Apparatus and method for real-time urban flood simulation of dual drainage model based on webgpu

The WebGPU-based dual-drainage model addresses CPU limitations by integrating 2D surface runoff and 1D sewer networks in a web browser, providing real-time, cross-platform, and user-friendly urban flood simulation with accelerated GPU processing and direct data handling.

KR103047147B1Active Publication Date: 2026-11-11LAMILAB LNC
View PDF 6 Cites 0 Cited by

Patent Information

Application Number
KR1020260065927
Authority / Receiving Office
KR · KR
Patent Type
Patents
Current Assignee / Owner
Filing Date
2026-04-13
Publication Date
2026-11-11
Estimated Expiration
2046-04-13

AI Technical Summary

Technical Problem

Conventional urban flood simulation technologies face limitations in processing speed due to CPU-based computation, lack of a dual drainage model integrating surface runoff and sewer network drainage, platform dependency and compatibility issues, data preprocessing requirements, and user accessibility challenges, making real-time simulation and intuitive visualization difficult.

Method used

A WebGPU-based dual-drainage model that bidirectionally links 2D surface runoff and 1D sewer network models in a web browser environment, utilizing GPU parallel processing for accelerated calculations, cross-platform compatibility, direct processing of real-world data, and an intuitive user interface for real-time interaction and simulation adjustments.

Benefits of technology

Achieves real-time simulation with 5 to 20 times faster computation speed, cross-platform compatibility, direct data processing without normalization, and intuitive visualization, enabling immediate reflection of terrain changes and disaster prevention effects.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 112026044468682-PAT00032_ABST
    Figure 112026044468682-PAT00032_ABST
Patent Text Reader

Abstract

The present invention relates to an apparatus and method capable of simulating and predicting the process of flood occurrence and diffusion across an entire city by utilizing WebGPU technology to run a dual-drainage model in real time that bidirectionally links a 2D surface runoff model and a 1D sewer network model in a web browser environment. The problem that the present invention aims to solve is to resolve issues such as the slow processing speed of conventional CPU-based simulations, the lack of bidirectional linkage between surface runoff and sewer network drainage, platform dependency, complex data preprocessing, and the requirement for high user expertise. As a means to solve this, the present invention includes a terrain data automatic collection unit, a SWMM-based sewer network modeling unit, a WebGPU initialization unit, a numerical analysis unit that implements shallow water equations and advection equations using GPU compute shaders, a simulation execution unit including a dual-drainage coupling unit that performs bidirectional water exchange at junction points by synchronizing 2D surface runoff calculations and 1D sewer network calculations, and a 3D visualization and user real-time interaction unit. According to the present invention, through GPU parallel processing, performance improvement of 5 to 20 times or more compared to a CPU is achieved, web browser-based cross-platform compatibility is secured, real-world values ​​are processed directly without normalization, simulation is possible by inputting only rainfall intensity without specialized knowledge, and disaster prevention effects can be analyzed immediately by applying terrain changes such as the installation of cutoff walls and the collapse of embankments in real time during the simulation.
Need to check novelty before this filing date? Find Prior Art

Description

Technology Field

[0001] The present invention relates to urban flood simulation technology, and more specifically, to an apparatus and method capable of simulating and predicting the occurrence and spread of flooding throughout an entire city due to rainfall by utilizing WebGPU technology to run a dual drainage model in real time that bidirectionally links a 2D surface runoff model and a 1D sewer network model in a web browser environment. Background Technology

[0002] Urban flooding damage is increasing due to the progression of urbanization and the frequent occurrence of torrential downpours caused by climate change. Urban flooding occurs as a result of the complex interaction between surface runoff, where rainwater flows along the terrain and surroundings of buildings, roads, and the groundwater drainage system. In particular, when rainfall exceeds the drainage capacity of the sewer network, a bidirectional water exchange takes place where sewage backflows to the surface through manholes, while surface rainwater flows into the sewer pipes through manholes. Accurately simulating this bidirectional interaction between the surface and underground—known as dual drainage—is a key challenge in urban flood prediction.

[0004] Conventional urban flood simulation technologies have the following problems.

[0006] First, there is a limitation in processing speed due to CPU-based computation. Conventional 2D flood simulation tools operate on a CPU basis and sequentially calculate water depths and flow velocities for hundreds of thousands of grid cells, requiring a waiting period of tens of minutes to several hours to check the simulation results. Consequently, it is virtually impossible to immediately reflect changing field conditions, such as levee collapse or the installation of cutoff walls, during the simulation and re-simulate. In this regard, Chinese registered patent CN115795939B (June 13, 2023) presented a method for parallel simulation of unstructured grid-based 2D flood processes using multiple GPUs to improve computation speed; however, this operates in a native C++ / CUDA environment, which does not provide accessibility in web browser environments or cross-platform compatibility, nor does it include integration with sewer networks.

[0008] Second, there is a problem of the absence of a dual drainage model that integrates surface runoff and sewer network drainage, or incomplete coupling. Many existing flood simulation tools calculate only 2D surface runoff or perform only 1D sewer network simulations, failing to accurately reflect the bidirectional water exchange between the surface and underground. Australian Innovation Patent AU2020101587A4 disclosed a technology for calculating the exchange flow between the surface and underground pipe networks by presenting a method that couples a 1D river and pipe network model with a 2D hydrodynamic model based on spatial topology. However, this technology does not utilize GPU-accelerated computation, making real-time simulation impossible, and is limited to standalone systems independent of the web browser environment.

[0010] Third, there are issues of platform dependency and compatibility. Conventional GPU-based simulation technologies rely on specific graphics APIs such as CUDA, Direct3D, and OpenGL, requiring the installation of separate drivers and libraries for each operating system and graphics card manufacturer to run them. This platform dependency causes compatibility problems across different operating system environments (Windows, Linux, Mac, etc.) and graphics card environments (NVIDIA, AMD, etc.), incurring significant costs for system setup and maintenance.

[0012] Fourth, there are issues regarding data compatibility and preprocessing. In order to apply input data measured in the real world, such as digital elevation models (DEM), precipitation, surface roughness coefficients, sewer network data, and infiltration rates, to simulations, complex preprocessing work is required, such as normalizing all data to values ​​between 0 and 1. This hinders the rapid execution of simulations for new regions or new conditions.

[0014] Fifth, there are issues regarding user accessibility and learning costs. Using conventional flood simulation tools requires significant expertise in hydrology, fluid dynamics, and sewer network design, as well as the manual setting of numerous parameters. Furthermore, since simulation results are often displayed only in the form of text or two-dimensional contour lines, it is difficult to intuitively understand flood conditions, and it is impossible for users to immediately reflect conditions such as terrain changes or the installation of disaster prevention facilities during the simulation and verify their effects.

[0016] With the recent standardization of WebGPU technology, a foundation has been laid to utilize the parallel computing capabilities of GPUs in web browsers regardless of the operating system or type of graphics card; however, a real-time urban flood simulation technology based on a dual drainage model that bidirectionally links 2D surface runoff and 1D sewer network drainage using this has not yet been presented. Prior art literature

[0017] Chinese Registered Patent CN115795939B (June 13, 2023) Australian Innovation Patent AU2020101587A4 (September 10, 2020) The problem to be solved

[0018] The present invention aims to solve the problems of the aforementioned prior art and to provide a high-performance simulation device and method capable of simulating and predicting the process of flooding occurrence and spread throughout a city by utilizing WebGPU technology to run a dual drainage model in real time that bidirectionally links a 2D surface runoff model and a 1D sewer network model in a web browser environment.

[0020] Specifically, the problem that the present invention aims to solve is as follows.

[0022] First, by accelerating 2D surface runoff calculations through graphics processing unit (GPU) parallel processing and synchronizing and linking them with 1D sewer network calculations, it achieves a calculation speed improvement of 5 to 20 times or more compared to existing CPU-only calculations, thereby providing a real-time simulation environment that enables re-simulation by immediately reflecting changing field conditions, such as levee collapse or cutoff wall installation, during the simulation.

[0024] Second, it is to provide simulation results that correspond to the actual mechanism of urban flooding by implementing a dual drainage model that accurately simulates the bidirectional water exchange between the surface and underground, in which rainwater from the surface flows into the sewer pipe through manholes, backflows to the surface through manholes when the water level rises, and discharges the pipe water to the surface through outlets.

[0026] Third, by utilizing WebGPU technology to run in a web browser environment, it provides cross-platform compatibility on devices equipped with graphics cards without the need for separate software installation, regardless of the operating system (Windows, Linux, Mac), graphics card manufacturer (NVIDIA, AMD), or graphics API (CUDA, Direct3D, OpenGL).

[0028] Fourth, the system can directly process real-world values ​​without preprocessing to normalize input data such as digital elevation models (DEM), precipitation, surface roughness coefficients, sewer network data, and infiltration rates to values ​​between 0 and 1, thereby ensuring the speed of simulation for new regions or conditions.

[0030] Fifth, input data such as numerical elevation models, surface roughness coefficients, sewer networks, and infiltration rates are embedded in the system, allowing the user to run a simulation by simply inputting the rainfall intensity, thereby providing an intuitive interface that can be easily used without specialized knowledge of hydrology or fluid dynamics.

[0032] Sixth, the goal is to implement a real-time interactive simulation environment where users can pause, restart, and adjust playback speed during the simulation, as well as apply terrain changes in real time—such as installing cutoff walls, collapsing levees, adding terrain, or adding open channels—and immediately analyze the resulting drainage and disaster prevention effects. means of solving the problem

[0033] To solve the above-mentioned problem, the WebGPU-based dual-drainage model real-time urban flood simulation device according to the present invention simulates and predicts the process of flood occurrence and diffusion throughout the city by running a dual-drainage model in real time that bidirectionally links a 2D surface runoff model and a 1D sewer network model in a web browser environment. This comprises a terrain data collection unit, a sewer network modeling unit, a WebGPU initialization unit, a numerical analysis unit, a simulation execution unit, and a visualization and interaction unit.

[0035] Specifically, the terrain data collection unit collects satellite imagery and Digital Elevation Model (DEM) data of the simulation target area, and generates a terrain model by integrating Digital Strand Model (DSM), cover map data, and river data. At this time, the rectangular boundaries of the simulation target area are converted into tile numbers based on a Quadtree Tile System in the format of a Web Map Tile Service (WMTS), and satellite imagery corresponding to each tile number is sequentially collected and combined to generate the texture of the terrain model.

[0037] The sewer network modeling unit receives an INP file of the EPA SWMM stormwater management model (EPA Storm Water Management Model) and generates a sewer network model including junction, conduit, and outfall elements. In this process, the C language source code of the SWMM engine is compiled into WebAssembly to prepare for performing one-dimensional sewer network operations in a web browser environment.

[0039] The WebGPU initializer creates a WebGPU context in the web browser and configures compute shaders and rendering pipelines for simulation. This provides a cross-platform environment that allows the utilization of GPU parallel computing capabilities on devices equipped with graphics cards without the need for separate software installation, regardless of the operating system (Windows, Linux, Mac), graphics card manufacturer (NVIDIA, AMD), or graphics API (CUDA, Direct3D, OpenGL).

[0041] The numerical analysis unit implements 2D surface runoff calculations using GPU compute shaders, which include flow velocity calculation based on the shallow water equation and the advection equation, surface friction processing considering the Manning roughness coefficient, and determination of the water level of each cell. In particular, by utilizing the RGBA channels of the texture to store the flow velocity in the X direction in the Red channel, the flow velocity in the Y direction in the Green channel, the water level in the Blue channel, and the elevation in the Alpha channel, computational efficiency is maximized by simultaneously reading the four physical quantities of the corresponding cell with a single memory access for a single pixel.

[0043] The simulation execution unit synchronizes and drives the 2D surface runoff calculation by the numerical analysis unit and the 1D sewer network calculation by the SWMM engine according to their respective time intervals. Specifically, the time interval for the 2D surface runoff calculation is set to the first time interval, and the time interval for the 1D sewer network calculation is set to the second time interval, which is larger than the first time interval, so that the 1D sewer network calculation is performed once whenever the 2D surface runoff calculation is performed multiple times. In performing the bidirectional double drainage coupling, the 2D grid cell data stored in the GPU memory is copied to the CPU memory, and then all junction points of the SWMM engine are traversed. The water level of the surface cell where each junction point is located is compared with the water level of the corresponding junction point to determine the inflow from the surface to the junction point or the outflow from the junction point to the surface. In addition, the flow rate of the conduit connected to the discharge port is calculated and reflected in the corresponding surface cell, thereby simulating the flow of water discharged from the pipeline to the surface.

[0045] The simulation execution unit performs the following steps repeatedly at every hour interval in detail. First, in the hydrology stage, the infiltration amount of each cell is calculated by applying the HOLTON infiltration formula based on cover data, and the water level corresponding to effective rainfall is assigned to each cell by subtracting the infiltration amount from the rainfall amount. Next, in the drainage coupling stage, the aforementioned bidirectional double drainage coupling is performed. Subsequently, in the flow velocity calculation stage, the flow velocity of each cell is calculated based on the water level, elevation, and flow velocity of adjacent cells. Then, in the water level update stage, the water level of each cell is updated based on the external inflow amount determined in the drainage coupling stage and the flow velocity calculated in the flow velocity calculation stage. Finally, in the time interval determination stage, the next time interval is dynamically determined based on the CFL (Courant-Friedrichs-Lewy) condition, but the numerical stability of the simulation is maintained by adjusting the value to be smaller than the initial time interval according to the maximum water depth within the grid cell.

[0047] The visualization and interaction unit visualizes the simulation results of the simulation execution unit in three dimensions and processes the user's pause, restart, and playback speed adjustments in real time during the simulation. In addition, it applies terrain changes corresponding to the installation of cutoff walls, collapse of embankments, addition of terrain, or addition of open channels based on user input during the simulation, and simulates drainage and disaster prevention effects in real time according to the changed terrain conditions.

[0049] In addition, the present invention improves usability by directly processing real-world values ​​without data normalization, achieves a computation speed improvement of 5 to 20 times or more compared to existing CPU-only computation through GPU parallel processing, and can achieve 60 frames per second (FPS) at 1x simulation speed based on 512×512 cells.

[0051] As another means to solve the above-mentioned problem, the WebGPU-based dual-drainage model real-time urban flood simulation method according to the present invention simulates flooding throughout the city by running a dual-drainage model in real time that bidirectionally links a 2D surface runoff model and a 1D sewer network model in a web browser environment. This comprises: (a) a terrain data collection step for creating a terrain model by collecting satellite imagery, a Digital Elevation Model (DEM), Digital Building Model (DSM), cover map data, and river data of the simulation target area; (b) a sewer network modeling step for creating a sewer network model including junction, conduit, and outlet elements from the INP file of the EPA SWMM rainfall-runoff model; (c) a WebGPU initialization step for creating a WebGPU context in a web browser and setting up compute shaders and a rendering pipeline; (d) a numerical analysis algorithm implementation step for implementing 2D surface runoff calculations using GPU compute shaders, including flow velocity calculations based on shallow water equations and advection equations, surface friction processing considering the Manning roughness coefficient, and determination of the water level of each cell; and (e) synchronously executing the 2D surface runoff calculations by the numerical analysis algorithm and the 1D sewer network calculations by the SWMM engine according to their respective time intervals, wherein the water level of the surface at each junction and the junction's The system comprises a real-time simulation execution step including a bidirectional double drainage coupling that determines inflow from the surface to the junction point or outflow from the junction point to the surface point by comparing water levels and reflects the flow rate of the conduit connected to the discharge port to the corresponding surface cell, and (f) a visualization and interaction step that visualizes the simulation results in three dimensions and processes real-time user interaction.

[0053] The above real-time simulation execution step (e) includes a hydrological step of assigning a water height due to effective rainfall to each cell by applying an infiltration formula based on cover map data, a drainage coupling step of performing bidirectional double drainage coupling by synchronizing the time of a 2D surface runoff calculation and a 1D sewer network calculation, a flow velocity calculation step of calculating the flow velocity of each cell based on the water height elevation and flow velocity of adjacent cells, a water height update step of updating the water height of each cell based on the external inflow amount determined in the drainage coupling step and the flow velocity calculated in the flow velocity calculation step, and a time interval determination step of dynamically determining the next time interval based on CFL conditions, and the above steps are repeated every time interval.

[0055] The above sewer network modeling step (b) includes the step of compiling the C language source code of the SWMM engine into WebAssembly to prepare for performing one-dimensional sewer network operations in a web browser environment. Effects of the invention

[0056] First, through a dual drainage model that links two-dimensional surface runoff and one-dimensional sewer network drainage in both directions, it is possible to provide precise simulation results that correspond to the actual mechanism of urban flooding.

[0058] Second, through graphics processing unit (GPU) parallel processing, performance improvement of 5 to 20 times or more compared to conventional central processing unit (CPU) single-processing is achieved, enabling real-time re-simulation that immediately reflects changing conditions such as terrain changes during simulation.

[0060] Third, by utilizing WebGPU technology, cross-platform compatibility can be provided in a web browser environment without the need for separate software installation, regardless of the operating system, graphics card type, or graphics API.

[0062] Fourth, real-world values ​​are processed directly without normalization of input data, and simulation is possible by simply inputting rainfall intensity, making it easy for even users without specialized knowledge to use.

[0064] Fifth, topographical changes such as the installation of cutoff walls, levee collapse, and the addition of open channels can be applied in real time during the simulation to immediately analyze drainage and disaster prevention effects. Brief explanation of the drawing

[0065] FIG. 1 is a block diagram showing the configuration of a WebGPU-based double drainage model real-time urban flood simulation device according to one embodiment of the present invention. FIG. 2 is a flowchart showing the overall flow of a WebGPU-based double drainage model real-time urban flood simulation method according to an embodiment of the present invention. FIG. 3 is a conceptual diagram showing a double drainage bidirectional coupling structure between a two-dimensional surface runoff model and a one-dimensional sewer network model according to an embodiment of the present invention. FIG. 4 is a flowchart showing the detailed steps of the real-time simulation loop of a simulation execution unit according to one embodiment of the present invention. FIG. 5 is a structural diagram showing a two-dimensional grid and RGBA texture data structure according to one embodiment of the present invention. FIG. 6 is a diagram showing the operation flow and interrelationships between key formulas of a numerical analysis unit according to one embodiment of the present invention. FIG. 7 is a diagram showing the process of collecting and combining satellite image tiles of a terrain data collection unit according to one embodiment of the present invention. FIG. 8 is a diagram showing the preprocessing pipeline flow of raw spatial data according to one embodiment of the present invention. FIG. 9 is a graph showing the results of a performance comparison between GPU-based computation and CPU-based computation according to grid resolution in accordance with one embodiment of the present invention. FIG. 10 is an illustration showing the overall operation concept of rainfall, surface runoff, sewer network interlocking, flooding occurrence, and disaster prevention response in an urban environment according to one embodiment of the present invention. FIG. 11 is a diagram showing the screen configuration of a web browser-based simulation user interface according to one embodiment of the present invention. Specific details for implementing the invention

[0066] Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the attached drawings. In describing the present invention, detailed descriptions of related known technologies are omitted if it is determined that such detailed descriptions may unnecessarily obscure the essence of the present invention. Furthermore, while preferred embodiments of the present invention will be described below, it is understood that the technical concept of the present invention is not limited thereto and can be modified and implemented in various ways by those skilled in the art.

[0068] The terms used in this specification have been selected to be as widely used as possible, taking into account their functions in the present invention; however, these terms may vary depending on the intent or convention of those skilled in the art, the emergence of new technologies, etc. Additionally, in specific cases, terms have been arbitrarily selected by the applicant, and in such cases, their meanings will be described in detail in the relevant description of the invention. Therefore, the terms used in this invention should be defined not merely by their names, but based on their meanings and the overall content of the invention.

[0070] In this specification, terms such as “comprising” or “having” are intended to indicate the existence of the features, numbers, steps, actions, components, parts, or combinations thereof described in the specification, and should be understood as not excluding in advance the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof.

[0072] In this specification, "Dual Drainage Model" refers to an integrated simulation system that simultaneously operates a two-dimensional (2D) surface runoff model for calculating rainwater flow on the surface and a one-dimensional (1D) sewer network model for calculating water flow inside an underground sewer network when simulating urban flooding phenomena, and calculates bidirectional water exchange between the two, such as the inflow of rainwater through manholes (junction points), backflow of sewage, and surface runoff of pipe discharge water through discharge outlets.

[0074] In this specification, "cell" refers to the smallest unit when a simulation target area is divided into a two-dimensional grid, and corresponds one-to-one with a single pixel of a texture when computed by a graphics processing unit (GPU).

[0076] As used herein, terms such as “part,” “module,” etc. refer to software or hardware components, each of which performs specific roles. However, these components are not limited to software or hardware. Each of these may be configured to reside in an addressable storage medium or configured to operate one or more processors.

[0078] Now, the configuration and operation according to an embodiment of the present invention will be described in detail with reference to the attached drawings.

[0080] FIG. 1 is a block diagram schematically showing the internal configuration of a WebGPU-based double drainage model real-time urban flood simulation device (100) according to one embodiment of the present invention.

[0082] Referring to FIG. 1, the device (100) according to the present invention utilizes the parallel processing capability of a graphics processing unit (GPU) in a web browser environment to operate a dual drainage model that links two-dimensional surface runoff and one-dimensional sewer network drainage in both directions in real time, thereby precisely simulating the process of flooding occurrence and spread throughout the city and providing the user with intuitive three-dimensional visualization results and a real-time interaction environment.

[0084] To this end, as illustrated in FIG. 1, the device (100) according to the present invention comprises a terrain data collection unit (110), a sewer network modeling unit (120), a WebGPU initialization unit (130), a numerical analysis unit (140), a simulation execution unit (150), and a visualization and interaction unit (160). These components are organically connected to exchange data and control signals and systematically perform the entire simulation process. The arrows in FIG. 1 indicate the direction of data flow between each component.

[0086] First, the terrain data collection unit (110) is responsible for generating base data to construct a virtual space where the entire simulation will take place. This component collects satellite image data and Digital Elevation Model (DEM) data from an external server for a specific area designated by the user. In addition, it integrates building height data (DSM, Digital Surface Model), cover map data (including surface roughness coefficient, minimum infiltration capacity, and maximum infiltration capacity), and river data to generate a terrain model that serves as the basis for the simulation.

[0088] The terrain model generated by the terrain data collection unit (110) is in the form of a two-dimensional grid with a scale of 512×512 cells, and real-world values ​​such as elevation above sea level, building height, surface roughness coefficient, and infiltration capacity are directly stored in each cell without normalization. This solves the problem in conventional technology where all input data values ​​had to be normalized to decimal values ​​between 0 and 1, thereby enabling rapid simulation of a new area. In particular, regarding the collection of satellite images, a quadtree tile system in the form of a Web Map Tile Service (WMTS) is utilized to convert the rectangular range of the simulation target area into tile numbers, and satellite images corresponding to each tile number are sequentially collected and combined to efficiently generate the texture of the terrain model. The specific details of this satellite image collection process will be described later with reference to FIG. 8. Additionally, the preprocessing process of various input data such as DEM, building DSM, rivers, and cover maps will be described later with reference to FIG. 9. The terrain model generated by the terrain data collection unit (110) is transmitted to the simulation execution unit (150). This corresponds to the arrow pointing from the terrain data collection unit (110) to the simulation execution unit (150) in Fig. 1, labeled "terrain model".

[0090] The sewer network modeling unit (120) performs the role of generating a sewer network structure that determines underground water flow. This component receives an INP file of the EPA SWMM stormwater runoff model and generates a sewer network model that includes key elements of the sewer network such as junctions, conduits, and outfalls.

[0092] Here, a junction is a point in a sewer network where multiple conduits meet or diverge, corresponding to a manhole in the real world, and has a depth from the surface to the invert elevation. A conduit is a sewer pipe connecting junctions to junctions or to an outlet, forming a path for water movement. An outlet refers to the final discharge point where water collected through the sewer network is released into an external water system, such as a river.

[0094] The most critical technical feature of the sewer network modeling unit (120) is that it compiles the C language source code of the SWMM engine into WebAssembly to prepare for performing one-dimensional sewer network calculations in a web browser environment. The SWMM engine was originally implemented in C language and used as a plugin in a CLI (Command Line Interface) environment or hydrological tool, but in the present invention, all source files of SWMM are compiled into a single WebAssembly module using the EMCC (Emscripten Compiler) tool. Through this, the SWMM engine can be run within a web browser without separate installation, and this becomes the core means responsible for one-dimensional sewer network calculations in the double drainage model of the present invention. The sewer network model generated in the sewer network modeling unit (120) is transmitted to the simulation execution unit (150). This corresponds to the arrow pointing from the sewer network modeling unit (120) to the simulation execution unit (150) in FIG. 1, labeled "Sewer Network Model."

[0096] The WebGPU initialization unit (130) is responsible for setting up and preparing the control environment of a graphics processing unit (GPU) within a web browser so that the present invention can perform high-performance parallel computation. This ensures cross-platform compatibility that is not dependent on operating systems (Windows, Linux, Mac), graphics card manufacturers (NVIDIA, AMD), or graphics APIs (CUDA, Direct3D, OpenGL) by utilizing WebGPU technology, which is the latest web graphics standard. Specifically, the WebGPU initialization unit (130) creates a context that allows access to the GPU device from the web browser, and sets up and compiles a compute shader to process complex physics computations in parallel and a rendering pipeline to render simulation results to the screen. Once this initialization process is completed, the GPU is ready to perform large-scale parallel computation, and the set computation environment information is transmitted to the numerical analysis unit (140). This corresponds to the arrow pointing from the WebGPU initialization unit (130) to the numerical analysis unit (140) in FIG. 1, labeled "GPU computation environment."

[0098] The numerical analysis unit (140) is the core brain of the simulation, implementing an algorithm that mathematically calculates fluid dynamic movements, such as flooding phenomena, based on physical laws. This component operates on a GPU computing environment prepared through the WebGPU initialization unit (130).

[0100] There are three main core algorithms implemented in the numerical analysis unit (140). First, there are shallow-water equations that efficiently simulate the movement of water in a shallow environment. Second, there are advection equations that calculate changes in physical quantities according to the flow of water. Third, there is surface friction processing based on Manning's roughness coefficient that considers frictional resistance according to the roughness of the surface. These algorithms are implemented as compute shader code optimized for the graphics processing unit.

[0102] In particular, the numerical analysis unit (140) utilizes the four RGBA channels of the texture as a data storage to maximize computational efficiency. The Red channel stores the flow velocity in the X direction, the Green channel stores the flow velocity in the Y direction, the Blue channel stores the water height, and the Alpha channel stores the elevation above sea level, so that the four physical quantities of the cell can be read simultaneously with a single memory access for a single pixel. The specific details of this data structure will be described later with reference to FIG. 5, and the core formulas used by the numerical analysis unit (140) will be described later with reference to FIG. 7. The numerical analysis algorithm implemented in this way is called by the simulation execution unit (150) to perform actual calculations. This corresponds to the arrow pointing from the numerical analysis unit (140) to the simulation execution unit (150) in FIG. 1, which is labeled "Numerical Analysis Algorithm".

[0104] The simulation execution unit (150) acts as a central control unit that manages all previously prepared components to run the actual flood simulation in real time. As indicated by the arrow in Fig. 1, the simulation execution unit (150) receives terrain model data from the terrain data collection unit (110), receives sewer network model data from the sewer network modeling unit (120), and receives a numerical analysis algorithm (compute shader) capable of calculating physical phenomena from the numerical analysis unit (140).

[0106] The most important technical feature of the simulation execution unit (150) is that it performs a two-dimensional surface runoff calculation and a one-dimensional sewer network calculation by synchronizing them according to their respective time intervals, and determines the inflow from the surface to the junction or the outflow from the junction to the surface by comparing the water level of the surface at each junction with the water level at the junction, and performs a bidirectional double drainage coupling that reflects the flow rate of the conduit connected to the discharge port to the corresponding surface cell.

[0108] Specifically, the time interval for the 2D surface runoff calculation is set to a first time interval (e.g., 0.25 seconds), and the time interval for the 1D sewer network calculation is set to a second time interval (e.g., 2 seconds) that is larger than the first time interval, so that the 1D sewer network calculation is performed once whenever the 2D surface runoff calculation is performed multiple times (eight times in the above example). In performing this bidirectional double drainage coupling, the 2D grid cell data stored in GPU memory is copied to CPU memory, and then all junction points of the SWMM engine are traversed to determine inflow or outflow by comparing the water level of the surface cell where each junction point is located with the water level of the corresponding junction point. This process accurately reproduces the actual mechanism of urban flooding, namely, the phenomenon where rainwater from the surface flows into the sewer pipe through manholes when there is little rain, and sewage flows back to the surface through manholes when there is a lot of rain and the water level in the sewer pipe rises. The specific mechanism of this double-double coupling will be described later with reference to FIG. 3, and the detailed step-by-step flow of the simulation loop will be described later with reference to FIG. 4.

[0110] When a user inputs scenario data such as rainfall intensity, the simulation execution unit (150) issues a computation command to the GPU at each time step based on this input value. The GPU simultaneously executes numerical analysis algorithms in parallel for each of the hundreds of thousands of simulation grid cells to calculate the water level and flow velocity changes for the next time step. As this process is repeated dozens of times per second, the entire process of rainwater accumulating on the surface, flowing through sewer pipes, and the expansion of flooded areas over time is simulated in real time. The calculated final result data is transmitted to the visualization and interaction unit (160). This corresponds to the arrow pointing from the simulation execution unit (150) to the visualization and interaction unit (160) in FIG. 1 labeled "simulation result data".

[0112] The visualization and interaction unit (160) converts complex simulation result data into visual information that the user can intuitively understand and provides it, and acts as an interface that responds to the user's operation in real time. This component receives water level data that is updated every frame from the simulation execution unit (150) and renders the height of the water as realistic graphics on a 3D terrain model. Through this, the user can clearly identify which areas are flooded and to what depth through a three-dimensional screen.

[0114] In addition, the visualization and interaction unit (160) provides functions that allow the user to pause or restart the simulation, adjust the playback speed from 1x to 50x, and change the amount of precipitation to reflect it immediately. In particular, during the simulation, terrain changes such as the installation of a cutoff wall, collapse of an embankment, addition of terrain, and addition of an open channel can be applied by user input, and drainage and disaster prevention effects according to the changed terrain conditions can be simulated in real time. The specific configuration of this user interface will be described later with reference to FIG. 11. In FIG. 1, the reverse arrow pointing from the visualization and interaction unit (160) to the simulation execution unit (150) labeled "User input (rainfall, terrain change)" indicates a bidirectional interaction structure in which user operations are reflected in the simulation in real time.

[0116] In FIG. 1, the dotted box surrounding the entire device (100) and labeled "web browser environment" indicates that all components of the device (100) according to the present invention operate within a web browser. This indicates that real-time urban flood simulation according to the present invention can be performed using only a web browser on any terminal equipped with a graphics card, without the need to install separate software on the user's terminal. As an example of the hardware specifications required for operation, in an environment equipped with a CPU of Intel 12th generation 24 cores or higher, a GPU of NVIDIA GeForce RTX 3060 or higher, and RAM of 32GB or higher, 60 frames per second (FPS) can be achieved at 1x simulation speed based on 512×512 cells.

[0118] In conclusion, each component illustrated in FIG. 1 systematically divides and organically links a series of processes including data collection (110), sewer network modeling (120), GPU computation environment setup (130), numerical analysis algorithm implementation (140), real-time simulation execution including double drainage coupling (150), and 3D visualization and real-time interaction (160), thereby overcoming the limitations of speed, compatibility, and accessibility of conventional CPU-based simulation technology and implementing a high-performance double drainage model-based real-time urban flood simulation device that anyone can easily use on a web browser.

[0119] Next, the overall flow of the method according to the present invention will be explained with reference to FIG. 2.

[0121] FIG. 2 is a flowchart showing the overall flow of a WebGPU-based double drainage model real-time urban flood simulation method according to an embodiment of the present invention.

[0123] Referring to FIG. 2, the method according to the present invention is performed sequentially, including a terrain data collection step (S100), a sewer network modeling step (S200), a WebGPU initialization step (S300), a numerical analysis algorithm implementation step (S400), a real-time simulation execution step (S500), and a visualization and interaction step (S600). Each of these steps corresponds to each component of the device described in FIG. 1, namely, a terrain data collection unit (110), a sewer network modeling unit (120), a WebGPU initialization unit (130), a numerical analysis unit (140), a simulation execution unit (150), and a visualization and interaction unit (160).

[0125] First, in the terrain data collection step (S100), the foundational environment necessary for the simulation is established. In this step, satellite images, a digital elevation model (DEM), building height data (DSM), cover map data, and river data of the simulation target area are collected to create a terrain model.

[0127] Specifically, the collection of satellite imagery is carried out through the following process. First, the geographical area to be simulated is defined in the form of a rectangle. For example, assuming the area around Dunsan-dong, Seo-gu, Daejeon Metropolitan City is set as the target area for flood simulation, the system sets the minimum longitude, maximum longitude, minimum latitude, and maximum latitude of the smallest rectangle that completely encompasses that area.

[0129] Next, the defined rectangular range is converted into tile numbers based on the quadtree tile system. The world map is divided into a grid according to the zoom level; at zoom level 0, the world is a single tile, but for every 1 increase in the zoom level, each tile is divided into a total of 4 tiles, with 2 tiles in each row and 2 tiles in each column. Based on the zoom level desired by the user, for example, zoom level 16, the system calculates the unique X, Y coordinate numbers of all tiles spanning that range.

[0131] Next, using the calculated tile number, each tile image is requested one by one from a map service server, such as a VWorld server, in accordance with the Web Map Tile Service (WMTS) standard. That is, a satellite image corresponding to 'Zoom Level=16, X=54321, Y=27180' (e.g., a PNG file with a size of 256×256 pixels) is requested and downloaded, and then the image for the next tile number is requested, thereby collecting all necessary tiles sequentially. At this time, to avoid placing an excessive load on the map server, a delay of 0.1 seconds is applied each time a tile image is requested.

[0133] Finally, all downloaded tile images are combined. The system references the X and Y coordinate values ​​of each tile to connect them in the correct order at precise locations, much like fitting together a puzzle. Through this process, dozens to hundreds of small tile images are ultimately combined into a single high-resolution satellite image representing the entire simulation area.

[0135] For the collection of DEM data, rectangular boundaries of the simulation target area are specified from the original DEMs stored on the server by province, and a 512×512 cell size DEM is generated using the gdal_translate tool of GDAL (Geospatial Data Abstraction Library). At this stage, a bilinear interpolation algorithm is applied to ensure data continuity. Building height data (DSM) is generated by converting building information stored in vector format in a PostgreSQL database into raster format using the gdal_rasterize tool of GDAL. In this process, if building height information is unavailable, the height is calculated by multiplying the number of floors by the average height per floor (e.g., 2.4 meters). River data is generated by merging publicly available river SHP (Shapefile) files existing in South Korea, cropping them into rectangular boundaries, and converting them into raster files. Cover map data is generated by combining data included within a rectangular range from SHP files by city and province, and then converting three variables—surface roughness coefficient (roughness coefficient), minimum infiltration capacity (mm / h), and maximum infiltration capacity (mm / h)—into raster files.

[0137] Next, in the sewer network modeling step (S200), underground water flow is modeled. In this step, an INP file of the EPA SWMM stormwater runoff model is input, and a sewer network model is created that includes key elements such as junctions, conduits, and outfalls.

[0139] The most important task in this stage is to compile the C source code of the SWMM engine into WebAssembly to prepare it for performing one-dimensional sewer network calculations in a web browser environment. To achieve this, the EMCC (Emscripten Compiler) tool is used, and during compilation, all SWMM source files (swmm5.c, climate.c, controls.c, culvert.c, datetime.c, dwflow.c, dynwave.c, etc.) are integrated into a single WebAssembly module. As compilation options, optimization level O1, WASM enabled, and file system forced inclusion are applied, and runtime methods such as ccall, cwrap, getValue, and setValue are exported so that they can be called externally. This enables one-dimensional sewer network simulations to be performed by directly calling the core functions of the SWMM engine from JavaScript code within the web browser.

[0141] In addition, in this step, the XY coordinate values ​​and coordinate systems of all junctions and outlets included in the SWMM INP file are read to determine the surface cell location of each element. This coordinate information is subsequently used in the simulation execution step (S500) to identify the surface cell where each junction is located during bidirectional double drainage coupling.

[0143] Next, in the WebGPU initialization step (S300), preparations are made to perform high-performance graphics operations in a web browser. In this step, a WebGPU context is created to allow access to the GPU device from the web browser, and a compute shader to process the core physics operations of the simulation in parallel and a rendering pipeline to display the simulation results on the screen are set up.

[0145] The technical significance of this stage lies in securing a cross-platform environment that is not dependent on the operating system or graphics card type by utilizing the latest web standard known as WebGPU. Conventional GPU-based simulation technologies relied on specific graphics APIs such as CUDA (NVIDIA exclusive), Direct3D (Windows exclusive), and OpenGL, which led to compatibility issues between platforms. However, WebGPU provides an abstraction layer on top of these graphics APIs, allowing the web browser to automatically select the appropriate underlying API. Consequently, users can perform GPU-accelerated simulations using only a web browser on any device equipped with a graphics card.

[0147] Subsequently, in the numerical analysis algorithm implementation stage (S400), the physical laws of the simulation are implemented in code. In this stage, complex fluid dynamics calculations, such as shallow-water equations, advection equations, and surface friction processing using Manning's roughness coefficient, are implemented using compute shaders optimized for parallel processing on graphics processing units (GPUs).

[0149] Explain the key formulas implemented in this step.

[0151] The shallow water equation is a two-dimensional fluid behavior equation that forms the basis of the simulation, and consists of a continuity equation for updating the water height (water level) and a momentum equation for updating the flow velocity. The equation for updating the water height is ∂η / ∂t + v·∇η = -η∇·v, and the equation for updating the flow velocity is ∂v / ∂t + v·∇v = g∇h. Here, η represents the water height, v represents the horizontal flow velocity vector, g represents the acceleration due to gravity, and h represents the total hydrostatic head, which is the sum of the water level and the terrain elevation. In the implementation of the present invention, a semi-implicit scheme is applied, but the value of the weight parameter θ is fixed at 0.9 to ensure a balance between numerical stability and accuracy.

[0153] The friction gradient formula based on Manning's law is Sf = n²·v|v| / h^(4 / 3). Here, Sf represents the friction gradient, n represents the local Manning roughness coefficient, v represents the flow velocity vector, and h represents the water depth. This formula differentially calculates the frictional force hindering water flow depending on the type of surface (roads, green spaces, areas around buildings, etc.), allowing roughness coefficient values ​​included in the cover data to be applied cell by cell.

[0155] The equation determining the water level is d(t+Δt_2D) = d(t) + d_ext(t) + ΣQ / (ΔxΔy)·Δt_2D. Here, d(t) represents the current water depth, d_ext(t) represents the external water depth added or reduced by the 1D sewer network calculation (drainage coupling stage), Q represents the flow rate to neighboring cells, Δx and Δy represent the grid resolution, and Δt_2D represents the simulation time interval. In this equation, the d_ext(t) term is the core of the double drainage coupling, and the amount of water exchanged from the underground sewer network to the surface (backflow or discharge) is reflected in the 2D surface runoff calculation through this term.

[0157] The formula for determining the time interval to maintain the numerical stability of the simulation is determined as Δt_2D = a·min(Δx,Δy) / √(g·d_max) based on the CFL (Courant-Friedrichs-Lewy) condition, where a is the CFL safety factor and d_max is the maximum water depth within the grid cell. Through this, the time interval can be lowered from the initial value (e.g., 0.25 seconds) depending on the water level, which serves as a safety mechanism to prevent divergence of the simulation.

[0159] The HOLTON equation for calculating surface infiltration capacity is f = fc + (f0 - fc)·e^(-K·t). Here, f represents the infiltration rate at time t, fc represents the minimum infiltration rate, f0 represents the maximum infiltration rate, and K represents the damping coefficient. The value of K is calculated as (MIN_K - MAX_K) × infiltration efficiency + MAX_K, and in this embodiment, MIN_K is set to 2, MAX_K to 9, and infiltration efficiency to 1.0 (100%). This formula simulates the phenomenon in which the surface infiltration capacity gradually decreases as simulation time elapses.

[0161] The numerical analysis unit (140) utilizes the four RGBA channels of the texture as a data storage unit to implement these formulas in a form optimized for the GPU. When the simulation target area is divided into a grid, each cell corresponds one-to-one with a single pixel of the texture. A single pixel generally has four channel values: Red (R), Green (G), Blue (B), and Alpha (A), but in the present invention, these are utilized for storing physical data rather than image color information. The flow velocity in the X direction of the corresponding cell is stored in the Red channel, the flow velocity in the Y direction in the Green channel, the height of the water in the Blue channel, and the elevation above sea level in the Alpha channel. Through this, when the GPU wants to calculate the physical phenomenon of a specific cell, it can simultaneously retrieve four key pieces of information by reading the corresponding pixel of the texture only once. Thousands of computing cores of the GPU simultaneously read information for thousands of pixels (cells), simultaneously calculate physical equations, and then write the results back to a new texture.

[0163] Then, in the real-time simulation execution stage (S500), actual flooding phenomena are simulated. In this stage, rainfall data is received from the user, and the water level change is calculated in real time and the flooding spread process is simulated using the previously implemented numerical analysis algorithm and the parallel processing power of the graphics processing unit (GPU). In particular, the core of this stage is to perform bidirectional double drainage coupling by synchronizing the 2D surface runoff calculation and the 1D sewer network calculation.

[0165] In FIG. 2, the block representing the real-time simulation execution step (S500) is depicted with a larger and thicker border than other blocks, indicating that this step is the most critical step in the method of the present invention. Additionally, the notation "(simulation loop, see FIG. 4)" inside the S500 block signifies that within this step, there exists a cyclic loop in which the hydrology step (S510), drainage interlocking step (S520), flow velocity calculation step (S530), water level update step (S540), and time interval determination step (S550) are repeatedly performed at every time interval, and the details thereof will be described later with reference to FIG. 4.

[0167] Finally, in the visualization and interaction step (S600), the simulation results are displayed to the user and user interactions are processed. In this step, the calculated flood results are visualized on a 3D terrain model so that the user can understand them intuitively, and user operations such as pausing, restarting, adjusting playback speed, and changing terrain during the simulation are reflected in real time.

[0169] In FIG. 2, the dotted arrow returning from the visualization and interaction step (S600) to the real-time simulation execution step (S500) is labeled "Reflect User Input," which means that if the user changes rainfall intensity or performs terrain changes such as installing a cutoff wall or collapsing a levee during the simulation, the changes are immediately reflected in the real-time simulation execution step (S500), and the flooding process under the changed conditions is continuously simulated. This real-time feedback structure is a key technical feature of the present invention, in contrast to conventional simulation tools, where results had to be checked after the simulation was completed and input conditions had to be changed to re-execute from the beginning.

[0171] Meanwhile, the method according to the present invention has the additional advantage that, although an error rate may exist due to the characteristics of 2D grid simulation, the area where an error occurred can be easily identified visually through 3D visualization, making it easy to re-execute with adjusted variables. This solves the problem of difficulty in identifying errors in conventional text-based result output.

[0172] Next, with reference to FIG. 3, the specific mechanism of the double-drainage coupling, which is the most essential technical feature of the present invention, will be explained.

[0174] FIG. 3 is a conceptual diagram showing a bidirectional water exchange mechanism between a two-dimensional surface runoff (GPU computation) and a one-dimensional sewer network (CPU / SWMM computation) according to the present invention, i.e., a double drainage coupling structure.

[0176] Referring to FIG. 3, the drawing is composed of two sections, upper and lower. The upper section is a "surface (2D, GPU computation)" area in which a 2D surface grid (210) is arranged, and the lower section is a "sewer network (1D, CPU / SWMM computation)" area in which a 1D sewer network (220) is arranged.

[0178] First, the two-dimensional surface grid (210) at the top is described. The flood simulation of the present invention begins by dividing the target area into a grid in the form of a checkerboard, and this grid is the two-dimensional surface grid (210). Each cell of the grid serves as a surface cell (211) and becomes the minimum unit of the simulation. Physical quantities such as elevation above sea level, water height, flow velocity in the X direction, and flow velocity in the Y direction are stored in each surface cell (211), and hundreds of thousands of cells are processed simultaneously in parallel by the compute shader of the GPU. In this embodiment, the size of the grid is 512×512 cells, that is, it consists of a total of 262,144 cells.

[0180] Next, the one-dimensional sewer network (220) at the bottom is described. The one-dimensional sewer network (220) is an underground pipe network that is computed by the CPU using the SWMM engine. The sewer network (220) consists of three key elements: a junction (221), a conduit (222), and a discharge port (223).

[0182] The junction point (221) is a point in the sewer network where multiple pipes meet or diverge, corresponding to a manhole in the real world. The junction point (221) has a depth from the surface to the invert elevation and is a passage that vertically connects the surface and the underground pipe network. In the simulation, the junction point (221) is a key node where the inflow and outflow of water are determined, serving as a reference point for calculating the phenomenon where rainwater from the surface flows into the sewer pipe, or conversely, the water level in the sewer pipe rises, causing water to backflow out of the manhole and induce surface flooding.

[0184] The conduit (222) is a sewer pipe connecting the junction (221) and the junction (221) or the junction (221) and the discharge port (223), forming a path for the movement of water. In this embodiment, the conduit (222) is simulated in a circular cross-sectional shape, and the flow velocity, flow rate, and water level inside the pipe are calculated by the dynamic wave method within the SWMM engine.

[0186] The discharge port (223) is the final outlet through which water collected via the sewer network (220) is discharged into an external water system, such as a river. Unlike the junction point (221), the discharge port (223) is not modeled as a separate physical structure and functions as a passage connecting the sewer network system and the surface system. In the simulation, the coordinates of the surface cell where the discharge port (223) is located are determined, and the flow rate (m³ / s) of the conduit (222) connected to the discharge port (223) is calculated and treated as a source supplying water to the corresponding surface cell.

[0188] Now, the bidirectional water exchange mechanism between the upper surface area and the lower sewer network area, which is the core of Fig. 3, will be explained.

[0190] In FIG. 3, a specific surface cell (211) of a two-dimensional surface grid (210) and a junction point (221) of a one-dimensional sewer network (220) are connected by a vertical dotted line. This vertical dotted line represents a structure in the real world where a manhole vertically connects the surface and the underground network. At this connection point, bidirectional water exchange occurs.

[0192] First, the inflow direction arrow (A1) indicates the direction in which water flows from the surface cell (211) to the junction point (221). In FIG. 3, it is indicated by a downward straight arrow, and next to it is labeled "Inflow: Surface water level > Junction point water level". This represents the flow of water that occurs when the water level of the corresponding surface cell (211) is higher than the water level at the junction point (221). In other words, it simulates the phenomenon where rainwater from the surface naturally flows into the sewer pipe through the manhole when there is little rain or when there is spare capacity in the sewer pipe.

[0194] Second, the outflow direction arrow (A2) indicates the direction in which water flows backward from the junction point (221) to the surface cell (211). In FIG. 3, it is indicated by an upward straight arrow, and next to it, it is labeled "Backflow: Junction point water level > Surface water level." This is the flow of water that occurs when the water level in the sewer network rises rapidly due to heavy rainfall, etc., causing the water level at the junction point (221) to be higher than the water level at the surface cell (211). In other words, it simulates the phenomenon where, if rainfall exceeding the drainage capacity of the sewer pipe occurs, sewage flows backward to the surface through the manhole, causing urban flooding. Accurate simulation of this backflow phenomenon is one of the most critical factors in urban flood prediction.

[0196] Third, the discharge direction arrow (A3) indicates the direction in which water is discharged from the discharge port (223) to the surface cell (211). In FIG. 3, it is indicated by an upward straight arrow pointing from the discharge port (223) toward the corresponding cell of the upper grid, and next to it is labeled "Discharge: Conduit flow rate → Surface". This simulates the phenomenon of water flowing through the sewer network being discharged to the surface through the discharge port (223). In the simulation, the amount of water discharged during a given time is calculated by multiplying the flow rate (m³ / s) of the discharge port (223) connecting conduit (222), calculated by the SWMM engine, by the time interval (dt), and this is added as the water height to the surface cell where the discharge port (223) is located.

[0198] In order to perform such bidirectional water exchange, information exchange between 2D surface data computed on the GPU and 1D sewer network data computed on the CPU is essential. The GPU→CPU data copy arrow (A4) and CPU→GPU data reflection arrow (A5) shown on the right side of Fig. 3 represent this process.

[0200] Specifically, in the GPU→CPU data copying process (A4), 2D grid data (water level, elevation, etc. of each cell) of 512×512 cells stored in the GPU memory in the form of a texture is copied to the CPU memory. Once this copying is complete, the CPU iterates through all junction points (221) of the SWMM engine, reads the water level of the surface cell (211) where each junction point is located, and determines the inflow (A1 direction) or outflow (A2 direction) by comparing it with the water level of the corresponding junction point. At the same time, for all discharge points (223), the flow rate of the connected conduit (222) is calculated to determine the amount of water to be reflected in the corresponding cell.

[0202] Next, in the CPU→GPU data reflection (A5) process, when the result of the coupling operation, that is, the inflow or outflow at each junction point (221) and the discharge amount at each discharge port (223), is determined, this result is reflected in the GPU memory. The reflected data is subsequently included in the calculation of the water level of each cell as the d_ext(t) term in the water level update step (S540, described later in FIG. 4).

[0204] This bidirectional water exchange mechanism is explained with a simple example. It is assumed that light rainfall of 30 millimeters per hour occurs in the simulation target area. In this case, the water level accumulating in the surface cell (211) is low, and the water level at the manhole, which is the junction point (221), is also low. Since the water level in the surface cell is higher than the water level at the junction point, water flows into the sewer pipe through the manhole according to the inflow direction arrow (A1). The water flowing through the sewer pipe passes through the conduit (222) and is discharged into the river through the discharge port (223). Under this situation, urban flooding does not occur.

[0206] On the other hand, it is assumed that heavy rainfall of 100 millimeters per hour occurs in the same area. Initially, rainwater flows into the sewer pipe through the manhole as described above, but if the inflow continues to exceed the drainage capacity of the sewer pipe, the water level at the junction point (221) rises rapidly. At some point, when the water level at the junction point (221) rises above the water level of the surface cell (211), the direction of water flow is reversed, and sewage flows back to the surface through the manhole according to the outflow direction arrow (A2). The backflowed water spreads outward according to the terrain (elevation, building height, etc.) above the surface, causing urban flooding. The dual drainage model of the present invention accurately reproduces the actual mechanism of urban flooding by repeatedly calculating this bidirectional water exchange for all junction points at every hourly interval.

[0208] A notable feature of the double-drainage coupling structure illustrated in Fig. 3 is that the 2D surface runoff operation is performed on the GPU, while the 1D sewer network operation is performed on the CPU. This is because the characteristics of the two operations are different. Since the 2D surface runoff operation involves simultaneously applying the same physical equation to hundreds of thousands of grid cells, it is suitable for large-scale parallel processing on the GPU. On the other hand, since the 1D sewer network operation is structured so that the SWMM engine sequentially traverses each node of the network to perform calculations, sequential processing on the CPU is suitable. Through this collaborative structure between heterogeneous computing devices, the present invention maximizes the advantages of each computing device while organically linking the two through GPU→CPU data copying (A4) and CPU→GPU data reflection (A5).

[0210] In addition, the 2D surface runoff calculation and the 1D sewer network calculation are performed at different time intervals. In this embodiment, the time interval for the 2D surface runoff calculation (first time interval) is 0.25 seconds, and the time interval for the 1D sewer network calculation (second time interval) is 2 seconds. Therefore, the 1D sewer network calculation is synchronized so that it is performed once for every 8 times the 2D surface runoff calculation is performed. This asymmetric time interval design is based on the hydraulic characteristic that surface runoff is a rapidly changing phenomenon and is therefore calculated at short time intervals, whereas the flow inside the sewer network changes relatively slowly and can be calculated at long time intervals to ensure sufficient accuracy. Through this, unnecessary calculations are reduced, thereby increasing the efficiency of the overall simulation.

[0211] Next, with reference to FIG. 4, the detailed steps of the simulation loop that are repeatedly performed within the real-time simulation execution step (S500) are described.

[0213] FIG. 4 is a flowchart showing in detail the internal iteration loop of the real-time simulation execution step (S500) according to the present invention.

[0215] Referring to FIG. 4, the simulation loop includes a hydrological step (S510), a drainage interlocking step (S520), a flow velocity calculation step (S530), a water level update step (S540), and a time interval determination step (S550), and these steps are repeated at every time interval until a termination condition is satisfied by a simulation termination determination (S560). As this loop is repeated tens of times per second, the entire process of rainwater accumulating on the surface, flowing through sewer pipes, and the flooded area spreading over time is simulated in real time.

[0217] First, in the hydrological stage (S510), the effective rainfall supplied to each cell by rainfall is calculated to assign a water height. In this stage, the infiltration amount of each cell is calculated by applying the HOLTON infiltration formula based on cover data, and the water height due to effective rainfall is assigned to each cell by subtracting the infiltration amount from the rainfall amount.

[0219] Specifically, the penetration capacity (inf) for each cell is calculated as follows: inf = minimum penetration rate + ((maximum penetration rate - minimum penetration rate) × e^(-K × simTime / 60)). Here, the minimum penetration rate and the maximum penetration rate are values ​​read from the coverage data of the corresponding cell, K is the attenuation coefficient, and simTime is the simulation operation time (seconds). The attenuation coefficient K is calculated as (MIN_K - MAX_K) × penetration efficiency + MAX_K, where in this embodiment, MIN_K is set to 2, MAX_K to 9, and penetration efficiency to 1.0. Therefore, K = (2 - 9) × 1.0 + 9 = 2.

[0221] Once the infiltration capacity is calculated, the new water level in each cell is calculated as follows: h_new = water level + (rain_val - inf) × dt / 3600 / 1000, where rain_val is the precipitation value (mm / h) entered by the user, inf is the infiltration capacity (mm / h) calculated above, and dt is the current time interval (seconds). Dividing by 3600 converts the time unit to seconds, and dividing by 1000 converts the millimeter unit to meters. As a result of this calculation, if the value of (rain_val - inf) is positive, it means that rainwater exceeds the infiltration amount and accumulates on the surface; if it is negative, it means that all the rainwater seeps into the ground and does not accumulate on the surface.

[0223] This is explained with specific numerical examples. Immediately after the simulation starts (simTime Assume that rainfall of 100 millimeters per hour is occurring, and that the minimum infiltration rate of a certain cell is 5 mm / h and the maximum infiltration rate is 20 mm / h. At the beginning of the simulation, e^(-K × 0 / 60) = 1, so inf = 5 + (20 - 5) × 1 = 20 mm / h. Therefore, the effective rainfall is 100 - 20 = 80 mm / h, and if dt is 0.25 seconds, the height of water added to this cell is 80 × 0.25 / 3600 / 1000 It becomes 0.0000056 meters, or approximately 0.0056 millimeters. If the simulation has elapsed 30 minutes (simTime = 1800 seconds), e^(-2 × 1800 / 60) = e^(-60) Since it is 0, the infiltration capacity converges to the minimum value as inf = 5 + (20 - 5) × 0 = 5 mm / h. At this point, the effective rainfall increases to 100 - 5 = 95 mm / h, causing more water to accumulate on the surface. This accurately reflects the real-world phenomenon where, when it rains for a long time, the ground becomes saturated, reducing infiltration and increasing surface runoff.

[0225] Next, in the drainage coupling step (S520), the 2D surface runoff calculation and the 1D sewer network calculation are time-synchronized to perform bidirectional double drainage coupling. In FIG. 4, the drainage coupling step (S520) is enclosed in a dotted box to indicate that the synchronization judgment and detailed sub-steps are included within this step.

[0227] The internal flow of the drainage interlocking step (S520) is as follows. First, in the 2D-1D synchronization determination (S521), it is determined whether the cumulative number of 2D surface runoff operations performed so far satisfies the synchronization condition with the 1D sewer network operation. This corresponds to the notation "2D-1D synchronization condition satisfied?" inside the diamond symbol (S521) in Fig. 4. In this embodiment, since the time interval (first time interval) of the 2D surface runoff operation is 0.25 seconds and the time interval (second time interval) of the 1D sewer network operation is 2 seconds, the synchronization condition is satisfied every time the 2D operation is performed 8 times.

[0229] If the synchronization condition is not met (the "No" branch in Fig. 4), the drainage interlocking step (S520) is skipped and the process proceeds directly to the flow velocity calculation step (S530). This means that in the corresponding loop cycle, only surface runoff calculations are performed, and sewer network calculations are not performed.

[0231] When the synchronization condition is satisfied (the "Yes" branch in Fig. 4), the following three sub-steps are performed sequentially.

[0233] First, in the SWMM engine step 1 execution (S522), the swmm_step function of the SWMM engine compiled into WebAssembly is called to advance the 1D sewer network simulation by a second time interval (2 seconds). During this process, the SWMM engine internally calculates the routing step and updates the water level at all junctions, and the flow rate and velocity of all conduits.

[0235] Second, in the apply_linkage execution (S523), the 2D grid data of 512×512 cells stored in GPU memory is copied to CPU memory, and then bidirectional water exchange is determined by iterating through all junctions of the SWMM engine. Specifically, for each junction, the water level of the surface cell where the junction is located is compared with the water level of the junction. If the water level of the surface cell is higher than the water level of the junction, it is determined that water flows from the surface to the junction, so the water in the cell decreases and the water level of the junction rises. Conversely, if the water level of the junction is higher than the water level of the surface cell, it is determined that water flows out (backflow) from the junction to the surface, so water is added to the cell and the water level of the junction falls. This process is the actual operation of bidirectional water exchange corresponding to the inflow direction arrow (A1) and outflow direction arrow (A2) described in FIG. 3.

[0237] Third, in the outfall_step execution (S524), the flow rate (Flow, m³ / s) of the conduit connected to all outfalls of the SWMM engine is calculated. The calculated flow rate is multiplied by the time interval (dt) to calculate the amount of water discharged through the outfalls during that time, and this is converted into a water height to be reflected in the surface cell where the outfalls are located. This process is the calculation of the surface outflow of the pipe discharge water corresponding to the discharge direction arrow (A3) described in Fig. 3.

[0239] The amount of water level change for each cell determined in apply_linkage(S523) and outfall_step(S524) is not reflected immediately, but is reflected collectively in the subsequent water level update step (S540). This is a design intended to maintain numerical consistency in the simulation.

[0241] The flow of this drainage linkage step (S520) is explained with specific numerical examples. Assume that the simulation proceeds and the 2D surface runoff calculation is performed for the 8th time (simulation time = 2 seconds). At this point, since the synchronization condition is satisfied (S521, "Yes"), drainage linkage is performed. Assume that after the SWMM engine executes Step 1 (S522), it iterates through a specific linkage point J1 in apply_linkage (S523). If the water level of the surface cell where J1 is located is 0.05 meters (5 centimeters) and the water level of J1 is -2 meters from the surface (i.e., the low water level near the bottom of the pipe), water flows from the surface into J1 because the surface water level is higher than the linkage point water level. On the other hand, if heavy rain continues and the water level of J1 exceeds the height of the ground surface, for example, if the water level of J1 is 0.1 meters higher than the ground surface, the water flows back from J1 to the ground surface, and the water level of the corresponding cell increases.

[0243] Next, in the flow velocity calculation step (S530), the water flow velocity is calculated for each cell. The calculation of the flow velocity requires not only the water height, elevation, and flow velocity of the corresponding cell, but also the water height, elevation, and flow velocity values ​​of the four adjacent cells above, below, left, and right.

[0245] The physical principles of flow velocity calculation are explained. Water has a tendency to flow from a higher total height (elevation above sea level + water level) to a lower total height. Therefore, the difference in total height between a cell and an adjacent cell, i.e., the slope, becomes a key factor in determining the flow velocity of water. In this invention, the flow velocity is calculated based on the momentum equation of the shallow water equation, while ensuring numerical stability by applying the weighting parameter θ = 0.9 of the semi-implicit scheme. Additionally, the friction slope formula Sf = n²·v|v| / h^(4 / 3), which considers the Manning roughness coefficient, is applied to differentially reflect the surface frictional resistance according to the coverage (roads, green spaces, surroundings of buildings, etc.) of each cell.

[0247] The average flow velocity of adjacent cells is also considered during the flow velocity calculation process. This is intended to reflect the physical phenomenon where the flow velocity of a given cell is influenced by interaction with surrounding flow patterns. Additionally, the direction of the flow velocity is determined by the difference in water level with adjacent cells, and the flow velocities in the X direction (east-west) and Y direction (north-south) are calculated independently.

[0249] This flow velocity calculation is explained with a simple example. Assume there are three grid cells A, B, and C arranged horizontally, and that the elevation of all three cells is the same at 10 meters above sea level. Assume that water exists at a height of 0.5 meters only in cell B, while there is no water in cells A and C. In this case, the total height of cell B is 10.5 meters (elevation 10 + water height 0.5), and the total height of cells A and C is 10 meters. Since there is a height difference (slope) of 0.5 meters in the direction from B to A and from B to C, the water tends to spread out to the sides from B due to the influence of gravity. The GPU performs these calculations simultaneously for all cells in the 512×512 grid.

[0251] After that, in the water level update step (S540), the water level of each cell is updated based on the external inflow amount determined in the drainage interlocking step (S520) and the flow velocity calculated in the flow velocity calculation step (S530). The formula applied in this step is d(t+Δt_2D) = d(t) + d_ext(t) + ΣQ / (ΔxΔy)·Δt_2D.

[0253] Each term of this formula is explained. d(t) is the water depth of the cell at the current time step. d_ext(t) is the external inflow amount determined in the drainage linkage step (S520), which is the sum of the inflow or outflow amount at the junction point by apply_linkage (S523) and the discharge amount at the outlet by outfall_step (S524). This term is the key path through which the result of the double drainage coupling is reflected in the 2D surface runoff calculation. If the value of d_ext(t) is positive, water is added from the sewer network to the surface (backflow or discharge), and if it is negative, water is out of the surface into the sewer pipe (inflow). In a loop cycle where the drainage linkage step is not performed, d_ext(t) = 0. The term ΣQ / (ΔxΔy)·Δt_2D is calculated by dividing the sum of the flow rates (Q) with adjacent four-directional cells calculated in the flow rate calculation step (S530) by the cell area (ΔxΔy) and multiplying by the time interval (Δt_2D), and represents the amount of change in water depth due to horizontal water exchange with adjacent cells. If the sum of the flow rates is positive, water flows in from surrounding cells and the water depth increases, and if it is negative, water flows out to surrounding cells and the water depth decreases.

[0255] Continuing from the previous example of the flow velocity calculation step, the water level update step is entered after the flow velocity spreading outward from cell B to both sides (A, C) has been calculated. Since water is flowing out of cell B on both sides, the ΣQ value becomes negative, causing the water depth to decrease. For example, if d(t) was 0.5 meters, it may drop to 0.48 meters in the next time step. Conversely, since cells A and C are receiving water from cell B, the ΣQ value becomes positive, causing the water depth, which was previously 0, to increase to 0.01 meters. As this calculation is repeated for all cells at every time interval, complex flooding phenomena are naturally implemented, where water flows along the terrain to lower areas, pools in puddles, and bypasses buildings.

[0257] Finally, in the time interval determination step (S550), the time interval of the next loop cycle is dynamically determined based on the CFL (Courant-Friedrichs-Lewy) condition. The formula applied is Δt_2D = a × min(Δx, Δy) / √(g × d_max). Here, a is the CFL safety factor (a constant less than or equal to 1), min(Δx, Δy) is the minimum side length of the grid cell, g is the acceleration due to gravity (approx. 9.81 m / s²), and d_max is the deepest water depth in the entire current simulation area.

[0259] The physical meaning of this equation is explained. In the shallow water equation, water waves propagate at a speed of √(g × d). The CFL condition ensures that water waves do not travel beyond a single grid cell during a single time interval. If there are cells with very deep water, the wave propagation speed increases, so the time interval must be reduced to prevent the simulation from diverging. Conversely, if the water depth is generally shallow, a larger time interval can be used, improving computational efficiency.

[0261] For example, at the beginning of the simulation, when there is almost no water on the surface (d_max is very small), the time interval is maintained at the initial value of 0.25 seconds. However, if heavy rainfall persists and water accumulates to a depth of more than 1 meter in specific low-lying areas, √(9.81 × 1.0) A wave propagation speed of 3.13 m / s occurs, and the time interval can be automatically adjusted to a value smaller than 0.25 seconds, for example, 0.1 seconds, by CFL conditions. This slightly reduces the speed of the simulation but ensures numerical stability, preventing the simulation from outputting unrealistic values ​​or diverging.

[0263] Once the time interval is determined, the process enters the simulation termination determination (S560). This corresponds to the "Simulation Termination" labeled inside the diamond symbol (S560) in FIG. 4. If the total simulation time set by the user is reached or the user requests to stop the simulation, the loop terminates according to the "Yes" branch. Otherwise, it returns to the hydrology step (S510) according to the "No (Repeat)" branch and starts the next loop cycle.

[0265] In conclusion, the simulation loop illustrated in FIG. 4 forms an organically connected five-stage cyclic structure, which calculates the effective rainfall amount due to rainfall in the hydrological stage (S510), performs bidirectional water exchange between the surface and the sewer network in the drainage coupling stage (S520), calculates the flow velocity of each cell in the flow velocity calculation stage (S530), updates the water depth of each cell by synthesizing all results in the water height update stage (S540), and dynamically determines the time interval of the next cycle in the time interval determination stage (S550). Through the repeated execution of this loop, the occurrence, spread, and drainage processes of urban flooding over time are simulated in real time based on physical laws. In particular, the bidirectional double drainage coupling performed in the drainage coupling stage (S520) executes the mechanism described in FIG. 3 at every synchronization point, thereby precisely reproducing the actual mechanism of urban flooding in which water flows above and below ground mutually influence each other.

[0266] FIG. 5 is a diagram showing the two-dimensional grid and RGBA texture data structure used in the numerical analysis unit (140) of the present invention.

[0268] In the present invention, the two-dimensional surface runoff model performs calculations on a grid (230) that divides the simulation target area into 512 horizontal and 512 vertical square cells, totaling 262,144 cells. Each cell corresponds to a real-world coordinate system, and the size of a single cell corresponds to the value obtained by dividing the total area of ​​the target area by the number of grid cells. For example, if the target area is 5.12 km in width and length, a single cell represents a real-world area of ​​approximately 10 m × 10 m.

[0270] In each cell, four physical quantities required for the simulation are stored in a single RGBA texture pixel (231). Specifically, the X-direction flow velocity (m / s) of the cell is stored in the R channel (232), the Y-direction flow velocity (m / s) is stored in the G channel (233), the water height (m) of the cell is stored in the B channel (234), and the elevation (m) of the cell above sea level is stored in the A channel (235). At this time, all values ​​are recorded in a floating-point (float32) format in their real-world physical units without being normalized to a range of 0 to 1. This is to eliminate the loss of precision and computational overhead that occurs in the process of normalizing texture values ​​to 0 to 1 and then converting them back in conventional GPU-based simulations.

[0272] By adopting this RGBA texture structure, the GPU compute shader can acquire four physical quantities simultaneously with a single texture fetch command. In the conventional method, flow velocity, water height, and elevation must each be stored in separate buffers or textures and read individually, requiring at least four memory accesses; however, the present invention reduces this to one, thereby reducing memory bandwidth usage by approximately 75%.

[0274] The entire arrangement of the grid (230) is displayed in the left area of ​​FIG. 5, and a detailed view of an arbitrary cell is displayed on the right. In the detailed view, the RGBA texture pixel (231) of the cell is separated into four channels (232 to 235), and the name and unit of the physical quantity corresponding to each channel are indicated.

[0276] To explain through a specific example, assume that at a specific point in time, the RGBA values ​​of the cell located at the coordinates (i=256, j=256) of the grid (230) are as follows. If R channel (232) = 0.35, it means that a flow velocity of 0.35 m / s is occurring in the X direction of the cell, and if G channel (233) = -0.12, it means that a flow velocity of 0.12 m / s is occurring in the Y direction, that is, in the south. If B channel (234) = 0.45, it means that there is water at a depth of 0.45 m above the surface, and if A channel (235) = 12.8, it means that the elevation of the cell is 12.8 m above sea level. From this, the water level elevation of the cell is calculated as 13.25 m, which is the sum of the A channel value and the B channel value.

[0278] In each stage of the simulation loop, namely the hydrology stage (S510), the flow velocity calculation stage (S530), and the water level update stage (S540), the GPU compute shader reads RGBA texture pixels (231) in parallel for all cells of the grid (230), performs calculations according to the aforementioned formulas, and then writes the updated values ​​back to the same texture. At this time, a double buffering technique is applied to prevent read-write conflicts by writing to the texture of the previous time stage while reading the texture of the current time stage.

[0280] In addition, when data transfer between the GPU and CPU is required during the multiplier linkage step (S520), the entire RGBA texture is copied into a single contiguous memory block, so the transfer efficiency is higher compared to transferring individual variables sporadically. Specifically, 512 × 512 × 4 channels × 4 bytes (float32) = approximately 4 MB of data is transferred in a single GPU→CPU copy command.

[0282] In addition, the surface roughness coefficient (Manning's n) and infiltration-related coefficients (minimum infiltration rate, maximum infiltration rate) derived from the cover map are stored in a separate auxiliary texture and are referenced together with the main texture (231) when the compute shader performs calculations. Thus, the numerical analysis unit (140) can perform a flood simulation with high physical precision that reflects different surface characteristics on a cell-by-cell basis for the entire grid (230).

[0284] As explained above, the two-dimensional grid and RGBA texture data structure illustrated in FIG. 5 corresponds to a core data architecture that enables the numerical analysis unit (140) of the present invention to process real-world physical quantities without loss while making full use of the advantages of GPU parallel processing.

[0285] FIG. 6 is a diagram showing the operation flow and interrelationships of key formulas performed by the GPU compute shader in the numerical analysis unit (140) of the present invention.

[0287] The numerical analysis unit (140) of the present invention is at each time step A number of formulas are applied sequentially and in parallel to all cells of the grid (230). FIG. 6 schematically shows the order in which these formulas are called and how the inputs and outputs of each formula are connected.

[0289] First, the shallow-water equations, denoted by Equation 1, are placed at the top of Fig. 6. The shallow-water equations consist of two systems of partial differential equations: the first is the continuity equation regarding the rate of change of water height, and the second is the momentum equation regarding the rate of change of flow velocity. The continuity equation is as follows.

[0290]

[0291] Here, η is the water height (m) of the corresponding cell, and is the flow velocity vector (m / s). This equation represents the conservation of mass of water and implies that the water level changes due to the difference between the flow rate entering the cell and the flow rate exiting it.

[0292] The momentum equation is as follows.

[0293]

[0294] Here, g is the acceleration due to gravity (9.81 m / s²), and h is the elevation of the water surface, that is, the sum of the elevation above sea level and the water level. This equation represents the relationship in which the slope of the water surface due to gravity changes the flow velocity. Since water flows from a high place to a low place, the flow velocity accelerates in the direction opposite to the direction of the slope (∇h).

[0295] In the present invention, a semi-implicit scheme is applied when discretizing the above shallow water equation, and a weight θ = 0.9 is used. This is because it is suitable for real-time simulation as it offers higher numerical stability compared to the fully implicit scheme (θ = 0) and lower computational cost compared to the fully implicit scheme (θ = 1). Specifically, the flow velocity update from time t to t + Δt₂D is performed as follows.

[0296]

[0297] Here, F is a right-hand side function containing advection and gravitational gradient terms. Since θ = 0.9, a weight of 90% is applied to the value of the new time step and a weight of 10% to the value of the current time step to ensure a balance between stability and accuracy.

[0299] In the middle left of Fig. 6, the Manning Friction Slope formula, denoted by Equation 2, is placed. The Manning Friction Slope represents the flow velocity attenuation caused by surface roughness and is defined as follows.

[0300]

[0301] Here, n is Manning's roughness coefficient, which is set on a cell-by-cell basis from the cover data. For example, in the case of an asphalt road, n It is 0.015, and if it is a lawn, n It is 0.035, and in the case of a dense forest area, n It has a value of 0.10 to 0.15. h is the water depth (m) of the corresponding cell. Friction slope is added as a damping term to the right-hand side of the momentum equation (Equation 1b) to reduce the flow velocity. As the water depth h decreases, the denominator becomes smaller, increasing the frictional effect, which is consistent with the physical phenomenon that the influence of surface friction is dominant in shallow water.

[0303] A water level update formula denoted by Equation 3 is placed on the right side of the middle of Fig. 6. This formula is applied in the water level update step (S540) and is as follows.

[0304]

[0305]

[0306]

[0307] In other words, it is possible to quantitatively simulate the pattern in which the water level increases due to backflow in the sewer network, but is partially offset by outflow to adjacent cells.

[0309] In the lower left of Fig. 6, a CFL-based adaptation time interval determination formula denoted by Equation 4 is placed. This formula is applied in the time interval determination step (S550) and is as follows.

[0310]

[0311] Here, a is the Courant safety factor, having a value of 1 or less, and in one embodiment of the present invention, a = 0.5 is applied. dmax is the maximum water depth (m) among all cells of the grid (230). This formula is based on the Courant-Friedrichs-Lewy (CFL) condition and serves to automatically adjust the time interval so that a wave does not propagate beyond a single cell during one time step. For example, if the cell size is 10 m and the maximum water depth is 1.0 m, Δt₂D = 0.5 × 10 / √(9.81 × 1.0) It becomes 1.60 s, but if the maximum water depth increases to 4.0 m, Δt₂D = 0.5 × 10 / √(9.81 × 4.0) It decreases to 0.80 s. As the wave propagation speed increases as the immersion depth increases, the time interval is automatically reduced to prevent numerical divergence. The initial time interval of the present invention is set to approximately 0.25 s and is dynamically adjusted as the simulation progresses.

[0312] In the lower right corner of Fig. 6, the Holton Infiltration Formula, denoted by Equation 5, is placed. This formula is applied in the hydrological stage (S510) and is as follows.

[0313]

[0314]

[0315]

[0316] In one embodiment of the present invention, applying MIN_K = 2, MAX_K = 9, and efficiency = 1.0 results in K = (2 - 9) × 1.0 + 9 = 2. This simulates a phenomenon where, at the beginning of the simulation, the surface is dry and the infiltration capacity is close to the maximum infiltration rate, but as time passes and the soil becomes saturated, the infiltration capacity decreases exponentially and eventually converges to the minimum infiltration rate. For example, if the minimum infiltration rate of a cell is 5 mm / h and the maximum infiltration rate is 30 mm / h, the infiltration capacity 1 hour after the start of the simulation is f = 5 + (30 - 5) × e^(-2×1) = 5 + 25 × 0.135 It becomes 8.38 mm / h. If the rainfall intensity at that time is 50 mm / h, water equivalent to 50 - 8.38 = 41.62 mm / h that did not infiltrate remains on the surface and leads to runoff. In the cover data, asphalt paved areas have minimum and maximum infiltration rates close to zero, so most of the rainfall is converted into surface runoff, whereas green areas or soil areas have relatively high infiltration rates, reducing the amount of runoff.

[0317]

[0318] As explained above, the formula system illustrated in FIG. 6 constitutes a mathematical basis for the numerical analysis unit (140) of the present invention to reproduce urban flooding phenomena in a manner that is physically accurate and suitable for GPU parallel processing.

[0319] FIG. 7 is a diagram showing the process of collecting and combining satellite image tiles performed by the terrain data collection unit (110) of the present invention.

[0321] In the present invention, the accuracy of the two-dimensional surface runoff simulation depends directly on the quality of the terrain data of the target area. The terrain data collection unit (110) automatically collects satellite images, a digital elevation model (DEM), a digital surface model (DSM), a cover map, and river data required for the simulation to generate an integrated dataset corresponding to the grid (230). FIG. 7 describes the procedure for collecting and combining satellite image tiles during this process.

[0323] A WMTS (Web Map Tile Service) server (240) is shown at the top of FIG. 7. The WMTS server (240) is a standard web service that provides satellite images of the world in a tile format by pre-dividing them by zoom level. In one embodiment of the present invention, zoom level 16 is used, in which case the entire Earth is divided into approximately 65,536 × 65,536 tiles, and each individual tile corresponds to a real-world area of ​​approximately 610 m × 610 m.

[0325] When the user specifies the center coordinates and range of the simulation target area, the terrain data collection unit (110) calculates the X and Y indices of the tiles containing the area. In FIG. 7, this process is indicated as the tile index calculation step (241). For example, if the target area is the Gangnam-gu area of ​​Seoul (latitude approx. 37.50°, longitude approx. 127.03°) and the simulation range is approximately 5 km in width and height, then at zoom level 16, approximately 9 tiles in width and approximately 9 tiles in height, totaling approximately 81 tiles, are required.

[0327] When a tile index is calculated, the terrain data collection unit (110) sequentially requests individual tile images from the WMTS server (240). This process corresponds to the tile download step (242) of FIG. 7. Each tile request URL includes a zoom level, an X index, and a Y index as parameters. For example, in the case of a tile with a zoom level of 16, X = 54321, and Y = 27180, a 256 × 256 pixel satellite image corresponding to the coordinates is returned. At this time, to prevent server overload, the request interval between tiles is set to approximately 0.1 seconds. The total download time for 81 tiles is approximately 8 to 10 seconds.

[0329] The downloaded individual tiles are merged into a single continuous high-resolution image in the tile merging step (243) of FIG. 7. Since the 9 × 9 tiles are each 256 × 256 pixels, the total resolution of the combined image becomes 2,304 × 2,304 pixels. During merging, precise coordinate alignment at the pixel level is performed so that no seams occur at the boundaries between adjacent tiles. The combined satellite image is used as a texture for 3D terrain rendering in the visualization and interaction unit (160), allowing the user to observe the simulation results while intuitively checking the appearance of the actual terrain.

[0331] At the bottom of FIG. 7, a combined image post-processing step (244) is illustrated. In the post-processing step (244), the combined satellite image is resampled to match the resolution of the grid (230), i.e., 512 × 512 pixels. During resampling, bilinear interpolation is applied to minimize information loss during the resolution reduction process. This process is applied equally to DEM, DSM, cover map, and river raster data as well as to satellite images.

[0333] The data collection and preprocessing process other than satellite imagery is described below as an additional explanation of Fig. 7.

[0335] In the case of a digital elevation model (DEM), the terrain data collection unit (110) receives the original DEM file (e.g., GeoTIFF format), cuts it into a range of latitude and longitude of the target area using the gdal_translate command, and resamples it into 512 × 512 cells. At this time, bilinear interpolation is applied, and the output file is saved as a 32-bit floating-point GeoTIFF. Each cell value represents the elevation (m) of the corresponding point, and this value is directly recorded in the A channel (235) of the RGBA texture pixel (231).

[0337] In the case of the Digital Surface Model (DSM), the building outline (SHP format) is input and rasterized using the gdal_rasterize command. The average building height (in one embodiment of the present invention, the average floor height of 2.4 m multiplied by the number of floors) is assigned to cells where buildings exist, and 0 is assigned to cells where no buildings exist. The DSM value is added to the DEM to reflect the effect of blocking water flow caused by the building. For example, if the DEM of a cell is 15.0 m and a 5-story building exists at that location, DSM = 2.4 × 5 = 12.0 m is added, making the effective elevation of that cell 27.0 m. As a result, during the simulation, water flows around that cell, and the impact of a building on the runoff path in a real urban environment can be realistically simulated.

[0338]

[0339] For river data, the river centerline (SHP format) is input and rasterized, and the initial water level is set in the cells corresponding to the river. This allows for the reflection of the state in which water already exists in the river from the start of the simulation, and enables the continuous simulation of the river water level rise and flooding processes caused by rainfall.

[0341] Once all of the above data is collected and preprocessed, the terrain data collection unit (110) generates an integrated dataset for each cell of the grid (230) that includes elevation (DEM + DSM), Manning roughness coefficient, minimum and maximum infiltration rates, initial river water level, and satellite image texture coordinates. This integrated dataset is transmitted to the WebGPU initialization unit (130) and uploaded as a GPU texture and buffer, after which the numerical analysis unit (140) and the simulation execution unit (150) perform calculations by directly referencing the GPU memory without separate file input / output.

[0343] To summarize the entire process through a specific example, when a user designates the Gangnam-gu area of ​​Seoul as the target area, the terrain data collection unit (110) downloads 81 satellite photo tiles for about 8 to 10 seconds and combines them into a 2,304 × 2,304 pixel image, and separately preprocesses DEM, DSM, cover map, and river data into 512 × 512 cells each, and then composes them into a single integrated dataset and uploads it to the GPU. This entire process is completed within about 15 to 30 seconds, and thereafter, the user can start the simulation immediately by simply inputting the rainfall intensity.

[0345] As explained above, the satellite image tile collection and combination process and the additional terrain data preprocessing pipeline illustrated in FIG. 7 correspond to the core procedure in which the terrain data collection unit (110) of the present invention automatically integrates various spatial data to generate input data in a consistent format required for simulation execution. Through this, the user can prepare for simulation by specifying only the coordinates and range of the target area without having to separately perform preprocessing tasks such as format conversion of raw data or coordinate system alignment.

[0346] FIG. 8 is a diagram showing the detailed flow of a data preprocessing pipeline in which the terrain data collection unit (110) of the present invention converts raw spatial data into a simulation input format.

[0348] Although the process of collecting and combining satellite image tiles was explained in Figure 7, in order to run an actual simulation, multiple spatial data with heterogeneous formats, coordinate systems, and resolutions, such as digital elevation models (DEM), digital surface models (DSM), rivers, and cover maps, in addition to satellite images, must be normalized into a single consistent grid system. Figure 8 illustrates the entire pipeline of this preprocessing process step by step.

[0350] A raw data input unit (251) is positioned on the far left of FIG. 8. The raw data input unit (251) receives five types of spatial data regarding the simulation target area. The first is a digital elevation model (DEM) original file (GeoTIFF format), which is grid-type elevation data provided by the National Geographic Information Institute or an overseas public institution. The second is building outline data (SHP format), which includes planar boundary polygons and floor count attributes for each building. The third is river centerline data (SHP format), which includes the linear path and river width attributes of the river. The fourth is cover map data (SHP to raster format), which includes land use classification codes (roads, buildings, grasslands, forests, water bodies, etc.). The fifth is sewer network data (SWMM INP format), which includes coordinates, specifications, and connection information of junctions, conduits, and outlets. Of these, the sewer network data is processed separately in the sewer network modeling unit (120), so FIG. 8 describes the preprocessing flow of the remaining four types of data.

[0352] The first processing block of FIG. 8 is the coordinate system unification step (252). The raw data may use different coordinate reference systems (CRS). For example, the DEM may be provided in UTM-K (EPSG:5178), the building outline in WGS84 latitude and longitude (EPSG:4326), and the cover map in GRS80 central origin (EPSG:5186). In the coordinate system unification step (252), all data is converted to a single reference coordinate system. In one embodiment of the present invention, WGS84 latitude and longitude (EPSG:4326) is used as the reference coordinate system to ensure compatibility with WMTS tile services and web browser-based map services. The ogr2ogr to gdalwarp commands of the Geospatial Data Abstraction Library (GDAL) are used for coordinate conversion.

[0354] The second processing block of FIG. 8 is the area cutting step (253). In each data set with a unified coordinate system, only the portion corresponding to the latitude and longitude range of the simulation target area (from the bottom-left coordinate to the top-right coordinate) is cut. In the case of a DEM, the raster of the corresponding latitude and longitude range is extracted by specifying the -projwin option to the gdal_translate command, and in the case of vector data (building outline, river centerline, cover map), only the features within the corresponding range are extracted by specifying the -clipdst option to the ogr2ogr command. Through this process, subsequent processing is performed only on the area required for simulation from the entire dataset, thereby reducing unnecessary computation and memory usage.

[0356] The third processing block of FIG. 8 is a raster conversion step (254). In this step, spatial data in vector format is converted into a raster with a resolution of 512 × 512, identical to the grid (230).

[0358] The process of rasterizing building outline data is described in detail as follows. Each building polygon is recorded in a raster cell using the gdal_rasterize command, and the recorded value is set to the building height (m) obtained by multiplying the number of floors of the building by the average floor height (2.4 m in one embodiment of the present invention). For example, 2.4 × 7 = 16.8 m is recorded in a cell where a 7-story building is located. 0 is recorded in a cell where no building exists. This raster corresponds to a Digital Surface Model (DSM) and is added to the DEM value to constitute the effective elevation of each cell.

[0360] The rasterization process of river centerline data is explained as follows. The river alignment is buffered according to the channel width attribute and then rasterized to identify the cells corresponding to the river. An initial water level value is assigned to each river cell, which is set based on the river's normal water level data. At the start of the simulation, this initial water level is recorded in the water height (Channel B) of the river cell, thereby reflecting the state in which water exists in the river prior to rainfall.

[0361]

[0362] The fourth processing block of FIG. 8 is the resolution unification step (255). If the resolution of the rasters generated in the previous step is not 512 × 512, all rasters are resampled to 512 × 512 cells by applying the -ts 512 512 option and the bilinear interpolation (-r bilinear) option to the gdalwarp command. This ensures that the DEM, DSM, river, and cover map rasters correspond exactly one-to-one with the cells of the grid (230). During resampling, bilinear interpolation is applied to the elevation data (DEM, DSM) to maintain smooth elevation changes, and nearest neighbor interpolation is applied to the classification data (cover map) to prevent mixing of classification codes.

[0363]

[0364] When an integrated dataset is generated, the WebGPU initialization unit (130) uploads it to GPU memory in the form of textures and buffers. Subsequently, during the simulation process, the numerical analysis unit (140) and the simulation execution unit (150) reference data directly from GPU memory without CPU-side file input / output, thereby minimizing data access delay.

[0366] The entire pipeline can be summarized as follows through a specific example. It is assumed that the user designates the area of ​​Haeundae-gu, Busan (from latitude 35.15°, longitude 129.12° in the lower left to latitude 35.19°, longitude 129.17° in the upper right) as the target area. In the coordinate system unification step (252), the DEM (UTM-K) and the building outline (GRS80) are converted to WGS84, and in the area cutting step (253), data corresponding to the above latitude and longitude range is extracted. In the raster conversion step (254), the building outline is converted to a DSM raster, the river centerline to an initial water level raster, and the cover map to a counting raster, respectively. In the resolution unification step (255), all rasters are resampled to 512 × 512, and in the data integration step (256), the effective elevation, initial water level, roughness coefficient, infiltration rate, etc. are combined cell by cell and uploaded to the GPU. In the area, low-lying cells near Haeundae Beach have DEM elevations of approximately 2 to 5 m, while high-rise building-dense areas have DSM values ​​reaching tens of m. Beach sand sections have high infiltration rates, while urban asphalt sections have infiltration rates close to zero. These differences in topography and land characteristics at the cell level are reflected in the simulation results, allowing for the reproduction of actual urban flooding patterns where flooding is concentrated in low-lying and impermeable areas even under the same rainfall conditions.

[0368] Meanwhile, the GDAL commands used in the data preprocessing pipeline of the present invention are open-source libraries that can be used without separate licensing costs. It is possible to configure the system to pre-execute on the server side and deliver only the resulting raster to the client (web browser), or to compile into WebAssembly and execute directly within the browser. The present invention does not limit the location of preprocessing execution to a specific configuration and includes both server-side and client-side execution.

[0370] As described above, the data preprocessing pipeline illustrated in FIG. 8 automatically collects, transforms, and integrates multiple raw spatial data with heterogeneous formats, coordinate systems, and resolutions, and systematically performs the process of assigning all physical attributes required for simulation to each cell of the grid (230) in a consistent format. Through this, the user does not need to manually perform complex preprocessing tasks such as format conversion of individual data, coordinate system alignment, and resolution adjustment; instead, the user can complete the preparation for simulation by simply specifying the coordinates of the target area.

[0371] Figure 9 is a graph showing the results of a performance comparison between GPU-based 2D surface runoff operation and CPU-based sequential operation according to the present invention, categorized by grid resolution.

[0373] One of the key technical effects of the present invention is to achieve a significant performance improvement compared to conventional CPU sequential computation through GPU parallel computation utilizing WebGPU. In order to quantitatively present this performance difference, Figure 9 shows the results of measuring the processing time per step of GPU computation and CPU computation under the same simulation conditions while changing the grid resolution to three values: 256 × 256, 512 × 512, and 1024 × 1024.

[0375] The horizontal axis of Fig. 9 represents the grid resolution and is divided into three sections: 256 × 256 (total 65,536 cells), 512 × 512 (total 262,144 cells), and 1024 × 1024 (total 1,048,576 cells). The vertical axis represents the time (milliseconds, ms) required to process one simulation step (one time interval Δt₂D). In each resolution section, bars representing GPU computation results and bars representing CPU computation results are arranged in parallel to enable direct comparison.

[0377] The measurement environment is based on the hardware specifications recommended in one embodiment of the present invention. The CPU is an Intel 12th generation processor (24 cores), the GPU is an NVIDIA GeForce RTX 3060 (3,584 CUDA cores, 12 GB VRAM), and the system memory is 32 GB. The web browser used was the latest version of Google Chrome supporting WebGPU, and the operating system was Windows 11. The simulation conditions were set identically with a rainfall intensity of 50 mm / h, a Manning roughness coefficient n = 0.03 (city average), and HOLTON infiltration parameters MIN_K = 2, MAX_K = 9, and efficiency = 1.0.

[0379] For a 256 × 256 grid (65,536 cells), CPU computation takes approximately 45 to 55 ms per step. The CPU sequentially iterates through the 65,536 cells, applying the shallow water equation (Equation 1), Manning friction (Equation 2), and water level update (Equation 3) to each cell, and determines the CFL time interval (Equation 4) by finding the maximum water depth of all cells. On the other hand, GPU computation takes approximately 3 to 5 ms per step. Since the GPU compute shader divides the 65,536 cells into workgroups and thousands of threads perform computations simultaneously, a speed improvement of approximately 10 to 15 times compared to the CPU is observed.

[0381] For a 512 × 512 grid (262,144 cells), CPU computation takes approximately 180 to 220 ms per step. Since the number of cells has increased fourfold compared to 256 × 256, CPU computation time also increases linearly by roughly four times. In this case, processing just one step takes about 0.2 seconds, making it virtually impossible to achieve real-time visualization (60 FPS, i.e., within about 16.7 ms per frame). On the other hand, GPU computation takes approximately 5 to 8 ms per step. Even though the number of cells has increased fourfold, the increase in GPU computation time is limited to about 1.5 to 2 times, because the GPU's massively parallel processing architecture exhibits sub-linear scaling characteristics with respect to the increase in the number of cells. At this resolution, the speed ratio of GPU to CPU reaches approximately 25 to 35 times, and since the 1-step processing time is less than 16.7 ms, 60 FPS real-time visualization is possible. This is the reason why the basic embodiment of the present invention adopts a 512 × 512 grid.

[0383] For a 1024 × 1024 grid (1,048,576 cells), CPU computation takes approximately 750 to 900 ms per step. Since each step takes nearly 1 second, simulating a 10-minute rainfall scenario (approximately 2,400 steps based on a simulation time of 600 seconds and Δt₂D = 0.25 seconds) requires about 30 to 36 minutes of real-time processing. This makes its application for disaster response or decision support purposes extremely limited. On the other hand, GPU computation takes approximately 15 to 25 ms per step. Simulating the same 10-minute rainfall scenario requires about 36 to 60 seconds of real-time processing. The speed ratio of GPU to CPU reaches approximately 35 to 50 times. However, since the processing time for one step may exceed 16.7 ms, it may be difficult to maintain a stable 60 FPS. In this case, measures such as lowering the visualization frame rate to 30 FPS or adjusting the rendering cycle to once every two steps may be applied.

[0385] The above measurement results are summarized in a table format as follows.

[0386]

[0387] A notable point in the graph of Figure 9 is the tendency for the performance gap between the CPU and the GPU to widen as the grid resolution increases. The speed ratio, which was about 10 to 15 times at 256 × 256, increases to about 35 to 50 times at 1024 × 1024. This is because the computation time of the CPU increases linearly (O(N)) in proportion to the number of cells, whereas the GPU increases quasi-linearly with respect to the increase in the number of cells as long as there is a sufficient number of parallel processing units. In other words, the benefits of GPU parallel computation become more pronounced as the simulation target area expands or the grid resolution increases.

[0389] Meanwhile, the above performance measurement is based solely on 2D surface runoff calculations, and additional computational overhead occurs when double-double coupling is enabled. Specifically, in the double linkage step (S520), GPU→CPU data copying (approx. 4 MB, based on 512 × 512), SWMM engine 1-step execution, apply_linkage (S523) calculation, and CPU→GPU data copying are added. In one embodiment of the present invention, when including double-double coupling based on a 512 × 512 grid, the total time required per step was measured to be approximately 8 to 12 ms. This corresponds to an increase of approximately 3 to 4 ms compared to when coupling is not included (5 to 8 ms), and is still within the 60 FPS real-time visualization standard (16.7 ms). The reason this result is possible is that SWMM calculations are executed only at 2-second intervals (second time intervals) and are not performed at every step, and 1D calculations are performed only once out of 8 2D steps. Therefore, when converted to the average processing time of 8 steps, the average additional overhead per step due to coupling is only about 0.4 to 0.5 ms.

[0391] In addition, since the SWMM engine in the present invention compiles C source code into WebAssembly using EMScripten (EMCC), a performance degradation of about 1.2 to 1.5 times compared to native C execution may occur. However, since the absolute scale of 1D sewer network calculations is smaller than that of 2D surface calculations (hundreds to thousands of junction points versus hundreds of thousands of grid cells) and the calculation cycle is 8 times longer, the impact of this performance degradation on the overall simulation performance is minimal.

[0393] Additional information regarding the performance characteristics of the present invention is described at the bottom of FIG. 9. First, GPU computational performance may vary depending on the GPU model, and additional performance improvement is possible by using a higher-end model (e.g., RTX 4070 to RTX 4090) than the RTX 3060 used in this embodiment. Second, while the performance of the WebGPU API may differ slightly depending on the level of implementation optimization by browser vendors (Google Chrome, Mozilla Firefox, Apple Safari, etc.), basic cross-platform compatibility is guaranteed because the WebGPU standard provides a hardware abstraction layer. Third, although there may be some differences in compute shader performance compared to NVIDIA GPUs in the case of AMD GPUs, parallel computation acceleration effects can be obtained regardless of the GPU vendor because WebGPU operates on native graphics APIs such as Vulkan (Linux), Metal (macOS), and Direct3D 12 (Windows).

[0395] As explained above, the performance comparison results shown in FIG. 9 quantitatively demonstrate that the present invention achieves a performance improvement of approximately 10 to 50 times depending on the grid resolution through WebGPU-based GPU parallel computing compared to conventional CPU sequential computing, and secures a processing speed capable of 60 FPS real-time visualization even with double coupling based on a 512 × 512 grid. These performance characteristics serve as the technical basis for the present invention to be practically utilized as a real-time urban flood simulation and disaster prevention decision support tool.

[0396] FIG. 10 is an illustration that intuitively illustrates the overall operation concept of a WebGPU-based double drainage model real-time urban flood simulation device according to the present invention.

[0398] FIG. 10 illustrates, in a single integrated scene, the process of flooding occurring due to rainfall in an urban environment and the way each component of the present invention responds to it, so that even a non-expert can easily understand the technical configuration and operating principle of the present invention. Below, each component of FIG. 10 will be described in order from top to bottom and from left to right.

[0400] A rain cloud (310) is positioned at the top of FIG. 10, and raindrops (311) are falling from the rain cloud (310) toward the urban surface. The raindrops (311) represent rainfall intensity (mm / h), which is an input parameter of the simulation, and is a value that can be adjusted in real time by the user through the rainfall intensity slider (333) of the visualization and interaction unit (160). The thickness or density of the raindrops (311) visually represents the magnitude of the rainfall intensity.

[0402] Below the rain clouds (310), an urban landscape unfolds. A group of high-rise buildings (312) is positioned on the left side of the urban landscape, a road (313) in the center, and a green space (314) on the right. The group of high-rise buildings (312) corresponds to cells assigned high effective elevations by the Digital Surface Model (DSM), indicating that during simulation, water cannot penetrate the buildings and flows by bypassing the roads and low-lying areas between the buildings. The road (313) corresponds to asphalt paving, having a low Manning roughness coefficient (n 0.015) is an area where the infiltration rate is close to 0, so most of the rainfall is converted into surface runoff. Greenery (314) corresponds to grassland or forest, and has a relatively high roughness coefficient (n It is an area where the outflow is reduced by having a permeability rate ($f_c$ = 5 to 10 mm / h, $f_0$ = 30 to 50 mm / h).

[0404] A shallow water flow (315) is illustrated on the surface of the road (313) with an arrow. The water flow (315) represents surface runoff calculated by the numerical analysis unit (140) of the present invention based on the shallow water equation (Equation 1) and Manning friction (Equation 2). The direction of the arrow corresponds to the direction of the flow velocity vector ($\vec{v}$), and the thickness of the water corresponds to the water height (B channel value) of the corresponding cell. It is illustrated that a flooded area (316) is formed where water is concentrated in a topographically low section of the road (313), such as the entrance to an underpass or the lowest point of an intersection. The flooded area (316) is represented in different colors according to the water depth in the visualization and interaction section (160), for example, 0 to 0.1 m is light blue, 0.1 to 0.5 m is blue, 0.5 to 1.0 m is dark blue, and 1.0 m or more is red.

[0406] Below the surface of Fig. 10, an underground sewer network (317) is shown in cross-sectional form. The underground sewer network (317) corresponds to a one-dimensional model constructed from a SWMM INP file in the sewer network modeling section (120). As elements constituting the underground sewer network (317), a conduit (318) with a circular cross-section buried under the road (313) is extended horizontally, a junction point (319) is arranged in the form of a vertical structure (manhole) at the intersection or branch point of the conduit (318), and a discharge port (320) connected to a river or sea is arranged at the end of the conduit (318).

[0408] A bidirectional arrow (321) is shown between the junction point (319) and the ground surface. The bidirectional arrow (321) indicates the bidirectional water exchange between the ground surface and the sewer pipe by the core mechanism of the double drainage coupling described in detail in FIG. 3, namely, apply_linkage (S523). The downward arrow indicates the direction in which water infiltrates into the sewer pipe through the manhole when the ground surface water level is higher than the junction point water level, and the upward arrow indicates the direction in which water in the sewer pipe exfiltrates back to the ground surface through the manhole when the junction point water level is higher than the ground surface water level. In actual urban flooding, backflow from the manhole due to exceeding the sewer pipe capacity is one of the key factors that rapidly worsen flooding, and the present invention can simulate this in real time.

[0410] At the discharge port (320), the flow of water inside the conduit (318) being discharged to the outside (river or sea) is indicated by a unidirectional arrow. This flow is calculated by the outfall_step (S524), and the discharged flow rate is reflected in the water level of the surface cell corresponding to the discharge port (320).

[0412] In the upper right corner of FIG. 10, a user's web browser screen (322) is shown. On the web browser screen (322), the simulation results are displayed rendered in colors corresponding to water depth on a three-dimensional terrain, and a control panel is positioned on the left side of the screen. The control panel includes a rainfall intensity slider (333), a simulation speed controller (334) (1x to 50x speed), a pause / restart button (335), and a terrain editing tool (336). This indicates that the user can adjust parameters or change the terrain in real time while observing the simulation through the web browser screen (322).

[0414] In the lower right corner of FIG. 10, a scene is illustrated in which a user installs a water barrier (323) on a road (313) using a terrain editing tool (336). The water barrier (323) corresponds to a disaster prevention structure that blocks the flow of water by artificially raising the elevation of a specific cell during the simulation execution. As described in FIG. 4, when the visualization and interaction unit (160) receives the user's terrain change input, it updates the RGBA texture A channel (235) value of the corresponding cell, and from the next time step, the changed elevation is reflected in the shallow water equation calculation. After the water barrier (323) is installed, the water flow (315) is blocked by the water barrier (323), so that water accumulates on the upstream side and flooding is reduced on the downstream side. Through this, the user can compare and verify the effects of various disaster prevention scenarios (changing the height and location of the water barrier, adding embankments, installing open channels, etc.) in real time.

[0416] Additionally, a levee collapse scenario (324) is additionally illustrated in the lower left corner of FIG. 10. A scene is shown in which a section of the levee constructed in the section adjacent to the river collapses, causing river water to overflow into the city. This is implemented by the user artificially lowering the elevation of a specific cell using a terrain editing tool (336). When the levee collapse (324) occurs, the effective elevation of the cell becomes lower than the river water level, so an outflow from the river toward the city occurs immediately according to the shallow water equation, and the resulting flooding spread process is simulated in real time. This function can be utilized for disaster prevention tasks such as river flood risk assessment, verification of levee design suitability, and establishment of emergency evacuation plans.

[0418] The correspondence between each element illustrated in FIG. 10 and the components of the present invention is summarized as follows. Rain clouds (310) and rain streaks (311) correspond to simulation input parameters (rainfall intensity). High-rise building groups (312), roads (313), and green spaces (314) correspond to DEM, DSM, and cover map data generated by the terrain data collection unit (110). Water flow (315) and flooded areas (316) correspond to the calculation results of the numerical analysis unit (140) and the simulation execution unit (150). Underground sewer network (317), conduits (318), junction points (319), and discharge outlets (320) correspond to the one-dimensional model constructed by the sewer network modeling unit (120). The bidirectional arrow (321) corresponds to the apply_linkage (S523) and outfall_step (S524) calculations of the drainage linkage step (S520). The web browser screen (322) corresponds to the visualization and interaction unit (160) and operates on a cross-platform environment set by the WebGPU initialization unit (130). The water barrier (323) and the embankment collapse (324) correspond to the real-time terrain change function of the visualization and interaction unit (160).

[0420] As explained above, FIG. 10 illustrates, in a single integrated scene, how the overall technical configuration of the present invention—rainfall input, integration of topographic and pipe network data, bidirectional linkage of 2D surface runoff and 1D sewer network, GPU parallel real-time computation, and web browser-based visualization and interaction—operates organically in a real urban environment to enable the analysis of flood occurrence, spread, and response. This demonstrates that the technical concept of the present invention is not merely limited to a combination of numerical algorithms, but can be implemented as a practical tool that supports real-time decision-making at actual disaster sites.

[0422] FIG. 11 is a diagram showing the configuration of a web browser-based simulation user interface (UI) provided to a user by the visualization and interaction unit (160) of the present invention.

[0424] The present invention is characterized by utilizing a WebGPU to run a real-time urban flood simulation within a web browser without the need for separate software installation, and a prerequisite for practical application is providing an interface that allows the user to intuitively execute, observe, and manipulate the simulation. Figure 11 describes in detail the screen layout of this user interface and the functions of each component.

[0426] The screen of Fig. 11 is divided into three main areas: a 3D simulation viewport (331), a left control panel (332), and a bottom timeline bar (340).

[0428] The 3D simulation viewport (331) is the main display area that occupies most of the center or right side of the screen. In the viewport (331), a 3D terrain model based on DEM / DSM generated by the terrain data collection unit (110) is rendered along with a satellite image texture, and the water level of each cell calculated by the simulation execution unit (150) is superimposed on it as a translucent water surface layer. The color of the water surface layer changes stepwise according to the water depth, allowing the flood depth to be visually and intuitively understood. In one embodiment of the present invention, the water depth range of 0 to 0.1 m is represented in light sky blue, the range of 0.1 to 0.3 m in blue, the range of 0.3 to 0.5 m in navy blue, the range of 0.5 to 1.0 m in dark blue, and the range of 1.0 m or more in red. These color ranges can be correlated with the flood risk grade standards of the Ministry of the Interior and Safety, enabling intuitive use in disaster prevention practice.

[0430] The viewport (331) supports viewpoint manipulation via mouse or touch input. Camera rotation (orbit) is performed by dragging the left mouse button, camera movement (pan) is performed by dragging the right mouse button, and zooming in and out (zoom) is performed by scrolling the mouse wheel. Through this, the user can freely switch between a viewpoint that provides an overview of the entire target area and a viewpoint that closely observes specific road intersections or alleys between buildings, thereby assessing the flooding situation. Additionally, if any cell within the viewport (331) is clicked, detailed information such as the coordinates of the cell, elevation, current water depth, X·Y flow velocity, Manning roughness coefficient, and infiltration rate is displayed in a pop-up format.

[0432] The left control panel (332) is an area where input elements for controlling the execution of the simulation and adjusting parameters are arranged vertically. A rainfall intensity slider (333) is placed at the top of the control panel (332). The rainfall intensity slider (333) is a horizontal slider that can adjust rainfall intensity in real time within a range of 0 mm / h to 200 mm / h, and the current setting value is displayed as a number to the right of the slider. When the user operates the slider, the value is immediately reflected as the rainfall input for the hydrological stage (S510), and flooding simulation according to the changed rainfall intensity is performed from the next time stage. For example, if the rainfall intensity is set to 30 mm / h at the start of the simulation to simulate a normal rain situation, and then adjusted upward to 100 mm / h during the simulation, the transition to a heavy rain situation is immediately reflected, and the process of rapidly increasing flooded area and depth can be observed.

[0434] A simulation speed controller (334) is positioned below the rainfall intensity slider (333). The simulation speed controller (334) provides the function of controlling the progress speed of the simulation in the range of 1x to 50x speed. 1x speed is a real-time mode where 1 second of the real world corresponds to 1 second of the simulation, and 50x speed is an acceleration mode where 50 seconds of the simulation proceed during 1 second of the real world. In acceleration mode, GPU computational resources are concentrated on the simulation computation by adjusting the frame update cycle of the visualization rendering. Through this, the user can complete the simulation of a rainfall scenario lasting several hours within minutes, and can also lower the speed to 1x speed at a specific point to observe the progress of the flooding in detail.

[0436] A pause / restart button (335) is positioned below the simulation speed controller (334). When the pause button is pressed, the simulation loop stops at the current time step, and the flooded state at the time of the pause is maintained in the viewport (331). In the paused state, the user can freely manipulate the viewpoint of the viewport (331), view cell information, or perform terrain modification operations. When the restart button is pressed, the simulation resumes from the state at the time of the pause. This function is useful when performing a precise analysis at the point where a specific flooding threshold is reached during the simulation, or when observing the effects of terrain modification step by step.

[0438] A terrain editing tool (336) is placed below the pause / restart button (335). The terrain editing tool (336) is a set of tools that allows the user to change the elevation of specific cells of the grid (230) while the simulation is running or stopped. The terrain editing tool (336) includes a plurality of sub-tools.

[0440] First, the barrier installation tool (337) performs the function of raising the elevation of the cells included in the path by a specified height when the user specifies a linear path with a mouse over the viewport (331). For example, if a barrier with a height of 1.0 m is installed over the road (313), the RGBA texture A channel (235) value of the cells is updated to a value that adds 1.0 m to the existing elevation. When the simulation resumes, the effect of water flow being blocked because the cells have a higher elevation than the surroundings in the shallow water equation calculation, water remaining on the upstream side, and flooding being reduced on the downstream side appears immediately.

[0442] Second, the embankment installation and collapse tool (338) performs the function of adding an embankment in a section adjacent to a river or removing a section of an existing embankment. When adding an embankment, it raises the elevation similarly to the cutoff wall installation tool (337), and when collapsing an embankment, it lowers the cell elevation of the designated section to the original DEM value or lower. When embankment collapse is applied, the process of river water overflowing into the urban area in that section is immediately calculated by the shallow water equation, and flooding spread begins.

[0444] Third, the open channel installation tool (339) performs the function of creating an artificial waterway (open channel) by lowering the cell elevation along a linear path specified by the user. When the open channel is installed, a cell path with a lower elevation than the surroundings is formed, resulting in a drainage effect in which water is preferentially guided along the path. Through this, the user can verify the flood reduction effect in advance when installing a temporary drainage channel.

[0446] Since elevation changes by the terrain editing tool (336) are recorded directly in the A channel (235) of the RGBA texture, they are reflected immediately from the next time step without separate data reprocessing or simulation restart. This immediacy is one of the core practical values ​​of the present invention, as in conventional simulation tools, input data had to be modified and the entire simulation restarted from the beginning when terrain was changed, whereas in the present invention, terrain can be changed during the simulation and the effect can be checked in real time.

[0448] At the bottom of FIG. 11, a timeline bar (340) is horizontally positioned across the entire width of the screen. The timeline bar (340) is a horizontal progress bar indicating the time progression of the simulation, with the left end corresponding to the simulation start time (t = 0) and the right end to the simulation end time (e.g., t = 3,600 seconds, i.e., 1 hour). Above the timeline bar (340), a playhead indicating the current simulation time is displayed, and as the simulation progresses, the playhead moves from left to right. To the right of the timeline bar (340), the current simulation elapsed time is displayed as a number in the format "hh:mm:ss". Additionally, above the timeline bar (340), event markers such as the time when the user changed the rainfall intensity or edited the terrain are displayed as small icons, allowing one to clearly see on the time axis what operations were performed at what point during the simulation process.

[0450] Current simulation status information is displayed on the left side of the timeline bar (340). The displayed information includes the current time interval (Δt₂D, e.g., 0.25 s), current maximum water depth ($d_{max}$, e.g., 1.23 m), grid resolution (512 × 512), and current frame rate (FPS, e.g., 60). This information is updated every frame, allowing the user to monitor the numerical status of the simulation and system performance in real time. In particular, the process of Δt₂D automatically decreasing when the water depth increases due to the CFL-based adaptation time interval (Equation 4) can be confirmed through the status information on the timeline bar (340), which serves to visually assure the user that numerical stability is being maintained in real time.

[0452] Meanwhile, the user interface illustrated in FIG. 11 corresponds to an embodiment of the present invention, and the screen layout, color range, arrangement of control elements, etc., may be changed according to the purpose of implementation. For example, modifications are possible to consider use on mobile devices, such as changing the control panel (332) to a top foldable menu or optimizing the touch gestures of the viewport (331) for a mobile environment. Since the present invention can be run on any web browser where a WebGPU operates, the same simulation function can be used on various devices such as desktop PCs, laptops, and tablets.

[0454] Additionally, the visualization and interaction unit (160) may provide additional functions for quantitatively analyzing simulation results. For example, when a user specifies two points in the viewport (331), a function may be included to display a topographic cross-section and a water surface profile of the corresponding section as a graph, a function to calculate the average and maximum flood depth and flood area within a specific area when a specific area is selected as a rectangle, and a function to display the distribution of maximum flood depth over the entire simulation period as a heatmap. These analysis functions can be directly utilized for decision-making, such as establishing disaster prevention plans, estimating the scale of damage, and setting evacuation routes.

[0456] As described above, the user interface illustrated in FIG. 11 corresponds to a user contact point where the visualization and interaction unit (160) of the present invention provides integrated intuitive visualization of simulation results, real-time parameter adjustment, terrain editing, and disaster prevention scenario analysis. Since even users without specialized knowledge of hydrology and fluid dynamics can adjust rainfall intensity with a slider, check flooding results by color on a three-dimensional terrain, and immediately observe the effects of installing watertight walls or open ditches, the present invention functions as a general-purpose flood simulation platform that can be utilized in the practical work of various user groups, such as local government disaster prevention officials, urban planners, and construction site managers, as well as a professional research tool.

[0458] Below, technical details that were not fully explained in detail in FIGS. 1 to 11 are described in addition.

[0459] The sewer network modeling unit (120) of the present invention converts the C source code of the Storm Water Management Model (SWMM) released by the US Environmental Protection Agency (US EPA) into a WebAssembly binary using the EMScripten (EMCC) compiler. In this process, the core computational modules of the SWMM engine, namely the dynamic wave-based conduit flow rate calculation, junction point water level estimation, and discharge boundary condition processing, can be executed at a speed comparable to native on the JavaScript runtime of a web browser.

[0460] To explain the specific compilation process, SWMM's C source files are input into EMCC with the optimization level set to O1, the output format specified as WASM, and file system emulation enabled to enable reading of INP files. The outputs generated as a result of compilation are two types: a WebAssembly binary file (.wasm) and JavaScript glue code (.js) that loads and executes it. The JavaScript glue code exposes the main functions of the SWMM engine, namely initialization (swmm_open), step 1 execution (swmm_step), result retrieval (swmm_get), and termination (swmm_close), as wrapper functions that can be called in a web environment.

[0461] The advantages of porting the SWMM engine to WebAssembly are as follows: First, cross-platform compatibility is guaranteed because there is no need to separately install the SWMM executable file on the user's local system. Second, since one-dimensional sewer network calculations are performed on the client (web browser) side, there is no delay in real-time communication with the server. Third, the execution performance of WebAssembly is only about 1.2 to 1.5 times that of native C execution, so one step (2-second interval) for an urban sewer network with hundreds to thousands of junction points can be processed within a few milliseconds.

[0462] When the sewer network modeling unit (120) parses the SWMM INP file to obtain real-world coordinates (longitude, latitude) of each junction point (221) and discharge point (223), a coordinate mapping process is performed to convert these into cell indices (i, j) of the grid (230). The mapping is performed by a simple linear proportional calculation based on the latitude and longitude range of the target area and the grid resolution (512 × 512). For example, if the longitude range of the target area is 127.00° to 127.05° and the longitude of a specific junction point is 127.025°, the i index of the corresponding junction point is calculated as (127.025 - 127.00) / (127.05 - 127.00) × 512 = 256. The j index is calculated in the same way for the latitude. The (i, j) index mapped in this way is used to identify the surface cell corresponding to the junction point during the apply_linkage(S523) operation in the multiplication linkage step (S520).

[0463] When multiple junction points are mapped to a single cell, the external inflow and outflow ($d_{ext}^{t}$) of that cell is calculated by summing the exchange amounts of the multiple junction points. Conversely, when a single junction point spans multiple adjacent cells, it is mapped to the nearest cell, but a weighted distribution method to adjacent cells may be applied if necessary.

[0464] The double buffering technique mentioned in the description of FIG. 5 is explained in more detail. The numerical analysis unit (140) of the present invention maintains two textures of the same size (512 × 512 × RGBA), which are referred to as Texture A and Texture B, respectively. At any time step n, the compute shader reads the current state (flow rate, water height, altitude) from Texture A, performs calculations according to Equations 1 through 5, and then writes the updated state to Texture B. At the next time step n+1, the read / write targets are swapped, and data is read from Texture B and written to Texture A. Through this alternating structure, a race condition caused by simultaneous reading and writing to the same texture is fundamentally prevented. In the WebGPU API, this structure is implemented by separating and specifying read-only textures and write-only textures when setting the bind group of the compute pass.

[0465] The edge cells of the grid (230), namely the cells corresponding to i = 0, i = 511, j = 0, and j = 511, require separate boundary condition processing because some of the four adjacent cells are located outside the grid range. In one embodiment of the present invention, an open boundary condition is applied to allow the flow rate directed outside the grid range to flow out freely. Specifically, when referencing the value of an adjacent cell in the direction outside the grid from an edge cell, the value of the edge cell itself is duplicated and used (Neumann boundary condition), thereby making the rate of change of the flow velocity in the normal direction at the boundary surface zero, allowing the water to naturally flow out of the boundary. This is intended to prevent unrealistic flooding caused by artificial water retention at the edge of the simulation target area. However, if physical barriers such as embankments or mountains exist at the boundary of the target area, a closed boundary effect naturally occurs because the elevation of the cell is already set high.

[0466] The initial conditions set by the simulation execution unit (150) before initiating the simulation loop are further explained. For each cell of the grid (230), the effective elevation (DEM + DSM) calculated in the data integration step (256) is set in the A channel (235), the initial river water level in the B channel (234) if it is a river cell, and 0 if it is an other cell, and 0 (no initial flow velocity) is set in the R channel (232) and G channel (233), respectively. In the SWMM engine, the initial water level of each junction, the initial flow rate of the conduit, and the boundary conditions of the discharge port described in the INP file are loaded by the swmm_open function. The simulation start time is set to t = 0, the initial time interval Δt₂D is set to 0.25 s, the first time interval (2D) is set to 0.25 s, and the second time interval (1D) is set to 2.0 s, respectively. The initial value of the rainfall intensity is the value specified by the user through the rainfall intensity slider (333) of the control panel (332).

[0467] The simulation execution unit (150) of the present invention is equipped with multiple safety devices to prevent numerical divergence. First, the time interval is automatically reduced when the water depth increases by the CFL-based adaptive time interval (Equation 4). Second, the stability of the time integral is improved compared to the explicit method by the semi-negative method (θ = 0.9). Third, in cells where the water depth is less than a preset minimum threshold value (0.001 m in one embodiment of the present invention), the flow velocity calculation is omitted and the flow velocity of the cell is set to 0, thereby avoiding numerical singularity that may occur when the denominator (h^{4 / 3}) of the Manning friction term approaches 0 in extremely shallow water. Fourth, if the water height of the cell is calculated as a negative number in a single time step, it is clamped to 0 to prevent a physically impossible state.

[0468] The visualization and interaction unit (160) of the present invention can provide a function to store and compare multiple simulation results performed under different conditions (rainfall intensity, topographic change, sewer network capacity, etc.) for the same target area. For example, the maximum flood depth distribution of a scenario without a cutoff wall and a scenario with a cutoff wall can be simultaneously displayed on a left-right split screen, or the difference in flood depth between the two scenarios can be visualized as a separate heat map. This function can be utilized in decision-making processes such as cost-effectiveness analysis of disaster prevention alternatives and determination of optimal cutoff wall placement.

[0469] The basic embodiment of the present invention targets a 512 × 512 grid, but the grid resolution can be adjusted within the range of 256 × 256 to 1024 × 1024 depending on the scale of the target area and the required precision, and higher resolutions can also be applied as GPU hardware advances. Furthermore, the dual-drainage coupling structure of the present invention can be extended by linking with one-dimensional sewer network models other than SWMM (e.g., MOUSE, InfoWorks ICM, etc.), in which case the engine of the model can be ported to WebAssembly or a configuration can be applied to transmit results executed on the server side in real time via WebSocket, etc. Moreover, if rainfall input is linked with real-time observation data or forecast data from the Korea Meteorological Administration, it is possible to expand it into a real-time flood prediction system for actual rainfall events.

[0470] Although the present invention has been described above with reference to preferred embodiments, those skilled in the art will understand that various modifications and changes can be made to the invention without departing from the spirit and scope of the invention as described in the following claims. Accordingly, the scope of the present invention is not limited to the embodiments described above but is determined by the matters described in the claims, and should be interpreted to include all variations within the scope equivalent to the matters described in the claims. Explanation of the symbols

[0471] 110: Terrain Data Collection Unit 120 : Sewer network modeling section 130 : WebGPU Initialization Section 140 : Numerical Analysis Department 150 : Simulation Execution Unit 160 : Visualization and Interaction Section 210 : 2D surface grid layer 220 : 1D sewer network layer 221 : junction 222 : conduit 223 : Outfall 230 : Grid (512 × 512) 231 : RGBA texture pixel 232 : R channel (X-direction flow velocity) 233 : G Channel (Y-direction flow velocity) 234 : B Channel (Water Level) 235 : A Channel (elevation) 240 : WMTS server 241: Tile Index Calculation Step 242: Tile Download Steps 243: Tile Joining Step 244: Combined Image Post-processing Step

Claims

Claim 1 In a WebGPU-based dual-drainage model real-time urban flood simulation device, the device comprises: a terrain data collection unit that collects terrain data of a simulation target area and generates a terrain model; a sewer network modeling unit that receives an INP file of an EPA SWMM rainfall-runoff model and generates a sewer network model including junction, conduit, and outlet elements, and compiles the C language source code of the SWMM engine into WebAssembly to prepare for performing one-dimensional sewer network calculations in a web browser environment; a WebGPU initialization unit that creates a WebGPU context in a web browser and configures compute shaders and a rendering pipeline; and a GPU compute shader that parallelizes two-dimensional surface runoff calculations, including flow velocity calculation based on shallow water equations and advection equations, surface friction processing considering Manning roughness coefficients, and determination of water height in each cell, wherein the flow velocity in the X direction is in the Red channel of the texture, the flow velocity in the Green channel is in the Y direction, and the water height is in the Blue channel, A numerical analysis unit that stores elevations in the Alpha channel respectively and simultaneously reads four physical quantities of a cell with a single memory access for a single pixel; a simulation execution unit that time-synchronizes and executes a 2D surface runoff calculation by the numerical analysis unit and a 1D sewer network calculation by the sewer network modeling unit, and includes a double drainage coupling that performs bidirectional water exchange between the surface and the sewer network junction point, wherein the time interval of the 2D surface runoff calculation is set to a first time interval and the time interval of the 1D sewer network calculation is set to a second time interval greater than the first time interval, so that the 1D sewer network calculation is performed once whenever the 2D surface runoff calculation is performed multiple times; and a visualization and interaction unit that visualizes simulation results and processes real-time user interaction;A WebGPU-based double-drainage model real-time urban flood simulation device characterized by including Claim 2 In a WebGPU-based dual-drainage model real-time urban flood simulation method, the method comprises: (a) a terrain data collection step of generating a terrain model by collecting satellite images, a Digital Elevation Model (DEM), Digital Statistical Model (DSM), cover map data, and river data of a simulation target area, wherein the rectangular boundaries of the simulation target area are converted into tile numbers based on a quadtree tile system in the format of a Web Map Tile Service (WMTS), and satellite images corresponding to each tile number are sequentially collected and combined to generate a texture for the terrain model; (b) a sewer network modeling step of generating a sewer network model including junction, conduit, and outlet elements from an INP file of an EPA SWMM rainfall-runoff model, wherein the C language source code of the SWMM engine is compiled into WebAssembly to prepare for performing one-dimensional sewer network calculations in a web browser environment; (c) a WebGPU initialization step of creating a WebGPU context in a web browser and configuring compute shaders and a rendering pipeline; and (d) flow velocity based on the shallow water equation and the advection equation A numerical analysis algorithm implementation step in which a 2D surface runoff operation including calculation, surface friction processing considering the Manning roughness coefficient, and determination of the water level of each cell is implemented using a GPU compute shader, wherein the flow velocity in the X direction is stored in the Red channel of the texture, the flow velocity in the Y direction is stored in the Green channel, the water level in the Blue channel, and the elevation above sea level is stored in the Alpha channel, respectively;(e) A real-time simulation execution step comprising a bidirectional double drainage coupling that synchronizes and drives the 2D surface runoff calculation by the numerical analysis algorithm and the 1D sewer network calculation by the SWMM engine according to respective time intervals, wherein the time interval of the 2D surface runoff calculation is set as the first time interval and the time interval of the 1D sewer network calculation is set as the second time interval greater than the first time interval, so that the 1D sewer network calculation is performed once whenever the 2D surface runoff calculation is performed multiple times, and determines the inflow from the surface to the junction or the outflow from the junction to the surface by comparing the surface water height and the water level at each junction, and reflects the flow rate of the conduit connected to the discharge port to the corresponding surface cell, wherein a hydrological step that calculates the infiltration amount of each cell by applying the HOLTON infiltration capacity formula based on cover map data and assigns a water height due to effective rainfall to each cell by subtracting the infiltration amount from the rainfall amount, a drainage coupling step that performs the bidirectional double drainage coupling, and the adjacent cell's A real-time simulation execution step that repeatedly performs, at every time interval, a flow velocity calculation step that calculates the flow velocity of each cell based on water level, elevation, and flow velocity; a water level update step that updates the water level of each cell based on the external inflow amount determined in the drainage interlocking step and the flow velocity calculated in the flow velocity calculation step; and a time interval determination step that dynamically determines the next time interval based on CFL (Courant-Friedrichs-Lewy) conditions, but adjusts it to a value smaller than the initial time interval according to the maximum water depth within the grid cell; and (f) a visualization and interaction step that visualizes the simulation results in three dimensions, processes the user's pause, restart, and playback speed adjustments in real time during the simulation, applies terrain changes corresponding to the installation of cutoff walls, embankment collapse, terrain addition, or open channel addition based on user input during the simulation, and simulates drainage and disaster prevention effects according to the changed terrain conditions in real time;A WebGPU-based double-drainage model real-time urban flood simulation method characterized by including Claim 3 In claim 1, the terrain data collection unit collects satellite images, a digital elevation model (DEM), building height data (DSM), cover map data, and river data to generate a terrain model, wherein the terrain model is in the form of a two-dimensional grid with a scale of 512×512 cells, and each cell directly stores real-world values ​​including elevation above sea level, building height, surface roughness coefficient, and infiltration capacity without preprocessing to normalize them to values ​​between 0 and 1; and the numerical analysis unit applies a semi-implicit scheme in discretizing the shallow water equation, setting the weighting parameter θ value to 0.9, and the friction slope considering the Manning roughness coefficient It is applied cell by cell according to the relationship (wherein the gradient formula is expressed in ASCII as "Sf = n²·v|v| / h^(4 / 3)", where Sf is the friction gradient, n is the local Manning roughness coefficient, v is the flow velocity vector, and h is the water depth), and a double buffering technique is applied using two textures of the same size to write to the texture of the previous time step while reading the texture of the current time step; the simulation execution unit, in performing the bidirectional water exchange, copies the 2D grid cell data stored in GPU memory to CPU memory, then iterates through all junction points of the SWMM engine and compares the water height of the surface cell where each junction point is located with the water level of the corresponding junction point to determine the inflow from the surface to the junction point or the outflow from the junction point to the surface, and calculates the flow rate of the conduit connected to the discharge port and reflects it in the corresponding surface cell; the visualization and interaction unit applies terrain changes corresponding to the installation of a cutoff wall, embankment collapse, addition of terrain, or addition of an open channel based on user input during the simulation, wherein the terrain A WebGPU-based dual-drainage model real-time urban flood simulation device characterized by recording elevation changes caused by changes in the alpha channel of the texture and reflecting them in the shallow water equation calculation from the next time step to simulate drainage and disaster prevention effects according to changed terrain conditions in real time.

Citation Information

Patent Citations

  • Shielding elimination method, equipment, storage medium and device

    CN115937416A

  • Judgement system and method for flooding road area using GIS and information model of sewer pipe network

    KR101815581B1

  • Sewerage network modeling and optimization system

    KR102812206B1

  • Apparatus and mehtod for real-time web-based urban flood simulation

    KR102865868B1

  • WebGPU-based rendering method, electronic equipment and storage medium

    CN114494550A