A Watershed Flood Simulation and Optimization Method Based on Physical Information Neural Networks

By introducing physical information neural networks and partial differential equations into the watershed flood model, and combining them with the urban flood spatiotemporal simulation model, the problems of insufficient simulation accuracy and computational efficiency in existing technologies are solved, and efficient and reliable flood process simulation is achieved.

CN120470914BActive Publication Date: 2026-11-10NANJING NORMAL UNIVERSITY
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202510577437.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2025-05-06
Publication Date
2026-11-10
Estimated Expiration
2045-05-06

AI Technical Summary

Technical Problem

Existing watershed flood models are insufficient in terms of simulation accuracy and computational efficiency. Physics-based models are computationally complex and time-consuming, while machine learning-based models require a large number of free parameters and their prediction mechanisms are difficult to explain.

Method used

We employ Physical Information Neural Networks (PINNs) combined with Partial Differential Equations (PDEs) as the loss function. By modifying the loss function to incorporate physical knowledge and constrain the model parameter space, we construct a watershed flood model based on CNN neural networks. We then combine this model with a spatiotemporal simulation model of urban flooding to build the dataset and train the model.

Benefits of technology

It achieves high-precision flood process simulation, improves calculation speed by more than 300 times, reduces model uncertainty, and provides more reliable flood prediction results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120470914B_ABST
    Figure CN120470914B_ABST
Patent Text Reader

Abstract

This invention relates to a watershed flood simulation optimization method based on Physical Information Neural Networks (PINNs). The aim is to construct an accelerated solution for watershed flood simulation based on PINNs, ensuring the continuity of model operation and the physical property of mass conservation. The method includes the following steps: simulating watershed floods using an urban flood spatiotemporal simulation model to obtain the dataset required for constructing the physical neural network; integrating physical laws and simplifying them into constraints to facilitate integration into the neural network structure; establishing partial differential equations (PDEs) to describe the evolution of river floods, using them as the loss function of the physical information neural network model; constructing and training a PINNs watershed flood model based on a CNN neural network; and establishing and evaluating the model's performance using evaluation metrics. The beneficial effect of this invention is to improve the speed and accuracy of watershed flood simulation.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of urban flooding technology in the field of hydrology and water resources, and specifically relates to a watershed flood simulation and optimization method based on physical information neural networks. Background Technology

[0002] Watershed floods, characterized by their wide coverage, long duration, and destructive power, threaten the lives and property of the people, and also severely impact urban infrastructure such as transportation, electricity, and communications, disrupting normal production and daily life. Current research on watershed flood models falls into two categories: physical model simulations and machine learning (ML)-based numerical simulations.

[0003] Physics-based model simulations focus on the fundamental characteristics of river floods, validating and applying a large amount of prior physical knowledge. While achieving high simulation accuracy, they also suffer from computational complexity and time consumption. The problem with machine learning-based numerical simulation methods is that they require a large number of free parameters to accurately capture the potential complexity of the data. Furthermore, they are prone to extrapolation and generalization problems, and often fail to explain the reasoning and underlying mechanisms behind specific predictions.

[0004] Physically Information Neural Networks (PINNs) introduce physics-based regularization to neural networks. They constrain the parameter space of the learnable model by modifying the loss function to include terms encoding underlying physics. A common approach is to incorporate the residuals of partial differential equations (PDEs) into the loss function, aiming to create a more physically consistent model. By using a loss function to increase prediction errors that violate fundamental physics, the training process produces a model that minimizes data fitting errors while remaining consistent with fundamental physics. This physics-based machine learning approach has been successfully applied to various modeling tasks, such as prospective problems—inferring potential solutions to systems governed by known equations.

[0005] In this regard, this application extends the application of PINNs to complex real-world problems in fluid dynamics, and constructs an accelerated solution method based on PINNs for numerical hydrodynamic simulators in flood models, ensuring physical properties such as mass conservation during model operation. Summary of the Invention

[0006] The purpose of this invention is to provide a watershed flood simulation optimization method based on physical information neural networks to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A watershed flood simulation and optimization method based on physical information neural networks includes the following steps:

[0009] S1: Simulate watershed flooding using the Spatio-temporal flood simulation model for urban (STFS-Urban) to obtain the dataset required for constructing the physical neural network;

[0010] S2: Integrate physical laws and simplify them into constraints to facilitate integration into neural network structures;

[0011] S3: Establish a partial differential equation (PDE) to describe the evolution of river floods, and use it as the loss function of the physical information neural network model;

[0012] S4: Construct and train the PINNs watershed flood model based on the CNN neural network to obtain the optimized parameters;

[0013] S5: Establish model evaluation indicators and conduct evaluation.

[0014] Further, S1 includes the following steps:

[0015] S11: Based on topographic maps and underlying surface conditions, a refined model of the watershed's river channels is constructed. River direction, channel width, and confluence sequence are determined using remote sensing images and a digital elevation model (DEM). The river roughness is determined by referencing elevation differences, and the initial river depth is designed using measured data to assist in determining the river's depth. In the urban pipe network surface modeling, the stormwater pipe network system within the study area undergoes topological checking and generalization. Basic pipe segment parameters (such as pipe roughness) are set, and an underground pipe network model is constructed. The DEM data required for surface runoff calculation is input, and surface roughness and infiltration parameters are set with reference to land use data. It is considered that when the river passes through towns, some rainwater enters the river channel through residential drainage outlets, increasing the additional flow. To more accurately simulate the watershed's hydrological conditions, the impact of urban drainage on river flow must be considered. During the modeling process, the overflow flow at pipe points and the flow at drainage outlets along the river, calculated by coupling the pipe network with the surface, are used as a supplement to the river flow. The urban stormwater discharge flow is added to the river flow to complete the overall modeling of the watershed. Based on the above modeling steps, the watershed flood is simulated using the Spatio-temporal flood simulation model for urban (STFS-Urban) to construct the dataset required for the physical neural network.

[0016] S12: Construct a training dataset based on the simulated watershed flood results, including variables t (time), Y (water depth), (i, j) (pixel location), s (flow rate), riverbed elevation (h), and rainfall (P).

[0017] Further, S2 includes the following steps:

[0018] S21: The Shallow Water Equations (SWE) are a simplified form of the Navier-Stokes equations, used to describe fluid motion in shallow water (such as rivers, lakes, and estuaries) that is uniform along the depth axis. The partial differential equations for flood evolution described in this invention employ the Shallow Water Equations:

[0019]

[0020] In the above equations, u is the velocity vector, t is time, ν is the kinematic viscosity coefficient, p is the fluid pressure, ρ0 is the fluid density, and g is the acceleration due to gravity. These equations represent the continuity equation for mass conservation and the momentum equation for momentum conservation, respectively. For a vector differential operator (or Hami sub-operator), satisfying:

[0021]

[0022] in, These are the time rates of change of the three coordinate axes (x, y, z) representing spatial location, i.e. the components of flow velocity on the three coordinate axes. The optimization simulation method used in this invention will calculate the interaction of fluids within the watershed based on the above equations.

[0023] S22: For the above system of equations, this invention expresses it as follows:

[0024] Δ(Storage) = Δ(Source - Sink)

[0025] That is, the storage capacity within the cell is equal to the flow difference between the source and sink. Based on this, the PINNs model under the optimization method predicts the fluid depth at cell location (i, j) at time t. and true fluid depth This can be expressed as:

[0026]

[0027] and Let Δx and Δy refer to the inflow and outflow of pixel (i, j) at time t, respectively, and Δx and Δy refer to the pixel width in the x and y directions, respectively. The above formula shows that the total flow volume within the predicted time t minus the actual volume at time t-1 equals the total inflow minus the total outflow. Therefore, this formula can be rewritten as:

[0028]

[0029] because Represents the total outflow. This always holds true; therefore, we can deduce that:

[0030]

[0031] This is equivalent to limiting the total predicted amount at time t so that it cannot exceed the sum of the water volume at time t-1 and the net inflow from time t-1 to time t;

[0032] S23: Continuing to integrate the continuity equations, the following constraints are obtained:

[0033]

[0034] Let (i, j) be the inflow at time t+1. The complete continuity equation is shown below:

[0035]

[0036] By integrating physical laws and simplifying them into constraints, it is easier to integrate neural network structures and constrain the gradient of neural networks.

[0037] Further, S3 includes the following steps:

[0038] S31: The PDE used in the linearized wave equation model calculation method is represented as follows:

[0039]

[0040] Where c is the wave velocity, and u(x,t) represents the displacement of the actual wave. For the Laplace operator,

[0041] S32: Estimation using a neural network Then, through backpropagation, we obtain... This process is repeated to obtain the nth derivative, which is then added to the loss function to optimize the watershed flood simulation method based on the CNN framework.

[0042]

[0043] The first term is the data fitting error term, and the second term is the PDE residual regularization term;

[0044] S33: After integrating the physical information from S2, the physical information loss function is defined as:

[0045]

[0046] The first term is the difference between the actual output water depth z and the predicted output water depth. The mean squared error between them, i.e., the data fit term, It is the actual volume at time t-1. It is the actual volume at time t+1. It is the predicted volume at time t. ReLu(x) = max(x, 0) refers to the inflow of pixel (i, j) at times t and t-1, and refers to the ramp function in mathematical terms, which is responsible for penalizing violations of physical laws; the above formula shows that if the shallow water equations are violated, the loss function will include additional values.

[0047] S34: To ensure the effectiveness of the above physical information loss function, scaling and regularization are applied to the loss function so that the expected MSE is less than 10. -1 m 2 :

[0048]

[0049] Where, n x With n y The n represents the total number of pixels in the modeling region along the x and y directions. x ×n y Represents the total area of ​​the modeling region;

[0050] Further, S4 includes the following steps:

[0051] S41: The obtained watershed flood evolution data corresponding to each rainfall scenario are used as training samples, and a neural network architecture is built based on the CNN model framework and the hyperparameters and weights are initialized; the parameters include: number of iterations (epoch), learning rate (learn_rate), batch size (batch), convolution kernel size (filter), pooling size (pooling), etc.

[0052] S43: Perform dataset partitioning; divide the dataset into training and test sets;

[0053] S42: The loss function uses the result calculated in S3; the Adam optimizer is selected.

[0054] S43: In each iteration, the training set is traversed, the loss is calculated, and the parameters are updated; whether to continue the loop is determined based on whether the loss function L < ε or whether the preset total number of iterations has been reached, where ε is a constant. When the above conditions are met, the loop stops.

[0055] Further, S5 includes the following steps:

[0056] S5: To evaluate the predictive performance of the model, this application uses two metrics:

[0057] S51: The first metric is the root mean square error between the true depth and the predicted depth.

[0058]

[0059] S52: Use mRMSE for a more realistic estimate of the true error:

[0060]

[0061] Where Ω refers to the set of pixels. This refers to the set of all pixels whose actual pixel fluid depth and predicted pixel fluid depth are not zero. and This refers to the measured and predicted values ​​of fluid depth within the Ω pixel set.

[0062] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0063] 1. Based on artificial intelligence technology, it has broad applicability and cutting-edge nature in the context of big data and cloud computing;

[0064] 2. By "embedding" physical knowledge into deep learning models, we can obtain physically consistent simulation and prediction results, thereby simulating the process of river floods more accurately and with high credibility.

[0065] 3. Compared to hydrological and hydrodynamic models, it can output forecast results much faster, and the calculation speed after training can be increased by more than 300 times;

[0066] 4. By employing physical constraint deep learning parameter optimization methods, model uncertainty is reduced. Attached Figure Description

[0067] Figure 1 This is a framework diagram of the present invention;

[0068] Figure 2 This is a schematic diagram of the network structure of PINNs. Detailed Implementation

[0069] The present invention will be further described in detail below with reference to the accompanying drawings.

[0070] Example: A watershed flood simulation and optimization method based on physical neural networks, the process is as follows: Figure 1 As shown, it includes the following steps:

[0071] S1: Simulate watershed flooding using the Spatio-temporal flood simulation model for urban (STFS-Urban) and construct the dataset required for the physical neural network;

[0072] S2: Integrate physical laws, simplify them into constraints, and integrate them into the physical neural network structure;

[0073] S3: Establish a partial differential equation (PDE) to describe the evolution of river floods, and use it as the loss function of the physical information neural network model;

[0074] S4: Construct and train the PINNs watershed flood model based on the CNN neural network to obtain the optimized parameters;

[0075] S5: Establish model evaluation indicators and conduct evaluation;

[0076] Further, S1 includes the following steps:

[0077] S111: Based on topographic maps and underlying surface conditions, a refined model of the watershed's river channels is constructed. River direction, channel width, and confluence sequence are determined using remote sensing images and a digital elevation model (DEM). The river roughness is determined by referencing elevation differences, and the initial river depth is designed using measured data to assist in determining the river's depth. In the urban pipe network surface modeling, the stormwater pipe network system within the study area undergoes topological checking and generalization. Basic pipe segment parameters (such as pipe roughness) are set, and an underground pipe network model is constructed. The DEM data required for surface runoff calculation is input, and surface roughness and infiltration parameters are set with reference to land use data. It is considered that when the river passes through towns, some rainwater enters the river channel through residential drainage outlets, increasing the additional flow. To more accurately simulate the watershed's hydrological conditions, the impact of urban drainage on river flow must be considered. During the modeling process, the overflow flow of pipe points and the flow of drainage outlets along the river are calculated by coupling the pipe network with the ground surface as a supplement to the river flow. The urban rainwater discharge flow is added to the river flow to complete the overall modeling of the watershed.

[0078] S112: Design different flood scenarios and design rainfall patterns (Chicago rainfall pattern) based on different rainfall return periods (1 year, 5 years, 10 years, 20 years, 50 years, 100 years). The rainfall peak coefficient is 0.3, 0.5, and 0.7. The duration of each rainfall event is 6 hours. The river source flow is designed based on measured data. The data is output every half hour. 12 inundation depth maps are generated for each rainfall event.

[0079] S113: Simulate watershed flooding using the Spatio-temporal flood simulation model for urban (STFS-Urban). Inputs include: elevation (DEM), rainfall, land use data, source flow (QFIX), and infiltration coefficient. Input data requirements are as follows:

[0080] Table 1: Data Requirements

[0081] Data Name Data format Data Description Digital Elevation Data (DEM) TIFF Experimental area DEM data Land use types TIFF Experimental area DEM data water depth txt Historical measured water depth data of the experimental area Rainfall txt Measured water depth data from rainfall in the experimental area

[0082] S113: Conduct flood simulations under multiple rainfall scenarios after parameter calibration using historical water depth and precipitation datasets;

[0083] S12: Construct a framework for variables t (time) and Z based on the simulated watershed flood results. t-1,...,t-n (Water depth), (i, j) (pixel spatiotemporal location), s (flow rate), riverbed elevation (h), rainfall (P) 1,2,...,m Z t Training dataset for (water depth);

[0084] Further, S2 includes the following steps:

[0085] S21: The Shallow Water Equations (SWE) are a simplified form of the Navier-Stokes equations, used to describe fluid motion in shallow water (such as rivers, lakes, and estuaries) that is uniform along the depth axis. The partial differential equations for flood evolution described in this invention employ the Shallow Water Equations:

[0086]

[0087] In the above equations, u is the velocity vector, t is time, ν is the kinematic viscosity coefficient, p is the fluid pressure, ρ0 is the fluid density, and g is the acceleration due to gravity. These equations represent the continuity equation for mass conservation and the momentum equation for momentum conservation, respectively. For a vector differential operator (or Hami sub-operator), satisfying:

[0088]

[0089] in, These represent the time variation of the three coordinate axes (x, y, z) representing the spatial location, i.e., the components of the flow velocity on the three coordinate axes;

[0090] S22: For the above shallow water equations, this invention expresses them as follows:

[0091] Δ(Storage) = Δ(Source - Sink)

[0092] That is, the storage capacity within the cell is equal to the flow difference between the source and sink. Based on this, the PINNs model under the optimization method predicts the fluid depth at cell location (i, j) at time t. and true fluid depth This can be expressed as:

[0093]

[0094] and Let Δx and Δy refer to the inflow and outflow of pixel (i, j) at time t, respectively, and Δx and Δy refer to the pixel width in the x and y directions, respectively. The above formula shows that the total flow volume within the predicted time t minus the actual volume at time t-1 equals the total inflow minus the total outflow. Therefore, this formula can be rewritten as:

[0095]

[0096] because Represents the total outflow. This always holds true; therefore, we can deduce that:

[0097]

[0098] This is equivalent to limiting the total predicted amount at time t so that it cannot exceed the sum of the water volume at time t-1 and the net inflow from time t-1 to time t.

[0099] S23: Similarly, integrating the continuity equations, we obtain the following constraints:

[0100]

[0101] Let (i, j) be the inflow at time t+1. The complete continuity equation is shown below:

[0102]

[0103] Further, S3 includes the following steps:

[0104] S31: The PDE used in the linearized wave equation model calculation method is represented as follows:

[0105]

[0106] Where c is the wave velocity, and u(x,t) represents the displacement of the actual wave. For the Laplace operator,

[0107] S32: Estimation using a neural network Then, through backpropagation, we obtain... This process is repeated to obtain the nth derivative, which is then added to the loss function to optimize the watershed flood simulation method based on the CNN framework.

[0108]

[0109] The first term is the data fitting error term, and the second term is the PDE residual regularization term. A schematic diagram of the network structure is shown below. Figure 2 As shown.

[0110] S33: After integrating the physical information from S2, the physical information loss function is defined as:

[0111]

[0112] The first term is the difference between the actual output water depth z and the predicted output water depth. The mean squared error between them, i.e., the data fit term, It is the actual volume at time t-1. It is the actual volume at time t+1. It is the predicted volume at time t. The inflow at time t and t-1 refers to the inflow at pixel (i, j), and ReLu(x) = max(x, 0) refers to the ramp function in mathematics, which is responsible for penalizing violations of physical laws; the above formula shows that if the shallow water equations are violated, the loss function will include additional values.

[0113] S34: To ensure the effectiveness of the above physical information loss function, scaling and regularization are applied to the loss function so that the expected MSE is less than 10. -1 m 2 :

[0114]

[0115] Where, n x With n y The n represents the total number of pixels in the modeling region along the x and y directions. x ×n y Represents the total area of ​​the modeling region.

[0116] Further, S4 includes the following steps:

[0117] S41: The obtained watershed flood evolution data corresponding to each rainfall scenario are used as training samples, and a neural network architecture is built based on the CNN model framework and the hyperparameters and weights are initialized; the parameters include: number of iterations (epoch), learning rate (learn_rate), batch size (batch), convolution kernel size (filter), pooling size (pooling), etc.

[0118] S43: Perform dataset partitioning; divide the dataset into training and test sets;

[0119] S42: The loss function uses the result calculated in S3; the Adam optimizer is selected.

[0120] S43: In each iteration, the training set is traversed, the loss is calculated, and the parameters are updated; whether to continue the loop is determined based on whether the loss function L < ε or whether the preset total number of iterations has been reached, where ε is a constant. When the above conditions are met, the loop stops.

[0121] Further, S5 includes the following steps:

[0122] S5: To evaluate the predictive performance of the model, this application uses two metrics:

[0123] S51: The first metric is the root mean square error between the true depth and the predicted depth.

[0124]

[0125] S52: Use mRMSE for a more realistic estimate of the true error:

[0126]

[0127] Where Ω refers to the set of pixels. This refers to the set of all pixels whose actual pixel fluid depth and predicted pixel fluid depth are not zero. and This refers to the measured and predicted values ​​of fluid depth within the Ω pixel set.

[0128] The above description is intended only to help understand the method and core concepts of the present invention. Furthermore, those skilled in the art may make appropriate modifications in specific implementations and applications based on the ideas of the present invention. Therefore, the content of this specification should not be considered as a limitation of the present invention.

Claims

1. A watershed flood simulation and optimization method based on physical information neural networks, characterized in that, The method includes the following steps: S1: Simulate watershed flooding using an urban flood spatiotemporal simulation model to obtain the dataset needed to construct a physical neural network; S2: Integrate physical laws, simplify them into constraints, and integrate the neural network structure; S3: Establish a partial differential equation to describe the evolution of river floods and use it as the loss function of the physical information neural network model; S4: Construct and train the PINNs watershed flood model based on the CNN neural network to obtain the optimized parameters; S5: Establish model evaluation indicators and conduct evaluation; Step S1 is as follows: S11: Simulate watershed flooding using a spatiotemporal simulation model of urban flooding, taking into account the additional flow brought by urban drainage when the watershed flows through the city; S12: Construct a training dataset based on the simulated watershed flood results, including variables time t, water depth z, pixel location (i, j), flow rate s, riverbed elevation h, and rainfall P. S2 includes the following steps: S21: The Shallow Water Equations (SWE) are a simplified form of the Navier-Stokes equations, used to describe fluid motion at shallow depths and uniform along the depth axis. The partial differential equations for flood evolution employ the Shallow Water Equations: In the above formula, u is the velocity vector, and t is time. Where is the kinematic viscosity coefficient, and p is the fluid pressure. Let be the fluid density, and g be the acceleration due to gravity; the above equations are the continuity equation representing the conservation of mass and the momentum equation representing the conservation of momentum, respectively. The vector differential operator, also known as the Hami sub-operator, satisfies: in, , , These represent the time variability of the three coordinate axes (x, y, z) representing spatial location, i.e., the components of the flow velocity on the three coordinate axes. S22: The above shallow water equations can be expressed as: That is, the storage capacity within the cell is equal to the flow difference between the source and sink. Based on this, the PINNs model under the optimization method predicts the fluid depth at cell location (i, j) at time t. and true fluid depth Expressed as: and These refer to the inflow and outflow at time t for pixel (i, j), respectively. and The pixel width refers to the pixel width in the x and y directions. The above formula shows that the total flow volume within the predicted time t minus the actual volume at time t-1 equals the total inflow minus the total outflow. Therefore, this formula can be restated as: because Represents the total outflow. The value ≥ 0 is always true; therefore, it can be deduced that: This is equivalent to limiting the total predicted amount at time t so that it cannot exceed the sum of the water volume at time t-1 and the net inflow from time t-1 to time t; S23: Continuing to integrate the continuity equations, the following constraints are obtained: Let (i, j) be the inflow at time t+1. The complete continuity equation is shown below: 。 2. The watershed flood simulation and optimization method based on physical information neural network according to claim 1, characterized in that, S3 includes the following steps: S31: The PDE used in the linearized wave equation model calculation method is represented as follows: Where c is the wave velocity, and u(x,t) represents the displacement of the actual wave. For the Laplace operator, S32: Estimation using a neural network Then backpropagation yields , This process is repeated to obtain the nth derivative, which is then added to the loss function to optimize the watershed flood simulation method based on the CNN framework. The first term is the data fitting error term, and the second term is the PDE residual regularization term. S33: After integrating the complete continuity equation of S2, the physical information loss function is defined as: The first term is the difference between the actual output water depth z and the predicted output water depth. The mean squared error between them, i.e., the data fit term, It is the actual volume at time t-1. It is the actual volume at time t+1. It is the predicted volume at time t. , ReLu(x) = max(x,0) refers to the inflow rate at time t and t+1 for pixel (i,j). It represents the ramp function in mathematics, responsible for penalizing violations of physical laws. S34: To ensure the effectiveness of the above physical information loss function, scaling and regularization are applied to the loss function so that the expected MSE is less than 10. -1 m 2 : , in, and This represents the total number of pixels in the modeling region along the x and y directions. This represents the total area of ​​the modeled region.

3. The watershed flood simulation and optimization method based on physical information neural network according to claim 2, characterized in that, S4 includes the following steps: S41: The obtained watershed flood evolution data corresponding to each rainfall scenario are used as training samples, and a neural network architecture is built based on the CNN model framework and the hyperparameters and weights are initialized; the parameters include: number of epochs, learning rate, batch size, kernel size (filter), and pooling size (pooling). S43: Perform dataset partitioning; divide the dataset into training and test sets; S42: The loss function uses the result calculated in S3; the Adam optimizer is selected. S43: In each iteration, the training set is traversed, the loss is calculated, and the parameters are updated; based on the judgment loss function... Alternatively, a preset total number of iterations can be used to determine whether to continue the loop. Take a constant value, and stop the loop when the above conditions are met.

4. The watershed flood simulation and optimization method based on physical information neural network according to claim 3, characterized in that, S5 includes the following steps: S5: To evaluate the model's predictive performance, two metrics are used: S51: The first metric is the root mean square error between the true depth and the predicted depth. S52: Use MRMSE to make a more realistic estimate of the true error: in, Refers to the set of pixels. This refers to the set of all pixels whose actual pixel fluid depth and predicted pixel fluid depth are not zero. and In Measured and predicted values ​​of fluid depth within a pixel set.

5. An electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, it implements a watershed flood simulation optimization method based on a physical information neural network as described in any one of claims 1 to 4.

6. A computer-readable storage medium storing computer instructions thereon, characterized in that: When executed by the processor, the computer instructions implement a watershed flood simulation optimization method based on a physical information neural network as described in any one of claims 1-4.

Citation Information

Patent Citations

  • Flood forecasting method based on SCM-PINNs

    CN116975513A