A flood inundation extent prediction method and apparatus

Abstract

This invention discloses a method and apparatus for predicting flood inundation range. The method includes: firstly, acquiring first and second data from different types of a first watershed, and then aligning and fusing each first data based on spatiotemporal characteristics to extract corresponding feature vector data; then, training multiple graph neural network models to be trained based on each feature vector data to obtain multiple trained prediction models; next, determining third and fourth data in the target watershed; then, determining the optimal prediction model based on the fourth data of the target watershed and the second data corresponding to each prediction model; finally, inputting the third data into the optimal prediction model to obtain the prediction result. This method mainly uses graph neural network models, determines multiple prediction models, and selects the optimal prediction model to predict the target watershed, which can avoid excessive reliance on manual methods, reduce the time for predicting flood inundation range, and improve the versatility of the prediction.

Description

Technical Field

[0001] This invention belongs to the field of water conservancy technology, specifically relating to a method and device for predicting the extent of flood inundation. Background Technology

[0002] With extreme rainfall events becoming more frequent, the frequency and intensity of floods are also on the rise. Therefore, the ability to quickly and accurately predict the extent and severity of flood inundation is of paramount importance for flood disaster early warning, emergency response, and loss reduction.

[0003] Current methods for predicting or simulating flood inundation extent mainly rely on two-dimensional hydrodynamic models based on shallow water equations, such as HEC-RAS, MIKE 21, and IFMS. Existing schemes can simulate the evolution of floods relatively accurately by mathematically describing the physical processes of water flow. However, existing schemes have the following limitations:

[0004] 1. Model building and parameter setting are complex, requiring a variety of input parameters such as accurate high-resolution digital elevation models, river cross sections, low-bed roughness, and boundary conditions, which rely heavily on human experience.

[0005] 2. Long calculation time: Due to the complexity of the model and the large amount of computation, it takes a long time to obtain the final result, which affects the execution of subsequent emergency response.

[0006] 3. Poor versatility; cannot be easily deployed to regions with scarce data or new areas.

[0007] Therefore, how to reduce reliance on manual labor and quickly and accurately predict the extent of flood inundation, thereby improving the versatility of the prediction, is a technical problem that needs to be solved by those skilled in the art. Summary of the Invention

[0008] The purpose of this invention is to solve the technical problems of existing technologies, such as reliance on manual labor, long prediction time for flood inundation range, and low versatility in flood inundation range prediction.

[0009] To achieve the above-mentioned technical objectives, in one aspect, the present invention provides a method for predicting the extent of flood inundation, the method comprising:

[0010] First and second data from different types of first watersheds are obtained, and the first data are aligned and fused based on spatiotemporal features to extract the corresponding feature vector data.

[0011] Multiple training graph neural network models are obtained by training multiple training prediction models based on each feature vector data.

[0012] The third and fourth data points in the target watershed were identified;

[0013] The optimal prediction model is determined based on the fourth data of the target watershed and the second data corresponding to each prediction model.

[0014] The third data is input into the optimal prediction model to obtain the prediction result.

[0015] Furthermore, the first data specifically includes hydrological data, high-resolution digital elevation model, slope map, aerial image, and historical flood patch data corresponding to the first watershed; the second data specifically includes topographic and hydrological structure data corresponding to the first watershed; the third data specifically includes hydrological data, high-resolution digital elevation model, slope map, and aerial image of the target watershed; and the fourth data specifically includes topographic and hydrological structure data of the target watershed.

[0016] Furthermore, the determination of the optimal prediction model based on the fourth data of the target watershed and the second data corresponding to each prediction model specifically includes:

[0017] Based on the fourth data and each of the second data, the cosine similarity between the target watershed and each of the first watersheds is determined;

[0018] The prediction model corresponding to the highest cosine similarity is taken as the best prediction model.

[0019] Furthermore, the temporal feature in the spatiotemporal features specifically refers to an hourly sliding window, and the spatial feature in the spatiotemporal features specifically refers to a 30m resolution. The step of aligning and fusing each set of first data based on the spatiotemporal features to extract the corresponding feature vector data specifically includes:

[0020] The data at different spatial resolutions in the first dataset are synchronized using spatial sampling and interpolation methods.

[0021] Construct nodes with a grid cell as the graph structure, where the grid cell is established based on spatial features;

[0022] Based on the hourly sliding window, the synchronized data are time-aligned and mapped to the raster cells.

[0023] Furthermore, the first constraint function of the graph neural network model to be trained during training is specifically shown in the following formula:

[0024] ,

[0025] In the formula, For the node water depth, For rainfall, Evaporation amount The inflow of the adjacent edge. This represents the outflow of the adjacent edge.

[0026] Furthermore, the second constraint function of the graph neural network model to be trained during training is specifically shown in the following equation:

[0027] ,

[0028] In the formula, Let be the flow velocity of the graph edge. The roughness coefficient, For hydraulic radius, This represents the slope of the edge of the graph.

[0029] Furthermore, the edge weights of the graph neural network to be trained are determined by the following formula:

[0030] ,

[0031] In the formula, Let be the edge weight between node i and node j at time t. For the sigmoid function, As the first learning parameter, The water depth at the current node. The water depth of the nearest node, As the second learning parameter, Let be the slope of the edge of the graph. As the third learning parameter, The rainfall intensity at the current node. This is the fourth learning parameter. This refers to land use type.

[0032] Furthermore, during the training process of the graph neural network model to be trained, if the water depth difference or slope on any side is lower than a preset threshold, the corresponding side is considered invalid at the current time step. In each propagation time step, the edge weight is recalculated based on the currently predicted water depth and rainfall intensity.

[0033] On the other hand, the present invention also provides a flood inundation range prediction device, the device comprising:

[0034] The acquisition module is used to acquire first and second data from different types of first watersheds, and extract the corresponding feature vector data after aligning and fusing each first data based on spatiotemporal features.

[0035] The training module is used to train multiple graph neural network models to be trained based on each feature vector data to obtain multiple trained prediction models.

[0036] The first determining module is used to determine the third and fourth data in the target watershed;

[0037] The second determination module is used to determine the best prediction model based on the fourth data of the target watershed and the second data corresponding to each prediction model.

[0038] The results module is used to input the third data into the best prediction model to obtain the prediction results.

[0039] This invention provides a method and apparatus for predicting flood inundation range. Compared with existing technologies, this method first acquires first and second data from different types of first watersheds, and then aligns and fuses each first data based on spatiotemporal characteristics to extract corresponding feature vector data. Next, it trains multiple graph neural network models based on these feature vector data to obtain multiple trained prediction models. Then, it determines the third and fourth data in the target watershed. Finally, it determines the optimal prediction model based on the fourth data of the target watershed and the second data corresponding to each prediction model. Finally, it inputs the third data into the optimal prediction model to obtain the prediction result. This method primarily uses graph neural network models, determines multiple prediction models, and selects the optimal prediction model to predict the target watershed. This avoids excessive reliance on manual methods, reduces the time required for flood inundation range prediction, and improves the versatility of the prediction. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in the embodiments or prior art of this specification, the drawings used in the description of the embodiments or prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 The diagram shown is a flowchart illustrating the flood inundation range prediction method provided in the embodiments of this specification.

[0042] Figure 2 The diagram shown is a structural schematic of the flood inundation range prediction device provided in the embodiments of this specification;

[0043] Figure 3 The diagram shown is a schematic representation of the process for fusing the first data in an embodiment of this specification. Detailed Implementation

[0044] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0045] like Figure 1 The diagram illustrates a flow chart of the flood inundation range prediction method provided in the embodiments of this specification. Although this specification provides the method operation steps or device structure shown in the following embodiments or figures, based on convention or without creative effort, the method or device may include more or fewer operation steps or module units after partial merging. In steps or structures where there is no logically necessary causal relationship, the execution order of these steps or the module structure of the device are not limited to the execution order or module structure shown in the embodiments or figures of this specification. When the method or module structure is applied in actual devices, servers, or terminal products, it can be executed sequentially or in parallel according to the method or module structure shown in the embodiments or figures (e.g., in a parallel processor or multi-threaded processing environment, or even in a distributed processing or server cluster implementation environment).

[0046] The flood inundation range prediction method provided in the embodiments of this specification can be applied to terminal devices such as clients and servers, and is applicable to scenarios such as urban flood control, reservoir dam failure early warning, and flash flood disaster response. Figure 1 As shown, the method specifically includes the following steps:

[0047] Step S101: Obtain first data and second data from different types of first watersheds, and extract the corresponding feature vector data by aligning and fusing each first data according to spatiotemporal features.

[0048] Specifically, the first data includes hydrological data, high-resolution digital elevation model, slope map, aerial image, and historical flood patch data corresponding to the first watershed; the second data includes topographic and hydrological structure data corresponding to the first watershed; the third data includes hydrological data, high-resolution digital elevation model, slope map, and aerial image of the target watershed; and the fourth data includes topographic and hydrological structure data of the target watershed.

[0049] Furthermore, the monitoring station sensor data acquisition involves: rationally deploying monitoring stations for water level and rainfall within the watershed, and using sensors to collect basic hydrological data such as river water level and rainfall in real time; DEM and slope maps, as well as UAV aerial imagery; GEE historical flood patch data acquisition: leveraging the remote sensing big data advantages of the Google Earth Engine (GEE) platform, calling its historical flood patch extraction function, filtering images from the flood occurrence period, and extracting patch data of the historical flood inundation range as historical sample data for model training; spatiotemporal data fusion and feature encoding: using spatial sampling and interpolation methods to unify data of different spatial resolutions, constructing nodes with raster units as the graph structure; aligning various types of data to the same time step according to a fixed 1-hour sliding window; using Transformer to encode spatial features and stitch together standardized temporal hydrological features to form composite feature vectors for each node at each time step, such as... Figure 3 The diagram shows the process of fusing the first data, i.e., multi-source data. First, multi-source data is collected, including but not limited to GEE historical flood patches, radar rainfall, water level gauges and DEM; then, resampling and interpolation space unification; next, sliding space unification; then Transformer feature enhancement and module fusion; finally, spatiotemporal features of each node are fused.

[0050] Specifically, taking a flood simulation scenario in a certain watershed as an example, the multi-source data available includes rainfall grid data with a spatial resolution of 1km and recorded every 10 minutes, water level monitoring station data recorded every 5 minutes, remote sensing image flood patch data with a spatial resolution of 30m and acquired every hour, and DEM data with a resolution of 15m.

[0051] The time alignment method for rainfall grid data is to calculate the cumulative rainfall within 1 hour, and the spatial alignment method is to refine the 1km resolution rainfall data to 30m resolution through bilinear interpolation. The time alignment method for water level data is to take the average water level within 1 hour as the representative water level for that window, and the spatial alignment method is to interpolate the point water level monitoring data to a 30m grid using an inverse distance weighting method. The time alignment method for remote sensing imagery is to directly assign the window if the time exists, and to use linear interpolation to fill in the missing data using imagery data from the preceding and following times. The 15km resolution DEM is converted to 30km resolution through aggregation.

[0052] The aligned data is uniformly mapped into 30m×30m grid cells. Each grid cell serves as a node in the graph neural network, and its corresponding feature vector includes cumulative rainfall, average water level, elevation, and remote sensing inundation marker.

[0053] Step S102: Train multiple graph neural network models to be trained based on each feature vector data to obtain multiple trained prediction models.

[0054] Specifically, the prediction model structure includes a model input layer, a model graph convolutional layer, a physical constraint module, and a model output layer. Node features include water depth, rainfall intensity, and roughness; edge features include slope and water depth difference between adjacent nodes. The physical constraint module introduces water conservation constraints and Manning's formula constraints during training to ensure that the prediction results conform to hydrodynamic laws. The model output layer is used to output the node water depth and edge flow velocity at the next moment, and thereby derive the inundation range.

[0055] The specific training process for the prediction model is as follows:

[0056] Step 1: Prepare training data including DEM, rainfall time series, water level monitoring data, and land use data, and map them into a uniform 30m×30m grid cell to construct a graph structure input;

[0057] Step 2: If historical map data exists in the target area, the measured flood map data is used as a monitoring signal; if the target area lacks measured map data, pseudo-labels are generated using the source domain model as monitoring signals.

[0058] Step 3: Perform forward propagation; the node module predicts the node water depth, and the edge module predicts the flow velocity.

[0059] Step 4: Construct the loss function, including supervised loss, physical constraints and consistency loss, and optimize the parameters α, β, γ, δ and network weights using backpropagation;

[0060] Step 5: Iterate the training until convergence to obtain the prediction model for the first watershed matching of the corresponding type.

[0061] The specific applications of the prediction model are shown below:

[0062] Step 1: Input DEM, rainfall forecast, and water level monitoring data;

[0063] Step 2: Align the data spatiotemporally and map it into a graph structure;

[0064] Step 3: Use the trained model of the target region to predict the water depth at nodes and the flow velocity at each time step;

[0065] Step 4: When the water depth of the current node exceeds the threshold of 0.05m, it is determined that the area is flooded, and the time-series flooding range map is output.

[0066] In this embodiment of the application, the first constraint function of the graph neural network model to be trained during training is specifically shown in the following formula:

[0067] ,

[0068] In the formula, For the node water depth, For rainfall, Evaporation amount The inflow of the adjacent edge. This represents the outflow of the adjacent edge.

[0069] The second constraint function of the graph neural network model to be trained during training is shown in the following formula:

[0070] ,

[0071] In the formula, Let be the flow velocity of the graph edge. The roughness coefficient, For hydraulic radius, This represents the slope of the edge of the graph.

[0072] The edge weights of the neural network to be trained are determined by the following formula:

[0073] ,

[0074] In the formula, Let be the edge weight between node i and node j at time t. For the sigmoid function, As the first learning parameter, The water depth at the current node. The water depth of the nearest node, As the second learning parameter, Let be the slope of the edge of the graph. As the third learning parameter, The rainfall intensity at the current node. This is the fourth learning parameter. This refers to land use type.

[0075] During the training process of the graph neural network model to be trained, if the water depth difference or slope on any side is lower than a preset threshold, the corresponding side is considered invalid at the current time step. In each propagation time step, the edge weight is recalculated based on the currently predicted water depth and rainfall.

[0076] Specifically, the graph neural network model to be trained proposed in this application introduces physical mechanisms on the traditional graph neural network architecture to enhance the model's ability to simulate hydrodynamics, enabling the prediction and physical interpretability of water depth, flow velocity, and inundation range. Specifically:

[0077] Dual-channel prediction structure: Node channels predict the water depth of each graph node at future time steps. Used to derive water depth and flooding extent; edge channels predict the flow velocity of each graph edge. It is used to reconstruct the water flow propagation path and velocity field. The water depth change of the node is bidirectionally coupled with the flow velocity output of the adjacent edge. The network structure embeds a "node-edge-node" cyclic propagation path, which allows water depth information to regulate flow velocity updates.

[0078] Water balance constraint mechanism: During the model training phase, a water continuity equation is introduced as a loss function constraint to ensure that the prediction process satisfies the local conservation law. The formula is as follows: ,in For the node water depth, For rainfall, For evaporation, The inflow of the adjacent edge. This represents the outflow of the adjacent edge.

[0079] Slope and flow velocity relationship constraint mechanism: Slope of map edges calculated using DEM. By setting the Manning formula as the loss term, the flow velocity conforming to terrain constraints is calculated using the following formula: ,in, Let be the flow velocity of the graph edge. The roughness coefficient, For hydraulic radius, This represents the slope of the edge of the graph.

[0080] Dynamic inundation range derivation: Setting inundation thresholds based on nodal water depth predictions. The value is set to 0.05m, and the values ​​are binarized to obtain the raster flooded area, i.e., values ​​greater than 0.05 are assigned 1, and others are assigned 0. By predicting the water depth sequence at different time steps, the dynamic evolution map of the flooding process can be reconstructed.

[0081] Furthermore, traditional graph neural networks often employ graphs with fixed edge weights, which does not accurately reflect the dynamic changes in water flow paths due to rainfall, topography, and water depth during flood propagation. This module improves the ability to characterize complex terrain structures and dynamic water flow paths in flood simulation by introducing a variable edge weight modeling mechanism that combines multi-scale spatial representation with dynamic physical information. Specifically:

[0082] Multi-scale map structure construction: fine-scale maps use 15m resolution grids to describe local slope changes and minor topographic waterlogging phenomena; mesoscale maps depict the main water catchment paths within a region at the sub-basin level; and coarse-scale maps provide a comprehensive overview of water flow evolution trends for the larger watershed structure. Nodes in each scale map represent a spatial unit, while edges represent water flow channels between adjacent units.

[0083] A variable edge weight mechanism based on dynamic physical information: The dynamic calculation formula for edge weights is as follows: ,in Let be the edge weight between node i and node j at time t; The direction of water flow is represented by the difference in water depth between the current node and its neighboring nodes; The slope between nodes, i.e. the slope of the graph edges, is extracted from the DEM. The rainfall intensity at the current node; Land use type; For the sigmoid function, As the first learning parameter, As the second learning parameter, As the third learning parameter, This is the fourth learning parameter.

[0084] The graph structure dynamic update mechanism is as follows: when the water depth difference or slope on a certain edge is lower than the threshold, the edge is considered invalid in the current time step, and the weight of the edge participating in the propagation approaches 0; in each propagation time step, the edge weight is recalculated based on the current predicted water depth and rainfall intensity.

[0085] The prediction model provided in this application offers flexible input, integrating multi-source information such as remote sensing data and sensor data to improve data adaptability. Secondly, the introduction of a physical constraint mechanism enhances the model's dynamic evolution capability and physical interpretability. It also supports dynamic graph structure modeling, capturing the spatiotemporal variation characteristics of flood propagation paths and achieving a higher fitting capability to the real hydrodynamic propagation process. The output of the prediction model includes water depth, extent, and flow velocity, providing refined decision support for scientific prevention and control.

[0086] Step S103: Determine the third and fourth data in the target watershed.

[0087] Specifically, when predicting the flood inundation range of a target watershed, it is necessary to obtain the third and fourth data of the target watershed. As mentioned above, the third data specifically includes the hydrological data, high-resolution digital elevation model, slope map and aerial image of the target watershed, and the fourth data specifically includes the topographic and hydrological structure data of the target watershed.

[0088] Step S104: Determine the optimal prediction model based on the fourth data of the target watershed and the second data corresponding to each prediction model.

[0089] Specifically, after determining the third and fourth data of the target watershed, since the topography and hydrology data of each watershed are not the same, it is necessary to determine the best prediction model among multiple prediction models based on the fourth data of the target watershed, that is, the topography and hydrology structure data of the target watershed, so as to use the best prediction model to predict the flood inundation range of the target watershed.

[0090] In this embodiment of the application, the determination of the optimal prediction model based on the fourth data of the target watershed and the second data corresponding to each prediction model specifically includes:

[0091] Based on the fourth data and each of the second data, the cosine similarity between the target watershed and each of the first watersheds is determined;

[0092] The prediction model corresponding to the highest cosine similarity is taken as the best prediction model.

[0093] Specifically, the similarity of topographic and hydrological structures between the target area and the trained source domain is determined, cosine similarity is calculated, and the closest model parameters are selected from the training library based on similarity.

[0094] Furthermore, the calculation process for cosine similarity can be as follows:

[0095] ① Calculate the average slope S: The average slope of each grid cell within the area is calculated using the DEM and is 0.12.

[0096] ② Calculate the elevation distribution (Ele): The standard deviation of the elevation within the calculation area is 50m;

[0097] ③ Calculate the multi-year average rainfall intensity: Statistically analyze historical rainfall data, and take the annual maximum hourly rainfall intensity as 75 mm / h;

[0098] ④ Calculate the annual runoff coefficient: estimated to be 0.55 using an empirical formula;

[0099] ⑤ Calculate the Manning roughness: The overall roughness is 0.035;

[0100] After normalization using the Min-Max normalization method, the eigenvector is VA = (0.35, 0.30, 0.64, 0.50, 0.375). The source domain eigenvector is calculated using the same method, and then the cosine similarity is calculated using the formula: The cosine similarity value between the target region and each source region is calculated, and the larger the result, the stronger the similarity.

[0101] It should be noted that the above cosine similarity calculation process is only one specific implementation process adopted by this scheme. Those skilled in the art can flexibly set other cosine similarity calculation schemes according to the actual situation.

[0102] In this embodiment of the application, the solution further includes enabling rapid transfer and adaptation of the prediction model through an unsupervised pseudo-label training mechanism when real flood patches are lacking in the target area. This involves using the source domain model to generate pseudo patches (pseudo-labels) based on the topographic and hydrological features of the target area, and using these pseudo-labels as supervisory signals. The specific process is as follows:

[0103] (1) Based on the trained source domain model, input DEM, rainfall sequence and other data into the target area to automatically generate the predicted flood range and use it as a pseudo label for the target area;

[0104] (2) Using pseudo-labels as supervision signals, the prediction model for the target area determined in the above steps is trained. During the training process, loss functions with consistency constraints and water balance physical constraints are introduced to ensure the spatial continuity and stability of the prediction results at different time steps, avoid model oscillations caused by pseudo-label bias, and ensure that the prediction results meet physical laws such as water conservation, correct the errors in pseudo-labels, and make the prediction results conform to hydrodynamic laws.

[0105] (3) The backpropagation algorithm is used to continuously update the model parameters, and the output of the target model is used to replace the pseudo-patterns in multiple iterations to continue training.

[0106] To avoid model oscillations caused by label errors, a temporal consistency constraint formula is introduced as follows:

[0107] ,

[0108] in, Indicates the model at time step Predicted water depth distribution, This represents the time evolution operator based on the hydrodynamic continuity equation.

[0109] To ensure that the prediction results conform to hydrodynamics, physical constraints based on water balance are introduced:

[0110] ,

[0111] in, For time step Time node The predicted water depth For the corresponding grid area, This refers to the amount of rainfall during a given period. For outgoing traffic.

[0112] Simultaneously, the Manning formula loss is introduced to establish an empirical relationship between the predicted flow velocity and slope, hydraulic radius, and roughness for each edge, thereby ensuring the physical rationality of the flow velocity and enhancing the physical interpretability of the model. The formula is as follows:

[0113] ,

[0114] in, The flow rate predicted by the model, For hydraulic radius, For slope, For roughness, These are optional weights.

[0115] In pseudo-label training, the loss function can be written in the form of a total loss function:

[0116] ,

[0117] in, The loss for pseudo-label supervision is the mean squared error. For time-series consistency loss, For water balance loss, For Manning's constraint loss, This is the loss weight hyperparameter.

[0118] During training, the predicted output of the target region model will replace the initial pseudo-label, and the model will be adaptively corrected through multiple rounds of iterative training.

[0119] This mechanism enables rapid migration and adaptation of flood inundation prediction models under unlabeled or poorly labeled conditions.

[0120] Step S105: Input the third data into the optimal prediction model to obtain the prediction result.

[0121] Specifically, after determining the optimal prediction model through the above steps, the third data of the target watershed is input into the optimal prediction model to obtain the prediction results output by the optimal prediction model.

[0122] The above scheme uses an improved graph neural network model as the main prediction model. Compared with existing hydrodynamic models such as HEC-RAS, MIKE 21, and IFMS, it reduces the complexity of model building and parameter setting, does not rely too much on manual labor, and requires less computation than existing hydrodynamic models. It can output prediction results in a timely and effective manner, enabling disaster management personnel to make management plans more quickly. At the same time, matching the most suitable prediction model according to the hydrological and topographic features of the target watershed can improve the versatility of the prediction.

[0123] Based on the above-described method for predicting flood inundation range, one or more embodiments of this specification also provide a platform or terminal for predicting flood inundation range. This platform or terminal may include devices, software, modules, plug-ins, servers, clients, etc., using the methods described in the embodiments of this specification, combined with necessary hardware implementation. Based on the same innovative concept, the systems in one or more embodiments provided in this specification are as described in the following embodiments. Since the implementation schemes and methods for solving the system problem are similar, the specific system implementation in the embodiments of this specification can refer to the implementation of the aforementioned methods. Repeated descriptions will not be repeated. The terms "unit" or "module" used below can refer to a combination of software and / or hardware that performs a predetermined function. Although the systems described in the following embodiments are preferably implemented in software, hardware implementation, and a combination of software and hardware, are also possible and contemplated.

[0124] Specifically, Figure 2 This is a schematic diagram of the module structure of one embodiment of the flood inundation range prediction device provided in this specification, as shown below. Figure 2 As shown, the flood inundation range prediction device provided in this specification includes:

[0125] The acquisition module 201 is used to acquire first data and second data in the first watershed of different types, and extract the corresponding feature vector data after aligning and fusing each first data according to spatiotemporal features.

[0126] Training module 202 is used to train multiple graph neural network models to be trained based on each feature vector data to obtain multiple trained prediction models;

[0127] The first determining module 203 is used to determine the third and fourth data in the target watershed;

[0128] The second determining module 204 is used to determine the best prediction model based on the fourth data of the target watershed and the second data corresponding to each prediction model.

[0129] The result module 205 is used to input the third data into the optimal prediction model to obtain the prediction result.

[0130] It should be noted that the system described above may include other implementation methods based on the description of the corresponding method embodiments. The specific implementation methods can be referred to the description of the corresponding method embodiments above, and will not be elaborated here.

[0131] This application also provides an electronic device, including:

[0132] processor;

[0133] Memory used to store the processor's executable instructions;

[0134] The processor is configured to perform the methods provided in the embodiments described above.

[0135] The electronic device provided in this application stores executable instructions of the processor in a memory. When the processor executes the executable instructions, it can first acquire first data and second data from different types of first watersheds, and then extract corresponding feature vector data by aligning and fusing each first data based on spatiotemporal features. Then, it trains multiple graph neural network models to be trained based on each feature vector data to obtain multiple trained prediction models. Next, it determines the third data and fourth data in the target watershed. Then, it determines the best prediction model based on the fourth data of the target watershed and the second data corresponding to each prediction model. Finally, it inputs the third data into the best prediction model to obtain the prediction result. This method mainly uses graph neural network models, determines multiple prediction models and selects the best prediction model to predict the target watershed, which can avoid excessive reliance on manual labor, reduce the time for predicting the flood inundation range, and improve the versatility of the prediction.

[0136] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0137] The methods or apparatus described in the embodiments provided in this specification can implement business logic through a computer program and record it on a storage medium. The storage medium can be read and executed by a computer to achieve the effects of the solutions described in the embodiments of this specification, such as:

[0138] First and second data from different types of first watersheds are obtained, and the first data are aligned and fused based on spatiotemporal features to extract the corresponding feature vector data.

[0139] Multiple training graph neural network models are obtained by training multiple training prediction models based on each feature vector data.

[0140] The third and fourth data points in the target watershed were identified;

[0141] The optimal prediction model is determined based on the fourth data of the target watershed and the second data corresponding to each prediction model.

[0142] The third data is input into the optimal prediction model to obtain the prediction result.

[0143] The storage medium can include physical devices for storing information, typically digitizing the information and then storing it using electrical, magnetic, or optical methods. The storage medium can include: devices that store information using electrical energy, such as various types of memory, like RAM and ROM; devices that store information using magnetic energy, such as hard disks, floppy disks, magnetic tapes, magnetic core memory, bubble memory, and USB flash drives; and devices that store information using optical methods, such as CDs or DVDs. Of course, there are other readable storage media, such as quantum memories and graphene memories.

[0144] The embodiments in this specification are not limited to conforming to industry communication standards, standard computer resource data update and data storage rules, or the situations described in one or more embodiments of this specification. Slightly modified implementations based on certain industry standards or custom methods or embodiments can also achieve the same, equivalent, or similar, or predictable, implementation effects as described above. Embodiments that utilize these modified or modified methods for data acquisition, storage, judgment, and processing still fall within the scope of optional implementations of the embodiments in this specification.

[0145] The controller can be implemented in any suitable manner. For example, it can take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro)processor, logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers, and embedded microcontrollers. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicon Labs C8051F320. A memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art will also recognize that, in addition to implementing the controller in purely computer-readable program code form, the same functionality can be achieved by logically programming the method steps to make the controller take the form of logic gates, switches, ASICs, programmable logic controllers, and embedded microcontrollers. Therefore, such a controller can be considered a hardware component, and the means included therein for implementing various functions can also be considered as structures within the hardware component. Alternatively, the means for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0146] The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or plug-ins may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, indirect coupling or communication connection between devices or units, and may be electrical, mechanical, or other forms.

[0147] These computer program instructions can also be loaded onto a computer or other programmable resource data updating device, causing a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable device 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.

[0148] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, system embodiments are basically similar to method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. In the description of this specification, the terms "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this specification. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described can be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0149] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical teachings disclosed in this invention without departing from the spirit of the invention, and these modifications and combinations are still within the scope of protection of this invention.

Claims

1. A method for predicting the extent of flood inundation, characterized in that, The method includes: First and second data from different types of first watersheds are obtained, and the first data are aligned and fused based on spatiotemporal features to extract the corresponding feature vector data. Multiple training graph neural network models are obtained by training multiple training prediction models based on each feature vector data. The third and fourth data points in the target watershed were identified; The optimal prediction model is determined based on the fourth data of the target watershed and the second data corresponding to each prediction model. The third data is input into the optimal prediction model to obtain the prediction result; Specifically, the first data includes hydrological data, high-resolution digital elevation model, slope map, aerial image, and historical flood patch data corresponding to the first watershed; the second data includes topographic and hydrological structure data corresponding to the first watershed; the third data includes hydrological data, high-resolution digital elevation model, slope map, and aerial image of the target watershed; and the fourth data includes topographic and hydrological structure data of the target watershed. The first constraint function of the graph neural network model to be trained during training is specifically shown in the following formula: ； In the formula, For the node water depth, For rainfall, Evaporation amount The inflow of the adjacent edge. The outflow of adjacent edges; The second constraint function of the graph neural network model to be trained during training is shown in the following formula: ； In the formula, Let be the flow velocity of the graph edge. The roughness coefficient, For hydraulic radius, The slope of the edge of the graph; The edge weights of the neural network to be trained are determined by the following formula: ； In the formula, Let be the edge weight between node i and node j at time t. For the sigmoid function, As the first learning parameter, The water depth at the current node. The water depth of the nearest node, As the second learning parameter, Let be the slope of the edge of the graph. As the third learning parameter, The rainfall intensity at the current node. This is the fourth learning parameter. This refers to land use type.

2. The flood inundation range prediction method as described in claim 1, characterized in that, The optimal prediction model is determined based on the fourth data from the target watershed and the second data corresponding to each prediction model, specifically including: Based on the fourth data and each of the second data, the cosine similarity between the target watershed and each of the first watersheds is determined; The prediction model corresponding to the highest cosine similarity is taken as the best prediction model.

3. The flood inundation range prediction method as described in claim 1, characterized in that, The temporal feature in the spatiotemporal features specifically refers to an hourly sliding window, and the spatial feature in the spatiotemporal features specifically refers to a 30m resolution. The step of extracting the corresponding feature vector data after aligning and fusing each set of first data based on the spatiotemporal features specifically includes: The data at different spatial resolutions in the first dataset are synchronized using spatial sampling and interpolation methods. Construct nodes with a grid cell as the graph structure, where the grid cell is established based on spatial features; Based on the hourly sliding window, the synchronized data are time-aligned and mapped to the raster cells.

4. The flood inundation range prediction method as described in claim 1, characterized in that, During the training process of the graph neural network model to be trained, if the water depth difference or slope on any side is lower than a preset threshold, the corresponding side is considered invalid at the current time step. In each propagation time step, the edge weight is recalculated based on the currently predicted water depth and rainfall intensity.

5. A device for predicting the extent of flood inundation, characterized in that, The device includes: The acquisition module is used to acquire first and second data from different types of first watersheds, and extract the corresponding feature vector data after aligning and fusing each first data based on spatiotemporal features. The training module is used to train multiple graph neural network models to be trained based on each feature vector data to obtain multiple trained prediction models. The first determining module is used to determine the third and fourth data in the target watershed; The second determination module is used to determine the best prediction model based on the fourth data of the target watershed and the second data corresponding to each prediction model. The results module is used to input the third data into the best prediction model to obtain the prediction results; Specifically, the first data includes hydrological data, high-resolution digital elevation model, slope map, aerial image, and historical flood patch data corresponding to the first watershed; the second data includes topographic and hydrological structure data corresponding to the first watershed; the third data includes hydrological data, high-resolution digital elevation model, slope map, and aerial image of the target watershed; and the fourth data includes topographic and hydrological structure data of the target watershed. The first constraint function of the graph neural network model to be trained during training is specifically shown in the following formula: ； In the formula, For the node water depth, For rainfall, Evaporation amount The inflow of the adjacent edge. The outflow of adjacent edges; The second constraint function of the graph neural network model to be trained during training is shown in the following formula: ； In the formula, Let be the flow velocity of the graph edge. The roughness coefficient, For hydraulic radius, The slope of the edge of the graph; The edge weights of the neural network to be trained are determined by the following formula: ； In the formula, Let be the edge weight between node i and node j at time t. For the sigmoid function, As the first learning parameter, The water depth at the current node. The water depth of the nearest node, As the second learning parameter, Let be the slope of the edge of the graph. As the third learning parameter, The rainfall intensity at the current node. This is the fourth learning parameter. This refers to land use type.

Images