Method for urban underlay waterlogging risk assessment based on machine learning

Abstract

This application proposes a machine learning-based method for assessing urban surface flooding risk, comprising: determining the surface characteristics of the target geographic area based on high-resolution remote sensing imagery, lidar point cloud data, and drainage network data; using the surface characteristics, precipitation time series, and initial surface water depth before precipitation as input information for a water depth prediction model, outputting the first predicted water depth for each location unit within the target geographic area; simulating the water diffusion process in the target geographic area to obtain the second predicted water depth for each location unit at a specified prediction time; and determining the flooding risk level of the target geographic area based on the second predicted water depth. This method achieves real-time, efficient, and accurate prediction of urban surface flooding risk for complex underlying surfaces.

Description

Technical Field

[0001] This application relates to the field of disaster prevention technology, and in particular to a machine learning-based method for assessing urban surface flooding risk. Background Technology

[0002] Urban flooding is a major natural disaster that threatens urban safety. Hydrological and hydrodynamic models in related technologies use complex physical equations to solve when assessing flooding risks. This often requires a lot of computing resources and time, making it difficult to meet the timeliness requirements of urban emergency early warning.

[0003] Meanwhile, the models in related technologies that can assess the risk of urban flooding often overgeneralize the complex underlying surface of cities, making it difficult to accurately capture the spatial heterogeneity of micro-land features such as building complexes, green spaces, and road networks and their impact on surface runoff. This can lead to a large discrepancy between the simulated water accumulation results and the actual water accumulation situation.

[0004] Therefore, how to avoid the shortcomings of the current urban underlying surface waterlogging risk assessment in terms of timeliness and accuracy, and improve the accuracy of urban underlying surface waterlogging risk assessment, has become an urgent technical problem to be solved. Summary of the Invention

[0005] This application provides a machine learning-based method for assessing urban surface waterlogging risk, aiming to solve the technical problems of lack of timeliness and accuracy in related technologies.

[0006] In a first aspect, embodiments of this application provide a method for assessing urban surface flooding risk based on machine learning, including: Based on high-resolution remote sensing images, lidar point cloud data, and drainage network data of the target geographic area, the underlying surface characteristics of the target geographic area are determined. The underlying surface characteristics are used to reflect the physical properties of the surface of the target geographic area after being affected by geographic and human factors. The underlying surface characteristics of the target geographic region, the precipitation time series, and the initial surface water depth of the target geographic region before precipitation are used as input information for the water depth prediction model. The water depth prediction model outputs the first predicted water depth of each location unit in the target geographic region. Based on the first predicted water depth, the water diffusion process of the target geographical area is simulated by cellular automata to obtain the second predicted water depth of each location unit in the target geographical area at a specified prediction time. The waterlogging risk level of the target geographical area is determined based on the second predicted water depth of each location unit within the target geographical area at a specified prediction time.

[0007] Optionally, in one embodiment of this application, the underlying surface characteristics include the building density, impervious surface ratio, and green coverage of each location unit within the target geographic area. Then, determining the underlying surface characteristics of the target geographic area based on high-resolution remote sensing imagery, lidar point cloud data, and drainage network data includes: Using the high-resolution remote sensing image of the target geographic area as input information for the land cover type classification model, the land cover type of each pixel in the high-resolution remote sensing image is determined by the land cover type classification model. For each location unit within the target geographic region. The building density of the location unit is determined based on the number of pixels in the location unit whose land cover type belongs to the building type and the total number of pixels in the location unit. Based on the number of pixels in the location unit whose surface cover type is impermeable, and the total number of pixels in the location unit, the proportion of impermeable surface in the location unit is determined, wherein the impermeable type is the superordinate type of the building type; The green coverage rate of the location unit is determined based on the number of pixels in the location unit whose land cover type is green space, and the total number of pixels in the location unit.

[0008] In one embodiment of this application, optionally, the underlying surface performance characteristics include the surface roughness of each location unit within the target geographic area. Then, determining the underlying surface performance characteristics of the target geographic area based on high-resolution remote sensing imagery, lidar point cloud data, and drainage network data of the target geographic area includes: Determine the pixel percentage of each land cover type within the location unit; The surface roughness of the location unit is obtained by weighting the pixel proportion of each of the aforementioned surface cover types by using a predetermined Manning roughness coefficient as the weight.

[0009] Optionally, in one embodiment of this application, the underlying surface characteristics may further include the hydrotopographic features and urban drainage network density of each location unit within the target geographic area. Then, determining the underlying surface characteristics of the target geographic area based on high-resolution remote sensing imagery, lidar point cloud data, and drainage network data includes: Based on the lidar point cloud data of the target geographic area, ground points are extracted to construct a digital elevation model of the target geographic area. The digital elevation model is subjected to terrain analysis processing according to a predetermined terrain analysis method to determine the hydrological and topographic features of each location unit within the target geographic area. These hydrological and topographic features include: average elevation, slope, slope length, aspect, topographic relief, runoff path, runoff direction, catchment area, regional volume, and maximum depth. Based on the drainage network data of the target geographical area, the total length of the drainage pipes and the area of ​​each location unit within the target geographical area are determined, and the ratio of the total length of the drainage pipes to the area of ​​the unit is used as the urban drainage network density of the location unit.

[0010] In one embodiment of this application, optionally, determining the underlying surface characteristics of the target geographic area based on high-resolution remote sensing imagery, lidar point cloud data, and drainage network data of the target geographic area includes: Based on the high-resolution remote sensing imagery of the target geographic area, the lidar point cloud data, and the drainage network data, the building density, impermeable surface ratio, green coverage, surface roughness, hydro-topographic features, and urban drainage network density of each location unit within the target geographic area are determined. A comprehensive feature matrix of the underlying surface corresponding to each location unit is constructed based on the building density, impermeable surface ratio, green coverage, surface roughness, hydro-topographic features, and urban drainage network density of each location unit. The step of using the underlying surface characteristics of the target geographic region, precipitation time series, and the initial surface water depth of the target geographic region before precipitation as input information for the water depth prediction model, and outputting the first predicted water depth for each location unit within the target geographic region through the water depth prediction model, includes: For each location unit within the target geographic area, the comprehensive feature matrix of the underlying surface corresponding to the location unit, the precipitation time series, and the initial surface water depth of the location unit before precipitation are used as the input information of the water depth prediction model. The water depth prediction model outputs the first predicted water depth of the location unit.

[0011] In one embodiment of this application, optionally, the element in the m-th row and n-th column of the comprehensive feature matrix of the underlying surface corresponding to the location unit is the weighted average of the covariance and normalized product of the m-th and n-th features among the building density, impervious surface ratio, green coverage, surface roughness, hydrotopographic features, and urban drainage network density of the location unit, wherein, The weights of the covariance and the weights of the normalized product sum to 1. The weights of the covariance are used to reflect the accuracy of the covariance in describing the correlation between the m-th feature and the n-th feature in the historical urban surface waterlogging risk assessment. The weights of the normalized product are used to reflect the accuracy of the normalized product in describing the correlation between the m-th feature and the n-th feature in the historical urban surface waterlogging risk assessment business.

[0012] In one embodiment of this application, optionally, simulating the water diffusion process of the target geographical area using a cellular automata approach based on the predicted water depth includes: In the prediction period starting from the prediction time corresponding to the first predicted water depth, a second predicted water depth at a subsequent prediction time is determined using a cellular automata approach during the chronological water diffusion process, based on the known predicted water depth at the previous prediction time. , The known predicted water depth of the i-th location unit in the target geographic area at time t. The second predicted water depth for the i-th location unit at time t+1. Let be the time step between time t and time t+1. The inflow rate of the i-th position unit, The outflow rate of the i-th position unit is... The water exchange value between the i-th location unit and the adjacent j-th location unit is given.

[0013] In one embodiment of this application, optionally, before simulating the water diffusion process of the target geographical area using a cellular automata approach based on the predicted water depth, the method further includes: Based on the drainage network data, the drainage network nodes and edges, as well as the real-time operating status features of the drainage network nodes and the attribute features of the edges, are encoded into graph structure data. The graph structure data includes: an adjacency matrix representing the topology of the drainage network, a feature matrix representing the real-time operating status features of the drainage network nodes, and an edge feature matrix representing the physical properties of the pipeline. Using the graph structure data as input information for a graph neural network, the dynamic drainage capacity value of each location unit in the target geographic area at the current moment is determined by the graph neural network. Based on the dynamic drainage capacity value of each location unit at the current moment, the inflow and outflow of each location unit are calculated.

[0014] In one embodiment of this application, optionally, determining the flood risk level of the target geographical area based on the second predicted water depth of each location unit within the target geographical area at a specified prediction time includes: Based on the second predicted water depth of each location unit within the target geographical area at a specified prediction time, the probability that the water depth of each location unit exceeds a preset disaster threshold is calculated. Based on the aforementioned probability, the waterlogging risk level of the target geographical area is determined, wherein... , This indicates the risk of urban flooding in the target geographical area. Indicates the location of the target geographic area The water depth at the specified prediction time τ, To preset the disaster threshold, Indicates the position The vulnerability index, P(·), represents the probability that the water depth exceeds the preset disaster threshold.

[0015] In a second aspect, embodiments of this application provide a computer device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being configured to perform the method described in the first aspect above.

[0016] Thirdly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions for performing the method described in the first aspect above.

[0017] The above technical solution addresses the lack of timeliness and accuracy in urban surface flooding risk assessment in related technologies. By integrating multi-source data such as high-resolution remote sensing imagery, lidar point clouds, and drainage networks, it constructs surface performance characteristics that characterize the physical properties of complex urban surfaces. A water depth prediction model then predicts the water depth in a target geographic area under these surface performance characteristics and current rainfall intensity. Furthermore, the prediction results are used to simulate water diffusion, determining the water depth at each location unit within the target geographic area at a specified prediction time, serving as the basis for calculating the flooding risk of the target geographic area. This overcomes the shortcomings of related technologies, such as complex hydrodynamic model calculations, coarse surface characterization, and difficulty in comprehensively considering drainage system responses. It achieves real-time, efficient, and accurate prediction of urban surface flooding risk for complex surfaces, significantly improving the timeliness and reliability of flood disaster early warning, and providing reliable technical support for urban disaster prevention and mitigation as well as emergency management. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 A flowchart is shown for a machine learning-based urban surface flooding risk assessment method according to an embodiment of this application; Figure 2 This illustration shows a schematic diagram of intelligent extraction and fusion of multi-dimensional features of urban underlying surface according to an embodiment of this application; Figure 3 A flowchart of a deep neural network processing based on a multilayer perceptron according to an embodiment of this application is shown; Figure 4 A schematic diagram illustrating a cellular automata modeling process for simulating the propagation of urban flooding risk according to an embodiment of this application is shown. Figure 5 A block diagram of a computer device according to one embodiment of this application is shown; Figure 6 A block diagram of a computer device according to another embodiment of this application is shown. Detailed Implementation

[0020] 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, not all, of the embodiments of the present invention. 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.

[0021] Figure 1 A flowchart of a machine learning-based urban surface flooding risk assessment method according to an embodiment of this application is shown.

[0022] like Figure 1 As shown, a machine learning-based urban surface flooding risk assessment method according to an embodiment of this application includes: Step 102: Based on high-resolution remote sensing images, lidar point cloud data, and drainage network data of the target geographic area, determine the underlying surface characteristics of the target geographic area.

[0023] The target geographic area is the object of urban flooding assessment, including but not limited to any specified geographic area such as cities and street areas. Conducting urban flooding risk assessment on the target geographic area can provide a reliable basis for decision-making in urban planning, drainage system design, disaster early warning and emergency response.

[0024] The high-resolution remote sensing imagery includes, but is not limited to, satellite imagery, UAV aerial imagery, and any image covering the entire geographic area of ​​the target geographic region. The ground sampling distance of the high-resolution remote sensing imagery is less than a specified distance. Optionally, the ground sampling distance of the high-resolution remote sensing imagery is less than 1 meter. Of course, the specified distance can be any value that meets the actual accuracy requirements for urban flooding assessment, and is not limited to the examples given in this application. The lidar point cloud data reflects the three-dimensional fine structure of the terrain within the target geographic region and is the basis for extracting the hydro-topographic features of the target geographic region. The drainage network data reflects the layout of artificial drainage facilities within the target geographic region and can be used to quantify the drainage capacity of the target geographic region. The high-resolution remote sensing imagery, lidar point cloud data, and drainage network data effectively describe the complex underlying surface conditions of the city from three dimensions: the surface, the above-ground three-dimensional area, and infrastructure, respectively, and can serve as the basis for urban flooding risk assessment.

[0025] Among them, the underlying surface characteristics obtained from high-resolution remote sensing images, lidar point cloud data and drainage network data of the target geographical area can be used to reflect the physical properties of the surface of the target geographical area after being affected by geographical and human factors. These physical properties can reflect the target geographical area's ability to generate, collect and drain rainfall, and are a deep basis for predicting urban flooding risk.

[0026] In one possible design, based on the high-resolution remote sensing imagery, lidar point cloud data, and drainage network data of the target geographical area, the building density, impervious surface ratio, green space coverage, surface roughness, hydrotopographic features, and urban drainage network density of each location unit within the target geographical area can be determined. Then, a comprehensive feature matrix of the underlying surface corresponding to each location unit can be constructed using these parameters. This comprehensive feature matrix reflects the performance of each location unit in the target geographical area across multiple dimensions of urban flooding assessment, including building density, impervious surface ratio, green space coverage, surface roughness, hydrotopographic features, and urban drainage network density, and thus serves as an important basis for urban flooding assessment.

[0027] Wherein, the element in the m-th row and n-th column of the comprehensive feature matrix of the underlying surface corresponding to the location unit is the weighted average of the covariance and normalized product of the m-th feature and the n-th feature among the building density, impermeable surface ratio, green coverage, surface roughness, hydro-topographic features and urban drainage network density of the location unit.

[0028] The covariance of the m-th and n-th features reflects the statistical regularity of their correlation and synergistic change in the historical dataset. The normalized product of the m-th and n-th features reflects the proportional amplification or suppression effect of their values ​​within the current location unit. The weighted average of the covariance and normalized product of the m-th and n-th features comprehensively reflects the balance between the historical statistical regularity and the local numerical characteristics within the current location unit, thus providing a more comprehensive characterization of the impact of these two features on the flood risk of the location unit.

[0029] Wherein, the weight of the covariance and the weight of the normalized product sum to 1. The weight of the covariance is used to reflect the accuracy of the description of the correlation between the m-th feature and the n-th feature by the covariance in the historical urban surface waterlogging risk assessment business. The weight of the normalized product is used to reflect the accuracy of the description of the correlation between the m-th feature and the n-th feature by the normalized product in the historical urban surface waterlogging risk assessment business.

[0030] In historical urban surface flooding risk assessment, the higher the accuracy of the covariance in describing the correlation between the m-th and n-th features, the greater its effective contribution to accurately assessing flooding risk. Similarly, the higher the accuracy of the normalized product in describing the correlation between the m-th and n-th features, the greater its effective contribution to accurately assessing flooding risk. Therefore, the accuracy of both in describing the correlation between the m-th and n-th features can be used as their respective weights. The capabilities demonstrated by both in historical urban surface flooding risk assessment can be incorporated into the current flooding risk assessment criteria, ensuring that the current flooding risk assessment effectively aligns with the performance patterns of historical urban surface flooding risk assessment and improving the accuracy of the current assessment.

[0031] Step 104: Using the underlying surface characteristics of the target geographic area, the precipitation time series, and the initial surface water depth of the target geographic area before precipitation as input information for the water depth prediction model, the water depth prediction model outputs the first predicted water depth of each location unit in the target geographic area.

[0032] The underlying surface characteristics of the target geographic area determine the internal static influencing factors of its surface hydrological response, while precipitation time series, a set of precipitation intensity data arranged chronologically, describes the change in precipitation intensity over time within a specific period, representing the external dynamic influencing factor of the target geographic area's surface hydrological response. Furthermore, the initial surface water depth of the target geographic area before precipitation reflects the initial water conditions before the start of rainfall, i.e., the initial surface water depth of each location unit in the target geographic area at the start of the simulation. Combining these three factors as input information for the water depth prediction model integrates the key conditions determining urban flooding risk from three aspects: initial surface water depth of each location unit in the target geographic area, external influencing factors, and internal influencing factors. This allows the water depth prediction model to effectively learn the correlation between the key conditions determining urban flooding risk and the resulting water depth, thereby facilitating accurate prediction of the first predicted water depth for each location unit within the target geographic area.

[0033] In one possible design, for each location unit within the target geographic area, the comprehensive feature matrix of the underlying surface corresponding to the location unit, the precipitation time series, and the initial surface water depth of the location unit before precipitation are used as input information for the water depth prediction model, and the first predicted water depth of the location unit is output through the water depth prediction model.

[0034] In a practical scenario, a deep neural network constructed using a multilayer perceptron based on the TensorFlow framework can be used as a water depth prediction model. Additionally, a precipitation time series can be constructed from the average precipitation intensity every 10 minutes over the past 6 hours to predict the water depth within the next hour. Furthermore, the initial surface water depth of the specified location unit before precipitation can be obtained from historical monitoring data. If the initial surface water depth of the specified location unit before precipitation is not available in the historical monitoring data, it can be directly set to 0.

[0035] Step 106: Based on the first predicted water depth, simulate the water diffusion process of the target geographical area using a cellular automata approach to obtain the second predicted water depth of each location unit within the target geographical area at a specified prediction time.

[0036] Cellular automata is a computational modeling framework based on discrete space, discrete time, and local interaction rules. In this application, cellular automata is used to simulate the entire process of surface water diffusion, confluence, and receding in various location units of a target geographical area according to predefined local hydrophysical rules. Through this simulation, the real physical process of water evolution in each location unit over time and topography can be dynamically depicted. The static, instantaneous precipitation response-based first predicted water depth can be simulated and extrapolated into a second predicted water depth that includes spatial interaction and temporal accumulation, thus providing a reliable dynamic basis for subsequent accurate assessment of urban flooding risk.

[0037] In one possible design, during a prediction period starting from the prediction time corresponding to the first predicted water depth, a second predicted water depth for a subsequent prediction time adjacent to the previous prediction time is determined using a cellular automata approach during the chronological water diffusion process. This determination is based on the known predicted water depth at the previous prediction time. , The known predicted water depth of the i-th location unit in the target geographic area at time t. The second predicted water depth for the i-th location unit at time t+1. Let be the time step between time t and time t+1. The inflow rate of the i-th position unit, The outflow rate of the i-th position unit is... The water exchange value between the i-th location unit and the adjacent j-th location unit is given.

[0038] In other words, the water depth of a location unit at the next moment is equal to the water depth of the location unit at the current moment plus the net change in inflow, outflow, and water exchange between adjacent units per unit time.

[0039] It should be added that, before step 106, based on the drainage network data, the drainage network nodes and edges, as well as the real-time operating status features of the drainage network nodes and the attribute features of the edges, need to be encoded into graph structure data; using the graph structure data as input information for the graph neural network, the dynamic drainage capacity value of each location unit in the target geographical area at the current time is determined by the graph neural network; based on the dynamic drainage capacity value of each location unit at the current time, the inflow and outflow of each location unit are calculated.

[0040] The graph structure data includes: an adjacency matrix representing the topology of the drainage network, a feature matrix representing the real-time operational status of the nodes in the drainage network, and an edge feature matrix representing the physical properties of the pipelines. In essence, the graph neural network reflects the correlation between the graph structure data of the drainage network in the target geographical area and the drainage capacity corresponding to that network.

[0041] The assessment of dynamic drainage capacity can evaluate the drainage system's ability to absorb runoff in real time and take into account the dynamic operating status of the pipe network. In a practical scenario, a graph neural network based on the PyTorch Geometric library can be used to model the urban drainage pipe network as a graph, in which drainage wells, pumping stations, and storage tanks are nodes, and pipes are edges.

[0042] , This represents the dynamic drainage capacity value of the i-th drainage network node in the target geographical area. That is, the adjacency matrix that represents the topology of the drainage pipe network. The feature matrix is ​​used to characterize the real-time operating status of the i-th drainage network node at time t. The real-time operating status features include, but are not limited to, the real-time water level, flow rate, pump station operating status, and water storage capacity of the regulating reservoir at time t. The edge feature matrix is ​​used to characterize the physical properties of a pipeline. The edge features include, but are not limited to, fixed attributes such as the pipe diameter, material, length, and slope.

[0043] It can be said that the dynamic drainage capacity value of a drainage network node reflects the instantaneous maximum drainage efficiency of the node under the combined effects of the current topology, real-time load, and pipeline attributes. Therefore, the dynamic drainage capacity value of a drainage network node can be calculated in the above way, thereby providing a basis for the calculation of the inflow and outflow of each location unit in the cellular automata model.

[0044] Step 108: Determine the flood risk level of the target geographical area based on the second predicted water depth of each location unit within the target geographical area at a specified prediction time.

[0045] For each location unit within the target geographical area, the second predicted water depth at a specified prediction time reflects the possibility that the water in the location unit may cause flooding in the future after dynamic diffusion and pipe network absorption. Therefore, the risk of flooding in the entire target geographical area can be calculated by the second predicted water depth of each location unit at a specified prediction time.

[0046] In one possible design, based on the second predicted water depth of each location unit within the target geographical area at a specified prediction time, the probability that the water depth of each location unit exceeds a preset disaster-causing threshold is calculated; based on the probability, the waterlogging risk level of the target geographical area is determined, wherein... , This indicates the risk of urban flooding in the target geographical area. Indicates the location of the target geographic area The water depth at the specified prediction time τ, To preset the disaster threshold, Indicates the position The vulnerability index, P(·), represents the probability that the water depth exceeds the preset disaster threshold.

[0047] In essence, the probability that the water depth of a location unit exceeds a preset disaster-causing threshold reflects the likelihood of a hazardous flooding event occurring at that location unit at a specified prediction time. Based on this, the probability of the water depth exceeding the preset disaster-causing threshold for each location unit can be spatially weighted and integrated with its corresponding socioeconomic vulnerability index to obtain the overall flooding risk level of the target geographical area at the specified prediction time τ. This overall risk level comprehensively reflects the probability of flooding occurring in the physical dimension, or in other words, it comprehensively reflects the severity of the potential flooding disaster consequences in the target geographical area.

[0048] In summary, the above technical solutions, by integrating multi-source data such as high-resolution remote sensing imagery, lidar point clouds, and drainage network data, construct underlying surface characteristics that characterize the physical properties of complex urban underlying surfaces. This allows for the use of a water depth prediction model to predict the water depth in a target geographical area under these underlying surface characteristics and current precipitation intensity. Furthermore, the prediction results are used to simulate water diffusion, determining the water depth at each location unit within the target geographical area at a specified prediction time, serving as the basis for calculating the urban flooding risk of the target geographical area. This overcomes the shortcomings of related technologies, such as complex hydrodynamic model calculations, coarse underlying surface characterization, and difficulty in comprehensively considering drainage system responses. It achieves real-time, efficient, and accurate prediction of urban flooding risk on complex underlying surfaces, significantly improving the timeliness and reliability of flood disaster early warning, and providing reliable technical support for urban disaster prevention and mitigation as well as emergency management.

[0049] In one possible design, the underlying surface characteristics include the building density, impermeable surface ratio, and green coverage of each location unit within the target geographic area. Building density reflects the likelihood that buildings within a location unit will obstruct the ground, impede surface runoff, and potentially create localized waterlogging spaces. The impermeable surface ratio reflects the degree to which the surface within a location unit inhibits rainwater infiltration. Green coverage reflects the surface's ability to retain rainwater through vegetation interception and soil infiltration. These three factors reflect the potential waterlogging level of a location unit from different perspectives; in other words, they reflect the likelihood of waterlogging occurring within a location unit from different dimensions.

[0050] Based on this, step 102 includes: using the high-resolution remote sensing image of the target geographic area as input information for the land cover type classification model, and determining the land cover type to which each pixel of the high-resolution remote sensing image belongs through the land cover type classification model.

[0051] The land cover type classification model reflects the relationship between pixel representation and the land cover type to which the pixel belongs. Optionally, the land cover type classification model is a semantic segmentation network such as a convolutional neural network or U-Net. Land cover types include, but are not limited to, any content that may cover the land surface, such as buildings (roofs, walls), different types of green space (lawns, shrubs, trees), various impermeable surfaces (asphalt pavement, concrete pavement, plazas), bare soil, and water bodies.

[0052] Next, step 102 further includes: for each location unit within the target geographic area, determining the building density of the location unit based on the number of pixels in the location unit whose surface cover type belongs to the building type and the total number of pixels in the location unit; determining the impermeable surface ratio of the location unit based on the number of pixels in the location unit whose surface cover type belongs to the impermeable type and the total number of pixels in the location unit, wherein the impermeable type is the superordinate type of the building type; and determining the green coverage rate of the location unit based on the number of pixels in the location unit whose surface cover type belongs to the green space type and the total number of pixels in the location unit.

[0053] Optionally, the building density of the location unit is the ratio of the number of pixels whose surface cover type belongs to the building type to the total number of pixels in the location unit; the impermeable surface ratio of the location unit is the ratio of the number of pixels whose surface cover type belongs to the impermeable type to the total number of pixels in the location unit; and the green coverage rate of the location unit is the ratio of the number of pixels whose surface cover type belongs to the green space type to the total number of pixels in the location unit.

[0054] In one embodiment of this application, a U-Net++ deep learning network implemented based on the PyTorch framework is used as the land cover type classification model. Its training data comes from a manually labeled sample set of high-resolution remote sensing images. The preprocessed high-resolution remote sensing images are input into the PyTorch-based classification model to classify each pixel. Finally, the percentage of each land cover type within each 5m × 5m location unit of the target geographic area is determined, and then the building density, impervious surface ratio, and green space coverage rate of each location unit are calculated.

[0055] In another possible design, the underlying surface performance characteristics also include the surface roughness of each location unit within the target geographic area. In this case, step 102 includes: determining the pixel percentage of each type of land cover within the location unit; and weighting and summing the pixel percentages of each type of land cover using a predetermined Manning roughness coefficient as the weight to obtain the surface roughness of the location unit.

[0056] The surface roughness of each location unit reflects the overall resistance level of surface cover types within that unit to surface runoff, directly reflecting the velocity and diffusion capacity of surface water. The predetermined Manning roughness coefficient corresponding to each surface cover type reflects the inherent physical property of that type of surface in terms of flow resistance in hydrodynamics; therefore, it can be used as a weight for each surface cover type to quantify its contribution to the overall resistance level of surface runoff.

[0057] In one embodiment of this application, a U-Net++ deep learning network based on the PyTorch framework is used as a land cover type classification model to perform classification. After classification, empirical Manning roughness coefficients are assigned to different land cover types, and the land surface roughness is obtained by weighting each 5m × 5m location unit in the target geographic area using the Manning roughness coefficients as weights.

[0058] In another possible design, the underlying surface characteristics also include the hydro-topographic features and urban drainage network density of each location unit within the target geographic area. In this case, step 102 includes: extracting ground points based on the lidar point cloud data of the target geographic area to construct a digital elevation model of the target geographic area; and performing terrain analysis processing on the digital elevation model according to a predetermined terrain analysis method to determine the hydro-topographic features of each location unit within the target geographic area.

[0059] LiDAR point cloud data reflects the detailed three-dimensional structure of the terrain within the target geographic area. Based on this data, hydro-topographic features such as average elevation, slope, slope length, aspect, topographic relief, runoff path, runoff direction, catchment area, regional volume, and maximum depth can be extracted for each location unit within the target geographic area. These hydro-topographic features reflect the influence of the landform of the target geographic area on precipitation convergence, flow path, and water accumulation distribution, and are one of the bases for effectively assessing the risk of urban flooding. When performing terrain analysis processing on the digital elevation model according to a predetermined terrain analysis method, the predetermined terrain analysis method includes, but is not limited to, using geographic information system software such as ArcGIS or QGIS.

[0060] Drainage network data reflects the layout of artificial drainage facilities within a target geographical area, providing a quantitative representation of the area's drainage capacity. Based on this data, the total length and area of ​​drainage pipes for each location unit within the target geographical area can be determined. The ratio of the total drainage pipe length to the unit area is then used as the urban drainage network density for that location unit. In short, urban drainage network density reflects the density of artificial drainage infrastructure per unit area within a target geographical area. The higher the density of artificial drainage infrastructure per unit area within an area, the stronger the area's ability to quickly remove surface water through the drainage network system.

[0061] In the training process of the water depth prediction model, in addition to the traditional prediction error loss, a constraint term based on physical knowledge can be introduced to form a composite loss function as follows.

[0062] , The prediction error loss is the result of the composite loss function. It is the error between the model's predicted value and the actual value. This represents a water conservation constraint, ensuring that the inflow and outflow predicted by the model satisfy the law of conservation of mass. For example, during training, penalties are imposed on the predicted rates of change in water accumulation and runoff to make them conform as closely as possible to the physical water balance. This reflects the terrain gradient constraint, ensuring that the water flow direction predicted by the model is consistent with the actual terrain gradient direction, and avoiding the physical inconsistency of water flowing uphill. To constrain extreme cases, a small number of extreme historical flood event samples are added to the training data, and special weighting or loss function design is used to ensure that the model can accurately simulate urban flooding when facing extreme precipitation, avoiding prediction failure under boundary conditions.

[0063] This allows the trained deep learning model to not only fit the data but also follow basic physical laws, improving the model's interpretability, stability, and generalization ability under unknown and extreme conditions.

[0064] exist Figure 1 Based on the implementation examples, Figure 2 A schematic diagram illustrating intelligent extraction and fusion of multi-dimensional features of urban underlying surface according to an embodiment of this application is shown.

[0065] like Figure 2 As shown, a comprehensive feature matrix of the urban underlying surface is constructed. The process is divided into three main stages.

[0066] The first stage involves multi-source data input and intelligent processing. High-resolution remote sensing imagery undergoes a data training phase, followed by feature extraction using convolutional neural networks or U-Net semantic segmentation techniques. LiDAR point cloud data is processed through a linear discrimination phase using point cloud processing algorithms. Urban drainage network data undergoes feature parameter extraction and is dynamically represented based on graph neural network (GNN) modeling.

[0067] The second stage involves generating key feature parameters, including extracting building density from high-resolution remote sensing images. , Green coverage rate impermeable surface density Extract refined DEMz (elevation model) and surface roughness n from LiDAR point cloud data; extract drainage network density (or dynamic status) from drainage network data. .

[0068] The third stage involves multi-dimensional feature fusion, which integrates the six feature parameters mentioned above and performs data normalization to ultimately construct a unified comprehensive feature matrix of the urban underlying surface. .

[0069] This enables the automated and refined extraction and fusion of multidimensional physical properties of urban underlying surfaces, providing high-quality input features for subsequent urban flooding risk assessment.

[0070] Figure 3 A flowchart of a deep neural network processing method based on a multilayer perceptron according to an embodiment of this application is shown.

[0071] like Figure 3 As shown, the graph neural network for predicting dynamic drainage capacity and the water depth prediction model for predicting water depth can be implemented in a deep neural network based on a multilayer perceptron, respectively.

[0072] First, based on deep neural networks ( In predicting urban flooding depth, a nonlinear mapping relationship between precipitation and flooding depth is established. This is based on precipitation intensity time series. , underlying surface feature vector and initial conditions for The input data is divided into training and testing sets, and the final output is the predicted depth of urban flooding. .

[0073] Among these methods, the accuracy of the model's prediction results can be verified using the test set, and hyperparameters can be optimized. That is, based on the verification results, the key hyperparameters of the model can be optimized and adjusted, such as the learning rate, batch size, and number of iterations.

[0074] Next, in the dynamic drainage capacity assessment based on graph neural networks (GNNs), the drainage capacity of the drainage network under real-time conditions is evaluated. This is done using an adjacency matrix. Drainage pipe network topology and node status characteristics Real-time water level and edge features ( Pipe diameter, slope, and material are used as input data for a graph neural network. An ensemble averaging method is used to fuse the calculation results of the graph neural network after multiple optimizations to obtain the dynamic drainage capacity. .

[0075] Ultimately, based on the predicted depth of flooding and dynamic drainage capacity By combining Bayesian statistical methods, dynamic assessment of drainage capacity can be achieved.

[0076] Figure 4 A schematic diagram of a cellular automata modeling process for the propagation of urban flooding risk is shown according to an embodiment of this application.

[0077] like Figure 4 As shown, firstly, the cell water accumulation depth is initialized. , that is, the water depth of the i-th position unit at time t, and the initial water state at the start of the simulation is set for each position unit.

[0078] Then, the iterative loop begins, which is the core process of cellular automata (CA) risk propagation modeling. Step 1: Update the inflow. Step 1: Determine the amount of water entering each cell within the current time step; Step 2: Update the outflow rate. (Considering terrain and) The process calculates the water discharge volume for each cell, taking into account both topographical factors and dynamic drainage capacity. Step 3: Simulate water exchange between adjacent units. Step 4: Iteratively update the water depth using the core formula. Calculate the water depth of each cell at the next time step. .

[0079] Ultimately, a visual output is achieved. Through the aforementioned cyclical simulation, a real-time water depth map can be generated, intuitively displaying the spatial distribution of water accumulation at different times. Simultaneously, meteorological forecast information can be integrated, inputting different forecast rainfall scenarios into the aforementioned CA model and combining it with the regional vulnerability index V to conduct multi-scenario risk pre-assessment, thereby predicting potential future urban flooding risks.

[0080] In summary, by dynamically simulating water accumulation and diffusion using cellular automata and combining forecast information for multi-scenario analysis, a visualized assessment of the initial water situation and future risks was achieved.

[0081] In another embodiment, this application provides a computer device, which may be a server, and its internal structure diagram may be as follows. Figure 5 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with external clients via a network connection. When the computer program is executed by the processor, it can implement the methods described in any of the above embodiments.

[0082] In one embodiment, this application also provides a computer device, which can be a client, and its internal structure diagram can be as follows: Figure 6 As shown, the computer device includes a processor, memory, network interface, display screen, and input device connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface is used to communicate with an external server via a network connection. When the computer program is executed by the processor, it can implement the methods described in any of the above embodiments.

[0083] Any of the computer devices described in the embodiments of this application exist in various forms, including but not limited to: (1) Mobile communication devices: These devices are characterized by their mobile communication capabilities and primarily aim to provide voice and data communication. These terminals include smartphones, multimedia phones, feature phones, and low-end phones.

[0084] (2) Ultra-mobile personal computer devices: These devices fall under the category of personal computers, possessing computing and processing capabilities, and generally also have mobile internet access features. These terminals include PDAs, MIDs, and UMPCs, etc.

[0085] (3) Portable entertainment devices: These devices can display and play multimedia content. This category includes: audio and video players, handheld game consoles, e-books, as well as smart toys, wearable devices, and portable car navigation devices.

[0086] (4) Server: A device that provides computing services. The components of a server include a processor, hard disk, memory, system bus, etc. Servers are similar to general computer architectures, but because they need to provide highly reliable services, they have higher requirements in terms of processing power, stability, reliability, security, scalability, and manageability.

[0087] (5) Other electronic devices with data interaction functions.

[0088] Additionally, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which are used to perform the following steps: Based on high-resolution remote sensing images, lidar point cloud data, and drainage network data of the target geographic area, the underlying surface characteristics of the target geographic area are determined. The underlying surface characteristics are used to reflect the physical properties of the surface of the target geographic area after being affected by geographic and human factors. The underlying surface characteristics of the target geographic region, the precipitation time series, and the initial surface water depth of the target geographic region before precipitation are used as input information for the water depth prediction model. The water depth prediction model outputs the first predicted water depth of each location unit in the target geographic region. Based on the first predicted water depth, the water diffusion process of the target geographical area is simulated by cellular automata to obtain the second predicted water depth of each location unit in the target geographical area at a specified prediction time. The waterlogging risk level of the target geographical area is determined based on the second predicted water depth of each location unit within the target geographical area at a specified prediction time.

[0089] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or computer device described above can be referred to the relevant descriptions in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0090] The technical solution of this application has been described in detail above with reference to the accompanying drawings. This technical solution integrates multi-source data such as high-resolution remote sensing images, lidar point clouds, and drainage pipe networks to construct underlying surface characteristics that characterize the physical properties of complex urban underlying surfaces. A water depth prediction model is then used to predict the water depth in a target geographical area under these underlying surface characteristics and the current precipitation intensity. Furthermore, the prediction results are used to simulate water diffusion, simulating the water depth of each location unit within the target geographical area at a specified prediction time, serving as the basis for calculating the urban flooding risk of the target geographical area. This overcomes the shortcomings of related technologies, such as complex hydrodynamic model calculations, coarse underlying surface characterization, and difficulty in comprehensively considering drainage system responses. It achieves real-time, efficient, and accurate prediction of urban flooding risk on complex underlying surfaces, significantly improving the timeliness and reliability of flood disaster early warning, and providing reliable technical support for urban disaster prevention and mitigation work and emergency management.

[0091] It should be understood that the term "and / or" used in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0092] It should be understood that although the terms "first," "second," etc., may be used to describe the predicted water depth in the embodiments of this application, these predicted water depths should not be limited to these terms. These terms are only used to distinguish the predicted water depths from one another. For example, without departing from the scope of the embodiments of this application, the first predicted water depth may also be referred to as the second predicted water depth, and similarly, the second predicted water depth may also be referred to as the first predicted water depth.

[0093] Depending on the context, the word "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0094] The terminology used in the embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. The singular forms “a,” “the,” and “the” used in the embodiments of this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise.

[0095] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0096] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0097] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0098] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A machine learning-based method for assessing urban surface flooding risk, characterized in that, include: Based on high-resolution remote sensing images, lidar point cloud data, and drainage network data of the target geographic area, the underlying surface characteristics of the target geographic area are determined. The underlying surface characteristics are used to reflect the physical properties of the surface of the target geographic area after being affected by geographic and human factors. The underlying surface characteristics of the target geographic region, the precipitation time series, and the initial surface water depth of the target geographic region before precipitation are used as input information for the water depth prediction model. The water depth prediction model outputs the first predicted water depth of each location unit in the target geographic region. Based on the first predicted water depth, the water diffusion process of the target geographical area is simulated by cellular automata to obtain the second predicted water depth of each location unit in the target geographical area at a specified prediction time. The waterlogging risk level of the target geographical area is determined based on the second predicted water depth of each location unit within the target geographical area at a specified prediction time.

2. The method according to claim 1, characterized in that, The underlying surface characteristics include the building density, impermeable surface ratio, and green coverage of each location unit within the target geographic area. Therefore, determining the underlying surface characteristics of the target geographic area based on high-resolution remote sensing imagery, lidar point cloud data, and drainage network data includes: Using the high-resolution remote sensing image of the target geographic area as input information for the land cover type classification model, the land cover type of each pixel in the high-resolution remote sensing image is determined by the land cover type classification model. For each location unit within the target geographic region. The building density of the location unit is determined based on the number of pixels in the location unit whose land cover type belongs to the building type and the total number of pixels in the location unit. Based on the number of pixels in the location unit whose surface cover type is impermeable, and the total number of pixels in the location unit, the proportion of impermeable surface in the location unit is determined, wherein the impermeable type is the superordinate type of the building type; The green coverage rate of the location unit is determined based on the number of pixels in the location unit whose land cover type is green space, and the total number of pixels in the location unit.

3. The method according to claim 2, characterized in that, The underlying surface characteristics include the surface roughness of each location unit within the target geographic area. Therefore, determining the underlying surface characteristics of the target geographic area based on high-resolution remote sensing imagery, lidar point cloud data, and drainage network data includes: Determine the pixel percentage of each land cover type within the location unit; The surface roughness of the location unit is obtained by weighting the pixel proportion of each of the aforementioned surface cover types by using a predetermined Manning roughness coefficient as the weight.

4. The method according to claim 3, characterized in that, The underlying surface characteristics also include the hydrotopographic features and urban drainage network density of each location unit within the target geographic area. Therefore, determining the underlying surface characteristics of the target geographic area based on high-resolution remote sensing imagery, lidar point cloud data, and drainage network data includes: Based on the lidar point cloud data of the target geographic area, ground points are extracted to construct a digital elevation model of the target geographic area. The digital elevation model is subjected to terrain analysis processing according to a predetermined terrain analysis method to determine the hydrological and topographic features of each location unit within the target geographic area. These hydrological and topographic features include: average elevation, slope, slope length, aspect, topographic relief, runoff path, runoff direction, catchment area, regional volume, and maximum depth. Based on the drainage network data of the target geographical area, the total length of the drainage pipes and the area of ​​each location unit within the target geographical area are determined, and the ratio of the total length of the drainage pipes to the area of ​​the unit is used as the urban drainage network density of the location unit.

5. The method according to any one of claims 1 to 4, characterized in that, The determination of the underlying surface characteristics of the target geographic area based on high-resolution remote sensing imagery, lidar point cloud data, and drainage network data includes: Based on the high-resolution remote sensing imagery of the target geographic area, the lidar point cloud data, and the drainage network data, the building density, impermeable surface ratio, green coverage, surface roughness, hydro-topographic features, and urban drainage network density of each location unit within the target geographic area are determined. A comprehensive feature matrix of the underlying surface corresponding to each location unit is constructed based on the building density, impermeable surface ratio, green coverage, surface roughness, hydro-topographic features, and urban drainage network density of each location unit. The step of using the underlying surface characteristics of the target geographic region, precipitation time series, and the initial surface water depth of the target geographic region before precipitation as input information for the water depth prediction model, and outputting the first predicted water depth for each location unit within the target geographic region through the water depth prediction model, includes: For each location unit within the target geographic area, the comprehensive feature matrix of the underlying surface corresponding to the location unit, the precipitation time series, and the initial surface water depth of the location unit before precipitation are used as the input information of the water depth prediction model. The water depth prediction model outputs the first predicted water depth of the location unit.

6. The method according to claim 5, characterized in that, The element in the m-th row and n-th column of the comprehensive feature matrix of the underlying surface corresponding to the location unit is the weighted average of the covariance and normalized product of the m-th and n-th features among the building density, impervious surface ratio, green coverage, surface roughness, hydrotopographic features, and urban drainage network density of the location unit. The weights of the covariance and the weights of the normalized product sum to 1. The weights of the covariance are used to reflect the accuracy of the covariance in describing the correlation between the m-th feature and the n-th feature in the historical urban surface waterlogging risk assessment. The weights of the normalized product are used to reflect the accuracy of the normalized product in describing the correlation between the m-th feature and the n-th feature in the historical urban surface waterlogging risk assessment business.

7. The method according to claim 5, characterized in that, The step of simulating the water diffusion process in the target geographical area based on the predicted water depth using cellular automata includes: In the prediction period starting from the prediction time corresponding to the first predicted water depth, a second predicted water depth at a subsequent prediction time is determined using a cellular automata approach during the chronological water diffusion process, based on the known predicted water depth at the previous prediction time. ， The known predicted water depth of the i-th location unit in the target geographic area at time t. The second predicted water depth for the i-th location unit at time t+1. Let be the time step between time t and time t+1. The inflow rate of the i-th position unit, The outflow rate of the i-th position unit is... The water exchange value between the i-th location unit and the adjacent j-th location unit is given.

8. The method according to claim 7, characterized in that, Before simulating the water diffusion process of the target geographical area using cellular automata based on the predicted water depth, the method further includes: Based on the drainage network data, the drainage network nodes and edges, as well as the real-time operating status features of the drainage network nodes and the attribute features of the edges, are encoded into graph structure data. The graph structure data includes: an adjacency matrix representing the topology of the drainage network, a feature matrix representing the real-time operating status features of the drainage network nodes, and an edge feature matrix representing the physical properties of the pipeline. Using the graph structure data as input information for a graph neural network, the dynamic drainage capacity value of each location unit in the target geographic area at the current moment is determined by the graph neural network. Based on the dynamic drainage capacity value of each location unit at the current moment, the inflow and outflow of each location unit are calculated.

9. The method according to claim 7, characterized in that, The determination of the flood risk level of the target geographical area based on the second predicted water depth of each location unit within the target geographical area at a specified prediction time includes: Based on the second predicted water depth of each location unit within the target geographical area at a specified prediction time, the probability that the water depth of each location unit exceeds a preset disaster threshold is calculated. Based on the aforementioned probability, the waterlogging risk level of the target geographical area is determined, wherein... ， This indicates the risk of urban flooding in the target geographical area. Indicates the location of the target geographic area The water depth at the specified prediction time τ, To preset the disaster threshold, Indicates the position The vulnerability index, P(·), represents the probability that the water depth exceeds the preset disaster threshold.

10. A computer device, characterized in that, include: At least one processor; And, a memory communicatively connected to the at least one processor; The memory stores instructions executable by the at least one processor, the instructions being configured to cause the processor to perform the method according to any one of claims 1 to 9.

11. A computer-readable storage medium, characterized in that, The device stores computer-executable instructions configured to perform the method as described in any one of claims 1 to 9.

Images