A method and device for preventing intelligent decision-making by combining cesium to realize flood simulation

Abstract

This invention relates to the field of disaster prevention and management, specifically providing a method and device for intelligent decision-making in flood simulation and prevention, combined with Cesium. The method comprises the following steps: S1, geographic data access; S2, real-time monitoring and analysis of hydrological conditions; S3, flood simulation model construction, including hydrodynamic model construction, model parameter setting and optimization, and a real-time update mechanism; S4, Unreal Engine's Chaos physics engine handles the physical interaction of water flow, linking Chaos physical data through modular logic; S5, integrating an LSTM time-series prediction model with real-time hydrological rainfall to realize flood evolution trends; S6, a HeatMap displays key disaster areas and disaster water levels; S7, identifying nearby evacuation points and scientifically and rationally deploying emergency routes. Compared with existing technologies, this invention can provide intelligent decision-making management for pre-disaster simulation prevention and post-disaster relief and recovery.

Description

Technical Field

[0001] This invention relates to the field of disaster prevention and management technology, specifically providing a method and device for intelligent decision-making in flood simulation and prevention, which combines Cesium. Background Technology

[0002] Floods, as highly destructive natural disasters, cause enormous loss of life and property to human society. With the acceleration of global climate change and urbanization, the frequency and intensity of floods are on the rise, making accurate prediction and effective prevention of flood disasters particularly important.

[0003] Traditional flood simulation methods have several limitations. Firstly, regarding model accuracy, some models struggle to accurately reflect complex terrain and the intricate changes in water flow within the actual environment, leading to significant discrepancies between simulation results and reality. For instance, some simple hydrodynamic models cannot accurately simulate the changes in flow velocity and discharge under different terrain conditions, such as valleys and hillsides, when faced with complex mountainous terrain. Secondly, in terms of real-time performance, many existing flood simulation systems cannot provide real-time updates and rapid calculations of flood velocity and discharge, failing to meet the needs of emergency decision-making during disasters. When floodwaters rise rapidly, the lack of timely updates to simulation data prevents relevant departments from obtaining accurate flood information, thus impacting disaster response and rescue efforts.

[0004] Furthermore, existing flood simulation systems suffer from shortcomings in data visualization and interactivity. For decision-makers and rescue personnel, a clear and intuitive understanding of the dynamic changes in floods is crucial. However, the visualizations generated by traditional systems are often not realistic enough, making it difficult for users to quickly and accurately grasp the development trend of floods. Moreover, the lack of convenient and efficient operation methods for user interaction with the simulation system causes numerous inconveniences for users when obtaining the necessary information and conducting relevant analyses.

[0005] For example, an existing patent (publication number CN119203856B) discloses a dynamic simulation method for digital twin flood inundation in small and medium-sized watersheds based on UE5. It includes the following steps: (1) data collection, construction of hydrodynamic model, calculation and output of grid nodes and triangular unit GeoJSON data; (2) creation of UE5 visualization flood inundation rendering plugin; (3) creation of flood roles, construction of flood model triangular grid primitives, and construction of flood hydrological flow mode method; (4) parsing of grid nodes and triangular unit GeoJSON data and inputting it into UE5 to complete the dynamic simulation of the flood inundation process. Although the above scheme can achieve efficient simulation, the system is lacking in intelligent decision-making. In addition, this invention is insufficient in large-area three-dimensional scenes and data loading. Another example is a visualization rendering method for digital twin flood flow field disclosed (publication number CN117237567A). Although this method can achieve simulation, it lacks the ability to perform some visual effects of flood particle flow and cannot accurately simulate the changes in flow velocity and flow rate of floods due to changes in terrain conditions. Summary of the Invention

[0006] This invention addresses the shortcomings of the prior art by providing a highly practical intelligent decision-making method for flood simulation and prevention, which combines Cesium.

[0007] A further technical objective of this invention is to provide a rationally designed, safe, and applicable intelligent decision-making device for flood simulation and prevention that incorporates Cesium.

[0008] The technical solution adopted by this invention to solve its technical problem is:

[0009] A method for intelligent decision-making in flood simulation and prevention, combined with Cesium, comprises the following steps:

[0010] S1. Geographic data access includes deep integration of Cesium terrain data and Uneal's Landscape tool;

[0011] S2. Real-time hydrological data integration collects, integrates, and manages real-time rainfall data from rain gauges and river water level data in a unified manner, enabling real-time monitoring and analysis of hydrological conditions.

[0012] S3. Flood simulation model construction includes hydrodynamic model construction, model parameter setting and optimization, and real-time update mechanism;

[0013] S4 and Unreal Engine's Chaos physics engine handle the physical interactions of water flow, ensuring that particle motion conforms to the laws of physics. Niagara is responsible for rendering the visual effects of water flow and associates Chaos's physical data through modular logic.

[0014] S5. Integrating an LSTM time series prediction model with real-time hydrological rainfall to realize the flood evolution trend;

[0015] S6. Based on Unreal Engine flood evolution simulation, the HeatMap displays key disaster areas and disaster water levels.

[0016] S7. Intelligent decision-making uses technology to identify the nearest refuge points and scientifically and rationally deploy emergency routes.

[0017] Furthermore, in step S1, high-precision terrain data and ground oblique photography data acquired by UAV or satellite remote sensing are then processed by denoising and smoothing. Gaussian filtering algorithm is used to smooth the terrain data. By setting filtering parameters, noise points in the terrain data are removed to make the terrain surface smoother. The processed data is published as a local service through CesiumLab and the service is connected to Unreal Engine through the CeaiumForUnreal plugin.

[0018] Sensor nodes are deployed around rivers and lakes in the simulated area. The sensors transmit the collected water level and flow velocity data to the server of the data processing center in real time through wireless communication modules. A data interface is established with the meteorological department to obtain rainfall data of the simulated area in real time. The meteorological department collects rainfall information and sends the data to the data processing center of this system in a standard format.

[0019] At the data processing center, the collected hydrological data are calibrated and quality controlled.

[0020] Furthermore, in step S3, the hydrodynamic model construction includes:

[0021] (1) Construction of a one-dimensional river channel model;

[0022] For the main rivers in the basin, a one-dimensional hydrodynamic model is constructed using the Saint-Venant equations, employing the continuity equation αA / αt+αQ / αx=0 and the momentum equation αQ / αt+α(Q² / A) / αx+gAαh / αx+gASf=0.

[0023] Where A is the cross-sectional area of ​​the water passage, Q is the flow rate, h is the water level, Sf is the friction gradient, and g is the acceleration due to gravity.

[0024] The Preissmann four-point implicit difference scheme is used for discretization, with the time step set to a fixed time and the spatial step dynamically adjusted according to the width of the river channel.

[0025] Boundary conditions: The upstream boundary uses the flow process curve, and the downstream boundary uses the water level-flow relationship curve. When flooding occurs, it automatically switches to a free outflow boundary.

[0026] (2) Construction of a two-dimensional watershed model;

[0027] For the floodplains, low-lying areas and urban areas around the river, a two-dimensional shallow water equation model is constructed.

[0028] The two-dimensional shallow water equations are adopted with water depth average, including the continuity equation αh / αt+α(hu) / αx+α(hv) / αy=0 and the momentum equations in the x and y directions;

[0029] The solution is obtained by combining the finite volume method with the Roe scheme. Spatial discretization is performed by adaptively fitting unstructured triangular meshes to the boundaries of complex terrain and buildings.

[0030] Furthermore, step S4 includes:

[0031] S4-1, the water flow physics interaction of the Chaos physics engine, adopts a fluid physics model and momentum transfer and watershed interaction algorithm;

[0032] The fluid physics model uses the SPH algorithm as its basic framework, discretizing the water flow into tiny particles. Each particle carries mass, velocity, and density attributes. The Chaos engine calculates the interaction force between particles by solving the Navier-Stokes equations, where the gravitational acceleration is perpendicular to the Cesium terrain surface, ensuring that the water flows naturally along the terrain slope.

[0033] The resistance calculation introduces a dynamic viscosity coefficient, which is automatically adjusted according to the water flow velocity to simulate the viscosity of low-speed water flow and the turbulence characteristics of high-speed water flow. The collision detection adopts a combination of hierarchical bounding box and precise geometric detection. When static collision objects come into contact, the Chaos engine completes the collision response calculation and generates a reflection velocity vector.

[0034] The momentum transfer and watershed interaction algorithm achieves momentum transfer between particles through pressure gradient force. When the density difference between adjacent particles exceeds a threshold, the Chaos engine calculates the acceleration force generated by the pressure difference, which propels the particles from high-density areas to low-density areas, simulating the diffusion effect of water flow.

[0035] With support for interaction with dynamic objects, when water particles collide with floating objects, Chaos calculates the magnitude of the impact force and transmits it to the object, causing the object to produce corresponding displacement and rotation; at the same time, the motion of the object will in turn disturb the water flow, forming local eddies. The particle velocity vector in the eddy region will produce a circular offset, and the offset angle is positively correlated with the object's motion speed.

[0036] S4-2, visual rendering of water flow effects using the Niagara particle system;

[0037] The basic particles use a hybrid approach of surface sprites and mesh particles: low-speed water flow uses circular surface sprites, with texture animation to simulate water surface ripples; high-speed water flow uses mesh particles and enables distortion to represent the turbulent and breaking effects of the water flow.

[0038] Particle lifecycle is tied to physical state: When Chaos detects that the particle speed exceeds a certain speed, Niagara automatically triggers the "wave generation" event, generating 5-8 micro wave particles behind the particle trajectory. The lifecycle of the wave particles increases with the speed and generates random offsets in the direction perpendicular to the speed.

[0039] Furthermore, step S5 includes:

[0040] S5-1, Construction of basic feature dataset;

[0041] Historical hydrological sequences are extracted from the system database using hourly hydrological data of a simulated area over a period of time to form a basic time series. In the data preprocessing stage, cubic spline interpolation is used to fill in missing values, and Z-score standardization is used to eliminate the influence of dimensions.

[0042] Geographic feature parameters are extracted from the Cesium topographic database as static inputs for watershed feature parameters. Principal component analysis is used to reduce the dimensionality of the 12-dimensional geographic parameters to 5-dimensional principal components, thereby reducing the model complexity.

[0043] S5-2, Real-time data access;

[0044] Hydrological sensor data is uploaded to the automatic monitoring stations deployed in the watershed at regular intervals. After being denoised by Kalman filtering, the data is pushed to the input of the LSTM model in JSON format.

[0045] Meteorological data is integrated with the National Meteorological Data Network API to obtain rainfall forecast data for a period of time in the future, while also accessing the rainfall data that has already occurred. The rainfall data is converted into a watershed grid rainfall sequence using the inverse distance weighted interpolation method, which serves as the key dynamic input for the LSTM model.

[0046] S5-3, Data Timing Alignment and Window Division;

[0047] Multi-source data synchronization uses a timestamp alignment algorithm to unify data with different sampling frequencies into a 1-hour interval sequence. For high-frequency data, a sliding window mean is used, and for low-frequency data, linear interpolation is used for expansion.

[0048] The input window is constructed using a sliding time window mechanism to generate model input samples. Each sample contains historical hydrological data, rainfall data, and static geographic parameters, and the corresponding output is the flood evolution prediction result for a future period of time.

[0049] Furthermore, in step S6, a piecewise nonlinear mapping function is used to convert the water level value to the RGB color space, and different sensitivity parameters are set for different water level ranges.

[0050] The low water level section smoothly transitions from pure cyan to sky blue, the medium water level section transitions from pure yellow to orange to highlight the warning of medium danger areas, and the high water level section transitions from pure red to dark red. The higher the water level, the more saturated the display becomes, which strengthens the visual impact of extremely dangerous areas.

[0051] Then, a regional importance weight coefficient (W) is introduced to correct the base color, with the weight value dynamically calculated based on regional attributes; flow velocity data is incorporated into HeatMap as an auxiliary parameter, and dual-parameter visualization is achieved through color saturation modulation; finally, multi-dimensional query interaction is performed.

[0052] Furthermore, in step S7, firstly, a basic database of refuge points is constructed. Based on spatial indexing technology and Cesium terrain data, two major categories of information are associated: static attributes and dynamic status. The analytic hierarchy process is used to construct a refuge point scoring system to quantify the applicability of refuge points from multiple dimensions.

[0053] In the scientific layout of emergency routes, Cesium road vector data is integrated to construct a directed graph network, where nodes represent road intersections and edges represent road segments. The initial weight of each edge is set to the segment length / design speed. An improved Dijkstra algorithm combined with a genetic algorithm is used to simultaneously optimize three objective functions:

[0054] Minimum travel time T: minT = sum(segment length / actual travel speed);

[0055] Maximum safety S: maxS=1-(length of high-risk road section / total length), where high-risk road section refers to road section with water level ≥0.2m or slope >10°;

[0056] Minimum evacuation pressure P: minP = Number of people to be evacuated along the route / Traffic capacity of the road segment;

[0057] By using Pareto optimality, 3-5 candidate paths are generated for decision-makers to choose from based on the actual situation.

[0058] A smart decision-making device for flood simulation and prevention using Cesium, comprising: at least one memory and at least one processor;

[0059] The at least one memory is used to store a machine-readable program;

[0060] The at least one processor is configured to invoke the machine-readable program to execute a flood simulation and prevention intelligent decision-making method that incorporates Cesium.

[0061] Compared with existing technologies, the present invention provides a method and apparatus for intelligent decision-making in flood simulation and prevention, which combines Cesium and has the following significant advantages:

[0062] This invention, by combining high-precision topographic data from Cesium with an advanced hydrodynamic model built on Unreal Engine, can accurately simulate the variations in flood velocity and flow rate under complex terrain conditions, improving simulation accuracy and providing reliable data support for flood disaster prediction and assessment. The real-time update mechanism ensures that the system can adjust simulation parameters promptly based on the latest hydrological and meteorological data, quickly calculating flood velocity and flow rate, meeting the urgent need for real-time information during flood disasters, and buying valuable time for emergency decision-making.

[0063] Leveraging Unreal Engine's powerful rendering capabilities, the system achieves 3D visualization of flood simulation results, showcasing the dynamic changes of floods in an intuitive and realistic way. This allows users to quickly and accurately understand the development trend of floods, improving the efficiency of information dissemination. Rich interactive features enable users to easily query information and set simulation scenario parameters, enhancing user-system interactivity and providing convenient conditions for decision analysis. A comprehensive early warning mechanism and decision support analysis functions can promptly release early warning information to relevant departments and the public, and provide decision-makers with scientific and reasonable decision-making suggestions, helping to improve flood disaster response capabilities and reduce loss of life and property. Attached Figure Description

[0064] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0065] Figure 1 This is a flowchart illustrating a method for intelligent decision-making in flood simulation and prevention that combines Cesium.

[0066] Figure 2 This is a method for intelligent decision-making in flood simulation and prevention that combines Cesium with video recording. Figure 1 ;

[0067] Figure 3 This is a method for intelligent decision-making in flood simulation and prevention that combines Cesium with video recording. Figure 2 ;

[0068] Figure 4 This is a method for intelligent decision-making in flood simulation and prevention that combines Cesium with video recording. Figure 3 ;

[0069] Figure 5 This is a method for intelligent decision-making in flood simulation and prevention that combines Cesium with video recording. Figure 4 ;

[0070] Figure 6 This is a method for intelligent decision-making in flood simulation and prevention that combines Cesium with video recording. Figure 5 ;

[0071] Figure 7 This is a method for intelligent decision-making in flood simulation and prevention that combines Cesium with video recording. Figure 6 ;

[0072] Figure 8 This is a method for intelligent decision-making in flood simulation and prevention that combines Cesium with video recording. Figure 7 . Detailed Implementation

[0073] To enable those skilled in the art to better understand the present invention, the present invention will be further described in detail below with reference to specific embodiments. Obviously, the described embodiments are merely some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0074] The following is a preferred embodiment:

[0075] like Figure 1-8 As shown in this embodiment, a method for intelligent decision-making in flood simulation and prevention, combined with Cesium, has the following steps:

[0076] S1. Geographic data access includes deep integration of Cesium terrain data and Uneal's Landscape tool;

[0077] High-precision terrain data and oblique ground photography data are acquired through UAV or satellite remote sensing. The terrain data then undergoes denoising and smoothing. A Gaussian filtering algorithm is used to smooth the terrain data; by setting filtering parameters (filter radius, standard deviation, etc.), noise points in the terrain data are removed, making the terrain surface smoother and preventing abnormal fluctuations in the terrain data from affecting flood simulation results. The processed data is then published as a local service through CesiumLab, and the service is integrated into Unreal Engine (e.g., using the CesiumForUnreal plugin). Figure 2 , Figure 3 (As shown).

[0078] Hydrological data acquisition and processing methods:

[0079] Sensor nodes are strategically deployed around rivers, lakes, and other water bodies in the simulated area. These sensors transmit the collected water level and flow velocity data in real time to the server at the data processing center via wireless communication modules (such as LoRa and NB-IoT). A data interface is established with the meteorological department to obtain real-time rainfall data for the simulated area. The meteorological department collects rainfall information using equipment such as weather radar and rain gauges and sends the data to the system's data processing center in a standard format (such as JSON).

[0080] At the data processing center, the collected hydrological data undergoes calibration and quality control. For data collected by current meters and water level gauges, calibration is performed using historical calibration data and sensor characteristic parameters to eliminate sensor errors. Simultaneously, a data fusion algorithm is employed to combine data from multiple sensors, improving data accuracy and reliability. Finally, hydrological data is integrated through Unreal Engine's access to the hydrological backend data API interface and local JSON data.

[0081] S2. Real-time hydrological data integration collects, integrates, and manages real-time rainfall data from rain gauges and river water level data in a unified manner, enabling real-time monitoring and analysis of hydrological conditions.

[0082] S3. Flood simulation model construction includes hydrodynamic model construction, model parameter setting and optimization, and real-time update mechanism;

[0083] The following steps are involved in constructing a hydrodynamic model:

[0084] (1) Construction of a one-dimensional river channel model;

[0085] For the main rivers and major tributaries in the basin, a one-dimensional hydrodynamic model is constructed using the Saint-Venant equations.

[0086] The governing equations are: the continuity equation αA / αt+αQ / αx=0 and the momentum equation αQ / αt+α(Q² / A) / αx+gAαh / αx+gASf=0 (where A is the cross-sectional area of ​​the water passage, Q is the flow rate, h is the water level, Sf is the friction gradient, and g is the acceleration due to gravity).

[0087] Numerical solution: The Preissmann four-point implicit difference scheme is used for discrete solution. The time step is set to 10 seconds, and the spatial step is dynamically adjusted according to the width of the river channel (50-100 meters for wide channels and 10-20 meters for narrow channels) to ensure the calculation accuracy at complex river conditions (such as bends and branching channels).

[0088] Boundary conditions: The upstream boundary uses the flow process curve (from real-time monitoring of hydrological stations or LSTM prediction), and the downstream boundary uses the water level-flow relationship curve (fitted based on historical data). When flooding occurs, it automatically switches to a free outflow boundary.

[0089] (2) Construction of two-dimensional watershed model:

[0090] For the floodplains, low-lying areas and urban areas around the river, a two-dimensional shallow water equation model is constructed.

[0091] Governing equations: A two-dimensional shallow water equation set averaged by water depth is adopted, which includes the continuity equation αh / αt+α(hu) / αx+α(hv) / αy=0 and the momentum equations in the x and y directions (considering Coriolis force, turbulent viscous force, etc.).

[0092] Numerical Method: The finite volume method combined with the Roe scheme is used for solution. Spatial discretization employs an unstructured triangular mesh (minimum mesh size 5 meters, maximum 50 meters), which can adaptively fit complex terrain and building boundaries. The flood inundation and receding processes are simulated using dry and wet boundary treatment techniques, with a dry riverbed threshold set at 0.05 meters.

[0093] Roughness coefficient: A spatially distributed roughness field is set according to land use type (e.g., 0.03-0.05 for cultivated land, 0.015-0.025 for urban areas, and 0.05-0.08 for forest land), and the roughness values ​​of different regions can be quickly retrieved by looking up tables.

[0094] S4 and Unreal Engine's Chaos physics engine handle the physical interactions of water flow, ensuring that particle motion conforms to the laws of physics. Niagara is responsible for rendering the visual effects of water flow and associates Chaos's physical data through modular logic.

[0095] Specifically, it includes:

[0096] S4-1, Water flow physics interaction of the Chaos physics engine;

[0097] The Chaos physics engine is responsible for building the underlying physical behavior model of flood flow. It simulates the collision, gravity response, resistance and momentum transfer effects of water flow in complex environments through high-precision numerical calculations, ensuring that particle motion strictly follows the laws of fluid mechanics.

[0098] Fluid physics model algorithm:

[0099] The SPH (Smoothed Particle Hydrodynamics) algorithm is used as the basic framework to discretize the water flow into a massive number of tiny particles, each carrying physical properties such as mass, velocity, and density. The Chaos engine calculates the interaction forces between particles by solving the Navier-Stokes equations, where the gravitational acceleration is set to 9.81 m / s², perpendicular to the Cesium terrain surface (the direction is dynamically adjusted through the terrain normal vector) to ensure that the water flows naturally along the terrain slope. A dynamic viscosity coefficient is introduced for resistance calculation, automatically adjusted according to the water flow velocity (0.8 for velocities < 1 m / s, decreasing to 0.3 for velocities > 3 m / s), simulating the viscosity of low-speed water flow and the turbulent characteristics of high-speed water flow. Collision detection uses a combination of hierarchical bounding box (AABB) and precise geometric detection. When water particles come into contact with static collision objects (3D models loaded by the Cesium service) such as terrain and buildings, the Chaos engine completes the collision response calculation within 0.01 seconds and generates a reflection velocity vector. For example, when water hits a vertical wall, it will produce a rebound trajectory symmetrical to the wall normal. The rebound coefficient is dynamically adjusted according to the wall material (0.3 for concrete walls and 0.1 for soil walls).

[0100] Momentum transfer and watershed interaction algorithms:

[0101] Momentum transfer between particles is achieved through pressure gradient force. When the density difference between adjacent particles exceeds the threshold (0.05 g / cm³), the Chaos engine calculates the acceleration force generated by the pressure difference, which propels the particles from the high-density region to the low-density region, simulating the diffusion effect of water flow.

[0102] Support for interaction with dynamic objects, such as the interaction between floating objects (trees, vehicles) and water flow: When water particles collide with floating objects, Chaos calculates the magnitude of the impact force (F=Δp / Δt, where Δp is the change in particle momentum) and transmits it to the object, causing the object to produce corresponding displacement and rotation; at the same time, the motion of the object will in turn disturb the water flow, forming local eddies. The particle velocity vector in the eddy region will produce a circular offset, and the offset angle is positively correlated with the object's motion speed (maximum offset 30°).

[0103] S4-2, visual rendering of water flow effects using the Niagara particle system;

[0104] Particle morphology and lifecycle management:

[0105] The basic particles use a hybrid approach of Billboard and mesh particles: low-speed water flow (<1m / s) mainly uses circular Billboard to simulate water surface ripples through texture animation; high-speed water flow (>2m / s) uses mesh particles (miniature water units composed of triangular patches) and enables distortion deformation (based on a noise function) to represent the turbulent and breaking effect of the water flow.

[0106] Particle lifecycle is tied to physical state: When Chaos detects a particle velocity exceeding 2.5 m / s, Niagara automatically triggers a "wave generation" event, generating 5-8 micro-wave particles (1 / 3 the size of the original particle) behind the particle's trajectory. The lifecycle of the wave particles increases with velocity (0.5 seconds at 2.5 m / s, and 1.2 seconds at 4 m / s), and they generate random offsets along the direction perpendicular to the velocity.

[0107] Simulation of material and optical properties:

[0108] Dynamic transparency adjustment: Based on particle density data (ρ) provided by Chaos, transparency (α) is calculated through material nodes using the formula α=1-exp(-k×ρ), where k is the scene lighting coefficient (k=0.8 for sunny days and k=0.5 for cloudy days). High-density areas (such as deep water, ρ>1.0g / cm³) exhibit high opacity (α≈0.9), while low-density areas (such as shallow water or waves, ρ<0.3g / cm³) have transparency reduced to 0.3-0.5, creating a natural transition between light and dark areas.

[0109] Reflection and Refraction Effects: A combination of environment mapping (Cubemap) and real-time reflection is used. Water surface particles reflect images of the surrounding environment (Cesium terrain, buildings), and the reflection intensity varies with the angle between the particle normal and the light source (the reflection intensity increases by 30% when facing the sun). At the same time, the refraction offset node simulates the refraction effect of light passing through the water body. The offset amount is positively correlated with the water depth (maximum offset of 3 pixels), which makes the underwater terrain appear distorted.

[0110] Enhanced dynamic effects:

[0111] Foam and foam simulation: In the area where water and solids collide (such as the bottom of the riverbed and around the bridge piers), Niagara generates foam particles based on the collision impulse (I) of Chaos. The greater the impulse (I>5N·s), the higher the foam density (100-200 particles per square meter). The foam particles are made of white semi-transparent material with a lifespan of 1-3 seconds, gradually dissipating over time (transparency decreases from 0.6 to 0).

[0112] Enhanced flow rate visualization: The flow rate is mapped by particle color, using the HSV color space to transition from blue (0 m / s) to red (5 m / s), consistent with the color system of the HeatMap module, allowing users to directly associate flow rate with physical state visually.

[0113] S5. Integrating an LSTM time series prediction model with real-time hydrological rainfall to realize the flood evolution trend;

[0114] include:

[0115] S5-1, Construction of basic feature dataset;

[0116] Historical hydrological series were extracted from the system database using hourly hydrological data for the simulated area over the past 10 years, including water level (unit: m), cross-sectional velocity (unit: m / s), and discharge (unit: m³ / s), forming the basic time series. In the data preprocessing stage, cubic spline interpolation was used to fill in missing values ​​(missing value rate controlled within 5%), and Z-score standardization was used to eliminate the influence of dimensions (standardization formula: x'=(x-μ) / σ, where μ is the mean and σ is the standard deviation).

[0117] Geographic feature parameters are extracted from the Cesium topographic database as static inputs, including watershed area (km²), average slope (°), river meander coefficient, and soil permeability coefficient (mm / h). Principal component analysis (PCA) is used to reduce the 12-dimensional geographic parameters to 5-dimensional principal components, thereby reducing the model complexity.

[0118] S5-2, Real-time data access mechanism;

[0119] Hydrological sensor data is deployed at 32 automatic monitoring stations within the watershed, uploading real-time data every 5 minutes, including current water level, instantaneous flow velocity, water temperature, etc. The data is transmitted to edge computing nodes via a 5G private network, and after being denoised by Kalman filtering, it is pushed to the input end of the LSTM model in JSON format.

[0120] Meteorological data is integrated with the National Meteorological Data Network API to obtain 72-hour rainfall forecast data (spatial resolution 1km×1km, temporal resolution 1 hour), while also incorporating past rainfall data (collected from 68 rain gauges distributed throughout the watershed, updated every 10 minutes). The rainfall data is converted into a watershed grid rainfall sequence using the inverse distance weighted (IDW) interpolation method, serving as a key dynamic input to the LSTM model.

[0121] S5-3, Data Timing Alignment and Window Division;

[0122] Multi-source data synchronization uses a timestamp alignment algorithm to unify data with different sampling frequencies into a 1-hour interval sequence. For high-frequency data (such as water level readings every 5 minutes), a sliding window mean is used for processing, while for low-frequency data (such as daily soil moisture), linear interpolation is used to expand the data, ensuring that the time granularity of the input sequence is consistent.

[0123] The input window is constructed using a sliding time window mechanism to generate model input samples. The window length is set to 72 hours (i.e., 72 time steps). Each sample contains 72 hours of historical hydrological data, rainfall data, and static geographic parameters. The corresponding output is the flood evolution prediction result for the next 24 hours. The window sliding step is 1 hour, achieving hourly updates to the prediction.

[0124] S6. Based on Unreal Engine flood evolution simulation, the HeatMap displays key disaster areas and disaster water levels.

[0125] include:

[0126] S6-1, Refined Implementation of Data Mapping Rules:

[0127] Establish a water level-color depth mapping model;

[0128] A piecewise nonlinear mapping function is used to convert water level values ​​to the RGB color space, and different sensitivity parameters are set for different water level ranges.

[0129] Low water level section (0-1m): Using the linear formula color=(0,255-127×h,255) (where h is the water level value), the color smoothly transitions from pure cyan (0,255,255) to sky blue (0,128,255). The color change difference is controlled within 12-15 RGB units for every 0.1m, ensuring that subtle water level changes in the shallow water area are discernible.

[0130] Medium water level section (1-3m): The exponential function mapping color=(255,255-63×(h-1)^1.5,0) is activated, transitioning from pure yellow (255,255,0) to orange (255,165,0). When the water level exceeds 2m, the rate of color change increases by 30%, highlighting the warning of the medium danger area.

[0131] High water level section (above 3m): A linear attenuation of color=(255-31×(h-3),0,0) is used to transition from pure red (255,0,0) to dark red (128,0,0), while an alpha=0.9-0.05×(h-3) transparency adjustment is superimposed. The higher the water level, the more saturated the display becomes, enhancing the visual impact of the extremely dangerous area.

[0132] Geographic weighting factor fusion:

[0133] A regional importance weighting coefficient (W) is introduced to modify the base color. The weight value is dynamically calculated based on the regional attributes.

[0134] Residential area: W=1.2 (color intensity increased by 20%);

[0135] Industrial plant area: W=1.1 (color intensity increased by 10%)

[0136] Agricultural area: W=0.9 (color intensity reduced by 10%);

[0137] No man's land: W=0.7 (color intensity reduced by 30%);

[0138] The revised formula is final_color=base_color×W, ensuring that densely populated or economically valuable areas are more prominent in the HeatMap under the same water level conditions.

[0139] Multi-parameter superposition mapping:

[0140] Flow velocity data is incorporated as an auxiliary parameter into HeatMap, and dual-parameter visualization is achieved through color saturation modulation.

[0141] Flow velocity < 1 m / s: Saturation remains at the baseline value (70%).

[0142] 1-3 m / s: Saturation increases linearly with flow velocity to 100%.

[0143] >3m / s: Saturation remains at 100% with a 2px white border added.

[0144] S6-2, Regional Query and Details Display Function:

[0145] Point query: Supports precise mouse clicks and long presses on the touchscreen to trigger 3D raycasting and return the result for that point.

[0146] Real-time data: current water level (accurate to 0.01m), flow velocity, inundation duration, and estimated receding time;

[0147] Historical data: Water level change curve over the past 24 hours (including hourly record points);

[0148] Related information: Population density of the area, types of key facilities (such as schools and hospitals);

[0149] The query results are displayed as floating information cards, with the background color of the card matching the color of the HeatMap, enhancing the sense of spatial connection.

[0150] Select a polygonal area by dragging the mouse, and the system will automatically calculate the result.

[0151] Regional statistics: average water level, location of the highest water level, and standard deviation of water level (reflecting the uniformity of water level within the region);

[0152] Impact assessment: Inundated area (accurate to 0.01 km²), estimated affected population (based on census data), and predicted direct economic losses (in ten thousand yuan).

[0153] The results are displayed in a dynamic statistics panel, with the selected area boundary marked by a yellow dashed line on the HeatMap, which fades out automatically after 10 seconds.

[0154] Historical Retrospection and Comparison:

[0155] Timeline control: Provides a slider and key time point buttons (such as "1 hour ago" and "peak time"), and supports retracing the HeatMap status at any time within 24 hours.

[0156] Dual-screen comparison: Enable the "history vs. current" split-screen mode. The left side displays the HeatMap of the selected historical moment, and the right side displays the current status. The system automatically marks areas where the water level is rising (red) and falling (green) by calculating the difference. Areas with a difference ≥ 0.5m are highlighted with a flashing effect.

[0157] S7. Intelligent decision-making uses technology to identify the nearest refuge points and scientifically and rationally deploy emergency routes;

[0158] include:

[0159] S7-1, Construction of the Basic Database of Refuge Points:

[0160] The database covers two main categories of information: static attributes and dynamic status of refuge sites. It is linked to Cesium terrain data through spatial indexing technology.

[0161] Static attributes include the latitude and longitude coordinates of the refuge site (accurate to 0.0001°), the area (m²), the number of people it can accommodate (calculated based on an average area of ​​2m² per person), the structural safety level (three levels: A / B / C, with level A able to withstand a 50-year flood), the facility configuration (such as whether there is an emergency medical station, material storage warehouse, and drinking water supply system), and the year of construction (used to assess the degree of structural aging).

[0162] Dynamic Status: Real-time updates on the current number of evacuees (collected via an infrared counter at the entrance), remaining capacity (total capacity - current number of evacuees), resource consumption (e.g., remaining food and medicine), and accessibility status (whether the area is inaccessible due to flooding of surrounding roads). Data is updated every 5 minutes to ensure the timeliness of status information.

[0163] Multifactor weighted scoring model:

[0164] An evacuation point rating system was constructed using the Analytic Hierarchy Process (AHP) to quantify the applicability of evacuation points from multiple dimensions.

[0165] Distance factor (weight 30%): Calculated based on the straight-line distance (d) between the current disaster point and the refuge point, using score_distance=100×exp(-0.001×d) (d is in meters). The closer the distance, the higher the score. The score drops to below 37 points when the distance is more than 1000 meters.

[0166] Capacity factor (weight 25%): Calculated based on the ratio of remaining capacity (c) to the number of people to be evacuated (n). score_capacity=min(100,(c / n)×100). Full marks are awarded when the remaining capacity is greater than or equal to the number of people to be evacuated.

[0167] Safety factor (weight 20%): The difference between the structural safety level (A=100 points, B=80 points, C=50 points) and the current water level and the elevation of the refuge point (5 points for every 1 meter of elevation, up to a maximum of 20 points), score_safety=structural score+min(20,5×h_diff).

[0168] Facility Factor (Weight 15%): Scored based on the completeness of facilities such as the number of medical stations and the amount of supplies, with a maximum score of 100 points. 20 points are deducted for each missing key facility.

[0169] Accessibility factor (weight 10%): dynamically adjusted from 0 to 100 points based on real-time road conditions and flooding situation, with 0 points awarded when completely inaccessible.

[0170] The total score is calculated as score_total = sum(score of each factor × weight). Refuges with a score ≥ 60 are included in the candidate list and recommended in descending order of score.

[0171] Dynamic adaptation and real-time adjustment mechanism:

[0172] Disaster response adjustment: When the flood level is raised (e.g., from blue alert to orange alert), the safety factor weight is automatically increased to 30%, the distance factor weight is reduced to 20%, and safer refuge points are prioritized.

[0173] Group-specific adaptation: The scoring model is adjusted for different evacuation groups (such as ordinary residents, the elderly and children, and patients). For example, for patients, the weight of medical stations in the facility factor is increased to 50%, and refugees with sufficient medical resources are given priority.

[0174] Real-time exclusion mechanism: When the remaining capacity of the refuge point is less than 80% of the number of people to be evacuated, or it is expected to be flooded within 2 hours (based on LSTM prediction), it is automatically excluded from the candidate list and a secondary search within a 3-kilometer radius is triggered.

[0175] S7-2, Scientific Layout of Emergency Routes;

[0176] Path network model construction:

[0177] Basic road network data: Integrating Cesium's road vector data, including attributes such as road type (expressway, national highway, county road, rural road), width, design speed, number of lanes, etc., to construct a directed graph network, where nodes are road intersections and edges are road segments, and the initial weight of the edges is set to the road segment length / design speed (representing travel time).

[0178] Real-time weight adjustment: The weights of the edges are dynamically adjusted based on flood evolution data;

[0179] Submerged road sections (water level ≥ 0.3m): weight set to infinite (unpassable);

[0180] Flooded sections (0.1m ≤ water level < 0.3m): weight multiplied by 5 (passage time increased by 5 times).

[0181] Multi-objective path optimization algorithm:

[0182] An improved Dijkstra algorithm combined with a genetic algorithm is used to optimize three objective functions simultaneously.

[0183] Minimum travel time (T): minT = sum(segment length / actual travel speed);

[0184] Maximum safety (S): maxS=1-(length of high-risk section / total length), where high-risk section refers to section with water level ≥0.2m or slope >10°;

[0185] Minimum evacuation pressure (P): minP = Number of people to be evacuated along the route / Traffic capacity of the road segment;

[0186] By using Pareto optimality, 3-5 candidate paths are generated for decision-makers to choose from based on the actual situation.

[0187] F3, Intelligent Decision Support and Visual Presentation;

[0188] Evacuation point allocation table: Recommended evacuation points, estimated evacuation numbers, and arrival time windows for each disaster-stricken area;

[0189] Route allocation map: number of people to be evacuated for each route, and estimated start and end times;

[0190] Resource requirements list: number of guides, transportation, and emergency supplies needed;

[0191] Risk assessment: Potential risks during implementation (such as the probability of the path being flooded) and corresponding countermeasures;

[0192] The simulation function allows for the rehearsal of the evacuation process within Unreal Engine, calculating indicators such as evacuation completion rate and average evacuation time to assist decision-makers in assessing the feasibility of the plan.

[0193] Visualization and interactive features:

[0194] 3D path roaming: Supports roaming paths in Unreal Engine from a first-person or third-person perspective, allowing you to intuitively view the terrain, water accumulation, obstacles, and other conditions along the path;

[0195] Path comparison analysis: Overlay the coverage of different paths on the HeatMap and show the advantages and disadvantages of each path through difference analysis.

[0196] Based on the above method, an intelligent decision-making device for flood simulation and prevention combined with Cesium in this embodiment includes: at least one memory and at least one processor;

[0197] The at least one memory is used to store a machine-readable program;

[0198] The at least one processor is configured to invoke the machine-readable program to execute a flood simulation and prevention intelligent decision-making method that incorporates Cesium.

[0199] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent decision-making in flood simulation and prevention, combined with Cesium, characterized in that, It has the following steps: S1. Geographic data access includes deep integration of Cesium terrain data and Uneal's Landscape tool; S2. Real-time hydrological data integration collects, integrates, and manages real-time rainfall data from rain gauges and river water level data in a unified manner, enabling real-time monitoring and analysis of hydrological conditions. S3. Flood simulation model construction includes hydrodynamic model construction, model parameter setting and optimization, and real-time update mechanism; S4 and Unreal Engine's Chaos physics engine handle the physical interactions of water flow, ensuring that particle motion conforms to the laws of physics. Niagara is responsible for rendering the visual effects of water flow and associates Chaos's physical data through modular logic. include: S4-1, the water flow physics interaction of the Chaos physics engine, adopts a fluid physics model and momentum transfer and watershed interaction algorithm; The fluid physics model uses the SPH algorithm as its basic framework, discretizing the water flow into tiny particles. Each particle carries mass, velocity, and density attributes. The Chaos engine calculates the interaction force between particles by solving the Navier-Stokes equations, where the gravitational acceleration is perpendicular to the Cesium terrain surface, ensuring that the water flows naturally along the terrain slope. The resistance calculation introduces a dynamic viscosity coefficient, which is automatically adjusted according to the water flow velocity to simulate the viscosity of low-speed water flow and the turbulence characteristics of high-speed water flow. The collision detection adopts a combination of hierarchical bounding box and precise geometric detection. When static collision objects come into contact, the Chaos engine completes the collision response calculation and generates a reflection velocity vector. The momentum transfer and watershed interaction algorithm achieves momentum transfer between particles through pressure gradient force. When the density difference between adjacent particles exceeds a threshold, the Chaos engine calculates the acceleration force generated by the pressure difference, which propels the particles from high-density areas to low-density areas, simulating the diffusion effect of water flow. With support for interaction with dynamic objects, when water particles collide with floating objects, Chaos calculates the magnitude of the impact force and transmits it to the object, causing the object to produce corresponding displacement and rotation; at the same time, the motion of the object will in turn disturb the water flow, forming local eddies. The particle velocity vector in the eddy region will produce a circular offset, and the offset angle is positively correlated with the object's motion speed. S4-2, visual rendering of water flow effects using the Niagara particle system; The basic particles use a hybrid approach of surface sprites and mesh particles: low-speed water flow uses circular surface sprites, with texture animation to simulate water surface ripples; high-speed water flow uses mesh particles and enables distortion to represent the turbulent and breaking effects of the water flow. Particle lifecycle is tied to physical state: When Chaos detects that the particle speed exceeds a certain speed, Niagara automatically triggers the "wave generation" event, generating 5-8 micro wave particles behind the particle trajectory. The lifecycle of the wave particles increases with the speed and generates random offsets in the direction perpendicular to the speed. S5. Integrating an LSTM time series prediction model with real-time hydrological rainfall to realize the flood evolution trend; S6. Based on Unreal Engine flood evolution simulation, the HeatMap displays key disaster areas and disaster water levels. S7. Intelligent decision-making uses technology to identify the nearest refuge points and scientifically and rationally deploy emergency routes; First, a basic database of refuge sites is constructed. Based on spatial indexing technology and Cesium terrain data, two major categories of information are associated: static attributes and dynamic status. An analytic hierarchy process is used to construct a refuge site scoring system to quantify the applicability of refuge sites from multiple dimensions. In the scientific layout of emergency routes, Cesium road vector data is integrated to construct a directed graph network, where nodes represent road intersections and edges represent road segments. The initial weight of each edge is set to the segment length / design speed. An improved Dijkstra algorithm combined with a genetic algorithm is used to simultaneously optimize three objective functions: Minimum travel time T: minT = sum(segment length / actual travel speed); Maximum safety S: maxS=1-(length of high-risk road section / total length), where high-risk road section refers to road section with water level ≥0.2m or slope >10°; Minimum evacuation pressure P: minP = Number of people to be evacuated along the route / Traffic capacity of the road segment; By using Pareto optimality, 3-5 candidate paths are generated for decision-makers to choose from based on the actual situation.

2. The method for intelligent decision-making in flood simulation and prevention based on Cesium as described in claim 1, characterized in that, In step S1, high-precision terrain data and ground oblique photography data are acquired by UAV or satellite remote sensing. Then, the terrain data is denoised and smoothed. Gaussian filtering algorithm is used to smooth the terrain data. By setting the filtering parameters, noise points in the terrain data are removed to make the terrain surface smoother. The processed data is published as a local service through CesiumLab and the service is connected to Unreal Engine through the CesiumForUnreal plugin. Sensor nodes are deployed around rivers and lakes in the simulated area. The sensors transmit the collected water level and flow velocity data to the server of the data processing center in real time through wireless communication modules. A data interface is established with the meteorological department to obtain rainfall data of the simulated area in real time. The meteorological department collects rainfall information and sends the data to the data processing center of this system in a standard format. At the data processing center, the collected hydrological data are calibrated and quality controlled.

3. The method for intelligent decision-making in flood simulation and prevention based on Cesium according to claim 2, characterized in that, In step S3, the hydrodynamic model construction includes: (1) Construction of a one-dimensional river channel model; For the main rivers in the basin, a one-dimensional hydrodynamic model is constructed using the Saint-Venant equations, employing the continuity equation αA / αt+αQ / αx=0 and the momentum equation αQ / αt+α(Q² / A) / αx+gAαh / αx+gASf=0. Where A is the cross-sectional area of ​​the water passage, Q is the flow rate, h is the water level, Sf is the friction gradient, and g is the acceleration due to gravity. The Preissmann four-point implicit difference scheme is used for discretization, with the time step set to a fixed time and the spatial step dynamically adjusted according to the width of the river channel. Boundary conditions: The upstream boundary uses the flow process curve, and the downstream boundary uses the water level-flow relationship curve. When flooding occurs, it automatically switches to a free outflow boundary. (2) Construction of a two-dimensional watershed model; For the floodplains, low-lying areas and urban areas around the river, a two-dimensional shallow water equation model is constructed. The two-dimensional shallow water equations are adopted with water depth average, including the continuity equation αh / αt+α(hu) / αx+α(hv) / αy=0 and the momentum equations in the x and y directions; The solution is obtained by combining the finite volume method with the Roe scheme. Spatial discretization is performed by adaptively fitting unstructured triangular meshes to the boundaries of complex terrain and buildings.

4. The method for intelligent decision-making in flood simulation and prevention based on Cesium according to claim 3, characterized in that, Step S5 includes: S5-1, Construction of basic feature dataset; Historical hydrological sequences are extracted from the system database using hourly hydrological data of a simulated area over a period of time to form a basic time series. In the data preprocessing stage, cubic spline interpolation is used to fill in missing values, and Z-score standardization is used to eliminate the influence of dimensions. Geographic feature parameters are extracted from the Cesium topographic database as static inputs for watershed feature parameters. Principal component analysis is used to reduce the dimensionality of the 12-dimensional geographic parameters to 5-dimensional principal components, thereby reducing the model complexity. S5-2, Real-time data access; Hydrological sensor data is uploaded to the automatic monitoring stations deployed in the watershed at regular intervals. After being denoised by Kalman filtering, the data is pushed to the input of the LSTM model in JSON format. Acquire rainfall forecast data for a period of time in the future, and simultaneously access the rainfall data that has already occurred. The rainfall data is converted into a watershed grid rainfall sequence using the inverse distance weighted interpolation method, which serves as the key dynamic input to the LSTM model. S5-3, Data Timing Alignment and Window Division; Multi-source data synchronization uses a timestamp alignment algorithm to unify data with different sampling frequencies into a 1-hour interval sequence. For high-frequency data, a sliding window mean is used, and for low-frequency data, linear interpolation is used for expansion. The input window is constructed using a sliding time window mechanism to generate model input samples. Each sample contains historical hydrological data, rainfall data, and static geographic parameters, and the corresponding output is the flood evolution prediction result for a future period of time.

5. The intelligent decision-making method for flood simulation and prevention based on Cesium as described in claim 4, characterized in that, In step S6, a piecewise nonlinear mapping function is used to convert the water level value to the RGB color space, and different sensitivity parameters are set for different water level ranges. The low water level section smoothly transitions from pure cyan to sky blue, the medium water level section transitions from pure yellow to orange to highlight the warning of medium danger areas, and the high water level section transitions from pure red to dark red. The higher the water level, the more saturated the display becomes, which strengthens the visual impact of extremely dangerous areas. Then, a regional importance weight coefficient (W) is introduced to correct the base color, with the weight value dynamically calculated based on regional attributes; flow velocity data is incorporated into HeatMap as an auxiliary parameter, and dual-parameter visualization is achieved through color saturation modulation; finally, multi-dimensional query interaction is performed.

6. A smart decision-making device for flood simulation and prevention, combined with Cesium, characterized in that, include: At least one memory and at least one processor; The at least one memory is used to store a machine-readable program; The at least one processor is configured to invoke the machine-readable program to perform the method according to any one of claims 1 to 5.

Images