A waterlogging prediction method based on behavioral heat map and infiltration dynamic modeling

Abstract

This invention provides a method for predicting urban flooding based on behavioral heat maps and dynamic infiltration modeling. This method first divides spatial regions based on terrain data, extracts surface type, soil properties, and vegetation information, and constructs an infiltration capacity model that dynamically adjusts with rainfall and time. It then combines surface temperature and meteorological data to establish an evaporation model that accounts for the urban heat island effect. Furthermore, it uses crowd trajectory data to generate behavioral heat maps and extract behavioral feedback factors that reflect abnormal movement patterns. Finally, it integrates rainfall, infiltration, evaporation, drainage capacity, and behavioral feedback factors to construct a machine learning model for predicting the probability of urban flooding in each region. The method optimizes the loss function by introducing a weighting mechanism for behavioral feedback factors, improving the model's sensitivity to risk areas. This method combines physical mechanisms with social behavioral characteristics, improving the accuracy and timeliness of urban flooding predictions in complex urban environments.

Description

Technical Field

[0001] The present invention belongs to the technical field of urban waterlogging prediction, and in particular relates to an urban waterlogging prediction method based on behavioral heat maps and infiltration dynamic modeling. Background Art

[0002] With the accelerating pace of urbanization, urban waterlogging is becoming an increasingly prominent problem, a significant factor impacting the safe operation of cities and the quality of life of residents. Waterlogging not only causes traffic disruptions and property damage, but also poses a serious threat to public safety. Especially in the context of frequent extreme weather events, preventing and controlling waterlogging is particularly urgent.

[0003] Existing flooding prediction methods primarily rely on meteorological data and drainage system information, assessing flooding risk by analyzing rainfall patterns and drainage capacity. However, due to the complex and varied urban topography and significant variations in land cover types, models that fail to fully account for these factors often suffer from reduced accuracy in flooding simulations. Differences in surface hardening and soil infiltration capacity across regions directly influence the flow path and accumulation rate of rainwater. Ignoring these factors can lead to significant deviations in model predictions of flooding risk.

[0004] Furthermore, localized microclimate changes in urban environments, such as the urban heat island effect, caused by factors like dense buildings and scarce green spaces, can affect the evaporation and infiltration of surface water, thus influencing the formation and resolution of urban waterlogging. If prediction models fail to account for these environmental dynamics, it will be difficult to accurately predict the spatial and temporal distribution and severity of urban waterlogging.

[0005] In the prior art, Chinese patent CN119378448A discloses a method for constructing an urban waterlogging prediction model, an urban waterlogging prediction method, and an apparatus, relating to the field of data processing technology. The method for constructing an urban waterlogging prediction model includes: constructing an urban waterlogging process model coupled with an emergency drainage mode based on regional topographic data, hydrological and meteorological data, and flood control response data; inputting rainfall data under different scenarios into the urban waterlogging process model to obtain waterlogging characteristic data; performing a correlation analysis between the rainfall data and the waterlogging characteristic data, and determining rainfall parameters related to flood risk from the rainfall data; constructing sample data based on the rainfall parameters and the waterlogging characteristic data, and training a neural network model based on the sample data to obtain an urban waterlogging prediction model. Although this method takes into account factors such as topography, hydrology, and drainage facilities to a certain extent, its model does not adequately consider differences in surface cover, local environmental changes, and real-time dynamic responses, which can easily lead to insufficient adaptability of the prediction results to complex actual situations.

[0006] In addition, existing methods are mainly based on preset flood control response measures and drainage facility scheduling, and lack real-time reflection of urban dynamic changes (such as population distribution), which affects the real-time performance and accuracy of the prediction model.

[0007] Therefore, existing technologies still have shortcomings in comprehensive consideration of multiple factors and dynamic adaptability of models, and it is difficult to meet the high-precision and real-time requirements of urban waterlogging risk management. Summary of the Invention

[0008] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and provide a waterlogging prediction method based on behavioral thermogram and infiltration dynamic modeling.

[0009] The purpose of the present invention can be achieved by the following technical solutions:

[0010] The present invention provides a waterlogging prediction method based on behavioral heat maps and infiltration dynamic modeling, comprising the following steps:

[0011] Collect multi-source data in the target area, including environmental data, meteorological data, terrain data, drainage network data, crowd trajectory data, and surface temperature data;

[0012] Divide spatial regions based on terrain data, extract the surface cover type, soil properties, and vegetation coverage of each spatial region, and construct a rainwater infiltration capacity model that dynamically adjusts with time and rainfall conditions;

[0013] Based on surface temperature data and environmental data, an evaporation model reflecting the urban heat island effect is established to calculate the evaporation capacity of each spatial area;

[0014] Use crowd trajectory data to generate a behavior heat map that changes over time, and use the degree of crowd aggregation and abnormal behavior changes in the behavior heat map as behavior feedback factors;

[0015] A waterlogging prediction model was constructed by combining rainwater infiltration capacity models, evaporation models, behavioral feedback factors, meteorological data, and drainage network data;

[0016] According to the waterlogging prediction model, the probability of waterlogging occurrence in each spatial area in the future period is output.

[0017] Furthermore, the construction of a rainwater infiltration capacity model that is dynamically adjusted with time and rainfall conditions specifically includes:

[0018] dividing the target area into a number of spatial regions according to the terrain data;

[0019] For each spatial region, based on the terrain data, the main surface types in the spatial region are extracted, including bare soil, grass, concrete, asphalt, and masonry pavement, and the area proportion of each surface type in the spatial region is calculated. ;

[0020] For each surface type j, obtain the corresponding infiltration related parameters, including: saturated hydraulic conductivity , capillary suction , saturated volume moisture content , Volume moisture content in the previous time period ; Among them, saturated hydraulic conductivity Refers to surface type j The amount of water that can penetrate per unit time under fully saturated conditions, capillary suction Refers to the type of moisture on the surface j The suction force when adsorbed in the capillary pores of the soil, the saturated volume water content Refers to surface type j The maximum amount of water that a unit volume of soil can hold in a fully saturated state, the volumetric moisture content in the previous time period Refers to surface type j The actual soil moisture content at the previous moment; saturated hydraulic conductivity , capillary suction , saturated volume moisture content Obtained through geological data and soil database, volumetric moisture content in the previous time period Obtained through measured soil moisture sensor data;

[0021] Using the improved algorithm based on the Green-Ampt model, according to the infiltration related parameters of each spatial area, a rainwater infiltration capacity model of each spatial area with dynamic evolution over time is constructed to calculate the infiltration capacity of each spatial area at different times. t The instantaneous infiltration rate , calculated as follows:

[0022]

[0023] in, Indicates the i A spatial region in time t The instantaneous infiltration rate, indicating the i A spatial region in time t At time , the rate at which rainwater per unit area is absorbed by the surface of the area, Represents the time from the initial moment to the t , in i In the spatial region j The cumulative infiltration depth of the surface type is j The instantaneous infiltration rate of the surface type and whether there is precipitation are integrated over time. is the total number of surface types.

[0024] Furthermore, the target area is divided into several spatial areas according to the terrain data, specifically including:

[0025] According to the digital elevation model in the target area terrain data, the slope and slope direction of each grid cell in the digital elevation model are calculated;

[0026] Based on the slope and aspect, combined with the flow path analysis, the catchment area of ​​each grid cell is calculated;

[0027] By setting the watershed area threshold and slope threshold, the spatial areas that meet the conditions are screened:

[0028]

[0029] in, The set of spatial regions obtained by filtering and dividing the resulting terrain data, Represents a grid cell The catchment area, is the watershed area threshold, Represents a grid cell The slope, Indicates the slope threshold.

[0030] Furthermore, the evaporation model reflecting the urban heat island effect is established based on the surface temperature data and environmental data, specifically including:

[0031] Collect surface temperature data and environmental data of each spatial area in the target area. Environmental data include surface temperature, air temperature, wind speed, relative humidity and solar radiation;

[0032] Based on the surface temperature data and environmental data, a modified Penman-Monteith model was used to construct an evaporation model that takes into account the urban heat island effect, and the potential evaporation rate of each spatial area in the future time period was calculated. , the formula is as follows:

[0033]

[0034] in, Indicates the i Spatial region in time t potential evaporation rate; Indicates the i The slope of the saturated vapor pressure curve for a spatial region, calculated based on the air temperature; Indicates the i Net radiation of a spatial region, estimated from solar radiation and surface temperature; Indicates the i Soil heat flux in spatial regions; is the humidity constant; Indicates the i the average air temperature of a spatial region; Indicates thei The wind speed in the space area at a height of 2 meters above the ground, 、 Respectively represent i Saturated vapor pressure and actual vapor pressure in a spatial region.

[0035] Furthermore, the method of generating a behavior heat map that changes over time using crowd trajectory data specifically includes:

[0036] Collecting crowd trajectory data in the target area The trajectory data is obtained by mobile signals, WiFi detection or video monitoring, and is used to represent the individual's time t spatial location;

[0037] Map the personnel trajectory data to multiple spatial regions divided by the target area, and count the number of people in each spatial region at any time. t number of people inside ;

[0038] According to each spatial region at any time t number of people inside The area of ​​each spatial region Calculate the i A spatial region in time t Crowd density :

[0039]

[0040] Gaussian kernel function is used to calculate the crowd density Perform kernel density estimation and generate the moment t Behavior heat map:

[0041]

[0042] in, Indicates spatial location At the moment t The crowd behavior heat map value, Indicates the i The coordinates of the centroid of the spatial region, represents the total number of spatial regions, is the kernel function, is the bandwidth parameter.

[0043] Furthermore, the behavior feedback factor calculation process is:

[0044] Calculate the average density value of the historical behavior heat map , as a baseline reference;

[0045] Based on the average density value of the current moment behavior heat map and the historical behavior heat map of the same period , calculate the behavioral abnormality factor:

[0046]

[0047] in, Indicates spatial location At the moment t behavioral abnormalities;

[0048] Setting thresholds , when behavioral abnormality factor When , the crowd gathering intensity is extracted as the risk weighting factor:

[0049]

[0050] Risk-weighted factors Integrate and average by spatial region to obtain i Behavioral feedback factor for each spatial region:

[0051]

[0052] in, Indicates the i A spatial region in time t behavioral feedback factors.

[0053] Furthermore, the historical period refers to a historical time period corresponding to the current time point in terms of time cycle.

[0054] Furthermore, the waterlogging prediction model is constructed by combining the rainwater infiltration capacity model, evaporation model, behavioral feedback factor, meteorological data and drainage network data, specifically including:

[0055] Get each spatial area i In time t Rainfall intensity , instantaneous infiltration rate , potential evaporation rate , drainage capacity , estimate each spatial region i Surface water accumulation index per unit area :

[0056]

[0057] in, Represents a spatial region i The concentration index of surface water per unit area is is the time interval, that is, the length of time from the previous time step to the current time step, 、 Represents spatial regions i rainfall intensity, instantaneous infiltration rate and potential evaporation rate, For the spatial region i Drainage capacity, according to each space area i Drainage network data acquisition; For the spatial region i The rainfall intensity is obtained based on meteorological data; Represents a spatial region i area;

[0058] The training samples are constructed using the characteristic information of each spatial region at each time step as input. The characteristic information includes the estimated surface water accumulation index, behavioral feedback factors and meteorological data, where the meteorological data includes the current rainfall intensity and the accumulated rainfall in the previous period.

[0059] Whether waterlogging has occurred in each spatial area is used as a supervised learning label, and the label is determined based on historical waterlogging event data, and the historical waterlogging event data is obtained based on remote sensing images and water level sensor monitoring data;

[0060] Constructing a waterlogging prediction model based on a classification model to predict the probability of waterlogging, wherein the classification model is a probabilistic model, including a gradient boosting decision tree model;

[0061] The waterlogging prediction model takes the characteristic information of each spatial area as input, outputs a waterlogging probability prediction value at the corresponding moment, and trains the waterlogging prediction model through a loss function according to the output waterlogging probability prediction value to obtain a trained waterlogging prediction model.

[0062] Furthermore, the drainage capacity The drainage network data of each spatial area is calculated using the Manning formula, which is:

[0063]

[0064] in, For the spatial region i The number of drainage pipes, Indicates the j The hydraulic radius of the drainage pipe, Indicates the j The hydraulic slope of the drainage pipe, Indicates the j The flow cross-sectional area of ​​the drainage pipe, Indicates the j Manning's roughness coefficient of the drainage pipe.

[0065] Furthermore, the loss function is:

[0066]

[0067] in, is the loss function, is the weight coefficient, To adjust the parameters, For the spatial region i behavioral feedback factor, Represents a spatial region i At the moment t The real label of waterlogging, 1 means waterlogging has occurred, 0 means no waterlogging has occurred, and the label is determined based on historical waterlogging event data; The output of the waterlogging prediction model is i A spatial region at time t The predicted value of the probability of waterlogging; represents the total number of spatial regions, Represents the number of time steps in the training sample.

[0068] Compared with the prior art, the present invention has the following advantages:

[0069] (1) The present invention calculates the slope and slope direction of each grid cell by using the digital elevation model in the terrain data, and calculates the catchment area based on the slope and water flow path analysis. It uses the watershed area threshold and slope threshold to screen and divide the target area, solving the problem that the existing technology does not effectively divide the key waterlogging-prone areas in the target area, resulting in a large amount of calculation for waterlogging prediction and difficulty in accurately identifying waterlogging risks. By screening out spatial areas with large catchment areas and steeper slopes, the key waterlogging-prone areas in the target area are accurately located, avoiding redundant calculations for areas with less impact on water flow convergence or less prone to waterlogging, thereby significantly reducing computing resource consumption and improving the computational efficiency and response speed of the waterlogging prediction model. At the same time, this division method ensures that the waterlogging prediction model can focus on areas with active hydrological processes, improve the pertinence and accuracy of the prediction, and enhance the model's ability to identify waterlogging risks and its practical application value.

[0070] (2) The present invention calculates the instantaneous infiltration rate, aiming to solve the problems of rough modeling of rainwater infiltration process, insufficient dynamic response capability, and insufficient consideration of surface heterogeneity in traditional waterlogging simulation. During actual rainfall, the surface infiltration capacity is constantly changing, affected by the combined influence of surface type, soil moisture state, and rainfall intensity. However, the existing technology does not take this into account. The present invention introduces a dynamic infiltration capacity modeling method based on the Green-Ampt model to dynamically update and calculate the instantaneous infiltration rate at each time point. Through the refined and dynamic calculation of the instantaneous infiltration rate, the present invention achieves an accurate simulation of the surface infiltration response to rainfall, can more accurately judge the time point and intensity of surface runoff during rainfall, enhances the model's response capability to the dynamic behavior of rainwater, significantly improves the physical rationality of surface runoff simulation, reduces the error accumulation caused by inaccurate infiltration estimation, and thus improves the accuracy and reliability of the urban waterlogging prediction system as a whole, providing a more timely and spatially targeted decision-making basis for urban drainage planning, waterlogging warning, and emergency response.

[0071] (3) The present invention aims to solve the problem of ignoring or roughly treating the evaporation process in traditional waterlogging simulations by constructing an evaporation model under the urban heat island effect based on surface temperature data and environmental data. This is especially true in areas with significant urban heat island effects, where evaporation capacity varies dramatically and has a significant impact on the surface water balance. However, the existing technology does not take rainwater evaporation into account when considering waterlogging, resulting in insufficient simulation accuracy of key processes in the surface water cycle. The present invention collects multi-source environmental data such as surface temperature, air temperature, wind speed, relative humidity and solar radiation, and adopts a modified Penman-Monteith model to establish an evaporation model that reflects the urban heat island effect. It dynamically calculates the potential evaporation rate of each spatial region in the future time period. By introducing high-precision dynamic evaporation modeling, the present invention effectively enhances the simulation capability of the surface water evaporation process, providing key support for accurately estimating the net surface water accumulation, judging local runoff generation and waterlogging evolution. This method improves the physical completeness and parameter adaptability of the entire urban flood simulation system, significantly enhances the model's prediction accuracy and spatiotemporal resolution in complex urban environments, and provides more reliable technical support for urban water environment management, extreme rainfall warning, and drainage scheduling decisions.

[0072] (4) The behavioral feedback factor calculated by the present invention is an indicator for quantifying the degree of abnormal behavior of crowds, reflecting the degree of deviation of crowds from historical behavior patterns within a specific time and space range. This factor is based on the behavioral heat map generated by the crowd trajectory. By comparing the crowd density distribution at the current moment with the baseline density distribution at the same historical period, it identifies abnormal behaviors such as crowd gathering or avoidance, thereby inferring that the area may be experiencing abnormal environmental or social events. In extreme weather or sudden flooding, crowd behavior often responds before physical changes, such as detours, gatherings, detentions, and centralized evacuations. Therefore, the behavioral feedback factor has strong sensitivity and foresight in spatial and temporal dimensions, and can timely capture areas in the city where risk events may be occurring or about to occur. Using it as part of the flooding prediction model can effectively reflect the location and evolution trend of potential waterlogging risk areas, make up for the shortcomings of traditional hydrological models in real-time and sudden identification, improve the model's ability to respond to urban operation dynamics, and help achieve more timely and accurate risk warnings and scheduling responses.

[0073] (5) The present invention solves the problems of incomplete physical mechanism, insufficient response to dynamic changes, and lack of human behavior perception in existing waterlogging prediction by constructing a waterlogging prediction model that integrates rainwater infiltration capacity model, evaporation model, behavioral feedback factor, meteorological data and drainage network data. Traditional models often only consider rainfall intensity and drainage capacity, while ignoring the dynamic evolution of key water cycle processes such as surface infiltration and evaporation, resulting in inaccurate estimation of water surplus and prone to prediction bias. The present invention introduces high-resolution spatiotemporal dynamic instantaneous infiltration rate and potential evaporation rate, which effectively supplements the physical elements of surface water budget process modeling and improves the authenticity and sensitivity of surface water accumulation indicators per unit area. At the same time, the behavioral feedback factor reflects the response characteristics of the population in sudden waterlogging events, realizes enhanced recognition of abnormal areas, and helps capture signs of waterlogging that are difficult to directly obtain through ground perception signals.

[0074] (6) The loss function designed by the present invention solves the problems of sample imbalance and insufficient identification of risk areas in traditional flooding prediction by introducing a weight coefficient based on the behavioral feedback factor. The probability of flooding events is usually low, resulting in an extremely unbalanced distribution of positive and negative samples, which easily causes the model to predict the "no flooding" category, thereby ignoring potential high-risk areas. By incorporating the behavioral feedback factor into the weight, the sample loss weight corresponding to areas with a high probability of flooding or abnormal population movement is amplified, and the model pays more attention to the prediction accuracy of these key time and space points during the training process. The adjustment parameter α in the weight coefficient can flexibly control the influence of the behavioral feedback factor on the loss function, so as to achieve the model's sensitive response to risk areas guided by abnormal population behavior. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] Figure 1 This is a diagram of a method for predicting urban waterlogging according to an embodiment of the present invention;

[0076] Figure 2 This is a model diagram of the waterlogging prediction system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0077] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0078] This embodiment provides a waterlogging prediction method based on behavior heat map and infiltration dynamic modeling. Figure 1 As shown, the following steps are included:

[0079] Step S1: Collect multi-source data of the target area, including environmental data, meteorological data, terrain data, drainage network data, crowd trajectory data and surface temperature data; among them, environmental data include meteorological elements such as air temperature, wind speed, relative humidity and solar radiation, as well as relevant environmental characteristics reflecting the urban heat island effect. Environmental data can be obtained through multiple channels such as meteorological monitoring stations, environmental monitoring equipment, remote sensing satellites and ground sensors; meteorological data mainly include the current rainfall intensity and the accumulated rainfall in the past period, which are usually collected in real time through meteorological radar, rain gauge, meteorological satellite and meteorological forecast system; terrain data include digital elevation model (DEM), slope, slope aspect and catchment area information. Using these data, the target area can be divided into multiple spatial regions, and the surface cover type (such as bare soil, grass, concrete, asphalt) in each spatial region can be further extracted. Topographic data such as paving (green, brick and stone paving), soil properties, and vegetation cover are typically obtained through LiDAR, aerial photogrammetry, remote sensing satellite imagery, and geographic information system (GIS) databases. Drainage network data includes information such as the number, location, diameter, cross-sectional area, hydraulic radius, hydraulic gradient, and Manning's roughness coefficient of drainage pipes. This data is obtained from urban drainage network design drawings, network management systems, field measurements, and urban infrastructure databases. Crowd trajectory data represents the spatial locations of individuals within the target area at different points in time, reflecting the flow and aggregation of people. This data is collected through mobile communication signals, Wi-Fi detection devices, and video surveillance systems. Surface temperature data is collected through remote sensing satellite infrared thermal imaging equipment, ground-based thermal cameras, and environmental monitoring stations to provide surface thermal information. By collecting these multi-source data, we can provide a detailed understanding of the target area's natural environment, meteorological conditions, topographic characteristics, drainage capacity, and human behavior, providing a comprehensive and accurate data foundation for the subsequent construction of dynamic rainwater infiltration capacity models, evaporation models, behavioral feedback factors, and waterlogging prediction models.

[0080] Step S2: Divide the spatial regions according to the terrain data, extract the surface cover type, soil properties, and vegetation coverage of each spatial region, and construct a rainwater infiltration capacity model that is dynamically adjusted with time and rainfall conditions;

[0081] Step S2 specifically includes:

[0082] Based on the terrain data, the target area is divided into several spatial regions, including:

[0083] According to the digital elevation model in the target area terrain data, the slope and slope direction of each grid cell in the digital elevation model are calculated;

[0084] Based on the slope and aspect, combined with the flow path analysis, the catchment area of ​​each grid cell is calculated;

[0085] By setting the watershed area threshold and slope threshold, the spatial areas that meet the conditions are screened:

[0086]

[0087] in, The set of spatial regions obtained by filtering and dividing the resulting terrain data, Represents a grid cell The catchment area, is the watershed area threshold, Represents a grid cell The slope, Indicates the slope threshold.

[0088] For each spatial region, based on the terrain data, the main surface types in the spatial region are extracted, including bare soil, grass, concrete, asphalt, and masonry pavement, and the area proportion of each surface type in the spatial region is calculated. ;

[0089] For each surface type j , obtain the corresponding infiltration related parameters, including: saturated hydraulic conductivity , capillary suction , saturated volume moisture content , Volume moisture content in the previous time period ; Among them, saturated hydraulic conductivity Refers to surface type j The amount of water that can penetrate per unit time under fully saturated conditions, capillary suction Refers to the type of moisture on the surface j The suction force when adsorbed in the capillary pores of the soil, the saturated volume water content Refers to surface type j The maximum amount of water that a unit volume of soil can hold in a fully saturated state, the volumetric moisture content in the previous time period Refers to surface type j The actual soil moisture content at the previous moment; saturated hydraulic conductivity , capillary suction , saturated volume moisture content Obtained through geological data and soil database, volumetric moisture content in the previous time period Obtained through measured soil moisture sensor data;

[0090] Using the improved algorithm based on the Green-Ampt model, according to the infiltration related parameters of each spatial area, a rainwater infiltration capacity model of each spatial area with dynamic evolution over time is constructed to calculate the infiltration capacity of each spatial area at different times. t The instantaneous infiltration rate , calculated as follows:

[0091]

[0092] in, Indicates the i A spatial region in time t The instantaneous infiltration rate, indicating the i A spatial region in time t At time , the rate at which rainwater per unit area is absorbed by the surface of the area, Represents the time from the initial moment to the t , in i In the spatial region j The cumulative infiltration depth of the surface type is j The instantaneous infiltration rate of the surface type and whether there is precipitation are integrated over time. is the total number of surface types.

[0093] Because the target area is typically vast and has complex and diverse terrain, directly predicting flooding for the entire region requires detailed simulation of rainwater infiltration and hydrological processes for each spatial unit. This full-area, high-resolution modeling approach is extremely computationally intensive, consuming significant computing resources during model training and prediction, making real-time or rapid prediction difficult to achieve, limiting the practicality and responsiveness of the flood warning system. To address this technical challenge, step S2 of this embodiment first utilizes high-precision terrain data from a digital elevation model (DEM) to rationally divide the target area. By calculating the slope and aspect of each grid cell and combining this with analysis of water flow paths and catchment areas, spatial regions with significant hydrological impacts are identified, while regions with a low likelihood of flooding are eliminated. This partitioning strategy effectively reduces the spatial dimensionality of the model, minimizes redundant computations, and significantly reduces data processing scale and computational burden. Within each divided spatial region, the improved Green-Ampt infiltration model is used to dynamically calculate rainwater infiltration capacity, taking into account the surface cover type (e.g., bare soil, grassland, concrete), soil physical properties, and vegetation cover. The model comprehensively considers parameters such as saturated hydraulic conductivity, capillary suction, and volumetric moisture content for each surface type, and introduces the soil moisture status of the previous time period to achieve a dynamic response of the infiltration process to time and rainfall conditions. By weighting the area of ​​different surface types, it can accurately reflect the heterogeneity of rainwater infiltration characteristics within a spatial region. Through spatial partitioning based on terrain data, the number of spatial units that the model needs to process is significantly reduced, reducing the overall computational complexity. Secondly, the dynamic rainwater infiltration capacity model can capture changes in rainfall intensity and soil moisture status in real time, improving the timeliness and accuracy of the infiltration process. Finally, by comprehensively considering the heterogeneity of terrain, hydrology, and surface cover, the spatial resolution and prediction accuracy of the waterlogging prediction model are effectively improved. Overall, it solves the technical bottleneck of excessive computational complexity in waterlogging modeling within a large target area, improving prediction efficiency and application feasibility.

[0094] Step S3: Based on the surface temperature data and environmental data, an evaporation model reflecting the urban heat island effect is established to calculate the evaporation capacity of each spatial area;

[0095] Step S3 specifically includes:

[0096] Collect surface temperature data and environmental data of each spatial area in the target area. Environmental data include surface temperature, air temperature, wind speed, relative humidity and solar radiation;

[0097] Based on surface temperature data and environmental data, a modified Penman-Monteith model was used to construct an evaporation model that takes into account the urban heat island effect and calculate the potential evaporation rate of each spatial area in the future time period. , the formula is as follows:

[0098]

[0099] in, Indicates the i Spatial region in time t potential evaporation rate; Indicates the i The slope of the saturated vapor pressure curve for a spatial region, calculated based on the air temperature; Indicates the i Net radiation of a spatial region, estimated from solar radiation and surface temperature; Indicates the i Soil heat flux in spatial regions; is the humidity constant; Indicates the i the average air temperature of a spatial region; Indicates the i The wind speed in the space area at a height of 2 meters above the ground, 、 Respectively represent i Saturated vapor pressure and actual vapor pressure in a spatial region.

[0100] Step S3 accurately reflects the impact of the urban heat island effect on the evaporation process, thereby improving the evaporation prediction accuracy of the hydrological model. The urban heat island effect causes surface temperatures in urban areas to be generally higher than those in surrounding suburbs. This temperature difference significantly affects the evaporation rate and is directly related to the simulation of the rainwater cycle and soil moisture dynamics. The target area is spatially extensive and has complex surface types. Surface temperature and environmental conditions vary significantly across the land. Simply adopting a unified evaporation model cannot accurately reflect the evaporation characteristics of different spatial regions. Therefore, this embodiment collects high-resolution surface temperature data and environmental data (including air temperature, wind speed, relative humidity, and solar radiation) to achieve a detailed characterization of the environmental conditions in each spatial region. Based on this data, a dynamic evaporation model is established using a modified Penman-Monteith model, incorporating corrections for surface temperature and radiation due to the urban heat island effect. This model not only considers traditional meteorological factors but also incorporates the impact of the urban heat island effect on net radiation and soil heat flux, making the calculation of potential evaporation rates more realistic in urban environments. By accurately calculating the potential evaporation rate in each spatial region, the model can reflect the differences in evaporation under different terrain and cover types, improving the responsiveness of hydrological simulations to water evaporation processes. By combining detailed spatial zoning with multi-source environmental data, the spatial resolution and temporal dynamic response of the evaporation model are improved. Furthermore, the introduction of corrections for the urban heat island effect addresses the errors caused by traditional models that ignore high temperature effects in urban evaporation calculations. Overall, the accuracy of rainwater cycle simulations is enhanced, providing more scientific support for urban water resource management and waterlogging risk prediction.

[0101] Step S4: Generate a behavior heat map that changes over time using crowd trajectory data, and use the degree of crowd aggregation and abnormal behavior changes in the behavior heat map as behavior feedback factors;

[0102] Step S4 specifically includes:

[0103] Collecting crowd trajectory data in the target area , trajectory data is obtained by mobile signals, WiFi detection or video monitoring, and is used to represent the individual's t spatial location;

[0104] Map the personnel trajectory data to multiple spatial regions divided by the target area, and count the number of people in each spatial region at any time. t number of people inside ;

[0105] According to each spatial region at any time t number of people inside The area of ​​each spatial region Calculate the i A spatial region in timet Crowd density :

[0106]

[0107] Gaussian kernel function is used to calculate the crowd density Perform kernel density estimation and generate the moment t Behavior heat map:

[0108]

[0109] in, Indicates spatial location At the moment t The crowd behavior heat map value, Indicates the i The coordinates of the centroid of the spatial region, represents the total number of spatial regions, is the kernel function, is the bandwidth parameter.

[0110] The calculation process of behavioral feedback factor is:

[0111] Calculate the average density value of the historical behavior heat map , as a baseline reference; the historical period refers to the historical time period corresponding to the current time point in terms of time cycle.

[0112] Based on the average density value of the current moment behavior heat map and the historical behavior heat map of the same period , calculate the behavioral abnormality factor:

[0113]

[0114] in, Indicates spatial location At the moment t behavioral abnormalities;

[0115] Setting thresholds , when behavioral abnormality factor When , the crowd gathering intensity is extracted as the risk weighting factor:

[0116]

[0117] Risk-weighted factors Integrate and average by spatial region to obtain i Behavioral feedback factor for each spatial region:

[0118]

[0119] in, Indicates thei A spatial region in time t behavioral feedback factors.

[0120] The core purpose of calculating the behavioral feedback factor in step S4 of this embodiment is to indirectly reflect the potential risk of urban flooding and its scope of impact by dynamically monitoring the degree of crowd gathering and abnormal behavior within the target area. When urban flooding occurs, ground waterlogging, traffic blockages, and damaged public facilities often cause people to concentrate in limited safe areas or exhibit abnormal evacuation and retention behaviors. Traditional urban flooding risk assessments rely heavily on static meteorological, hydrological, and topographic data, making it difficult to timely reflect the complex dynamics of crowd responses triggered by urban flooding, thereby hindering the effective formulation and implementation of prevention and control measures. By collecting and analyzing real-time crowd trajectory data and calculating the behavioral feedback factor, it is possible to keenly capture abnormal crowd gathering or dispersion behaviors induced by urban flooding. For example, in areas with severe urban flooding, the behavioral feedback factor will exhibit abnormal peaks that deviate significantly from historical normal levels. This abnormal change not only reflects the crowd response but also serves as an important indicator of the occurrence and development of urban flooding. The introduction of the behavioral feedback factor overcomes the shortcomings of urban flooding impact assessments that rely solely on environmental and hydrological models, enabling dynamic coupled monitoring of the human-water-environment system. The calculation method of the behavioral feedback factor can accurately locate the crowd gathering hotspots and abnormal areas in high-risk areas of urban flooding through kernel density estimation and smoothing processing; through the comparison of historical data for the same period, the risk threshold is dynamically adjusted to reduce the false alarm rate and improve the timeliness and accuracy of detection. These feedback factors are integrated into the urban flooding risk assessment model as important behavioral input parameters to assist in judging the spatial distribution and severity of urban flooding events, and to improve the scientific nature and practicality of the overall risk warning. The behavioral feedback factors of this embodiment not only achieve accurate quantification of the dynamic behavior of the crowd, but more importantly, by reflecting the behavioral anomalies caused by urban flooding, they establish the intrinsic connection between crowd behavior and urban flooding risk, effectively solving the problem of insufficient dynamic response of traditional models to the impact of urban flooding, significantly enhancing the ability to identify, warn and respond to urban flooding risks, and improving the overall level and management efficiency of urban flood control and disaster reduction.

[0121] Step S5: Constructing a waterlogging prediction model by combining the rainwater infiltration capacity model, evaporation model, behavioral feedback factors, meteorological data, and drainage network data;

[0122] Step S5 specifically includes:

[0123] Get each spatial area i In time t Rainfall intensity , instantaneous infiltration rate , potential evaporation rate , drainage capacity , estimate each spatial region iSurface water accumulation index per unit area :

[0124]

[0125] in, Represents a spatial region i The concentration index of surface water per unit area is is the time interval, that is, the length of time from the previous time step to the current time step, 、 Represents spatial regions i rainfall intensity, instantaneous infiltration rate and potential evaporation rate, For the spatial region i Drainage capacity, according to each space area i Drainage network data acquisition; For the spatial region i The rainfall intensity is obtained based on meteorological data; Represents a spatial region i area;

[0126] Drainage capacity The drainage network data of each spatial area is calculated using the Manning formula, which is:

[0127]

[0128] in, For the spatial region i The number of drainage pipes, Indicates the j The hydraulic radius of the drainage pipe, Indicates the j The hydraulic slope of the drainage pipe, Indicates the j The flow cross-sectional area of ​​the drainage pipe, Indicates the j Manning's roughness coefficient of the drainage pipe.

[0129] The training samples are constructed by taking the characteristic information of each spatial region at each time step as input. The characteristic information includes the estimated surface water accumulation index, behavioral feedback factors and meteorological data, where the meteorological data includes the current rainfall intensity and the accumulated rainfall in the previous period.

[0130] Whether waterlogging has occurred in each spatial area is used as a supervised learning label. The label is determined based on historical waterlogging event data, which is obtained from remote sensing images and water level sensor monitoring data.

[0131] Construct a waterlogging prediction model based on a classification model to predict the probability of waterlogging. The classification model is a probabilistic model, including a gradient boosting decision tree model.

[0132] The waterlogging prediction model takes the characteristic information of each spatial area as input and outputs the predicted value of waterlogging probability at the corresponding moment. The waterlogging prediction model is trained through the loss function according to the output predicted value of waterlogging probability to obtain the trained waterlogging prediction model.

[0133] The loss function is:

[0134]

[0135] in, is the loss function, is the weight coefficient, To adjust the parameters, For the spatial region i behavioral feedback factor, Represents a spatial region i At the moment t The real label of waterlogging, 1 means waterlogging has occurred, 0 means no waterlogging has occurred, and the label is determined based on historical waterlogging event data; The output of the waterlogging prediction model is i A spatial region at time t The predicted value of the probability of waterlogging; represents the total number of spatial regions, Represents the number of time steps in the training sample.

[0136] The present invention calculates the surface water accumulation index (SWI) primarily to accurately reflect the dynamic changes in surface moisture within each spatial region over a specific timeframe, thereby providing a scientific basis for predicting urban flooding risk. Urban flooding essentially occurs when surface water exceeds the drainage system's carrying capacity and cannot be promptly removed, resulting in accumulated water that inundates the ground. Therefore, the accumulation of surface water is directly linked to the formation mechanism of urban flooding and is a key physical quantity for assessing urban flooding risk. Specifically, the SWI quantitatively expresses the net increase in surface water per unit area per unit time by comprehensively considering four key factors: rainfall input (rainfall intensity), infiltration loss (instantaneous infiltration rate), evaporation loss (potential evaporation rate), and drainage system removal capacity (drainage capacity). This index captures the inflow and outflow of rainfall water into the urban surface and can dynamically capture water surpluses and deficits. During urban flooding, if the SWI for a spatial region is consistently positive and high, it indicates that rainfall exceeds the combined infiltration capacity, evaporation loss, and drainage capacity of that region, and water begins to accumulate on the surface, a precursor to urban flooding. By real-time monitoring and calculation of this indicator, we can discover areas and time points of abnormal water accumulation at an early stage, and provide quantitative early warning signals for urban flooding risks from a physical mechanism perspective.

[0137] While the accumulation of surface water is indeed a direct physical mechanism for the formation of urban flooding, direct assessment of urban flooding based solely on surface water accumulation indicators has significant limitations. Urban flooding is influenced not only by the amount of surface water but also by a variety of complex factors, including the state of the drainage system, topography, sewer blockage, soil carrying capacity, and human activities. A single water quantity indicator cannot fully capture the interplay of these complex factors, making it difficult to accurately distinguish between urban flooding risks under different conditions. While the surface water accumulation indicator itself reflects the net accumulation of water per unit time, the specific manifestations and severity of urban flooding also depend on dynamic processes such as water flow convergence, drainage pathways, surface structure, and temporary water flow obstructions. Simple threshold assessments often overlook these nonlinearities and spatiotemporal complexities, leading to misjudgments or omissions.

[0138] The use of probabilistic prediction models can quantify urban flooding risks and provide prediction results for different risk levels, making it easier for urban managers to formulate differentiated emergency plans and preventive measures based on actual needs, rather than simply making a binary judgment of "occurrence" or "non-occurrence", thereby improving the scientific nature and practical value of urban flood prevention and control.

[0139] The behavioral feedback factor reflects significant differences between the spatial distribution and mobility patterns of people within a region and historical norms. When flooding occurs, surface water impedes normal traffic and movement, forcing people to change their planned routes, even leading to congestion, detours, or collective evacuation. This abnormal mobility pattern is not only a direct reflection of the occurrence of flooding but also a concrete manifestation of its impact on urban space use. Therefore, by analyzing abnormal mobility changes in crowd trajectory data, it is possible to indirectly identify spatial environmental changes and traffic disruptions caused by flooding. In other words, the behavioral feedback factor reveals the abnormal flow of people caused by flooding. This abnormal mobility pattern serves as a "social perception signal" of flooding, supplementing flooding predictions based solely on hydrometeorological and drainage system parameters, enabling the model to more comprehensively and accurately capture flooding risks, especially in rapidly changing and complex urban environments. This invention adds the behavioral feedback factor to the weight coefficient of the loss function. Since flooding events typically have a low probability of occurrence, the distribution of positive and negative samples is extremely unbalanced, which can easily bias the model towards predicting the "no flooding" category and overlook potentially high-risk areas. By incorporating behavioral feedback into the weights, the sample loss weights corresponding to areas with a high probability of flooding or unusual crowd movement are amplified, placing greater emphasis on the prediction accuracy of these key spatiotemporal points during model training. The adjustment parameter α in the weight coefficients flexibly controls the influence of the behavioral feedback factor on the loss function, ensuring the model's sensitive response to risk areas driven by unusual crowd behavior.

[0140] Step S6: Outputting the probability of waterlogging in each spatial area in the future period according to the waterlogging prediction model;

[0141] Step S6 specifically includes:

[0142] The model inputs current and historical characteristic information for each spatial region, including rainfall intensity, rainwater infiltration capacity, potential evaporation rate, drainage capacity, behavioral feedback factors, and relevant meteorological data. The model outputs a corresponding probability of urban flooding for each spatial region within a preset future time period. This probability reflects the likelihood of urban flooding in that region within a specific time window due to the combined effects of multiple factors, including rainfall, hydrological conditions, drainage status, and human behavior.

[0143] By analyzing the probability values ​​of consecutive time steps, dynamic monitoring and early warning of urban waterlogging risks can be achieved, supporting decision makers to take timely and targeted measures in urban management and disaster prevention and mitigation, and effectively reducing the losses caused by urban waterlogging.

[0144] This embodiment also provides a waterlogging prediction system based on behavioral heat map and infiltration dynamic modeling. Figure 2 As shown, it includes the following main modules:

[0145] The data acquisition module is used to obtain multi-source data of the target area in real time, including terrain data, soil and vegetation cover information, surface temperature and environmental meteorological data, crowd trajectory data, rainfall data, and drainage network information;

[0146] The spatial division and parameter extraction module divides the spatial regions into several areas based on the terrain data, extracts the surface cover type, soil properties and vegetation conditions of each area, and dynamically adjusts the rainwater infiltration capacity model based on the sensor data to construct an infiltration dynamic model that reflects the changes in rainfall and time.

[0147] The evaporation calculation module uses the modified Penman-Monteith model based on the collected surface temperature and environmental data to establish an evaporation model that takes into account the urban heat island effect and calculates the potential evaporation rate of each spatial area;

[0148] The behavior heat map generation module uses crowd trajectory data through kernel density estimation technology to generate a behavior heat map that changes over time, and then extracts behavior feedback factors to reflect abnormal crowd movement;

[0149] The waterlogging prediction model construction module integrates infiltration models, evaporation models, behavioral feedback factors, meteorological data, and drainage network information, and trains the waterlogging prediction model based on machine learning methods to achieve dynamic prediction of the probability of waterlogging in each spatial area;

[0150] The prediction result output module provides the probability of urban flooding risk in each spatial area in the future period based on the output of the urban flooding prediction model, supporting urban managers in early warning and emergency response.

[0151] Through multi-source data fusion and dynamic modeling, the system effectively improves the accuracy and timeliness of urban flooding predictions. Especially in complex urban environments, it can comprehensively reflect changes in hydrological processes and human behavior, thereby improving the ability to identify and warn of urban flooding risks.

[0152] If the above functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program code, such as USB flash drives, mobile hard drives, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0153] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and such modifications or substitutions are intended to be within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. A waterlogging prediction method based on behavioral heat map and infiltration dynamic modeling, characterized by: The following steps are involved: Collect multi-source data in the target area, including environmental data, meteorological data, terrain data, drainage network data, crowd trajectory data, and surface temperature data; Divide spatial regions based on terrain data, extract the surface cover type, soil properties, and vegetation coverage of each spatial region, and construct a rainwater infiltration capacity model that dynamically adjusts with time and rainfall conditions; Based on surface temperature data and environmental data, an evaporation model reflecting the urban heat island effect is established to calculate the evaporation capacity of each spatial area; Use crowd trajectory data to generate a behavior heat map that changes over time, and use the degree of crowd aggregation and abnormal behavior changes in the behavior heat map as behavior feedback factors; A waterlogging prediction model was constructed by combining rainwater infiltration capacity models, evaporation models, behavioral feedback factors, meteorological data, and drainage network data; Output the probability of waterlogging in each spatial area in the future period based on the waterlogging prediction model; The evaporation model reflecting the urban heat island effect is established based on surface temperature data and environmental data, specifically including: Collect surface temperature data and environmental data of each spatial area in the target area. Environmental data include surface temperature, air temperature, wind speed, relative humidity and solar radiation; Based on the surface temperature data and environmental data, a modified Penman-Monteith model was used to construct an evaporation model that takes into account the urban heat island effect, and the potential evaporation rate of each spatial area in the future time period was calculated. , the formula is as follows: in, Indicates the i Spatial region in time t potential evaporation rate; Indicates the i The slope of the saturated vapor pressure curve for a spatial region, calculated based on the air temperature; Indicates the i Net radiation of a spatial region, estimated from solar radiation and surface temperature; Indicates the i Soil heat flux in spatial regions; is the humidity constant; Indicates the i the average air temperature of a spatial region; Indicates the i The wind speed in the space area at a height of 2 meters above the ground, 、 Respectively represent i Saturated vapor pressure and actual vapor pressure of a spatial region; The method of generating a behavior heat map that changes over time by using crowd trajectory data specifically includes: Collecting crowd trajectory data in the target area The trajectory data is obtained by mobile signals, WiFi detection or video monitoring, and is used to represent the individual's time t spatial location; Map the personnel trajectory data to multiple spatial regions divided by the target area, and count the number of people in each spatial region at any time. t number of people inside ; According to each spatial region at any time t number of people inside The area of ​​each spatial region Calculate the i A region of space in time t Crowd density : Gaussian kernel function is used to calculate the crowd density Perform kernel density estimation and generate the moment t Behavior heat map: in, Indicates spatial location At the moment t The crowd behavior heat map value, Indicates the i The coordinates of the centroid of the spatial region, represents the total number of spatial regions, is the kernel function, is the bandwidth parameter.

2. The method for predicting waterlogging based on behavioral heat map and infiltration dynamic modeling according to claim 1 is characterized in that: The construction of a rainwater infiltration capacity model that is dynamically adjusted over time and rainfall conditions specifically includes: dividing the target area into a number of spatial regions according to the terrain data; For each spatial region, based on the terrain data, the main surface types in the spatial region are extracted, including bare soil, grass, concrete, asphalt, and masonry pavement, and the area proportion of each surface type in the spatial region is calculated. ; For each surface type j , obtain the corresponding infiltration related parameters, including: saturated hydraulic conductivity , capillary suction , saturated volume moisture content , Volume moisture content in the previous time period ; Among them, saturated hydraulic conductivity Refers to surface type j The amount of water that can penetrate per unit time under fully saturated conditions, capillary suction Refers to the type of moisture on the surface j The suction force when adsorbed in the capillary pores of the soil, the saturated volume water content Refers to surface type j The maximum amount of water that a unit volume of soil can hold in a fully saturated state, the volumetric moisture content in the previous time period Refers to surface type j The actual soil moisture content at the previous moment; saturated hydraulic conductivity , capillary suction , saturated volume moisture content Obtained through geological data and soil database, volumetric moisture content in the previous time period Obtained through measured soil moisture sensor data; Using the improved algorithm based on the Green-Ampt model, according to the infiltration related parameters of each spatial area, a rainwater infiltration capacity model of each spatial area with dynamic evolution over time is constructed to calculate the infiltration capacity of each spatial area at different times. t The instantaneous infiltration rate , calculated as follows: in, Indicates the i A region of space in time t The instantaneous infiltration rate, indicating the i A region of space in time t At time , the rate at which rainwater per unit area is absorbed by the surface of the area, Represents the time from the initial moment to the t , in i In the spatial region j The cumulative infiltration depth of the surface type is j The instantaneous infiltration rate of the surface type and whether there is precipitation are integrated over time. is the total number of surface types.

3. The method for predicting waterlogging based on behavioral heat map and infiltration dynamic modeling according to claim 2 is characterized in that: The target area is divided into several spatial areas according to the terrain data, specifically including: According to the digital elevation model in the target area terrain data, the slope and slope direction of each grid cell in the digital elevation model are calculated; Based on the slope and aspect, combined with the flow path analysis, the catchment area of ​​each grid cell is calculated; By setting the watershed area threshold and slope threshold, the spatial areas that meet the conditions are screened: in, The set of spatial regions obtained by filtering and dividing the resulting terrain data, Represents a grid cell The catchment area, is the watershed area threshold, Represents a grid cell The slope, Indicates the slope threshold.

4. The method for predicting waterlogging based on behavioral heat map and infiltration dynamic modeling according to claim 1, characterized in that: The behavior feedback factor calculation process is: Calculate the average density value of the historical behavior heat map , as a baseline reference; Based on the average density value of the current moment behavior heat map and the historical behavior heat map of the same period , calculate the behavioral abnormality factor: in, Indicates spatial location At the moment t behavioral abnormalities; Setting thresholds , when behavioral abnormality factor When , the crowd gathering intensity is extracted as the risk weighting factor: Risk-weighted factors Integrate and average by spatial area to obtain i Behavioral feedback factor for each spatial region: in, Indicates the i A spatial region in time t behavioral feedback factors.

5. The method for predicting waterlogging based on behavior heat map and infiltration dynamic modeling according to claim 4 is characterized in that: The historical period refers to the historical time period corresponding to the current time point in terms of time cycle.

6. The method for predicting waterlogging based on behavior heat map and infiltration dynamic modeling according to claim 1, characterized in that: The waterlogging prediction model is constructed by combining the rainwater infiltration capacity model, evaporation model, behavioral feedback factors, meteorological data and drainage network data, specifically including: Get each spatial area i In time t Rainfall intensity , instantaneous infiltration rate , potential evaporation rate , drainage capacity , estimate each spatial region i Surface water accumulation index per unit area : in, Represents a spatial region i The concentration index of surface water per unit area is is the time interval, that is, the length of time from the previous time step to the current time step, 、 Represents spatial regions i rainfall intensity, instantaneous infiltration rate and potential evaporation rate, For the spatial region i Drainage capacity, according to each space area i Drainage network data acquisition; For the spatial region i The rainfall intensity is obtained based on meteorological data; Represents a spatial region i area; The training samples are constructed using the characteristic information of each spatial region at each time step as input. The characteristic information includes the estimated surface water accumulation index, behavioral feedback factors and meteorological data, where the meteorological data includes the current rainfall intensity and the accumulated rainfall in the previous period. Whether waterlogging has occurred in each spatial area is used as a supervised learning label, and the label is determined based on historical waterlogging event data, and the historical waterlogging event data is obtained based on remote sensing images and water level sensor monitoring data; Constructing a waterlogging prediction model based on a classification model to predict the probability of waterlogging, wherein the classification model is a probabilistic model, including a gradient boosting decision tree model; The waterlogging prediction model takes the characteristic information of each spatial area as input, outputs a waterlogging probability prediction value at the corresponding moment, and trains the waterlogging prediction model through a loss function according to the output waterlogging probability prediction value to obtain a trained waterlogging prediction model.

7. The method for predicting waterlogging based on behavioral heat map and infiltration dynamic modeling according to claim 6, characterized in that: The drainage capacity The drainage network data of each spatial area is calculated using the Manning formula, which is: in, For the spatial region i The number of drainage pipes, Indicates the j The hydraulic radius of the drainage pipe, Indicates the j The hydraulic slope of the drainage pipe, Indicates the j The flow cross-sectional area of ​​the drainage pipe, Indicates the j Manning's roughness coefficient of the drainage pipe.

8. The method for predicting waterlogging based on behavior heat map and infiltration dynamic modeling according to claim 6, characterized in that: The loss function is: in, is the loss function, is the weight coefficient, To adjust the parameters, For the spatial region i behavioral feedback factor, Represents a spatial region i At the moment t The real label of waterlogging, 1 means waterlogging has occurred, 0 means no waterlogging has occurred, and the label is determined based on historical waterlogging event data; The output of the waterlogging prediction model is i A spatial region at time t The predicted value of the probability of waterlogging; represents the total number of spatial regions, Represents the number of time steps in the training sample.

Images