An edge-computing-based distributed waterlogging digital map prediction system

Abstract

The application discloses a kind of distributed waterlogging accumulation digital map prediction systems based on edge computing, it is related to urban flood control intelligent decision-making technical field.The system includes distributed intelligent acquisition module, edge integration processing module, multistage collaborative analysis module and visualization and decision output module.Through the acquisition of multiple source heterogeneous data of multiple sensing devices;Call local digital elevation map slice and light hydrological model, generate local waterlogging state information and self-calibrate model parameters;Based on information entropy variation and water depth change rate dual threshold screening key information, fusion multi-monitoring point data and drainage pipe network topological relationship, realize regional waterlogging situation analysis and accumulation water recession trend prediction;Generate and update the distributed water accumulation digital map of whole region, output graded early warning and auxiliary decision information.The application combines distributed acquisition and edge computing, improves monitoring real-time and precision, effectively solves the problems of traditional monitoring, such as high delay, low precision and fuzzy decision.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of urban flood control intelligent decision-making, more specifically, the present application relates to a distributed waterlogging digital map prediction system based on edge computing. BACKGROUND

[0002] In the current urban flood control management and waterlogging prevention and control field, the mainstream technical solution adopts a basic framework of fixed-point sensing collection and central storage. Single type monitoring devices such as water level meters and rain gauges are deployed in waterlogged areas to collect hydrological data, which is transmitted to a remote server to only realize basic data statistics or simple early warning prompts, providing preliminary reference for flood control work. Although some solutions introduce electronic maps and simple numerical models, data processing only stays at the basic level of de-duplication and filtering, without deep integration of core management elements such as drainage pipe network topology and terrain parameters, and without improving data reliability through dynamic model calibration, resulting in a lack of spatio-temporal synchronization and confidence assessment of multi-source data such as water level, rainfall, flow rate and images, making it difficult to convert into decision-making information supporting pump station regulation and control, personnel transfer and other actual management scenarios. A few advanced solutions introduce video monitoring for auxiliary observation, but the devices operate independently and the data are not associated with each other, which are not deeply adapted to the needs of flood control decision-making, and cannot form regional-level overall research and judgment and precise scheduling solutions.

[0003] But there are still many outstanding defects in the actual flood control management decision application: first, the data value conversion is insufficient, various types of sensor data are stored and run independently, and the structured decision data is not formed through standardized preprocessing and cross-validation, which leads to the disconnection between data and pump station start, gate control and other scheduling needs, and it is difficult to serve the regional overall management; second, the decision model has poor adaptability, the traditional experience type or fixed parameter model is not combined with historical water accumulation data for dynamic self-calibration, which cannot adapt to different terrain, surface coverage type and drainage conditions, resulting in large calculation error of decision basis, which is difficult to support accurate decision; third, the decision feedback is seriously lagging behind, the massive raw data depends on the centralized processing of the remote center, which not only occupies a lot of communication bandwidth, but also is easily affected by network delay, which cannot meet the core needs of real-time scheduling and rapid response of flood control management, and it is difficult to deal with waterlogging emergency; fourth, the information screening mechanism is missing, a large amount of invalid data and low-quality data under stable state occupy transmission and calculation resources, resulting in low decision efficiency, and it is difficult to focus on rapid judgment of high-risk situation; fifth, the regional overall decision is missing, the core parameters such as pipe diameter, flow direction and design flow of the drainage pipe network topology relationship are not associated with the terrain grid data, which cannot identify the water accumulation spreading path and pipe network blockage risk point through path algorithm, and it is difficult to form a cross-regional situation judgment and resource optimization configuration scheme; sixth, the decision scheme lacks pertinence, the early warning system is rough with two or three levels, the trigger condition only depends on the single index of water depth, the auxiliary decision information is mostly macro-guiding suggestion, and there is no operable scheme for specific management scenarios such as pump station start power calculation, gate control strategy, partition emergency disposal such as road closure and personnel transfer, which is difficult to adapt to the management needs of city flood control refinement and actual combat. SUMMARY

[0004] In order to overcome the above defects of the prior art, the embodiments of the present application provide a distributed waterlogging accumulation digital map prediction system based on edge computing, which solves the problems of single data collection, high processing delay, low monitoring accuracy, one-sided situation analysis, inaccurate prediction and insufficient decision support in the background art through the following scheme.

[0005] In order to achieve the above purpose, the present application provides the following technical scheme: a distributed waterlogging accumulation digital map prediction system based on edge computing, comprising a distributed intelligent collection module: distributed in urban prone-to-flooding monitoring points, collecting multi-source heterogeneous data reflecting waterlogging state through preset hydrological and environmental sensing equipment, and generating original data set through preprocessing;

[0006] The multi-source heterogeneous data includes water level information, rainfall information, flow rate information and image information;

[0007] Edge integration processing module: for receiving the original data set, and calling the pre-stored local digital elevation map slice and embedded lightweight hydrological model, generating local waterlogging state information through calculation, and self-calibrating the model parameters by using historical water accumulation data;

[0008] Multi-level collaborative analysis module: for value evaluation and screening of the local waterlogging state information according to the double threshold of information entropy change and water depth change rate, obtaining key state information; and fusing the key state information of multiple monitoring points on the center side, combining the pre-set drainage pipe network topology, performing regional waterlogging situation analysis and accumulated water recession trend prediction;

[0009] Visualization and decision output module: for generating and updating a distributed water accumulation digital map covering the whole region according to the results of the regional waterlogging situation analysis and accumulated water recession trend prediction, and generating hierarchical warning information and auxiliary decision information based on the map water accumulation data.

[0010] Technical effects and advantages of the present application:

[0011] 1、The present application significantly improves the real-time and accuracy of waterlogging state calculation by constructing an edge integration processing architecture and designing a model dynamic self-calibration mechanism. The system combines local storage of digital elevation map slices and embedded deployment of lightweight hydrological models, allowing edge nodes to directly complete data processing and waterlogging state calculation, greatly reducing dependence on the center server and data transmission delay, and realizing rapid response of waterlogging information; at the same time, a calibration mode triggered by timing and events is adopted, and the core parameters of the model are iteratively optimized combined with historical effective data, so that the model can adapt to different topography, land cover types and drainage conditions, solving the problem of poor adaptability and large calculation error of traditional fixed parameter models, and embodying the creative integration of edge computing and dynamic modeling;

[0012] 2、The present application optimizes system operation efficiency and realizes accurate judgment of global waterlogging situation by establishing a double threshold screening mechanism of information entropy change and water depth change rate and a regional level collaborative analysis mode. The system scientifically screens high-value key state information through double threshold, filters invalid data in stable state, reduces resource occupation, and improves transmission and calculation efficiency; on the center side, through spatio-temporal fusion of multi-monitoring point data, combined with the topology of drainage pipe network, water accumulation spreading path identification, risk zoning and pipe network blockage warning are carried out, and multi-class features are fused for accumulated water recession trend prediction, creatively breaking through the limitations of traditional single-point monitoring analysis, realizing overall planning and judgment from local data to regional level global situation, and providing scientific basis for early prevention and control;

[0013] 3. This invention provides intuitive and efficient technical support for urban flood control by constructing a full-area distributed digital map visualization system for waterlogging and a precise decision-making output mechanism. The system maps real-time flood data and prediction results onto a geographic base map, overlaying key information such as risk zones and spread paths. A dynamic update mode ensures map timeliness, making the overall flood status intuitively perceptible. Simultaneously, it establishes a tiered early warning system supported by multi-dimensional indicators and generates personalized solutions for different risk areas, including drainage scheduling strategies and emergency response suggestions. These solutions cover specific measures such as pump station control, gate operation, road management, and personnel evacuation. Its innovation lies in the deep integration of visualization and practical decision-making, solving the problems of general early warnings and lack of specific decision-making suggestions in traditional systems, and comprehensively meeting the needs of refined urban flood control. Attached Figure Description

[0014] Fig. 1 This is a schematic diagram of the overall structure of the present invention;

[0015] Fig. 2 This is a schematic diagram of the process for obtaining local flooding status information according to the present invention;

[0016] Fig. 3 This is a functional flowchart of the multi-level collaborative analysis module of the present invention;

[0017] Fig. 4 This is a functional flowchart of the visualization and decision output module of the present invention. Detailed Implementation

[0018] 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. 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.

[0019] As attached Figs. 1 to 4 The distributed digital map prediction system for urban flooding based on edge computing, shown below, includes:

[0020] Distributed intelligent data acquisition module: Collects multi-source heterogeneous data reflecting the state of urban flooding through preset hydrological and environmental sensors, and generates raw datasets through preprocessing;

[0021] The multi-source heterogeneous data includes water level information, rainfall information, flow velocity information, and image information;

[0022] It should be specifically noted that multi-source heterogeneous data reflecting the state of urban flooding are collected through pre-set hydrological and environmental sensing equipment. The specific process is as follows:

[0023] Water level information: ultrasonic water level meter is used as the core collection equipment, the type selection meets the technical indicators of measurement range 0 to 5 meters, accuracy positive and negative 0.5% FS, response time not more than 50 milliseconds, with IP67 and above protection level to adapt to harsh environment; The device is deployed at the low-lying place of the road, the entrance of the underground passage and other key positions in the city prone to flooding, installed with a diameter of 89 mm galvanized steel pipe stand, the sensor probe is 0.3 meters away from the ground reference surface, and a 0.5m*0.5m*0.6m concrete base is poured at the bottom of the stand to ensure stability; The installation position avoids the direct overhead of the drain, and the distance from the curbstone is not less than 0.8 meters, avoiding external interference affecting the measurement accuracy; The collection frequency is dynamically adjusted according to the water accumulation state, and when the water depth change rate is not more than 0.05 meters per minute, it is 1 time per second, and when the water depth change rate exceeds 0.05 meters per minute, it is automatically upgraded to 2 times per second; The collected data directly outputs the water depth value, unit: meters, with the device unique ID and collection timestamp.

[0024] Rainfall information: tipping bucket rain gauge is selected, with a resolution of 0.2 mm, a measurement range of 0 to 4 mm per minute, and a rain intensity error of not more than plus or minus 4%, also with IP67 protection level and working temperature adaptation range of minus 20 to 60 degrees Celsius; The device is horizontally installed at the top of a 1.5-meter-high stand around the monitoring point, and a 0.8-meter-diameter wind shield is added to the stand to ensure that the rain falls vertically into the tipping bucket, with an installation level error of not more than 0.5 degrees to avoid measurement deviation; During the collection process, the cumulative rainfall and instantaneous rainfall intensity are recorded in real time, the cumulative rainfall unit is millimeter, the instantaneous rainfall intensity unit is millimeter per hour, and the collection interval is fixed at 1 time per second, and under no rain conditions, it keeps low power consumption running, when the rainfall intensity is not less than 0.5 mm per hour, the intelligent wake-up mechanism of other sensing devices is triggered immediately, and the overall collection frequency is synchronized.

[0025] Flow rate information: non-contact radar flow meter is used, with a measurement range of 0.1 to 10 meters per second, an accuracy of plus or minus 1% FS, and a blind area of not more than 0.3 meters, supporting stable work in complex water flow environment. The device is wall-mounted on the outlet of the drainage pipe network or the side wall of the river through an L-shaped bracket, the installation height is set to 1.2 times the water depth of the measurement section, the lens is vertically aligned with the water flow direction, and the deviation from the water flow center line is not more than 5 degrees to ensure effective signal reception; The collection frequency is synchronized with the water level information, 1 time per second under normal conditions and 2 times per second under abnormal conditions, the collected data includes instantaneous flow rate and cumulative flow, the instantaneous flow rate unit is meters per second, and the cumulative flow unit is cubic meters, the data output automatically filters the instantaneous abnormal values caused by water flow vortex, preliminarily ensuring data reliability.

[0026] Image information: Configure high-definition camera group, resolution up to 1920x1080, frame rate 25 frames per second, support night vision distance not less than 30 meters, with IP67 protection level and automatic wiper function; The equipment is installed on an 8-meter-high fixed support, the lens has a 30-degree to 45-degree downward angle, ensuring complete coverage of the prone to flooding monitoring area without obstruction and blind area; The lens automatic wiper starts once every 10 seconds when it rains to avoid rainwater and dust obstruction affecting image clarity; The collection frequency is dynamically adjusted according to the working conditions, normal working conditions 3 frames per second, abnormal working conditions up to 5 frames per second, the collected image format is JPEG, each frame of image is attached with collection timestamp, device ID and shooting angle information, automatically switches to infrared night vision mode at night to ensure all-weather collection effect.

[0027] It should be further explained that the original data set is generated by preprocessing, and the specific process is as follows:

[0028] The edge collection gateway first calibrates all sensor data in space and time, ensures timestamp consistency through GPS and NTP dual-source time calibration, controls time deviation within ten milliseconds, and realizes data space correlation combined with pre-stored monitoring point coordinates; Then filter the outliers in numerical data such as water level, rainfall and flow rate based on the Laplace criterion, i.e. 3σ criterion, filter fuzzy image frames through image sharpness evaluation, trigger self-check and alarm mechanism for devices that output abnormal data for three times in a row; Then standardize the effective data in a unified format to generate the original data set S raw (t), and assign a confidence label Conf based on signal strength and historical stability to each piece of data. The confidence label value ranges from 0 to 1, and marks low-quality data with a confidence label value below 0.3 and reduces its upload priority.

[0029] Edge integrated processing module: used to receive the original data set, and call pre-stored local digital elevation map slices and embedded lightweight hydrological model to generate local waterlogging state information through calculation, and use historical waterlogging data to self-calibrate model parameters;

[0030] It should be specifically explained that the pre-stored local digital elevation map slices and embedded lightweight hydrological model are called to generate local waterlogging state information through calculation, and the specific process is as follows:

[0031] The edge node adopts a "center distribution + local on-demand storage" map management mode. The center platform pre-distributes the corresponding full-area high-precision digital elevation map (DEM) of the jurisdictional area to each edge node. The data format is GeoTIFF, and the resolution is 1 meter. The edge node cuts the full-area DEM into 100m x 100m vector slices through the map slicing engine, and only retains the slice data and corresponding terrain parameter matrix within its own jurisdiction, including ground elevation D(x, y), pipe or river bottom elevation Z(x, y), and surface cover type core parameters. When receiving the original data set, the edge node automatically matches and calls the corresponding regional map slice and terrain parameter matrix according to the monitoring point coordinate information attached to the data.

[0032] The edge node receives the original data set S raw After (t), the secondary cleaning and completion processing is performed first: for the valid data with confidence label Conf ≥ 0.3 but with missing data, the linear interpolation method is used to complete the data together with the same period data of adjacent monitoring points in the same area; for the data with Conf lower than 0.3, it is directly excluded to ensure that the data validity of the input model is not less than 95%; after processing, the purified data set S clean (t) is standardized and stored in the structure of "timestamp-coordinate-data type-value-confidence", where the numerical data retains 3 decimal places, and the image data extracts key feature such as water accumulation boundary pixel coordinates as auxiliary verification information.

[0033] The embedded lightweight hydrological model is called: taking S clean (t) as input, the embedded lightweight rainfall-water level-terrain (R-W-T) model is called to calculate the water accumulation depth h(x, y, t) of any coordinate (x, y) in the monitoring area at time t, and the core calculation process and formula are as follows:

[0034] Basic parameter initialization: according to the surface cover type in the map slice, including asphalt, concrete, green land, bare soil, etc., the dynamic runoff coefficient a is obtained by table lookup, with a value range of 0.4 to 0.9, 0.9 for asphalt pavement and 0.4 for green land; the basic value of infiltration rate is determined based on soil type data, and the infiltration rate I s (τ) is obtained by combining real-time rainfall, with the infiltration correction coefficient β fixed at 0.85 (participating in model self-calibration); according to the drainage pipe network design parameters and flow velocity information, the drainage flow rate Q out (τ) is calculated, and the drainage efficiency coefficient γ is initially set to 0.75;

[0035] Water accumulation depth core calculation: through integral operation to integrate rainfall, terrain, drainage and other factors, the formula is: where D(x, y) denotes the ground elevation at coordinate (x, y) in meters; Z(x, y) denotes the bottom elevation of the pipe or river at coordinate (x, y) in meters; R'(τ) = R(τ) x 10 -3 / 3600 (R(τ) is the instantaneous rainfall intensity at time τ, extracted from S clean (t), converted to meters per second); I s (τ) is the infiltration rate in meters per second; q out (τ) = Q out (τ) / A (Q out (τ) is the drainage flow in cubic meters per second; A is the area of the drainage region in square meters, converted to meters per second); t0 is the starting time for calculation, and if the integral result is negative, it is taken as 0, indicating no ponding; the model uses the Euler method to simplify the integral operation, and the integral step is consistent with the data collection interval, which is 1 second under normal conditions and shortened to 0.5 seconds under abnormal conditions, with a single grid calculation time not exceeding 1 millisecond;

[0036] Result cross-validation: compare the ponding depth h(x, y, t) calculated by the model with the measured ponding depth collected by the ultrasonic water level meter at the same coordinate. If the absolute error is not more than 0.05 meters, it is directly output; if the error exceeds 0.05 meters, the ponding boundary recognition result of the same period image data is called to assist in correction, ensuring that the deviation between the calculated value and the actual working condition does not exceed 0.1 meters.

[0037] It needs to be further explained that the local waterlogging state information is obtained as follows:

[0038] Based on DEM gridded data with a grid resolution of 1 meter x 1 meter, spatial integration is performed on h(x, y, t) to calculate the local waterlogging core parameters:

[0039] Ponding volume V e (t): the sum of the ponding depth of all grids and the grid area, which is 1 square meter, and the formula is: , in cubic meters, where V e (t) specifically refers to the local ponding volume calculated by the edge node;

[0040] Ponding area A e (t): the number of grids with h(x, y, t) exceeding 0.01 meters is counted, and multiplied by the grid area to obtain A , where δ(z) is an indicator function, δ(z) = 1 when z exceeds 0.01 meters, otherwise δ(z) = 0; A e (t) specifically refers to the local ponding area calculated by the edge node, in square meters;

[0041] Derived parameter calculation: the regional average ponding depth havg (t) = V e (t) / A e (t), unit: meter; water slope ∇h(t) = (h max (t) - h min (t)) / L, where h max (t) is the maximum water depth in the area, h min (t) is the minimum water depth in the area, and L is the longest distance in the water area, unit: meter;

[0042] The above calculation results are integrated into a standard information package Info pack (t), which contains core parameters: water volume V e (t), water area A e (t), average water depth h avg (t), water slope ∇h(t), and rasterized water depth matrix h(x, y, t), while additional processing status codes are attached, 0 indicating normal calculation, 1 indicating calculation after data completion, and 2 indicating temporary model correction calculation.

[0043] It needs to be specifically explained that the model parameters are self-calibrated using historical water data, and the specific process is as follows:

[0044] The edge node locally caches valid historical data for nearly 6 months, including historical raw data and preprocessing records of distributed intelligent collection modules, historical calculation results of local waterlogging state information, and historical waterlogging verification data synchronized by the center platform. Effective data with data integrity not less than 90%, rainfall intensity not less than 1 mm per hour, and continuous monitoring time not less than 2 hours are selected to form not less than 30 sets of calibration data set. If 30 sets of effective data cannot be met within 6 months, the center platform remote calibration mechanism is triggered;

[0045] The calibration adopts a "timing + event" double-triggering mode. The system automatically starts full calibration at low load time at 2 am every day. When the average absolute error between model calculation value and measured value exceeds 0.1 meter or the error exceeds 0.08 meters for 3 times in a row during a single waterlogging process, emergency calibration is triggered. The least squares method is used to construct an error function with the minimum error between model calculation value and historical measured value as the target: Wherein, the meanings of each parameter are as follows: E(α, β, γ) is the total error square sum; n is the number of calibration data sets, i.e. the total number of effective historical data after screening; h meas (i) is the i-th group of historical measured water depth, which comes from ultrasonic water level meter measurement or center platform synchronized manual measurement data; h(x, y, t) i is the water depth calculated based on the current parameters α, β, γ for the i-th group;

[0046] The minimum value of the error function is solved by iterative gradient descent method, the iteration step is set to 0.01, and the iteration is stopped until the parameter change amount of adjacent two times is less than 0.001 or the iteration number reaches 100 times; after calibration, 10 groups of historical data not participating in calibration are selected for verification, and the average absolute error is not more than 0.05 meters to determine the effectiveness, the optimal parameters are stored according to the "region-surface type" classification and synchronized to the center platform; if the error is more than 0.08 meters, the data is reselected and calibrated repeatedly, and if it still does not meet the standard after 3 times, the parameter abnormality alarm is triggered.

[0047] Multi-level collaborative analysis module: for value evaluation and screening of the local waterlogging state information according to the double thresholds of information entropy change and water depth change rate, to obtain key state information; and in the center side, the key state information of multiple monitoring points is fused, combined with the preset drainage pipe network topology relationship, to perform regional waterlogging situation analysis and accumulated water recession trend prediction;

[0048] It should be specifically pointed out that the value of the local waterlogging state information is evaluated and screened according to the double thresholds of information entropy change and water depth change rate, and the key state information is obtained, and the specific process is as follows:

[0049] Information entropy change is used to quantify the uncertainty change of local waterlogging state in unit time, and the more stable the waterlogging state is, the smaller the information entropy change is; when the waterlogging intensifies or recedes, the information entropy change increases; the calculation steps are as follows:

[0050] The accumulated water area A e (t) is divided into 4 intervals according to 0-50㎡, 50-200㎡, 200-500㎡ and above 500㎡, and the accumulated water volume V e (t) is divided into 4 intervals according to 0-100 cubic meters, 100-500 cubic meters, 500-1000 cubic meters and above 1000 cubic meters, and the average accumulated water depth h avg (t) is divided into 4 intervals according to 0-0.1 meters, 0.1-0.3 meters, 0.3-0.5 meters and above 0.5 meters;

[0051] The accumulated water area A e (t), the accumulated water volume V e (t), and the average accumulated water depth h avg (t) are respectively fallen into the corresponding interval probability p A (t), p V (t), p h (t), and the interval probability of a single parameter is the ratio of the historical appearance frequency of the parameter in the interval to the total statistical number in the last 3 months;

[0052] The waterlogging state information entropy H(t) and H(t-1) at t time and t-1 time are calculated by using Shannon entropy formula, and the formula is: In the formula, H(t) is the entropy of the waterlogging status at time t, in bits; Let be the probability that the water area, volume, and average depth fall into the i-th interval at time t, respectively. When the probability is 0, the limit value is 0, i.e., 0・log20=0.

[0053] Information entropy change calculation: ΔH(t)=|H(t)-H(t-1)|, which represents the change range of the waterlogging state between adjacent time points.

[0054] The rate of change in water depth directly reflects the speed at which urban flooding intensifies or recedes. It is calculated based on the average water depth at time t and time t-1, using the following formula: In the formula, Δt is the time interval between two consecutive data packet transmissions, consistent with the data acquisition interval: 1 second under normal operating conditions and 0.5 seconds under abnormal operating conditions. h (t) is in meters per second; h avg (t) and h avg (t-1) represent the average regional water depth at time t and time t-1, respectively, directly from Info pack Extracted from (t).

[0055] It should be further explained that the specific method for setting the dual thresholds is as follows:

[0056] The dual threshold adopts a "baseline threshold + dynamic correction" mode. The baseline threshold is determined based on historical flooding data statistics, and the dynamic correction coefficient is adjusted by combining real-time rainfall intensity and drainage flow velocity.

[0057] Information entropy threshold ΔH th Based on historical data from the past year, the average value of ΔH is 0.02 bits when there is no flooding and 0.08 bits when there is slight flooding. A baseline threshold ΔH is set. base =0.05 bits;

[0058] Introducing a correction factor k H , , where R 实时 The current instantaneous rainfall intensity is from S clean (t) Extraction, R 历史均值 The average rainfall intensity in the region during the same period is pre-stored locally; the final threshold ΔH th =ΔH base ×k H It is controlled within the range of 0.05-0.2 bits.

[0059] Threshold r of water depth change rate hth The threshold is divided into three levels based on the risk of urban flooding, and dynamically adjusted based on real-time flow velocity information.

[0060] Level 1 Threshold (Low Risk): r h1 =10-5 m / s, corresponding to slow changes in water accumulation, correction factor k h1 = 0.8 (when the drainage flow rate ≥ 1 m / s) or 1.2 (when the drainage flow rate < 0.5 m / s);

[0061] Secondary threshold (medium risk): r h2 = 5 x 10 -5 m / s, correction factor k h2 = 0.9 (drainage flow rate ≥ 1 m / s) or 1.1 (drainage flow rate < 0.5 m / s);

[0062] Tertiary threshold (high risk): r h3 = 10 -4 m / s, correction factor k h3 = 1.0, not affected by flow rate, as the high rate of change already characterizes an emergency state.

[0063] It should be further noted that the screening method of the key state information is as follows:

[0064] Index calculation: for each edge node transmission Info pack (t), calculate ΔH(t) and r h (t) synchronously;

[0065] Threshold comparison: if ΔH(t) ≥ ΔH th and r h (t) ≥ r h1 , it is determined that the "state is effectively changed", and the complete information package core parameters are retained; if ΔH(t) ≥ ΔH th and r h (t) ≥ r h2 , it is marked as "medium risk change"; if ΔH(t) ≥ ΔH th and r h (t) ≥ r h3 , it is marked as "high risk change";

[0066] Invalid filtering: if ΔH(t) < ΔH th and r h (t) < r h1 , it is determined that the "state is stable", and only the device ID, timestamp and h avg (t) three core parameters are retained;

[0067] Key state information output: the information screened with risk marks is integrated into the key state information package Key Info (t), the format is "device ID-time stamp-core parameter-risk mark-treatment status code", and is transmitted to the center side in real time.

[0068] It needs to be specifically pointed out that the key state information of the center side fusion of multiple monitoring points is combined with the preset drainage pipe network topological relationship to perform regional waterlogging situation analysis and accumulated water trend prediction, and the specific process is as follows:

[0069] The center side receives the Key of each edge node Info (t), spatio-temporal alignment and data fusion are performed: in time, all data are calibrated to millisecond level based on NTP timestamp, and data with deviation exceeding 10 milliseconds are corrected by linear interpolation; in space, local accumulated water data are mapped to 1 meter x 1 meter grid in combination with pre-stored monitoring point coordinates and drainage pipe network topological map to form regional accumulated water grid matrix H region (t); for multi-source data of the same grid, weighted fusion method is adopted to calculate the final accumulated water depth H fuse (x,y,t)=Σ[w1×w2×h i (x,y,t)] / Σ(w1×w2), wherein w1 is a data confidence weight, i.e., a Conf label value, w2 is a device accuracy weight, an ultrasonic water level meter corresponds to 0.8, a model calculation value corresponds to 0.6, and manual verification data corresponds to 1.0.

[0070] It needs to be further pointed out that the specific way of performing regional waterlogging situation analysis is as follows:

[0071] In combination with the preset drainage pipe network topological relationship, including pipe network direction, pipe diameter, slope, inspection well position, drainage capacity and other parameters, the “grid-pipe network association” method is adopted to perform regional waterlogging situation analysis, and regional waterlogging situation analysis results are obtained;

[0072] The drainage pipe network is abstracted as an undirected graph G=(V,E), wherein the nodes V include monitoring points, inspection wells and pipe network intersection points, and the edges E are pipe sections connecting the nodes; each pipe section is assigned with attributes: pipe diameter d, length L p , design flow Q d , current flow Q c calculated from radar flow meter data, and a pipe network topological adjacency matrix M is constructed, wherein M[i][j] represents the pipe section attribute between nodes i and j;

[0073] The waterlogging situation core analysis includes:

[0074] Accumulated water spreading path identification: based on the pipe network flow direction determined by the pipe section slope and H fuse (x,y,t), the shortest path algorithm (Dijkstra) is adopted to calculate the accumulated water spreading path from the high water level area to the low water level area, and the pipe network blockage risk point is highlighted, and when Q c ≥0.9×Q d , it is determined as blockage warning;

[0075] Regional waterlogging risk classification: calculate the total water accumulation volume V total (t)=∑∑H fuse (x,y,t)×1×1, average water accumulation depth H totalavg =V total (t) / A total (t), A total (t) is the total area of water accumulation in the region, combined with the spread path, the region is divided into high-risk areas, medium-risk areas, and low-risk areas, wherein the high-risk area is H fuse (x,y,t)≥0.5 meters or located upstream of the blockage warning point, the medium-risk area is 0.3 meters≤H fuse (x,y,t)<0.5 meters and located on the spread path, and the low-risk area is H fuse (x,y,t)<0.3 meters and away from the spread path.

[0076] Finally, the regional waterlogging situation analysis results are obtained: including the rasterized water accumulation data after time and space fusion, the risk zoning map, the water accumulation spread path, the pipe network blockage warning information, and the core indicators, including the total water accumulation volume V total (t), the average water accumulation depth H totalavg , the area and proportion of each risk area.

[0077] It should be further explained that the specific way to perform the water accumulation trend prediction is as follows:

[0078] The long short-term memory network (LSTM) is selected as the basic prediction model, the input features are selected as the historical time series features, the driving factor features, and the pipe network features, wherein the historical time series features are the time series of H fuse (x,y,t), V e (t), and A e (t) in the past 1 hour, the driving factor features are the real-time rainfall intensity R(τ), the drainage flow rate v(τ), and the infiltration rate I s (τ), and the pipe network features are the current flow Q c and the design flow Q d of the pipe section; the model output is H fuse (x,y,t+ΔT) in the future 36 time steps, ΔT=10 minutes, covering the next 6 hours, the model is trained by the historical data in the past 3 months, the mean square error (MSE) is used as the loss function, and the model is deployed locally after training;

[0079] Combined with the embedded hydrological model principle: if the meteorological department pushes the future rainfall warning, the pre-stored interface is obtained, then the prediction result is corrected according to the rainfall intensity increase, the increase is ; if there is a blockage warning in the pipe network, multiply the predicted water accumulation depth of the corresponding region by the correction coefficient k Q =1.1(Q c≥ 0.9 x Q d ) when the blockage has been processed, k Q = 0.95;

[0080] Corrected prediction value: H pred (x, y, t + ΔT) = H model (x, y, t + ΔT) x k R x k Q , where H model is the original output of the LSTM model;

[0081] Finally, the accumulated water trend prediction result is obtained: a regional average accumulated water depth trend curve is displayed by time step, a regional accumulated water prediction graph for the next 6 hours, a trend state, and a prediction confidence. The trend state is marked as "intensifying", "stable", or "subsiding", and the prediction confidence is calculated based on historical prediction accuracy, ranging from 0 to 1.

[0082] The visualization and decision output module is used to generate and update a distributed accumulated water digital map covering the entire region based on the results of the regional waterlogging situation analysis and accumulated water trend prediction, and to generate hierarchical warning information and auxiliary decision information based on the map accumulated water data.

[0083] It should be specifically noted that the distributed accumulated water digital map covering the entire region is generated and updated as follows:

[0084] The basic geographic base map uses the city electronic map pre-stored by the central platform, with a coordinate system of WGS84 and a resolution of 1 meter x 1 meter, containing static elements such as roads, buildings, drainage networks, and monitoring points.

[0085] When the map is generated, H fuse (x, y, t) and H pred (x, y, t + ΔT) are mapped to the basic base map according to coordinates to form a real-time accumulated water layer and a predicted accumulated water layer; then a risk zoning layer is superimposed, with high-risk areas filled with red, medium-risk areas filled with orange, and low-risk areas filled with yellow. Blockage warning points are marked with red triangular symbols, and accumulated water spreading paths are marked with yellow dashed lines.

[0086] The visualization rendering is color-coded according to accumulated water depth, with 0-0.1 meters as light blue, 0.1-0.3 meters as blue, 0.3-0.5 meters as orange, and no less than 0.5 meters as red. The color depth increases with the accumulated water depth.

[0087] The map update adopts a "timed + event triggered" dual mode, with real-time data synchronized every 1 second under normal conditions, and the data is updated every 10 seconds under abnormal conditions, i.e., r h (t) ≥ r h2is updated every 0.5 seconds; when the area of the high-risk zone increases by no less than 10%, a new blockage warning point is added, or the state of the water accumulation trend changes, an instant update is triggered, and the update delay is no more than 1 second.

[0088] It should be specifically pointed out that the hierarchical warning information and the auxiliary decision information are generated based on the map waterlogging data, and the specific process is as follows:

[0089] The hierarchical warning information is generated based on the average waterlogging depth H total_avg , the proportion of the risk zone, and the water accumulation trend in the region in the map, and a four-level warning system is established.

[0090] The first-level warning (red) is determined when H total_avg ≥ 0.5 meters, the proportion of the high-risk zone is no less than 30%, the trend is "intensifying", or the number of blockage points is no less than 3.

[0091] The second-level warning (orange) is determined when 0.3 meters ≤ H total_avg < 0.5 meters, the proportion of the high-risk zone is 15%-30%, the trend is "intensifying", or the number of blockage points is 1-2.

[0092] The third-level warning (yellow) is determined when 0.1 meters ≤ H total_avg < 0.3 meters, the proportion of the medium-risk zone is no less than 40%, or the trend changes from "stable" to "intensifying".

[0093] The fourth-level warning (blue) is determined when H total_avg < 0.1 meter, but the local waterlogging depth is no less than 0.3 meters.

[0094] The warning information includes the warning level, the triggering index, the affected area, and the trend prediction.

[0095] The auxiliary decision information includes the drainage dispatching scheme and the emergency disposal suggestion:

[0096] The drainage dispatching scheme is generated based on the pipe network topology relationship in the map, the current flow Q c , and the design flow Q d , and the number and power of the pump stations are calculated according to the principle of "high-risk zone first, upstream of the blockage point first", and the formula is , where ρ = 1000 kg / m3 (density of water), g = 9.8 m / s ² (gravity acceleration), Q = V total (t) × r h (t) / h avg(t) (discharge flow, unit: cubic meters / second), H is the discharge lift (m, pre-stored pipe network design lift), η=0.7 (pump efficiency, adjusted according to actual working condition), 1.3 is the discharge safety factor, and gate control suggestions are given, such as closing the gate of the risk area of accumulated water backflow and opening the downstream diversion gate;

[0097] The emergency disposal suggestions are made for different risk areas, the high-risk area is suggested to block the surrounding roads, and the personnel and equipment in the low-lying area are transferred, the medium-risk area is suggested to strengthen road patrol and clean up the sundries in the drainage outlet, the low-risk area is suggested to lay temporary drainage facilities and preposition emergency water pumps, and meanwhile, emergency shelters and rescue routes are marked to support rapid disposal.

[0098] Secondly: the drawings in the disclosed embodiments of the application only involve the structures involved in the disclosed embodiments, other structures can refer to the usual design, and in the case of no conflict, the same embodiment and different embodiments of the application can be combined with each other;

[0099] Finally: the above only describes the preferred embodiments of the application and is not used to limit the application, and any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the application should be included in the protection scope of the application.

Claims

1. A distributed digital map prediction system for urban flooding based on edge computing, characterized in that, include: Distributed intelligent data acquisition module: Deployed in a decentralized manner at urban flood-prone monitoring points, it collects multi-source heterogeneous data reflecting the flooding status through preset hydrological and environmental sensors, and generates raw datasets through preprocessing; The multi-source heterogeneous data includes water level information, rainfall information, flow velocity information, and image information; Edge integrated processing module: used to receive the original dataset, call the pre-stored local digital elevation map tiles and embedded lightweight hydrological model, generate local waterlogging status information through calculation, and use historical water accumulation data to self-calibrate the model parameters; The local digital elevation map tiles are accessed in the following way: a central distribution plus local on-demand storage mode is adopted. The central platform distributes digital elevation maps to edge nodes, and the edge nodes cut them into vector tiles through the tiling engine, retaining only the tile data and terrain parameter matrix within their jurisdiction. After receiving the original dataset, the edge nodes automatically match and call the map tiles and terrain parameter matrix of the corresponding area according to the coordinates of the monitoring points. The embedded lightweight hydrological model is invoked as follows: First, the edge nodes perform secondary cleaning and completion on the original dataset to generate a purified dataset and extract key image features as auxiliary verification information; then, the precipitation-water level-topography model is invoked with the purified dataset as input to obtain the dynamic runoff coefficient, correct the infiltration rate, and calculate the drainage flow. The water depth is calculated grid-by-grid by integrating multiple factors; finally, the water depth calculated by the model is compared with the measured water depth. When the error exceeds the set value, the water boundary recognition results of the image data from the same period are invoked to assist in correction. Multi-level collaborative analysis module: It is used to evaluate and filter the local waterlogging status information based on the dual thresholds of information entropy change and water depth change rate to obtain key status information; and integrate the key status information from multiple monitoring points at the central side, combined with the preset drainage network topology, to perform regional waterlogging situation analysis and water receding trend prediction. The key status information is obtained as follows: the water accumulation area, water accumulation volume, and average water accumulation depth are divided into four intervals, the probability of each parameter falling into the corresponding interval is calculated, the Shannon entropy formula is used to calculate the waterlogging status information entropy and information entropy change at adjacent times, and the water accumulation depth change rate is calculated based on the average water accumulation depth and time interval at adjacent times. The dual threshold adopts a baseline threshold plus a dynamic correction mode. The information entropy change threshold is adjusted in combination with real-time and historical rainfall intensity, and the water depth change rate threshold is divided according to risk level and adjusted in combination with drainage flow velocity. Parameters are selected and risk levels are marked according to the comparison results of the dual threshold. Visualization and Decision Output Module: Based on the results of the regional flood situation analysis and water receding trend prediction, it generates and updates a distributed digital map of waterlogging covering the entire region, and generates hierarchical early warning information and auxiliary decision-making information based on the map waterlogging data.

2. The distributed digital map prediction system for urban flooding based on edge computing according to claim 1, characterized in that: The multi-source heterogeneous data is collected in the following ways: Water level information: Data is collected using ultrasonic water level gauges, which are deployed at key locations in flood-prone monitoring points, and the data output is the water depth value; Rainfall information: A tipping bucket rain gauge was used to collect data, which was installed horizontally on the top of the poles around the monitoring point. The data output included the cumulative rainfall and the instantaneous rainfall intensity. Flow velocity information: Data is collected using a non-contact radar flow meter, which is installed at the outlet of the drainage network and on the side wall of the river. The data output includes instantaneous flow velocity and cumulative flow. Image information: Acquired by a high-definition camera array mounted on a fixed bracket, covering flood-prone monitoring areas.

3. The distributed digital map prediction system for urban flooding based on edge computing according to claim 1, characterized in that: The method for obtaining local flooding status information is as follows: Based on the rasterized data of digital elevation map, spatial integration is performed on the water depth. The water volume is the sum of the raster water depth and the raster area. The water area is the number of raster cells whose water depth exceeds the set value and multiplied by the raster area. The average water depth and water slope of the area are calculated simultaneously. The water slope is the difference between the maximum and minimum water depth of the area divided by the longest distance of the waterlogged area.

4. The distributed digital map prediction system for urban flooding based on edge computing according to claim 1, characterized in that: The specific self-calibration process is as follows: The calibration adopts a dual-trigger mode of timed and event-triggered calibration, with the goal of minimizing the error between the model calculated value and the historical measured value. The error function is constructed by the least squares method, and the optimal parameters are solved iteratively by the gradient descent method. After calibration, historical data that did not participate in the calibration are selected for verification. If the error meets the requirements, the optimal parameters are stored and synchronized to the central platform. Otherwise, recalibration is performed. If the error fails to meet the requirements multiple times, an alarm for abnormal parameters is triggered.

5. A distributed digital map prediction system for urban flooding based on edge computing as described in claim 1, characterized in that: The specific process of the waterlogging situation analysis is as follows: After receiving key status information, the central side spatially maps local water accumulation data to a grid to form a regional water accumulation grid matrix. For multi-source data of the same grid, a weighted fusion method is used to calculate the final water accumulation depth. Combined with the drainage network topology, a network topology adjacency matrix is ​​constructed. The shortest path algorithm is used to identify the water accumulation spread path. When the current flow of a pipe section reaches the set proportion of the design flow, it is determined as a blockage warning. The total water accumulation volume and average water accumulation depth of the area are calculated. Risk areas are divided in combination with the spread path.

6. The distributed digital map prediction system for urban flooding based on edge computing according to claim 1, characterized in that: The specific process of predicting the water accumulation and receding trend is as follows: a long short-term memory network is selected as the prediction model to output the predicted water accumulation depth for multiple future time steps, covering the next six hours; the prediction results are corrected by combining rainfall warnings and pipeline blockage status. When rainfall warnings are issued, the correction is made according to the increase in rainfall intensity. When pipeline blockage occurs, the correction is multiplied by the corresponding correction coefficient. If the blockage has been dealt with, another correction coefficient is used. The corrected prediction results are then obtained.

Images