KR101855652B1

KR101855652B1 - System of total cloud cover retrieval from satellite observation using machine learning and method thereof

Info

Publication number: KR101855652B1
Application number: KR1020180017469A
Authority: KR
Inventors: 이환우; 류근혁
Original assignee: 대한민국
Priority date: 2018-02-13
Filing date: 2018-02-13
Publication date: 2018-06-25

Abstract

In the satellite-based full-scale calculation system using the machine learning of the present invention, the satellite-based full-scale calculation archive server 20 using the machine learning uses the satellite- A preprocessing unit 110 for generating input data of a machine learning model in which cloud false pixels such as snowfall are classified using brightness temperature and reflectivity; A light determination unit 120 for sequentially driving a machine learning model trained for each cluster through the machine learning model input data created in the preprocessing unit 110 to perform a function of determining the total amount of light; And an image display and point-based tallying unit (130) for performing a function of displaying the result of the calculation of the traffic light by the lightness determining unit (120) as aggregate information for each image and point.

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a system for calculating a satellite-

The present invention relates to a system and a method for calculating an electric current by applying a sight observed on the ground using satellite observations to machine learning, and more particularly, to a system and method for estimating electric power by mitigating volatility using time- The present invention relates to a system and a method for calculating a satellite-based full-scale cloud computing system using machine learning,

The machine learning used in the invention randomly learns a plurality of decision trees through a random forest (RF) method to construct a plurality of decision trees to classify or predict when input data comes in. Here, the decision tree is an optional branch used to make a decision, and is a technique for dividing a complex problem into a hierarchical structure composed of simple problems. The most important characteristic of RF is that it consists of trees with slightly different characteristics due to randomness, which makes the predictions of each tree to be de-correlated and consequently improves the generalization performance.

Until now, the cloudiness has been determined through the eyes of the observer. The accuracy of the observed cloudiness is comparatively accurate, but it has the disadvantage of the limitation of the observation area of the point observation and the subjective decision reflecting the human psyche.

Patent Literature 1 discloses a method and system for calculating the total amount of light using RGB color of the sky image data of all sky. Patent Document 1 discloses an image processing apparatus including a video information receiving unit, a calculation reference setting unit, a shielding removal processing unit, a video classification processing unit, a pixel boundary value setting processing unit, a sunlight removal processing unit, a validity checking processing unit, And a cloud computing system. The system includes an image image collection step, a shielded image removal processing step, an image classification processing step, a pixel boundary value setting step, a solar light removal processing and validation step And a total cost calculation step. Patent Literature 1 discloses a technique of removing shielding from an image of a sky image RGB (Red Green Blue) image taken through a skyview, classification according to GBR (Green Blue Ratio) pixel distribution, cloud pixel classification considering RGB (Red Green Blue) Although it is configured to calculate sunlight by removing sunlight and validation in classified cloud pixels, observation data and observation area are limited because it utilizes images taken over the observatory with Skyview device, , The image obtained through the image image acquisition step is substituted into the shield removal processing unit provided in the electric-power calculation processing apparatus to restrict the shooting angle of the video image to the range of the sun's zenith angle of 75 ° or less and the viewing angle of 150 ° or less, The shielding image to be removed is limited.

In addition, since 2011, the computation and live observation of the Cheonryan satellite have been carried out through observation data of Cheollian satellite and the explicit algorithm. However, since the cloud cloud of Cheollian satellite is affected by the cloud data of Cheollian satellite, The accuracy of cloud determination is different. In addition, the algorithm uses the cloud shape ratio to calculate the electric current. Since the condition does not reflect the characteristics of the satellite observation data, The accuracy of 1 ~ 9% of the lightness in the light is not good. Since cloudiness is of great importance as forecast data for weather forecasts, forecasts for minimum temperature and radiation, and as a key climatic variable, an accurate and objective calculation of the electric power is necessary in a wide area.

KR 10-1709860 B1

An object of the present invention is to provide a system for providing precise weather information to a user and expressing a weather forecast by expressing it as a sky level (sunny, cloudy, cloudy, cloudy) using machine learning .

Another object of the present invention is to provide a method and system for estimating satellite cloud characteristics by using machine learning to solve the problem of the use of the preceding products and the problem of ineffective reflection of the satellite observation characteristics by the explicit condition (application of cloud shape, etc.) And to provide a satellite-based computation technique.

Yet another object of the present invention is to provide a satellite-based computation method using machine learning that solves limitations on point observation and subjective determination of the observed cloud.

Another object of the present invention is to provide a satellite-based computation method using machine learning using a sequential cluster model in order to increase the accuracy of the whole cloud determination.

In order to accomplish the above object, in a satellite-based full-scale calculation system using machine learning according to the present invention, a satellite-based full-scale calculation server using machine learning calculates a brightness A preprocessing unit which performs input and output of the input data of a machine learning model in which cloud mispredicted pixels such as snowfall are classified using temperature and reflectivity; A shade decision unit for sequentially driving a machine learning model trained by the clusters through the machine learning model input data created in the preprocessing unit to perform a function of determining the electric power; And an image display and point-by-point tallying unit for performing a function of displaying the result of the calculation of the traffic light by the lightness determining unit as aggregate information for each image and point.

Here, the machine learning is to randomly learn a plurality of decision trees in a random forest (RF) manner to form a plurality of decision trees to classify or predict when input data is received, May be the total cost calculation software through the machine learning model utilizing the observational satellite Himawari-8 meteorological satellite observation data.

Here, the brightness determination unit may use a brightness temperature difference (BTD) of the infrared region channels observed in the Himawari-8 weather satellite of the observation satellite.

Here, the preprocessor may divide the snowfall and the cloud area with a 1.6 탆 reflectance through the observation value of the Himawari-8 weather satellite of the observation satellite.

Here, the pre-processing unit may use a brightness temperature difference of 3.9 μm and 11.2 μm at night through observation values of the Himawari-8 weather satellite of the observation satellite.

Here, the preprocessing unit may use Equation 1, which is a normalized reflectivity conversion formula.

Here, the pre-processing unit may use Equation (2), which is a normalization operator presented by Li and Shibata.

Here, the parameters used in the RF model training are largely the combination of the average infrared channel brightness temperature difference (18x18 km ² ) and the combination of 3.9 μm in the 18 × 18 km ² region and the window channel (10.4 μm, 11.2 μm, 12.3 [micro] m) may be one that uses the specific pixel number category specified by the brightness temperature difference condition.

Here, the number of pixels specified by the brightness temperature difference condition of 3.9 탆 within the variable category 18 × 18 km ² and the window channels (10.4 袖 m, 11.2 쨉 m, 12.3 쨉 m) in the variable category adjusts the total light determination accuracy And may be a factor used to distinguish undifferentiated snow in the preprocessing process.

In order to accomplish another object of the present invention, there is provided a satellite-based total-cost calculation method using machine learning, comprising the steps of: (a) Performing a function of creating input data of a machine learning model in which cloud false pixels such as snowfall are classified using the brightness temperature and the reflectivity transmitted from the observation satellite; (b) performing a function of sequentially driving a machine learning model trained by the clusters through the machine learning model input data created in the preprocessing unit by using the calculator for calculating the amount of cloud, And (c) performing a function of displaying the total lightness result calculated by the lightness determination unit by the image display and the point-by-point aggregate calculation software using image and point-by-point aggregate information. . &Lt; / RTI >

The step (b) may further include the step of classifying the total amount of the cloud decision unit into a clear, cloudy or cloudy cloudy level.

The step (b) may further include the step of classifying the amount of the cloudiness to 0 or 1 to 5, and the step of classifying the cloudiness to 1 or 2 or 3 or 4 or 5 degrees .

Here, the step (b) may further include the step of classifying the light quantity of the light quantity determination unit 6 to 8 or 9 to 10.

Here, the step (b) may further include the step of classifying the light quantity into 6, 7, or 8 halo.

The step (b) may further include the step of classifying the light quantity of the light quantity determination unit as 9 or 10.

In order to achieve another object of the present invention, a satellite-based computerized calculation method using machine learning may be implemented by a computer program stored in a computer-readable recording medium.

In order to achieve the above and other objects of the present invention, there is provided a satellite-based computer-readable storage medium storing a computer-readable recording medium storing a program.

Specific details of other embodiments are included in the " Detailed Description of the Invention "and the accompanying drawings.

BRIEF DESCRIPTION OF THE DRAWINGS The advantages and / or features of the present invention and the manner of achieving them will be apparent by reference to various embodiments described in detail below with reference to the accompanying drawings.

However, the present invention is not limited to the configurations of the embodiments described below, but may be embodied in various other forms, and each embodiment disclosed in this specification is intended to be illustrative only, It will be understood by those of ordinary skill in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims.

According to the present invention, the user can calculate and use the electric information about the desired area more precisely and conveniently by calculating the objective electric current data at the ground level and expressing the image based on the sky state And the information can be used for weather forecasting.

In addition, machine learning can search for features that are difficult to define with explicit conditions as part of artificial intelligence, and can apply the searched features to the analysis target prediction.

In addition, it is expected that it will be useful for automation of cloud observation by a method that is specialized for electric cloud observation.

It is also expected to be useful for forecasting weather forecasts nationwide or regionally because it can be applied as an input data to the short - term predictive numerical model.

In addition, because the weather forecast is an important factor for forecasting weather and climate data, many applications are expected and a large reduction of input budget and human resources related to terrestrial weather observation is expected.

1 is a block diagram illustrating a satellite-based total-cost calculation system using the machine learning of the present invention.
FIG. 2 is a configuration diagram of the satellite-based full-scale calculation archive server of FIG. 1. FIG.
Fig. 3 is a graph showing the frequency of the clouding of the eyes. Fig.
Fig. 4 is a schematic diagram showing the criterion and classification of the sky status classification.
FIG. 5 (a) is a graph showing a space-averaged distribution of brightness temperature differences among infrared channels of channels 14 to 16 according to the entire traffic class, FIG. 5 (b) FIG. 2 is a graph showing a distribution of brightness temperature differences among infrared channels. FIG.
6 (b) is a photograph of brightness temperature difference between channel 15 and channel 16 in the snow and cloud region, and FIG. 6 (c) This is a picture of channel 5.
Fig. 7 is a graph showing the result of the calculation of the total amount of air.
FIG. 8 is a diagram showing a physical analysis of a color of a final image. FIG.
Fig. 9 (a) is a chart for each point, and Fig. 9 (b) is a time series chart graph.
FIG. 10 is a diagram illustrating an example of a software implementation of the electric vehicle calculation system according to the present invention.
11 is a schematic diagram showing a satellite-based total-cost calculation method using machine learning using a satellite-based total-cost calculation system using the machine learning of the present invention.
FIG. 12 is an operation diagram illustrating a satellite-based electric-power calculation method using machine learning using a satellite-based electric-power calculation system using the machine learning of FIG.
FIG. 13 is a flowchart illustrating a satellite-based total-cost calculation method using machine learning using the satellite-based total-cost calculation system using the machine learning of FIG.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings.

Before describing the present invention in detail, terms and words used herein should not be construed as being unconditionally limited in a conventional or dictionary sense, and the inventor of the present invention should not be interpreted in the best way It is to be understood that the concepts of various terms can be properly defined and used, and further, these terms and words should be interpreted in terms of meaning and concept consistent with the technical idea of the present invention.

That is, the terms used herein are used only to describe preferred embodiments of the present invention, and are not intended to specifically limit the contents of the present invention, It should be noted that this is a defined term.

Furthermore, in this specification, the singular forms "a", "an", and "the" include plural referents unless the context clearly dictates otherwise, and it should be understood that they may include singular do.

Where an element is referred to as "comprising" another element throughout this specification, the term " comprises " does not exclude any other element, It can mean that you can do it.

Further, when it is stated that an element is "inside or connected to" another element, the element may be directly connected to or in contact with the other element, A third component or means for fixing or connecting the component to another component may be present when the component is spaced apart from the first component by a predetermined distance, It should be noted that the description of the components or means of 3 may be omitted.

On the other hand, it should be understood that there is no third component or means when an element is described as being "directly connected" or "directly connected" to another element.

Likewise, other expressions that describe the relationship between the components, such as "between" and "immediately", or "neighboring to" and "directly adjacent to" .

In this specification, terms such as "one side", "other side", "one side", "other side", "first", "second" Is used to clearly distinguish one element from another element, and it should be understood that the meaning of the element is not limited by such term.

It is also to be understood that terms related to positions such as "top", "bottom", "left", "right" in this specification are used to indicate relative positions in the drawing, Unless an absolute position is specified for these positions, it should not be understood that these position-related terms refer to absolute positions.

Furthermore, in the specification of the present invention, the terms "part", "unit", "module", "device" and the like mean units capable of handling one or more functions or operations, , Or a combination of hardware and software.

In this specification, the same reference numerals are used for the respective components of the drawings to denote the same reference numerals even though they are shown in different drawings, that is, the same reference numerals throughout the specification The symbols indicate the same components.

In the drawings attached to the present specification, the size, position, coupling relationship, and the like of each constituent element of the present invention may be partially or exaggerated or omitted or omitted for the sake of clarity of description of the present invention or for convenience of explanation May be described, and therefore the proportion or scale may not be rigorous.

Further, in the following description of the present invention, a detailed description of a configuration that is considered to be unnecessarily blurring the gist of the present invention, for example, a known technology including the prior art may be omitted.

1 is a block diagram illustrating a satellite-based total-cost calculation system using the machine learning of the present invention.

FIG. 2 is a configuration diagram of the satellite-based full-scale calculation archive server of FIG. 1. FIG.

Fig. 3 is a graph showing the frequency of the clouding of the eyes. Fig.

Fig. 4 is a schematic diagram showing the criterion and classification of the sky status classification.

FIG. 5 (a) is a graph showing a space-averaged distribution of brightness temperature differences among infrared channels of channels 14 to 16 according to the entire traffic class, FIG. 5 (b) FIG. 2 is a graph showing a distribution of brightness temperature differences among infrared channels. FIG.

6 (b) is a photograph of brightness temperature difference between channel 15 and channel 16 in the snow and cloud region, and FIG. 6 (c) This is a picture of channel 5.

Fig. 7 is a graph showing the result of the calculation of the total amount of air.

FIG. 8 is a diagram showing a physical analysis of a color of a final image. FIG.

Fig. 9 (a) is a chart for each point, and Fig. 9 (b) is a time series chart graph.

FIG. 10 is a diagram illustrating an example of a software implementation of the electric vehicle calculation system according to the present invention.

11 is a schematic diagram showing a satellite-based total-cost calculation method using machine learning using a satellite-based total-cost calculation system using the machine learning of the present invention.

FIG. 12 is an operation diagram illustrating a satellite-based electric-power calculation method using machine learning using a satellite-based electric-power calculation system using the machine learning of FIG.

FIG. 13 is a flowchart illustrating a satellite-based total-cost calculation method using machine learning using the satellite-based total-cost calculation system using the machine learning of FIG.

In the satellite-based full-scale calculation system using the machine learning of the present invention, the satellite-based full-traffic calculation archive server 20 uses the machine learning to calculate the brightness A preprocessing unit 110 for generating input data of a machine learning model in which cloud mispredicted pixels such as snowfall are classified using temperature and reflectivity; A light determination unit 120 for sequentially driving a machine learning model trained for each cluster through the machine learning model input data created in the preprocessing unit 110 to perform a function of determining the total amount of light; And an image display and point-based tallying unit (130) for performing a function of displaying the result of the calculation of the traffic light by the lightness determining unit (120) as aggregate information for each image and point.

Referring to FIG. 1, a configuration diagram of a satellite-based total-cost calculation system using machine learning of the present invention is disclosed.

One embodiment of the satellite-based electric-power calculation system using the machine learning of the present invention is a satellite data broadcasting service HimawariCast through communication satellite 16 and a Himawari cloud service providing data through the Internet through a cloud server (18).

Currently, the Japan Weather Satellite Center (MSC) is operating the Himawari cast by Himawari satellite data reception service. Direct transmission of satellite images to High Rate Information Transmission (HRIT) and Low Rate Information Transmission (LRIT) receiving stations (MDUS and SDUS) that were formerly Himawari 6 (MTSAT-1R) Service was completed in November 2015, and data transmission service via communication satellite was started in January 2015 instead. In Himawari No. 8, the observation frequency is greatly improved compared with Himawari No. 7 (MTSAT-2), and the light bulb is observed for 10 minutes, so the satellite image is transmitted every 10 minutes at Himawari cast.

Himawari cast began transmitting satellite images of Himawari 8 from 11 o'clock on July 3, 2015 (Japan time) before the satellite 10 changed to Himawari 8. The communication satellite 16 used in the Himawari cast was scheduled to use JCSAT-2A (JCSAT-8) satellite at the start of service and JCSAT-2B (JCSAT-14) from 4th quarter of 2015, but was postponed for about six months. Currently, satellite images provided by Himawari cast transmit data in the light area in a format compatible with the HRIT / LRIT service of Himawari No. 6 and No. 8. If the satellite being operated is Himawari 8, it is transmitted at intervals of 10 minutes, but if Himawari 7 is in operation, it is transmitted according to the observation schedule. The HRIT file contains 5 channels for all observation channels during operation of Himawari No. 7, and 14 channels (visible 1 and infrared 13) of 16 channels for all observation channels when operating the Himawari No. 8. ). In addition, Himawari No. 8 transmits only high-resolution infrared image channels at night for efficient transmission.

The SATAID format of the Himawari cast is used primarily to transmit numerical forecast grid point values (GPV) and observational data to meteorological agencies in developing countries. These data can be superimposed on satellite images with SATAID software.

(DVB-S2 (Digital Video Broadcasting-Satellite-Second Generation) receiver 28, a low noise block downconverter (LNB) 30 is required, and a countermeasure such as inserting a band-pass filter (BPF) is required when there is interference between the ground microsystem and the fourth generation transmission communication system (IMT-Advanced system) Do. The service through the receiver can receive the satellite data to the C-band antenna 32 without any restriction on the ground through the receiver and some equipment even when the terrestrial circuit environment is poor.

In the case where the terrestrial circuit environment is not good as in a developing country, the Himawari cast service is used via the communication satellite 16 via the CTS operator 14, but in other cases, data can be provided through the Himawari cloud. The Himawari Cloud is able to receive full-range observations of 16 channels (3 visible and 13 infrared) at native resolutions (0.5 ~ 1km of visible and 2km of infrared) through satellite registration. In addition, the Himawari cloud provides a web service based on the HTTP 1.1 transmission specification, and a client must guarantee a transmission speed of at least 25 Mbps in order to receive the bulb image. The Japan Meteorological Agency (12) plans various service distribution methods according to the purpose of use. The overseas weather organizations will be provided directly to the Himawari cloud service (18) at the Japan Meteorological Agency (12), and the data will be obtained from DIAS, NICT, JAXA and Chiba University through the Himawari Cloud Service (18). In case of Information and Communication Research Organization (NICT), real time data of Himawari 8 is released through FTP server. To use the service, you can access it by using the account and password that is issued after you apply for the account, and access in the form of browser-based access, command line base access, FTP client application (WinsCP) access. OS (Operating System) is available for Windows, Mac, and Linux. According to the user 24, a real-time Web is used when the user is a general person. The latest public image (Japanese region image, hemisphere image) provided by the Meteorological Agency can be viewed in various environments such as a smart phone, a tablet or a PC. You can also browse past weather event videos.

In the embodiment of the satellite-based electric-power calculation system using the machine learning of the present invention, the Himawari satellite data archive server 20 is used for the researcher and the general public through the real-time web and the WSDBank Web application developed in the NICT Science cloud It is possible to acquire data of a certain period of time. The target data are Himawari no. 1 to 7, GOES-9, and Himawari no. 8 (however, in case of Himawari no. 8 data, 24 hours elapsed data is added). For researchers, real-time data and historical data of Himawari 8 can be received from NICT Science Cloud FTP server. The past data is the observation data after March 20, 2015, and the real time data is the data immediately after observation (about 10 minutes later). To use historical data, you need to apply for an account on the Web.

In addition, a time-series display web application can be used. NIKT Science Cloud's Scalable Time Series Data Disclosure Technology (STARStouch) enables you to view the Himawari satellite data retention period and visual images in a time series. Another providing site, DIAS, is the data integration and analysis system of Tokyo University, which is the main site for research use. It creates meta data in a uniform format for each data set and creates a document converted into readable formats (HTML and PDF) . The data transmitted through the cloud includes Himawari standard data (full disk, Japanese area, startup observation area), netCDF data (Japan area, startup observation area), color image data (full disk, Japan area, The service contents are different for the general public and the researcher such as NICT. In order to use the service for the researcher, the account acquisition and the procedure are necessary. With DIAS service, you can obtain the corrected image you need by setting the period, data format setting, observation area, RGB setting, gamma correction, mask overlay, channel and so on.

Such a Himawari cloud service (18) is a service in Japan. If you wish to provide Himawari data from abroad, it is a principle to ask each national weather agency, but direct acquisition from DIAS and NICT is also possible. In addition, the Japan Meteorological Agency (JMA) (12) on the means of using Himawari cloud service (18) does not distinguish between users assumed in NICT and DIAS, and it is decided according to user convenience only. For each site, basic images that can perform basic operations such as time designation and enlargement / reduction can be used only by URL access without login.

In addition, in real-time, it supports 30 languages in addition to displaying the current position using GPS function, and since it takes a URL preservation method for a specific period of video that a specific user was watching on the web, The user can also view the same image. In relation to the web, browsing information can be shared with various SNSs such as e-mail, Facebook, Twitter, LINE, etc., and the images can download the basic images and videos of the designated time through the WSDB (World Science Data Bank) . These images are used for individual users, research utilization and education, exhibition of science museum, magazine, newspaper, web exhibition, international event display. With the provision of Himawari data through such a web service, there is a one-stop service because of its use in the research community, sufficient data storage capacity, and other related scientific data and products.

In one embodiment of the satellite-based computation system using machine learning of the present invention, machine learning is introduced. This is to solve the problem of the use of the preceding products and the limitation of the problem of inefficient reflection of the satellite observation characteristics and calculation of the overaccumulation decision by the explicit condition (application of the cloud shape, etc.) in the Chunlian satellite satellite cloud algorithm. Machine learning is a part of artificial intelligence that can be searched for features that are difficult to define with explicit conditions and can be applied to explored analytical predictions.

Referring to FIG. 2, in the satellite-based total-power calculation system using the machine learning of the present invention, the satellite-based total-power calculation server 20 using the machine learning uses the satellite- A preprocessing unit 110 for generating input data of a machine learning model in which cloud mispredicted pixels such as snowfall are classified using the brightness temperature and reflectivity transmitted after observation; A light determination unit 120 for sequentially driving a machine learning model trained for each cluster through the machine learning model input data created in the preprocessing unit 110 to perform a function of determining the total amount of light; And an image display and point-by-point tallying unit (130) for performing the function of displaying the total lightness result calculated by the lightness determination unit (120) as aggregation information for each image and point.

The preprocessing unit 110 generates input data of a machine learning model in which cloud false pixels such as snowfall are classified using the brightness temperature and the reflectivity transmitted from the Himawari-8 satellite of the observation satellite 10 Function.

The lightness determining unit 120 sequentially drives the machine learning models trained by the clusters through the machine learning model input data created in the preprocessing unit 110, thereby performing the function of determining the total lightness.

The image display and point-by-point tallying unit 130 performs a function of displaying the result of the total calculation calculated by the lightness determination unit 120 as aggregation information for each image and point.

In the satellite-based full-scale calculation system using the machine learning of the present invention, a machine learning technique called Random Forest (hereinafter, referred to as RF) is applied and observation data of each channel of the Himawari- And used in the training parameters and the lightning determination input data of the RF model included in the light source determination unit 110 and the light amount determination unit 120.

Since the observing satellite 10 observes only the reflectance and the brightness temperature value, the level of the corresponding value is often similar even in other phenomena. However, since the recently launched Himawari-8 weather satellite in Japan observes 16 wavelengths, it can detect more meteorological phenomena than Chollian satellites. In other words, when preparing the input data of the trained RF model, it is possible to improve the accuracy of the determination of the lightness of the RF model by performing the preprocessing process of distinguishing pixels that can be mistaken as clouds. Table 1 summarizes the boundary conditions and boundary values of the preprocessing unit 110. The boundary conditions shown in Table 1 refer to the Himawari-8 fog detection algorithm developed by the National Weather Service Center of Japan Meteorological Agency. The 1.6 ㎛ reflectivity observed by Himawari-8 meteorological satellite is theoretically and empirically distinguishable because it has high reflectivity for clouds and low reflectivity for snow areas. Also, since the reflectivity changes according to the solar zenith angle in view of the reflectivity observation characteristic, the following normalized reflectivity conversion equation (1) is used for stable calculation. The normalization operator uses Equation 2 presented by Li and Shibata in 2006.

channel(

)
Day / Night Boundary value NIR (1.6)
weekly 33% SWIR (3.9) -IR (11.2)
Nighttime -1.8K to 0.7K IR (11.2) -IR (8.6)
common 1.5K IR (11.2) -IR (13.3)
common 8.5K

Since no 1.6 탆 reflectivity was observed at night, the brightness temperature difference of 3.9 탆 and 11.2 탆 was used. It is difficult to remove completely but most boundary conditions are used because it shows -1.8K ~ 0.7K in most snow areas. Since the Cirrus region can be removed when the normalized 1.6 ㎛ reflectivity and the brightness temperature difference of 3.9 urn and 11.2 탆 are applied, the brightness temperature difference between 11.2 탆 and 8.6 탆 and the brightness temperature difference between 11.2 탆 and 13.3 탆 Use to recover the cirrus area.

In an embodiment of the satellite-based computation system using the machine learning of the present invention, the ensemble technique of randomly learning a plurality of decision trees by machine learning applied to the Himawari-8 cloud is described in Random Forest (RF; Breiman, 2001 ; H2O.ai team, 2016) was used. The data applied to the training and verification of the RF model were measured at the brightness of 3.9 ㎛, 6.2 ㎛, 7.0 ㎛, 7.3 ㎛, 8.6 ㎛, 9.6 ㎛, 10.4 ㎛, 11.2 ㎛, 12.3 ㎛ and 13.3 ㎛, which were observed at Himawari - And Cloud Amount (CA) observed at the weather station. In general, 1, 5, and 9 are not declared well, so the RF model can be trained to a certain amount of cloudiness. To compensate for this, weighted average cloudiness (CA ') is applied as a target variable , And the calculation formula is as shown in Equation (3).

Here, h is on-time. However, RF model verification was performed at the time of day. The reason for limiting the RF model training to weekly cases is because the nighttime accuracy is low. The infrared channel brightness temperature difference (IRBTD) used as a cloud decision variable is a space-averaged value for a 9 × 9 pixel region corresponding to about 18 × 18 km ² which is a region of the eye. Also, we applied a condition variable (Count Value of BTD 3.9 μm-window) that utilizes difference of 3.9 μm and waiting window channel for determination error of snowing area and determination of cloudiness for nightly underwash.

In the training set, the frequency distribution of the observed cloud in 2016 shows an uneven distribution of 0 or 10% cloudiness, so we created an RF model for each cluster based on sky conditions. Himawari-8 cloudiness is calculated by preprocessing process (removal of cloud-error pixels, etc.) and trained six RF models, and the performance is calculated by comparing PC and Bias, RMSE, and Correlation with observed cloudiness. Bias and RMSE are closer to 0 and Correlation and PC (± 0 / ± 1 / ± 2) are closer to 1, which means better coincidence.

As a result of the Himawari-8 cloud output through the RF model, it shows generally high qualitative agreement with infrared images. As a result of applying the RF model for January 2017, 71.5% of the PC (difference between the satellite and the middles) ± 1, 87.2% of the PC ± 2 (about 90% of the PC ± 2 between the ground and the airbase) and RMSE of 1.74 , Corr. Is about 90%, which is better than most of the proved indicators for the same period.

Referring to FIG. 3, a satellite-based total cloud computing system using the machine learning of the present invention applies a sequential cluster model in order to increase the accuracy of the total cloud determination. As shown in FIG. 3, the whole-cloud observation data used as a target value of the machine learning model shows a relatively uneven distribution of the majority of the cloudiness of 0 and 10%. Since machine learning is dependent on the number of data of the target value, when the result is reflected in the machine learning without refining, the result is focused on 0 or 10% of the cloud, and the accuracy of determination for the cloud of 1 ~ 9 is weakened Lt; / RTI >

Referring to FIG. 4, a machine learning model for each cluster is sequentially generated in consideration of the number of class data of the whole traffic on the basis of the sky state prescribed by the Korea Meteorological Administration, and the system is constructed to improve the accuracy of the range of 1 to 9 hours.

The satellite-based computation system using the machine learning of the present invention uses a brightness temperature difference (BTD) of infrared region channels observed in a satellite. This is to solve the cloud decision subjectivity and the observation area limitation.

For example, the Himawari-8 weather satellite will be described as an example of a satellite-based total-power calculation system using the machine learning of the present invention. The Himawari-8 meteorological satellite observation data can observe the full disk area in real time, so it can solve the area limitation of the spot observation. And 16 channels (0.46, 0.51, 0.64, 0.86, 1.6, 2.3, 3.9, 6.2, 7.0, 7.3, 8.6, 9.6, 12.3 ㎛, and 13.3 ㎛) have intrinsic physical characteristics for each wavelength band. In general, the temperature difference between infrared channels is used to minimize the seasonal satellite observation variability and determine the specific weather phenomenon.

FIG. 5 (a) is a graph showing a spatial average distribution of brightness temperature differences between channels 14 (11.2 탆) to 16 (13.3 탆) for the entire traffic class, and FIG. 5 (b) Averaged infrared channel brightness temperature difference distribution of channel 13 (10.4 占 퐉) to channel 14 (11.2 占 퐉).

5 (a) and 5 (b), since a specific tendency is shown in the brightness-temperature difference between the infrared channels of the space-averaged region in the same region as the cloud resolution of each cloud class, a machine learning model It can be used as a training variable and it can guarantee the objectivity of the decision of the traffic.

The satellite-based full-scale calculation system using the machine learning of the present invention applies a preprocessing process for distinguishing pixels that can be mistakenly mistaken as clouds in order to prevent erroneous determination. For example, the Himawari-8 weather satellite will be described as an example of a satellite-based total-power calculation system using the machine learning of the present invention. Generally, because the brightness temperature differences between infrared channels are similar for certain weather phenomena (eg, clouds, snow), the erroneous results can be calculated from the full-range decision of the machine learning model.

6 (b) is a photograph of the brightness temperature difference between the channel 15 (11.2 m) and the channel 16 (13.3 m) in the snowy and rolling region, and Fig. 6 c) is a photograph of channel 5 (1.6 탆) in the snow and cloud areas.

6 (a) and 6 (b), it is assumed that the snowfall (dark blue sky area in FIG. 6 (a)) and the cloud image (white area in FIG. 6 Though the brightness temperature difference between channels is generally similar, it is possible to distinguish snow and cloud areas with 1.6 ㎛ reflectivity. In other words, if the machine learning model is applied to distinguish parts that can be mistaken as cloud through the observations of Himawari-8 weather satellite, the accuracy of the calculation results of the whole clouds can be increased. The boundary value of the preprocessing process is based on the boundary values of the Himawari-8 fog detection products of the National Weather Satellite Center of Japan, as shown in Table 1.

Referring to FIG. 7, the comparison between the result of the calculation and the result of the calculation is shown in FIG. 7. As a general rule, the tendency of the calculation before the calculation is in good agreement with the tendency of the observation, and the accuracy of the calculation is improved.

8, the satellite based around the cloud cover on the basis of 18 x 18 km ² area on each pixel in the observed result of the entire cloud cover image of the calculation system white the cloudiness to 0-2, light aqua using a machine learning of the present invention It means clear, clear, cloudy, cloudy, cloudy, and cloudy, respectively, with a range of 3 ~ . The interpretation of these colors is based on the sky conditions prescribed by the Korea Meteorological Administration.

9 (a) and 9 (b), an aggregate table and a time series chart, which are aggregated on a point-by-point basis at intervals of one hour, are used to calculate the total cloudiness results calculated by the satellite-based cloud computing system using the machine learning of the present invention .

This is to support live weather monitoring and weather forecasting for the branch office, and it is possible to add branches and adjust the aggregation time interval. In addition, it is possible to store the traffic data at each point in a specific time range.

Referring to FIG. 10, the satellite-based total-cost calculation method using the machine learning of the present invention includes steps S100 to S600.

Step S100 is a step of observing the atmospheric state of the earth at the observing satellite 10 of the satellite-based computation system using the machine learning.

The step S200 is a step of producing the data observed in the step S100 by the satellite-based computerized calculation system using the machine learning.

In step S300, the image of the data produced in step S200 is processed by the satellite-based computerized calculation system using the machine learning.

In step S400, the image processed in step S300 is processed by the satellite-based computation system using machine learning using the preprocessing input data, auxiliary data, and sensor response functions.

Step S500 is a step of verifying the data processed in step S400 with various weather observation data.

The step S600 is a step of calculating, distributing and storing the data obtained in step S500 by the satellite-based computation system using machine learning.

Referring to FIG. 11, the satellite-based total-cost calculation method using the machine learning of the present invention includes steps S1000 to S3000.

In step S1000, the satellite-based total-cost calculation method using the machine learning of the present invention is performed in the satellite-based total-cost calculation archive server 20 by (a) the preprocessing unit 110, using a machine learning model using meteorological satellite observation data, This is the step of creating the input data of the machine learning model in which the clouds are misidentified pixels such as snowfall using the brightness temperature and the reflectance, which are observed and then transmitted from the meteorological satellite, using the computation software.

In step S2000, the satellite-based full-scale calculation method using the machine learning of the present invention is performed in the satellite-based full-scale calculation archive server 20, (b) the cloud determination unit 120 calculates a machine learning model using meteorological satellite observation data And a step of sequentially driving the machine learning models trained by the clusters through the machine learning model input data created in the preprocessing unit 110 by using the total amount calculation software to perform the function of determining the total load.

In step S3000, the satellite-based full-scale calculation method using the machine learning of the present invention is performed in the satellite-based full-scale calculation archive server 20, (c) the image display and point- And a step of displaying the result of the calculation of the total light amount by the light amount determination unit 120 using the total light amount calculation software through the machine learning model as aggregated information for each image and point.

The above-described machine learning is to randomly learn a plurality of decision trees in a random forest (RF) manner to form a plurality of decision trees to classify or predict when input data is received. Himawari-8 It can be a total-cost calculation software through a machine learning model using meteorological satellite observation data.

The operation of the satellite-based full-scale calculation method using the machine learning using the satellite-based full-scale system using the machine learning of the present invention will be described with reference to the drawings.

Referring to FIG. 12, an operation method of the cloud determination unit 120, which is a process for sequentially determining the cloudiness through the RF model trained for each cluster, using the Himawari-8 weather satellite observation data, is described in detail.

The total number of RF models trained by each cluster shown in Fig. 12 is six, and each model is trained in consideration of the unevenness of the number of data of the class of the target cloud, which is the target value, and the sky condition prescribed by the Korea Meteorological Administration. RF1 models are 0 to 5 and 6 to 10, RF2 models are 0 and 1 to 5, RF3 models are 6 to 8 and 9 to 10, RF4 models are 1, 2, 3, 4 The RF5 model was trained as an eigenvalue assigned to a cluster divided into 6, 7, 8, and RF6 models divided into 9, 10, and 10 halos.

The parameters used in each RF model training are largely the spatial temperature (18x18 km ² ) average of the combination of the infrared channel-to-channel brightness temperature difference and the brightness temperature of 3.9 μm in the 18 × 18 km ² region and the air window channels (10.4 μm, 11.2 μm, 12.3 μm) The specific pixel count category specified by the difference condition was used. As mentioned above, since the brightness-temperature difference between the spatial averaged infrared channels was specific for each cloud class, it was used as a training variable. However, the combination of brightness temperature difference variables related to 13.3 ㎛ was excluded because the importance of the variables in the model was high, but it was sensitive to the snowy area and recognized as cloud. Also, the combinations of the water vapor channels 6.2 μm, 7.0 μm, and 7.3 μm were used only for the RF models that determine the 6 to 10% cloudiness, because the importance was low.

The specific number of pixels defined by the brightness temperature difference condition of 3.9 μm in the second variable category 18 × 18 km ² and the atmosphere window channels (10.4 μm, 11.2 μm, and 12.3 μm) increases the accuracy of the total light determination for nighttime subgrade, Is a factor used to distinguish snowflakes that are not distinguished from snowflakes. The brightness temperature difference has the following empirical characteristics and is used as a training variable by counting the number of pixels based on this characteristic.

Snow and sunshine: -3K <corresponding brightness temperature difference <3K (night) or 10K (weekly)

Night Fog and Downstairs: The corresponding brightness temperature difference <-3K

Daytime Cloudy Clouds Including: 10K <corresponding brightness temperature difference

Table 2 shows the electric conductivity verification index, which is the result of the verification between the electric power calculation system and the observation light. Because of the subjectivity of the observer, the accuracy of the measurement is often 87 ± 90%, which is mainly determined by the PC ± 2 (± 2 accuracy between the eye and the pre-calculation). This is similar to the PC ± 2 result (over 85%) between the terrestrial weather station and the aeronautical observatory, and the calculation result of the present invention can be judged to be a human observation level.

PC ± 0 PC ± 1 PC ± 2 Mean Bias RMSE Corr. 2015.9-12. 43.9% 74.5% 89.3% -0.05 1.58 91% 2017.1. 43.5% 71.5% 87.2% -0.22 1.74 90% 2017.7. 41.4% 71.2% 87.2% 0.44 1.79 82%

Referring to FIG. 13, there is shown an example of software for implementing a satellite-based total-cost calculating method using machine learning of a total-amount calculation system through a machine learning model using Himawari-8 meteorological satellite observation data according to the present invention have.

The satellite-based traffic calculation method using the machine learning of the present invention can be implemented by a computer program stored in a computer-readable recording medium.

The computer program of the satellite-based network calculation method using the machine learning of the present invention can be stored and implemented in a computer-readable recording medium.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is evident that many alternatives, modifications and variations will be apparent to those skilled in the art. It will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims.

In addition, since the present invention can be embodied in various other forms, the present invention is not limited by the above description, and the above description is intended to be a complete description of the present invention, It will be understood by those of ordinary skill in the art that the present invention is only provided to fully inform the person skilled in the art of the scope of the present invention and that the present invention is only defined by the claims of the claims.

10: Observation satellite (meteorological satellite)
12: Meteorological Office
14: CTS operator
16: Communication satellite
18: Cloud Services
20: Satellite-based network calculation server
22: NMHSs
24: User
26: PC with software
28: DVB-S2 Receiver
30: LNB
32: C-band antenna
110:
120:
130: Video presentation and aggregation by branch

Claims

1. A satellite-based electric-power calculation system using machine learning,
The satellite-based full-traffic calculation archive server 20,
Using the brightness and temperature calculations of the observed satellite 10 and the reflectivity, the computer 10 performs the function of creating the input data of the machine learning model in which cloud false pixels such as snowfall are classified A preprocessing unit 110;
A lightness determining unit (120) that performs a function of sequentially driving a machine learning model trained by the clusters through the machine learning model input data created in the preprocessing unit (110) by using the total light amount calculating software to determine the total light amount; And
And an image display and point-by-point tallying unit (130) for performing the function of displaying the total traffic result calculated by the lightness determination unit (120) by the image and point-by-point aggregation information by using the total lightness calculation software ,
A satellite - based computation system using machine learning.

The method according to claim 1,
The machine learning is to randomly learn a plurality of decision trees through a random forest (RF) method to construct a plurality of decision trees to classify or predict when input data is received.
Characterized in that the electric-power calculation software is electric-power calculation software through a machine-learning model using the Himawari-8 weather satellite observation data of the observation satellite (10)
A satellite - based computation system using machine learning.

The method according to claim 1,
Wherein the brightness determining unit 120 uses a brightness temperature difference (BTD) of infrared region channels observed from a Himawari-8 weather satellite of the observation satellite 10. [
A satellite - based computation system using machine learning.

The method of claim 3,
Wherein the preprocessing unit 110 divides snowfall and cloud regions with a 1.6 占 퐉 reflection rate through observations of the Himawari-8 weather satellite of the observation satellite 10. [
A satellite - based computation system using machine learning.

The method of claim 3,
Wherein the preprocessing unit (110) uses a brightness temperature difference of 3.9 탆 and 11.2 탆 at night through observation values of Himawari-8 weather satellite of the observation satellite (10)
A satellite - based computation system using machine learning.

The method according to claim 1,
The preprocessing unit 110 uses Equation 1, which is a normalized reflectance conversion formula.
A satellite - based computation system using machine learning.
(1)

The method according to claim 6,
The preprocessing unit 110 uses Equation 2, which is a normalization operator given by Li and Shibata.
A satellite - based computation system using machine learning.
(2)

The method according to claim 1,
The lightness determining unit 120
The parameters used in each RF model training are largely the spatial temperature (18x18 km ² ) average of the combination of the infrared channel-to-channel brightness temperature difference and the brightness temperature of 3.9 μm in the 18 × 18 km ² region and the air window channels (10.4 μm, 11.2 μm, 12.3 μm) And the specific pixel number category specified by the difference condition is used.
A satellite - based computation system using machine learning.

The method according to claim 1,
The lightness determining unit 120
The specific number of pixels defined by the brightness temperature difference condition of 3.9 μm in the 18 × 18 km ² variable area and the window channels (10.4 μm, 11.2 μm, and 12.3 μm) in the variable category increases the accuracy of the total light determination for the nighttime subgrade, Which is a factor used to distinguish between snow and snow,
A satellite - based computation system using machine learning.

A method for calculating a satellite-based electric vehicle using machine learning,
In the satellite-based full-traffic calculation archive server 20
(a) The preprocessing unit 110 uses the brightness temperature and the reflectivity transmitted from the meteorological satellite using the computation software through a machine learning model using meteorological satellite observation data to calculate cloudiness such as snowfall Performing a function of creating input data of a machine learning model in which pixels are classified;
(b) The cloud determination unit 120 acquires the machine learning model training data on the basis of the machine learning model input data created in the preprocessing unit 110 by using the computerized learning model software using the meteorological satellite observation data, Performing a function of sequentially driving the model to determine the electric current; And
(c) The image display and point-by-point tallying unit 130 calculates and displays the result of the calculation of the lightness calculated by the lightness determination unit 120 using the computerized calculation software through the machine learning model using the weather satellite observation data, Performing a function of displaying the classified aggregate information; / RTI >
Computation of Satellite - based Electricity Using Machine Learning.

11. The method of claim 10,
The machine learning is to randomly learn a plurality of decision trees through a random forest (RF) method to construct a plurality of decision trees to classify or predict when input data is received.
Wherein said electric power calculation software is electric power calculation software through a machine learning model using Himawari-8 weather satellite observation data,
Computation of Satellite - based Electricity Using Machine Learning.

11. The method of claim 10,
The step (b)
Further comprising the step of classifying the traffic light into fine clouds, cloudy clouds or cloudy clouds,
Computation of Satellite - based Electricity Using Machine Learning.

13. The method of claim 12,
The step (b)
The lightness determining unit 120 classifies the electric current as 0 or 1 to 5 halo; And
Further comprising the step of causing the light quantity determination unit (120) to classify the electric quantity as 1 or 2 or 3 or 4 or 5,
Computation of Satellite - based Electricity Using Machine Learning.

13. The method of claim 12,
The step (b)
The light quantity determining unit 120 classifies the electric quantity as 6 to 8 or 9 to 10; &Lt; / RTI >
Computation of Satellite - based Electricity Using Machine Learning.

15. The method of claim 14,
The step (b)
The light quantity determining unit 120 classifies the electric wave as 6, 7, or 8 halo; &Lt; / RTI >
Computation of Satellite - based Electricity Using Machine Learning.

15. The method of claim 14,
The step (b)
Further comprising the step of the light quantity determining unit (120) classifying the electric wave as 9 or 10 halo.
Computation of Satellite - based Electricity Using Machine Learning.

A computer program stored in a computer-readable medium for implementing the method of any one of claims 10 to 16.

17. A computer-readable recording medium on which a program for implementing the method of any one of claims 10 to 16 is stored.