KR102082341B1 - Traffic state prediction system based on bayesian network and operating method thereof - Google Patents

Abstract

Bayesian network-based traffic condition prediction system according to an embodiment of the present invention, the traffic information input unit for receiving traffic information from a plurality of measuring equipment, and the Bayesian network structure by setting the traffic information as an input variable, and the learning data It may include a traffic state prediction unit for learning a Bayesian network parameter based on the prediction, and probabilistically predict the traffic state using the Bayesian network structure and the Bayesian network parameters.

Description

TRAFFIC STATE PREDICTION SYSTEM BASED ON BAYESIAN NETWORK AND OPERATING METHOD THEREOF

The present invention relates to a Bayesian network-based traffic condition prediction system and its operation method.

In terms of traffic operation, traffic condition prediction is an important issue. One of the main objectives of the traffic operation is to prevent traffic flow reconciliation. Therefore, it is necessary to anticipate the dynamically changing traffic conditions and to correspondingly reduce the traffic congestion. However, traffic flow conditions are complex and simultaneous influenced by various factors (geometry, drivers, weather, accidents, etc.), and there is a lack of accurate prediction and understanding. Although parametric methodologies (time series analysis, Kalman filter, regression analysis, etc.) have been tried to accurately predict traffic conditions, the prediction accuracy is not high. Parametric methodology assumes a linear relationship between variables, which may limit the prediction of traffic conditions with complex and nonlinear characteristics. To improve this, the machine learning based methodology (Neural Network, Support Vector Machine, Nearest Neighbors, etc.) is applied to improve the predictive power. In other words, it is difficult to analyze which variables influence how much and in what direction, and further analysis is needed through multiple trials and errors or variable selection methodologies in order to create an improved model. The methodology used to predict traffic conditions may result in the inability or convenience of results if some of the input variables included in the model are missing. In order to prevent this, a methodology for generating and correcting missing information is considered, but there is a limit to accurate correction. In particular, vehicle detectors (loops, images, radars, etc.) installed on actual roads may have relatively high missing information, so a traffic condition prediction methodology is required to produce a result that is not convenient for missing information.

Patent Registration: 10-1370735, Registered Date: 2014-02-27, Title: Computer-based Adaptive Evaluation System and Method using Bayesian Network Korean Patent Application Publication No. 10-2013-0089685, Publication Date: 2013-08-13, Title: Apparatus and method for allocating virtual machine using conditional probabilistic reasoning for traffic information service in cloud environment United States Patent: US 9,371,009, Registered: 2016-06-21, Title: Modular Intelligent Transportation System Japanese Patent Publication: JP 2009-529186, Publication date: 2009-08-13, Title: Dynamic time series prediction of future traffic conditions

It is an object of the present invention to provide a novel Bayesian network-based traffic condition prediction system for predicting traffic conditions and a method of operating the same.

An operation method of a traffic state prediction system according to an exemplary embodiment of the present invention includes: receiving traffic information from measurement equipment of a target point, an upstream point, a downstream point, an entry route, and an entry route; And estimating a traffic condition using a Bayesian network set from the traffic information, wherein the traffic information is a continuous variable, the traffic condition is a discrete variable, between the continuous variable and the discrete variable, and Conditional probabilistic relationships between discrete variables can be established.

In an embodiment, the measurement equipment may include a point detector for measuring the traffic information having a traffic volume, speed, and occupancy based on a predetermined time.

In an embodiment, the measuring equipment may include a section detector for measuring the traffic information having a section passage speed, a passage time, and a traffic volume based on a predetermined time.

In an embodiment, the measurement equipment may include a GPS device in a vehicle measuring the traffic information having a link travel time and a travel speed.

In an embodiment, the estimating the traffic state may further include setting a Bayesian network structure using the collected traffic information as an input variable.

In an embodiment, the estimating the fraudulent traffic state may further include verifying the set Bayesian network structure.

In an embodiment, the predicting the traffic state may further include learning a Bayesian network parameter based on learning data.

In an embodiment, the estimating the traffic state may further include analyzing the set Bayesian network structure and the Bayesian network parameters by using an analysis index.

In an embodiment, the estimating the traffic state may further include changing the Bayesian network structure or parameter when deriving an improvement direction of the Bayesian network as a result of analysis using the analysis index. have.

In some embodiments, the analysis indicator may include Cost-of-Omission (CoC), Minimum and Maximum Beliefs (MMB), or Normalized Likelihood (NL) analysis indicators.

In an embodiment, the conditional probability between the discrete variables may be configured in a conditional probability table form, and the continuous variable and the discrete variable may be modeled using a distribution of Gaussians (MoGs) distribution.

Bayesian network-based traffic condition prediction system according to an embodiment of the present invention includes: a traffic information input unit for receiving traffic information from a plurality of measurement equipment; And setting a Bayesian network structure using the traffic information as an input variable, learning a Bayesian network parameter based on learning data, and probably predicting a traffic state using the Bayesian network structure and the Bayesian network parameter. It may include a prediction unit.

Bayesian network-based traffic condition prediction system and its operation method according to an embodiment of the present invention, using the Bayesian network with the traffic information as a variable for probabilistic traffic condition prediction and various situations occurring on the road It can also be used for prediction.

Bayesian network-based traffic condition prediction system and its operation method according to an embodiment of the present invention can be used to predict the risk of accidents based on traffic information collected from roads and vehicles in the field of traffic safety.

BRIEF DESCRIPTION OF THE DRAWINGS The accompanying drawings are provided to assist in understanding the present embodiment, and provide embodiments with a detailed description. However, technical features of the present embodiment are not limited to the specific drawings, and the features disclosed in the drawings may be combined with each other to constitute a new embodiment.
1 is a view for explaining the necessity of the traffic state prediction system based on Bayesian network according to an embodiment of the present invention.
2 is a diagram illustrating a traffic state prediction system based on a Bayesian network according to an embodiment of the present invention.
3 is a diagram illustrating a method of operating a traffic state prediction system based on Bayesian network according to an embodiment of the present invention.
4 is a diagram illustrating a relationship between traffic conditions and measured traffic information in a Bayesian network structure according to an embodiment of the present invention.
5 is a diagram illustrating a Bayesian network structure between a traffic state and a point detector traffic information according to an embodiment of the present invention.
6 is a diagram illustrating a Bayesian network structure according to an embodiment of the present invention.
7 is a diagram illustrating a result through the CoO analysis index according to an embodiment of the present invention.
8 is a diagram illustrating a result through the MMB analysis index according to an embodiment of the present invention.
9 is a diagram illustrating a result through the NL analysis index according to an embodiment of the present invention.
10 is a table evaluating the prediction performance of the traffic condition prediction system according to an embodiment of the present invention.

DETAILED DESCRIPTION Hereinafter, the contents of the present invention will be described clearly and in detail so that those skilled in the art can easily carry out the drawings.

As the inventive concept allows for various changes and numerous embodiments, particular embodiments will be illustrated in the drawings and described in detail in the text. However, this is not intended to limit the present invention to the specific form disclosed, it should be understood to include all modifications, equivalents, and substitutes included in the spirit and scope of the present invention. Terms such as first and second may be used to describe various components, but the components should not be limited by the terms.

The terms may be used for the purpose of distinguishing one component from another component. For example, without departing from the scope of the present invention, the first component may be referred to as the second component, and similarly, the second component may also be referred to as the first component. When a component is referred to as being "connected" or "connected" to another component, it may be directly connected to or connected to that other component, but it may be understood that other components may exist in between. Should be. On the other hand, when a component is said to be "directly connected" or "directly connected" to another component, it should be understood that there is no other component in between.

Other expressions describing the relationship between components, such as "between" and "immediately between" or "neighboring to," and "directly neighboring to" should be interpreted as well. The terminology used herein is for the purpose of describing particular example embodiments only and is not intended to be limiting of the invention. Singular expressions include plural expressions unless the context clearly indicates otherwise.

In this application, the terms "comprise" or "have" are intended to indicate that there is a feature, number, step, action, component, part, or combination thereof that is implemented, and that one or more other features or numbers are present. It should be understood that it does not exclude in advance the possibility of the presence or addition of steps, actions, components, parts or combinations thereof. Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art. Terms such as those defined in the commonly used dictionaries should be construed as meanings consistent with the meanings in the context of the related art, and shall not be construed in ideal or excessively formal meanings unless expressly defined in this application. .

The present invention discloses a traffic state prediction system based on Bayesian network that probabilistically predicts future traffic conditions by using traffic information collected in real time. Bayesian network-based traffic condition prediction system can be divided into parameters, which are structures and probability distributions of largely illustrated variables. The structure of the plotted variables for traffic condition prediction indicates the relationship between the measured traffic information and the traffic condition. The structure of the plotted variables for traffic condition prediction may vary depending on the range and type of traffic information measured. When the structure of the plotted variables is completed, the relationship between each variable may be derived as a probability distribution by using the measured traffic information as learning data. The Bayesian network-based traffic condition prediction system having the structure and parameters of the derived variables can predict the future traffic condition based on traffic information collected in real time.

The accurate traffic condition prediction result derived from the Bayesian network-based traffic condition prediction system according to an embodiment of the present invention may be used as information for preventing future traffic congestion as important information in an intelligent transportation system (ITS).

1 is a view for explaining the necessity of the traffic state prediction system based on Bayesian network according to an embodiment of the present invention.

Traffic conditions are affected by many factors, such as traffic loads upstream, breakdowns and fluctuations downstream, confluence of vehicles entering the connecting road and shockwave transitions, or slowdowns due to lane changes of vehicles entering the main track. These factors are complex and concurrent with changes in traffic conditions. Thus, traffic state transitions (e.g., disruption) are probabilistic processes and only predictable in terms of probability. Therefore, it is necessary to adopt a methodology that reflects these probabilistic characteristics as the probability of traffic condition prediction.

Traffic condition prediction is an important issue in terms of traffic operation. One of the main objectives of the traffic operation is to prevent flow breakdown. Therefore, it is necessary to anticipate the dynamically changing traffic conditions and to correspondingly reduce the traffic congestion. A traffic state prediction system based on a Bayesian network according to an embodiment of the present invention may probably predict a traffic state by using traffic information collected in real time on a road as an input variable. In general, a Bayesian network has a set of directional edges between a set of variables, and each variable is a finite set of mutually exclusive states, and together with the edges each variable forms an acyclic directed graph (DAG). And a conditional probability table P (A | B ₁ , ..., B _N ) to each variable A together with the parents B ₁ , B _N.

To construct a Bayesian network, it is divided into the structure of the schematic variables and the parameter which is the probability distribution. The structure of Bayesian networks reflects the causal relationship between variables and can be built on expert knowledge or artificial intelligence.

The Bayesian network-based traffic condition prediction system according to an embodiment of the present invention proposes a network structure suitable for traffic condition prediction, and can correctly interpret the prediction model based on this. Once the structure of the Bayesian network is finally determined through verification and evaluation, the measured traffic information can be used as learning data to derive the relationship between the variables as a conditional probability distribution. The established Bayesian network can probabilistically predict future traffic conditions based on traffic information collected in real time. In addition, even if some missing information may result in uncomfortable results. The advantage of these Bayesian networks is that they enable model analysis based on sensitivity analysis. Bayesian-based traffic condition prediction system according to an embodiment of the present invention to analyze how the collected traffic information (variables) affect traffic condition prediction, cost-of-omission (CoO), minimum and maximum beliefs (MMB) In addition, three analytical indicators are available: normalized likelihood (NL). On the other hand, it should be understood that the analytical index of the present invention is not limited thereto.

2 is a diagram illustrating a traffic state prediction system based on a Bayesian network according to an embodiment of the present invention. Referring to FIG. 2, the traffic state prediction system 100 may include a traffic information input unit 110 through which real-time traffic information measured by traffic information measuring equipment in a road / vehicle is input and a traffic state predictor that predicts traffic conditions through the traffic information input unit 110. 120).

The real-time traffic information input to the traffic information input unit 110 may be variously classified according to the type of measurement equipment. In an embodiment, when the measurement equipment is a point detector, traffic volume, speed, occupancy, etc. based on a predetermined time (30 seconds, 1 minute, 5 minutes, 15 minutes, 30 minutes, etc.) may be measured. In an embodiment, when the measurement equipment is a section detector, section passage speed, passage time, traffic volume, etc. may be measured based on a predetermined time. In addition, link travel time, travel speed, etc. may be measured even in a vehicle based global positioning system (GPS) based traffic information. The traffic state prediction system 100 of the present invention may be configured by utilizing all types of traffic information. The traffic information thus input is classified according to the measured section, and the traffic state prediction system 100 may be configured by utilizing all of various traffic information (traffic volume, speed, share, etc.) in the same section.

When the structure and parameters of the Bayesian network of the traffic state prediction unit 120 are finally established, the data format of the real-time traffic information input unit may be configured to be suitable therefor.

The traffic state prediction unit 120 may be constructed based on a Bayesian network, and may probably predict a future traffic state through real-time traffic information of the traffic information input unit 110. The Bayesian network-based traffic state prediction unit 120 may predict the traffic state as follows.

3 is a diagram illustrating a method of operating a traffic state prediction system based on Bayesian network according to an embodiment of the present invention.

First, traffic information to be used as learning data and basic analysis data may be collected (S110). Next, the Bayesian network structure may be set using the collected traffic information as an input variable (S120). The basic structure of Bayesian network for traffic condition prediction is explained in detail below. After that, the Bayesian network structure may be verified (S130). Then, a Bayesian network parameter may be learned based on the learning data (S140). In an embodiment, the parameter learning may be batch learning or online learning. Batch learning can be used here to fully estimate the parameters only once. Online learning can update each data.

The Bayesian network with the constructed structure and parameters can be analyzed using the presented analysis indicators (S150). In an embodiment, the analysis index analysis may be performed based on a predetermined algorithm. In another embodiment, the analysis index analysis may be performed from a device equipped with artificial intelligence. Through this, the improvement direction of the traffic state prediction system may be derived (S160). The structure and parameters of the Bayesian network may change to reflect the direction of improvement. Finally, a Bayesian network for traffic condition prediction may be derived (S170).

The steps and / or actions according to the invention may occur simultaneously in different embodiments in different order, in parallel, or for other epochs, etc., as would be understood by one skilled in the art. Can be.

In some embodiments, some or all of the steps and / or actions may be directed to instructions, programs, interactive data structures, clients, and / or servers stored on one or more non-transitory computer-readable media. At least some may be implemented or performed using one or more processors. One or more non-transitory computer-readable media may be illustratively software, firmware, hardware, and / or any combination thereof. In addition, the functionality of the "module" discussed herein may be implemented in software, firmware, hardware, and / or any combination thereof.

One or more non-transitory computer-readable media and / or means for implementing / performing one or more operations / steps / modules of embodiments of the present invention may be used in application-specific integrated circuits (ASICs), standard integrated circuits, A controller that performs appropriate instructions, including a microcontroller, and / or an embedded controller, field-programmable gate arrays (FPGAs), complex programmable logic devices (CPLDs), and the like. Does not.

Meanwhile, the collected traffic information may have various forms such as a point detector, a section detector, and an in-vehicle GPS. A Bayesian network structure may be set based on the collected traffic information. Bayesian networks should be constructed to reflect logical causality. Logical causality must be ensured so that correct model analysis can be performed.

4 is a diagram illustrating a relationship between traffic conditions and measured traffic information in a Bayesian network structure according to an embodiment of the present invention. Referring to FIG. 4, the relationship between traffic conditions and measured traffic information shows a Bayesian network structure for a point.

5 is a diagram illustrating a Bayesian network structure between a traffic state and a point detector traffic information according to an embodiment of the present invention. Referring to FIG. 5, the measured traffic information is a result or sample revealed by the traffic state and cannot directly cause causality. In contrast, spatio-temporal traffic conditions can affect other spatio-temporal traffic conditions. The traffic condition can be observed by the measured traffic information. As an example of the traffic condition prediction system 100 according to an exemplary embodiment of the present invention, if the traffic volume, speed, and occupancy of time t can be obtained from a point detector of an area of a highway, the traffic state of time t and the structure shown in FIG. To have.

6 is a diagram illustrating a Bayesian network structure according to an embodiment of the present invention. If the traffic information of the target point, the upstream, the downstream, the entry road, and the exit road can be collected to perform the future traffic state prediction of the actual highway target point, the Bayesian network structure shown in FIG. 6 may be provided. Bayesian network structure that predicts traffic condition after 5 minutes using traffic information at time t.

Once the Bayesian network structure is created, verification procedures can be performed. Verification can be performed by logical judgment that takes into account conditional independence and dependencies. The relationship between two specific variables can be independent and dependent depending on the presence or absence of surrounding variables. In all situations, you will determine whether two specific variables have the correct logical relationship.

Bayesian networks that have completed the verification process for the structure can learn the parameters within a given structure. Traffic information measured in the traffic condition prediction system 100 according to an embodiment of the present invention is a continuous variable, traffic state will be defined as a discrete variable. Therefore, in the traffic state prediction system 100 according to an exemplary embodiment of the present invention, a conditional probabilistic relationship between a continuous variable and a discrete variable and between a discrete variable and a discrete variable can be constructed.

가 측정된 교통정보인 연속형 변수이며,

가 교통상태를 나타내는 이산형 변수라면

의 MoGs는 다음과 같이 계산될 수 있다.Conditional probabilities between discrete variables can simply be configured in the form of a conditional probability table. On the other hand, MoGs (Mixture of Gaussians) distribution can be used to model continuous and discrete variables. MoGs can be defined as the sum of finite Gaussian distributions with weights. if

Is a continuous variable that is measured traffic information.

Is a discrete variable representing a traffic condition

MoGs can be calculated as follows.

,

는 차례로 평균, 분산, 가중치 (

,

,

) here,

,

In turn average, variance, weight (

,

,

)

Depending on the type of variable, it has the following calculation method. Conditional probabilities may be learned by the previously collected learning data. The Bayesian network, with its structure and parameters, can derive future traffic conditions corresponding to the inputted information.

The established Bayesian network can analyze the traffic status and traffic condition through the following analysis indicators. The following analytical indicators provide quantitative results that can be used to interpret how each variable affects traffic conditions.

Cost-of-Omission (CoO)

Minimum and Maximum Beliefs (MMB)

,

,

Normalized Likelihood (NL)

Based on the analysis index results, the Bayesian network model can be derived and the Bayesian network model with new structure and parameters can be established. The derived Bayesian network model can play the role of a traffic condition prediction unit. The following are examples of analysis results through three analysis indicators.

7 is a diagram illustrating a result through the CoO analysis index according to an embodiment of the present invention. CoO evaluates the effect of each spatio-temporal traffic condition on the prediction of future traffic conditions. The larger the value, the higher the information on the prediction of future traffic conditions. FIG. 7 shows the effects of future traffic conditions on free-flow or congested prediction, and also sums them. As a result, the target point has the highest information, and the information by the entry or exit link is relatively low. For the sake of simplicity and efficiency of the model, points with low information about future traffic conditions may be excluded from the model.

8 is a diagram illustrating a result through the MMB analysis index according to an embodiment of the present invention. The MMB evaluates the range of future traffic conditions that can be changed by the input variables. The wider the range, the greater the influence of the input variables. As a result, the traffic volume, speed, and occupancy of the target site were found to have a great influence on the prediction of future traffic conditions. In addition, the upstream point can have a big influence when the future traffic conditions are heavy, and the downstream point can have a big influence when the future traffic conditions are smooth. Relatively, the impact of connecting road traffic on the prediction of future traffic conditions was small. For the sake of simplicity and efficiency of the model, points with low information about future traffic conditions may be excluded from the model.

9 is a diagram illustrating a result through the NL analysis index according to an embodiment of the present invention. NL is an indicator for analyzing the relationship between input variables and future traffic conditions. It can evaluate the effect of changes in input variables on the prediction of future traffic conditions. If the NL is close to 1, the input variable is evaluated as having no influence on the prediction of future traffic conditions.The smaller the value is, the larger the value is, the larger the value is. Has a big impact on predicting As a result, the faster the speed, the more likely the future traffic condition is, and the opposite of the occupancy rate. In the case of traffic volume, up to a certain level, the higher the traffic volume, the higher the probability of congestion, but after that, the influence of the traffic condition prediction becomes smaller. In addition, the impact of the traffic volume on the road link to the prediction of future traffic conditions was small. For the sake of simplicity and efficiency of the model, points with low information about future traffic conditions may be excluded from the model.

10 is a table evaluating the prediction performance of the traffic condition prediction system according to an embodiment of the present invention. Compared with other methodologies, the applicability of the traffic condition prediction system 100 according to the embodiment of the present invention was evaluated. Compared with the existing parametric methodology, Logistic regression, it has higher performance and shows similar results with ANN.

The traffic condition prediction system 100 according to an exemplary embodiment of the present invention may be used to predict traffic conditions and predict problems for various situations occurring on a road. In the field of traffic safety, it can be used as a technology for predicting the risk of accidents based on traffic information collected from roads and vehicles.

On the other hand, the contents of the present invention described above are only specific embodiments for carrying out the invention. The present invention will include not only specific and practically usable means per se, but also technical ideas as abstract and conceptual ideas that can be utilized in future technologies.

100: traffic condition prediction system
110: traffic information input unit
120: traffic condition prediction unit

Claims (12)

In the operation method of the traffic condition prediction system:
Receiving traffic information from the measurement equipment of the target point, the upstream point, the downstream point, the entry route, and the entry route; And
Predicting future traffic conditions using a Bayesian network set up from the traffic information;
The predicting of the future traffic conditions may include setting a Bayesian network structure using continuous variables of the collected traffic information and discrete variables of the traffic conditions observed by the traffic information as input variables, and based on learning data. Further comprising: learning a Bayesian network parameter, and analyzing the set Bayesian network structure and the Bayesian network parameter using analysis indicators,
A conditional probability relationship is established between the continuous variable and the discrete variable and between the discrete variable,
The analytical indicators include: Cost-of-Omission (CoC), which evaluates the effect of spatiotemporal traffic conditions on the prediction of future traffic conditions, and MMB (Minimum), which evaluates the extent to which the future traffic conditions can be changed by input variables. and Maximum Normalized Likelihood (NL) analytical indicators that assess the impact of changes in input variables or Maximum Beliefs on future traffic conditions. The method of claim 1,
The measuring equipment includes a point detector for measuring the traffic information having a traffic volume, speed and occupancy based on a predetermined time. The method of claim 1,
The measuring equipment includes a section detector for measuring the traffic information having a section passage speed, a passage time and the amount of traffic passing based on a predetermined time. The method of claim 1,
And the measuring equipment comprises a GPS device in a vehicle measuring the traffic information having a link travel time and a travel speed. delete The method of claim 1,
Predicting the future traffic conditions,
Verifying the established Bayesian network structure. delete delete The method of claim 1,
Predicting the future traffic conditions,
Changing the Bayesian network structure or parameter when a direction of improvement of the Bayesian network is derived as a result of analysis using the analysis indicator. delete The method of claim 1,
The conditional probability between the discrete variables is configured in the form of a conditional probability table,
The continuous variable and the discrete variable are modeled using a Mixture of Gaussians (MoGs) distribution. In Bayesian network based traffic condition prediction system:
A traffic information input unit configured to receive traffic information from a plurality of measurement equipments; And
A Bayesian network structure is set by using the continuous variables of the traffic information and the discrete variables of the traffic state observed by the traffic information as input variables, and learning Bayesian network parameters based on learning data. A traffic state prediction unit probabilistically predicting future traffic conditions using the Bayesian network parameter,
The traffic state prediction unit analyzes the set Bayesian network structure and the Bayesian network parameters by using an analysis index,
The analytical indicators include: Cost-of-Omission (CoC), which evaluates the effect of spatiotemporal traffic conditions on the prediction of future traffic conditions, and MMB (Minimum), which evaluates the extent to which the future traffic conditions can be changed by input variables. and Maximum Beliefs (NL) or Normalized Likelihood (NL) analysis indicators that assess the impact of changes in input variables on future traffic conditions prediction.

Images