A basin flood dynamic simulation system based on multi-source data fusion

Abstract

This invention relates to the field of flood disaster simulation technology, and in particular to a watershed flood dynamic simulation system based on multi-source data fusion. The system includes a multi-source data preprocessing module, a multi-source data fusion module, a digital twin watershed modeling module, a flood dynamic simulation module, a flood risk assessment module, and a visualization module. This solution proposes a spatiotemporal cross-modal interactive fusion method to fuse multi-source data. This method introduces a cross-modal interaction mechanism to capture intramodal dependencies and uses low-rank decomposition to efficiently fuse high-dimensional interaction tensors, taking into account the complementarity of spatial topology, temporal dynamics, and multi-source heterogeneous data, effectively improving the accuracy and robustness of hydrological process modeling. By constructing a distributed hydrological-hydrodynamic coupled model, it achieves dynamic simulation of the entire process from runoff generation to flood evolution in the river channel and floodplain.

Description

Technical Field

[0001] This invention relates to the field of flood disaster simulation technology, and in particular to a watershed flood dynamic simulation system based on multi-source data fusion, which is used to achieve high-precision, real-time watershed flood process simulation and risk assessment. Background Technology

[0002] With the acceleration of global climate change and urbanization, the frequent occurrence of extreme rainfall events has led to increasingly severe watershed flood disasters. Current watershed flood simulation systems generally suffer from problems such as single data sources, lagging updates, and static models. Existing multi-source data exhibit significant heterogeneity in terms of spatiotemporal resolution, accuracy, and semantic consistency, making it difficult to effectively integrate them using traditional interpolation or statistical fusion methods. Furthermore, existing flood simulation systems in small and medium-sized watershed flood simulations suffer from problems such as the disconnect between runoff generation and flow processes and the dynamics of flood evolution, and the difficulty in quantifying the hydrological effects of natural intervention measures. Therefore, there is an urgent need to construct a watershed flood simulation system that can integrate multi-source observation and forecast data, possess adaptive correction capabilities, and support dynamic simulation and visualization. Summary of the Invention

[0003] To address the above issues and overcome the shortcomings of existing technologies, this invention provides a watershed flood dynamic simulation system based on multi-source data fusion. Addressing the problems of single data sources, delayed updates, and static models in current watershed flood simulation systems, this solution employs multi-source data acquisition to monitor the watershed and integrates multi-source data. Recognizing the significant heterogeneity in spatiotemporal resolution, accuracy, and semantic consistency of multi-source data, which traditional interpolation or statistical fusion methods struggle to effectively integrate, this solution proposes a spatiotemporal cross-modal interactive fusion method to fuse multi-source data. This method constructs a graph node and spatial adjacency matrix based on the watershed topology. By using direction-aware feature decomposition, multimodal features in the horizontal and vertical directions are obtained separately. A cross-modal interaction mechanism is introduced to capture intramodal dependencies, and high-dimensional interaction tensors are efficiently fused using low-rank decomposition. The overall method takes into account the complementarity of spatial topology, temporal dynamics, and multi-source heterogeneous data, effectively improving the accuracy and robustness of hydrological process modeling. In view of the problems of existing flood simulation systems in small and medium-sized watershed flood simulation, such as the disconnect between runoff generation and flood evolution dynamics and the difficulty in quantifying the hydrological effects of natural intervention measures, this scheme constructs a distributed hydrological-hydrodynamic coupling model to realize the dynamic simulation of the entire process from runoff generation to flood evolution in the river channel and floodplain.

[0004] The present invention provides a watershed flood dynamic simulation system based on multi-source data fusion. The system includes a multi-source data preprocessing module, a multi-source data fusion module, a digital twin watershed modeling module, a flood dynamic simulation module, a flood risk assessment module, and a visualization module.

[0005] The multi-source data preprocessing module is used to access data from ground rain gauges, soil moisture sensors, weather radars, satellite remote sensing, and numerical weather prediction models, and to perform spatiotemporal alignment, quality control, outlier removal, and format standardization on the accessed data to obtain multi-source data.

[0006] The multi-source data fusion module uses a spatiotemporal cross-modal interactive fusion method to fuse multi-source data and obtain a fused feature vector;

[0007] The digital twin watershed modeling module constructs a spatial distribution map and a BIM model of engineering facilities based on DEM topography, river network structure, and NFM model (neural factor decomposition machine), generating a three-dimensional visualization scene of the watershed.

[0008] The flood dynamics simulation module constructs a distributed hydrological-hydraulic coupling model, and simulates the water flow changes in the basin based on fused feature vectors and a three-dimensional visualization scene of the basin to obtain simulation results.

[0009] The flood risk assessment module assesses flood risk based on the simulation results of the flood dynamics simulation module and obtains the risk level.

[0010] The visualization module provides an API interface to support integration with emergency management, water conservancy scheduling, and urban flooding early warning systems. It combines population and infrastructure disaster-bearing data to generate dynamic risk heat maps, which are then visualized and used for early warning pushes through a Web GIS platform.

[0011] Furthermore, the multi-source data fusion module uses a spatiotemporal cross-modal interactive fusion method to fuse multi-source data, specifically including the following steps:

[0012] Step A1: Remote sensing data feature extraction. Use ResNet101 to extract deep semantic features of remote sensing precipitation images from multi-source data to obtain remote sensing features, and obtain meteorological features from meteorological radar and numerical weather prediction models in multi-source data.

[0013] Step A2: Ground data fusion. Ground hydrological sensor data is obtained from multi-source data. Bayesian dynamic bias correction method is used to correct the ground hydrological sensor data. Kalman filtering is combined to synergistically assimilate remote sensing precipitation monitoring and ground observation to obtain hydrological characteristics.

[0014] Step A3: Watershed node definition and spatial adjacency matrix construction. Define all monitoring points in the watershed as graph nodes, and calculate the spatial weights between nodes based on the actual watershed hydrological topology to obtain the spatial adjacency matrix. Construct a multimodal feature vector for each node, which includes remote sensing features, hydrological features and meteorological features.

[0015] Step A4: Feature decomposition. The direction-aware feature decomposition method is used to decompose the multimodal feature vector along the spatial dimension to obtain the horizontal and vertical features of the multimodal features.

[0016] Step A5: Feature fusion. The horizontal and vertical features of the multimodal features of each node are concatenated to obtain the horizontal fused feature and the vertical fused feature.

[0017] Step A6: Spatiotemporal graph construction. Combine the spatial adjacency matrix, horizontal features, and vertical features to construct a spatiotemporal graph, generating a temporal feature graph and a spatial feature graph.

[0018] Step A7: Cross-modal interactive fusion. Perform cross-modal interaction and low-rank multimodal fusion on the temporal feature map and spatial feature map to obtain the fused feature vector.

[0019] Furthermore, the flood dynamics simulation module constructs a distributed hydrological-hydrodynamic coupled model, specifically including the following steps:

[0020] Step B1: Distributed runoff generation and confluence simulation. A distributed watershed hydrological model is used as the upstream runoff generation and confluence model. The input fused feature vector is processed to simulate the rainfall-runoff process in the upstream watershed and output a runoff map.

[0021] Step B2: Spatial distribution update. Coupled sections are set at the confluence of tributaries and the main channel within the watershed. The runoff map is segmented using a rasterization method. From the segmented runoff map, a representative grid is selected for each confluence point to extract the hourly total runoff sequence.

[0022] Step B3: Initialize the Flood Modeller hydrodynamic domain, perform two-dimensional hydrodynamic simulation using the Flood Modeller hydrodynamic model, obtain the simulation area and spatial distribution map in the three-dimensional visualization scene, and set the initial water depth;

[0023] Step B4: Couple the hydrodynamic model, use the obtained hourly total runoff sequence as the inflow boundary condition of the Flood Modeller hydrodynamic model, and use the two-dimensional Saint-Venant equations of the alternating direction implicit solver to calculate the water depth and flow rate, simulating the entire process from rainfall to inundation.

[0024] Step B5: Parameter adjustment. Based on the spatial distribution map, the real-time spatial distribution of the watershed is dynamically updated, and the parameters of the watershed hydrological model and the Flood Modeller hydrodynamic model are adjusted to optimize the flood simulation process.

[0025] Step B6: Model output, set the maximum number of iterations, obtain historical watershed flood data to construct a calibration dataset, use the calibration dataset to iterate through steps B1 to B5 until the maximum number of iterations is reached, and output the calibrated model as the final distributed hydrological-hydrodynamic coupling model.

[0026] Step B7: Simulation output. Use the final distributed hydrological-hydrodynamic coupling model to simulate the water flow changes in the basin and obtain the simulation results.

[0027] The beneficial effects achieved by adopting the above solution are as follows:

[0028] (1) In view of the problems of single data source, delayed updates and static model in the current watershed flood simulation system, this scheme adopts a multi-source data acquisition method to monitor the watershed and integrates multi-source data for subsequent analysis and processing;

[0029] (2) In view of the significant heterogeneity in spatiotemporal resolution, accuracy and semantic consistency of multi-source data, traditional interpolation or statistical fusion methods are difficult to integrate effectively. This scheme proposes a spatiotemporal cross-modal interactive fusion method to fuse multi-source data. This method constructs graph nodes and spatial adjacency matrix based on watershed topology, dynamically adjusts the importance of different modes in different regions, avoids information loss or noise introduction caused by simple splicing, obtains multimodal features in horizontal and vertical directions through direction-aware feature decomposition, introduces cross-modal interaction mechanism to capture intramodal dependencies, and designs cross-modal diffusion attention to realize spatiotemporal feature interaction. Finally, it uses low-rank decomposition to efficiently fuse high-dimensional interaction tensors. The overall method takes into account the complementarity of spatial topology, temporal dynamics and multi-source heterogeneous data, effectively improving the accuracy and robustness of hydrological process modeling.

[0030] (3) In view of the problems that existing flood simulation systems have in flood simulation of small and medium watersheds, such as the disconnect between runoff generation and flood evolution and the difficulty in quantifying the hydrological effects of natural intervention measures, this scheme constructs a high-resolution distributed hydro-hydrodynamic coupling model. It uses a grid to finely depict the spatial distribution of watershed land cover and NFM measures. Through a one-way coupling mechanism of time synchronization and spatial matching, the generated dynamic gridded runoff is used as the boundary condition to drive the FloodModeller two-dimensional hydrodynamic model, thereby realizing the full process of rainfall runoff generation, river evolution to flood inundation and high spatiotemporal resolution dynamic simulation, providing scientific support for flood risk assessment and optimization of natural flood control measures in small and medium watersheds. Attached Figure Description

[0031] Figure 1 This is a schematic diagram of a watershed flood dynamic simulation system based on multi-source data fusion proposed in this invention;

[0032] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0033] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0034] Example 1, see Figure 1 The present invention provides a watershed flood dynamic simulation system based on multi-source data fusion. The system includes a multi-source data preprocessing module, a multi-source data fusion module, a digital twin watershed modeling module, a flood dynamic simulation module, a flood risk assessment module, and a visualization module.

[0035] The multi-source data preprocessing module is used to access data from ground rain gauges, soil moisture sensors, weather radars, satellite remote sensing, and numerical weather prediction models, and to perform spatiotemporal alignment, quality control, outlier removal, and format standardization on the accessed data to obtain multi-source data.

[0036] The multi-source data fusion module uses a spatiotemporal cross-modal interactive fusion method to fuse multi-source data and obtain a fused feature vector;

[0037] The digital twin watershed modeling module constructs a spatial distribution map and a BIM model of engineering facilities based on DEM topography, river network structure, and NFM model (neural factor decomposition machine), generating a three-dimensional visualization scene of the watershed.

[0038] The flood dynamic simulation module constructs a distributed hydrological-hydraulic coupling model, acquires multi-source data in the basin in real time, and performs dynamic simulation of water flow changes in the basin based on the fused feature vector obtained from the multi-source data and the three-dimensional visualization scene of the basin to obtain simulation results.

[0039] The flood risk assessment module assesses flood risk based on the simulation results of the flood dynamics simulation module and obtains the risk level.

[0040] The visualization module provides an API interface to support integration with emergency management, water conservancy scheduling, and urban flooding early warning systems. It combines population and infrastructure disaster-bearing data to generate dynamic risk heat maps, which are then visualized and used for early warning pushes through a Web GIS platform.

[0041] To address the problems of single data source, delayed updates, and static models in current watershed flood simulation systems, this solution adopts a multi-source data acquisition approach to monitor the watershed and integrates the multi-source data for subsequent analysis and processing.

[0042] Example 2, based on the above examples, describes a multi-source data fusion module that uses a spatiotemporal cross-modal interactive fusion method to fuse multi-source data, specifically including the following steps:

[0043] Step A1: Remote sensing data feature extraction. Use ResNet101 to extract deep semantic features of remote sensing precipitation images from multi-source data to obtain remote sensing features, and obtain meteorological features from meteorological radar and numerical weather prediction models in multi-source data.

[0044] Step A2: Ground data fusion. Ground hydrological sensor data is obtained from multi-source data. Bayesian dynamic bias correction method is used to correct the ground hydrological sensor data. Kalman filtering is combined to synergistically assimilate remote sensing precipitation monitoring and ground observation to obtain hydrological characteristics.

[0045] Step A3: Watershed node definition and spatial adjacency matrix construction. Define all monitoring points in the watershed as graph nodes, and calculate the spatial weights between nodes based on the actual watershed hydrological topology to obtain the spatial adjacency matrix. Construct a multimodal feature vector for each node, which includes remote sensing features, hydrological features and meteorological features.

[0046] Step A4: Feature decomposition. The direction-aware feature decomposition method is used to decompose the multimodal feature vector along the spatial dimension to obtain the horizontal and vertical features of the multimodal features.

[0047] The multimodal horizontal features include remote sensing horizontal features, horizontal hydrological features, and horizontal meteorological features; the multimodal vertical features include vertical remote sensing features, vertical hydrological features, and vertical meteorological features.

[0048] Step A5: Feature fusion. The horizontal and vertical features of the multimodal features of each node are concatenated to obtain the horizontal fused feature and the vertical fused feature.

[0049] Step A6: Spatiotemporal graph construction. Combine the spatial adjacency matrix, horizontal features, and vertical features to construct a spatiotemporal graph, generating a temporal feature graph and a spatial feature graph.

[0050] Step A7: Cross-modal interactive fusion. Perform cross-modal interaction and low-rank multimodal fusion on the temporal feature map and spatial feature map to obtain the fused feature vector.

[0051] Example 3, based on the above examples, includes the following steps in step A7 for cross-modal interaction fusion:

[0052] Step A71: Intramodal self-attention, calculate the self-similarity matrix for the temporal feature map and the spatial feature map respectively, and normalize them to obtain the normalized temporal self-similarity matrix and spatial self-similarity matrix;

[0053] Step A72: Cross-modal diffusion attention is performed, constructing a bidirectional cross-modal similarity matrix. Through value matrix propagation, a cross representation is generated. The original temporal and spatial self-similarity matrices are then concatenated with the cross representation to obtain optimized temporal and spatial feature maps. The formula used for the cross-modal similarity matrix is ​​as follows:

[0054] ;

[0055] In the formula, For hyperparameters, and Let these represent the unnormalized and normalized time self-similarity matrices, respectively. and Let these represent the unnormalized and normalized spatial self-similarity matrices, respectively. For transpose operation, This is a cross-modal similarity matrix;

[0056] Step A73: Construct a high-dimensional interaction tensor, reconstruct the optimized temporal and spatial feature maps into a high-dimensional tensor, and process the high-dimensional tensor using a linear layer. Perform a linear transformation through the weight matrix W to generate a high-dimensional vector representation.

[0057] Step A74: Low-rank decomposition dimensionality reduction. In step A73, the weight matrix W is decomposed into m pairs of low-rank factors. The fusion vector is calculated using the decomposed low-rank factors, and the final fusion feature vector is output. The formula used is as follows:

[0058] ;

[0059] In the formula, This represents the number of low-rank factor pairs obtained by decomposing the weight matrix W. Indexes for low-rank factors. This represents the low-rank weight vector of the spatial feature map. This represents the low-rank weight vector of the time feature map. and As a pair of low-rank factors, Represents the optimized spatial feature map. This represents the optimized time feature map. Represents element-wise product. This is the final fused feature vector.

[0060] By performing the aforementioned operations, this solution addresses the significant heterogeneity in spatiotemporal resolution, accuracy, and semantic consistency of multi-source data, which traditional interpolation or statistical fusion methods struggle to effectively integrate. To address this challenge, a spatiotemporal cross-modal interactive fusion method is proposed. This method constructs graph nodes and spatial adjacency matrices based on the watershed topology, dynamically adjusting the importance of different modes in different regions to avoid information loss or noise introduction caused by simple splicing. Through direction-aware feature decomposition, multimodal features in the horizontal and vertical directions are obtained separately. A cross-modal interaction mechanism is introduced to capture intramodal dependencies, and cross-modal diffusion attention is designed to achieve spatiotemporal feature interaction. Finally, low-rank decomposition is used to efficiently fuse high-dimensional interaction tensors. The overall method considers the complementarity of spatial topology, temporal dynamics, and multi-source heterogeneous data, effectively improving the accuracy and robustness of hydrological process modeling.

[0061] Example 4, based on the above examples, describes a digital twin watershed modeling module that uses a digital elevation model (DEM) as the watershed topographic skeleton. Based on the DEM, the D∞ algorithm is used to automatically extract the watershed flow direction, cumulative discharge, and river network structure through slope, aspect, mountain shadows, and contour lines to restore the watershed topographic features. At the same time, based on hydrological analysis methods (such as depression filling, flow direction, and calculation of cumulative runoff), the river network structure is automatically extracted from the DEM, and topology correction and network connectivity optimization are performed in combination with measured river vector data.

[0062] The NFM model, as an intelligent spatial modeling engine, is used to generate continuous spatial distribution maps of key hydrological variables, including runoff coefficient, infiltration rate, and inundation sensitivity. This method encodes rasterized multi-source variables (such as NDVI vegetation index, slope, impervious surface ratio, and historical flood frequency) into sparse high-dimensional feature vectors. The sparse features are then mapped into low-dimensional dense vectors through an embedding matrix, preserving semantic relationships. The embedded low-dimensional dense vectors are then multiplied element-wise and summed with weights to explicitly model high-order nonlinear interactions between features, and regression is used to generate spatial distribution maps within the watershed.

[0063] For key water conservancy facilities such as dams, pumping stations, and reservoirs, BIM modeling technology with LOD2.0 and above accuracy is used to restore their geometric shape, equipment parameters and operating logic at a 1:1 scale and accurately embed them into the three-dimensional geographic scene. Through the 3D engine, the terrain, river network, and NFM-driven dynamic layers are integrated with the BIM model to construct a digital twin watershed 3D visualization scene that supports real-time rendering, attribute query, section analysis and simulation pre-visualization.

[0064] Example 5, based on the above examples, describes a flood dynamics simulation module that constructs a distributed hydrological-hydrodynamic coupled model, specifically including the following steps:

[0065] Step B1: Distributed runoff generation and confluence simulation. A distributed watershed hydrological model is used as the upstream runoff generation and confluence model. The input fused feature vector is processed to simulate the rainfall-runoff process in the upstream watershed and output a runoff map. The upstream runoff generation and confluence refers to the entire process in the upstream area of ​​the watershed from the occurrence of rainfall to the formation of runoff and its collection at the outlet section. The runoff map is a thematic map or chart used to represent the distribution and changes of surface or subsurface runoff in terms of time, space or quantity characteristics, including a total runoff map, a runoff depth map, a runoff coefficient map and a runoff process line map.

[0066] Step B2: Spatial distribution update. Coupled sections are set at the confluence of tributaries and the main channel within the watershed. The runoff map is segmented using a rasterization method. From the segmented runoff map, a representative grid is selected for each confluence point to extract the hourly total runoff sequence.

[0067] Step B3: Initialize the Flood Modeller hydrodynamic domain, perform two-dimensional hydrodynamic simulation using the Flood Modeller hydrodynamic model, obtain the simulation area and spatial distribution map in the three-dimensional visualization scene, and set the initial water depth;

[0068] Step B4: Couple the hydrodynamic model, use the obtained hourly total runoff sequence as the inflow boundary condition of the Flood Modeller hydrodynamic model, and use the two-dimensional Saint-Venant equations of the alternating direction implicit solver to calculate the water depth and flow rate, simulating the entire process from rainfall to inundation.

[0069] Step B5: Parameter adjustment. Based on the spatial distribution map, the spatial distribution of the watershed is dynamically updated. The parameters of the watershed hydrological model and the Flood Modeller hydrodynamic model are adjusted using the Manning roughness coefficient to optimize the flood simulation process.

[0070] Step B6: Model output, set the maximum number of iterations, obtain historical watershed flood data to construct a calibration dataset, use the calibration dataset to iterate through steps B1 to B5 until the maximum number of iterations is reached, and output the calibrated model as the final distributed hydrological-hydrodynamic coupling model.

[0071] Step B7: Simulation output. Use the final distributed hydrological-hydrodynamic coupling model to simulate the water flow changes in the basin and obtain the simulation results.

[0072] Example 6, based on the above examples, includes the following specific steps in step B2 for updating the spatial distribution:

[0073] Based on the river network topology of the watershed, the confluence points of all tributaries flowing into the main channel are identified. At each confluence point, a virtual cross-section line (i.e., a coupled cross-section) is drawn perpendicular to the flow direction, serving as the data exchange interface for the hydrological-hydrodynamic model. The runoff map is segmented using a rasterization method near each confluence point, and the grid cell closest to that confluence point and located on the tributary outlet confluence path is selected as the representative grid. For the selected representative grid, the total runoff at each time step is extracted from the time series of the runoff map.

[0074] In step B5, the application of parameter adjustments includes the following:

[0075] The seepage dams in the spatial distribution map automatically generate high resistance during the rising water stage to delay the flood peak; during low flow, they only pass through the bottom orifice to maintain the ecological base flow; after the flood overflows the top, the resistance decreases slightly, but is still higher than that of the natural river channel.

[0076] Riverbank vegetation and woodland in the spatial distribution map reduce flood flow velocity, promote water storage, change the direction of water flow, and guide floods to spread to low-resistivity areas through high Manning values ​​(0.10–0.16).

[0077] The Manning value refers to the numerical value of the Manning roughness coefficient, usually represented by the symbol n. It is a dimensionless empirical parameter in hydraulics used to characterize the influence of the roughness of the water flow boundary on the flow resistance. It is widely used in the calculation of velocity and flow rate in open channels, rivers, drainage pipes and floodplains.

[0078] By performing the aforementioned operations, this scheme addresses the problems of existing flood simulation systems in simulating floods in small and medium-sized watersheds, such as the disconnect between runoff generation and flood evolution dynamics and the difficulty in quantifying the hydrological effects of natural intervention measures. It constructs a high-resolution distributed hydro-hydrodynamic coupling model that meticulously depicts the spatial distribution of watershed land cover and NFM measures in a grid format. Through a one-way coupling mechanism of time synchronization and spatial matching, the generated dynamic gridded runoff is used as a boundary condition to drive the Flood Modeller two-dimensional hydrodynamic model. This enables high spatiotemporal resolution dynamic simulation of the entire process from rainfall runoff generation and river evolution to floodplain inundation, providing scientific support for flood risk assessment and optimization of natural flood control measures in small and medium-sized watersheds.

[0079] Example 7, based on the above examples, presents a three-dimensional visualization scene of the watershed constructed by the digital twin watershed modeling module. This enables dynamic and immersive visualization of the entire flood process, drives real-time flood evolution animation, accurately displays the spatial expansion process of the inundation area, the hourly water depth change heat map, and the time series of the flood peak reaching each key section. Based on the risk assessment module, it synchronously displays a multi-level flood risk zoning map (including low, medium, high, and extremely high risk areas, using a red-orange-yellow-blue color scheme). At the same time, users can interactively explore the details of flood evolution by dragging the timeline, rotating the view, and cutting the profile.

[0080] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover 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 process, method, article, or apparatus.

[0081] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

[0082] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A watershed flood dynamic simulation system based on multi-source data fusion, characterized in that: The system includes a multi-source data preprocessing module, a multi-source data fusion module, a digital twin watershed modeling module, a flood dynamics simulation module, a flood risk assessment module, and a visualization module; The multi-source data preprocessing module is used to access data from ground rain gauges, soil moisture sensors, weather radars, satellite remote sensing, and numerical weather prediction models, and to perform spatiotemporal alignment, quality control, outlier removal, and format standardization on the accessed data to obtain multi-source data. The multi-source data fusion module uses a spatiotemporal cross-modal interactive fusion method to fuse multi-source data and obtain a fused feature vector; The multi-source data fusion module uses a spatiotemporal cross-modal interactive fusion method to fuse multi-source data, specifically including the following steps: Step A1: Remote sensing data feature extraction. Use ResNet101 to extract deep semantic features of remote sensing precipitation images from multi-source data to obtain remote sensing features, and obtain meteorological features from meteorological radar and numerical weather prediction models in multi-source data. Step A2: Ground data fusion. Ground hydrological sensor data is obtained from multi-source data. Bayesian dynamic bias correction method is used to correct the ground hydrological sensor data. Kalman filtering is combined to synergistically assimilate remote sensing precipitation monitoring and ground observation to obtain hydrological characteristics. Step A3: Watershed node definition and spatial adjacency matrix construction. Define all monitoring points in the watershed as graph nodes, and calculate the spatial weights between nodes based on the actual watershed hydrological topology to obtain the spatial adjacency matrix. Construct a multimodal feature vector for each node, which includes remote sensing features, hydrological features and meteorological features. Step A4: Feature decomposition. The direction-aware feature decomposition method is used to decompose the multimodal feature vector along the spatial dimension to obtain the horizontal and vertical features of the multimodal features. Step A5: Feature fusion. The horizontal and vertical features of the multimodal features of each node are concatenated to obtain the horizontal fused feature and the vertical fused feature. Step A6: Spatiotemporal graph construction. Combine the spatial adjacency matrix, horizontal features, and vertical features to construct a spatiotemporal graph, generating a temporal feature graph and a spatial feature graph. Step A7: Cross-modal interactive fusion, perform cross-modal interaction and low-rank multimodal fusion on the temporal feature map and spatial feature map to obtain the fused feature vector; The digital twin watershed modeling module constructs a spatial distribution map and a BIM model of engineering facilities based on DEM topography, river network structure, and neural factor decomposition machine model, generating a three-dimensional visualization scene of the watershed. The flood dynamics simulation module constructs a distributed hydrological-hydraulic coupling model, and simulates the water flow changes in the basin based on fused feature vectors and a three-dimensional visualization scene of the basin to obtain simulation results. The flood dynamics simulation module constructs a distributed hydrological-hydrodynamic coupled model, specifically including the following steps: Step B1: Distributed runoff generation and confluence simulation. A distributed watershed hydrological model is used as the upstream runoff generation and confluence model. The input fused feature vector is processed to simulate the rainfall-runoff process in the upstream watershed and output a runoff map. Step B2: Spatial distribution update. Coupled sections are set at the confluence of tributaries and the main channel within the watershed. The runoff map is segmented using a rasterization method. From the segmented runoff map, a representative grid is selected for each confluence point to extract the hourly total runoff sequence. Step B3: Initialize the Flood Modeller hydrodynamic domain, perform two-dimensional hydrodynamic simulation using the Flood Modeller hydrodynamic model, obtain the simulation area and spatial distribution map in the three-dimensional visualization scene, and set the initial water depth; Step B4: Couple the hydrodynamic model, use the obtained hourly total runoff sequence as the inflow boundary condition of the Flood Modeller hydrodynamic model, and use the two-dimensional Saint-Venant equations of the alternating direction implicit solver to calculate the water depth and flow rate, simulating the entire process from rainfall to inundation. Step B5: Parameter adjustment. Based on the spatial distribution map, the real-time spatial distribution of the watershed is dynamically updated, and the parameters of the watershed hydrological model and the Flood Modeller hydrodynamic model are adjusted to optimize the flood simulation process. Step B6: Model output, set the maximum number of iterations, obtain historical watershed flood data to construct a calibration dataset, use the calibration dataset to iterate through steps B1 to B5 until the maximum number of iterations is reached, and output the calibrated model as the final distributed hydrological-hydrodynamic coupling model. Step B7: Simulation output. Use the final distributed hydrological-hydrodynamic coupling model to simulate the water flow changes in the basin and obtain the simulation results. The flood risk assessment module assesses flood risk based on the simulation results of the flood dynamics simulation module and obtains the risk level. The visualization module provides an API interface to support integration with emergency management, water conservancy scheduling, and urban flooding early warning systems. It combines population and infrastructure disaster-bearing data to generate dynamic risk heat maps, which are then visualized and used for early warning pushes through a Web GIS platform.

Images