KR102480985B1 - Typhoon Dismissal Prediction System Using Deep Learning Model - Google Patents

Abstract

The present invention relates to a typhoon surge height prediction system using a deep learning model, comprising: a scenario generator for generating virtual typhoon information based on typhoon data pre-stored in a database through TCRM; a tidal wave generating unit for generating tidal wave height information at a coast based on the virtual typhoon information through an ADCIRC storm surge model; a learning unit that generates learned tidal height prediction data that performs deep learning by putting the generated tidal height information as an input of an artificial neural network; an application unit that generates improved tidal wave prediction data by matching the tidal wave height prediction data generated by deep learning of the learning unit with the tidal wave height data observed during an actual typhoon attack; and a tidal height prediction unit inputting tidal height information generated based on improved tidal height prediction data during an actual typhoon attack to the learning unit and outputting tidal height prediction information generated through deep learning of the learning unit.
According to one embodiment of the present invention as described above, virtual typhoons are randomly generated based on observed typhoon data using TCRM for virtual typhoon scenarios, and the produced observation-based virtual typhoon scenario is input into the ADCIRC storm surge model By calculating the height of the tidal wave at the coast by using it as data, it is possible to predict the height of the tidal wave at the coast due to the invasion of the typhoon.

Description

Typhoon Dismissal Prediction System Using Deep Learning Model}

The present invention relates to a typhoon surge height prediction system using a deep learning model, and more specifically, to randomly generate a virtual typhoon based on observed typhoon data using a tropical cyclone risk model (TCRM) for a virtual scenario of a typhoon, , The produced observation-based virtual typhoon scenario is used as input data to the ADCIRC (ADvanced CIRCulation) storm surge model to calculate the tidal wave height at the coast, and the calculated tidal wave height is calculated in real time in the event of an actual typhoon attack through a deep learning model. It's about predictive technology.

Unpredictable phenomena are appearing all over the world to the extent that the words climate change and abnormal climate are not awkward. Heavy rains in South Asia, heat waves, storms and floods in the eastern part of the United States, as well as the heavy rains in August last year in Korea, can be cited as representative examples. Such meteorological phenomena are significantly different from those of the 20th century. In particular, they have never been observed in the past and generally do not occur according to a regular cycle or pattern. there is.

Recently, along with global warming and rapid urbanization and industrialization, disasters that occur along the coast due to two or more complex causes such as sea level rise, typhoons, or concentrated rainfall have frequently occurred.

According to IPCC AR5, it has been studied that the sea level will rise by 40cm to 63cm by 2100, and in particular, it is predicted that frequent inundation damage caused by Baekjungsari in low-lying areas along the coast will occur (maximum high tide).

In addition, as the intensity of typhoons increases due to climate warming, the frequency of super typhoons attacking the Korean Peninsula is increasing, resulting in increased tidal waves and waves, increasing the possibility of complex disasters. In addition, the possibility of localized heavy rain and extreme rainfall is also increasing in the rainfall pattern, and it is predicted that the rainfall on the Korean Peninsula will increase by 3% in the short term and by 10% in the long term.

The Korea Meteorological Administration uses supercomputers and various meteorological data to provide predicted information on the typhoon's path, central pressure, moving speed, maximum wind speed, strong wind radius, size, and direction. Although the model has been performed to predict the height of surge generated off the coast, the establishment and actual application of the numerical model of storm surge takes a long time, and it is difficult to accurately predict the height of surge generated off the coast, and the accuracy is also low. to be.

Accordingly, the present applicant randomly generates virtual typhoon scenarios based on observed typhoon data using TCRM, and uses the produced virtual typhoon scenarios as input data to the ADCIRC storm surge model to determine the height of surge at the coast. In addition, we intend to propose a system that predicts the height of a tidal wave in real time more quickly in the event of an actual typhoon through a deep learning model using the calculated multiple heights of a tidal wave in real time.

Korean Patent Publication No. 10-2012-0000716 (2012.01.04)

The purpose of the present invention is to randomly generate virtual typhoon scenarios based on observed typhoon data using TCRM, and use the produced virtual typhoon scenarios as input data to the ADCIRC storm surge model By calculating the height of the tidal wave, it is to predict the height of the tidal wave at the coast due to the invasion of the typhoon.

An object of the present invention is to quickly improve the accuracy of predicting the height of a tidal wave in real time during an actual typhoon attack by learning a plurality of heights of a tidal wave calculated through a virtual scenario of a typhoon through a deep learning model.

The typhoon surge height prediction system using the deep learning model of the present invention to achieve this technical problem includes a scenario generator for generating virtual typhoon information based on typhoon data pre-stored in a database through TCRM; a tidal wave generating unit for generating tidal wave height information at a coast based on the virtual typhoon information through an ADCIRC storm surge model; a learning unit that generates learned tidal height prediction data that performs deep learning by putting the generated tidal height information as an input of an artificial neural network; an application unit that generates improved tidal wave prediction data by matching the tidal wave height prediction data generated by deep learning of the learning unit with the tidal wave height data observed during an actual typhoon attack; and a tidal height prediction unit inputting tidal height information generated based on improved tidal height prediction data during an actual typhoon attack to the learning unit and outputting tidal height prediction information generated through deep learning of the learning unit.

Preferably, it is characterized by further comprising a variable setting unit for generating a Radius of Maximum Wind (RMW) variable that replaces a variable for generating tidal height information.

The RMW variable is a variable that sets the maximum wind radius value, which is the distance to the point where the wind speed is strong, and is characterized in that it is generated by [Equation 1].

[Equation 1]

The learning unit includes an input module for inputting the tsunami height information received from the tsunami height generation unit into a long short-term memory models (LSTM) artificial neural network; A learning module that performs deep learning learning so that tidal height information input to the LSTM artificial neural network corresponds to a set category; and an output module outputting the tsunami height prediction data learned from the tsunami height information by the learning module.

According to the present invention as described above, a virtual typhoon is randomly generated based on the observed typhoon data using TCRM for a virtual typhoon scenario, and the produced observation-based virtual typhoon scenario is used as input data to the ADCIRC storm surge model. It has the effect of quickly predicting the height of a tidal wave in the event of a typhoon by calculating the height of a tidal wave in and applying the height of a tidal wave to a deep learning model.

According to the present invention, by learning a plurality of heights of tidal waves calculated through virtual scenarios of typhoons through a deep learning model, there is an effect of improving the accuracy of predicting the heights of tidal waves in real time during an actual typhoon attack.

1 is a block diagram showing a typhoon tidal height prediction system using a deep learning model according to an embodiment of the present invention.
2 is an exemplary view showing that a learning unit of a typhoon tidal height prediction system using a deep learning model according to an embodiment of the present invention learns all typhoon information that has affected Korea.
3 is an exemplary view showing that a learning unit of a typhoon surge height prediction system using a deep learning model according to an embodiment of the present invention learns by dividing sea areas.
4 is an exemplary view showing that the learning unit of the typhoon tidal height prediction system using a deep learning model according to an embodiment of the present invention learns using only variables other than the maximum wind speed and the maximum wind radius as input data.
5 is a configuration diagram showing a variable setting unit of a typhoon tidal height prediction system using a deep learning model according to an embodiment of the present invention.
6 is a flowchart illustrating a method for predicting a typhoon surge height using a deep learning model according to an embodiment of the present invention.
7 is a flowchart showing the detailed process of step S606 of the typhoon surge height prediction system using a deep learning model according to an embodiment of the present invention.

Specific features and advantages of the present invention will become more apparent from the following detailed description based on the accompanying drawings. Prior to this, the terms or words used in this specification and claims do not reflect the technical spirit of the present invention based on the principle that the inventor can appropriately define the concept of terms in order to best explain his or her invention. It should be interpreted in accordance with the meaning and concept. In addition, when it is determined that a detailed description of a known function and its configuration related to the present invention may unnecessarily obscure the gist of the present invention, it should be noted that the detailed description is omitted.

Referring to FIG. 1, a typhoon surge height prediction system (S) using a deep learning model according to an embodiment of the present invention provides virtual typhoon information based on typhoon data previously stored in a database through a Tropical Cyclone Risk Model (TCRM) a scenario generating unit 100 that generates, a tidal wave generating unit 200 that generates tidal height information at the coast based on virtual typhoon information through an ADCIRC (ADvanced CIRCulation) storm surge model, and generated tidal wave information The learning unit 300 that generates the learned tidal height prediction data that performs deep learning by putting it as an input of the LSTM (Long Short-Term Memory) artificial neural network, and the tidal wave height generated by the deep learning learning of the learning unit 300 An application unit 400 that generates improved tidal height prediction data by matching predicted data with tidal height data observed during an actual typhoon attack, and tidal wave height information generated based on the improved tidal wave height prediction data during an actual typhoon attack It is configured to include a tsunami height prediction unit 500 that inputs information to the learning unit 300 and outputs tsunami height prediction information generated through deep learning of the learning unit 300 .

Specifically, the scenario generator 100 includes an indexing module 102 for indexing typhoon data previously stored in the database, a TCRM loading module 104 for inputting the indexed typhoon data into TCRM, and typhoon data input to TCRM. It is configured to include a typhoon information generating module 106 that generates virtual typhoon information based on the typhoon information.

In this way, when observation is performed based on typhoon information generated by TCRM, a plurality of pieces of information for a period set according to a scenario can be generated, and the limitations of deep learning can be supplemented only with conventionally observed typhoon information.

In addition, the tidal wave generating unit 200 includes the ADCIRC loading module 202 that inputs the typhoon information authorized from the scenario generating unit 100 into the ADCIRC storm surge model, and the coastal area based on the typhoon information input to the ADCIRC storm surge model. It is configured to include a tsunami height generating module 204 for generating tsunami height information.

In addition, the learning unit 300 includes an input module 302 for inputting the tidal height information received from the tidal height generating unit 200 to the LSTM artificial neural network, and the tidal height information input to the LSTM artificial neural network to correspond to a set category. It is configured to include a learning module 304 that performs deep learning learning, and an output module 306 that outputs tsunami height prediction data learned from the tsunami height information by the learning module 304.

At this time, the category set during deep learning learning includes any one of 'learning all typhoon information', 'learning strong typhoon', 'learning regional typhoon', or 'learning single variable', and learning strong typhoon includes the set central pressure range. It is desirable to understand that only typhoon information belonging to is input to the artificial neural network and learned.

In addition, as shown in FIG. 2, the learning unit 300 selects and learns only typhoons that have affected Korea from among all virtual typhoons performed in the TCRM model and predicts the height of the tidal wave, or as shown in FIG. 3, Divide the affected sea area according to the path of the typhoon (south coast or west coast), select only the typhoon passing through the relevant sea area and predict the height of the tidal wave after learning, or as shown in FIG. 4, input typhoon data (progress path, center Atmospheric pressure, maximum wind speed, and maximum wind radius), only the variables except the maximum wind speed and maximum wind radius can be used as input data and learned, and then the height of the tidal wave can be predicted.

In addition, referring to FIG. 5, the typhoon surge height prediction system (S) using a deep learning model according to an embodiment of the present invention, RMW (RMW) that replaces variables for generating tidal height information to suit the environment of East Asia around the Korean Peninsula It is configured to further include a variable setting unit 600 that generates a Radius of Maximum Wind variable.

는 태풍의 중심 기압이다.At this time, the RMW variable generated by the variable setting unit 600 (a variable for setting the maximum wind radius value, which is the distance to the point where the wind speed is strong) is combined with the typhoon information generated by the scenario generating unit 100 and the tidal wave generating unit. It is input as (200) and set by [Equation 1]. here,

is the central pressure of the typhoon.

[Equation 1]

As described above, according to the typhoon surge height prediction system (S) using the deep learning model according to an embodiment of the present invention, a virtual typhoon based on observed typhoon data is randomly generated by using TCRM for a virtual scenario of a typhoon. , and by using the generated observation-based virtual typhoon scenario as input data to the ADCIRC storm surge model to calculate the surge height at the coast, it is possible to predict the surge height at the coast due to the invasion of the typhoon.

Hereinafter, referring to FIG. 6, a method for predicting a typhoon surge height using a deep learning model according to an embodiment of the present invention is as follows.

First, the scenario generation unit 100 generates virtual typhoon information based on typhoon data pre-stored in the database through TCRM (S602).

Next, the tidal wave generating unit 200 generates tidal wave height information at the coast based on virtual typhoon information through the ADCIRC storm surge model (S604).

Subsequently, the learning unit 300 inputs the tsunami height information as an input to the LSTM artificial neural network to generate learned tsunami height prediction data for performing deep learning (S606).

Next, the application unit 400 generates improved tidal wave height prediction data by matching the tidal wave height prediction data with the tidal wave height data observed during an actual typhoon attack (S608).

Subsequently, the tidal height prediction unit 500 inputs the tidal height information generated based on the improved tidal height prediction data in the event of an actual typhoon attack to the learning unit 300 (S610).

Then, the tsunami height prediction unit 500 outputs the tsunami height prediction information generated through the deep learning of the learning unit 300 (S612).

Hereinafter, referring to FIG. 7, the detailed process of step S606 of the method for predicting the height of a typhoon surge using a deep learning model according to an embodiment of the present invention is as follows.

After step S604, the learning unit 300 inputs the tidal height information received from the tidal height generating unit 200 into the LSTM artificial neural network (S702).

Subsequently, the learning unit 300 performs deep learning learning so that the tidal height information input to the LSTM artificial neural network corresponds to the set category (S704).

And, the learning unit 300 outputs the tsunami height prediction data learned from the tsunami height information (S706).

Although the above has been described and illustrated in relation to preferred embodiments for illustrating the technical idea of the present invention, the present invention is not limited to the configuration and operation as shown and described in this way, and does not deviate from the scope of the technical idea. It will be readily apparent to those skilled in the art that many changes and modifications can be made to the present invention without Accordingly, all such appropriate alterations and modifications and equivalents should be regarded as falling within the scope of the present invention.

S: Typhoon surge height prediction system using deep learning model
100: scenario generator
102: indexing module
104: TCRM loading module
106: typhoon information generation module
200: Tsunami height generating unit
202: ADCIRC loading module
204: Tsunami height generation module
300: learning unit
302: input module
304: learning module
306: output module
400: application unit
500: tidal wave prediction unit

Claims (4)

a scenario generating unit that generates virtual typhoon information based on typhoon data pre-stored in a database through TCRM;
a tidal wave generating unit for generating tidal wave height information at a coast based on the virtual typhoon information through an ADCIRC storm surge model;
a learning unit that generates learned tidal height prediction data that performs deep learning by putting the generated tidal height information as an input of an artificial neural network;
an application unit that generates improved tidal wave height prediction data by matching the tidal wave height prediction data generated by the learning unit's deep learning with the tidal wave height data observed during an actual typhoon attack; and
Including a tidal height prediction unit that inputs tidal height information generated based on improved tidal height prediction data during an actual typhoon attack to the learning unit and outputs tidal height prediction information generated through deep learning of the learning unit,
The learning unit inputs the tsunami height information received from the tsunami height generation unit to the LSTM artificial neural network, the learning module performs deep learning learning so that the tsunami height information input to the LSTM artificial neural network corresponds to the set category, and the learning Including an output module that outputs the tidal height prediction data learned from the tidal height information by the module,
The category set during deep learning of the learning module includes any one of 'learning all typhoon information', 'learning strong typhoons', 'learning about typhoons by region', or 'learning a single variable',
The learning unit selects and learns only a specific typhoon from among the virtual typhoons of the scenario generator according to the category, and then first learning to predict the height of the tidal wave, or divides the affected sea area according to the typhoon path and selects only the typhoon that has passed through the sea area After learning, either the second learning to predict the height of the tidal wave or the third learning to predict the height of the tidal wave after learning the progress path excluding the maximum wind speed and the maximum wind radius and the central air pressure as input data Dip characterized in that Typhoon surge height prediction system using a learning model.
According to claim 1,
A variable setting unit for generating an RMW variable that replaces the variable for generating the tidal height information
Typhoon surge height prediction system using a deep learning model, characterized in that it further comprises.
According to claim 2,
The RMW variable is,
It is a variable that sets the maximum wind radius value, which is the distance to the point where the wind speed is strong, and is generated by [Equation 1]
Typhoon surge height prediction system using a deep learning model, characterized in that being.
[Equation 1]

delete

Images