Sensor-based urban flood prediction system

Abstract

The present invention comprises a plurality of flood measurement units (110) for generating flood measurement data; a rainfall data collection unit (120) for collecting rainfall data including rainfall amount, cumulative rainfall amount, rainfall duration, and rainfall distribution by time period; a spatial information database unit (130) for storing geospatial information including a digital elevation model for an urban flood-predicted area, administrative district information, watershed boundary information, imperviousness rate, average elevation, relative elevation, slope, drainage facility information, and past flood damage history; a preprocessing unit (140) for generating prediction input data by performing missing value correction and outlier removal on flood measurement data and rainfall data; a flood prediction model unit (150) composed of an LSTM-based time series prediction model for predicting flood depth for multiple time points; a risk determination unit (160) for calculating a flood risk grade of a specific area using the predicted flood depth by the flood prediction model unit (150), rainfall data, and geospatial information of the spatial information database unit (130); and a flood impact range and A sensor-based urban flood prediction system is disclosed, which implements a real-time flood depth prediction and risk warning provision system for urban flood vulnerable points, including a flood information provision unit (170) that generates flood trace maps and risk warning information and provides them to an administrator terminal, a control system, or a field warning means (400).

Description

Technology Field

[0001] The present invention relates to a sensor-based urban flood prediction system that implements a system for providing real-time flood depth prediction and risk warnings for urban flood-vulnerable points. Background Technology

[0003] As is well known, urban flooding damage is recurring due to the increased frequency and intensity of torrential rains caused by climate change. In particular, localized sudden downpours, which concentrate on a small area within a short period, have become established as unpredictable meteorological phenomena. Furthermore, urban areas densely populated with underground structures such as underpasses, semi-basement homes, underground parking lots, and underground shopping malls are highly vulnerable to flooding, making the establishment of a rapid detection, prediction, and warning system urgently required.

[0004] However, conventional flood prediction technology has primarily relied on rainfall forecast data from the meteorological agency or numerical runoff analysis models.

[0005] This method has fundamental limitations, such as insufficient spatial and temporal resolution for localized rainfall and the inability to directly utilize measured water level data from flood-vulnerable points for training the prediction model. Furthermore, when rainfall occurs intensively at points far from the observation station, it is difficult to adequately reflect the flooding patterns at those locations using only rainfall information from a single station. Consequently, this leads to problems such as reduced accuracy in predicting maximum flood depth or discrepancies between the time of flooding occurrence and the time of prediction.

[0006] In addition, data collected from urban flood measurement sensors frequently contains outliers and missing values ​​caused by vehicle traffic, pedestrians, fog during periods without rainfall, condensation, and communication failures. If this data is input into a prediction model without proper processing, it significantly degrades learning performance and undermines the reliability of the prediction results. In particular, noisy missing values, such as water level values ​​being irregularly measured as 0 cm in flood measurement zones, are a major cause of hindering the model's time-series learning.

[0007] Furthermore, conventional systems adopt a simple step alarm method based on a single rainfall threshold, resulting in an excessively high rate of false alarms. Conversely, they also harbor inherent problems such as missing data, which prevents the dissemination of warnings when flood risks actually occur. Additionally, the absence of a composite standard method that simultaneously considers real-time observed rainfall and weather-predicted rainfall leads to limitations such as low usability of forecast information and reduced reliability of warnings.

[0008] Furthermore, the lack of a function to visualize the extent of flood impact or generate flood trace maps by linking predicted flood depths at flood sites with geospatial information such as Digital Elevation Models (DEMs) limits the ability to quickly assess the scale of flood damage and guide evacuation routes at disaster response sites. Prior art literature

[0010] Korean Published Patent Application No. 10-2023-0102492 Korean Registered Patent Application No. 10-2615686 The problem to be solved

[0011] The technical problem that the concept of the present invention aims to solve is to provide a sensor-based urban flood prediction system capable of enabling differentiated preemptive response measures for each prediction point by continuously predicting flood depth at multiple time points through the application of actual flood measurement data and rainfall data collected from multiple flood measurement units installed at urban flood-vulnerable points to a time-series prediction model. means of solving the problem

[0013] To achieve the aforementioned objective, an embodiment of the present invention comprises: a plurality of flood measurement units each installed at a plurality of urban flood-vulnerable points to measure the flood depth at the installation point during rainfall and generate flood measurement data; a rainfall data collection unit that collects rainfall data including rainfall amount, cumulative rainfall amount, rainfall duration, and rainfall distribution by time period from a weather station server or an external weather database; a spatial information database unit that stores geospatial information including a digital elevation model for an urban flood-predicted area, administrative district information, watershed boundary information, imperviousness ratio, average elevation, relative elevation, slope, drainage facility information, and history of past flood damage; a preprocessing unit that synchronizes the flood measurement data and the rainfall data based on the same time axis and generates prediction input data by performing missing value correction and outlier removal on the flood measurement data and the rainfall data; and a flood prediction model unit composed of an LSTM-based time series prediction model that receives the prediction input data and predicts the flood depth for multiple time points after a specific point in time. A sensor-based urban flood prediction system is provided, comprising: a risk determination unit that calculates a flood risk grade for a specific area using the predicted flood depth by the flood prediction model unit, the rainfall data, and the topographic spatial information of the spatial information database unit; and a flood information provision unit that generates a flood impact range, a flood trace map, and risk alert information by reflecting the predicted flood depth and the flood risk grade, and provides them to an administrator terminal, a control system, or an on-site alarm means.

[0015] Here, the flood measurement unit may include a flood depth sensor comprising at least one of an ultrasonic sensor, a radar sensor, a laser sensor, and a contact-type water level sensor; a control module that calculates the flood depth by analyzing a reference installation height and the flood measurement value of the flood depth sensor; a data logger that stores flood measurement data of the flood depth, flood measurement time, and sensor identification information from the flood depth sensor; an LTE Cat.M1-based wireless communication module that transmits the flood measurement data to the outside; and a power module composed of a battery and a solar charging module for power supply.

[0017] Additionally, the preprocessing unit may be composed of a time alignment module that converts the flood measurement data into time series data in units of preset time intervals, a missing value correction module that performs at least one of linear interpolation, moving average interpolation, or adjacent point-in-time reference interpolation on missing flood measurement values ​​or rainfall data, and an outlier processing module that removes or corrects outliers such as abnormal communication values, rapidly intermittent fluctuation values, or mismeasurement values ​​outside the reference range.

[0019] In addition, the flood prediction model unit comprises an LSTM model including at least one of a many-to-one structure that receives flood measurement data and rainfall data from a past time interval and outputs a flood depth at a specific future point in time, and a many-to-many structure that outputs a flood depth time series for multiple future points in time, and the LSTM model can calculate a predicted flood depth for at least one future point in time among 1 minute later, 10 minutes later, 30 minutes later, and 60 minutes later.

[0021] In addition, the above-mentioned risk determination unit refers to a risk standard database that stores a threshold rainfall amount, rainfall standard values ​​by duration, predicted flood depth standard values, correction values ​​based on past flood damage history, and composite standard threshold values ​​corresponding to each administrative district of the city or each installation point of the flood measurement unit, and if at least one of the following applies—where the rainfall data is greater than or equal to the threshold rainfall amount, where the predicted flood depth is greater than or equal to the predicted flood depth standard value, or where the rainfall data and the predicted flood depth simultaneously satisfy the composite standard threshold value—the flood risk is determined to be one of the flood risk grades of interest, caution, warning, and severe, and the standard values ​​of the above-mentioned risk standard database can be updated using the flood measurement data, CCTV-based flood verification data, and actual damage occurrence data.

[0023] Additionally, the flood information providing unit may include an impact range calculation module that calculates the flood impact range using the predicted flood depth corresponding to the installation points of the plurality of flood measurement units and the digital elevation model stored in the spatial information database unit; a flood trace map generation module that visualizes the distribution of flood range and flood depth on a map by combining the digital elevation model grid of the spatial information database unit and the flood impact range; and a risk notification generation module that generates risk notification information including a flood risk grade calculated by the risk judgment unit, a predicted time, a predicted flood depth, a target area, and a warning message. Effects of the invention

[0025] According to the present invention, actual flood measurement data and rainfall data collected from multiple flood measurement units installed at urban flood-vulnerable points are applied to an LSTM-based time series prediction model to continuously predict flood depth at multiple time points, thereby enabling differentiated preemptive response measures for each prediction time point before flood damage intensifies; flood risk grades are determined based on a multi-factor composite risk index combining rainfall intensity, predicted flood depth, topographic wetness index, and antecedent rainfall index, as well as region-specific weighting coefficients; and a risk standard database is periodically updated using actual data and CCTV verification data to overcome the limitations of a single threshold method and implement a dynamic risk assessment system that adaptively responds to climate change and changes in the urban development environment. Furthermore, by integrally performing the calculation of flood impact range linked with a digital elevation model, time series visualization of flood trace maps at multiple future time points, calculation of dynamic safe evacuation routes based on road network geographic information, and linkage of barriers, drainage pumps, etc. through control trigger signals, a flood response system can be established without manual intervention by an administrator. Brief explanation of the drawing

[0027] Figure 1 illustrates a configuration diagram of a sensor-based urban flood prediction system according to an embodiment of the present invention. Figure 2 illustrates the flood measurement section of the sensor-based urban flood prediction system of Figure 1. Figure 3 illustrates an exploded view of the flood measurement unit of Figure 2. Figure 4 illustrates the configuration of the preprocessing unit of the sensor-based urban flood prediction system of Figure 1. Figure 5 illustrates the configuration of the flood information providing unit of the sensor-based urban flood prediction system of Figure 1. Figure 6 illustrates a flood trace map provided by the flood information provision unit of Figure 5. Specific details for implementing the invention

[0028] Hereinafter, embodiments of the present invention having the aforementioned features will be described in more detail with reference to the attached drawings.

[0030] A sensor-based urban flood prediction system according to an embodiment of the present invention comprises: a plurality of flood measurement units (110) each installed at a plurality of urban flood-vulnerable points and generating flood measurement data by measuring the flood depth at the installation point during rainfall; a rainfall data collection unit (120) that collects rainfall data including rainfall amount, cumulative rainfall amount, rainfall duration, and rainfall distribution by time period from a weather station server (200) or an external weather database (300); a spatial information database unit (130) that stores geospatial information including a digital elevation model for an urban flood-predicted area, administrative district information, watershed boundary information, imperviousness rate, average elevation, relative elevation, slope, drainage facility information, and past flood damage history; a preprocessing unit (140) that synchronizes flood measurement data and rainfall data based on the same time axis and generates prediction input data by performing missing value correction and outlier removal on flood measurement data and rainfall data; and an LSTM-based time series prediction model that receives prediction input data and predicts flood depth for multiple time points after a specific time point. The gist of the invention is to implement a real-time flood depth prediction and risk warning provision system for urban flood-vulnerable points by including a flood prediction model unit (150), a risk determination unit (160) that calculates a flood risk grade of a specific area using the predicted flood depth by the flood prediction model unit (150), rainfall data, and topographic spatial information of the spatial information database unit (130), and a flood information provision unit (170) that generates a flood impact range, a flood trace map, and risk warning information by reflecting the predicted flood depth and flood risk grade and provides them to an administrator terminal, a control system, or a field warning means (400).

[0032] Hereinafter, with reference to FIGS. 1 to 6, the sensor-based urban flood prediction system of the above-described configuration will be specifically described as follows.

[0034] First, the flood measurement unit (110) is composed of multiple units and installed at multiple urban flood-vulnerable points, respectively, to measure the flood depth at the installation point during rainfall, generate flood measurement data, and transmit it to the preprocessing unit (140).

[0035] That is, the flood measurement unit (110) is installed on the surface of a point with a history of flood damage or a high risk of flooding, such as an underground roadway, a semi-underground residential area, an underground parking lot entrance, a low-lying intersection, or a residential area for vulnerable groups within the city, so as to measure the flood depth of the point in real time when rainfall occurs and generate flood measurement data.

[0036] Here, the flood measurement unit (110) may include a flood depth sensor comprising at least one of an ultrasonic sensor, a radar sensor, a laser sensor, and a contact water level sensor, a control module (113) that calculates the flood depth by analyzing the flood measurement value of the reference installation height and the flood depth sensor, a data logger (not shown) that stores flood measurement data of the flood depth, flood measurement time, and sensor identification information from the flood depth sensor, an LTE Cat.M1-based wireless communication module (not shown) that transmits the flood measurement data to the outside, and a power module composed of a battery and a solar charging module (114) for power supply.

[0037] For example, as illustrated in FIGS. 2 and 3, the immersion depth sensor is composed of a first sensor module (111a) that measures the immersion depth in a non-contact manner by being placed in the internal space of the main body (112), and a second sensor module (111b) that measures the immersion depth in a contact manner by being formed extending in the height direction on the outer side of the main body (112). The first sensor module (111a) is placed in the internal space of the main body (112) to protect the core sensing element from physical impacts such as debris, interference by birds, heavy rain, and direct impact from muddy water. The control module (113) allows the first sensor module (111a) and the second sensor module (111b) to cross-measure the immersion depth independently of each other, so that even if temporary noise occurs in the first sensor module (111a) due to weather conditions such as severe turbulence on the water surface, large floating objects, or strong winds, the data of the second sensor module (111b) that directly detects physical water level changes By referencing and algorithmically correcting, it is possible to prevent measurement errors and produce high-precision measurement data.

[0038] Here, the lower part of the first sensor module (111a) may further be equipped with a self-cleaning and zero-point diagnostic unit to maintain the accuracy of non-contact measurement, and a heating coating layer is formed on the surface of the ultrasonic or laser transmitting and receiving lens of the first sensor module (111a) to prevent moisture and contaminants from adhering, and the control module (113) can perform a self-cleaning operation by applying a microcurrent to the heating coating layer to evaporate moisture on the lens surface when conditions for condensation to occur due to a sudden temperature difference between night and day are predicted or when the noise of the received signal increases above a preset threshold.

[0039] Additionally, the second sensor module (111b) may be composed of a vertical support extending in the vertical direction, a plurality of unit electrodes arranged at regular intervals along the vertical support, and a common electrode arranged in a continuous manner, so that when water comes into simultaneous contact with the unit electrodes and the common electrode due to a rise in water level, a closed circuit is formed, and the submersion depth at the corresponding height can be physically measured, thus forming a multi-point electrode structure.

[0040] Additionally, the main body (112) may be composed of a first main body plate (112a) which is formed upright so that a solar charging module (114) is coupled to the top and a first sensor module (111a) is disposed in the internal space and has an open bottom so that water flows when the water level rises, and a second main body plate (112b) which is coupled to one side of the first main body plate (112a) to protect the second sensor module (111b).

[0041] Additionally, the flood measurement unit (110) periodically diagnoses the connection status of the main communication network, the LTE Cat.M1 communication network, and if a communication failure is detected, it activates a backup communication module based on a built-in low-power long-range communication network or a short-range wireless communication network to perform multi-hop relay communication that bypasses the measurement data to an adjacent flood measurement unit or gateway. The control module (113) can dynamically adjust the measurement and transmission cycle of data by comprehensively analyzing the remaining battery level and the future rainfall forecast received from the weather station server (200). For example, it may operate in an ultra-low power sleep mode that minimizes the measurement cycle when there is no rainfall or when the remaining battery level is below the danger level, and switch to an intensive monitoring mode that shortens the measurement and transmission cycle to minutes by intensively consuming the remaining battery regardless of whether solar charging is possible when a risk level higher than the warning level is calculated by the risk judgment unit (160) or when heavy rainfall begins.

[0044] Next, the rainfall data collection unit (120), referring to FIG. 1, collects rainfall data including rainfall amount, cumulative rainfall amount, rainfall duration, and rainfall distribution by time period from a weather station server (200) or an external weather database (300) and transmits it to the preprocessing unit (140).

[0045] For example, the rainfall data collection unit (120) may selectively apply a real-time collection method and a periodic batch collection method, and when a rainfall event occurs, it switches to a real-time collection mode to shorten the collection cycle and ensure the timeliness of the data, and by considering the spatial distribution of rainfall data collected from multiple weather station servers (200), it performs interpolation processing between observation points to estimate rainfall data for areas with gaps in observation, thereby increasing the spatial completeness of the collected data, and primarily detects missing values ​​and outliers in the rainfall data, and for the detected missing values, it maintains data continuity by utilizing replacement values ​​from adjacent observation stations or external weather databases (300), and for outliers, it assigns a flag to the data and transmits it to the preprocessing unit (140) so that reliability-based weight adjustment can be performed in the subsequent processing stage.

[0046] Meanwhile, the rainfall data collection unit (120) structures and stores the collected rainfall data as metadata by linking it with rainfall type information such as observation point identifier, observation time, convective rainfall or orographic rainfall, thereby supporting a detailed analysis based on rainfall characteristics during slope stability analysis by the risk judgment unit (160).

[0047] For example, since convective rainfall has the characteristic of occurring intensively within a short period of time, the risk assessment unit (160) applies rainfall intensity per unit time as a priority analysis factor when the rainfall type is recorded in metadata to intensively evaluate the possibility of a rapid increase in pore water pressure within the slope, and since orographic rainfall has the characteristic of occurring continuously over a long period of time, the risk assessment unit (160) applies cumulative rainfall amount and rainfall duration as major analysis factors to evaluate whether stability is reduced due to changes in infiltration depth and saturation within the slope soil layer.

[0048] Additionally, the rainfall data collection unit (120) accumulates and stores time-series rainfall data for the same observation point in conjunction with metadata, thereby providing a data basis for the risk judgment unit (160) to calculate an Antecedent Rainfall Index (ARI) that reflects the history of preceding rainfall. This allows the risk level to be adjusted upward if the moisture content in the soil layer has already increased due to the influence of preceding rainfall, even if the rainfall is judged to be stable at the beginning of the rainfall. Furthermore, the metadata structured by rainfall type is utilized when applying the threshold rainfall amount standard value of the risk judgment unit (160), allowing for the application of a threshold value differentiated according to the rainfall type to further increase the accuracy of the slope collapse risk judgment.

[0050] Next, the spatial information database section (130), referring to FIG. 1, stores geospatial information including a digital elevation model (DEM) for an area expected to be flooded in the city, administrative district information, basin boundary information, imperviousness rate, average elevation, relative elevation, slope, drainage facility information, and history of past flood damage.

[0051] For example, the spatial information database section (130) stores and manages geospatial information including a digital elevation model of a 5m or 10m grid collected from the National Geographic Information Institute for urban flooded areas, administrative district information, watershed boundary information, imperviousness rate calculated from a land cover map, average elevation by administrative district and relative elevation compared to adjacent administrative districts, slope calculated from the digital elevation model, drainage facility information including rainwater catchment density, and flood damage history analyzed from past flood damage reports.

[0052] In addition, the spatial information database unit (130) is configured to store six basin characteristic values, such as basin slope, elevation, relative elevation, imperviousness ratio, catch basin density, and river length density, as basin characteristics that affect flooding, at the administrative district level. The basin slope, elevation, and relative elevation are analyzed using a digital elevation model, the imperviousness ratio is calculated from the land cover map provided by the environmental spatial information service, the catch basin density is calculated using underground facility data built by the local government, and the topographic spatial information can be updated periodically according to the update cycle.

[0053] That is, the spatial information database unit (130) can provide topographical context to the results of the deep learning-based flood prediction model unit (150) and combine watershed boundary and drainage facility information with the predicted flood depth data to support three-dimensional simulation of how water spreads into the actual urban area and where damage will be concentrated.

[0054] In addition, such topographic spatial information is provided as basic data for calculating customized limit rainfall amounts for each administrative district or drainage district in the risk judgment unit (160). That is, the risk judgment unit (160) can make it possible to determine a subdivided flood risk grade according to topographic conditions even if the same amount of rain falls by comprehensively linking and analyzing the history of past flood damage, the limit capacity of drainage facilities, and topographical vulnerability of relative elevation and slope.

[0055] In addition, local measurement data at specific point units collected by the flood measurement unit (110) is expanded into area unit information of the entire basin, and in particular, topographic characteristic data such as a digital elevation model and impermeability and slope allows for algorithmic tracking of the surface runoff and collection path of rainwater, thereby enabling the flood information providing unit (170) to visualize a precise flood trace map in a 2D or 3D form that reflects the curvature of the terrain.

[0057] Next, the preprocessing unit (140), referring to FIGS. 1 and FIGS. 4, synchronizes flood measurement data from the flood measurement unit (110) and rainfall data from the rainfall data collection unit (120) based on the same time axis, and performs missing value correction and outlier removal on the flood measurement data and rainfall data to generate predicted input data.

[0058] That is, the preprocessing unit (140) may be composed of a time alignment module (141) that converts flood measurement data into time series data in units of preset time intervals, a missing value correction module (142) that performs at least one of linear interpolation, moving average interpolation, or adjacent time point reference interpolation on missing flood measurement values ​​or rainfall data, and an outlier processing module (143) that removes or corrects outliers such as abnormal communication values, sudden intermittent fluctuation values, or mismeasurement values ​​outside the reference range.

[0059] Here, the missing value correction module (142) applies spline interpolation to the noisy missing values ​​that are irregularly measured as 0cm in the section where the flooding is measured, thereby interpolating into a natural water level rise and fall curve. Spline interpolation is a method of dividing the entire section into sub-sections and obtaining a function using a low-order polynomial, and when the data is discontinuous or irregular, it can produce a smoother and more precise curve shape than linear interpolation, making it suitable for interpolating missing values ​​of flooding depth.

[0060] Additionally, the outlier processing module (143) can identify data as outliers in which the flood depth is measured rapidly when the cumulative rainfall is very low, or the flood depth is calculated to be excessively high compared to the cumulative rainfall over time, thereby removing the corresponding rainfall-flooding event, and filter out low flood depth values ​​that are measured irregularly due to high humidity or fog occurring inside the main body (112) due to the temperature difference between day and night when there is no rainfall during night and dawn hours.

[0062] Next, the flood prediction model unit (150), referring to FIG. 1, is composed of a Long Short-Term Memory (LSTM) based time series prediction model that receives prediction input data from the preprocessing unit (140) and predicts flood depth for multiple time points after a specific time point.

[0063] That is, the flood prediction model unit (150) can continuously predict the trend of temporal change in flood depth over multiple time points, enabling preemptive response before flood damage becomes full-scale. In particular, the LSTM-based time series prediction model can simultaneously learn the long-term and short-term dependencies of time series input data including rainfall amount, cumulative rainfall amount, rainfall duration, and rainfall distribution by time period, thereby enabling high accuracy in predicting fluctuations in flood depth caused by rapid changes in rainfall patterns or continuous accumulation. The predicted flood depth values ​​for multiple time points are transmitted to the risk judgment unit (160) and used to calculate the risk level for each time point, thereby providing a decision-making basis that allows a manager or relevant agency to sequentially execute step-by-step response measures such as issuing evacuation orders, operating drainage pumps, and controlling vehicles with a margin of time.

[0064] For example, the flood prediction model unit (150) is provided with an LSTM model that includes at least one of a many-to-one structure that receives flood measurement data and rainfall data from a past time interval and outputs a flood depth at a specific future point in time, and a many-to-many structure that outputs a flood depth time series for multiple future points in time, and the LSTM model can calculate a predicted flood depth for at least one future point in time among 1 minute later, 10 minutes later, 30 minutes later, and 60 minutes later.

[0065] Here, the many-to-one structure can maximize the accuracy of flood depth prediction for a specific single point in time and can be utilized for intensively predicting the flood depth at an imminent risk point in the initial stage of flood occurrence, and the many-to-many structure can continuously output the trend of flood depth change for multiple future points in time to generate a flood depth time series profile including the rising section, the peak arrival point, and the falling section of the flood, so that the risk judgment unit (160) can be utilized to comprehensively evaluate the flood development pattern over time.

[0066] In addition, the predicted value after 1 minute calculated by the LSTM model is used to determine the immediate issuance of an alarm, the predicted values ​​after 10 minutes and 30 minutes are used to determine the deployment of on-site response personnel and the operation of drainage facilities, and the predicted value after 60 minutes is used to establish mid-to-long-term response measures such as securing evacuation routes and notifying relevant agencies, thereby enabling the implementation of a differentiated response system for each prediction time point. The flood prediction model unit (150) calculates the error between the predicted value and the actual value whenever additional actual flood depth data is collected, and applies an online learning method that gradually updates the weights of the LSTM model based on the calculated error, thereby adaptively responding to changes in terrain characteristics and rainfall patterns at each observation point and continuously improving the prediction accuracy.

[0067] Specifically, the LSTM model consists of 3 hidden layers and 64 nodes to learn patterns in long-term time series data, and compared to XGBoost or Random Forest models, it improves the performance of predicting maximum flood depth, enabling accurate prediction of high-level flood events of 30 cm or more. Additionally, by accumulating flood measurement data for each rainfall event and periodically retraining it, it can be updated to the latest LSTM model that reflects changes in flood characteristics and rainfall patterns at each location.

[0069] Next, the risk judgment unit (160), referring to FIG. 1, calculates the flood risk level of a specific area using the predicted flood depth by the flood prediction model unit (150), rainfall data, and topographic spatial information of the spatial information database unit (130).

[0070] Specifically, the risk determination unit (160) refers to a risk standard database that stores a limit rainfall amount, a rainfall standard value for duration corresponding to each administrative district of the city or each installation point of the flood measurement unit (110), a predicted flood depth standard value, a correction value based on past flood damage history, and a composite standard threshold value, and if at least one of the following applies: rainfall data is greater than or equal to the limit rainfall amount, predicted flood depth is greater than or equal to the predicted flood depth standard value, or rainfall data and predicted flood depth simultaneously satisfy the composite standard threshold value, the flood risk is determined to be one of the flood risk grades of interest, caution, warning, and severe, and can update the standard value of the risk standard database using flood measurement data, CCTV-based flood verification data, and actual damage occurrence data.

[0071] For example, the risk judgment unit (160) determines the flood risk level as one of the flood risk grades of interest, caution, warning, and severe if at least one of the following applies: when the rainfall data is greater than or equal to the limit rainfall amount, when the predicted flood depth is greater than or equal to the predicted flood depth threshold value, or when the rainfall data and the predicted flood depth simultaneously satisfy the composite threshold value. The composite standard method combines real-time observed rainfall data with predicted rainfall information from the Korea Meteorological Administration's ultra-short-term forecast to predict whether the limit rainfall amount is reached and reflects this in the risk grade determination, thereby minimizing the high false positive rate when only observed rainfall is used and improving the reliability of the risk warning.

[0072] In addition, the risk judgment unit (160) may periodically update the reference values ​​of the risk standard database using flood measurement data, CCTV-based flood verification data, and actual damage occurrence data, thereby continuously reflecting environmental changes such as changes in imperviousness due to urban development and improvements in drainage facilities into the risk standard, so as to continuously increase the prediction accuracy and reliability by synchronizing the changes in localized heavy rainfall patterns due to climate change and changes in rainfall-runoff characteristics due to changes in topography and infrastructure with the risk standard in real time.

[0073] For example, the risk judgment unit (160) may be equipped with an advanced judgment algorithm that determines risk by synthesizing multidimensional data, and may calculate a Multi-factor Composite Risk Index (MCRI) by utilizing rainfall data, predicted flood depth, and geospatial information as input variables, and determine the final flood risk grade by comparing the composite risk index with a threshold interval for each grade that is pre-set in a risk standard database.

[0074] Here, the multi-factor composite risk index (MCRI) can be calculated by the following [Equation 1].

[0075]

[0076] Here, RI (Rainfall Intensity) is the rainfall intensity per unit time derived from rainfall data, PFD (Predicted Flood Depth) is the predicted flood depth calculated by the flood prediction model unit (150), TWI (Topographic Wetness Index) is the topographic wetness index calculated based on topographic spatial information of the spatial information database unit (130), ARI (Antecedent Rainfall Index) is the antecedent rainfall index reflecting the prior saturation of the target soil, and α, β, γ, and δ are weighting coefficients reflecting the relative contribution of each factor to the flood risk, which can be determined regionally through multiple regression analysis using past flood damage history data by target administrative district or sensor installation point.

[0077] Additionally, the risk judgment unit (160) includes a weighted judgment logic to prevent underestimation errors in complex extreme weather event situations. When an extreme exceedance situation is detected where rainfall data and predicted flood depth simultaneously satisfy a complex standard threshold, the risk judgment unit (160) may apply a pre-set Risk Amplification Factor to the calculated MCRI value or raise the judgment grade by at least one step to preemptively reflect the risk of complex disaster situations that are difficult to identify using a single factor standard method.

[0079] Meanwhile, the risk judgment unit (160) may further include a real-time drainage capacity reduction estimation logic that recognizes sudden blockages in the rainwater catch basin and drainage network occurring at the actual site and reflects this in the judgment. Specifically, the risk judgment unit (160) continuously monitors the error between the predicted flood depth under normal drainage conditions calculated by the flood prediction model unit (150) and the actual flood depth collected in real-time by the flood measurement unit (110). When a pattern is detected where the actual flood depth exceeds the predicted flood depth by more than a preset error tolerance range and the water level rises rapidly, the risk judgment unit (160) estimates that the rainwater catch basin or sewer pipe at the relevant point is physically blocked by fallen leaves or sediment. The risk judgment unit (160) temporarily lowers the drainage facility information parameter among the topographic spatial information of the relevant point stored in the spatial information database unit (130) to 0 or a minimum value, and additionally assigns a poor drainage weight when calculating the multi-factor composite risk index to the risk grade. By raising the height, it is possible to prevent misjudgment of flooding by reflecting in the prediction model in real time sudden drainage paralysis situations at the site, which cannot be known solely from topographic information on the drawings.

[0081] Next, the flood information providing unit (170), referring to FIG. 1, generates flood impact range, flood trace map, and risk notification information by reflecting the predicted flood depth and flood risk grade, and provides them to the warning means (400) of the manager terminal, control system, or field alarm means.

[0082] Specifically, the flood information providing unit (170) may include an impact range calculation module (171) that calculates the flood impact range using a predicted flood depth corresponding to the installation point of a plurality of flood measurement units (110) and a digital elevation model stored in the spatial information database unit (130), a flood trace map generation module (172) that visualizes a flood trace map (see FIG. 6) of the flood range and flood depth distribution on a map by combining the digital elevation model grid of the spatial information database unit (130) and the flood impact range, and a risk notification generation module (173) that generates risk notification information including a flood risk grade calculated by the risk judgment unit (160), a predicted time, a predicted flood depth, a target area, and a warning message.

[0083] For example, the flood trace map generation module (172) can display the process of the flood range gradually expanding by reflecting the predicted flood depth at multiple future time points, such as 10 minutes, 30 minutes, and 60 minutes later, derived from the flood prediction model unit (150), in the form of a time-series animation or multiple overlapping layers, and can further include logic to calculate a dynamic safe evacuation route that bypasses the flooded area where the water level rises over time in real time by linking the calculated predicted flood trace map with the road network geographic information of the spatial information database unit (130), thereby supporting disaster control personnel and citizens to intuitively understand the direction of flooding progression and evacuate preemptively.

[0084] In addition, the danger alert generation module (173) can generate a control trigger signal in conjunction with danger alert information to immediately activate an underpass entry barrier, a road electronic display, an audible alarm device, an automatic drainage pump, etc., without manual intervention by a manager.

[0085] Accordingly, the flood information provision unit (170) can perform the calculation of the flood impact range by combining the predicted flood depth and the digital elevation model, the time-series visualization of flood trace maps at multiple future points in time, and the calculation of a dynamic safe evacuation route linked with road network geographic information, thereby providing a spatial and temporal decision-making basis that allows disaster control personnel and citizens to intuitively grasp the direction of flood progress and the extent of spread, and to evacuate preemptively before flood damage becomes full-scale, thereby minimizing human and property damage caused by delayed response and linking the flood response system.

[0087] Meanwhile, it may further include a video flood verification module (not shown) that receives real-time video data from a CCTV installed at or near the installation point of the flood measurement unit (110) and independently verifies whether there is flooding and the depth of flooding based on video analysis.

[0088] Specifically, the video flood verification module analyzes the flood coverage area of ​​reference structures such as road surfaces, curbs, and drains from CCTV video using a deep learning-based semantic segmentation algorithm to estimate the video-based flood depth, and compares the estimated video-based flood depth with the sensor measurement value of the flood measurement unit (110). If the deviation between the two exceeds a preset allowable range, it assigns a reliability reduction flag to the flood measurement data and transmits it to the preprocessing unit (140) so that a secondary correction processing is performed by the outlier processing module (143). The actual occurrence of flood and the flood depth confirmed through video analysis are provided as CCTV-based flood verification data to update the reference value of the risk standard database of the risk judgment unit (160), thereby improving the accuracy and reliability of the risk standard data compared to the sensor-only measurement method.

[0090] Therefore, through the configuration of the sensor-based urban flood prediction system described above, actual flood measurement data and rainfall data collected from multiple flood measurement units installed at urban flood-vulnerable points are applied to an LSTM-based time-series prediction model to continuously predict flood depths at multiple time points, enabling differentiated preemptive response measures for each prediction point before flood damage intensifies. Furthermore, flood risk grades are determined based on a multi-factor composite risk index combining rainfall intensity, predicted flood depth, topographic wetness index, and antecedent rainfall index, as well as region-specific weighting coefficients. By periodically updating the risk criteria database using actual data and CCTV verification data, the limitations of the single threshold method are overcome, and a dynamic risk assessment system that adaptively responds to climate change and changes in the urban development environment can be implemented. Additionally, by integrally performing the calculation of flood impact ranges linked with a digital elevation model, time-series visualization of flood trace maps at multiple future time points, calculation of dynamic safe evacuation routes based on road network geographic information, and the integration of barriers, drainage pumps, etc., via control trigger signals, a flood response system can be established without manual intervention by an administrator. It can be built.

[0092] The embodiments described in this specification and the configurations illustrated in the drawings are merely the most preferred embodiments of the present invention and do not represent all of the technical ideas of the present invention; therefore, it should be understood that various equivalents and modifications that can replace them may exist at the time of filing this application. Explanation of the symbols

[0094] 110: Submersion measurement unit 111a: First sensor module 111b : Second sensor module 112 : Main body 112a : 1st main body plate 112b : 2nd main body plate 113: Control module 114: Solar charging module 120: Rainfall Data Collection Unit 130: Spatial Information Database Unit 140: Preprocessor 141: Time Alignment Module 142: Missing Value Correction Module 143: Outlier Handling Module 150: Flood Prediction Model Section 160: Risk Assessment Section 170: Flood Information Provision Unit 171: Impact Range Calculation Module 172: Flood Trace Map Generation Module 173: Risk Alert Generation Module 200: Weather station server 300: External weather database 400 : Warning means

Claims

Claim 1 A plurality of flood measurement units, each installed at multiple urban flood-vulnerable points to measure the flood depth at the installation point during rainfall and generate flood measurement data; a rainfall data collection unit that collects rainfall data including rainfall amount, cumulative rainfall amount, rainfall duration, and rainfall distribution by time period from a weather station server or an external weather database; a spatial information database unit that stores geospatial information including a digital elevation model for urban flood-predicted areas, administrative district information, watershed boundary information, imperviousness ratio, average elevation, relative elevation, slope, drainage facility information, and history of past flood damage; a preprocessing unit that synchronizes the flood measurement data and the rainfall data based on the same time axis and generates prediction input data by performing missing value correction and outlier removal on the flood measurement data and the rainfall data; a flood prediction model unit composed of an LSTM-based time series prediction model that receives the prediction input data and predicts flood depth for multiple time points after a specific point in time; and the predicted flood depth by the flood prediction model unit, the rainfall data, and the spatial information database unit A risk determination unit that calculates a flood risk grade for a specific area using geospatial information; and a flood information provision unit that generates a flood impact range, a flood trace map, and risk alert information by reflecting the predicted flood depth and the flood risk grade, and provides them to an administrator terminal, a control system, or a field alarm means; wherein the flood measurement unit includes a flood depth sensor comprising an ultrasonic sensor, a radar sensor, a laser sensor, and a contact-type water level sensor; a control module that calculates the flood depth by analyzing a reference installation height and the flood measurement value of the flood depth sensor; a data logger that stores flood measurement data including the flood depth, flood measurement time, and sensor identification information from the flood depth sensor; an LTE Cat.M1-based wireless communication module that transmits the flood measurement data externally; and a power module composed of a battery and a solar charging module for power supply, wherein the flood depth sensor is composed of a first sensor module that measures the flood depth in a non-contact manner and a second sensor module that measures the flood depth in a contact manner.The control module above enables the first sensor module and the second sensor module to cross-measure flood depth independently of each other, and further comprises a self-cleaning and zero-point diagnostic unit at the bottom of the first sensor module to maintain the accuracy of non-contact measurement, and a heating coating layer is formed on the surface of the ultrasonic or laser transmitting / receiving lens of the first sensor module to prevent the attachment of moisture and contaminants, and the preprocessing unit is composed of a time alignment module that converts the flood measurement data into time-series data in units of preset time intervals, a missing value correction module that performs at least one of linear interpolation, moving average interpolation, or adjacent point-in-time reference interpolation for missing flood measurement values ​​or rainfall data, and an outlier processing module that removes or corrects outliers such as abnormal communication values, rapidly intermittent fluctuation values, or mismeasurement values ​​outside the reference range, and the risk determination unit is based on the threshold rainfall amount corresponding to each administrative district of the city or the installation point of the flood measurement unit, rainfall reference values ​​by duration, predicted flood depth reference values, and past flood damage history Referring to a risk standard database storing correction values ​​and composite standard thresholds, if at least one of the following applies—where the rainfall data is greater than or equal to the limit rainfall amount, where the predicted flood depth is greater than or equal to the predicted flood depth standard value, or where the rainfall data and the predicted flood depth simultaneously satisfy the composite standard threshold—the flood risk is determined to be one of the flood risk grades of Interest, Caution, Warning, and Severe, and the standard values ​​of the risk standard database are updated using the flood measurement data, CCTV-based flood verification data, and actual damage occurrence data; the flood measurement unit periodically diagnoses the connection status of the LTE Cat.M1 communication network, and if a communication failure is detected, activates a backup communication module based on a built-in low-power long-range communication network or short-range wireless communication network to perform multi-hop relay communication that bypasses the measurement data to an adjacent flood measurement unit or gateway, and the risk determination unit further includes a real-time drainage capacity degradation estimation logic,The above-mentioned flood prediction model unit continuously monitors the error between the predicted flood depth under normal drainage conditions calculated by the above-mentioned flood prediction model unit and the actual flood depth collected in real-time by the above-mentioned flood measurement unit; if the actual flood depth exceeds the predicted flood depth by more than a preset error tolerance range and a pattern of rapid water level rise is detected, the above-mentioned risk judgment unit presumes that the storm drain or sewer pipe at the relevant point is physically blocked by fallen leaves or sediment; the above-mentioned risk judgment unit temporarily lowers the drainage facility information parameter among the topographic spatial information of the relevant point stored in the above-mentioned spatial information database unit to 0 or a minimum value, and additionally assigns a poor drainage weight when calculating the multi-factor composite risk index to raise the flood risk grade; and further includes a video flood verification module that receives real-time video data from a CCTV installed at or near the installation point of the above-mentioned flood measurement unit and independently verifies the presence of flooding and the flood depth based on video analysis; the above-mentioned video flood verification module analyzes the flood cover area of ​​reference structures such as road surfaces, curbs, and drains from the CCTV video based on deep learning A sensor-based urban flood prediction system that estimates image-based flood depth by analyzing with a semantic segmentation algorithm, compares the estimated image-based flood depth with the sensor measurement value of the flood measurement unit, assigns a reliability degradation flag to the flood measurement data if the deviation between the two exceeds a preset allowable range, transmits it to the preprocessing unit so that secondary correction processing is performed by the outlier processing module, and provides the actual occurrence of flood and flood depth confirmed through image analysis as CCTV-based flood verification data for updating the reference value of the risk standard database of the risk judgment unit. Claim 2 delete Claim 3 delete Claim 4 A sensor-based urban flood prediction system according to claim 1, wherein the flood prediction model unit comprises an LSTM model including at least one of a many-to-one structure that receives flood measurement data and rainfall data from a past time interval and outputs a flood depth at a specific future point in time, and a many-to-many structure that outputs a flood depth time series for multiple future points in time, and wherein the LSTM model calculates a predicted flood depth for at least one future point in time among 1 minute later, 10 minutes later, 30 minutes later, and 60 minutes later. Claim 5 delete Claim 6 A sensor-based urban flood prediction system according to claim 1, wherein the flood information providing unit comprises: an impact range calculation module that calculates the flood impact range using the predicted flood depth corresponding to the installation points of the plurality of flood measurement units and the digital elevation model stored in the spatial information database unit; a flood trace map generation module that visualizes the distribution of flood range and flood depth on a map by combining the digital elevation model grid of the spatial information database unit and the flood impact range; and a risk notification generation module that generates risk notification information including a flood risk grade calculated by the risk judgment unit, a predicted time, a predicted flood depth, a target area, and a warning message.

Images