KR20230104011A - Method for predicting volume of fine particulate matter using model based on artificial intelligence, program and apparatus therefor - Google Patents

Abstract

An apparatus for predicting the volume of ultrafine dust is disclosed. The present device collects sensing data of one or more particulate matter and environmental factors using a communication unit, an input unit, and sensors disposed in a plurality of regions, and collects characteristic information having a predetermined correlation between the collected sensing data and ultrafine dust. and generate information on the proximity relationship between sensors based on the distances between sensors disposed in a plurality of regions, input selected feature information and information on the generated proximity relationship to a pre-learned prediction model, and pass a predetermined time It may include a control unit for estimating the volume of ultrafine dust for each of the plurality of areas later. By providing this device, the volume of ultrafine dust can be predicted in chronological order.

Description

Method for predicting volume of ultrafine dust using artificial intelligence-based model, program and device therefor

The present disclosure relates to techniques for predicting contamination conditions. More specifically, the present disclosure relates to a method for predicting the volume of ultrafine dust using an artificial intelligence-based model, and a program and apparatus therefor.

Dust refers to particulate matter that floats or is blown down in the air. There are natural sources of fine dust, but in general, soot generated when fossil fuels such as coal and oil are burned in boilers and power generation facilities, vehicle exhaust gas, It can be said to be artificial, such as flying dust from construction sites, raw materials in the form of powder in factories, powder components in the handling process of subsidiary materials, and smoke from incineration plants. Fine dust generated in this way floats in the air and is deposited on roads, etc., and has a characteristic of re-dispersion.

Such dust is classified into Total Suspended Particles (TSP), which is less than 50 μm, and Particulate Matter (PM), which has a very small particle size, depending on the particle size. Fine dust is divided into fine dust with a diameter of less than 10 μm (hereinafter referred to as PM10) and ultrafine dust with a diameter of less than 2.5 μm (hereinafter referred to as PM2.5).

If PM10 is about 1/5 to 1/7 smaller than the diameter of a human hair (50 to 70㎛), PM2.5 is very small, about 1/20 to 1/30 of a human hair. Since fine dust is so small that it is invisible to the naked eye, it may have a bad effect on health by staying in the air and penetrating the lungs through the respiratory tract or moving into the body along the blood vessels.

The World Health Organization (WHO) has been presenting air quality guidelines for fine dust (PM10, PM2.5, etc.) since 1987, and in 2013, the International Agency for Research on Cancer (IARC) under the World Health Organization ) designated fine dust as a Group 1 carcinogen that has been confirmed to be carcinogenic to humans (Group 1), and interest in it has recently increased due to health risks such as respiratory diseases, cardiovascular diseases, and asthma caused by fine dust.

Therefore, fine dust that poses a risk to health occurs more and more frequently, and weather forecasts also present information on fine dust by expressing PM10 as fine dust and PM2.5 as ultrafine dust. However, while the climate continuously changes, information on fine dust is not presented quickly, and accurate information is not obtained because the measurement location is different from the area where individual people are located.

Therefore, it is necessary to quickly and accurately predict the volume (concentration) of ultrafine dust.

Publication No. 10-2021-0081600 (Publication date: 2021.07.02.)

An object of the present disclosure is to provide a method for predicting the volume of ultrafine dust using sensing data collected from a plurality of regions.

The problems to be solved by the present disclosure are not limited to the problems mentioned above, and other problems not mentioned will be clearly understood by those skilled in the art from the description below.

A method for predicting the volume of ultrafine dust performed by a computing device according to an aspect of the present disclosure for achieving the above-described technical problem is a measurement of one or more particulate matter and environmental factors using sensors disposed in a plurality of regions. Collecting sensing data; selecting characteristic information having a predetermined correlation between the collected sensing data and ultrafine dust; generating information about a proximity relationship between sensors based on distances between sensors disposed in the plurality of regions; and predicting the volume of ultrafine dust for each of the plurality of regions after a predetermined time has elapsed by inputting the selected feature information and information on the generated neighbor relationship into a pre-learned prediction model.

The one or more particulate matter has a diameter of less than or equal to 0.1 μm, particulate matter greater than 0.1 μm and less than or equal to 0.3 μm, particulate matter greater than 0.3 μm and less than or equal to 0.5 μm, particulate matter greater than 0.5 μm and less than or equal to 1.0 μm, greater than 1.0 μm and less than or equal to 2.5 μm. It includes particulate matter of less than μm and particulate matter of more than 2.5 μm and less than or equal to 5.0 μm, wherein the environmental factor is carbon monoxide (CO), nitrogen dioxide (NO ₂ ), ozone (O ₃ ), sulfur dioxide (SO ₂ ), the plurality of regions It may include at least one of temperature, humidity, pressure, wind direction, wind speed, and precipitation.

The selecting of the feature information may include calculating a Pearson's correlation coefficient between the at least one particulate matter and the ultrafine dust and between the environmental factor and the ultrafine dust; and selecting feature information having a value of the calculated Pearson's correlation coefficient greater than or equal to a predetermined value.

In the generating of the information on the proximity relationship, the proximity relationship between the sensors is generated based on the distance between the sensors based on a two-dimensional matrix, and each element of the matrix is set to 0 or more and 1 or less, but the distance between the sensors is the criterion. A step of setting the adjacent distance value when the distance is smaller than the adjacent distance value when the distance is larger than the reference distance.

The predictive model may include: a first model learned to output time-sequential data by receiving the selected feature information and proximity relationship information according to distances between sensors disposed in the plurality of regions; and a second model that receives the output value of the first model and is trained to transform the output value of the first model through attention processing.

The first model includes a first structure for receiving the selected feature information and neighbor relationship information according to distances between sensors disposed in the plurality of regions and calculating them in a forward direction and a second structure for calculating them in a reverse direction, wherein the first model includes: It may include a configuration of concatenating the output value of the first structure and the output value of the second structure.

The second model may include a configuration for learning by placing a more important weight among the concatenated output values.

An apparatus for predicting the volume of ultrafine dust according to an aspect of the present disclosure for achieving the above-described technical problem includes a communication unit; input unit; and collecting sensing data of one or more particulate matter and environmental factors using sensors disposed in a plurality of regions through the communication unit or input unit, and providing characteristic information having a predetermined correlation between the collected sensing data and ultrafine dust. select, generate information about the proximity relationship between sensors based on the distances between sensors disposed in the plurality of regions, and input the selected feature information and information about the generated proximity relationship into a pre-learned predictive model, A control unit for estimating the volume of ultrafine dust for each of the plurality of regions after a predetermined time has elapsed may be included.

The control unit calculates Pearson correlation coefficients between the one or more particulate matter and the ultrafine dust and between the environmental factor and the ultrafine dust, and selects feature information having a value of the calculated Pearson correlation coefficient equal to or greater than a predetermined value. can be configured.

The control unit generates an adjacency relationship between sensors based on a distance between sensors based on a two-dimensional matrix, and sets each element of the matrix to 0 or more and 1 or less, but an adjacency distance value when the distance between sensors is smaller than the reference distance. It may be configured to set greater than the adjacent distance value when greater than the reference distance.

The predictive model may include: a first model learned to output time-sequential data by receiving the selected feature information and proximity relationship information according to distances between sensors disposed in the plurality of regions; and a second model that receives the output value of the first model and is trained to transform the output value of the first model through attention processing.

The first model includes a first structure for receiving the selected feature information and neighbor relationship information according to distances between sensors disposed in the plurality of regions and calculating them in a forward direction and a second structure for calculating them in a reverse direction, wherein the first model includes: It may include a configuration of concatenating the output value of the first structure and the output value of the second structure.

The second model may include a configuration for learning by placing a more important weight among the concatenated output values.

In addition to this, a computer program stored in a computer readable recording medium for execution to implement the present disclosure may be further provided.

In addition to this, a computer readable recording medium recording a computer program for executing a method for implementing the present disclosure may be further provided.

According to the above-described problem solving means of the present disclosure, the volume (concentration) of ultrafine dust can be accurately predicted over time based on sensing data collected from a plurality of regions, so that analysis efficiency can be improved.

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

1 is a diagram for explaining an environment to which an ultrafine dust volume prediction device according to the present disclosure is applied.
2 is a block diagram showing the configuration of an apparatus for predicting the volume of ultrafine dust according to the present disclosure.
3 is a sequence diagram illustrating a method for predicting the volume of ultrafine dust according to the present disclosure.
4 shows a model for predicting ultrafine dust volume according to the present disclosure.
5 is a diagram for explaining detailed processing of a model for predicting the volume of ultrafine dust according to the present disclosure.

Like reference numbers designate like elements throughout this disclosure. The present disclosure does not describe all elements of the embodiments, and general content or overlapping content between the embodiments in the technical field to which the present disclosure belongs is omitted. The term 'unit, module, member, or block' used in the specification may be implemented as software or hardware, and according to embodiments, a plurality of 'units, modules, members, or blocks' may be implemented as one component, It is also possible that one 'part, module, member, block' includes a plurality of components.

Throughout the specification, when a part is said to be "connected" to another part, this includes not only the case of being directly connected but also the case of being indirectly connected, and indirect connection includes being connected through a wireless communication network. do.

In addition, when a certain component is said to "include", this means that it may further include other components without excluding other components unless otherwise stated.

Throughout the specification, when a member is said to be located “on” another member, this includes not only a case where a member is in contact with another member, but also a case where another member exists between the two members.

Terms such as first and second are used to distinguish one component from another, and the components are not limited by the aforementioned terms.

Expressions in the singular number include plural expressions unless the context clearly dictates otherwise.

In each step, the identification code is used for convenience of description, and the identification code does not explain the order of each step, and each step may be performed in a different order from the specified order unless a specific order is clearly described in context. there is.

Hereinafter, the working principle and embodiments of the present disclosure will be described with reference to the accompanying drawings.

In this specification, the 'ultrafine dust volume prediction device according to the present disclosure' includes all various devices capable of providing results to users by performing calculation processing. For example, the device for predicting the volume of ultrafine dust according to the present disclosure may include a computer, a server device, and a portable terminal, or may be in any one form.

Here, the computer may include, for example, a laptop computer, a desktop computer, a laptop computer, a tablet PC, a slate PC, and the like equipped with a web browser.

The server device is a server that processes information by communicating with an external device, and may include an application server, a computing server, a database server, a file server, a game server, a mail server, a proxy server, and a web server.

The portable terminal is, for example, a wireless communication device that ensures portability and mobility, and includes a Personal Communication System (PCS), a Global System for Mobile communications (GSM), a Personal Digital Cellular (PDC), a Personal Handyphone System (PHS), and a PDA. (Personal Digital Assistant), IMT (International Mobile Telecommunication)-2000, CDMA (Code Division Multiple Access)-2000, W-CDMA (W-Code Division Multiple Access), WiBro (Wireless Broadband Internet) terminal, smart phone ) and wearable devices such as watches, rings, bracelets, anklets, necklaces, glasses, contact lenses, or head-mounted-devices (HMDs). can include

1 is a diagram for explaining an environment to which an ultrafine dust volume prediction apparatus 100 according to the present disclosure is applied.

The plurality of regions AR may include a first region AR1 , a second region AR2 , and a third region AR3 . Sensor groups S1 to S3 for collecting particulate matter and various environmental factors may be disposed in the plurality of regions AR. Here, the plurality of regions (AR) will be described as including Incheon (AR1), Seoul (AR2), and Gyeonggi (AR3) regions, but the embodiment is not limited thereto.

The sensor groups S1 to S3 disposed in the plurality of regions AR may collect sensing data of particulate matter having various sizes and environmental factors. The first sensor group S1 includes a plurality of sensors S1a to S1g disposed in the first area AR1, and the second sensor group S2 includes a plurality of sensors S2a to S1g in the second area AR2. S2d), and the third sensor group S3 may include a plurality of sensors S3a to S3g in the third area AR3. However, the number of sensors may vary according to embodiments.

In embodiments, the particulate material has a diameter of less than or equal to 0.1 μm, greater than 0.1 μm and less than or equal to 0.3 μm, greater than 0.3 μm and less than or equal to 0.5 μm, greater than 0.5 μm and less than or equal to 1.0 μm, greater than 1.0 μm and less than or equal to 2.5 μm. It may include particulate matter of less than μm and particulate matter of more than 2.5 μm and less than 5.0 μm, and particulate matter having a diameter of less than 2.5 μm may be referred to as ultrafine dust (hereinafter PM2.5).

In an embodiment, the environmental factor may include at least one of carbon monoxide (CO), nitrogen dioxide (NO ₂ ), ozone (O ₃ ), sulfur dioxide (SO ₂ ), temperature, humidity, pressure, wind direction, wind speed, and precipitation, Depending on the implementation example, data of various environmental factors may be collected.

The ultrafine dust volume prediction device 100 may analyze the sensing data collected using various sensors S1a to S3g disposed in a plurality of areas AR. After a predetermined time has elapsed based on the collected sensing data, The volume (concentration) of ultrafine dust in the area (AR) can be predicted.

The ultrafine dust volume prediction device 100 may select feature information having a predetermined correlation between the collected sensing data and the ultrafine dust, and based on the distance between the sensors disposed in the plurality of areas (AR), the proximity between the sensors Information about relationships can be created.

The ultrafine dust volume prediction device 100 can predict the volume of ultrafine dust for each of a plurality of regions (AR1 to AR3) after a predetermined time has elapsed based on the selected feature information and the generated information on the adjacent relationship between the sensors. .

In particular, the ultrafine dust volume prediction device 100 may predict the volume of ultrafine dust by reflecting regional characteristics. When yellow dust flows into Incheon (AR1) from China, the ultrafine dust volume prediction device 100 predicts the volume of ultrafine dust in Incheon (AR1) and at the same time, the volume of ultrafine dust in other regions (AR2, AR3). can be predicted over time by reflecting the adjacent relationship with Incheon (AR1).

2 is a block diagram showing the configuration of the ultrafine dust volume prediction device 100 according to the present disclosure.

Referring to FIG. 2 , the ultrafine dust volume prediction device 100 may include a communication unit 110, an input unit 120, a display 140, a memory 150, and a control unit 190.

Among the components, the communication unit 110 may include one or more components enabling communication with an external device, for example, a broadcast receiving module, a wired communication module, a wireless communication module, a short-distance communication module, and location information. At least one of the modules may be included.

The input unit 120 is for inputting image information (or signals), audio information (or signals), data, or information input from a user, and includes at least one of at least one camera, at least one microphone, and a user input unit. can do. Voice data or image data collected by the input unit 120 may be analyzed and processed as a user's control command.

The sensing unit (not shown) is described as not included, but when the sensing unit (not shown) is included, the sensing unit (not shown) may measure one or more particulate matter and environmental factors, and the information in the device, At least one of surrounding environment information surrounding the present device and user information is sensed, and a sensing signal corresponding thereto is generated. Based on these sensing signals, the controller 190 may control driving or operation of the device, or perform data processing, functions, or operations related to an application program installed in the device. In an embodiment, the sensing unit (not shown) may be implemented separately from the present device, and may collect sensing data through the communication unit 110 or the input unit 120.

As described above, the sensing unit includes an environmental sensor (eg, a barometer, a hygrometer, a thermometer, a radiation detection sensor, a heat detection sensor, and a gas detection sensor), a chemical sensor (eg, a healthcare sensor, a biometric sensor). recognition sensor, etc.), proximity sensor, illumination sensor, touch sensor, acceleration sensor, magnetic sensor, gravity sensor (G-sensor), gyroscope Sensors (gyroscope sensor), motion sensor (motion sensor), RGB sensor, infrared sensor (IR sensor), fingerprint recognition sensor (finger scan sensor), ultrasonic sensor (ultrasonic sensor), optical sensor (e.g. (e.g., camera), wind direction sensor, wind speed sensor, precipitation sensor, gas sensor for detecting gases (eg, carbon monoxide (CO), nitrogen dioxide (NO ₂ ), ozone (O ₃ ), sulfur dioxide (SO ₂ )), microphone may include at least one of them. Meanwhile, the present device may combine and utilize information sensed by at least two or more of these sensors.

The display 130 may implement a touch screen by forming a mutual layer structure or integrally with the touch sensor. Such a touch screen may function as a user input unit providing an input interface between the device and the user, and may provide an output interface between the device and the user.

The display 130 displays (outputs) information processed by the present device. For example, the display 130 may display execution screen information of an application program (for example, an application) running in the present device, or UI (User Interface) and GUI (Graphic User Interface) information according to such execution screen information. can

The memory 150 may store data supporting various functions of the device and programs for operation of the control unit, and may store input/output data (eg, music files, still images, moving images, etc.) , a plurality of application programs (application programs or applications) running in the present device, data for the operation of the present device, and instructions can be stored. At least some of these application programs may be downloaded from an external server through wireless communication.

The memory 150 may be a flash memory type, a hard disk type, a solid state disk type, a silicon disk drive type, or a multimedia card micro type. micro type), card type memory (eg SD or XD memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable EEPROM (EEPROM) It may include a storage medium of at least one type of a programmable read-only memory (PROM), a programmable read-only memory (PROM), a magnetic memory, a magnetic disk, and an optical disk. In addition, the memory is separate from the present device, but may be a database connected by wire or wireless.

The memory 150 may store a predictive model. The predictive model may be a pre-learned model, and may be stored in the memory 150 even during learning.

The control unit 190 includes a memory for storing data for an algorithm or a program for reproducing the algorithm for controlling the operation of components in the present device, and at least one processor for performing the above-described operation using the data stored in the memory ( not shown) may be implemented. In this case, the memory and the processor may be implemented as separate chips. Alternatively, the memory and the processor may be implemented as a single chip.

In addition, the control unit 190 may control any one or a combination of the components described above in order to implement various embodiments according to the present disclosure described in FIGS. 3 to 5 below on the present device.

At least one component may be added or deleted corresponding to the performance of the components shown in FIG. 2 . In addition, it will be easily understood by those skilled in the art that the mutual positions of the components may be changed corresponding to the performance or structure of the system.

Meanwhile, each component shown in FIG. 2 means software and/or hardware components such as a Field Programmable Gate Array (FPGA) and Application Specific Integrated Circuit (ASIC).

3 is a sequence diagram showing a method for predicting the volume of ultrafine dust according to the present disclosure, FIG. 4 shows a model for predicting the volume of ultrafine dust according to the present disclosure, and FIG. 5 shows the volume of ultrafine dust according to the present disclosure. It is a diagram for explaining detailed processing of a model for prediction. While describing FIG. 3, necessary parts will be described with reference to FIGS. 4 and 5 together.

The controller 190 may collect sensing data of one or more particulate matter and environmental factors using sensors disposed in a plurality of regions (S310).

The controller 190 receives sensing data for the volume (concentration) of particulate matter (PM0.3, PM0.5, PM1.0, PM2.5, PM5.0, PM10, etc.) having various sizes in each of a plurality of regions. It can be collected, and includes at least one of carbon monoxide (CO), nitrogen dioxide (NO ₂ ), ozone (O ₃ ), sulfur dioxide (SO ₂ ), temperature, humidity, pressure, wind direction, wind speed and precipitation in each of a plurality of regions. Sensing data of environmental factors can be collected.

After step S310, the controller 190 may select feature information having a predetermined degree of correlation between the collected sensing data and ultrafine dust (S320).

The controller 190 may apply Pearson's correlation coefficient for each item as shown in Equation [1] below in order to check the correlation between ultrafine dust (PM2.5) and collected data.

[Equation 1]

및

는 값의 평균일 수 있다. 즉, 비교되는 두 항목 간의 상관 관계가 피어슨 상관 계수에 의해 결정될 수 있다.Here, r is the Pearson correlation coefficient, a value representing the degree of correlation, from -1 to 1 can be calculated, x _i and y _i are values of the data set,

and

may be the average of the values. That is, a correlation between two items to be compared may be determined by a Pearson correlation coefficient.

When the value of the Pearson's correlation coefficient of ultrafine dust and each variable (eg, particulate matter having various sizes and environmental factors) is a predetermined value (eg, +0.5 or more or -0.5 or less), the controller 190 converts it to characteristic information can be selected with

Accordingly, the controller 190 may select one or more variables among the above-described particulate matter having various sizes and/or environmental factors as characteristic information according to the value of the Pearson's correlation coefficient. The controller 190 may select all variables as feature information or select some variables as feature information.

The controller 190 may generate information about a proximity relationship between sensors based on distances between sensors disposed in a plurality of regions (S330). In an embodiment, steps S320 and S330 may be performed in a different order.

The control unit 190 may generate an adjacency relationship between sensors based on a two-dimensional matrix. The distance between each sensor may be replaced with a value of 0 or more and 1 or less to set each element value of the matrix. For example, the control unit 190 may indicate the proximity relationship between sensor i and sensor j in the i and j th elements of the two-dimensional matrix representing the proximity relationship between sensors.

The controller 190 may set an adjacent distance value when the distance between sensors is smaller than the reference distance to be greater than an adjacent distance value when the sensor distance is larger than the reference distance. The controller 190 may set a reference distance. Based on a specific sensor, a sensor disposed farther than the reference distance may be set closer to 0 (for example, between 0 and 0.25) according to a distance ratio, and may be placed closer than the reference distance. The sensor may be set close to 1 (for example, between 0.75 and 1) according to the distance ratio, and the element value may be set to 1 for the distance between the same sensors.

Referring to the specific i-th sensor, the control unit 190 sets it closer to 0.25 (upper limit value) as it is closer to the specific i-th sensor in the case of a sensor disposed farther than the reference distance, and sets it to 0 ( lower limit) can be set close to. In addition, in the case of a sensor disposed closer than the reference distance, the control unit 190 sets it closer to 1 (upper limit value) as it is closer to the specific i-th sensor, and sets it closer to 0.75 (lower limit value) as it is disposed farther from the specific i-th sensor. there is. That is, different values may be set according to the distance ratio between the sensors.

In an embodiment, the control unit 190 adjusts the reference distance commonly applied to each sensor in real time after calculating all the distances between the sensors, and calculates the number of elements of the adjacent relationship disposed further than the reference distance to the adjacent elements disposed closer than the reference distance. It can be set smaller than the number of elements in the relationship. This is to reflect the adjacent relationship closer than the reference distance more importantly, and the matrix of the adjacent relationship may be reflected in weight calculation in the calculation of the predictive model.

After step S330, the control unit 190 may predict the volume of ultrafine dust for each of a plurality of regions after a predetermined time has elapsed by inputting the selected feature information and information on the generated neighbor relationship into a pre-learned prediction model. (S340).

Referring to FIG. 4 , the controller 190 may receive data in step ST1, and the input data may be formed as follows.

Here, N may be the number of samples, l may be a time period (n is the number of deployed sensors, and c is a feature dimension). N may be 5000, l may be 4 (n is tens to hundreds, and the feature dimension is the number of particulate matter and environmental factors described above). At this time, in the case of feature dimension (c), particulate matter with a diameter of less than 0.1 μm, particulate matter greater than 0.1 μm and less than 0.3 μm, particulate matter greater than 0.3 μm and less than 0.5 μm, particulate matter greater than 0.5 μm and less than 1.0 μm, particulate matter greater than 0.5 μm and less than 1.0 μm, 1.0 μm Particulate matter greater than 2.5 μm and less than or equal to 2.5 μm and less than or equal to 5.0 μm, carbon monoxide (CO), nitrogen dioxide (NO ₂ ), ozone (O ₃ ), sulfur dioxide (SO ₂ ), temperature, humidity, In the case of including pressure, wind direction, wind speed, and precipitation, the number may be 16, but the number may vary depending on the correlation with the above-mentioned ultrafine dust.

As described above, the control unit 190 may create a proximity relationship between sensors (ST2), and learn to output time-sequential data by receiving selected feature information and proximity relationship information according to distances between sensors disposed in a plurality of regions. A first model may be applied (ST3).

The first model includes a first structure (Forward LSTM) that receives selected feature information and adjacent relationship information according to distances between sensors disposed in a plurality of regions and calculates them in a forward direction and a second structure (Backward LSTM) that calculates them in a reverse direction. and may be configured to concatenate the output value of the first structure and the output value of the second structure. Since both the first and last LSTM operation values can be reflected through the first and second structures, the volume of ultrafine dust over time can be accurately predicted.

,

)가 weight 로 반영될 수 있다. 여기서, 가중치는 상술한 센서 간 인접 관계가 숫자 그대로 반영될 수 있다. 이때, 센서 간 거리가 기준 거리 내의 요소값이 보다 중요하게 연산될 수 있다. 기준 거리 밖의 요소값의 경우 0에 가까우므로, 연산에 영향이 적을 수 있다.Here, each gate (input gate, forget gate, output gate) of LSTM (long short term memory) has a weight (weight of adjacency relationship according to the distance between sensors)

,

) can be reflected as the weight. Here, the weight may reflect the above-described proximity relationship between sensors as a number. At this time, the distance between the sensors may be calculated as more important than the element value within the reference distance. Since element values outside the reference distance are close to 0, the calculation may be less affected.

), input gate(

), forget gate(

), output gate(

)에 센서 간 인접 관계가 weight 로 반영될 수 있는데, input node(

)는 t 시점의 입력 데이터, 가중치, 은닉 데이터, 제어 변수, 바이어스 등에 대한 시그모이드 함수를 이용하여 산출될 수 있으며, input gate(

), forget gate(

), output gate(

) 도 아래 식과 같이 유사하게 산출될 수 있다.

는 element-wise hyperbolic tangent linear activation function 일 수 있다.Specifically, referring to [Equation 2] below, the input node (

), input gate(

), forget gate(

), output gate(

), the proximity relationship between sensors can be reflected as weight, and the input node (

) can be calculated using a sigmoid function for the input data, weights, hidden data, control variables, biases, etc. at time t, and the input gate (

), forget gate(

), output gate(

) can also be similarly calculated as in the equation below.

may be an element-wise hyperbolic tangent linear activation function.

[Equation 2]

The controller 190 may apply the second model in step ST4, and the second model may be applied after receiving the output value of the first model to transform the input value through attention processing. In this case, the second model may include a structure for learning by placing a more important weight among concatenated output values.

)를 설정하여 보다 높은 가중치를 차등적으로 할당하며, 어텐션 wight와 연접된 출력값을 반영한 시그마 연산에 의해 컨텍스트 벡터(c_i)를 산출함으로써, 학습될 출력값(h₁~h_j) 중 결과에 영향을 크게 미치는 연접된 출력값을 강조할 수 있다. 제2 모델은 최종적으로 소정 시간 경과 후의 미세먼지의 볼륨(농도)를 산출할 수 있다.Referring to FIG _. 5, _the second model uses an attention weight (

) is set to assign higher weights differentially, and the context vector (c _i ) is calculated by sigma operation reflecting the output values concatenated with the attention wight, thereby affecting the result among the output values (h ₁ to h _j ) to be learned. It is possible to emphasize concatenated output values that greatly affect . The second model may finally calculate the volume (concentration) of fine dust after a predetermined time has elapsed.

The controller 190 may calculate the concentration of ultrafine dust over time through a predictive model in consideration of regional influences. For example, when yellow dust from China flows into Korea, the control unit 190 determines that the volume of ultrafine dust in Incheon is high and the volume of ultrafine dust increases in the direction of Seoul/Gyeonggi over time. It can be recognized as a change in the volume of ultrafine dust by

In this case, the controller 190 may deliver a message in the form of an alarm to a communication terminal in Gyeonggi/Seoul where the concentration of ultrafine dust is relatively low when ultrafine dust rapidly increases in Incheon.

Meanwhile, the disclosed embodiments may be implemented in the form of a recording medium storing instructions executable by a computer. Instructions may be stored in the form of program codes, and when executed by a processor, create program modules to perform operations of the disclosed embodiments. The recording medium may be implemented as a computer-readable recording medium.

Computer-readable recording media include all types of recording media in which instructions that can be decoded by a computer are stored. For example, there may be read only memory (ROM), random access memory (RAM), magnetic tape, magnetic disk, flash memory, optical data storage device, and the like.

As above, the disclosed embodiments have been described with reference to the accompanying drawings. Those skilled in the art to which the present disclosure pertains will understand that the present disclosure may be implemented in a form different from the disclosed embodiments without changing the technical spirit or essential features of the present disclosure. The disclosed embodiments are illustrative and should not be construed as limiting.

100: Ultrafine dust volume prediction device,
190: control unit.

Claims (13)

As a method for predicting the volume of ultrafine dust performed by a computing device,
collecting sensing data of one or more particulate matter and environmental factors using sensors disposed in a plurality of regions;
selecting characteristic information having a predetermined correlation between the collected sensing data and ultrafine dust;
generating information about a proximity relationship between sensors based on distances between sensors disposed in the plurality of regions; and
Predicting the volume of ultrafine dust for each of the plurality of regions after a predetermined time has elapsed by inputting the selected feature information and information on the generated neighbor relationship into a pre-learned prediction model, Volume forecasting method. According to claim 1,
The one or more particulate matter has a diameter of less than or equal to 0.1 μm, particulate matter greater than 0.1 μm and less than or equal to 0.3 μm, particulate matter greater than 0.3 μm and less than or equal to 0.5 μm, particulate matter greater than 0.5 μm and less than or equal to 1.0 μm, greater than 1.0 μm and less than or equal to 2.5 μm. It includes particulate matter of less than μm and particulate matter of more than 2.5 μm and less than or equal to 5.0 μm, wherein the environmental factor is carbon monoxide (CO), nitrogen dioxide (NO ₂ ), ozone (O ₃ ), sulfur dioxide (SO ₂ ), the plurality of regions Includes at least one of temperature, humidity, pressure, wind direction, wind speed and precipitation of,
The step of selecting the feature information,
calculating Pearson's correlation coefficients between the one or more particulate matter and the ultrafine dust and between the environmental factor and the ultrafine dust; and
A method for predicting the volume of ultrafine dust comprising the step of selecting feature information having a value of the calculated Pearson's correlation coefficient equal to or greater than a predetermined value. According to claim 2,
Generating information on the adjacency relationship,
An adjacency relationship between sensors is generated based on a two-dimensional matrix based on the distance between sensors, and each element of the matrix is set to 0 or more and 1 or less, but when the distance between sensors is smaller than the reference distance, the adjacency distance value is less than the reference distance. A method for predicting the volume of ultrafine dust, comprising the step of setting a larger value than the adjacent distance value in the case of a large case. According to claim 3,
The predictive model,
a first model learned to output time-sequential data by receiving the selected feature information and proximity relationship information according to distances between sensors disposed in the plurality of regions; and
A method for predicting the volume of ultrafine dust comprising a second model that receives an output value of the first model and is trained to transform the output value of the first model through attention processing. According to claim 4,
The first model,
a first structure for receiving the selected feature information and adjacent relationship information according to distances between sensors disposed in the plurality of regions and calculating them in a forward direction; and a second structure for calculating them in a reverse direction, wherein output values of the first structure and a second structure are calculated. A method for predicting the volume of ultrafine dust, including a configuration of concatenating the output values of the two structures. According to claim 5,
The second model,
A method for predicting the volume of ultrafine dust, including a configuration for learning with a more important weight among the concatenated output values. Combined with a computer, which is hardware, and stored in a medium to execute the method of any one of claims 1 to 6, ultrafine dust volume prediction program. As a device for predicting the volume of ultrafine dust,
communications department; input unit; and
One or more sensing data of particulate matter and environmental factors are collected using sensors disposed in a plurality of regions through the communication unit or input unit, and feature information having a predetermined correlation between the collected sensing data and ultrafine dust is selected. And, based on the distances between the sensors disposed in the plurality of regions, information on the proximity relationship between the sensors is generated, and the selected feature information and information on the generated proximity relationship are input to a pre-learned prediction model to determine a predetermined An apparatus for predicting volume of ultrafine dust comprising a control unit for predicting the volume of ultrafine dust for each of the plurality of regions after a lapse of time. According to claim 8,
The one or more particulate matter has a diameter of less than or equal to 0.1 μm, particulate matter greater than 0.1 μm and less than or equal to 0.3 μm, particulate matter greater than 0.3 μm and less than or equal to 0.5 μm, particulate matter greater than 0.5 μm and less than or equal to 1.0 μm, greater than 1.0 μm and less than or equal to 2.5 μm. It includes particulate matter of less than μm and particulate matter of more than 2.5 μm and less than or equal to 5.0 μm, wherein the environmental factor is carbon monoxide (CO), nitrogen dioxide (NO ₂ ), ozone (O ₃ ), sulfur dioxide (SO ₂ ), the plurality of regions Includes at least one of temperature, humidity, pressure, wind direction, wind speed and precipitation of,
The control unit,
A second configured to calculate a Pearson correlation coefficient between the one or more particulate matter and the ultrafine dust and between the environmental factor and the ultrafine dust, and select feature information for which the value of the calculated Pearson correlation coefficient is equal to or greater than a predetermined value. Fine dust volume prediction device. According to claim 9,
The control unit,
An adjacency relationship between sensors is generated based on a two-dimensional matrix based on the distance between sensors, and each element of the matrix is set to 0 or more and 1 or less, but when the distance between sensors is smaller than the reference distance, the adjacency distance value is less than the reference distance. An apparatus for predicting the volume of ultrafine dust configured to be set larger than the adjacent distance value in the case of a large case. According to claim 10,
The predictive model,
a first model learned to output time-sequential data by receiving the selected feature information and proximity relationship information according to distances between sensors disposed in the plurality of regions; and
An apparatus for predicting volume of ultrafine dust comprising a second model that receives an output value of the first model and is trained to transform the output value of the first model through attention processing. According to claim 11,
The first model,
a first structure for receiving the selected feature information and adjacent relationship information according to distances between sensors disposed in the plurality of regions and calculating them in a forward direction; and a second structure for calculating them in a reverse direction, wherein output values of the first structure and a second structure are calculated. An apparatus for predicting the volume of ultrafine dust, including a configuration for concatenating the output values of the two structures. According to claim 12,
The second model,
Ultrafine dust volume prediction device comprising a configuration for learning by placing a more important weight among the concatenated output values.

Images