KR102546540B1 - Method and apparatus for prediction of traffic congestion based on lstm - Google Patents

Abstract

The present invention relates to an LSTM-based traffic congestion prediction method, and more particularly, to an LSTM-based traffic congestion prediction method and apparatus for predicting LSTM-based traffic congestion and missing imputation using spatio-temporal feature extraction from traffic flow data. According to an embodiment of the present invention, the LSTM-based traffic congestion predicting apparatus may learn time-series characteristics of traffic data in consideration of the aspect of order or time. According to another embodiment of the present invention, the LSTM-based traffic congestion prediction apparatus can provide highly accurate prediction by reducing factors that act as variables during prediction because the outskirts of the city have a smaller influence on traffic flow due to external disturbances than the downtown area. can

Description

LSTM-based traffic congestion prediction method and apparatus {METHOD AND APPARATUS FOR PREDICTION OF TRAFFIC CONGESTION BASED ON LSTM}

The present invention relates to an LSTM-based traffic congestion prediction method, and more particularly, to an LSTM-based traffic congestion prediction method and apparatus for predicting LSTM-based traffic congestion and missing imputation using spatio-temporal feature extraction from traffic flow data.

Recently, smart cars are being produced in various forms based on the core technologies of the 4th industrial revolution. The role of the car is changing from a simple means of transportation to a living space, and it is changing to a form of infotainment that provides new convenience values. As the demand for smart cars increases, traffic information collection and processing for smooth road management is a very important area, so a qualitative rather than quantitative approach is required. To this end, research on an intelligent transportation system (ITS) using IT technology in the existing transportation system is being conducted.

The study of problems arising from the increase in demand for road resources is a major study that is the basis of transportation welfare that promotes user convenience. Transportation welfare consists of various factors such as operating cost, travel time, accident cost, parking cost, punctuality, and accessibility, and traffic congestion accounts for a high proportion. Accordingly, the function of the road traffic network is improved by collecting traffic information on all sections of the road in real time and providing various information such as congested sections, traffic volume, and traffic accident status for the intelligent traffic system.

In addition, ITS operates a real-time traffic information service. Since it is used in the form of providing the driver with the optimal route to prevent aggravation of congestion on congested roads and distributing the traffic volume on the road, the accuracy is relatively low due to the emphasis on speed. Therefore, in recent years, studies on real-time traffic situation pattern prediction through deep learning models and various prediction models to solve these problems have been actively conducted, and traffic volume prediction studies based on time-series data have been conducted.

In addition, a lot of data is being generated due to the Internet of Things and sensors, increased tracking and collection of customer data by companies, increased unstructured data due to the spread of social network services, and the development of storage media technology and price declines. Currently, among the representative technologies of the 4th industrial revolution, research on big data, artificial intelligence, machine learning, and deep learning, where data plays a key role, is becoming active. In these studies, the performance of the model is determined by the quantity and quality of the data. The quantitative aspect of the data is being resolved, but in the process of being collected in reality, there are many cases where values are missing or strange values are stored for a series of reasons, so the qualitative aspect of the data is degraded. This leads to difficulties in data analysis, unbalanced data structures, and reduced predictive performance of models. Therefore, it is necessary to apply algorithms to replace missing values and predictive techniques using statistical methods.

1. Korean Patent Publication No. 10-2013-0158919 “Traffic flow prediction system using space-time probability model” (published on June 30, 2015)

The present invention is an LSTM-based traffic congestion prediction method that predicts continuous traffic data with patterns preserved through an LSTM model suitable for the structure of time-series data by removing outliers from traffic data through missing imputation using spatio-temporal feature extraction from traffic data and devices.

According to one aspect of the present invention, an LSTM-based traffic congestion predicting device is provided.

An apparatus for predicting traffic congestion based on LSTM according to an embodiment of the present invention includes a data collection unit for collecting traffic data, a data correction unit for correcting the collected traffic data, and an adversarial auto encoder based on the corrected traffic data ( AAE) based missing imputation model and a prediction unit that predicts traffic congestion based on a time series based deep learning model (LSTM) based on the preprocessed traffic data.

According to one aspect of the present invention, a method for predicting traffic congestion based on LSTM and a computer program executing the same are provided.

An LSTM-based traffic congestion prediction method according to an embodiment of the present invention includes the steps of collecting traffic data, performing data correction on the collected traffic data, and adversarial auto encoder (AAE) based on the corrected traffic data. The method may include performing data preprocessing according to a missing imputation model and predicting traffic congestion through a time series-based deep learning model based on the preprocessed traffic data.

According to an embodiment of the present invention, the LSTM-based traffic congestion predicting apparatus may learn time-series characteristics of traffic data in consideration of the aspect of order or time.

According to another embodiment of the present invention, the LSTM-based traffic congestion prediction apparatus can provide highly accurate prediction by reducing factors that act as variables during prediction because the outskirts of the city have a smaller influence on traffic flow due to external disturbances than the downtown area. can

The effects of the present invention are not limited to the above effects, and should be understood to include all effects that can be inferred from the description of the present invention or the configuration of the invention described in the claims.

1 is a diagram for explaining a process of an intelligent transportation system according to an embodiment of the present invention.
2 is a diagram illustrating an apparatus for predicting traffic congestion based on LSTM according to an embodiment of the present invention.
3 and 4 are diagrams for explaining the elimination of outliers in the LSTM-based traffic congestion prediction method according to an embodiment of the present invention.
5 and 6 are diagrams for explaining missing value correction according to the spatial trend utilization method in the LSTM-based traffic congestion prediction method according to an embodiment of the present invention.
7 is a diagram for explaining missing value correction according to the time trend utilization method in the LSTM-based traffic congestion prediction method according to an embodiment of the present invention.
8 is a diagram for explaining an adversarial auto-encoder (AAE)-based missing data imputation model in the LSTM-based traffic congestion prediction method according to an embodiment of the present invention.
9 is a diagram for explaining a time-series-based deep learning model in an LSTM-based traffic congestion prediction method according to an embodiment of the present invention.
10 and 11 are diagrams for explaining an apparatus for predicting traffic congestion based on LSTM according to an embodiment of the present invention.
12 is a flowchart illustrating a method for predicting traffic congestion based on LSTM according to an embodiment of the present invention.

Hereinafter, the present invention will be described with reference to the accompanying drawings. However, the present invention may be embodied in many different forms and, therefore, is not limited to the embodiments described herein. And in order to clearly explain the present invention in the drawings, parts irrelevant to the description are omitted, and similar reference numerals are attached to similar parts throughout the specification.

Throughout the specification, when a part is said to be "connected (connected, contacted, combined)" with another part, this is not only "directly connected", but also "indirectly connected" with another member in between. "Including cases where 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.

Terms used in this specification are only used to describe specific embodiments, and are not intended to limit the present invention. Singular expressions include plural expressions unless the context clearly dictates otherwise. In this specification, terms such as "include" or "have" are intended to indicate that there is a feature, number, step, operation, component, part, or combination thereof described in the specification, but one or more other features It should be understood that the presence or addition of numbers, steps, operations, components, parts, or combinations thereof is not precluded.

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

1 is a diagram for explaining a process of an Intelligent Transport Systems (ITS) according to an embodiment of the present invention.

Referring to FIG. 1 , an intelligent transportation system (ITS) is a transportation system that improves the efficiency and safety of transportation through the scientific and automation of operation and management of the transportation system through technology convergence.

Intelligent transportation systems (ITS) include a bus stop arrival information system, a signal processing system according to traffic volume at an intersection, real-time traffic information of navigation, and high-pass.

An intelligent traffic system (ITS) may collect traffic information from traffic information collection devices installed on roads or roadsides. At this time, the traffic information collection device may include a CCTV, vehicle detection, GPS, and weather information system. Information collected by the traffic information collecting device may be transmitted to a traffic data management center through wired or wireless communication.

The traffic data management center secures, analyzes, and processes the collected information for use in traffic facilities. Analytically processed data may be provided for traffic management. The traffic data management center can control traffic conditions and deliver information to users. The Intelligent Transportation System (ITS) can collect and provide traffic information and services to means of transportation and transportation facilities using advanced technologies such as electronic control and communication.

Depending on the embodiment, the traffic data management center may provide information to a traffic information display board (VMS: Variable Message Sign) through the Internet. The traffic data management center can provide information to traffic signal controllers and mobile devices. In addition, traffic management can be performed through operation/logistics management.

2 is a diagram illustrating an apparatus for predicting traffic congestion based on LSTM according to an embodiment of the present invention.

Referring to FIG. 2 , the LSTM-based traffic congestion prediction apparatus 100 may include a data collection unit 110, a data correction unit 120, a pre-processing unit 130, and a prediction unit 140.

The data collection unit 110 may collect data for providing traffic conditions to the user. At this time, the traffic data may include node/link data and traffic data provided by an intelligent traffic system (ITS).

The node/link data may include data of a section of a road or a point where the road is connected.

Traffic data may be collected through traffic information collection equipment on roads or roadsides. Traffic data may include missing values and outliers due to the inability to collect information due to errors in the collection equipment or the occurrence of shadow areas caused by no cars on the road, and outliers generated due to speeding.

Depending on the embodiment, traffic data is collected every 5 minutes, and the collected data is interval missing if the entire data is not collected in the corresponding 5 minutes, and regional missing if there are parts not collected for each road. (Location Missing), and Missing Days (Days Missing), in which an entire day was not collected.

Depending on the embodiment, the traffic data may include outliers that are not missing values and represent excessively small or large values that deviate from the average traffic flow, although the average traffic flow exists for each road. In other words, traffic speed data has outliers that distort the flow of average traffic speed and missing data that have not been collected. An outlier represents an excessively small or large value that deviate from the average traffic flow even though there is an average traffic flow for each road.

Depending on the embodiment, the traffic data may include temporal missing if the traffic data should be collected every 5 minutes but the entire data corresponding to 5 minutes is not collected, and the traffic data is collected every 5 minutes but not collected by road. If there is a part, spatial missing may be included.

The data correction unit 120 may perform data preprocessing for extracting feature values for predicting traffic congestion through a deep learning model by processing outliers and missing values. To this end, the data correction unit 120 may remove outliers in the data through data filtering. In addition, the data correction unit 120 may correct missing values in traffic data having time-series and spatial characteristics.

According to the embodiment, the data compensator 120 may use the speed data of 5 minutes ago, the average speed of 5 minutes ago, the current speed, and the adjacent upstream speed data as input data during the construction of the deep learning model using the preprocessed data. , the output data can use the speed data after 5 minutes.

When the preprocessor 130 performs data missing correction, the data missing based on Adversarial Auto Encoder (AAE) using a time convolution layer and a graph convolution layer. Imputation models can be performed.

Adversarial Auto Encoder (AAE) is a model that can generate data through the structure of Encoder and Decoder, Variational Auto Encoder (VAE), Generator, and Discriminator. ), and may be a model created by combining the advantages of Generative Adversarial Networks (GANs).

The prediction unit 140 may predict traffic congestion through a time-series-based deep learning model with pre-processed data. In this case, the time series-based deep learning model may use LSTM (Long Short-Term Memory).

3 and 4 are diagrams for explaining the elimination of outliers in the LSTM-based traffic congestion prediction method according to an embodiment of the present invention.

The data correction unit 120 may first remove anomalies in order to correct anomalies and missing values in the traffic data. Methods for removing outliers may include median absolute deviation, trimmed average, and Winsorized average. The data correction unit 120 may use or combine appropriate methods according to characteristics of each road and traffic data.

The data correction unit 120 may perform the median absolute deviation method using the algorithm shown in FIG. 3 to remove outliers. The median absolute deviation is a type of absolute deviation and is a method of detecting abnormally large or small values using the median value of collected data. The data correction unit 120 may consider values detected in the data collected through the median absolute deviation method as outliers and remove them.

Traffic data in FIG. 4 is data in the form of a link-time matrix including outliers and each missing value. The vertical axis of the traffic data matrix represents the link ID (LINK_ID), and the horizontal axis represents the time (TIME).

The data correction unit 120 determines that the value of (t2, L4) is 74, the value of (t3, L1) is 80, and the value of (t3, L5) is 77 in the traffic data matrix, using the median value as abnormally large. Therefore, it can be judged as an outlier. Thereafter, the data correction unit 120 may remove (NA) the value of the corresponding outlier from the traffic data matrix.

5 and 6 are diagrams for explaining missing value correction according to the spatial trend utilization method in the LSTM-based traffic congestion prediction method according to an embodiment of the present invention.

The data correction unit 120 may perform missing value correction using the spatial trend utilization method according to the algorithm shown in FIG. 5 . The data compensator 120 may correct missing data on the premise that sections having similar traffic patterns, that is, the flow of traffic in the upstream area directly affects the flow of traffic in the downstream area. The data correction unit 120 may perform correction through an average of data collected from proximity traffic detectors, which are adjacent links, when data is omitted due to an error in the traffic detector.

Referring to FIG. 6, the data correction unit 120 calculates (t2,L4), (t3,L1), (t3,L4), (t3,L5), (t3,L4) by spatial missing in the traffic data matrix. , (t4,L3), (t4,Ln), (tn,L3), and (tn,Ln) data values can be corrected with an average value of data collected from adjacent traffic detectors.

7 is a diagram for explaining missing value correction according to the time trend utilization method in the LSTM-based traffic congestion prediction method according to an embodiment of the present invention.

The data correction unit 120 may perform missing value correction using a time trend utilization method. The data compensator 120 may estimate data up to the missing point by averaging n data from the point in time at which the missing data is detected through the time trend utilization method to the previous point in time.

The data correction unit 120 may estimate the missing value as shown in Equation 1 through a correction equation using a time trend.

은 기간 t-k의 검지 데이터, n은 과거 이용 데이터의 검지 주기수이다.Here, Ft is the missing data estimate of the current period t,

is the detection data of the period tk, and n is the number of detection cycles of past use data.

Referring to FIG. 7 , when a time point of t1 data is missing due to time missing in the traffic data matrix, the data correction unit 120 may correct the missing value based on data of a previous time point.

8 is a diagram for explaining an adversarial auto-encoder (AAE)-based missing data imputation model in the LSTM-based traffic congestion prediction method according to an embodiment of the present invention.

Adversarial autoencoder (AAE) can use the structure of adversarial autoencoder (AAE) model for missing imputation of traffic data. The pre-processing unit 130 adds a time convolution layer and a graph convolution layer to the encoder and decoder to learn the spatio-temporal characteristics of traffic data to the encoder and decoder, and In order to extract each feature from the convolution operation layer and the graph-based convolution operation layer, the operation can be performed through a block that combines features by stacking layers.

An adversarial autoencoder (AAE) can perform Hadamard multiplication of a mask vector, random noise (z), and a vector (x), which is missing data. The calculated value (X) can be input to the encoder as input data to model parameters of the latent distribution. The decoder can sample from the latent space to predict missing values. The discriminator distinguishes between the imputed data and the observed value and can proceed with learning.

를 복원한다. 적대적 오토 인코더(AAE)는 연속적인 잠재표현 벡터를 만들기 위해 사전분포 p(z)를 먼저 정의하고 판별자(Discriminator)를 사용하여 학습하고, 잠재표현 벡터 z를 입력으로 받아 사전분포 p(z)와 비교하며 학습을 진행할 수 있다.That is, in the preprocessor 130, the encoder part of the variational autoencoder (VAE) serves as a generator, receives data x, samples a latent vector z, and the generator's decoder regenerates from it.

restore The adversarial autoencoder (AAE) first defines a prior distribution p(z) to create a continuous latent expression vector, learns using a discriminator, and receives the latent expression vector z as input to obtain a prior distribution p(z) You can compare and learn.

9 is a diagram for explaining a time-series-based deep learning model in an LSTM-based traffic congestion prediction method according to an embodiment of the present invention.

Referring to FIG. 9 , the structure of an LSTM cell is composed of a memory cell and a gate. Input information is stored in the memory cell, and the gate may be used to control information stored in the cell.

:Input Gate), 포겟 게이트 (

:Forget Gate), 아웃풋 게이트(

:Output Gate)를 포함할 수 있다. 즉, 인풋 게이트는 현재 시간 t에서 받은 입력 데이터를 제어하고, 아웃풋 게이트는 현재 시간 t에서 출력되는 출력 데이터를 제어하고, 포겟 게이트는 현재 시간 t에서 데이터의 보존 여부를 제어할 수 있다. 따라서, LSTM 모델에 기반을 둔 교통 혼잡도 예측 방법은 인풋 게이트를 통해 입력 데이터를 제어하고, 아웃풋 게이트를 통해 출력 데이터를 제어하고, 포겟 게이트를 통해 데이터의 보존 여부를 제어하는 시계열 기반 딥러닝 모델을 통해 교통 혼잡도를 예측할 수 있다.At this time, the gate of the LSTM cell is the input gate (

:Input Gate), forget gate (

:Forget Gate), output gate (

:Output Gate). That is, the input gate can control input data received at the current time t, the output gate can control the output data output at the current time t, and the forget gate can control whether to preserve data at the current time t. Therefore, the traffic congestion prediction method based on the LSTM model is a time series-based deep learning model that controls input data through an input gate, output data through an output gate, and controls whether or not to preserve data through a forget gate. traffic congestion can be predicted.

,

,···,

)에 기초하여 아웃풋

을 출력할 수 있다. LSTM#2은 제2시간(t-1)의 데이터 (

,

···,

)에 기초하여 아웃풋

을 출력할 수 있다. LSTM#3은 제3시간(t)에 데이터

,

와 이전 LSTM의 출력인

및

에 어텐션(Attention) 메커니즘을 사용하여 예상 아웃풋을 출력할 수 있다.In the LSTM model, LSTM#1 is the data of the first time (t-|p|+2) (

,

,···,

) based on the output

can output LSTM#2 is the data of the second time (t-1) (

,

...,

) based on the output

can output LSTM#3 is the data at the third time (t)

,

and the output of the previous LSTM

and

Expected output can be output using the Attention mechanism.

Depending on the embodiment, the input data of the construction of the LSTM model uses the average speed of 10 minutes ago and the speed of 5 minutes ago, the current speed, and the speed data of the adjacent upstream. Output data can use speed data after 5 minutes.

10 and 11 are diagrams for explaining an apparatus for predicting traffic congestion based on LSTM according to an embodiment of the present invention.

As shown in FIG. 10, the prediction device may perform a preprocessing process for LSTM-based traffic congestion prediction.

To this end, the LSTM-based traffic congestion prediction apparatus 100 may display speed data for each section of the predicted section in a table. The LSTM-based traffic congestion prediction apparatus 100 may perform preprocessing according to the operation of the start button after a link ID (LINK_ID) is input, at least one required preprocessing process is selected from outlier removal and missing replacement.

In addition, the LSTM-based traffic congestion prediction apparatus 100 may display the data of the pre-processed section as a graph, and may also store the pre-processed data through a save button.

Referring to FIG. 11 , the LSTM-based traffic congestion prediction apparatus 100 may predict traffic congestion based on preprocessed data.

In the LSTM-based traffic congestion prediction apparatus 100, a section can be selected from the section selection window at the top left. The LSTM-based traffic congestion prediction device 100 corresponds to the selected section, the description of the selected section (LINK_ID, LINK_NAME, Velocity) in the link overview window, and the speed by date of the selected section in the data collection window Collection status can be displayed.

The LSTM-based traffic congestion prediction device 100 displays the overall congestion result of the corresponding section when the forecast after 5 minutes is selected in the Status window and the Predict button operates, and the congestion degree is smooth (green), complex (orange) ), can be expressed as very complex (red). In this case, the degree of congestion may be set to a criterion of 30 Km/h or more for smooth, 15 Km/h to 30 Km/h for complexity, and less than 15 Km/h for very complex. In addition, numerical information of sections in which congestion is expected may be displayed in a table.

12 is a flowchart illustrating a method for predicting traffic congestion based on LSTM according to an embodiment of the present invention.

Referring to FIG. 12 , in step S10, the LSTM-based traffic congestion prediction device may collect traffic data.

In step S20, the LSTM-based traffic congestion prediction apparatus 100 may remove outliers from the collected traffic data.

In step S30, the LSTM-based traffic congestion prediction apparatus 100 may correct missing values of the traffic data using at least one of a spatial trend utilization method and a temporal trend utilization method.

In step S40, the LSTM-based traffic congestion prediction apparatus 100 may perform an adversarial auto-encoder (AAE)-based missing imputation model based on the corrected traffic data. The LSTM-based traffic congestion prediction device 100 performs a spatial trend utilization method using data of sections with similar traffic patterns and a temporal trend utilization method that corrects the current value through past data, leaving spatial and temporal perspectives different and missing data in the data. Values can be compensated for each other.

In step S50, the LSTM-based traffic congestion prediction apparatus 100 may predict traffic congestion based on a time-series based deep learning model (LSTM; Long-Term Memory) based on the pre-processed traffic data.

The aforementioned LSTM-based traffic congestion prediction method may be implemented as computer readable code on a computer readable medium. The computer-readable recording medium may be, for example, a removable recording medium (CD, DVD, Blu-ray disc, USB storage device, removable hard disk) or a fixed recording medium (ROM, RAM, computer-equipped hard disk). can The computer program recorded on the computer-readable recording medium may be transmitted to another computing device through a network such as the Internet, installed in the other computing device, and thus used in the other computing device.

In the above, even though all the components constituting the embodiment of the present invention have been described as being combined or operated as one, the present invention is not necessarily limited to these embodiments. That is, within the scope of the object of the present invention, all of the components may be selectively combined with one or more to operate.

Although actions are shown in a particular order in the drawings, it should not be understood that the actions must be performed in the specific order shown or in a sequential order, or that all shown actions must be performed to obtain a desired result. In certain circumstances, multitasking and parallel processing may be advantageous. Moreover, the separation of the various components in the embodiments described above should not be understood as requiring such separation, and the described program components and systems may generally be integrated together into a single software product or packaged into multiple software products. It should be understood that there is

So far, the present invention has been looked at mainly by its embodiments. Those skilled in the art to which the present invention pertains will be able to understand that the present invention can be implemented in a modified form without departing from the essential characteristics of the present invention. Therefore, the disclosed embodiments should be considered from a descriptive point of view rather than a limiting point of view. The scope of the present invention is shown in the claims rather than the foregoing description, and all differences within the equivalent scope will be construed as being included in the present invention.

100: LSTM-based traffic congestion prediction device
110: data collection unit
120: data correction unit
130: pre-processing unit
140: prediction unit

Claims (11)

a data collection unit that collects traffic data;
a data correction unit that corrects the collected traffic data;
a pre-processing unit that performs a missing imputation model based on the corrected traffic data; and
A prediction unit for predicting traffic congestion based on a time series based deep learning model (LSTM) based on preprocessed traffic data,
The data correction unit
Remove outliers from the collected traffic data through data filtering,
Correcting the missing value of the traffic data using the spatial trend utilization method using data of sections with similar traffic patterns;
The missing value of the traffic data is corrected using the time trend utilization method that corrects the current value through past data,
The missing imputation model is
an adversarial autoencoder-based data-missing imputation model;
A temporal convolution operation layer and a graph-based convolution operation layer are added to the encoder and decoder, and the layers are stacked to extract each feature, and the features are combined to perform the operation,
The Hadamard product of the mask vector and the vector, which is random noise and missing data, is input to the encoder to model the parameters of the latent distribution, the decoder extracts a sample from the latent space for missing value prediction, The discriminator is an LSTM-based traffic congestion prediction device that learns by distinguishing between imputed data and observed values.
delete delete delete According to claim 1,
The time series-based deep learning model
an input gate controlling input data;
an output gate that controls output data; and
Including a forget gate that controls whether or not to preserve data,
Output each output based on the data of the first time and the second time,
An LSTM-based traffic congestion prediction device that outputs a predicted output using the attention mechanism to the data of the third time and the output.
collecting traffic data;
performing data correction on the collected traffic data;
performing data pre-processing according to a missing imputation model based on the corrected traffic data; and
Including predicting traffic congestion through a time series-based deep learning model based on preprocessed traffic data,
The step of performing the data correction is
removing outliers from the traffic data through data filtering;
Correcting the missing value of the traffic data using a time trend utilization method that corrects the current value through past data; and
Correcting missing values of the traffic data using a spatial trend utilization method using data of sections having similar traffic patterns,
The missing imputation model is
an adversarial autoencoder-based data-missing imputation model;
A temporal convolution operation layer and a graph-based convolution operation layer are added to the encoder and decoder, and the layers are stacked to extract each feature, and the features are combined to perform the operation,
The Hadamard product of the mask vector and the vector, which is random noise and missing data, is input to the encoder to model the parameters of the latent distribution, the decoder extracts a sample from the latent space for missing value prediction, The discriminator is an LSTM-based traffic congestion prediction method that learns by distinguishing imputed data and observed values.
delete delete delete According to claim 6,
The step of predicting traffic congestion through the time series-based deep learning model
controlling input data through an input gate;
controlling output data through an output gate; and
Including the step of controlling whether or not to preserve data through a forget gate,
Output each output based on the data of the first time and the second time,
An LSTM-based traffic congestion prediction method for outputting an expected output using an attention mechanism to data of a third time and the output.
A computer program recorded on a computer-readable recording medium for executing the LSTM-based traffic congestion prediction method according to any one of claims 6 and 10.

Images