KR102325137B1 - Apparatus and method for generating hybrid deep learning model for improving accuracy of fine dust prediction - Google Patents

Abstract

The present invention relates to an apparatus and method for generating a hybrid deep learning model to improve fine dust prediction accuracy. The apparatus collects fine dust measurement data, meteorological data, and gas data, pre-processes the collected data, analyzes how much each of the collected data affects a fine dust level using principal component analysis, calculates correlation coefficients between the collected data, selects data to be used for learning using the correlation coefficients between the collected data, learns with the selected data to build a hybrid deep learning model containing a one-dimensional convolutional neural network layer, a superposed autoencoder layer, and a superposed bidirectional LSTM layer, and predicts fine dust using the hybrid deep learning model.

Description

Apparatus and method for generating a hybrid deep learning model to improve fine dust prediction accuracy

The following embodiments relate to a technique for improving the accuracy of fine dust prediction.

As the level of fine dust that adversely affects health has increased in recent years, social interest in fine dust is increasing. Accordingly, many studies are being conducted on how to predict the fine dust level in advance to prepare for the increase or decrease of fine dust, rather than simply measuring the fine dust level.

One of them is a deep learning algorithm. A deep learning algorithm is a type of artificial intelligence algorithm that can predict future values by learning a neural network using learning data. However, existing deep learning algorithm models have a problem in that it is difficult to accurately predict the level of fine dust that is changed by various variables because it lacks consideration of various weather factors such as temperature, humidity, and wind speed.

There are two major problems with the deep learning models used in the existing fine dust level prediction research. The first is a very slow learning rate problem caused by large amounts of data being trained in a centralized deep learning structure. If a centralized structure is used, in the worst case, the learning rate cannot keep up with the changes in the data, so the model may need to be retrained. Second, it does not take into account the fact that there is noise in fine dust and meteorological data. Such noise can affect the accuracy of prediction because it interferes with extracting characteristics of the data.

An object of the present invention is to provide an apparatus and method for generating a distributed hybrid deep learning model for improving fine dust prediction accuracy.

According to an embodiment of the present invention, an apparatus for generating a hybrid deep learning model for improving fine dust prediction accuracy includes: a collecting unit configured to collect fine dust measurement data, weather data, and gas data; a pre-processing unit for pre-processing the collected data; Analyze how each of the collected data affects the fine dust level using the principal component analysis technique, calculate the correlation coefficient between the collected data, and use the correlation coefficient between the collected data to learn a data selection unit for selecting data to be used; a learning unit for learning with a hybrid deep learning model including a one-dimensional convolutional neural network layer, a nested autoencoder layer, and a nested bidirectional LSTM layer using the selected data; and a prediction unit for predicting fine dust using the hybrid deep learning model.

In this case, the preprocessor may preprocess by filling in the missing values in the collected data and standardizing the collected data.

At this time, when the data selection unit analyzes how much each of the collected data affects the fine dust level, it analyzes how much each of the collected data affects the fine dust level PM2.5 and PM10 can do.

In this case, if the correlation coefficient with PM2.5 or PM10, which is a value of fine dust among the collected data, is less than or equal to a preset value, the data selector may determine that the correlation is low and exclude it from data to be used.

In this case, the learning unit may use a preset amount of data among the selected data as learning data and use the remaining data as test data.

At this time, the learning unit processes fine dust measurement data among the selected data using a plurality of one-dimensional convolutional neural network layers, and receives the fine dust measurement data processed through the plurality of one-dimensional convolutional neural network layers as the input of the overlapping bidirectional LSTM layer. to perform interactive learning, learning in the overlapping autoencoder layer using fine dust measurement data of the selected data, weather data of the selected data, and gas data of the selected data, and learning result of the overlapping bidirectional LSTM layer and a hybrid deep learning model that predicts fine dust can be created by combining the learning results of the overlapping autoencoder layer.

A hybrid deep learning model generating method for improving fine dust prediction accuracy in a hybrid deep learning model generating apparatus according to an embodiment of the present invention includes: collecting fine dust measurement data, weather data, and gas data; pre-processing the collected data; analyzing how much each of the collected data affects the fine dust level by using a principal component analysis technique; calculating a correlation coefficient between the collected data; selecting data to be used for learning by using a correlation coefficient between the collected data; Learning with a hybrid deep learning model including a one-dimensional convolutional neural network layer, a nested autoencoder layer, and a nested bidirectional LSTM layer using the selected data; and predicting fine dust using the hybrid deep learning model.

In this case, the pre-processing of the collected data may include: filling in missing values in the collected data; and standardizing the collected data.

At this time, in the step of selecting data to be used for learning by using the correlation coefficient between the collected data, if the correlation coefficient with PM2.5 or PM10, which is a fine dust level, is less than a preset value, the correlation is low. It may include determining and excluding from data to be used.

At this time, the step of learning with a hybrid deep learning model including a one-dimensional convolutional neural network layer, an overlapping autoencoder layer, and an overlapping bidirectional LSTM layer using the selected data includes: processing using; performing interactive learning by using the fine dust measurement data processed through the plurality of one-dimensional convolutional neural network layers as an input of the overlapping bidirectional LSTM layer; learning in the superimposed autoencoder layer using fine dust measurement data of the selected data, weather data of the selected data, and gas data of the selected data; and generating a hybrid deep learning model for predicting fine dust by combining the learning result of the overlapping two-way LSTM layer and the learning result of the overlapping autoencoder layer.

The present invention predicts fine dust by generating a hybrid deep learning model including a one-dimensional convolutional neural network layer, an overlapping autoencoder layer, and an overlapping bidirectional LSTM layer using fine dust measurement data, weather data, and gas data. have

1 is a diagram illustrating the configuration of an apparatus for generating a hybrid deep learning model according to an embodiment of the present invention.
FIG. 2 is a diagram illustrating a configuration of generating a hybrid deep learning model for fine dust prediction using various types of deep learning techniques in the hybrid deep learning model generating apparatus according to an embodiment of the present invention.
3 is a diagram illustrating a degree of association between data collected in an apparatus for generating a hybrid deep learning model as a heat map according to an embodiment of the present invention.
4 is a diagram illustrating an example of an autoencoder according to an embodiment of the present invention.
5 is a diagram illustrating an example of an overlapping bidirectional LSTM layer according to an embodiment of the present invention.
6 is a flowchart illustrating a process of generating a hybrid deep learning model for improving fine dust prediction accuracy in an apparatus for generating a hybrid deep learning model according to an embodiment of the present invention.
7 is a flowchart illustrating a process of generating a hybrid deep learning model for fine dust prediction using various types of deep learning techniques in the hybrid deep learning model generating apparatus according to an embodiment of the present invention.

Hereinafter, embodiments will be described in detail with reference to the accompanying drawings. However, since various changes may be made to the embodiments, the scope of the patent application is not limited or limited by these embodiments. It should be understood that all modifications, equivalents and substitutes for the embodiments are included in the scope of the rights.

The terms used in the examples are used for the purpose of description only, and should not be construed as limiting. The singular expression includes the plural expression unless the context clearly dictates otherwise. In the present specification, terms such as “comprise” or “have” are intended to designate that a feature, number, step, operation, component, part, or combination thereof described in the specification exists, but one or more other features It should be understood that this does not preclude the existence or addition of numbers, steps, operations, components, parts, or combinations thereof.

Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which the embodiment belongs. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with the meaning in the context of the related art, and should not be interpreted in an ideal or excessively formal meaning unless explicitly defined in the present application. does not

In addition, in the description with reference to the accompanying drawings, the same components are given the same reference numerals regardless of the reference numerals, and the overlapping description thereof will be omitted. In the description of the embodiment, if it is determined that a detailed description of a related known technology may unnecessarily obscure the gist of the embodiment, the detailed description thereof will be omitted.

In addition, in describing the components of the embodiment, terms such as first, second, A, B, (a), (b), etc. may be used. These terms are only for distinguishing the components from other components, and the essence, order, or order of the components are not limited by the terms. When it is described that a component is “connected”, “coupled” or “connected” to another component, the component may be directly connected or connected to the other component, but another component is between each component. It will be understood that may also be "connected", "coupled" or "connected".

Components included in one embodiment and components having a common function will be described using the same names in other embodiments. Unless otherwise stated, a description described in one embodiment may be applied to another embodiment, and a detailed description in the overlapping range will be omitted.

Hereinafter, an apparatus and method for generating a distributed hybrid deep learning model for improving fine dust prediction accuracy according to an embodiment of the present invention will be described in detail with reference to FIGS. 1 to 7 .

1 is a diagram illustrating the configuration of an apparatus for generating a hybrid deep learning model according to an embodiment of the present invention.

Referring to FIG. 1 , the hybrid deep learning model generating apparatus 100 includes a controller 110 , a collector 111 , a preprocessor 112 , a data selector 113 , a learner 114 , and a predictor 115 . ), the communication unit 120 and the storage unit 130 may be configured to include.

The communication unit 120 is a communication interface device including a receiver and a transmitter, and the communication unit 120 transmits and receives data by wire or wirelessly. The communication unit 120 may receive fine dust measurement data, gas data, and meteorological data by communicating with a fine dust measuring device in each region or a place such as a meteorological office that measures meteorological data.

The storage unit 130 stores an operating system for controlling the overall operation of the hybrid deep learning model generating apparatus 100 , an application program, and data for storage. In addition, the storage unit 130 may store the fine dust measurement data, gas data, and weather data received according to the present invention for a preset period, and store the fine dust prediction data for a preset period.

The collection unit 111 may collect fine dust measurement data, weather data, and gas data received through the communication unit 120 . In this case, the meteorological data may include precipitation, temperature, humidity, and the like, and the gas data may include ozone, sulfur dioxide, and the like.

The preprocessor 112 may preprocess the collected data. In this case, the preprocessor 112 may preprocess by filling in the missing values in the collected data and standardizing the collected data. Here, as a method of filling in the value missing in the preprocessor 112 , a method of replacing the value with an average value within the same period may be used.

The data selection unit 113 analyzes how much each of the collected data affects the fine dust level by using a principal component analysis (PCA), calculates a correlation coefficient between the collected data, and collects Data to be used for learning can be selected by using the correlation coefficient between data sets.

When analyzing how much each of the collected data affects the fine dust level, the data selection unit 113 can analyze how much each of the collected data affects the fine dust level PM2.5 and PM10. have.

3 is a diagram illustrating a degree of association between data collected in an apparatus for generating a hybrid deep learning model as a heat map according to an embodiment of the present invention.

Referring to FIG. 3 , a correlation coefficient between characteristics of each of the collected data may be confirmed. Here, it can be seen that PM2.5 and PM10 have a high correlation, but since these two are different types of fine dust, in the present invention, when the model predicting PM2.5 is trained, PM10 is predicted without using PM10 data. PM2.5 data is not used when training the model.

It can be seen that the temperature (Temp) characteristic has a very low absolute value of the correlation coefficient. This means that temperature does not significantly affect fine dust levels. Accordingly, the temperature characteristic may not be used for learning.

In addition, since there is a low correlation between weather and gas data and each data is irreplaceable information, it can be used as it is for learning.

As described above, in the present invention, accuracy and learning speed of predicting fine dust levels can be improved by numerically accurately grasping and utilizing the effects of weather data and gas data on fine dust levels through the principal component analysis technique.

The data selection unit 113 may calculate a correlation coefficient between the collected data with reference to Equation 1 below.

[Equation 1]

Here, δ is the correlation coefficient, and a _i and b _i represent the data collected in the i-th monitoring area, respectively.

The closer the absolute value of the correlation coefficient δ to 1, the higher the correlation between the two variables.

If the correlation coefficient with PM2.5 or PM10, which is a value of fine dust among the collected data, is less than or equal to a preset value, the data selection unit 113 may determine that the correlation is low and exclude it from data to be used. The principal component analysis function of the data selection unit 113 has the effect of improving the accuracy of the deep learning learning model when the amount of data is not large enough. Since the environmental impact on fine dust varies from region to region, it is necessary to collect data for a very long time in that region to predict the level of fine dust in a certain region. Since this is practically impossible, the principal component analysis function is very useful.

The learning unit 114 may learn by using the selected data as a hybrid deep learning model including a one-dimensional convolutional neural network layer, a nested autoencoder layer, and an overlapped bidirectional LSTM layer.

The learning unit 114 may use a preset amount of data among the selected data as training data and use the remaining data as test data.

More specifically, the learning unit 114 processes fine dust measurement data from among the selected data using a plurality of one-dimensional convolutional neural network layers, and superimposes fine dust measurement data processed through a plurality of one-dimensional convolutional neural network layers. Input of the bidirectional LSTM layer to learn the characteristics of time series data in both directions, and extract the features using the superimposed autoencoder using the fine dust measurement data of the selected data, the weather data of the selected data, and the gas data of the selected data, and overlap the bidirectional LSTM layer. A hybrid deep learning model that predicts fine dust can be created by combining it with the learning results of

The learning unit 114 may generate a hybrid deep learning model for predicting fine dust as shown in the example of FIG. 2 below.

FIG. 2 is a diagram illustrating a configuration of performing learning for fine dust prediction using various types of deep learning techniques in the hybrid deep learning model generating apparatus according to an embodiment of the present invention.

Referring to FIG. 2 , the learning unit 114 may include a plurality of one-dimensional convolutional neural network layers 210 , an overlapping bidirectional LSTM layer 220 , an overlapping autoencoder layer 230 , and a combiner 240 . .

The overlapping autoencoder layer 230 extracts features of fine dust measurement data of the selected data, weather data of the selected data, and gas data of the selected data.

4 is a diagram illustrating an example of an autoencoder according to an embodiment of the present invention.

Referring to FIG. 4, the autoencoder is largely divided into two stages: the encoding unit 410 and the decoding unit 420. The encoding unit 410 converts the input data 401 to the hidden data 402 through unsupervised learning. By encoding , abstract features are extracted, and the decoding unit 420 can reconstruct the original data 403 through these features.

이고,

는 인코더 함수이다. 그리고 디코더 함수는

로 나타낼 수 있다. 특징 벡터 h는 원본 입력 데이터의 압축된 표현으로 아래 <수학식 2>와 같을 수 있다.The feature learning function of the autoencoder's network is

ego,

is the encoder function. And the decoder function is

can be expressed as The feature vector h is a compressed representation of the original input data, and may be as in Equation 2 below.

[Equation 2]

는 i번째 모니터링 지역의 가스 데이터와 기상 데이터의 특징이다. 따라서 인코딩 함수

는 입력 데이터의 숨겨진 PM2.5와 PM10의 핵심 정보에 대해 학습하였으므로 미세먼지 학습 모델링을 보조할 수 있다. here,

is a characteristic of the gas data and meteorological data of the i-th monitoring area. So the encoding function

learned about the core information of PM2.5 and PM10 hidden in the input data, so it can assist with fine dust learning modeling.

는 오토인코더의 가중치 제한 변수로서 PM2.5와 PM10의 예측에 관련이 적은 요소들이나 기상학적 정보를 적게 담고 있는 요소들의 가중치를 제한함으로써 그 특성들을 무시할 수 있다.

As a weight limiting variable of the autoencoder, its characteristics can be ignored by limiting the weights of factors that are not related to the prediction of PM2.5 and PM10 or that contain little meteorological information.

Conventionally, it has failed to simultaneously model PM2.5 time series data, weather data, and gas data, but in the present invention, it is modeled using an overlapping autoencoder layer. Through this, it is possible to learn more about the effects of weather data and gas data on fine dust time series data and to improve prediction accuracy.

The plurality of one-dimensional convolutional neural network layers 210 processes fine dust measurement data among selected data using a plurality of one-dimensional convolutional neural network layers.

The overlapped bidirectional LSTM layer 220 performs interactive learning by using fine dust measurement data processed through a plurality of one-dimensional convolutional neural network layers as an input of the overlapping bidirectional LSTM layer.

Since the fine dust data is time series data, a time-dependent modeling method is taken when implementing the prediction algorithm. Previously, feed-forward neural networks (FNN) models were widely used to predict PM2.5 values, but the FNN model did not consider the time dependence problem, so accurate prediction was impossible. Since then, RNN (Recuurent Neural Networks) algorithm has been proposed, but RNN also shows poor performance in prediction problems with large time series data. because problems arise.

Although fine dust data is time series data that flows in one direction, it has unit temporal patterns such as daily, weekly, and monthly units. If these characteristics are used for prediction, it will be more effective to predict fine dust on a specific day if not only the fine dust pattern of the previous day but also the fine dust pattern of the next day is used. However, since the basic LSTM structure learns in only one direction, this characteristic of fine dust data cannot be utilized. Therefore, learning in an additional direction is required for more accurate prediction.

The present invention may use the overlapping bidirectional LSTM layer of FIG. 5 below.

5 is a diagram illustrating an example of an overlapping bidirectional LSTM layer according to an embodiment of the present invention.

Referring to FIG. 5, the overlapping bidirectional LSTM layer uses forward LSTMs 521, 523, and 525 and backward LSTMs 531, 533, and 535 to transmit input data 511, 513, and 525 without affecting each other. Learning is conducted in both directions independently, and after learning is completed, the output data 541 , 543 , and 545 are generated by being integrated.

Through bidirectional learning as shown in FIG. 5 , a node can obtain information from past and future data, and through this, it is possible to utilize a pattern from the reverse direction rather than based only on the forward pattern. In this way, it is possible to solve the problems that occur in the basic LSTM of one direction and improve the accuracy of prediction.

The combiner 240 may generate a hybrid deep learning model for predicting fine dust by combining the learning result of the overlapping bidirectional LSTM layer and the learning result of the overlapping autoencoder layer. In other words, using the characteristics of gas data and weather data extracted by the autoencoder, gas data and weather data of a future point in time that have not yet arrived are created, and the fine dust prediction algorithm learned by LSTM based on this predicts fine dust.

The prediction unit 115 may predict fine dust using the generated hybrid deep learning model.

The controller 110 may control the overall operation of the hybrid deep learning model generating apparatus 100 . In addition, the control unit 110 may perform the functions of the collecting unit 111 , the preprocessing unit 112 , the data selecting unit 113 , the learning unit 114 , and the predicting unit 115 . The control unit 110 , the collection unit 111 , the preprocessor 112 , the data selection unit 113 , the learning unit 114 , and the prediction unit 115 are separately illustrated to describe each function separately. . Therefore, the control unit 110 is at least one processor configured to perform the respective functions of the collecting unit 111 , the preprocessing unit 112 , the data selecting unit 113 , the learning unit 114 , and the predicting unit 115 . may include. In addition, the control unit 110 is configured to perform at least some of the functions of the collection unit 111 , the preprocessor 112 , the data selection unit 113 , the learning unit 114 , and the prediction unit 115 , respectively. It may include one processor.

Hereinafter, a method according to the present invention configured as described above will be described with reference to the drawings below.

6 is a flowchart illustrating a process of generating a hybrid deep learning model for improving fine dust prediction accuracy in an apparatus for generating a hybrid deep learning model according to an embodiment of the present invention.

Referring to FIG. 6 , the hybrid deep learning model generating apparatus 100 collects fine dust measurement data, weather data, and gas data ( 610 ).

Then, the hybrid deep learning model generating apparatus 100 pre-processes the collected data ( 620 ). In this case, the hybrid deep learning model generating apparatus 100 may fill in the missing values in the collected data and standardize the collected data.

Then, the hybrid deep learning model generating apparatus 100 analyzes how much each of the data collected using the principal component analysis technique affects the fine dust level ( 630 ). At this time, the hybrid deep learning model generating apparatus 100 analyzes how much each of the collected data affects PM2.5 and PM10 which are fine dust values, and PM2.5 or PM10 which is the fine dust values among the collected data. If the correlation coefficient is less than or equal to a preset value, it is determined that the correlation is low and may be excluded from data to be used.

Then, the hybrid deep learning model generating apparatus 100 calculates a correlation coefficient between the collected data ( 640 ).

Then, the hybrid deep learning model generating apparatus 100 selects data to be used for learning by using a correlation coefficient between the collected data ( 650 ).

Then, the hybrid deep learning model generating apparatus 100 generates a hybrid deep learning model including a one-dimensional convolutional neural network layer, a nested autoencoder layer, and an overlapped bidirectional LSTM layer using the selected data ( 660 ). In this case, the hybrid deep learning model generating apparatus 100 may use a preset amount of data among the selected data as training data and use the remaining data as test data.

Then, the hybrid deep learning model generating apparatus 100 predicts fine dust using the generated hybrid deep learning model ( 670 ).

7 is a flowchart illustrating a process of generating a hybrid deep learning model for fine dust prediction using various types of deep learning techniques in the hybrid deep learning model generating apparatus according to an embodiment of the present invention.

Referring to FIG. 7 , the hybrid deep learning model generating apparatus 100 processes fine dust measurement data among selected data using a plurality of one-dimensional convolutional neural network layers ( 710 ).

Then, the hybrid deep learning model generating apparatus 100 performs interactive learning by using fine dust measurement data processed through a plurality of one-dimensional convolutional neural network layers as an input of the overlapping bidirectional LSTM layer (720).

Then, the hybrid deep learning model generating apparatus 100 learns from the superimposed autoencoder layer using the fine dust measurement data of the selected data, the weather data of the selected data, and the gas data of the selected data ( 730 ).

Then, the hybrid deep learning model generating apparatus 100 generates a hybrid deep learning model for predicting fine dust by combining the learning result of the overlapping bidirectional LSTM layer and the learning result of the overlapping autoencoder layer ( 740 ).

The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, etc. alone or in combination. The program instructions recorded on the medium may be specially designed and configured for the embodiment, or may be known and available to those skilled in the art of computer software. Examples of the computer-readable recording medium include magnetic media such as hard disks, floppy disks and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic such as floppy disks. - includes magneto-optical media, and hardware devices specially configured to store and carry out program instructions, such as ROM, RAM, flash memory, and the like. Examples of program instructions include not only machine language codes such as those generated by a compiler, but also high-level language codes that can be executed by a computer using an interpreter or the like. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

Software may comprise a computer program, code, instructions, or a combination of one or more thereof, which configures a processing device to operate as desired or is independently or collectively processed You can command the device. The software and/or data may be any kind of machine, component, physical device, virtual equipment, computer storage medium or device, to be interpreted by or to provide instructions or data to the processing device. , or may be permanently or temporarily embody in a transmitted signal wave. The software may be distributed over networked computer systems, and stored or executed in a distributed manner. Software and data may be stored in one or more computer-readable recording media.

As described above, although the embodiments have been described with reference to the limited drawings, those skilled in the art may apply various technical modifications and variations based on the above. For example, the described techniques are performed in a different order than the described method, and/or the described components of the system, structure, apparatus, circuit, etc. are combined or combined in a different form than the described method, or other components Or substituted or substituted by equivalents may achieve an appropriate result.

Therefore, other implementations, other embodiments, and equivalents to the claims are also within the scope of the following claims.

100: Hybrid deep learning model generation device
110: control unit
111: collection unit
112: preprocessor
113: data selection unit
114: study department
115: prediction unit
120: communication department
130: storage

Claims (10)

a collection unit for collecting fine dust measurement data, weather data, and gas data;
a pre-processing unit for pre-processing the collected data;
Analyze how each of the collected data affects the fine dust level using the principal component analysis technique, calculate the correlation coefficient between the collected data, and use the correlation coefficient between the collected data to learn a data selection unit for selecting data to be used;
a learning unit for learning with a hybrid deep learning model including a one-dimensional convolutional neural network layer, a nested autoencoder layer, and a nested bidirectional LSTM layer using the selected data; and
Prediction unit that predicts fine dust using the hybrid deep learning model
including,
The learning unit,
Process fine dust measurement data among the selected data using a plurality of one-dimensional convolutional neural network layers,
Bidirectional learning is performed by using the fine dust measurement data processed through the plurality of one-dimensional convolutional neural network layers as an input of the overlapping bidirectional LSTM layer,
Learning in the superimposed autoencoder layer using fine dust measurement data of the selected data, weather data of the selected data, and gas data of the selected data,
Creating a hybrid deep learning model that predicts fine dust by combining the learning result of the overlapping bidirectional LSTM layer and the learning result of the overlapping autoencoder layer
Hybrid deep learning model generation device to improve fine dust prediction accuracy.
According to claim 1,
The preprocessor is
Pre-processing by filling in the missing values in the collected data and standardizing the collected data
Hybrid deep learning model generation device to improve fine dust prediction accuracy.
According to claim 1,
The data selection unit,
When analyzing how much each of the collected data affects the fine dust level, it is a method to analyze how much each of the collected data affects the fine dust level PM2.5 and PM10.
Hybrid deep learning model generation device to improve fine dust prediction accuracy.
According to claim 1,
The data selection unit,
Among the collected data, if the correlation coefficient with PM2.5 or PM10, which is the value of fine dust, is less than a preset value, it is determined that the correlation is low and excluded from the data to be used.
Hybrid deep learning model generation device to improve fine dust prediction accuracy.
According to claim 1,
The learning unit,
Among the selected data, a preset amount of data is used as training data, and the remaining data is used as test data.
Hybrid deep learning model generation device to improve fine dust prediction accuracy.
delete collecting fine dust measurement data, weather data, and gas data;
pre-processing the collected data;
analyzing how much each of the collected data affects the fine dust level by using a principal component analysis technique;
calculating a correlation coefficient between the collected data;
selecting data to be used for learning by using a correlation coefficient between the collected data;
Learning with a hybrid deep learning model including a one-dimensional convolutional neural network layer, a nested autoencoder layer, and a nested bidirectional LSTM layer using the selected data; and
Predicting fine dust using the hybrid deep learning model
including,
The step of learning with a hybrid deep learning model including a one-dimensional convolutional neural network layer, a nested autoencoder layer, and a nested bidirectional LSTM layer using the selected data includes:
processing fine dust measurement data among the selected data using a plurality of one-dimensional convolutional neural network layers;
performing interactive learning by using the fine dust measurement data processed through the plurality of one-dimensional convolutional neural network layers as an input of the overlapping bidirectional LSTM layer;
learning in the superimposed autoencoder layer using fine dust measurement data of the selected data, weather data of the selected data, and gas data of the selected data; and
Generating a hybrid deep learning model for predicting fine dust by combining the learning result of the overlapping bidirectional LSTM layer and the learning result of the overlapping autoencoder layer
containing
Hybrid deep learning model generation method to improve fine dust prediction accuracy in hybrid deep learning model generation device.
8. The method of claim 7,
The step of pre-processing the collected data is,
filling in missing values in the collected data; and
Standardization of the collected data
A hybrid deep learning model generation method for improving the accuracy of fine dust prediction in a hybrid deep learning model generation device comprising a.
8. The method of claim 7,
The step of selecting data to be used for learning by using the correlation coefficient between the collected data includes:
Determining that the correlation with PM2.5 or PM10, which is a value of fine dust among the collected data, is less than a preset value, determining that the correlation is low, and excluding from the data to be used
A hybrid deep learning model generation method for improving the accuracy of fine dust prediction in a hybrid deep learning model generation device comprising a.
delete

Images