Computer vision-based waterlogging simulation optimization and intelligent monitoring method
By reshaping land use data using computer vision technology and algorithms, and combining it with a one-dimensional and two-dimensional surface coupling model, the accuracy and computational resource problems of traditional hydrological and hydrodynamic models in simulating urban flooding have been solved, achieving high-precision flooding simulation and intelligent monitoring.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HEBEI UNIV OF ENG
- Filing Date
- 2026-04-24
- Publication Date
- 2026-06-26
Smart Images
Figure CN122287250A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of flood model monitoring technology, and in particular relates to a computer vision-based method for urban flood simulation optimization and intelligent monitoring. Background Technology
[0002] With the acceleration of global climate change and urbanization, extreme rainfall events are becoming more frequent, and urban flooding is becoming increasingly serious. Traditional stormwater drainage network design standards are struggling to meet the drainage needs of high-density urban areas, leading to frequent problems such as excessive water depth in some areas and network overload. To address these issues, various stormwater models are now being used to simulate and assess rainfall risks, such as GIS-based stormwater flood models and one-dimensional stormwater models based on the Storm Runoff Management Model (SWMM).
[0003] Currently, hydrological and hydrodynamic models are the core tools for simulating urban flooding, but their application faces two major challenges: First, one-dimensional pipe network models are difficult to depict the interaction process between two-dimensional surface runoff and pipe network overflow, while high-precision two-dimensional models have problems such as high computational resource consumption and complex calibration parameters; Second, the model input data (such as surface land use types) mostly rely on remote sensing interpretation or manual surveys, with limited classification accuracy and lagging updates, resulting in significant deviations between simulation results and actual water accumulation distribution. Summary of the Invention
[0004] To address the aforementioned shortcomings in existing technologies, this invention provides a computer vision-based method for urban flooding simulation optimization and intelligent monitoring, which solves the problem of urban flooding modeling when digital elevation data and land use data are poor.
[0005] To achieve the above objectives, the technical solution adopted by this invention is: a method for urban flooding simulation optimization and intelligent monitoring based on computer vision, comprising the following steps: S1. Determine the study area, establish a one-dimensional hydrodynamic model of the water network based on the initial modeling data, and verify the simulation accuracy of the one-dimensional hydrodynamic model of the water network using computer vision technology based on extreme rainfall conditions; S2. Based on computer vision, the SegNet and U-Net algorithms are used to segment and extract land use data of the study area, reshape the underlying surface, construct new land use data, and calculate new uncertainty parameters; and the performance of the one-dimensional hydrodynamic model of the water network before and after updating the uncertainty parameters is compared and analyzed. S3. Based on the performance analysis results and simulation accuracy verification results, a one- or two-dimensional surface coupling model combining the land surface flux integrated flood forecasting model ITF-FLOOD is constructed on the basis of the one-dimensional hydrodynamic model of the water network. The one- or two-dimensional surface coupling model is used to simulate the surface water depth and water area in the study area, and to complete the simulation optimization and intelligent monitoring of urban flooding.
[0006] Further, S1 includes the following steps: Collect pipeline network data within the study area and process data from problematic pipeline networks; Based on the existing land use data, roads within the study area are classified and statistically analyzed, buffer analysis is performed on the existing road line layers, road widths are defined according to road classification, and the height of building facade areas is increased by generating precise road surfaces. Based on the weights of different land types in the catchment area within the study region, the uncertainty parameters of the one-dimensional hydrodynamic model of the water network are calculated, and the width ratio and slope ratio are calculated to complete the establishment of the one-dimensional hydrodynamic model of the water network.
[0007] Furthermore, the method of verifying the simulation accuracy of the one-dimensional hydrodynamic model of the water network using computer vision technology based on extreme rainfall conditions specifically involves: verifying the simulation accuracy of the one-dimensional hydrodynamic model of the water network using the Nash efficiency coefficient, peak occurrence time, and peak magnitude under the simulation of three actual rainfall events.
[0008] Furthermore, the Nash efficiency coefficient The expression is as follows:
[0009] in, Q obs，i Indicates the first i One observation value, Q sim，i Indicates the first i One simulated value, This represents the average of the observed values. n This represents the total number of observations.
[0010] Furthermore, S2 includes the following steps: Based on computer vision, the SegNet algorithm and U-Net algorithm were used to perform multi-land type segmentation and single-land type segmentation on satellite images of the study area, respectively. Based on the segmented image, reshape the underlying surface; Based on the reshaped underlying surface, the uncertainty parameters of the catchment area were recalculated, and the performance of the one-dimensional hydrodynamic model of the water network before and after updating the uncertainty parameters was compared and analyzed.
[0011] Furthermore, step S3 includes the following steps: Based on the performance analysis results, input files for a one-dimensional and two-dimensional surface coupling model were created, and an integrated land surface flux flood forecasting model, ITF-FLOOD, was constructed. The integrated flood forecasting model ITF-FLOOD based on land surface flux is coupled with a one-dimensional hydrodynamic model of the water network to obtain a two-dimensional coupled surface model. The simulation accuracy before and after coupling is compared and analyzed. Based on the simulation accuracy of comparative analysis, the simulated surface water area and surface water depth are recorded at the determined waterlogging points, and compared with the actual waterlogging situation at the waterlogging points to complete the construction of a one-dimensional and two-dimensional surface coupling model.
[0012] Furthermore, the integrated land surface flux flood forecasting model ITF-FLOOD includes: the Green-Ampt infiltration intensity equation, the two-dimensional unconfined groundwater equation, as well as the momentum equation and the Node continuity equation; The expression for the Green-Ampt infiltration intensity equation is as follows:
[0013] The amount of infiltration required for the saturated region to appear:
[0014] The infiltration rate is calculated using integration:
[0015]
[0016] in, f p Indicates time as p Infiltration rate at that time, in units of mm / h. K s This indicates the saturated hydraulic conductivity of the soil, expressed in mm / h. The soil suction force at a moist front is expressed in mm. F The cumulative infiltration amount represents the time from the start of infiltration to time [time value missing]. t The total amount of water that has entered the soil so far, in mm. θ d This indicates a deficiency in soil moisture. F s This represents the critical cumulative infiltration rate, in mm. i Indicates rainfall intensity, in mm / h. F 1 represents the cumulative infiltration amount before the time step. F 2 represents the cumulative infiltration after the time step, and C represents the time constant. Δt Indicates the time step; The expression for the two-dimensional diving equation is as follows:
[0017] in, h ( x,y,z () indicates water depth, in meters (m).u ( x,y,z )express x Flow velocity in the direction of travel, in m / s. v ( x, y,z )express y Flow velocity in the direction of travel, in m / s. t Time is expressed in seconds (s). g This represents the acceleration due to gravity, with units of m / s². x、 y This represents the horizontal coordinate, in meters (m). The momentum equation and the Node continuity equation are expressed as follows:
[0018]
[0019] in, Indicates the rate of change of flow over time. Represents the pressure gradient term. Represents the frictional resistance term. U This represents the average flow velocity across the cross-section, expressed in m / s. Q This represents the cross-sectional flow rate, in cubic meters (m³). 3 / s, A This represents the cross-sectional area of the water passage, in m². 2 , S f Indicates the gradient of hydraulic friction. H This indicates the water level, in meters (m). A SN This represents the water storage area of the node, in m². 2 , Σ A SL This represents the sum of the water storage areas of each river segment connecting the nodes, in meters (m²). 2 .
[0020] The beneficial effects of this invention are: This invention establishes a one-dimensional hydrodynamic model of a water network based on initial modeling data and the Rainstorm Runoff Management Model (SWMM). The simulation accuracy of the SWMM using computer vision technology is verified under extreme rainfall conditions. Comparison and analysis of the simulation results with measured data show that the SWMM has high simulation accuracy, facilitating data updates for the study area. Using computer vision, SegNet and U-Net algorithms are employed to segment and extract land use data from the study area, reshaping the underlying surface and constructing new land use data for simulation and comparative analysis. A one- and two-dimensional surface coupling model of the SWMM and the Integrated Land Flux Flood Forecasting Model (ITF-FLOOD) is constructed to accurately simulate surface water accumulation within the study area. The surface water depth and area are calculated using formulas such as the Green-Ampt infiltration intensity equation, the two-dimensional groundwater equation, the momentum equation, and the Node continuity equation. This invention improves the simulation accuracy of urban flooding models and provides a more intelligent and convenient monitoring method for flood-prone areas. Attached Figure Description
[0021] Figure 1 This is a flowchart of the computer vision-based method for simulating, optimizing, and intelligently monitoring urban flooding, as described in this invention.
[0022] Figure 2 This is a technical roadmap for the computer vision-based urban flooding simulation optimization and intelligent monitoring method in this embodiment.
[0023] Figure 3 This is a schematic diagram of a module applying the method in this embodiment. Detailed Implementation
[0024] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.
[0025] Example like Figures 1 to 2 As shown, this invention provides a computer vision-based method for simulating, optimizing, and intelligently monitoring urban flooding, the implementation of which is as follows: S1. The study area was determined, and a one-dimensional hydrodynamic model of the water network was established based on the initial modeling data. The simulation accuracy of the one-dimensional hydrodynamic model of the water network was verified by computer vision technology under extreme rainfall conditions. The simulation results of the one-dimensional hydrodynamic model of the water network were compared and analyzed with the measured data, which showed that the one-dimensional hydrodynamic model of the water network has high simulation accuracy and facilitates data updates in the study area. The implementation method is as follows: Collect pipeline network data within the study area and process data from problematic pipeline networks; Based on the existing land use data, roads within the study area are classified and statistically analyzed, buffer analysis is performed on the existing road line layers, road widths are defined according to road classification, and the height of building facade areas is increased by generating precise road surfaces. Based on the weights of different land types in the catchment area within the study region, the uncertainty parameters of the one-dimensional hydrodynamic model of the water network are calculated, and the width ratio and slope ratio are calculated to complete the establishment of the one-dimensional hydrodynamic model of the water network.
[0026] In this embodiment, the one-dimensional hydrodynamic model of the water network specifically applies Horton's infiltration formula, the expression of which is:
[0027] in, f t express t The infiltration rate at any given time, (mm / h); f c This represents the steady-state soil infiltration rate (mm / h). f 0 represents the initial soil infiltration rate (mm / h). k Indicates the attenuation coefficient. t Indicates time, (s).
[0028] After calibration and verification of the flooding model, it is shown that the established one-dimensional stormwater network model has high prediction accuracy.
[0029] In this embodiment, based on the initial modeling data and combined with the storm runoff management model SWMM, the expression of the storm runoff management model SWMM includes the surface runoff continuity equation, the Saint-Venant equations, and the infiltration Horton model. The expression of the surface runoff continuity equation is as follows:
[0030]
[0031] in, S This indicates the surface water storage capacity of the unit. t Indicates time, (s), P Indicates rainfall intensity. I Indicates infiltration intensity. E This indicates evapotranspiration (which can be ignored during short-term heavy rain). Q Indicates surface outflow intensity and net rainfall intensity. R=PI ( P > I hour), h This represents the average depth of water accumulation on the slope, in meters (m). xIndicates the horizontal coordinate along the process flow, (m). q This represents the unit width runoff flow rate, (m 2 / s), R The net rainwater inflow intensity after deducting infiltration is expressed as (m / s).
[0032] The Saint-Venant equations are expressed as follows:
[0033]
[0034] in, A Represents the cross-sectional area of the water passage, (m²) 2 ), Q Represents the instantaneous flow rate at the cross section, (m 3 / s), qL Indicates lateral single-width net rainwater inflow recharge, (m) 2 / s), g Let gravitational acceleration be (m) 2 / s), S 0 represents the bottom slope of a riverbed or hillside. S f The slope representing hydraulic friction can be solved iteratively using the Manning formula or the Chezy formula.
[0035] The expression for the infiltration Horton model is as follows:
[0036] in, f t express t The infiltration rate at any given time (mm / h). f c Indicates the steady-state soil infiltration rate (mm / h). f 0 represents the initial soil infiltration rate (mm / h). k Indicates the attenuation coefficient. t Indicates time.
[0037] Modeling includes: collecting pipeline network data within the study area, processing problem data, checking topological relationships, performing road classification, buffer analysis, and Kriging interpolation using construction drawings, increasing the height of building roof areas, optimizing digital elevation data for road and building roof areas to generalize the stormwater pipeline network, and calibrating the width, slope, and uncertainty parameters.
[0038] In this embodiment, a one-dimensional hydrodynamic model of the water network is established based on the initial modeling data and the storm runoff management model SWMM. Under the simulation of three actual rainfall events, the simulation accuracy of the one-dimensional hydrodynamic model of the water network is tested using the Nash efficiency coefficient, peak occurrence time, and peak size error as evaluation indicators.
[0039] In this embodiment, the verification of the one-dimensional hydrodynamic model of the water network includes: verifying the simulation accuracy of the urban flooding model using the Nash efficiency coefficient, peak occurrence time, and peak magnitude under simulations of three actual rainfall events; the expression for calculating the Nash efficiency coefficient is:
[0040] in, Q obs，i Indicates the first i One observation value, Q sim，i Indicates the first i One simulated value, This represents the average of the observed values. n This represents the total number of observations.
[0041] In this embodiment, the construction of the one-dimensional hydrodynamic model of the water network is based on the conventional one-dimensional stormwater pipe network modeling, including pipe network data collection, optimization of DEM (digital elevation data), calculation of uncertainty parameters, and model verification after calibration.
[0042] As a preferred approach, optimizing DEM accuracy employs a method for refining the elevation of road and building surface areas. Based on the existing land use data, roads within the study area are statistically classified, buffer analyses are performed on existing road line layers, road widths are defined according to road classification, and precise road surfaces are generated, thereby increasing the height of building surface areas.
[0043] As a preferred approach, the calculation of uncertain parameters in the one-dimensional hydrodynamic model of the water network needs to be based on the weights of different land types in the catchment area within the study region, and then on the values of empirical parameters. Uncertain parameters include impermeability percentage, impermeable surface roughness coefficient, permeable surface roughness coefficient, impermeable depression water storage, permeable depression water storage, maximum infiltration, and minimum infiltration.
[0044] Preferably, slope calibration uses elevation information from the DEM to calculate the average slope of the catchment area; width calibration assumes the catchment area is rectangular and estimates the catchment area width using the flow path length and catchment area; the calculation of uncertain parameters in the one-dimensional hydrodynamic model of the water network requires considering the weights of different land types in the catchment area within the study region, and then selecting values based on empirical parameters; the expression for slope calculation is:
[0045] in, Elevation represents the elevation difference between grid cells. Horizontal distance refers to the horizontal distance between grid cells.
[0046] The following describes the simulation process of three wells with three rainfall events in Zhongshan City. First, pipeline network data within the study area was collected, problematic data was processed, and topological relationships were checked. Road classification, buffer zone analysis, and Kriging interpolation were performed using construction drawings. The ground elevation of buildings in the study area was standardized to 50m, generating optimized DEM data. Under the simulation of three actual rainfall events, the accuracy of the one-dimensional hydrodynamic model of the water network was verified using the Nash efficiency coefficient, peak occurrence time, and peak value. The results show that the highest Nash efficiency coefficient for the three inspection wells under the three rainfall conditions was 0.917, and the lowest was 0.802. Detailed statistics of the one-dimensional hydrodynamic model verification indicators are shown in Table 1 below. Table 1 is the statistical table of model verification indicators.
[0047] Table 1
[0048] S2. Based on computer vision, the SegNet and U-Net algorithms are used to segment and extract land use data of the study area, reshape the underlying surface, construct new land use data, and calculate new uncertainty parameters. The performance of the one-dimensional hydrodynamic model of the water network before and after updating the uncertainty parameters is compared and analyzed. The results show that the accuracy of urban flooding simulation is significantly improved. This model is used to construct a one-dimensional and two-dimensional surface coupling model. The implementation method is as follows: Based on computer vision, the SegNet algorithm and U-Net algorithm were used to perform multi-land type segmentation and single-land type segmentation on satellite images of the study area, respectively. Based on the segmented image, reshape the underlying surface; Based on the reshaped underlying surface, the uncertainty parameters of the catchment area were recalculated, and the performance of the one-dimensional hydrodynamic model before and after updating the uncertainty parameters was compared and analyzed.
[0049] In this embodiment ， Land use data of the study area was extracted and segmented using the SegNet and U-Net algorithms. The underlying surface was reconstructed, new land use data was constructed, and simulation and comparative analysis were performed. Specifically, the satellite images of the study area were segmented into multiple land types and single land types using the SegNet and U-Net algorithms, respectively. The segmented images were used for new underlying surface data, and the uncertainty parameters of the catchment area were recalculated. The accuracy of the one-dimensional hydrodynamic model of the water network was verified under the same rainfall conditions.
[0050] In this embodiment, the experiment using the U-Net algorithm to segment and extract land use data uses precision, recall, mean average probability per unit area (mPA), and mean intersection-union ratio (mIoU) as evaluation metrics.
[0051]
[0052]
[0053]
[0054]
[0055]
[0056] in, TP , FP , FN and TN These represent the number of pixels correctly segmented and correctly predicted as positive, the number of pixels incorrectly predicted as positive, the number of pixels that were actually positive but incorrectly predicted as negative, and the number of pixels correctly predicted as negative, respectively. N Precision represents the number of categories. i Indicates the first i The accuracy of the class.
[0057] In this embodiment, land type image segmentation is performed on satellite image data based on the SegNet and U-Net algorithms. The underlying surface is analyzed, and a one-dimensional hydrodynamic model of the water network is trained. Multiple land type data are segmented to reconstruct the underlying surface data, optimize the uncertainty parameters of the one-dimensional hydrodynamic model of the water network, and verify the accuracy of the one-dimensional hydrodynamic model of the water network under the same rainfall conditions after updating the underlying surface data.
[0058] As a preferred approach, the building surfaces extracted by the SegNet and U-Net algorithms were selected based on their superior performance in extracting land use information. The SegNet algorithm preserves spatial detail through pooling indexing, making it suitable for fine-grained segmentation of high-resolution images. The U-Net algorithm, with its symmetrical structure and skip connections, effectively fuses shallow texture and deep semantic features, maintaining high robustness even with limited data.
[0059] As a preferred method, the building surfaces extracted based on the U-Net algorithm are transformed into a building surface layer, which is necessary for the modeling process, after vector transformation. The uncertainty parameters are then recalculated using the new building surfaces to verify the one-dimensional hydrodynamic model of the water network.
[0060] The following describes the simulation process of three rainfall events at three wells in Zhongshan City. SegNet and U-Net algorithms were used to segment satellite images of the study area to extract land use information. After comparison, the building surface segmentation result of the U-Net algorithm was selected as the new building surface layer to calculate new uncertainty parameters. The same three rainfall events were used for simulation. Model validation indicators on manholes A1, A2, and A3 were displayed. The statistical results of the model validation indicators after parameter updates are shown in Table 2 below. Table 2 is the statistical table of model validation indicators after parameter updates.
[0061] Table 2
[0062] S3. Based on the performance analysis results and simulation accuracy verification results, a one-dimensional and two-dimensional surface coupling model combining the land surface flux integrated flood forecasting model ITF-FLOOD is constructed on the basis of the one-dimensional hydrodynamic model of the water network. The one-dimensional and two-dimensional surface coupling model is used to simulate the surface water depth and water area in the study area, and to complete the simulation optimization and intelligent monitoring of urban flooding. The simulation results have been verified to be relatively accurate and meet the needs of actual monitoring and prediction. The results are used for the following comparative analysis, in which the surface water depth and water area in the study area are calculated using formulas such as the Green-Ampt infiltration intensity equation, the two-dimensional groundwater equation, the momentum equation, and the Node continuity equation.
[0063] In this embodiment, the two-dimensional model is obtained by constructing a two-dimensional stormwater network model ITF-FLOOD based on one-dimensional stormwater network data. Its hydraulic principles remain unchanged and its expression is consistent with that of the integrated flood forecasting model ITF-FLOOD based on land surface flux.
[0064] The expression for the Green-Ampt infiltration intensity equation is as follows:
[0065] The amount of infiltration required for the saturated region to appear:
[0066] The subsequent infiltration rate is calculated using integration:
[0067]
[0068] in, f p Indicates time as p Infiltration rate at that time, in units of mm / h. K s This indicates the saturated hydraulic conductivity of the soil, expressed in mm / h. The soil suction force at a moist front is expressed in mm. F The cumulative infiltration amount represents the time from the start of infiltration to time [time value missing]. t The total amount of water that has entered the soil so far, in mm. θ d This indicates a deficiency in soil moisture. F s The value represents the critical cumulative infiltration rate in mm, and i represents the rainfall intensity (infiltration water supply intensity) in mm / h. F 1 represents the cumulative infiltration amount before the time step. F 2 represents the cumulative infiltration after the time step, and C represents the time constant. Δt Indicates the time step.
[0069] The expression for the two-dimensional diving equation is as follows:
[0070] in, h ( x,y,z () indicates water depth, in meters (m). u ( x,y,z )express x Flow velocity in the direction of travel, in m / s; v ( x, y,z )express y Flow velocity in the direction of travel, in m / s; t Indicates time, in seconds; g This represents the acceleration due to gravity, and the unit is m / s². x、 y This represents the horizontal coordinate, in meters (m).
[0071] The momentum equation and the Node continuity equation are expressed as follows:
[0072]
[0073] in, Indicates the rate of change of flow over time. Represents the pressure gradient term. Represents the frictional resistance term. U This represents the average flow velocity across the cross-section, expressed in m / s. Q This represents the cross-sectional flow rate, in cubic meters (m³). 3 / s, A This represents the cross-sectional area of the water passage, in m². 2 , S f Indicates the gradient of hydraulic friction. H This indicates the water level, in meters (m). ASN This represents the water storage area of the node, in m². 2 , Σ A SL This represents the sum of the water storage areas of each river segment connecting the nodes, in meters (m²). 2 .
[0074] Among them, the one-dimensional and two-dimensional surface coupling model is used to simulate the depth and area of surface water accumulation in the study area, and its implementation method is as follows: Based on the performance analysis results, input files for a two-dimensional surface coupling model were created, and a two-dimensional hydrodynamic model, the land surface flux integrated flood forecasting model ITF-FLOOD, was constructed. The two-dimensional hydrodynamic model of land surface flux integrated flood forecasting model ITF-FLOOD was coupled with the one-dimensional hydrodynamic model of water network, and the simulation accuracy before and after coupling was compared and analyzed. Based on the simulation accuracy, the simulated surface water area and depth are recorded at the determined waterlogging points, and compared with the actual waterlogging situation at the waterlogging points to complete the construction of a one-dimensional and two-dimensional surface coupling model.
[0075] In this embodiment, a one- or two-dimensional surface coupled model of the storm runoff management model SWMM and the land surface flux integrated flood forecasting model ITF-FLOOD is constructed to accurately simulate the surface water accumulation state in the study area. Specifically, this includes: creating the input file for the one- or two-dimensional hydrodynamic model; constructing the land surface flux integrated flood forecasting model ITF-FLOOD and coupling it with the one-dimensional hydrodynamic model of the water network; comparing and analyzing the simulation accuracy of the model before and after coupling; recording the simulated surface water accumulation area and surface water accumulation depth at the determined waterlogging points and comparing and analyzing them with the actual water accumulation situation at the waterlogging points; wherein, the surface water accumulation depth and water accumulation area in the study area are calculated using formulas such as the Green-Ampt infiltration intensity equation, the two-dimensional groundwater equation, the momentum equation, and the Node continuity equation. In this embodiment, an integrated land surface flux flood forecasting model (ITF-FLOOD) is constructed based on one-dimensional stormwater network data to calculate the surface water depth and area in the study area. Input files for the one-dimensional and two-dimensional surface coupling model are created, simulation input parameters are adjusted, one-dimensional stormwater network data is coupled, and the simulation accuracy of the one-dimensional and two-dimensional surface coupling model before and after coupling is compared and analyzed. Simulated surface water accumulation at determined flood-prone points is recorded and compared with actual water accumulation conditions.
[0076] As a preferred approach, a one-dimensional stormwater pipe network coupled with a two-dimensional surface model is used to realize the dynamic interaction between the underground drainage system and the surface hydrodynamic process.
[0077] The following describes the simulation process of waterlogging points in Duxing West Road, Nan District, Zhongshan City, based on three rainfall events. An integrated land surface flux flood forecasting model (ITF-FLOOD) was constructed using one-dimensional stormwater network data. The Green-Ampt infiltration equation, two-dimensional groundwater equation, momentum equation, and Node continuity equation were employed to calculate surface water depth and area. A one-dimensional surface coupling model input file was prepared, parameters were adjusted, and one-dimensional network data was coupled. A rainfall event from 3:00 AM to 12:00 PM on June 3, 2024, in Zhongshan City was simulated. The water accumulation data was output and compared with the results before coupling, showing higher accuracy after coupling. Finally, under the conditions of three actual rainfall events, the simulation results were compared with the actual monitoring data of the waterlogging points studied in this invention. The results show that the error in water accumulation area is within 22%, and the error in water accumulation depth is within 4 cm. The simulation results and measured results of the waterlogging points under the three rainfall events are shown in Table 3 below. Table 3 is a comparison table of simulated and measured waterlogging points.
[0078] Table 3
[0079] In this embodiment, the YOLOv8 model based on the CBAM attention mechanism is used to monitor the water accumulation area of flood-prone points in real time, and the identification results are compared and analyzed with the water accumulation area simulated using a one- or two-dimensional surface coupling model. The implementation method is as follows: Collect images of waterlogged areas within the study region and construct a dataset; Preprocess the constructed dataset; Using the preprocessed dataset, we trained and compared YOLOv8 models based on the CBAM attention mechanism and those without the attention mechanism. Based on the comparison results, the model weight file generated during training is used to identify the water accumulation area at flood-prone points; The identification results were compared and analyzed with the simulated water accumulation area using a one- or two-dimensional surface coupling model.
[0080] In this embodiment, the YOLOv8 model based on the CBAM attention mechanism is used to monitor and verify the accuracy of the model in real time for the water accumulation area of flood-prone points. Specifically, this includes: collecting water accumulation images of flood points in the study area, constructing a dataset, performing preprocessing, creating a training environment, adjusting the training parameters, training the YOLOv8 model, comparing it with the YOLOv8 model trained without the attention mechanism, testing the water accumulation recognition effect, and then comparing and analyzing it with the simulated water accumulation area of the one-dimensional coupled model.
[0081] In this embodiment, images of waterlogged areas within the study region are collected, a dataset is constructed, preprocessed, a training environment is created, and training parameters are adjusted. The YOLOv8 model with and without the attention mechanism CBAM is trained and compared. The model weight file generated during training is used to identify road waterlogging. After testing the waterlogging identification effect, a comparative analysis is performed with the simulated waterlogging area of the one-dimensional surface coupling model.
[0082] Preferably, the dataset is constructed by augmenting the obtained data before training to increase the sample size, thereby solving the overfitting problem caused by insufficient samples; and on this basis, the robustness and generalization performance of the one-dimensional and two-dimensional surface coupling model are further improved. Overflow point images of the study area are collected, and after image preprocessing, the dataset is labeled using both manual and automatic annotation methods.
[0083] Preferably, the water accumulation area is calculated using perspective transformation and pixel value relationships. Perspective transformation positions the target area horizontally, eliminating the surrounding background and reducing the amount of model processing required.
[0084] The following is an illustration of the waterlogging process at a flood-prone area on Duxing West Road in Nan District, Zhongshan City. Based on the YOLOv8 algorithm, a CBAM attention mechanism was added to the neck structure network to enhance important features of the waterlogged area and suppress general features, thereby improving the accuracy of road waterlogging identification. Perspective transformation and pixel-based calculation of the waterlogged area were also employed. After training, the two-dimensional surface coupling model achieved a precision of 99%, a recall of 100%, and a mean squared error of 99.5% after convergence.
[0085] In this embodiment, the U-Net algorithm is used to monitor and verify the accuracy of water accumulation depth at flood-prone points, thereby completing the optimization and intelligent monitoring of flood simulation. The implementation method is as follows: Design a new type of water gauge based on basic physical water gauges; The U-Net algorithm is used to segment novel water level gauge binarized images; Based on the segmentation results, the surface water depth is output by calculating the pixel values of the binarized image and their relationship with the water gauge reading. The output water depth is compared and analyzed with the water area simulated using a one- or two-dimensional surface coupling model.
[0086] In this embodiment, the specific steps for monitoring and verifying the accuracy of water depth at flood-prone areas based on the U-Net algorithm include: designing a new type of water gauge based on a basic physical water gauge, collecting water gauge images of water accumulation at flood-prone areas, preprocessing the water gauge images, labeling the water gauge images, adjusting parameters to train a one-dimensional and two-dimensional surface coupling model, and designing a built-in pixel value and water gauge reading calculation program to process the segmented binarized images, output the water gauge readings, and compare and analyze them with the simulated water depth of the one-dimensional and two-dimensional coupling model.
[0087] In this embodiment, a new type of water gauge is designed based on the conventional water gauge. Based on the U-Net algorithm, the binarized image of the new water gauge is segmented, and the surface water depth is output by calculating the pixel values of the binarized image and the specific relationship with the water gauge reading.
[0088] The following example illustrates the monitoring of a flood-prone area on Duxing West Road in the Nan District of Zhongshan City. A new type of water gauge was designed based on conventional water gauges. Video footage of accumulated water was collected at the overflow point on Duxing West Road in the Nan District of Zhongshan City, and frames were extracted into images. To improve the segmentation effect of features, the water images underwent preprocessing operations such as grayscale conversion, median filtering, and histogram equalization. Then, the dataset was labeled and converted into a label file readable by a one-dimensional / two-dimensional surface coupling model. Training parameters were adjusted to train the one-dimensional / two-dimensional surface coupling model. Evaluation metrics after training showed that the one-dimensional / two-dimensional surface coupling model has good robustness and high accuracy. The segmented binarized images were input into the "built-in pixel value and water gauge reading calculation program" to output the water gauge reading.
[0089] In this embodiment, Figure 3 This is a schematic diagram of a computer vision-based urban flooding model optimization and intelligent monitoring method according to some embodiments of this specification. The computer vision-based urban flooding model optimization and intelligent monitoring system may include a model building module, a data refinement module, a model coupling module, and modules for identifying area and water depth. The model building module is used to verify the accuracy of computer vision technology in simulating a one-dimensional hydrodynamic model of a water network; The data refinement module is used to reshape the underlying surface and solve the problem of building a refined model when land use data is missing or poor during the modeling process; The coupled model module is used to accurately simulate the surface water accumulation state within the study area; The area identification module is used to achieve accurate and automated monitoring of the water accumulation area in flood-prone areas; The water depth identification module is used to automate the monitoring of water depth in flood-prone areas.
[0090] Urban flooding can be used to implement a computer vision-based urban flooding model optimization and intelligent monitoring method. For more details on computer vision technology simulation, please refer to the relevant description of a computer vision-based urban flooding model optimization and intelligent monitoring method, which will not be repeated here.
[0091] Compared with the prior art, this embodiment has at least the following beneficial effects: This invention optimizes the basic data for modeling the research area, analyzes and classifies road buffer zones before constructing a one-dimensional stormwater pipe network model, and improves the accuracy of building roof height and DEM, thereby enhancing the simulation accuracy of the one-dimensional hydrodynamic model of the water network.
[0092] This invention employs the SegNet and U-Net algorithms to segment high-precision satellite images of the study area into multiple land types and single land types, thereby reshaping the underlying surface, generating new land use data, and recalibrating parameters. This solves the problem of waterlogging modeling when basic data is missing or of poor quality, and significantly improves the accuracy of waterlogging prediction.
[0093] This invention employs the YOLOv8 algorithm to monitor water accumulation areas and incorporates a CBAM attention mechanism to enhance feature capture capabilities. It achieves precise and automated monitoring of water area through pixel coordinates and photographic principles. To improve depth recognition, a novel physical water level gauge is designed, using the U-Net algorithm for binarization segmentation. The water depth is output based on the relationship between pixel values and readings, thus achieving precise and automated depth monitoring.
[0094] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.
[0095] Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the above embodiments are only for the purpose of helping to understand the core ideas of this invention. Furthermore, those skilled in the art will recognize that, based on the ideas of this invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this invention.
Claims
1. A computer vision-based method for simulating, optimizing, and intelligently monitoring urban flooding, characterized in that, Includes the following steps: S1. Determine the study area, establish a one-dimensional hydrodynamic model of the water network based on the initial modeling data, and verify the simulation accuracy of the one-dimensional hydrodynamic model of the water network using computer vision technology based on extreme rainfall conditions; S2. Based on computer vision, the SegNet and U-Net algorithms are used to segment and extract land use data of the study area, reshape the underlying surface, construct new land use data, and calculate new uncertainty parameters. And compare and analyze the performance of the one-dimensional hydrodynamic model of the water network before and after updating the uncertainty parameters; S3. Based on the performance analysis results and simulation accuracy verification results, a one- or two-dimensional surface coupling model combining the land surface flux integrated flood forecasting model ITF-FLOOD is constructed on the basis of the one-dimensional hydrodynamic model of the water network. The one- or two-dimensional surface coupling model is used to simulate the surface water depth and water area in the study area, and to complete the simulation optimization and intelligent monitoring of urban flooding.
2. The method for urban flooding simulation optimization and intelligent monitoring based on computer vision according to claim 1, characterized in that, S1 includes the following steps: Collect pipeline network data within the study area and process data from problematic pipeline networks; Based on the existing land use data, the roads in the study area are classified and statistically analyzed, the existing road line layers are buffered, the road width is defined according to the road classification, and the height of the building surface area is improved by generating precise road surfaces. Based on the weights of different land types in the catchment area within the study region, the uncertainty parameters of the one-dimensional hydrodynamic model of the water network are calculated, and the width ratio and slope ratio are calculated to complete the establishment of the one-dimensional hydrodynamic model of the water network.
3. The method for urban flooding simulation optimization and intelligent monitoring based on computer vision according to claim 1, characterized in that, The method for verifying the simulation accuracy of the one-dimensional hydrodynamic model of the water network using computer vision technology based on extreme rainfall conditions is as follows: under the simulation of three actual rainfall events, the simulation accuracy of the one-dimensional hydrodynamic model of the water network is verified by the Nash efficiency coefficient, peak occurrence time, and peak value.
4. The method for urban flooding simulation optimization and intelligent monitoring based on computer vision according to claim 3, characterized in that, The Nash efficiency coefficient The expression is as follows: in, Q obs，i Indicates the first i One observation value, Q sim，i Indicates the first i One simulated value, This represents the average of the observed values. n This represents the total number of observations.
5. The method for urban flooding simulation optimization and intelligent monitoring based on computer vision according to claim 1, characterized in that, S2 includes the following steps: Based on computer vision, the SegNet algorithm and U-Net algorithm were used to perform multi-land type segmentation and single-land type segmentation on satellite images of the study area, respectively. Based on the segmented image, reshape the underlying surface; Based on the reshaped underlying surface, the uncertainty parameters of the catchment area were recalculated, and the performance of the one-dimensional hydrodynamic model of the water network before and after updating the uncertainty parameters was compared and analyzed.
6. The method for urban flooding simulation optimization and intelligent monitoring based on computer vision according to claim 1, characterized in that, S3 includes the following steps: Based on the performance analysis results, input files for a one-dimensional and two-dimensional surface coupling model were created, and an integrated land surface flux flood forecasting model, ITF-FLOOD, was constructed. The integrated flood forecasting model ITF-FLOOD based on land surface flux is coupled with a one-dimensional hydrodynamic model of the water network to obtain a two-dimensional coupled surface model. The simulation accuracy before and after coupling is compared and analyzed. Based on the simulation accuracy of comparative analysis, the simulated surface water area and surface water depth are recorded at the determined waterlogging points, and compared with the actual waterlogging situation at the waterlogging points to complete the construction of a one-dimensional and two-dimensional surface coupling model.
7. The method for simulating, optimizing, and intelligently monitoring urban flooding based on computer vision according to claim 6, characterized in that, The integrated land surface flux flood forecasting model ITF-FLOOD includes: the Green-Ampt infiltration intensity equation, the two-dimensional unconfined groundwater equation, as well as the momentum equation and the Node continuity equation; The expression for the Green-Ampt infiltration intensity equation is as follows: The infiltration rate required for the saturation zone to appear: The infiltration rate is calculated using integration: in, f p Indicates time as p Infiltration rate at that time, in units of mm / h. K s This represents the saturated hydraulic conductivity of the soil, expressed in mm / h. The soil suction force at a moist front is expressed in mm. F The cumulative infiltration amount represents the time from the start of infiltration to time [time value missing]. t The total amount of water that has entered the soil so far, in mm. θ d This indicates a deficiency in soil moisture. F s This represents the critical cumulative infiltration rate, in mm. i This indicates rainfall intensity, expressed in mm / h. F 1 represents the cumulative infiltration amount before the time step. F 2 represents the cumulative infiltration after the time step, and C represents the time constant. Δt Indicates the time step; The expression for the two-dimensional diving equation is as follows: in, h ( x,y,z () indicates water depth, in meters (m). u ( x,y,z )express x Flow velocity in the direction of travel, in m / s. v ( x,y,z )express y Flow velocity in the direction of travel, in m / s. t Time is expressed in seconds (s). g This represents the acceleration due to gravity, with units of m / s². x, y This represents the horizontal coordinate, in meters (m). The momentum equation and the Node continuity equation are expressed as follows: in, Indicates the rate of change of flow over time. Represents the pressure gradient term. Represents the frictional resistance term. U This represents the average flow velocity across the cross-section, expressed in m / s. Q This represents the cross-sectional flow rate, in cubic meters (m³). 3 / s, A This represents the cross-sectional area of the water passage, in m². 2 , S f Indicates the gradient of hydraulic friction. H This indicates the water level, in meters (m). A SN This represents the water storage area of the node, in m². 2 , Σ A SL This represents the sum of the water storage areas of each river segment connecting the nodes, in meters. 2 .