A water disaster dynamic monitoring and precise early warning method and system

Abstract

The application discloses a kind of water conservancy disaster dynamic monitoring precision early warning method, it is related to the technical field of intelligent water conservancy and emergency management, comprising the following steps: pre-multiple-source data fusion modeling: through high-resolution satellite, unmanned aerial vehicle LiDAR obtains high-precision DEM and point cloud data, and according to hydrological monitoring and communication signaling data, constructs the multidimensional database including terrain, hydrology, personnel distribution;Small watershed disaster evolution simulation is carried out.The application realizes multiple-source data fusion to improve the comprehensiveness and real-time of monitoring, solves the problem of traditional single data source monitoring range limitation and insufficient precision, while realizing high-precision positioning and trajectory prediction, realizing personnel coordinate centimeter level to 200m precision dynamic tracking, combined with state space model, the trajectory under signal shielding scene is smoothly predicted, avoids positioning interruption, provides real-time, continuous personnel distribution data support for disaster impact assessment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of smart water conservancy and emergency management, in particular to a water disaster dynamic monitoring and accurate early warning method. BACKGROUND

[0002] In the technical field of smart water conservancy and emergency management, the monitoring and early warning of water disasters such as floods and waterlogging have always been an important issue to protect people's lives and property safety and social stability. At present, the traditional water disaster monitoring and early warning method mainly relies on single type of monitoring data and simple model analysis, which has many limitations.

[0003] From the data collection level, the traditional method often relies on limited ground monitoring stations to obtain hydrological data, which has small data coverage, low precision and lagging update, and cannot fully and timely reflect the complex disaster evolution process. For example, in complex mountainous small watersheds, ground stations are easily affected by terrain obstruction, resulting in data missing or distortion, which cannot accurately capture the sudden disaster risk in local areas. At the same time, the traditional method lacks effective integration and dynamic monitoring of personnel distribution information, making it difficult to quickly assess the degree of personnel threat and develop targeted early warning and transfer schemes when disasters occur.

[0004] In terms of model simulation, existing hydrological models are mostly based on simplified assumptions and single physical processes, and the simulation accuracy of flood runoff, confluence and inundation process is insufficient, especially in dealing with the water dynamic coupling problem under complex terrain of small watersheds, it is difficult to accurately predict the evolution path and inundation range of flood. In addition, the traditional risk assessment method usually only considers single dimension of disaster intensity, ignoring social factors such as personnel density, resulting in inaccurate early warning level division and failing to meet the fine and personalized needs of emergency management.

[0005] In the aspect of early warning information pushing, the traditional method lacks efficient spatial index and base station matching algorithm, making it difficult to realize accurate delivery of early warning information. Often using broadcast pushing method, which not only leads to low information transmission efficiency, but also may cause resource waste and information confusion due to too much early warning information received by irrelevant areas, affecting the early warning effect of key areas.

[0006] In summary, the existing technology has problems such as incomplete data collection, insufficient model simulation accuracy, single risk assessment dimension and inaccurate early warning information pushing in the process of water disaster monitoring and early warning, and there is an urgent need for a new method that can integrate multi-source data, realize high-precision disaster simulation and accurate early warning, to improve the scientificity and effectiveness of water disaster emergency management.

[0007] In view of the problems in the related art, no effective solution has been proposed so far. SUMMARY

[0008] In view of the problems in the related art, the present application provides a water disaster dynamic monitoring and accurate early warning method to overcome the above technical problems existing in the prior art.

[0009] The technical scheme of the present application is implemented as follows:

[0010] A water disaster dynamic monitoring and accurate early warning method, comprising the following steps:

[0011] Step S1, pre-process multi-source data fusion modeling: obtain high-precision DEM and point cloud data through high-resolution satellites and unmanned aerial vehicle LiDAR, and construct a multi-dimensional database containing terrain, hydrology, and personnel distribution according to hydrological monitoring and communication signaling data;

[0012] Step S2, small watershed disaster evolution simulation, including: dividing the target small watershed into microgrids according to 500m*500m, calculating the terrain index and runoff rate based on TOPMODEL; at the same time, constructing a one-two-dimensional water dynamic coupling model by using a one-dimensional river model and a two-dimensional inundation model, realizing energy conservation through flow interpolation, simulating the flood evolution process, and outputting the inundation grid map and flow velocity distribution data;

[0013] Step S3, processing personnel coordinate data by using DPC clustering algorithm and KDE kernel density estimation method, generating personnel heat field distribution, and calculating personnel heat value;

[0014] Step S4, constructing a three-dimensional risk assessment matrix with water depth, flow velocity and personnel heat value as three dimensions, and determining the early warning level according to the three-dimensional risk assessment matrix;

[0015] Step S5, based on the R-tree spatial index algorithm, quickly matching the communication base stations associated with the disaster area, and using the TOPSIS algorithm to sort the associated base stations, and preferentially pushing the early warning information to the base stations covering the high-risk areas;

[0016] Step S6, constructing a weighted directed graph model, using an improved A* algorithm to generate personnel transfer paths by comprehensively considering terrain cost, flood parameters and personnel density, and monitoring personnel positions in real time, and triggering a secondary early warning for personnel staying in non-safe areas for more than a preset time.

[0017] The communication signaling data includes: analyzing the communication signaling to locate personnel, including the following steps:

[0018] The coordinates of the three base stations are represented as (x1, y1), (x2, y2), and (x3, y3), and the time difference of signal arrival is represented as Δt 21 , Δt 31 , the speed of light c=3*10 8 m / s, and the distance difference is calculated and represented as:

[0019] d 31 = c * Δt 31 ;

[0020] d 21 = c * Δt 21 ;

[0021] Constructing hyperbolic equation, obtaining user coordinates (x, y);

[0022] Wherein, base station 1 and base station 2 form a hyperbola, denoted as:

[0023]

[0024] Wherein, base station 1 and base station 3 form a hyperbola, denoted as:

[0025]

[0026] Further comprising: trajectory smoothing and movement trend prediction on personnel positioning data, including the following steps:

[0027] State space model construction, including: calibrating four-dimensional state vector Wherein, (x t , y t ) is the real-time coordinate obtained by analyzing the communication signaling, denoted as the initial value, Horizontal movement speed, state transition matrix F based on uniform motion, denoted as:

[0028]

[0029] Wherein, Δt = 60s, corresponding to high-frequency sampling interval 1 / minute, observation matrix H extracts coordinate information in state vector, used to match actual observation value, denoted as:

[0030]

[0031] Recursive prediction, including the following steps:

[0032] Calibrating the previous time state X t-1 , predicting the current state, denoted as:

[0033] Updating the prediction error covariance matrix, denoted as: P t - = F * P t-1 * F T + Q;

[0034] Wherein, Q is the process noise matrix, representing the motion model uncertainty, i.e. personnel speed change;

[0035] where, if a valid observation value Z t = [x t′ , y t′ ] T , the observation residual is calculated, denoted as:

[0036] The predicted value is fused with the observation value through the Kalman gain K t , denoted as:

[0037]

[0038] where, R = diag([200 2 , 200 2 ]) is the observation noise covariance matrix;

[0039] The state vector and covariance matrix are corrected, denoted as:

[0040]

[0041] When the signal is blocked, the predicted state is used as the current estimated value to avoid trajectory interruption.

[0042] The small watershed disaster evolution simulation includes: TOPMODEL simulation, including the following steps:

[0043] The small watershed is divided into 500m x 500m micro-grid, and the terrain index is calculated, denoted as:

[0044]

[0045] The critical terrain index TI c is determined, and the water balance equation is iteratively solved.

[0046] The runoff rate is calculated, denoted as:

[0047]

[0048] where, β is the slope, A is the catchment area, θ is the soil moisture, i is the rainfall intensity, K s is the saturated hydraulic conductivity.

[0049] The two-dimensional hydrodynamic coupling model simulation includes the following steps:

[0050] One-dimensional river model simulation: based on Saint-Venant equation set, Preissmann four-point implicit format is used to discretize the equation, the river is divided into N sections, time step Δt, space step Δx;

[0051] Equation discretization, expressed as:

[0052]

[0053] Where Q is the cross-sectional flow, A is the water cross-sectional area, S0 is the river bed slope, S f is the friction slope, expressed as: Where n is the Manning roughness coefficient, R is the hydraulic radius;

[0054] Where the two-dimensional submerged model simulates: including the following steps:

[0055] Calibration finite volume method discrete, divide the area into 500m×500m grid, Roe format is used to solve the flux, considering the ground roughness n=0.03;

[0056] Coupling interface, through the interpolation matching of one-dimensional river cross-sectional flow and two-dimensional grid boundary flow, realize the conservation of energy, expressed as:

[0057]

[0058] Where M is the number of two-dimensional grid boundary elements, u i ,h i is the grid flow velocity and water depth.

[0059] Where the calculation of human thermal value, including the following steps:

[0060] Density peak clustering algorithm is used to identify personnel aggregation core points, including:

[0061] Calibration base station signaling coordinate set as Cut-off distance d c =500m;

[0062] Calculate the local density, expressed as:

[0063]

[0064] Where ρ i is the local density, reflecting the number of personnel within 500m around point i, ρ i The larger the degree of aggregation is higher;

[0065] Calculate the distance index:

[0066] Where the core point is expressed as: δ i =min{d ij |ρ j >ρ i};

[0067] Where the edge point is expressed as: δi = max{d ij};

[0068] wherein, δ i is the distance index, the core point takes the minimum distance of the adjacent high-density point, the edge point takes the global maximum distance, and is used to identify the cluster center;

[0069] The cluster center identification is performed, and the points with high ρ i and δ i are selected as the cluster center to divide the personnel gathering area;

[0070] The kernel density estimation is performed, including the following steps:

[0071] The personnel coordinates after clustering are calibrated The bandwidth h = 500 m;

[0072] The heat value is calculated and is represented as:

[0073]

[0074] The base station covering the high-risk area is preferentially pushed with the early warning information, including the following steps:

[0075] The base station set B obtained through the R-tree query is input as a candidate base station set;

[0076] The evaluation index is calibrated and is respectively represented as: the number of covered users C1, the average signal strength C2, the historical early warning success rate C3, and the standardized matrix thereof;

[0077] The number of covered users C1 is the number of users covered by the base station, and is standardized as:

[0078]

[0079] The average signal strength C2 is the signal strength of the base station, and is standardized as:

[0080]

[0081] The historical early warning success rate C3 is a percentage value, and is standardized as:

[0082]

[0083] The weighted matrix and the ideal solution are calculated;

[0084] The weighted matrix is represented as: W = [0.5, 0.3, 0.2], wherein the weight of the number of users is the highest;

[0085] The positive ideal solution is represented as: Z + = (max(v i1 ), max(vi2 ),max(v i3 );

[0086] negative ideal solution, denoted as: Z - =(min(v i1 ),min(v i2 ),min(v i3 );

[0087] Distance calculation and sorting are performed, wherein the distances of the base station i from the positive and negative ideal solutions are respectively:

[0088]

[0089] The comprehensive score S i is obtained, and the base station priority for accurate push of early warning information is obtained by arranging in descending order of S i , and is denoted as:

[0090]

[0091] The method further comprises the following steps of triggering secondary early warning for the user staying for more than a preset time, comprising the following steps:

[0092] Real-time monitoring of personnel position change, wherein if the personnel stay in a non-safe area for more than a preset time and do not move along the planned path, it is determined as abnormal stay; secondary early warning is triggered.

[0093] The beneficial effects of the application are as follows:

[0094] 1. The application realizes multi-source data fusion to improve the comprehensiveness and real-time performance of monitoring, through space-air-ground-surface-signaling data cooperation: fusion of space-air data such as high-resolution satellite DEM, unmanned aerial vehicle LiDAR point cloud, and InSAR ground surface deformation monitoring, combined with high-frequency ground sampling data of NB-IoT rain gauge and Beidou water level gauge, and personnel positioning data analyzed from communication signaling, a multi-dimensional database containing terrain, hydrology, and personnel distribution is constructed, full-factor coverage from macro terrain to micro personnel dynamics is realized, and the problems of limited monitoring range and insufficient precision of traditional single data source monitoring are solved. At the same time, high-precision positioning and trajectory prediction are realized: through personnel positioning and Kalman filtering model, dynamic tracking of personnel coordinates with centimeter to 200-meter precision is realized, combined with state space model for trajectory smoothing prediction under signal shielding scene, positioning interruption is avoided, and real-time and continuous personnel distribution data support is provided for disaster impact assessment.

[0095] 2. This invention achieves refined model simulation to improve the accuracy of disaster prediction. It utilizes a TOPMODEL coupled hydrodynamic model: based on a 500m×500m microgrid, TOPMODEL calculates topographic indices and runoff rates, combined with a one-dimensional and two-dimensional hydrodynamic coupled model to simulate the flood evolution process, outputting high-precision inundation raster maps and flow velocity distribution data. This enables dynamic simulation of the entire process of flood generation, confluence, and inundation in small watersheds, improving the prediction accuracy of flood diffusion paths and inundation depths under complex terrain compared to traditional models (e.g., water depth error ≤ 0.5m, flow velocity error ≤ 0.2m / s). Simultaneously, by employing multi-physics coupling: the one-dimensional channel model uses the Saint-Venant equations to describe water flow, while the two-dimensional inundation model considers ground roughness and flux conservation, achieving energy conservation through flow interpolation. This effectively solves the problem of insufficient simulation of the coupling effect between the river channel and the overland area in traditional models, more realistically reflecting the propagation characteristics of floods between different geomorphic units.

[0096] 3. This invention achieves dynamic risk assessment and precise classification of early warning levels by constructing a three-dimensional risk assessment matrix: using water depth, flow velocity, and human thermal values ​​as the three dimensions of assessment, a discretized matrix containing 45 risk combinations is established. Through quantitative scoring (1-5 points) and level mapping (low risk, yellow warning, red warning), a shift from "one-way disaster intensity assessment" to "comprehensive assessment of human-disaster-environment interaction" is realized. For example, when the water depth is 2.5m (medium level), the flow velocity is 1.8m / s (medium level), and the population density is high (level 5 thermal value), a red warning can be accurately triggered, which is more in line with the actual risk than the traditional early warning mechanism based solely on water depth or flow velocity. Simultaneously, combined with dynamic thermal field generation: using DPC clustering and KDE kernel density estimation, the distribution of human thermal values ​​is updated every 5 minutes, reflecting changes in population aggregation in real time, ensuring that the risk assessment results are dynamically synchronized with the disaster site, and avoiding early warning deviations caused by lag in population movement.

[0097] 4. This invention achieves precise communication and route planning, improving emergency response efficiency. It employs R-tree spatial indexing and TOPSIS base station ranking: R-trees are used to quickly match base stations associated with disaster areas. TOPSIS ranking is then performed based on indicators such as the number of covered users, signal strength, and historical early warning success rate. This prioritizes the delivery of early warning information to high-risk areas (e.g., sending red warnings to base stations covering 1500 people), reducing the time spent transmitting invalid information by more than 50% compared to broadcast methods, and improving the speed of early warning response in key areas. Simultaneously, it enables personalized evacuation route generation: Based on a weighted directed graph model and an improved A* algorithm, it generates optimal evacuation routes by comprehensively considering terrain costs, flood parameters, and population density, avoiding dangerous areas with water depth ≥1.5m or current velocity ≥2m / s. It also provides estimated travel time and turning point coordinates, offering customized escape guidance for people in different locations. For example, the route planning for evacuating people from a village to a high-altitude shelter took only 25 minutes, a 30% reduction compared to traditional experience-based routes. It also introduces a secondary early warning mechanism: by monitoring the status of people staying in the area in real time, it automatically triggers voice calls or emergency contact notifications for abnormal personnel who stay in unsafe areas for more than 30 minutes, and links drones to provide guidance, effectively solving the problem of blind spots in the coverage of traditional early warning for people who "have not received information" or whose "movement is hindered". Attached Figure Description

[0098] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0099] Figure 1 This is a flowchart illustrating a method for precise early warning of dynamic monitoring of water disasters according to an embodiment of the present invention. Detailed Implementation

[0100] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention are within the scope of protection of the present invention.

[0101] According to an embodiment of the present invention, a method for dynamic monitoring and precise early warning of water conservancy disasters is provided.

[0102] like Figure 1 As shown, the method for precise early warning of dynamic monitoring of water conservancy disasters according to an embodiment of the present invention includes the following steps:

[0103] Step S1: Perform multi-source data fusion modeling in advance: acquire high-precision DEM (Digital Elevation Model) and point cloud data through high-resolution satellites and UAV LiDAR (Light Detection and Range), and construct a multi-dimensional database containing topography, hydrology, and personnel distribution based on hydrological monitoring and communication signaling data.

[0104] The acquisition of aerospace data includes the following steps:

[0105] A 5m resolution DEM was acquired using the Gaofen-7 satellite (GF-7), and terrain modeling with a planar accuracy of 0.5m and an elevation accuracy of 1.5m was achieved through the rational function model (RFM).

[0106] The drone LiDAR (such as DJI Mavic 3 Multispectral) is used to generate 1m precision point cloud data, and a 3D terrain model is constructed by a moving surface fitting algorithm, including parameters such as river curvature and vegetation coverage.

[0107] InSAR technology based on the Sentinel-1 SAR satellite enables millimeter-level monitoring of surface deformation (accuracy ±3 mm / year) through Small Baseline Sets (SBAS).

[0108] Ground data collection includes the following steps:

[0109] Deploy NB-IoT smart rain gauges (accuracy ±2%) and Beidou water level gauges (centimeter-level accuracy), and use time division multiplexing (TDM) technology to achieve high-frequency sampling once per minute, with data transmitted back through the LoRaWAN network;

[0110] This also includes communication signaling parsing for personnel location, including the following steps:

[0111] The coordinates of the three base stations are denoted as (x1, y1), (x2, y2), and (x3, y3), and the signal arrival time differences are denoted as Δt. 21 Δt 31 The speed of light, c = 3 × 10 8 m / s, calculate the distance difference, expressed as:

[0112] d 31 =c·Δt 31 ;

[0113] d 21 =c·Δt 21 ;

[0114] Construct the equation of the hyperbola and obtain the user's coordinates (x, y);

[0115] Among them, base station 1 and base station 2 form a hyperbola, which can be represented as:

[0116]

[0117] Among them, base station 1 and base station 3 form a hyperbola, represented as:

[0118]

[0119] Specifically, by solving the two sets of hyperbola equations mentioned above, the system of two quadratic equations in two variables is obtained, thus yielding the user's coordinates (x, y).

[0120] In addition, to address the signal obstruction problem caused by mountainous terrain, trajectory smoothing and movement trend prediction are performed on personnel positioning data. The specific steps are as follows:

[0121] State-space model construction includes: calibrating the four-dimensional state vector. Among them, (x t ,y t The coordinates are obtained through communication signaling parsing and are represented as initial values. Let F be the horizontal velocity. Based on uniform motion, the state transition matrix F is expressed as:

[0122]

[0123] Where Δt = 60s, corresponding to a high-frequency sampling interval of 1 time / minute, the observation matrix H extracts the coordinate information from the state vector to match the actual observed values, expressed as:

[0124]

[0125] Performing recursive prediction involves the following steps:

[0126] Calibrate the previous state X t-1 Predicting the current state is represented as:

[0127] The updated prediction error covariance matrix is ​​represented as: P t - =F·P t-1 ·F T +Q;

[0128] Where Q is the process noise matrix, which characterizes the uncertainty of the motion model, i.e., the speed variation of the personnel;

[0129] If a valid observation Z is received t =[x t′ ,y t′ ] T If the positioning error is ≤200m, calculate the observation residual, which is expressed as:

[0130] Through Kalman gain K t The combined predicted and observed values ​​are represented as follows:

[0131] K t =P t - ·H T ·(H·P t - ·H T +R) -1 ;

[0132] Where R = diag([200 2 200 2 ]) is the observation noise covariance matrix.

[0133] The corrected state vector and covariance matrix are expressed as follows:

[0134]

[0135] When signal obstruction leads to missing observations, a predicted state is used. As the current estimate, to avoid trajectory interruption.

[0136] Step S2, conduct small watershed disaster evolution simulation, including: based on TOPMODEL (Topography-Based Hydrological Model) and a one- or two-dimensional hydrodynamic coupling model, simulate the flood generation-confluence-inundation process, and output a 500m resolution inundation raster map;

[0137] The TOPMODEL simulation includes the following steps:

[0138] The small watershed is divided into 500m × 500m microgrids, and the topographic index is calculated and expressed as:

[0139]

[0140] Determining the critical topographic index TI c Through the water balance equation Iterative solution;

[0141] The flow rate is calculated and expressed as:

[0142]

[0143] Where β is the slope, A is the catchment area, θ is the soil moisture, i is the rainfall intensity, and K is the soil moisture content. s Saturated hydraulic conductivity;

[0144] The simulation of the one- or two-dimensional hydrodynamic coupling model includes the following steps:

[0145] One-dimensional river channel model simulation: including: based on the Saint-Venant equations, using the Preissmann four-point implicit scheme to discretize the equations, dividing the river channel into N cross sections, with a time step Δt and a spatial step Δx;

[0146] The equation is discretized and expressed as:

[0147]

[0148] Where Q is the cross-sectional discharge, A is the cross-sectional area, S0 is the riverbed slope, and S... f The friction slope is expressed as: Where n is the Manning roughness coefficient, which is 0.02 to 0.04 for natural rivers and 0.01 to 0.02 for artificial channels; R is the hydraulic radius.

[0149] The two-dimensional flooding model simulation includes the following steps:

[0150] The finite volume method was used for discretization, and the region was divided into a 500m×500m grid. The flux was solved using the Roe scheme, considering the ground roughness n=0.03.

[0151] By establishing a coupling interface and interpolating the one-dimensional river cross-sectional flow with the two-dimensional grid boundary flow, energy conservation is achieved, expressed as:

[0152]

[0153] Where M is the number of boundary elements in the two-dimensional mesh, u i ,h i For grid flow velocity and water depth;

[0154] Step S3, perform dynamic identification of personnel thermal fields, including: generating personnel thermal values ​​through DPC clustering and KDE kernel density estimation;

[0155] The generation of dynamic human thermal values ​​includes the following steps:

[0156] Density Peak Clustering (DPC) algorithm is used to identify core points of people clustering, including:

[0157] The set of base station signaling coordinates is as follows Cutoff distance d c =500m;

[0158] The local density is calculated as follows:

[0159]

[0160] Where, ρi ρ represents the local density, reflecting the number of people within a 500m radius of point i. i The larger the value, the higher the degree of aggregation.

[0161] Calculate distance metrics:

[0162] The core point is represented as: δ i =min{d ij |ρ j >ρ i};

[0163] Wherein, the edge point is represented as: δ i =max{d ij};

[0164] Where, δ i The distance metric is used to identify cluster centers. Core points are determined by the minimum distance to nearby high-density points, while edge points are determined by the maximum global distance.

[0165] Perform cluster center identification and select ρ i and δ i Points with higher average values ​​are used as cluster centers to delineate areas where people congregate.

[0166] Kernel density estimation (KDE) involves the following steps:

[0167] Personnel coordinates after clustering Bandwidth h = 500m;

[0168] The thermal value is calculated and expressed as:

[0169]

[0170] The thermal values ​​of a 500m×500m grid were obtained and normalized to be divided into 5 levels (level 1 is sparse, level 5 is dense).

[0171] In this technical solution, signaling data is fed in in real time based on the message queue Kafka, and the heat field is recalculated every 5 minutes.

[0172] Step S4 involves predicting the disaster level, including establishing a three-dimensional risk assessment matrix and obtaining the warning level based on disaster intensity (water depth, flow velocity) and population density (thermal value). Details are as follows:

[0173] Step S401: Pre-extract disaster intensity features and personnel density features, wherein disaster intensity includes water depth and flow velocity, and personnel density includes thermal value;

[0174] The water depth data comes from a 500m resolution inundation raster map output by a one-dimensional and two-dimensional hydrodynamic coupled model, with each grid recording the flood depth. The flow velocity data is calculated using a one-dimensional river channel model and a two-dimensional inundation model, reflecting the water flow velocity in the river channel and the inundated area.

[0175] Among them, the thermal values ​​of personnel are generated based on DPC clustering and KDE kernel density estimation. The study area is divided into 500m×500m grids, and each grid corresponds to thermal values ​​of 1-5 levels (level 1 is sparse and level 5 is dense).

[0176] Step S402, construct a three-dimensional risk assessment matrix, including: establishing a discretized risk assessment matrix with water depth, flow velocity, and personnel thermal values ​​as three-dimensional coordinate axes. The specific steps are as follows:

[0177] Pre-discretization of data includes:

[0178] Water depth is divided into three levels: low (<1m), medium (1-3m), and high (≥3m).

[0179] The flow velocity is divided into three levels: low (<0.5m / s), medium (0.5-2m / s), and high (≥2m / s).

[0180] The thermal values ​​of people are divided into 5 levels: Level 1 (sparse), Level 2, Level 3, Level 4, and Level 5 (dense).

[0181] Perform matrix dimension mapping to generate a three-dimensional risk assessment matrix, and obtain the current risk level based on the three-dimensional risk assessment matrix.

[0182] The three-dimensional risk assessment matrix is ​​shown in Table 1.

[0183] Table 1 Three-dimensional risk assessment matrix

[0184]

[0185]

[0186] Step S5 involves matching communication base stations, including using the R-tree spatial indexing algorithm to match base stations associated with disaster areas, and sorting them using the TOPSIS method for accurate push of early warning information.

[0187] Specifically, the R-tree spatial indexing algorithm is used to match base stations associated with disaster areas, including the following steps:

[0188] Perform R-tree spatial index construction, including spatial object modeling and hierarchical index construction;

[0189] Spatial object modeling involves abstracting each communication base station as a two-dimensional spatial point object with coordinates (x, y, y). i,y i ), and construct a minimum bounding rectangle (MBR) for each base station, which is the smallest rectangle containing that point.

[0190] The hierarchical index construction includes: storing the MBR (Master Browsing Register) and corresponding base station IDs for each base station. Each leaf node can hold a maximum of M base station entries (M is a node capacity threshold, typically a value adapted to the disk page size, such as 50-100). Non-leaf nodes store the set of MBRs for their child nodes. The parent node's MBR is the smallest bounding rectangle of all child node MBRs. Details are as follows:

[0191] Initialize by grouping all base stations according to their spatial location, generating a leaf node for each group, and calculating its MBR.

[0192] Aggregate upwards, taking leaf nodes as child nodes, and recursively build upper-level non-leaf nodes until the root node is generated.

[0193] Dynamic maintenance is implemented to maintain tree structure balance when a new base station is inserted or an old base station is deleted, through node splitting or merging operations. Node splitting strategy: If the insertion causes the leaf node to exceed the capacity M, the node is split according to the "minimum overlap priority" principle: the base stations within the node are divided into two groups to minimize the overlap area between the two groups of MBRs; if the overlap areas are the same, the splitting method with the smaller total area is selected.

[0194] Node splitting strategy: If insertion causes leaf nodes to exceed capacity M, the node is split according to the principle of "minimum overlap priority": the base stations within the node are divided into two groups to minimize the overlap area of ​​the two groups of MBRs; if the overlap area is the same, the division method with the smaller total area is selected.

[0195] To perform a search for base stations associated with a disaster area, the following steps are included:

[0196] The spatial extent of the disaster area, i.e., a polygonal or rectangular area, is denoted as the query window Q and used as input.

[0197] The root node traversal begins from the root node of the R-tree. It checks if the MBR of the current node intersects with the query window Q. If they do not intersect, skip the node and its subtrees; if they intersect, proceed to the following steps.

[0198] The child nodes are recursively processed. For intersecting child nodes, if they are leaf nodes, the base stations that intersect with Q are directly extracted, i.e., the point object intersects with Q and the point is within Q; if they are non-leaf nodes, the above steps are recursively executed.

[0199] The result set is generated by collecting all base station IDs that intersect with Q, forming a candidate base station set, represented as: B = {b1, b2, ..., b}. n}

[0200] The process of checking whether the MBR of the current node intersects with the query window Q includes: for two rectangles A(x) A1 ,y A1 ,x A2 ,y A2 ) and B(x B1 ,y B1 ,x B2 ,y B2 The intersection condition is expressed as:

[0201] That is, the condition of "complete separation" is not met in any dimension.

[0202] The determination of whether a point object intersects with Q, i.e., the base station coordinates (x, y) are located within the query window Q (polygon), is made using the ray method: a horizontal ray is emitted to the right from the point (x, y), and the number of intersections with the polygon boundary is counted. If the number is odd, the point is within the region.

[0203] The TOPSIS sorting method includes the following steps:

[0204] The base station set B obtained through R-tree query is used as the evaluation object of the TOPSIS method, that is, as the input of the candidate base station set;

[0205] The evaluation indicators are defined as follows: number of covered users C1, average signal strength C2, and historical early warning success rate C3, and their standardized matrix;

[0206] Wherein, the number of users covered, C1, is the number of users covered by the base station, standardized as follows:

[0207]

[0208] Mean signal strength C2, base station signal strength (unit: dBm), standardized as follows:

[0209]

[0210] Historical early warning success rate C3, percentage value, standardized as follows:

[0211]

[0212] Calculate the weighted matrix and the ideal solution;

[0213] The weighted matrix is ​​represented as: W = [0.5, 0.3, 0.2], where the number of users has the highest weight;

[0214] The ideal solution is represented as: Z + =(max(v i1 ),max(v i2 ),max(vi3 );

[0215] The negative ideal solution is represented as: Z - =(min(v) i1 ),min(v i2 ),min(v i3 );

[0216] Distance calculation and sorting are performed, where the distances between base station i and the positive and negative ideal solutions are respectively:

[0217]

[0218] Obtain the overall score S i And press S i The base station priority is obtained by sorting the base stations in descending order for accurate push of early warning information, represented as:

[0219]

[0220] Step S6: Perform transfer route planning and trigger a secondary warning for users who stay for more than the preset time.

[0221] This technical solution, after completing disaster level prediction and base station matching, constructs a personnel evacuation path planning model based on multi-source data, generates the optimal evacuation route by combining real-time personnel distribution and terrain features, and triggers a secondary early warning for abnormally stranded personnel. Specifically:

[0222] A weighted directed graph model is constructed based on the multidimensional database built in step S1, which includes high-precision DEM, point cloud data, and heatmap values ​​of personnel distribution. The geospatial area is abstracted as a weighted directed graph G(V,E), where: node V represents safe areas (such as high ground, shelters), key landmarks (such as bridges, road intersections), and the real-time location of personnel, with coordinates obtained from communication signaling parsing or GNSS positioning; edge E represents passable paths, such as roads and river embankments, and each edge is assigned a weight w. ij The following parameters are used for calculation:

[0223] Among them, terrain cost: based on DEM data, parameters such as slope and roughness are extracted, and then calculated using the formula w. terrain =α·slope + β·roughness (quantified), where α and β are empirical coefficients; the greater the slope and the higher the roughness, the higher the cost.

[0224] Among them, traffic efficiency: Combining real-time hydrological data, such as flow velocity and water depth, the travel time for water-crossing road sections is calculated using a hydrodynamic model, and is expressed as:

[0225] w flood =γ·depth + δ·velocity;

[0226] Specifically, in application, passage is prohibited when the water depth exceeds 1.5m or the flow velocity exceeds 2m / s;

[0227] Among them, personnel density: the heat value level generated by the associated steps, and the path average heat value HeatMap. avg The higher the level, such as in dense areas of level 4-5, the higher the weight, according to w. density =η·HeatMap avg The density weighting coefficient is η, which is incremented and ranges from 0.5 to 1.0.

[0228] Personalized route generation includes the following steps:

[0229] An improved A* algorithm is used to search for the optimal path in a weighted directed graph. Efficient pathfinding is achieved by combining the heuristic function h(n) and the actual cost g(n), as follows:

[0230] f(n)=g(n)+h(n)·(1+λ·HeatMap avg );

[0231] Wherein, the actual cost g(n) represents the cumulative weight from the starting point to the current node n, including the sum of terrain, flood, and density costs; the heuristic function h(n) uses Euclidean distance to estimate the straight-line distance from the current node to the endpoint, and combines it with the priority of safe areas, such as high ground priority, for correction.

[0232] Add the starting point, i.e. the current position of the personnel, to the open list, and set g(start) = 0, h(start) to the straight-line distance to the nearest safe zone;

[0233] Expand the nodes, traverse the nodes with the lowest cost in the open list, generate adjacent nodes and calculate f(n). If the adjacent node is a prohibited area, such as water depth ≥ 3m, then skip it.

[0234] Update the path. If the cost of reaching the adjacent node through the current node is lower, update the parent node of the adjacent node to the current node and add it to the open list.

[0235] When the endpoint node, i.e. the safe zone, is added to the open list, the optimal path is generated by backtracking the parent node, and the output includes a guidance instruction containing the coordinates of the turning point and the estimated time.

[0236] The process of triggering a secondary alert for users who remain in the area for more than a preset time includes the following steps:

[0237] Through communication signal parsing and trajectory prediction in step S1, real-time monitoring of personnel location changes is achieved.

[0238] If a person stays in an unsafe area, such as a flooded area or the vicinity of a high-risk river for more than 30 minutes and does not move along the planned route, it will be judged as "abnormal stay" and a secondary warning will be triggered, such as making a voice call or automatically extracting the emergency contact information corresponding to the user's anonymized ID and sending a help notification containing the location coordinates.

[0239] Furthermore, in a specific embodiment, taking a flood disaster early warning scenario in a small watershed as an example, the area of ​​a small watershed in a mountainous area is approximately 20 km². 2 The river basin has a length of 15 km and three villages (with a total population of approximately 2,000) along its banks. The terrain is mainly mountainous and hilly, with a vegetation coverage of about 60%. During the rainy season, the area experiences sudden heavy rainfall, as detailed below:

[0240] Step S1: Perform multi-source data acquisition and processing in advance;

[0241] Among the data, the aerospace data included DEM data acquired by the Gaofen-7 satellite, which showed an average slope of 15° and a minimum river channel curvature radius of 50m within the basin. Point cloud data generated by UAV LiDAR identified three landslide-prone areas (vegetation coverage <30%). InSAR monitoring showed a surface deformation rate of +2mm / year on the right bank of the river (a slight uplift, not currently posing a risk).

[0242] Among them, ground data: the NB-IoT rain gauge monitored the rainfall intensity in real time as 80 mm / h (rainstorm level), and the Beidou water level gauge showed that the river water level rose by 1.2m in 1 hour.

[0243] Among them, personnel location: through signal parsing of 3 base stations (coordinates A(0,0), B(1000,0), C(500,800) respectively), the coordinates of 2000 people were obtained, of which 1500 people were concentrated in the center of the village (heat value level 5), and 500 people were distributed in farmland (heat value level 3).

[0244] Step S2, conduct small watershed disaster evolution simulation:

[0245] Among them, TOPMODEL calculated that the average topographic index ln(A / tanβ) was 4.2, the critical topographic index was 3.8, and the runoff rate f = 0.05 m / s (significant infiltration runoff under heavy rainfall).

[0246] Among them, the one-dimensional and two-dimensional hydrodynamic coupling models: the one-dimensional river model calculates the maximum flow velocity as 3 m / s (downstream of the river), and the cross-sectional area of ​​the water passage is A = 20 m². 2 The two-dimensional flooding model outputs a flooding raster map, showing that the water depth in the downstream area of ​​the village reaches 2.5m (medium level) and the flow velocity is 1.8m / s (medium level).

[0247] Step S3: Perform dynamic identification of personnel thermal fields, including DPC clustering and truncation distance d. c =500m, 3 cluster centers were identified (corresponding to 3 villages), local density ρ i The values ​​are 200, 150, and 100 (people / 500m radius), respectively. KDE kernel density estimation generates 500m×500m grid thermal values, with the thermal value level of the village center area being level 5 and the farmland area being level 3.

[0248] Step S4, Disaster Level Prediction: Based on the three-dimensional risk assessment matrix, the downstream area of ​​the village has the following characteristics: water depth 2.5m (medium level), flow velocity 1.8m / s (medium level), and population density level 5 (dense). Referring to Table 1, the risk score is 4.5, corresponding to a red alert (immediate evacuation of personnel is required). The farmland area has the following characteristics: water depth 1.2m (low level), flow velocity 0.8m / s (medium level), and population density level 3 (medium density). The risk score is 2.5, corresponding to a yellow alert (intensified monitoring is required).

[0249] Step S5 involves matching communication base stations and sending early warnings, including the following steps:

[0250] The disaster area (a 2km×2km rectangle centered on the village) was matched to two base stations using an R-tree (base station 1 covers the village center, and base station 2 covers the farmland area).

[0251] The TOPSIS method was used for ranking. Base station 1 covered 1500 users (C1=1500), had a signal strength of -80dBm (C2=-80), and a historical early warning success rate of 95% (C3=95%). Base station 2 covered 500 users (C1=500), had a signal strength of -75dBm (C2=-75), and a historical early warning success rate of 90% (C3=90%). The overall scores were S1=0.56 and S2=0.44, meaning the priority was Base station 1 > Base station 2. Base station 1 was prioritized for pushing red early warning information to the village center area, while base station 2 pushed yellow early warning information to the farmland area.

[0252] Step S6 involves planning the transfer path and issuing a secondary early warning, as detailed below:

[0253] The optimal path generation involves transferring people from the village center to a highland shelter (1.5km in a straight line). The path avoids river areas with a water depth greater than 1.5m and prioritizes ridgelines with lower terrain costs. The estimated time is 25 minutes.

[0254] Meanwhile, monitoring detected that three users had been stranded in the flooded area for more than 40 minutes. The system automatically sent a distress notification containing coordinates (x=800, y=600) to their emergency contacts and activated a drone to provide guidance.

[0255] It should be noted that in this technical solution, the collected personnel location data is transmitted and stored using AES-256 encryption, and the user ID is anonymized using a hash algorithm to ensure compliance with relevant regulations. Communication signaling data is used solely for disaster early warning and not for other commercial purposes; the storage period does not exceed 72 hours.

[0256] In summary, by employing the above-described technical solution of the present invention, the following effects can be achieved:

[0257] 1. This invention achieves multi-source data fusion to enhance the comprehensiveness and real-time performance of monitoring. Through the collaborative integration of space-air, ground-based, and signaling data: it fuses space-air data such as high-resolution satellite DEM, UAV LiDAR point clouds, and InSAR surface deformation monitoring, combined with high-frequency ground sampling data from NB-IoT rain gauges and BeiDou water level gauges, and personnel positioning data from communication signaling parsing, constructing a multi-dimensional database encompassing topography, hydrology, and personnel distribution. This achieves full-element coverage from macro-topography to micro-personnel dynamics, solving the problems of limited monitoring range and insufficient accuracy associated with traditional single-source data sources. Simultaneously, it achieves high-precision positioning and trajectory prediction: through personnel positioning and a Kalman filter model, it enables dynamic tracking of personnel coordinates with centimeter-level to 200-meter accuracy. Combined with a state-space model, it smoothly predicts trajectories in signal-obstructed scenarios, avoiding positioning interruptions and providing real-time, continuous personnel distribution data support for disaster impact assessment.

[0258] 2. This invention achieves refined model simulation to improve the accuracy of disaster prediction. It utilizes a TOPMODEL coupled hydrodynamic model: based on a 500m×500m microgrid, TOPMODEL calculates topographic indices and runoff rates, combined with a one-dimensional and two-dimensional hydrodynamic coupled model to simulate the flood evolution process, outputting high-precision inundation raster maps and flow velocity distribution data. This enables dynamic simulation of the entire process of flood generation, confluence, and inundation in small watersheds, improving the prediction accuracy of flood diffusion paths and inundation depths under complex terrain compared to traditional models (e.g., water depth error ≤ 0.5m, flow velocity error ≤ 0.2m / s). Simultaneously, by employing multi-physics coupling: the one-dimensional channel model uses the Saint-Venant equations to describe water flow, while the two-dimensional inundation model considers ground roughness and flux conservation, achieving energy conservation through flow interpolation. This effectively solves the problem of insufficient simulation of the coupling effect between the river channel and the overland area in traditional models, more realistically reflecting the propagation characteristics of floods between different geomorphic units.

[0259] 3. This invention achieves dynamic risk assessment and precise classification of early warning levels by constructing a three-dimensional risk assessment matrix: using water depth, flow velocity, and human thermal values ​​as the three dimensions of assessment, a discretized matrix containing 45 risk combinations is established. Through quantitative scoring (1-5 points) and level mapping (low risk, yellow warning, red warning), a shift from "one-way disaster intensity assessment" to "comprehensive assessment of human-disaster-environment interaction" is realized. For example, when the water depth is 2.5m (medium level), the flow velocity is 1.8m / s (medium level), and the population density is high (level 5 thermal value), a red warning can be accurately triggered, which is more in line with the actual risk than the traditional early warning mechanism based solely on water depth or flow velocity. Simultaneously, combined with dynamic thermal field generation: using DPC clustering and KDE kernel density estimation, the distribution of human thermal values ​​is updated every 5 minutes, reflecting changes in population aggregation in real time, ensuring that the risk assessment results are dynamically synchronized with the disaster site, and avoiding early warning deviations caused by lag in population movement.

[0260] 4. This invention achieves precise communication and route planning, improving emergency response efficiency. It employs R-tree spatial indexing and TOPSIS base station ranking: R-trees are used to quickly match base stations associated with disaster areas. TOPSIS ranking is then performed based on indicators such as the number of covered users, signal strength, and historical early warning success rate. This prioritizes the delivery of early warning information to high-risk areas (e.g., sending red warnings to base stations covering 1500 people), reducing the time spent transmitting invalid information by more than 50% compared to broadcast methods, and improving the speed of early warning response in key areas. Simultaneously, it enables personalized evacuation route generation: Based on a weighted directed graph model and an improved A* algorithm, it generates optimal evacuation routes by comprehensively considering terrain costs, flood parameters, and population density, avoiding dangerous areas with water depth ≥1.5m or current velocity ≥2m / s. It also provides estimated travel time and turning point coordinates, offering customized escape guidance for people in different locations. For example, the route planning for evacuating people from a village to a high-altitude shelter took only 25 minutes, a 30% reduction compared to traditional experience-based routes. It also introduces a secondary early warning mechanism: by monitoring the status of people staying in the area in real time, it automatically triggers voice calls or emergency contact notifications for abnormal personnel who stay in unsafe areas for more than 30 minutes, and links drones to provide guidance, effectively solving the problem of blind spots in the coverage of traditional early warning for people who "have not received information" or whose "movement is hindered".

[0261] In summary, this invention breaks through the bottlenecks of traditional water disaster early warning in terms of data fusion, simulation accuracy, assessment dimensions, and response efficiency, and provides a high-precision, intelligent, and implementable solution for smart water conservancy and emergency management, with significant social value and engineering application prospects.

[0262] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Those skilled in the art, upon considering the disclosure in the specification and embodiments, will readily conceive of other embodiments of this disclosure. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and embodiments are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the claims.

[0263] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A method for precise early warning of dynamic monitoring of water conservancy disasters, characterized in that, Includes the following steps: Multi-source data fusion modeling is carried out in advance: high-precision DEM and point cloud data are obtained through high-resolution satellites and UAV LiDAR, and a multi-dimensional database containing topography, hydrology and personnel distribution is constructed based on hydrological monitoring and communication signaling data; The simulation of small watershed disaster evolution includes: dividing the target small watershed into microgrids, calculating the topographic index and runoff rate based on TOPMODEL; and constructing a one-dimensional and two-dimensional hydrodynamic coupling model using a one-dimensional channel model and a two-dimensional inundation model, achieving energy conservation through flow interpolation, simulating the flood evolution process, and outputting inundation raster map and flow velocity distribution data. The DPC clustering algorithm and KDE kernel density estimation method are used to process personnel coordinate data, generate personnel thermal field distribution, and calculate personnel thermal values. A three-dimensional risk assessment matrix is ​​constructed using water depth, flow velocity, and human thermal values ​​as three dimensions, and the warning level is determined based on the three-dimensional risk assessment matrix. Based on the R-tree spatial indexing algorithm, communication base stations associated with disaster areas are quickly matched. The TOPSIS algorithm is used to sort the associated base stations and push early warning information to base stations covering high-risk areas first. A weighted directed graph model is constructed to generate personnel transfer paths by integrating terrain costs, flood parameters, and personnel density, and the personnel location is monitored in real time. A secondary warning is triggered for personnel who stay in unsafe areas for more than a preset time. The step of prioritizing the delivery of early warning information to base stations covering high-risk areas includes the following steps: The base station set B obtained through R-tree query is used as the input candidate base station set; The evaluation indicators are defined as follows: number of covered users C1, average signal strength C2, and historical early warning success rate C3, and their standardized matrix; Wherein, the number of users covered, C1, is the number of users covered by the base station, standardized as follows: ； The mean signal strength C2, the base station signal strength, is standardized as follows: ； Historical early warning success rate C3, percentage value, standardized as follows: ； Calculate the weighted matrix and the ideal solution; The weighted matrix is ​​represented as W=[0.5,0.3,0.2], where the number of users has the highest weight. The ideal solution is expressed as: ; The negative ideal solution is expressed as: ; Distance calculation and sorting are performed, where the distances between base station i and the positive and negative ideal solutions are respectively: ； ； Get the overall score , and according to The base station priority is obtained by sorting the base stations in descending order for accurate push of early warning information, represented as: 。 2. The method for precise early warning of dynamic monitoring of water conservancy disasters according to claim 1, characterized in that, The communication signaling data includes: parsing the communication signaling and locating personnel, including the following steps: The coordinates of the three base stations are represented as follows: The signal arrival time difference is expressed as follows: speed of light The distance difference is calculated and expressed as: ； ； Construct the equation of the hyperbola and obtain the user's coordinates (x, y); Among them, base station 1 and base station 2 form a hyperbola, which can be represented as: ； Among them, base station 1 and base station 3 form a hyperbola, represented as: 。 3. The method for precise early warning of dynamic monitoring of water conservancy disasters according to claim 2, characterized in that, Also includes: Performing trajectory smoothing and movement trend prediction on personnel location data includes the following steps: State-space model construction includes: calibrating the four-dimensional state vector. ,in, The real-time coordinates obtained through communication signaling parsing are represented as initial values. Let F be the horizontal velocity. Based on uniform motion, the state transition matrix F is expressed as: ； in, The corresponding high-frequency sampling interval is 1 time / minute. The observation matrix H extracts the coordinate information from the state vector to match the actual observation value, as shown below: ； Performing recursive prediction involves the following steps: Calibrate the state of the previous moment Predicting the current state is represented as: ; The updated prediction error covariance matrix is ​​expressed as: ; Where Q is the process noise matrix, which characterizes the uncertainty of the motion model, i.e., the speed variation of the personnel; If valid observations are received The observed residuals are calculated and expressed as follows: ; Through Kalman gain K t The combined predicted and observed values ​​are represented as follows: ； in, To observe the noise covariance matrix; The corrected state vector and covariance matrix are expressed as follows: ； When signal obstruction leads to missing observations, a predicted state is used. As the current estimate, to avoid trajectory interruption.

4. The method for precise early warning of dynamic monitoring of water conservancy disasters according to claim 1, characterized in that, The simulation of small watershed disaster evolution includes: TOPMODEL simulation, which includes the following steps: The small watershed is divided into 500m × 500m microgrids, and the topographic index is calculated and expressed as: ； Determine the critical topographic index Through the water balance equation Iterative solution; The flow rate is calculated and expressed as: ； in, Slope, A is catchment area, Where i represents soil moisture and i represents rainfall intensity. It represents the saturated hydraulic conductivity.

5. The method for precise early warning of dynamic monitoring of water conservancy disasters according to claim 1, characterized in that, Also includes: One- or two-dimensional hydrodynamic coupling model simulation includes the following steps: One-dimensional river channel model simulation: including: based on the Saint-Venant equations, using the Preissmann four-point implicit scheme to discretize the equations, dividing the river channel into N cross-sections, and the time step. Spatial step size ; The equation is discretized and expressed as: ； ； Where Q is the cross-sectional discharge, A is the cross-sectional area, and S0 is the riverbed slope. The friction slope is expressed as: Where n is the Manning roughness coefficient and R is the hydraulic radius; The two-dimensional flooding model simulation includes the following steps: The finite volume method was used for discretization, and the region was divided into a 500m×500m grid. The flux was solved using the Roe scheme, considering the ground roughness n=0.

03. By establishing a coupling interface and interpolating the one-dimensional river cross-sectional flow with the two-dimensional grid boundary flow, energy conservation is achieved, expressed as: ； Where M is the number of boundary elements in the two-dimensional mesh, u i ,h i The grid flow velocity and water depth are given.

6. The method for precise early warning of dynamic monitoring of water conservancy disasters according to claim 1, characterized in that, The calculation of the person's thermal value includes the following steps: Density peak clustering algorithm is used to identify core points of people gathering, including: The set of base station signaling coordinates is as follows Cut-off distance ; The local density is calculated as follows: ； in, This represents local density, reflecting the number of people within a 500m radius of point i. The larger the value, the higher the degree of aggregation; Calculate distance metrics: The core point is represented as follows: ; The edge points are represented as follows: ; in, The distance metric is used to identify cluster centers. Core points are determined by the minimum distance to their nearest high-density points, while edge points are determined by the global maximum distance. Perform cluster center identification and select and Points with higher average values ​​are used as cluster centers to delineate areas where people congregate. Kernel density estimation includes the following steps: Personnel coordinates after clustering Bandwidth h = 500m; The thermal value is calculated and expressed as: 。 7. The method for precise early warning of dynamic monitoring of water conservancy disasters according to claim 1, characterized in that, The process of triggering a secondary warning for users who remain in the area for more than a preset time includes the following steps: The system monitors changes in personnel location in real time. If a person lingers in an unsafe area for more than a preset time and does not move along the planned path, it is considered an abnormal stay and a secondary warning is triggered.

Images