KR102710812B1 - Apparatus for predicting risk of inundation - Google Patents

Abstract

The present invention relates to a flood risk prediction device, comprising: a raw data acquisition module for acquiring first raw data corresponding to water level, tide level and rainfall data measured in real time through a water level sensor, a tide level sensor and a rainfall observation station installed in a prediction target area, and second raw data corresponding to CCTV image data captured in real time through a plurality of CCTV cameras installed in the prediction target area; a first prediction output unit for outputting a first map image output when applying water level, tide level and rainfall data at a current point in time based on the first raw data as input values to a first prediction model pre-trained to acquire a map image according to a simulation result of internal and external flood risk in the prediction target area, as internal and external flood prediction information predicted based on a current rainfall condition; a second prediction output unit for outputting a first area detection image output when applying CCTV image data at a current point in time based on the second raw data as input values to a second prediction model pre-trained to acquire an area detection image for an expected flood area in the prediction target area, as flood risk area information predicted based on a current CCTV situation; and the outputted It is characterized by including a prior prediction module that generates prior prediction information corresponding to an image visualized with at least one of the internal and external flood prediction information and the flood risk area information together with the first raw data or the second raw data at the current point in time, and displays the generated prior prediction information on a screen.
Accordingly, it is possible to provide real-time prediction information on the possibility of internal and external flooding and risk areas based on current rainfall conditions or current CCTV situations, which has the effect of preventing flood damage in advance and responding quickly.

Description

{APPARATUS FOR PREDICTING RISK OF INUNDATION}

The present invention relates to a flood risk prediction device that can predict the degree of inland and external flooding in a specific area in real time by utilizing observation data and nearby CCTV footage when rainfall occurs in that area.

In the 21st century, the scale of floods is becoming larger, and their frequency and intensity are also increasing. In addition, as global warming continues, extreme weather events are occurring all over the world. Korea is no exception, with temperatures rising by about 1.5℃ over the past 100 years, and damage from extreme weather events such as heavy rain and typhoons is becoming more severe day by day.

Our country has a large range of average annual precipitation, and due to the characteristics of the national terrain, the rivers have steep slopes and are vulnerable to flooding. Since heavy rain falls heavily in the summer, the heavy rain can be concentrated in river basins and cause flooding in cities.

In order to respond to floods occurring in riverine areas and cities, more accurate flood prediction and simulation are needed and flood control measures need to be established based on the results, but such devices or methods are completely absent.

KR 10-2023-0072876 A KR 10-1906858 B1

The purpose of the present invention is to solve the above problem, and the purpose is to provide a flood risk prediction device that can predict the degree of inland and external flooding in a specific area in real time by utilizing observation data and nearby CCTV footage when rainfall occurs in the specific area.

In order to achieve the above object, a flood risk prediction device according to one aspect of the present invention comprises: a raw data acquisition module for acquiring first raw data corresponding to water level, tide level, and rainfall data measured in real time through a water level sensor, a tide level sensor, and a rainfall observation station installed in a prediction target area; and second raw data corresponding to CCTV image data captured in real time through a plurality of CCTV cameras installed in the prediction target area; a first prediction output unit for outputting a first map image output when applying the current water level, tide level, and rainfall data based on the first raw data as input values to a first prediction model pre-trained to acquire a map image according to a simulation result of the internal and external flood risk of the prediction target area as internal and external flood prediction information predicted based on the current rainfall status; and a second prediction unit for outputting the first area detection image output when applying the current CCTV image data based on the second raw data as input values to a second prediction model pre-trained to acquire an area detection image for an expected flood area of the prediction target area as flood risk area information predicted based on the current CCTV situation. It is characterized by including an output section and a prior prediction module that generates prior prediction information corresponding to an image visualized with at least one of the output internal/external water flood prediction information and the flood risk area information together with the first raw data or the second raw data of the current point in time, and displays the generated prior prediction information on a screen.

Preferably, the method further includes a first dataset generation module that generates a first dataset including images labeling areas corresponding to the inundation depth and inundation range of each grid on the 3D map by simulating past external and internal flooding conditions according to the first raw data on a 3D map reflecting topographic information of the prediction target area, and a second dataset generation module that generates a second dataset including images labeling the inundated areas in each frame of the CCTV image data of a specific period of time by using metadata that time-synchronizes the first raw data and the second raw data of a specific period of time based on measured rainfall-runoff events according to the first raw data, and the first prediction model and the second prediction model are characterized in that they are pre-trained using the first dataset and the second dataset, respectively.

In addition, the first dataset creation module is characterized by including a terrain information acquisition unit that extracts and acquires terrain information on elevation, river cross-section, and soil map of the prediction target area from administrative district-specific digital elevation model data pre-stored in an external GIS server, a numerical model creation unit that generates a numerical model that simulates the inundation state of external and internal waters of the prediction target area by applying the past water level, tide level, and rainfall data to a preset rainfall-runoff analysis model based on a 3D map based on the terrain information, and a first dataset creation unit that generates and stores a first dataset including a plurality of numerical model images in which the grid-specific inundation depth and inundation range of the numerical model are mapped onto the 3D map and then the corresponding area is labeled according to a preset inundation depth legend.

Preferably, the second dataset generation module is characterized by including a metadata generation unit which derives an actual rainfall-runoff event according to a rainfall-water level or rainfall-water level-tide level correlation based on the first raw data and generates metadata by time-synchronizing the first raw data and the second raw data of a specific past period based on the metadata, and a second dataset generation unit which generates a second dataset including a plurality of frame images in which a flooded area is extracted from each frame of the second raw data acquired in the specific past period and labeled using a polygon technique.

Preferably, the advance forecast module is characterized in that, when the internal/external water flood prediction information is output, the first advance forecast information is generated by displaying, on the first map image, the coordinates and numbers of points adjacent to the flooded area in the first map image among a plurality of points numbered in the order of flood occurrence based on the topographical height of the prediction target area, and, when the flood risk area information is output, the module compares the frame image of the CCTV video data at the current point in time with the first area detection image, and, if a common object area exists, generates second advance forecast information by overlaying the corresponding object area on the frame image.

According to the present invention, it is possible to provide real-time prediction information on the possibility of internal and external flooding and risk areas based on current rainfall conditions or current CCTV situations, thereby enabling advance prevention of flood damage and rapid response.

The effects of the present invention are not limited to the effects mentioned above, and other effects not mentioned will be clearly understood by those skilled in the art from the description of the claims.

Figure 1 is a drawing schematically showing the configuration of a flood risk prediction device according to one embodiment of the present invention.
FIG. 2 is a block diagram illustrating in detail the internal configuration of a flood risk prediction device according to one embodiment of the present invention.
Figure 3 is a drawing showing the installation locations of water level, tide level and rainfall sensors that communicate with the weather observation server of Figure 2 and provide sensor data.
Figure 4 is a drawing showing the installation location of a CCTV camera that provides CCTV video data by communicating with the CCTV control server of Figure 2.
Figure 5 is a drawing showing an example of an image of the first dataset generated by the first dataset generation unit of Figure 2 and a legend classification of the immersion depth for the same.
Figure 6 is a drawing for explaining the process of generating metadata by the metadata generation unit of Figure 2.
FIG. 7 is a drawing showing an image of a second dataset generated by the second dataset generation unit of FIG. 2 and an example of attribute information included in the second dataset.
Figure 8 is a drawing showing the range of the detection reference area for determining the flood risk level by the risk level determination unit of Figure 2.
Figure 9 is a drawing showing an example of first advance prediction information generated by the first advance prediction unit of Figure 2.
Figure 10 is a flowchart showing a flood risk prediction method using a flood risk prediction device according to the present invention.

The specific details of the present invention, including the problems to be solved, the means for solving the problems, and the effects of the invention, are included in the embodiments and drawings described below. The advantages and features of the present invention, and the methods for achieving them, will become clear with reference to the embodiments described below in detail together with the accompanying drawings. Like reference numerals throughout the specification refer to like elements.

FIG. 1 is a drawing schematically showing the configuration of a flood risk prediction device according to one embodiment of the present invention, FIG. 2 is a block diagram showing the internal configuration of a flood risk prediction device according to one embodiment of the present invention in detail, FIG. 3 is a drawing showing the installation locations of water level, tide level, and rainfall sensors that communicate with the weather observation server of FIG. 2 to provide sensor data, FIG. 4 is a drawing showing the installation locations of CCTV cameras that communicate with the CCTV control server of FIG. 2 to provide CCTV image data, FIG. 5 is a drawing showing an example of an image of a first dataset generated by the first dataset generation unit of FIG. 2 and a flood depth legend classification for the same, FIG. 6 is a drawing for explaining a process of generating metadata by the metadata generation unit of FIG. 2, FIG. 7 is a drawing showing an example of an image of a second dataset generated by the second dataset generation unit of FIG. 2 and attribute information included in the second dataset, FIG. 8 is a drawing showing the range of a detection reference area for determining a flood risk level by the risk level determination unit of FIG. 2, and FIG. 9 is a drawing This is a drawing showing an example of first advance prediction information generated by the first advance prediction unit of Fig. 2, and Fig. 10 is a flowchart showing a flood risk prediction method using a flood risk prediction device according to the present invention.

Hereinafter, a flood risk prediction device according to a preferred embodiment of the present invention will be described with reference to the drawings described above.

Referring to FIG. 1, a flood risk prediction device according to one embodiment of the present invention is largely configured to include a raw data acquisition module (100), a first dataset generation module (200), a second dataset generation module (300), a first prediction model (400), a first training unit (450), a second prediction model (500), a second training unit (550), a first prediction output unit (610), a second prediction output unit (620), and a prior prediction module (700).

The raw data acquisition module (100) communicates with an external weather observation server (10) and CCTV control server (20) to acquire raw data related to the precipitation status of the prediction target area ( _AT ) (S100).

The raw data acquisition module (100) is composed of a first raw data acquisition unit (110) that acquires first raw data (r ₁ ) measuring the water level, tide level, and rainfall amount of the prediction target area (A _T ) when rainfall occurs, and a second raw data acquisition unit (120) that acquires second raw data (r ₂ ) corresponding to CCTV images of each surveillance area within the prediction target area (A _T ).

The first raw data acquisition unit (110) acquires first raw data ( _r 1 ) corresponding to water level, tide level and rainfall data measured in real time through a water level sensor ( ₁ ), a tide level sensor (3) and a rainfall observation station (5) installed in the prediction target area (A T ).

Here, the first raw data (r ₁ ) may include water level data and tide level data measured in real time by water level sensors (1) and tide level sensors (3) installed in rivers and coasts belonging to the administrative district of the _prediction target area (A T ), and rainfall data for each observation point measured in real time by multiple rainfall observation stations (5) installed and operated by the local government or the Korea Meteorological Administration that governs the prediction target area (A _T ).

The second raw data acquisition unit (120) acquires second raw data (r ₂ ) corresponding to CCTV image data captured in real time through multiple CCTV cameras (7) installed within the prediction target area (A _T ).

Here, it is preferable that multiple CCTV cameras (7) be installed around roads or rivers in multiple pre-designated flood risk areas within the predicted target area ( _AT ). The flood risk area may mean an area at risk of natural disasters such as flooding and loss due to overflow (outflow) of rivers and poor inland water drainage when heavy rain or typhoons occur.

Meanwhile, in the case of the first raw data acquisition unit (110) according to the present invention, although not shown in the drawing, in addition to the aforementioned river water level data, coastal tide level data, and rainfall data by observation point, rainwater gutter water level data measured by a water level sensor installed adjacent to a rainwater gutter placed in a drainage ditch of a road or alley in a flood-risk area can be additionally acquired.

The first dataset creation module (200) creates a first dataset (D ₁ ) based on the results of simulating the flooding status of the prediction target area ( _AT ) based on the first raw data (r ₁ ) (S200).

The first dataset creation module (200) can create a first dataset ( _{D 1} ) including images labeling areas corresponding to the inundation depth and inundation range of each grid on the 3D map by simulating past external and internal flooding conditions according to the first raw data (r ₁ ) on a 3D map reflecting topographic information of the prediction target area (AT ).

Specifically, the first dataset creation module (200) may include a terrain information acquisition unit (210), a numerical model creation unit (220), and a first dataset creation unit (230) as illustrated in FIG. 2.

The topographic information acquisition unit (210) extracts and acquires topographic information on the elevation, river section, and soil map of the prediction target area ( _AT ) from digital elevation model (DEM) data for each administrative district stored in an external GIS server.

Here, digital elevation model (DEM) data may mean two-dimensional spatial information including at least one of GPS-based map, aerial photograph, and satellite image based on the geographical location of administrative districts of the Republic of Korea, and DEM (Digital Elevation Model) information that converts the information into a three-dimensional height map format.

Additionally, the above-mentioned topographic information may include topographic data for river cross sections, digital elevations, land cover maps, and sewer network maps.

The numerical model generation unit (220) generates a numerical model (RUN) to simulate the flooding conditions for external and internal waters in the prediction target area ( _AT ).

The numerical model generation unit (220) can generate a numerical model (RUN) that simulates the flooding state of external and internal waters of the prediction target area (AT) by applying the first raw data (r ₁ ) of the past to a preset rainfall-runoff analysis model based on a 3D map based on the terrain information acquired by the terrain information acquisition unit ( ₂₁₀ ).

Here, the above rainfall-runoff analysis model can be applied by linking the HEC-RAS model and the XP-SWMM model. The HEC-RAS model simulates flood inundation and inundation of the prediction target area ( _AT ), and the XP-SWMM model divides the simulation data into spatial grid units and calculates the inundation depth and inundation range for each grid, thereby performing an analysis of the risk of internal and external flooding.

The HEC-RAS model is an improved version of the HEC2 model, which analyzes the water surface curve in the steady-state, gradually varying flow of artificial or natural rivers. It is used for design review of various river hydraulic facilities (bridges, embankments, dams, culverts, etc.) and river flow simulation. It has the characteristics of being able to calculate one-dimensional steady uneven flow and unsteady flow that changes over time, and perform two-dimensional flood analysis and similar simulations.

The XP-SWMM model can be implemented by a hydraulic-hydrological simulation software that integrates one-dimensional calculations for water flow in the basin-conduit-channel based on the EPA-SWMM engine and flood flow calculations based on the TUFLOW engine. The EPA-SWMM engine is a rainfall-runoff and surface runoff simulation model that can simulate various hydrological runoff situations and impose various hydraulic conditions on hydraulic structures. The TUFLOW engine is a two-dimensional numerical analysis model used for runoff analysis in rivers, ports, urban basins, and coasts, and can be used for numerical simulations of shallow water equations by the Stelling finite difference method and the ADI (alternating direction implicit) technique.

The first dataset generation unit (230) generates and stores a first dataset (D 1 ) including a plurality of numerical model images (I _S ) representing the inundation depth and inundation range based on the numerical model (RUN) generated by the numerical model generation unit ( ₂₂₀ ).

The first data set generation unit (230) can generate a first data set (D 1 ) including a plurality of numerical model images ( _IS ) that map the grid-by-grid inundation depth and inundation range of the numerical model (RUN) onto a 3D map based on topographic information of the prediction target area ( _AT ), and then label the corresponding area according to a preset inundation _depth legend, and then store the same.

Here, the flood depth legend may include a plurality of flood depth numerical range groups, such as a river and a lake, as illustrated in FIG. 5.

In this case, the first data set generation unit (230) may first classify 'rivers' and 'inland areas' based on the river locations indicated in the numerical model (RUN), and then classify each flooded area into one of the 'first group (0.3 m to less than 0.5 m)', 'second group (0.5 m to less than 1 m)', 'third group (1 m to less than 1.5 m)', 'fourth group (1.5 m to less than 2 m)', and 'fifth group (3 m or more)' based on the results of comparing the flood depth values for each flooded area calculated by the numerical model (RUN) with a plurality of preset reference values.

At this time, the above-mentioned flooded area may mean an area corresponding to a case where the grid-wise flooding depth on a 3D map based on a numerical model (RUN) is greater than a preset minimum threshold value (e.g., 0.3 m).

In addition, the first data set generation unit (230) can generate the first data set (D ₁ ) by labeling each classification with a different color according to the preset submergence depth legend classification as illustrated in FIG. 5.

The second dataset creation module (300) creates a second dataset (D _{2 ) that indicates a flood risk area for CCTV images of a flood risk area within a prediction target area (AT ) based on the second raw data (r 2 ) (} S200).

The second dataset creation module (300) can create a second dataset (D ₂ ) including images labeling the flooded area within each frame of CCTV video data of a specific period of time by using metadata that time-synchronizes the first raw data (r ₁ ) and the second raw data (r ₂ ) of a specific period of time based on the actual rainfall-runoff event according to the first raw data (r ₁ ).

Specifically, the second dataset creation module (300) may include a metadata creation unit (310), a second dataset creation unit (320), and a risk level determination unit (330) as illustrated in FIG. 2.

The metadata generation unit (310) derives an actual rainfall-runoff event based on the rainfall-water level or rainfall-water level-tide level correlation based on the first raw data (r ₁ ), and based on this, synchronizes the first raw data (r ₁ ) and the second raw data (r ₂ ) of a specific period in the past to generate metadata (M).

At this time, the first raw data (r ₁ ) of the specific period may be a set of rainfall data measured at preset intervals (e.g., 1 minute, 10 minutes, 1 hour, etc.), and the second raw data (r ₂ ) of the specific period may be a set of CCTV video data taken in an area closest to a rainfall observation station where the first raw data (r ₁ ) was measured.

Additionally, metadata (M) may be generated by time-synchronizing first raw data (r ₁ ) and second raw data (r ₂ ) that satisfy the conditions of identity of the measurement period and location adjacency.

The second dataset creation unit (320) creates and stores a second dataset (D ₂ ) that includes multiple frame images (I _A ) that indicate the submerged area in each frame of the second raw data (r ₂ ) used in creating the metadata (M).

The second dataset generation unit (320) can detect an object corresponding to an underwater area included in each frame of the second raw data (r ₂ ) through a preset feature map-based object detection algorithm, generate a second dataset (D ₂ ) including a plurality of frame images labeled using a polygon technique along the boundary portion of the detected object area, and then store the second dataset.

Here, the feature map can be generated by analyzing feature pattern components for the flooded area using a plurality of preset patterned masks targeting the second raw data (r ₂ ) acquired prior to the generation of metadata (M).

In this case, the object detection algorithm based on the aforementioned feature map may extract an object area corresponding to the submerged area based on the result of comparing the feature pattern components of the feature map with the feature pattern components of the video image of the second raw data (r ₂ ) used in generating the metadata (M) and determining whether they are similar or identical within a preset threshold range.

The second dataset generation unit (320) can generate a second dataset (D ₂ ) including a plurality of frame images labeled with flooded areas and corresponding attribute information and metadata (M).

Here, the attribute information may include dataset information (object_info) including the dataset name, creation date, detailed description and URL information, cumulative rainfall information by periodic unit, rainfall observation location information, river water level observation location information, and river water level bridge name information.

Additionally, the attribute information may include category information (object_categories) including category name and risk level information.

Additionally, the attribute information may include image information (object_image) including horizontal resolution, vertical resolution, and name information of the image and data collection tool information.

Additionally, the attribute information may include labeling information (object_annotations) including image ID, category ID, and labeling coordinate value information.

The risk level judgment unit (330) judges the flood risk level for whether or not there is internal or external flooding and the degree of flooding based on the result of comparing the labeled area corresponding to the flooded area in the image of the second data _set (D 2 ) with the preset detection reference area.

The risk level judgment unit (330) detects objects corresponding to river facilities or road facilities (e.g., bridges, streetlights, river curbs, road curbs, etc.) in the images of the second data set (D ₂ ), and then distinguishes multiple comparison target areas for judging the risk level based on the locations of the detected objects.

The risk level judgment unit (330) can set the boundary coordinates of a plurality of comparison target areas as detection reference areas for each risk level, and can judge the risk level based on whether the labeling area in the image of the second data set (D ₂ ) has crossed the boundary of at least one reference area.

The risk level judgment unit (330) first divides a plurality of comparison target areas in the image of the second data set (D ₂ ) into a river area (Q1), a riverbank area (Q2), and a road area based on the object locations detected in response to river facilities or road facilities, and then, if the image of the second data set (D ₂ ) includes a river area (Q1) and a riverbank area (Q2), divides the road area into an external flood area (Q3) and an external flood area (Q4) based on the proximity to the riverbank area (Q2), and if the image of the second data set (D ₂ ) does not include a river area (Q1) or a riverbank area (Q2), divides the road area into an initial flood area (Q5) and an inland flood area (Q6) based on the geographical proximity of the road area to the surrounding river.

The risk level determination unit (330) compares the labeled area in the image of the second data set (D ₂ ) with a plurality of preset detection reference areas (river area, embankment area, external water flood area, external water overflow area, initial flood area, inland water flood area) as illustrated in FIG. 8, and determines the risk level for each frame image of the second data set (D ₂ ) as one of the following: safe level (L1), embankment flood stage (L2), flood risk stage (L3), external water overflow stage (L4), initial flood stage (L5), and inland water overflow stage (L6).

For example, in the second dataset (D ₂ ) based on CCTV video data that captures a river and a road together, if the boundary of the labeling area in the second dataset (D ₂ ) is located within the river area (Q1), it can be determined as a 'safe stage (L1)', if the labeling area includes the river area (Q1) and the boundary of the labeling area is located within the riverbank area (Q2), it can be determined as a 'riverbank flood stage (L2)', if the labeling area includes the river area (Q1) and the riverbank area (Q2) and the boundary of the labeling area is located within the external flood area (Q3), it can be determined as a 'flood risk stage (L3)', and if the labeling area includes the river area (Q1), the riverbank area (Q2) and the external flood area (Q3) and the boundary of the labeling area is located within the external flood area (Q4), it can be determined as a 'external flood stage (L4)'.

In addition, in the second dataset (D ₂ ) based on CCTV video data that only captures roads around rivers, if the boundary of the labeling area in the second dataset (D ₂ ) is located within the initial flooding area (Q5), it can be determined as the 'initial flooding stage (L5)', and if the initial flooding area (Q5) is included in the labeling area and the boundary of the labeling area is located within the inland flooding area (Q6), it can be determined as the 'inland flooding stage (L6)'.

In this case, the second dataset creation unit (320) can match and store the flood risk level information determined by the risk level determination unit (330) with the corresponding image of the second dataset (D ₂ ), which may correspond to the risk level information included in the category information (object_categories) of the attribute information described above.

The first prediction model (400) is trained in advance using the first dataset (D ₁ ) to predict the level of risk of internal and external flooding according to rainfall conditions.

The first prediction model (400) has a recurrent neural network structure that processes a time series step by step and maintains an internal state between time steps to learn time dependency, but may include an LSTM (Long short-term memory) neural network model that can be learned by dividing short-term memory and long-term memory.

The first training unit (450) can train the first prediction model (400) using the first data set (D ₁ ) so that the first prediction model (400) outputs a numerical model-based map image (I _S ) predicted based on time series analysis of sensor data ( _{r 11} ) in response to input of any one rainfall condition-related sensor data (r 11 ).

The second prediction model (500) is trained in advance using the second dataset (D ₂ ) to predict flood risk areas according to CCTV situations.

The second prediction model (500) may include a neural network model based on the YOLO algorithm, which divides the original image into grids of the same size, predicts a number of bounding boxes, i.e. anchor boxes, which are designated in a predefined shape centered on the center of the grid for each grid, and calculates reliability based on this to select a location with high object reliability to identify the object category. Preferably, the neural network model may be implemented as a YOLOv7-Seg model.

The second training unit (550) can train the second prediction model (500) using the second dataset (D ₂ ) so that the second prediction model (500) outputs an image (I _A ) indicating a flood risk area predicted based on object analysis in the CCTV video data (r ₂₂ ) in response to input of one CCTV video data (r ₂₂ ).

The first prediction output unit (610) is for outputting prediction information on internal and external water inundation based on the prediction values of the first prediction model (400) based on the current water level, tide level, and rainfall data.

The first prediction output unit (610) can output the first map image (I _S1 ) as predicted internal and external flooding prediction information predicted based on the current rainfall conditions when the current water level, tide level, and rainfall data based on the first raw data (r ₁ ) are applied as input values to the first prediction model ( _LSTM ) that has been pre-trained to obtain a map image according to the simulation results of the internal and external flooding risk of the predicted target area (AT ).

The second prediction output unit (620) is for outputting the prediction value of the second prediction model (500) based on the CCTV image data at the current point in time as flood risk area information.

The second prediction output unit (620) can output the first area detection image (I _A1 ) as flood risk area information predicted based on the current CCTV situation when applying the CCTV image data of the current point in time based on the second raw data (r ₂ ) as an input value to the second prediction model ( _YOLO ) that has been pre-trained to obtain an area detection image for the flood expected area of the prediction target area (A T ).

The advance forecast module (700) generates and displays advance forecast information based on the outputs of the first prediction output unit (610) and the second prediction output unit (620).

The advance forecast module (700) can generate advance forecast information corresponding to an image visualizing at least one of the above-mentioned internal and external water flood forecast information and the above-mentioned flood risk area information and the corresponding first raw data or second raw data at the current point in time, and display the generated advance forecast information on the screen.

Specifically, the advance forecast module (700) may include a first advance forecast unit (710), a second advance forecast unit (720), and a forecast display unit (730) as illustrated in FIG. 2.

When the internal and external flood prediction information is output from the first prediction output unit (610), the first advance prediction unit ( ₇₁₀ ) can generate first advance prediction information that displays the coordinates and numbers of points adjacent to the flooded area in the first map image (I _S1 ) among a plurality of points numbered in the order of flood occurrence based on the topographical height of the prediction target area (AT), on the first map image (I _S1 ).

For example, if the labeling area corresponding to the flooded area in the first map image (I _S1 ) is as shown in FIG. 9, the first advance forecast unit (710) generates first advance forecast information that displays the numbers (①, ②, ③) of points adjacent to the flooded area at coordinates and adjacent locations of the corresponding points on the first map image (I _S1 ).

When flood risk area information is output from the second prediction output unit (610), the second advance prediction unit (720) can compare the frame image of the CCTV video data at the current point in time with the first area detection image (I _A1 ) and, if a common object area exists, generate second advance prediction information that overlays the corresponding object area on the frame image.

The forecast display unit (730) can display first advance forecast information or second advance forecast information on the screen of a pre-registered user terminal (not shown).

The forecast display unit (730) displays the corresponding prior prediction information as is on the screen of a user terminal (not shown) when only one of the first prior prediction information and the second prior prediction information is generated, and when both the first prior prediction information and the second prior prediction information are generated, the coordinates of each point displayed on the first map image (I _S1 ) and the geographical location coordinates of the CCTV camera from which the first area detection image (I _A1 ) was acquired are linked to additionally display the CCTV camera location corresponding to the first area detection image (I _A1 ) on the first map image (I _S1 ), and then display this together with the second prior prediction information on the screen of the user terminal (not shown).

Although the present invention has been described in detail through preferred embodiments, the present invention is not limited thereto and may be implemented in various ways within the scope of the claims.

In particular, since the foregoing has described the features and technical strengths of the present invention rather broadly in order to better understand the scope of the claims of the invention described later, it should be recognized by those skilled in the art that the concept and specific embodiments of the present invention described above can be immediately used as a basis for designing or modifying other shapes for carrying out similar purposes to the present invention.

In addition, it will be understood that the above-described embodiment is only one embodiment according to the present invention, and that it can be implemented in various modified and changed forms within the scope of the technical idea of the present invention by a person having ordinary skill in the art. Therefore, the disclosed embodiment should be considered from an illustrative point of view rather than a limiting point of view, and such various modifications and changes are also included in the scope of the technical idea of the present invention, which is indicated in the claims of the present invention described above, and all differences within the equivalent scope should be interpreted as being included in the present invention.

1: Water level sensor 3: Tide level sensor
5: Rainfall Observatory 7: CCTV
10: CCTV control server 20: Weather observation server
100: Raw data acquisition module 110: First raw data acquisition unit
120: Second raw data acquisition unit 200: First data set creation module
210: Terrain information acquisition section 220: Numerical model creation section
230: First dataset creation unit 300: Second dataset creation module
310: Metadata generation unit 320: Second dataset generation unit
330: Risk level judgment unit 400: First prediction model
450: 1st training unit 500: 2nd prediction model
550: 2nd Training Section 610: 1st Prediction Output Section
620: Second prediction output unit 700: Advance forecast module
710: 1st Advance Forecasting Department 720: 2nd Advance Forecasting Department
730: Forecast display

Claims (5)

A raw data acquisition module that acquires first raw data corresponding to water level, tide level and rainfall data measured in real time through water level sensors, tide level sensors and rainfall observation stations installed within a prediction target area, and second raw data corresponding to CCTV image data captured in real time through a plurality of CCTV cameras installed within the prediction target area;
A first dataset generation module that generates a first dataset including images labeling areas corresponding to the inundation depth and inundation range of each grid on the 3D map by simulating past external and internal flooding conditions according to the first raw data on a 3D map reflecting topographic information of the above-mentioned predicted target area;
A second dataset generation module that generates a second dataset including images labeling the flooded area within each frame of the CCTV video data of a specific period of time by using metadata that time-synchronizes the first raw data and the second raw data of the specific period of time based on the actual rainfall-runoff event according to the first raw data;
A first prediction output unit that outputs a first map image as predicted internal and external water inundation prediction information based on current rainfall conditions when applying current water level, tide level and rainfall data based on the first raw data as input values to a first prediction model pre-trained to obtain a map image according to simulation results on the risk of internal and external water inundation in the predicted target area;
A second prediction output unit that outputs the first area detection image as flood risk area information predicted based on the current CCTV situation when applying the CCTV image data at the current point in time based on the second raw data as an input value to the second prediction model that has been pre-trained to obtain an area detection image for the flood expected area of the above prediction target area; and
A prior prediction module that generates prior prediction information corresponding to an image visualizing at least one of the output internal and external flood prediction information and the flood risk area information and the first raw data or the second raw data of the current point in time, and displays it on a screen;
The above first prediction model and the above second prediction model are pre-trained using the first dataset and the second dataset, respectively.
The above second dataset creation module,
A metadata generation unit that derives actual rainfall-runoff events based on the rainfall-water level or rainfall-water level-tide level correlation based on the first raw data and generates metadata by time-synchronizing the first raw data and the second raw data of a specific period in the past based on the derived events;
A second dataset generation unit that generates and stores a second dataset including a plurality of frame images in which the submerged area is extracted from each frame of the second raw data acquired during the specific past period and labeled using a polygon technique;
A method of detecting an object corresponding to a river facility or a road facility in an image of the second dataset, and then distinguishing a plurality of comparison target areas for determining a risk level based on the location of the detected object, and setting the boundary coordinates of the plurality of comparison target areas as a detection reference area for each risk level, and including a risk level determination unit for determining a risk level based on whether a labeling area in an image of the second dataset exceeds the boundary of at least one detection reference area,
The above second dataset creation section
A flood risk prediction device characterized in that the risk level determined by the risk level determination unit is matched with an image of the second dataset and stored. delete In the first paragraph,
The above first dataset creation module,
A topographic information acquisition unit that extracts and acquires topographic information on elevation, river cross-section, and soil map of the predicted target area from digital elevation model data by administrative district stored in an external GIS server;
A numerical model generation unit that generates a numerical model that simulates the flooding state of external and internal waters of the prediction target area by applying the first raw data of the past to a preset rainfall-runoff analysis model based on a 3D map based on the above-mentioned terrain information; and
A flood risk prediction device characterized by comprising: a first dataset generation unit for generating and storing a first dataset including a plurality of numerical model images in which the grid-by-grid flood depth and flood range of the numerical model are mapped onto the 3D map and the corresponding area is labeled according to a preset flood depth legend. delete In the first paragraph,
The above advance forecast module,
When the above internal and external flood prediction information is output, first prior prediction information is generated to display the coordinates and number of a point adjacent to the flooded area in the first map image among a plurality of points numbered in the order of flood occurrence based on the terrain height of the predicted target area on the first map image.
A flood risk prediction device characterized in that when the above flood risk area information is output, the device compares the frame image of the CCTV video data at the current point in time with the first area detection image, and if a common object area exists, generates second preceding prediction information that overlays the corresponding object area on the frame image.

Images