step 3: Decouple the extracted flow deviation pattern from the natural flow restoration.

[0066] Obtain the daily sequence of numbers for the entire year of 2020 to be restored. ), fully utilizing the conditional mapping relationships already internalized in the unified network in step S2, construct two sets of virtual test feature sets respectively: the first set of features is (Simulating a natural, unreserved state), the second set of features is: (Simulating the storage and regulation state of the reservoir).

[0067] The two sets of features mentioned above are input in parallel into the trained unified flow pattern neural network, which outputs the baseline climatological predicted flow for 2020. and the climatological predicted flow during the impact period Calculating the difference response between the two yields the daily average flow deviation pattern, precisely extracted from the joint feature space by the neural network and driven solely by reservoir regulation. :

[0068]

[0069] Subsequently, using the measured flow rate in 2020 After eliminating this bias, the restored natural flow rate is obtained. :

[0070]

[0071] To prevent extreme outliers caused by neural network extrapolation, Set a minimum physical threshold limit, for example, set This ensures that the restored flow rate conforms to basic hydrological knowledge and that there are no negative flow rates.

[0072] Step 4: Construct a water level prediction neural network configured with a nonlinear hydrodynamic response operator.

[0073] Extract measured flow rates for the baseline period (1983-2003). Day ordinal number The two-dimensional feature matrix is ​​used as input, with the corresponding measured water level in the baseline period. As the output label, a third BP neural network is trained. ).

[0074] Once trained, the neural network is configured as a nonlinear hydrodynamic response operator. Compared to traditional tree-based models such as random forests, this BP network operator has a continuous nonlinear activation function, which can provide mathematical extrapolation capabilities that conform to the monotonically increasing law of fluid dynamics when encountering extreme flood peak flows that exceed historical extremes.

[0075] Step 5: Execute the water level projection mechanism that integrates physical boundary constraints.

[0076] This is the core step in overcoming the "peak-shaving" and "water level inversion" defects of purely data-driven models. Instead of directly using the model to predict absolute high water levels, the model is configured as an incremental calculator in the feature space.

[0077] The 2020 restored natural flow obtained in step S3 and the actual traffic volume in 2020 (Combined with the corresponding day sequence number) The inputs are respectively fed in parallel to the hydrodynamic response operator. In the process, the difference response between the two is calculated to obtain the dynamic water level increment. :

[0078]

[0079] This step extracts "simply due to traffic from..." Return to natural state The resulting dynamic rise in water level effectively filters out the systematic underestimation bias of the model in predicting extreme absolute water levels.

[0080] Step 6: Generate a physically constrained, restored natural water level.

[0081] The dynamic water level increment calculated in step S5 Strictly superimposed to the corresponding actual measured water level in 2020. Above (i.e., above the lower bound anchor point of the physical reference):

[0082]

[0083] Due to the reservoir's flood control and peak reduction, the natural flow... It must be greater than the traffic. Therefore, the operator output It is always a positive value. This formula, from an absolute mathematical and physical logic perspective, guarantees that "the natural water level will be restored during flood season." It must be higher than the measured water level ".

[0084] Step 7: Construct a two-dimensional hydrodynamic model of the entire region.

[0085] To achieve the leap from "single-point prediction" to "area evolution," high-precision underwater topographic data of the target watershed is obtained. Unstructured triangular meshes are generated based on the topographic data, and Manning's roughness parameter (n) is assigned to different land use types.

[0086] Based on the laws of conservation of mass and momentum, a solution engine for two-dimensional shallow water equations, including continuity equations and momentum equations, is constructed.

[0087] Step 8: Time-varying boundary conditions driving and reconstructing the natural flood process across the entire basin.

[0088] The high-precision single-point reconstruction results obtained in steps 3 and 6 using the neural network incremental method are transformed into the dynamic physical boundary of the two-dimensional hydrodynamic model:

[0089] 1) Reconstruct the natural flow process of typical upstream stations , is set as the upstream time-varying inflow boundary of the model;

[0090] 2) Restore the natural water level process of typical downstream stations , is set as the downstream time-varying water level control boundary of the model.

[0091] Driven by the aforementioned natural boundary conditions, the two-dimensional shallow water equations are solved explicitly using the finite volume method (FVM) with time-stepping. The model ultimately outputs the daily flood evolution process of the entire basin during the 2020 extreme flood without the influence of dam regulation, including but not limited to: the spatial distribution of water depth at each grid node, the dynamic changes of the velocity vector field, and the maximum flood inundation range.

[0092] This application provides a watershed flood restoration system, including:

[0093] The data processing module is used to acquire historical hydrological data of the target watershed, extract time features, and divide the dataset into baseline period and impact period datasets.

[0094] The flow restoration module is used to introduce database construction status identifier variables and train a unified conditional control neural network. By extracting the differential output of the same network under different identifier activation states, the climatological flow deviation pattern is accurately decoupled, and the natural flow is restored from the measured regulated flow accordingly.

[0095] The constrained water level extrapolation module has a built-in predictive neural network configured as a nonlinear hydrodynamic response operator. Under the physical lower bound constraint of the measured water level, it calculates the dynamic water level increment by inputting the characteristic difference between the natural flow and the measured flow, and maps it to the measured water level to output the restored natural water level.

[0096] The whole basin hydrodynamic calculation module has a built-in two-dimensional shallow water dynamic equation solving engine. It is used to receive the natural elements output by the flow restoration module and the restricted water level extrapolation module as time-varying physical boundaries, and calculate and output the spatiotemporal evolution process of natural floods in the whole basin.

[0097] This application provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the watershed flood restoration method described above.

[0098] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0099] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0100] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0101] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0102] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0103] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, like read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0104] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0105] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for reconstructing watershed floods, characterized in that, Includes the following steps: Step 1: Obtain the historical daily flow and water level sequences of typical control stations in the target watershed, extract the time features of the corresponding time, and divide the dataset into a baseline dataset without reservoir influence and an impact dataset with reservoir influence, with the reservoir construction and water storage node as the boundary. Step 2: Construct a unified flow pattern neural network with integrated status identifiers; introduce a database identifier variable, assign a first set value to the database identifier variable of all samples in the baseline period dataset, and assign a second set value to the database identifier variable of all samples in the influence period dataset; concatenate the two datasets to construct a training set with time features and database identifier variables as joint inputs and flow as output, and train the unified flow pattern neural network using a neural network. Step 3: Use the unified flow pattern neural network to decouple and extract the flow deviation pattern from the restored natural flow. The time characteristics of the extreme flood year to be restored are combined with the first and second set values ​​respectively, and input into the trained unified flow pattern neural network in parallel twice. The predicted flow difference output by the network under different flag activation is calculated, and the daily average flow deviation pattern affected by reservoir regulation in that year is obtained by precise decoupling. Then, the daily average flow deviation pattern is removed by actual measured regulation flow to obtain the restored natural flow. Step 4: Extract the flow rate and time features from the baseline dataset as input and the water level as output, train a water level prediction neural network with nonlinear extrapolation capability, and configure it as a nonlinear hydrodynamic response operator. Step 5: Construct a water level extrapolation mechanism that integrates physical boundary constraints to eliminate the mathematical truncation error of the pure data-driven model when extrapolating extreme floods: Set the measured water level as the lower physical boundary of the extrapolation, and input the restored natural flow and the measured flow into the nonlinear hydrodynamic response operator respectively; By extracting the difference response of the model in the feature space, calculate the dynamic water level increment constrained by the monotonicity of physical laws. Step 6: Strictly couple the dynamic water level increment to the measured water level, which serves as the physical reference boundary, to achieve the transfer mapping from the constrained state to the natural state and obtain the restored natural water level of the target station. Step 7: Construct a regional two-dimensional hydrodynamic model of the target watershed; Step 8: Use the restored natural flow rate obtained in Step 3 and the restored natural water level obtained in Step 6 as the time-varying boundary conditions of the two-dimensional hydrodynamic model to drive the two-dimensional hydrodynamic model to perform calculations and obtain the daily natural evolution process of floods in the entire basin.

2. The method for reconstructing watershed floods according to claim 1, characterized in that, The time feature in step 1 is specifically the day sequence feature DOY.

3. The method for reconstructing watershed floods according to claim 1, characterized in that, In step 3, the first setting value is 0, the second setting value is 1, and the calculation formula for the daily average flow deviation mode and the restored natural flow is as follows: ， in, To unify the neural network for traffic patterns; To restore natural traffic, To measure the actual flow rate, This refers to the daily average flow deviation pattern; and for A minimum physical threshold is set as a safety net, which is a set percentage of the historical minimum flow in the baseline period, to prevent abnormally low values ​​of the inverted flow that violate hydrological common sense.

4. The method for reconstructing watershed floods according to claim 1, characterized in that, The neural networks in steps 2 and 4 are both backpropagation (BP) feedforward neural networks, which utilize the nonlinear activation function of the hidden layer to provide continuous mathematical extrapolation capability when dealing with extreme flood peaks that exceed historical extreme values.

5. The method for reconstructing watershed floods according to claim 1, characterized in that, In step 5, the water level estimation mechanism that integrates physical boundary constraints specifically calculates the dynamic water level increment using the following difference mapping formula: ， in, For dynamic water level increment, For the nonlinear hydrodynamic response operator, The time characteristic is used; this mechanism forces the model output water level increment response to maintain a physical positive correlation with the positive perturbation of the input flow rate.

6. The method for reconstructing watershed floods according to claim 1, characterized in that, In step 6, the restored natural water level at the target site is calculated using the following confined state transition formula: ， in, To restore the natural water level, The measured water level is used as the reference; by strictly anchoring the actual measured value, the physical logic of restoring the natural water level to be always greater than the measured water level during the flood control peak reduction period is ensured.

7. The method for reconstructing watershed floods according to claim 1, characterized in that, In step 7, constructing the overall two-dimensional hydrodynamic model of the region requires acquiring and inputting underwater elevation data, digital elevation model (DEM) data, river cross-section topographic data, and surface roughness parameters of the target watershed in order to construct and solve the two-dimensional shallow water dynamic equations.

8. The method for reconstructing watershed floods according to claim 1, characterized in that, In step 8, the time-varying boundary conditions specifically include: using the restored natural flow rate of typical upstream control stations as the upstream inflow boundary of the two-dimensional hydrodynamic model, and using the restored natural water level of typical downstream control stations as the downstream water level control boundary; and outputting the spatial distribution of water depth, velocity vector field and flood inundation range of the grid nodes of the entire basin through boundary driving.

9. A watershed flood restoration system, characterized in that, include: The data processing module is used to acquire historical hydrological data of the target watershed, extract time features, and divide the dataset into baseline period and impact period datasets. The flow restoration module is used to introduce database construction status identifier variables and train a unified conditional control neural network. By extracting the differential output of the same network under different identifier activation states, the climatological flow deviation pattern is accurately decoupled, and the natural flow is restored from the measured regulated flow accordingly. The constrained water level extrapolation module has a built-in predictive neural network configured as a nonlinear hydrodynamic response operator. Under the physical lower bound constraint of the measured water level, it calculates the dynamic water level increment by inputting the characteristic difference between the natural flow and the measured flow, and maps it to the measured water level to output the restored natural water level. The whole basin hydrodynamic calculation module has a built-in two-dimensional shallow water dynamic equation solving engine. It is used to receive the natural elements output by the flow restoration module and the restricted water level extrapolation module as time-varying physical boundaries, and calculate and output the spatiotemporal evolution process of natural floods in the whole basin.

10. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the watershed flood restoration method as described in any one of claims 1 to 8.

Images