KR101926544B1 - Method, apparatus and computer program for meteorological observation using weather research forecasting model - Google Patents

Abstract

The present invention relates to a method, a system, and a computer program for meteorological observation. The meteorological observation method using a meteorological prediction model includes the steps of: receiving a predicted value at a specific time point provided by the meteorological prediction model from a server; receiving an actual value at the specific time point in an observation instrument; determining whether the difference between the predicted and actual values is equal to or greater than a preset reference value, determining whether or not the actual value is a value observed in a normal condition when the difference is equal to or greater than the reference value as a result of the determination; and correcting the actual value or the predicted value in accordance with the determination result. According to the present invention, it is possible to improve the accuracy of meteorological observation by using, for example, the meteorological prediction model and building thermal image information.

Description

[0001] METHOD, APPARATUS AND COMPUTER PROGRAM FOR METAOROLOGICAL OBSERVATION USING WEATHER RESEARCH FORECASTING MODEL [0002]

The present invention relates to a weather observation method, a system, and a computer program, and more particularly, to a weather observation method, a system, and a computer program using a weather prediction model.

Climate phenomena due to global warming and environmental pollution are emerging as a global problem, and the uncertainty of the weather prediction is also increasing with the increase of the weather change.

The current weather prediction method uses a numerical forecasting model computed on a high-performance computer to predict future weather from the current atmospheric state.

Atmospheric motion is governed by the laws of conservation of various trace gases such as energy conservation laws, momentum conservation laws, mass conservation laws, water vapor and carbon dioxide, and natural laws such as ideal gas equations. This group of equations is called the governing equation system. Weather forecasting is the process of obtaining the solution of this governing equations. Since the conservation laws are expressed in the same mathematical form as the partial differential equations, the governing equation system takes the form of a system of partial differential equations. The problem is how to solve the solution of this simultaneous equations. Since the differential equations that humans can solve are extremely limited, there is no way to solve this complicated governing equations. However, instead of solving this equation (exact solution), we can approximate the solution of this governing equations by temporally and spatially latticing the meteorological variables such as the temperature, wind and pressure to be obtained. We call this approximate solution a numerical value, which is a numerical forecasting data. Numerical calculations are needed to obtain the numerical solution. The numerical prediction model is called a numerical example and the code that enables the computer to recognize and operate the computation of the governing equation system is referred to as a numerical example. The smaller the lattice spacing, the closer to the analytical solution. Therefore, advanced computing devices such as supercomputers are used to reduce the time and spatial lattice spacing and to perform many calculations in a short time.

Numerical forecasting is an understanding of nature such as meteorological observations, remote sensing, high-speed communication, data assimilation, supercomputer, computational science, programming and graphic processing, and the determination of advanced science and IT technology. And is developing through convergence and exchange.

Meteorological observations are made through various methods such as ground observation, high-rise observations, weather radar observations, meteorological satellite observations, marine meteorological observations using vessels or buoys, and aerial observations. All of the data obtained here are digitized and used as input data for numerical forecasting models. In Korea, the ratio of satellite observations to initial values is about 80%, and the rest is used as other observation data.

Most of the observations are integrated into the central server of the Meteorological Agency at a specific point in time, and these data are shared with over 180 countries through the Global Weather Network (GTS), and at the same time we use observed weather data from these countries. Observations are entered as initial data for numerical forecasting.

Various observation equipments are used for collecting observation data. Examples of observation equipments include automatic weather observation equipment, mobile weather observation equipment, aeronautical observation equipment, high-rise weather observation equipment, and environmental measurement equipment.

Automatic meteorological equipment is a device that enables automatic meteorological observations using a computer. In recent years, the meteorological observatory has also been used to increase the accuracy of observations at low cost. The Korea Meteorological Administration (KMA) conducts terrestrial meteorological observations using not only manned stations but also more than 500 unmanned automated weather observation equipment.

The automatic weather observation equipment is mainly used for the production of disaster prevention weather information such as local real-time monitoring of the weather conditions. It has problems such as securing the installation place, artificial problems such as roads, surrounding buildings, automobile exhaust, power supply and communication network A part of which is installed on the roof.

The installation of the automatic weather observation equipment is in accordance with the regulation of the meteorological instrument of the World Meteorological Organization (WMO). The hygrometer should be placed inside a lawn or shading box made of grass and installed only inside the shading box Should be. In addition, regulations on wind direction, anemometer, precipitation, and barometer exist. Recently, however, the modernization of the city has intensified the heat island phenomenon in urban areas. When the automatic weather observation equipment is installed on the roof of a building, the large buildings block the wind, and the influence of radiant heat from hot winds such as outdoor units and cement or asphalt The temperature has been relatively high, and there has been a question about the regional representation of the observed data.

An object of the present invention is to improve the accuracy of meteorological observation by using a weather prediction model and thermal image information of a building to solve the above-described problems.

Another object of the present invention is to provide a meteorological observation method capable of coping with uncertainties caused by climate change.

Another object of the present invention is to generate objective and representative observation data by minimizing the influence of the environment in which the weather observation equipment is installed on the weather observation.

According to an aspect of the present invention, there is provided a meteorological observation method using a meteorological prediction model, the method comprising: receiving a predicted value at a specific time point provided by a weather prediction model from a server; Determining whether a difference between the predicted value and the measured value is equal to or greater than a preset reference value, determining whether the measured value is observed under a normal condition if the difference is equal to or greater than the reference value, And correcting the measured value or correcting the predicted value.

According to the present invention as described above, it is possible to improve the accuracy of meteorological observations using weather prediction models and thermal image information of buildings.

According to the present invention, it is possible to provide a weather observation method capable of coping with uncertainties caused by climate change.

According to the present invention, it is possible to generate objective and representative observation data by minimizing the influence of the environment in which the weather observation equipment is installed on the weather observation.

FIG. 1 illustrates a weather observation system according to an embodiment of the present invention. FIG.
2 is a flowchart for explaining a meteorological observation method according to an embodiment of the present invention,
3 is a flowchart for explaining a predicted value correction method according to an embodiment of the present invention.

The above and other objects, features, and advantages of the present invention will become more apparent by describing in detail exemplary embodiments thereof with reference to the attached drawings, which are not intended to limit the scope of the present invention. In the following description, well-known functions or constructions are not described in detail since they would obscure the invention in unnecessary detail. Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the drawings, the same reference numerals are used to designate the same or similar components, and all combinations described in the specification and claims can be combined in any manner. It is to be understood that, unless the context requires otherwise, references to singular forms may include more than one, and references to singular forms may also include plural forms.

1, a weather observation system according to an embodiment of the present invention includes a thermal imaging camera 10, a weather observation equipment 30, a server 50, and an analysis module 100, As shown in FIG.

The meteorological observation apparatus 200 according to an embodiment of the present invention may only refer to the analysis module 100 or may include the meteorological observation equipment 30 and the analysis module 100.

The weather observation equipment 30 is a device for measuring weather observation data and includes a sensor unit 31 including at least one of a weather vane, an anemometer, a hygrometer, a rainfall gauge, a sun, a barometer, A control unit 33 that is a processor that processes observation information, a communication unit 35 that transmits observation information measured through the wired / wireless communication module, and a storage unit 27 that stores observation information and software.

The analysis module 100 is a module for analyzing and processing observation information (actual measurement value) measured in the meteorological observation equipment 30 according to an embodiment of the present invention. The analysis module 100 can be included in the meteorological observation equipment 30, And can operate as a hardware device separately from the equipment 30.

Hereinafter, a weather observation device 200 according to an embodiment of the present invention will be described with reference to the components included in the analysis module 100. The weather observation device 30 includes an observation device 30, . In this specification, the observation equipment 30 will be described on the roof, but the present invention is not limited thereto.

The weather observing apparatus 200 includes an image collecting unit 110, a control unit 130 and a communication unit 150. The control unit 130 includes a learning unit 133, a determination unit 135, and a correction unit 137 .

The image collection unit 110 may collect the thermal image of the peripheral region of the observation equipment 30 from the thermal imaging camera 10. Since the thermal imaging camera 10 is rotatable and must collect the thermal image of the peripheral region of the observation equipment 30, the thermal imaging camera 10 can be positioned at a position where the peripheral region including the observation equipment 30 can be photographed, Lt; / RTI >

The image collecting unit 110 can receive the thermal image of the area around the observation equipment 30 from the thermal imaging camera 10 through the communication unit 150 and the thermal imaging camera 10 can receive the thermal image from the weather observation device 200, It is possible to directly acquire the thermal image photographed by the thermal imaging camera 10.

The image analysis unit 120 may extract an image feature of the surrounding area thermal image using an image analysis model including at least one convolution layer to construct an image feature database. The image analysis model is a neural network such as a CNN (Convolution Neural Network), which includes a plurality of convolution layers that produce feature maps of features of the thermal image, a plurality of convolution layers, (Not shown) may be formed. An image analysis model according to an embodiment of the present invention can extract different levels of features in a thermal image.

To this end, the image analysis model of the present invention needs to learn a plurality of thermal images corresponding to big data for feature extraction in a thermal image, and therefore, the image analysis unit 120 performs feature extraction So that a plurality of thermal image images taken by the thermal imaging camera 10 can be cumulatively collected.

The image analysis unit 120 extracts the characteristics of each thermal image every time the peripheral region thermal image is received using the learned image analysis model. And the image feature database can be constructed using the features of the surrounding region thermal image. The constructed image feature database can be used for re-learning and updating of the image analysis model.

The communication unit 150 receives a predicted value at a specific time point provided by the weather prediction model from the server 50 and receives an actual value at a specific point in time on the observation equipment 30. [ The communication unit 150 is connected to the observation equipment 30, the server 50, and the terminal 70 through a wired / wireless network.

The predicted value received from the server 50 can be understood to mean a numerical solution or a forecast value calculated by the server 50 by integrating the numerical forecast model in the form of a partial differential equation. That is, it can be understood that the weather forecasting model of the present invention means a numerical forecast model. The predicted value may be the predicted data itself obtained from the numerical forecast or the numerical forecast result obtained through post-processing in the server 50. The communication unit 150 can receive the weather data applied to the respective numerical prediction models from the server 50 and the predicted values calculated from the respective numerical forecast models.

For example, the numerical forecasting model includes a global model (GDAPS) used for global medium-term prediction, a global ensemble model (EPS) used for global medium-term probability prediction, a regional model (RDAPS, WRF) for East Asian short- (LENS) for forecasting the Korean Peninsula risk, KLAPS, VDAPS for the short-term model for short-term prediction of Korean peninsula, GWW3 for global marine forecasting, RWW3 for ocean marine forecasting , CWW3 for predicting offshore marine coastal waters, RTSM for forecasting Asian storm tsunami, ADAM2 for forecasting dredge transport, ADAM3 for fog forecasting, And a neighborhood forecasting / statistical model. The server 50 also stores and drives various numerical prediction models that enable the calculation of the governing equations to be performed in a lattice form. The communication unit 150 receives predicted values from the respective numerical prediction models of the server 50 .

Although not shown in the drawing, a communication unit (not shown) of the server is connected to a high-speed network such as a global weather network (GTS), and can quickly collect large-scale observation data (weather data) from all over the world in real time. In addition to the predicted values provided by the respective numerical forecasting models, the communication unit of the server may include observation data collected from other observation equipments by the weather forecasting apparatus 200, for example, visibility, cloudiness, heavy load, temperature, dew point temperature, It can transmit ground observation data such as humidity, wind direction, wind speed, sea surface, and air pressure, weather radar image, and satellite image.

The communication unit 150 can receive a measured value at a specific point in the observation equipment 30. [ The actual measurement value refers to an observation value measured by the sensor unit 31 of the observation equipment 30. When the observation equipment 30 is a ground observation equipment, the actual temperature, the wind direction, the wind speed, the air pressure, the humidity, the dew point temperature, and the like can be received. When the observation equipment 30 is a high-rise meteorological instrument such as Rawinsonde, Windprofiler, or Microwave Radiometer, the vertical position of air pressure, temperature, humidity, wind direction, Distribution, etc. will be received at the measured value.

The control unit 130 determines whether the difference between the predicted value and the measured value is equal to or greater than a preset reference value, identifies whether the measured value is observed under the normal condition when the determined difference is equal to or greater than the reference value, and corrects the measured value according to the identified result .

Identifying whether the measured value is observed under the normal condition is performed through a previously learned normal range judgment model. In other words, the control unit 130 calculates the normalized range determination model using the image characteristics collected during the predetermined period and the error information of the meteorological data observed in the observation equipment during the period, By applying the extracted image feature, it is possible to judge whether the measured value is observed under the normal condition.

The normal range decision model is also a deep neural network learned using a large number of data corresponding to big data. If the image analysis model was a machine learning framework for extracting image features by inputting an image, the normal range judgment model can classify whether the image feature corresponds to a normal condition or an abnormal condition by inputting the image characteristic There is a machine learning framework. In order to generate the normal range determination model, the control unit 130 according to an embodiment of the present invention includes a learning unit 133. [

The learning unit 133 learns the normal range determination model using the image characteristics collected during the predetermined period and the error information of the meteorological data observed in the observation equipment during the period. For this, the thermal image feature data classified into the normal condition and the thermal image feature data classified into the abnormal condition are required. The learning unit 133 can use the image feature database constructed by the image analysis unit 120.

More specifically, for example, if the surrounding area of the observing equipment excessively reflects or emits infrared rays with asphalt, cement, etc., and the color gamut of the thermal image of the surrounding area affects the observing equipment, The image can be classified into the thermal image corresponding to the condition. This may be classified by an expert and may be automatically classified by comparing the measured temperature at the observation equipment 30 at the time when each thermal image was photographed by the computer, Lt; / RTI >

For example, let's say that the observation equipment installed on the roof of building A in X area is called observation equipment _A and the observation equipment installed on the lawn in area X is observation equipment _B. If it is determined that the temperature measured in the observation equipment _A is higher than the average temperature measured in the other observation equipment in the area X on a specific date, it can be judged that the observation equipment _A operates in an abnormal condition, The peripheral region thermal image of the observation equipment_A obtained in the observation apparatus_B can be classified into the thermal image corresponding to the abnormal condition and the peripheral region thermal image of the observation equipment_B can be classified into the thermal image corresponding to the normal condition. The plurality of thermal images thus classified are extracted by the image analysis unit 120 in the image analysis unit 120. The extracted image characteristics include the temperature data (temperature and humidity) observed in the observation equipment (observation equipment A) , Atmospheric pressure, etc.), as well as the normal range judgment model learning.

The error information of the meteorological data is generated on the basis of observation data measured at various locations in the same area (X region). For example, the average value or the intermediate value a (B) is the difference between b and a, or b ± ε (error range) and a is the difference between a and b, It can be used as information.

The error information of the weather data is used to classify the thermal image into the normal condition / abnormal condition. The weather data and the representative value (the observation value ). If the error information (the difference between the weather data and the representative value) is greater than or equal to the error range, it corresponds to the abnormal condition, and if the error information is within the error range, it corresponds to the normal condition.

Therefore, if the normal range judgment model is learned by using the image feature of the thermal image and the error information of the weather data, the normal range judgment model of the learning, when the image feature is inputted, corresponds to the abnormal condition or the abnormal condition Can be classified.

The determination unit 135 can determine whether the measured value is observed under the normal condition by applying the first image feature extracted from the peripheral region thermal image obtained at the specific time to the normal range determination model in which the training has been completed have.

If the determination unit 135 determines that the measured value is not observed under the normal condition as a result of the determination by the determination unit 135 - if the normal range determination model classifies the first image characteristic as an abnormal condition, The controller 137 can calculate the correction observation value using the error information corresponding to the first image feature. The correction value refers to a value obtained by correcting the measured value. The correction unit 137 calculates error information corresponding to the first image characteristic, that is, the difference between the observed value and the representative value, The value can be corrected to the calibration observation value. For example, in the above-described embodiment, when | b - a | is error information corresponding to the first image feature, | b '(corrected observation value) - a B can be corrected to b 'so that? The calibrated observations can be sent to the server 50 or the terminal 70.

On the other hand, if the difference between the computed result predicted value and the measured value of the control unit 130 is equal to or greater than a preset reference value and the measured value is observed under normal conditions as a result of the determination by the determination unit 135, the predicted value may be incorrect. In this case, the correction unit 137 can correct the predicted value. For this, a step of generating and using a weather prediction correction model should be performed.

Although the present invention relates to a meteorological observation method, a method of correcting predicted values that can be performed by the analysis module 100 according to another embodiment of the present invention will be briefly described. The image capturing unit 110 may collect the weather radar image of the area where the observation equipment 30 is installed from the server 50 and the image analysis unit 120 may use the CNN (Convolution Neural Network) (Feature vector) of the meteorological image can be extracted from the image.

The control unit 130 may calculate the corrected predicted value by applying the weather-feature image feature vector at the first time point and the predicted value at the specific time point to the learned weather prediction correction model. At this time, the weather data of the past first point of view, the weather image feature vector of the first point of time and the predicted value of the specific point of time (predicted value that needs to be corrected) may be input to the weather prediction correction model.

The weather prediction correction model is a machine learning framework that shows the nonlinear relationship between the input vector and the output value, and it can perform learning and analysis using the in - depth neural network. For example, the weather prediction correction model may have Recurrent Neural Networks (RNN), Long Short-Term Memory models (LSTM), and Gated Recurrent Units (GRU) structures.

The weather forecasting correction model is composed of A weather data at A time point, weather image feature vector extracted from A weather wave radar image at time A, predicted value at time B calculated using A weather data, , And the weather data at the time point B as learning data. The control unit 130 uses the A-point weather data, the weather image feature vector extracted from the weather radar image at the A-point, and the B-point predicted value calculated using the A-point weather data for the learning of the weather prediction correction model Thereby generating one multidimensional learning input vector and inputting the learning input vector to the wake prediction correction model so that the output value corresponds to the actual wakeup data of the B view. As a result, well-trained weather forecasting correction models are learned in such a way as to reduce the error between the predicted value and the actual value of the numerical prediction model using the weather data and the weather image feature vector at a specific point in time, The feature vector, and the predicted value at the predicted time are input. Here, learning means training a depth-of-field neural network so that a large number of data corresponding to big data is input and an output value corresponds to an actual value. Since the process of learning (or training) is obvious to a general technician, It is omitted.

The correcting unit 137 can calculate the corrected predicted value using a sufficiently trained weather prediction correction model, which may differ depending on which prediction value is provided by a certain numerical prediction model. For example, the weather forecast predicted calibration model using the predicted values provided by the WRF model and the predicted weather forecast predicted models provided by the RDAPS model may be different. Therefore, the correction unit 137 can use another weather prediction correction model depending on the prediction value received from the server 50 by a certain numerical prediction model.

For example, if the predicted value received from the server 50 is a predictive value provided by the WRF model, the correcting unit 137 obtains the second viewpoint weather data used in the WRF model at the past second point of time, (Hereinafter, referred to as " WRF weather prediction correction model ") using the predicted value of the past third time calculated using the predicted value of the past third time.

In another embodiment, the weather prediction correction model may be generated for each region. For example, the WRF meteorological prediction model in Seoul, the RDAPS meteorological prediction model in Seoul, the WRF meteorological prediction model in Jeju, and the WRF meteorological prediction model in Jeju can be learned separately and independently. This is based on what data is used as learning data in generating each machine learning model. In other words, the WRF weather forecasting correction model in the Seoul area uses the predicted value of the third time point calculated by the WRF model and the weather image feature vector extracted from the weather radar image of the second time point using the second time weather data of the Seoul area, The weather data of the second time point and the weather data of the third time point are learned as the learning data. Therefore, when the terminal is designated as the area of interest by the terminal and the prediction time is within 66 hours The calculation unit 140 can correct the predicted value using the WRF weather prediction correction model in the Seoul area.

When the correction unit 127 corrects the predicted value, the corrected estimated value (corrected predicted value) is transmitted to the server 50 so that the server 50 can provide information so that the corrected predicted value can be utilized in the future numerical prediction model or the like have.

Referring to FIG. 2, a weather observation device according to an embodiment of the present invention receives a predicted value at a specific time point provided by a weather prediction model from a server (S10), and receives an actual value at a specific time point from an observation device (S20).

The weather prediction model according to an exemplary embodiment of the present invention may be any one of RDAPS, WRF, LDAPS, KENS, KLAPS, VDAPS, ADAM2, and ADAM3.

The meteorological observation apparatus determines whether the difference between the predicted value and the measured value is equal to or greater than a preset reference value (S30). If the difference is not less than the reference value, it is discriminated whether the measured value is observed under the normal condition (S40) The value can be corrected (S50) or the predicted value can be corrected (S60).

In step 40, the step of identifying whether the measured value is observed under the normal condition in step 40 includes a step of calculating a normal range determination model that is learned using the image characteristics collected during the predetermined period and the error information of the meteorological data observed in the observation equipment during the period By determining whether the measured value is observed under the normal condition by applying the image feature extracted from the peripheral region thermal image obtained at the specific time. Since specific embodiments of steps 40 to 50 are the same as those of the above-described example, duplicate descriptions are omitted. On the other hand, step 60 can be performed through the method shown in Fig.

3, the weather observing apparatus according to an embodiment of the present invention includes a weather data at a first time point corresponding to position information, a predicted value at a prediction time point provided by the numerical prediction model, The weather radar image can be collected (S200). It is obvious to those skilled in the art that the weather observation apparatus can collect the data in advance and selectively use the data corresponding to the weather prediction information request from among the information collected in advance in step 200.

Next, the meteorological observing apparatus can extract the meteorological image feature vector of the first viewpoint from the first viewpoint weather radar image (S300). The meteorological observing apparatus calculates the corrected predicted value by applying the meteorological data at the first time point, the meteorological image feature vector at the first time point, and the predicted value at the predicted time point to the learned meteorological prediction correction model (S400) The weather forecast information can be transmitted to the terminal (S500).

The meteorological prediction correction model used in step 400 includes the second viewpoint weather data used for the numerical forecast model at the second time point, the weather image feature vector extracted from the weather point radar image at the second point of time, the second time point weather data The predicted value of the past third point and the weather data of the third point, which have been calculated using the above-mentioned first and second points of time, as learning data.

The meteorological observing apparatus includes second point-in-time weather data used in the numerical forecast model at the second point in time, a predicted value at the third point in time at the past calculated using the second point-in-time weather data, The weather data at the third time point, and the weather radar image at the second time point are collected (S151). Then, the second time-of-day weather image feature vector is extracted from the weather radar image at the second time point, and the learning data is generated using the second time point weather data, the second time weather image feature vector, and the third time point prediction value S153). Learning data can be generated in vector form. Next, the meteorological observation apparatus learns the meteorological prediction correction model using the learning data (S155). If the iterative learning is performed through a sufficient number of learning data, the weight of the depth neural network constituting the meteorological prediction correction model is optimized , The trained weather prediction correction model may be used to calculate the corrected prediction value in step 400. [

Although not shown in the figure, the weather forecasting correction model may be generated for each region. When weather data at the prediction time point after step 500 is received, the weather observing apparatus calculates a first error between the predicted value at the prediction time point and the weather data at the prediction time point And calculate a second error between the corrected predicted value and the weather data at the prediction time to compare the magnitude of the first error and the second error. The meteorological observing apparatus can determine whether to apply the weather prediction correction model according to the result of the comparison.

The above-described weather observation method can be implemented through a weather observation application program stored in a computer-readable medium for executing any one of the methods described in this specification.

Some embodiments omitted in this specification are equally applicable if their implementation subject is the same. It is to be understood that both the foregoing general description and the following detailed description of the present invention are exemplary and explanatory and are intended to be exemplary and explanatory only and are not restrictive of the invention, The present invention is not limited to the drawings.

10: Thermal camera
30: Weather observation equipment
50: Server
70: terminal
100: Analysis module

Claims (4)

Receiving a predicted value at a specific time point provided by a weather forecasting model from a server;
Receiving an actual value of the specific time point from the observation equipment;
Determining whether a difference between the predicted value and the measured value is equal to or greater than a preset reference value;
Identifying whether the measured value is observed under normal conditions if the difference is greater than or equal to the reference value;
And correcting the actual value or correcting the predicted value according to the identification result.

The method according to claim 1,
The step of identifying whether the measured value is observed under normal conditions
An image feature extracted from the peripheral region thermal image acquired at the specific time point is applied to the learned normal range determination model using the image characteristics collected during the predetermined period and the error information of the weather data observed in the observation equipment during the period And determining whether the measured value is observed under a normal condition.

The method according to claim 1,
Wherein the weather prediction model is one of RDAPS, WRF, LDAPS, KENS, KLAPS, VDAPS, ADAM2 or ADAM3.
An image collecting unit for acquiring a thermal image of a peripheral region of the observation equipment using a rotatable thermal imaging camera;
An image analyzer for extracting features of the peripheral region thermal image using an image analysis model including at least one convolution layer to construct an image feature database;
A communication unit for receiving a predicted value at a specific time point provided by a weather prediction model from a server and receiving an actual value at the specific time point from an observation equipment;
Determining whether a difference between the predicted value and the actual value is equal to or greater than a preset reference value, discriminating whether the actual value is observed under a normal condition when the difference is equal to or greater than the reference value, And a control unit
The control unit
A learning unit for learning a normal range determination model using image characteristics collected during a predetermined period and error information of meteorological data observed in the observation equipment during the period;
A determination unit for determining whether the measured value is observed under normal conditions by applying a first image feature extracted from the peripheral region thermal image acquired at the specific time to the normal range determination model;
And a correction unit for calculating a correction observation value using the error information corresponding to the first image characteristic if the measured value is not observed under normal conditions.

Images