KR102038703B1 - Method for estimation on online multivariate time series using ensemble dynamic transfer models and system thereof - Google Patents

Abstract

A method and system for predicting real-time multivariate time series using a dynamic transition ensemble model are provided.
In the real-time multivariate time series prediction method using the dynamic transition ensemble model, the real-time multivariate time series prediction system selects a predetermined number of a plurality of dynamic transition models to define a dynamic transition ensemble model based on selected models, the real-time multivariate time series The prediction system receives real-time data and performs prediction in real time while updating the weight of each selection model or model coefficients of each selection model included in the dynamic transition ensemble model.

Description

Method for estimation on online multivariate time series using ensemble dynamic transfer models and system approximate method

The present invention relates to a method and system for predicting real-time multivariate time series using a dynamic transition ensemble model. More particularly, the present invention can effectively predict real-time time series in a multivariate environment using an ensemble model using a plurality of dynamic transition models. It relates to a method and a system thereof.

Attempts have been made to predict specific events (eg errors or failures) in real time through a variety of existing APM solutions, CCTV networks and other time series prediction models.

However, in a multivariate environment with many independent factors affecting the performance of the system, the conventional techniques are not merely predictable through real-time analysis but merely monitoring.

In addition, the VAR (Vector Autoregressive) model, which is known as a conventional multivariate time series prediction model, is a predictive model using various assumptions, and it is difficult to cope with various situations due to the feature of predicting variables at once using one model. It is difficult to predict one target variable while using multivariate data.

Accordingly, the present invention provides a method and system for improving the accuracy of prediction using an ensemble model combining a plurality of dynamic transition models.

In addition, by using a real-time multivariate time series data to provide a method and system that can improve the predictive power by setting the target response variable among many variables.

In addition, it provides a method and system that can select models through offline learning and make accurate predictions online in real time.

According to an aspect of the present invention, in the real-time multivariate time series prediction method using the dynamic transition ensemble model, the real-time multivariate time series prediction system selects a predetermined number of a plurality of dynamic transition models to define a dynamic transition ensemble model based on selected models. The real-time multivariate time series prediction system receives real-time data and performs prediction in real time while updating the weight of each selection model or model coefficients of each selection model included in the dynamic transition ensemble model. Include.

The real-time multivariate time series prediction method using the dynamic transition ensemble model may further include generating, by the real-time multivariate time series prediction system, the plurality of dynamic transition models through training data.

The generating of the plurality of dynamic transition models through the training data by the real-time multivariate time series prediction system includes linearly combining the autoregressive influence of the dependent variable Y and the influence of the independent variable x at the present time on the dependent variable Y. Generating the models.

The dynamic transition models may be generated by the following equation.

[Equation]

는 자기회귀계수,

는 회귀모델 상수,

는 독립변수 x에 대한 1차 회귀계수,

는 오차항, m은 독립변수의 개수를 나타낸다. Where Y is the dependent variable, B is the backshift operator, d is the difference factor, p is the autoregressive order, I is the display function,

Is the autoregressive coefficient,

Is the regression model constant,

Is the first-order regression coefficient for the independent variable x,

Is the error term and m is the number of independent variables.

개(여기서, q는 독립변수 선택개수)의 동적전이 모형들을 생성하는 것을 특징으로 할 수 있다.The real-time multivariate time series prediction system to generate the plurality of dynamic transition models through the training data,

Q, where q is the number of independent variable choices.

The step of defining a dynamic transition ensemble model based on selected models by selecting a predetermined number of the plurality of dynamic transition models generated by the real-time multivariate time series prediction system includes K excellent performances among the plurality of dynamic transition models. It can be characterized in that the selection to the selection models.

The dynamic transition ensemble model may be defined by the following equation.

Where w is the weight of each selection model.

The real-time multivariate time series prediction system may perform the prediction in real time while updating the weight of each of the selection models included in the dynamic transition ensemble model or the model coefficient of each of the selection models using the following equation. The weight may be updated.

The real-time multivariate time series prediction system performing the prediction in real time while updating the weight of each of the selection models included in the dynamic transition ensemble model or the model coefficient of each of the selection models is advanced using a predetermined window size. While updating, the model coefficients of each of the selected models may be updated through the RLS method.

The real-time multivariate time series prediction system performing the prediction in real time while updating the weight of each of the selection models included in the dynamic transition ensemble model or the model coefficient of each of the selection models is a previous point in time t ( The weight may be obtained by using the predicted error in t-1).

The real-time multivariate time series prediction method using the dynamic transition ensemble model for solving the technical problem is a step in which the real-time multivariate time series prediction system generates the plurality of dynamic transition models through the training data, the real-time multivariate time series prediction system Using a dynamic transition ensemble model based on selected models by selecting a predetermined number of the plurality of dynamic transition models, receiving real-time data and updating the weight of each of the selection models or the model coefficient of each of the selection models. Performing prediction in real time.

The method may be implemented by a computer program installed in a data processing apparatus and recorded on a recording medium.

The real-time multivariate time series prediction system using the dynamic transition ensemble model for solving the technical problem is a DB that stores information on each of the plurality of dynamic transition models, a selection model selected by selecting a predetermined number of the plurality of dynamic transition models Ensemble model module for defining a dynamic transition ensemble model based on a field, receiving real-time data and predicting in real time while updating the weight of each of the selection models included in the dynamic transition ensemble model or model coefficients of each of the selection models It includes a control module for performing the.

The real-time multivariate time series prediction system using the dynamic transition ensemble model may further include a dynamic transition model module configured to generate the plurality of dynamic transition models by performing learning through learning data.

A real-time multivariate time series prediction system using a dynamic transition ensemble model for solving the technical problem includes a processor and a memory for storing a computer program executed by the processor, wherein the computer program, when executed by the processor, The method can be performed.

According to the technical idea of the present invention, it is possible to increase the accuracy of prediction by using an ensemble model combining a plurality of dynamic transition models. In addition, it is possible to improve the real-time prediction performance while updating the weights and / or model coefficients of each of the plurality of dynamic transition models in real time.

In addition, the model can be selected through off-line learning to make accurate predictions online and in real time.

In addition, by using real-time multivariate time series data, it is possible to set a response variable among many variables to improve prediction power. Therefore, it can be used to predict failures in real time in a conventional application performance management solution or CCTV network, as well as real-time multivariate time series data generated in a server network system such as log data or performance data generated in various other databases or servers. It can be applied to industries that require analysis and prediction. If an appropriate threshold can be set, it can be applied to the field of outlier detection, and it can be applied to areas such as intrusion detection on the network or to detect diseases using data generated in the medical industry.

BRIEF DESCRIPTION OF THE DRAWINGS In order to better understand the drawings cited in the detailed description of the invention, a brief description of each drawing is provided.
1 illustrates an example of a real-time multivariate time series prediction system using a dynamic transition ensemble model according to the spirit of the present invention.
2 is a schematic flowchart for implementing a real-time multivariate time series prediction method using a dynamic transition ensemble model according to an embodiment of the present invention.
3 and 4 are views illustrating an embodiment of training data according to an embodiment of the present invention.
5 illustrates an example of a selection model selected from a plurality of dynamic transition models according to an embodiment of the present invention.
6 exemplarily shows a weight change of each of the selection prostheses in the ensemble model according to the embodiment of the present invention.
7 is a diagram illustrating a prediction result according to the technical spirit of the present invention.

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 written description. However, this is not intended to limit the present invention to specific embodiments, it should be understood to include all transformations, equivalents, and substitutes included in the spirit and scope of the present invention. In the following description of the present invention, if it is determined that the detailed description of the related known technology may obscure the gist of the present invention, the detailed description thereof will be omitted.

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 are used only for the purpose of distinguishing one component from another.

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 specification, the terms "comprise" or "having" are intended to indicate that there is a feature, number, step, operation, component, part, or combination thereof described in the specification, and one or more other. It is to be understood that the present invention does not exclude the possibility of the presence or the addition of features, numbers, steps, operations, components, parts, or a combination thereof.

In addition, in the present specification, when one component 'transmits' data to another component, the component may directly transmit the data to the other component, or through at least one other component. Means that the data may be transmitted to the other component. On the contrary, when one component 'directly transmits' data to another component, it means that the data is transmitted from the component to the other component without passing through the other component.

Hereinafter, exemplary embodiments will be described in detail with reference to the accompanying drawings. Like reference numerals in the drawings denote like elements.

1 illustrates an example of a real-time multivariate time series prediction system using a dynamic transition ensemble model according to the spirit of the present invention.

Referring to FIG. 1, in order to implement a real-time multivariate time series prediction method using a dynamic transition ensemble model according to the technical idea of the present invention, a real-time multivariate time series prediction system using a dynamic transition ensemble model (hereinafter, 'prediction system', 100) It may be provided.

The prediction system 100 may be installed in the form of a client in a terminal of a predetermined user or may be installed in a server side. The prediction system 100 is installed in a predetermined data processing apparatus, and the software (eg, an application) implemented to implement the technical idea of the present invention and the hardware of the data processing apparatus for driving the software are organically combined. It may be a system implemented.

The prediction system 100 may be a system for predicting time series data of a specific variable in an environment in which a plurality of independent variables exist, that is, in a multivariate environment, according to the spirit of the present invention. The prediction system 100 may be affected by one or a few specific variables (eg, independent variables) compared to a conventional multivariate time series prediction system (eg, Var, etc.) predicting all variables at once using a single model. It has a favorable effect on the prediction of the dependent variable.

The training data and models exemplified herein below are examples applied to an application performance management system, and the scope of the present invention is not limited thereto.

The prediction system 100 includes a control module 110 and an ensemble model module 120.

The prediction system 100 may further include a predetermined DB 130 and / or dynamic transition model module 140. According to an embodiment of the present invention, some of the above-described elements may not necessarily correspond to the elements necessary for the implementation of the present invention, and according to the embodiment, the prediction system 100 Of course, more components may be included.

The prediction system 100 may include hardware resources and / or software necessary to implement the technical idea of the present invention, and does not necessarily mean one physical component or one device. . That is, the prediction system 100 may mean a logical combination of hardware and / or software provided to implement the technical idea of the present invention, and if necessary, are installed in devices spaced apart from each other to perform respective functions. As a result, it may be implemented as a set of logical configurations for implementing the technical idea of the present invention.

In addition, the term "module" in the present specification may mean a functional and structural combination of hardware for performing the technical idea of the present invention and software for driving the hardware. For example, the module may mean a logical unit of a predetermined code and a hardware resource for performing the predetermined code, and does not necessarily mean a physically connected code or a kind of hardware. It can be easily inferred by the average expert in the technical field of the present invention.

The control module 110 may include other components included in the prediction system 100, for example, an ensemble model module 120, a DB 130, and / or a dynamic transition model module 140. Or you can control resources.

The prediction system 100 according to the technical spirit of the present invention has a plurality of dynamic transitions in order to compensate for a problem that many assumptions about multivariate time series prediction are difficult and prediction of one specific variable using multivariate data is not robust to various situations. It is characterized by the use of dynamic transition ensemble models that combine models.

To this end, the ensemble model module 120 may select a predetermined number of a plurality of predefined dynamic transition models. Then, the selected dynamic transition models, that is, the dynamic transition ensemble model based on the selection model can be defined. This dynamic transition ensemble model can be defined by the following equation.

[Equation 1]

는 k 번째 선택모형을 의미할 수 있다.Where w is the weight of each selection model,

May mean the kth selection model.

The ensemble model module 120 may select K dynamic transition models from among a plurality of dynamic transition models as a selection model. Of course, the ensemble model module 120 may select the K selection models in order of models having excellent predictive performance, and may select various methods.

Information about a plurality of dynamic transition models may be stored in the DB (130).

In addition, the DB 130 includes training data for learning the plurality of dynamic transition models, information about the dynamic transition models according to the learning results (that is, model coefficients and errors for defining the dynamic transition models), respectively. More information about the performance measurement results of the dynamic transition models may be stored.

The dynamic transition models may be multivariate time series models constructed through offline learning. In other words, each dynamic transition model can operate as a multivariate time series model.

The dynamic transition models may be learned through learning data, and coefficients for defining the dynamic transition model may be defined through the learning results.

The dynamic transition model module 140 may generate a plurality of such dynamic transition models. Information about the generated dynamic transition models and related information (eg, training data, performance, etc.) may be stored in the DB 130.

)들의 영향을 선형 결합하여 상기 동적전이 모형들을 생성할 수 있다.The dynamic transition model module 140 is an independent variable of the autoregressive influence portion and the current time point (

The dynamic transition models can be generated by linearly combining the effects of

These dynamic transition models can be generated through the following equation.

[Equation 2]

으로 표현될 수 있다. Where the regression impact

It can be expressed as.

)들의 영향은

로 표현될 수 있다.Also, the independent variable at this time (

)

It can be expressed as.

,

는 자기회귀 계수(autoregression coefficient),

는 회귀모델 상수,

는 독립변수에 대한 1차 회귀 계수(simple linear regression coefficient),

는 오차, m은 독립변수의 개수를 의미할 수 있다.Also, Y is the dependent variable, B is the backshift operator, d is the difference factor, p is the autoregressive order, and I is the indicator function.

,

Is the autoregression coefficient,

Is the regression model constant,

Is the linear linear regression coefficient for the independent variable,

Is an error, and m may mean the number of independent variables.

According to an embodiment, the data as shown in FIG. 3 may be learning data.

3 illustrates an example of learning data according to an embodiment of the present invention, and the learning data may be learning data used in an application performance management solution.

As shown in FIG. 3, the response speed of the learning data may be the dependent variable Y, and the remaining variables (eg, Totla_mem, Free_mem, Used_mem, Usage, sys_cpu, user_cpu, count, http, javad, and jdbc). All of these can be independent variables (x). Therefore, the dynamic transition ensemble model according to the technical idea of the present invention may be used to predict the response speed when the independent variables have a certain value and change in time series using the learning data.

The dynamic transition model module 140 may generate a plurality of dynamic transition models using Equation 2.

For example, the dynamic transition model module 140,

[Equation 3]

We can generate dynamic transition models, where q is the number of independent variable choices.

는 독립변수가 q개인 경우 전체 독립변수의 개수(m) 중에서 서로 다른 1개를 뽑을 수 있는 조합의 수부터 서로 다른 q개까지 뽑을 수 있는 조합의 수를 의미할 수 있다. 즉, [수학식 3]은 종속변수에 영향을 미치는 독립변수가 q개 이하라고 가정할 경우, 생성될 수 있는 모든 서로 다른 수학식 2의 개수가 될 수 있다.That is, when the number of independent variables affecting the possible autoregression order p, the possible difference order d, and the dependent variable, all possible equations 2 may be equation 3. here

When q is an independent variable, it may mean the number of combinations from which the number of combinations from which one can be extracted from the total number of independent variables (m) to q different from each other is determined. That is, Equation 3 may be the number of all the different equations 2 that can be generated, assuming that q or less independent variables affecting the dependent variable.

As described above, when a plurality of different dynamic transition models are generated and learned, the learned dynamic transition models may be stored in the DB 130.

According to an embodiment of the present disclosure, the dynamic transition model module 140 generates a dynamic transition model using some of the training data (eg, 4000) (eg, 3000), and generates a remaining portion (eg, 1000). You can update the generated dynamic transition model using. In addition, a predetermined measurement (eg, root meas square error) that may evaluate the performance of each model may be stored together in the DB 130.

For example, the data shown in FIG. 5 shows the identification numbers of the dynamic transition model sorted in ascending order based on a performance measure (eg, RMSE). That is, it may mean that the 1628 th dynamic transition model has the highest performance measurement (eg, 0.3679042).

Then, the ensemble model module 120 may select a predetermined number (eg, K) among different dynamic transition models stored in the DB 130 as a model to be used for the dynamic transition ensemble model, that is, a selection model. Of course, the ensemble model module 120 may select a selection model of K dynamic transition models in order of good performance evaluation results of each dynamic transition model. Here, the K pieces may be arbitrarily selected, or the K pieces suitable for the number of learning data or other various situations may be determined through a plurality of iterative simulations.

Then, the dynamic transition ensemble model defined by the ensemble model module 120 may be the same as Equation 1 described above.

As such, when the dynamic transition ensemble model is defined, the control module 110 receives real-time data and updates the weight of each of the selection models included in the dynamic transition ensemble model and / or model coefficients of each of the selection models. Predictions can be performed in real time. In other words, as the ensemble model becomes a dynamic transition, time series prediction can be performed in real time.

등)는 선택모형을 정의하는 파라미터 자체가 실시간으로 동적 전이됨을 의미할 수 있다. The updating of the weight of each of the selection models by the control module 110 may mean that the influence or performance of each selection model is dynamic. In addition, the model coefficients of each of the selection models (eg,

Etc.) may mean that the parameter itself defining the selection model is dynamically transitioned in real time.

As such, the control module 110 updates the dynamic transition ensemble model in real time as the real time data is input, thereby improving the predictive performance.

The control module 110 may update the weight of each of the selection models by using the following equation.

[Equation 4]

는 동적전이 모형에 의해 예측된 예측값을 의미할 수 있으며, 수학식 4의 왼쪽 식은 현재 시점의 다음시점부터 예측하고자 하는 시간 타임 개수(T)까지의 실제 데이터(

)와 예측값(

)의 차이의 제곱의 합에 반비례하여 해당 동적전이 모형의 가중치가 연산됨을 의미할 수 있다. 또한 수학식 4의 오른쪽 식은 선택모형들 각각의 가중치의 합은 1이 됨을 의미할 수 있으며 각각의 가중치는 전체 가중치의 합에서 차지하는 해당 가중치의 비율을 나타냄을 의미할 수 있다.here

May represent the predicted value predicted by the dynamic transition model, and the left formula of Equation 4 represents actual data (T) from the next time point to the current time point to be predicted.

) And the predicted value (

Inversely proportional to the sum of the squares of the difference of), it may mean that the weight of the dynamic transition model is calculated. In addition, the right expression of Equation 4 may mean that the sum of the weights of each of the selection models is 1, and each weight may represent the ratio of the corresponding weights to the sum of the total weights.

On the other hand, the control module 110 may update the model coefficients of each of the selection models, in this case, as is well known by performing a forward update using a predetermined window size through the Recursive Least Square (RLS) method You can update the model coefficients for each of the selected models.

을 이용하여 해당

번째 선택모형의 가중치를 구해야하지만 온라인(실시간)으로 순차적으로 데이터가 입력되는 상황이므로

이 존재하지 않게 된다. 따라서 이때에는 (t-1)의 시점에서의 오차 즉,

를 이용하여 가중치

를 구할 수도 있다.On the other hand, according to Equation 5, the control module 110 is a case of predicting the immediately next time point of the current time point,

Using

The weight of the first selection model needs to be calculated, but since the data is sequentially entered online (real time)

This will not exist. Therefore, at this time, the error at the time of (t-1),

Weight using

You can also get

를 이용하여 가중치

를 구할 수도 있다.Similarly, if the current view is t, then the weight of the subsequent view (e.g., t + n) is the error at the view of (tn)

Weight using

You can also get

The real-time multivariate time series prediction method through the dynamic transition ensemble model according to the technical idea of the present invention may be as shown in FIG. 2.

2 is a schematic flowchart for implementing a real-time multivariate time series prediction method using a dynamic transition ensemble model according to an embodiment of the present invention.

Referring to FIG. 2, the prediction system 100 according to the technical spirit of the present invention may receive learning data and perform learning offline (S100 and S110).

As a result, as described above, a plurality of dynamic transition models (eg, Equation 3) may be generated (S120). Each dynamic transition model may be defined by Equation 2.

Then, the prediction system 100 may build a dynamic transition ensemble model by selecting K of the plurality of dynamic transition models (S130). The dynamic transition ensemble model constructed may be the same as Equation 1.

Then, the prediction system 100 may obtain input data in real time (online) (S140), and thus perform prediction while updating weights and / or model coefficients of each selection model in real time (S150).

An embodiment according to the spirit of the present invention will be described with reference to FIGS. 3 to 7.

3 and 4 are views showing an embodiment of training data according to an embodiment of the present invention, Figure 5 shows an example of a selection model selected from a plurality of dynamic transition models according to an embodiment of the present invention, 6 is a view showing a weight change of each of the selection models in the ensemble model according to an embodiment of the present invention, and FIG. 7 is a diagram showing a prediction result according to the technical spirit of the present invention.

According to an exemplary embodiment of the present invention, prediction is applied to the application management solution of the application as described above using the type of learning data as shown in FIG. 3. The dependent variable to be predicted is the response speed.

4000 pieces of actual learning data were used, and the learning data used in this way is shown in FIG. 4.

Of these 4000 training data, the prediction system 100 constructed a plurality of dynamic transition models using 3000 training data, and updated each dynamic transition model using the remaining 1000 training data.

As shown in FIG. 5, K (eg, 40) selection models were selected using the performance values (eg, RMSE) of the respective dynamic transition models. 5 shows a selection model selected according to the predictive performance of t + 6 and performance values of the selection model.

We then defined a dynamic transition ensemble model using the selected K selection models and performed predictions by updating the weights and model coefficients.

Prediction was started after the initial 200 real-time data entered online, and as a result, as shown in FIG. 6, the weights of the K (40) selection models were changed.

In addition, Figure 7 shows the prediction result of t + 6 in the above environment, the solid black line is the actual data and the red solid line represents the prediction value through the dynamic transition ensemble model according to an embodiment of the present invention. And as shown in Figure 7 it can be seen that relatively excellent predictive performance is exhibited.

The real-time multivariate time series prediction method using the dynamic transition ensemble model according to the technical concept of the present invention is analyzed and predicted using real-time multivariate time series data generated in a server network system such as log data or performance data generated in various databases or servers. It may be applicable in the industry required. In addition, if an appropriate threshold value can be set, it can be applied to an anomaly detection field, and may be applied to an area such as an intrusion detection on a network or a disease determination using data generated in the medical industry.

Meanwhile, the real-time multivariate time series prediction method using the dynamic transition ensemble model according to an embodiment of the present invention may be implemented in a computer-readable program command form and stored in a computer-readable recording medium. The control program and the target program according to the present invention can also be stored in a computer-readable recording medium. The computer-readable recording medium includes all kinds of recording devices in which data that can be read by a computer system is stored.

The program instructions recorded on the recording medium may be those specially designed and constructed for the present invention, or may be known and available to those skilled in the software art.

Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tape, optical media such as CD-ROMs, DVDs, floppy disks, and the like. Included are hardware devices specifically configured to store and execute the same magneto-optical media and program instructions such as ROM, RAM, flash memory, and the like. The computer readable recording medium can also be distributed over network coupled computer systems so that the computer readable code is stored and executed in a distributed fashion.

Examples of program instructions include not only machine code generated by a compiler, but also devices that process information electronically using an interpreter, for example, high-level language code that can be executed by a computer.

The hardware device described above may be configured to operate as one or more software modules to perform the operations of the present invention, and vice versa.

The foregoing description of the present invention is intended for illustration, and it will be understood by those skilled in the art that the present invention may be easily modified in other specific forms without changing the technical spirit or essential features of the present invention. will be. Therefore, it should be understood that the embodiments described above are exemplary in all respects and not restrictive. For example, each component described as a single type may be implemented in a distributed manner, and similarly, components described as distributed may be implemented in a combined form.

The scope of the present invention is shown by the following claims rather than the detailed description, and all changes or modifications derived from the meaning and scope of the claims and their equivalents should be construed as being included in the scope of the present invention. .

Claims (15)

Selecting, by the real-time multivariate time series prediction system, a predetermined number of the plurality of dynamic transition models to define a dynamic transition ensemble model based on the selected selection models;
The real-time multivariate time series prediction system receives real-time data and performs prediction while updating the weight of each of the selection models included in the dynamic transition ensemble model or the model coefficient of each of the selection models. Real-time Multivariate Time Series Prediction Using Models.
The method of claim 1, wherein the real-time multivariate time series prediction method using the dynamic transition ensemble model comprises:
And generating, by the real-time multivariate time series prediction system, the plurality of dynamic transition models through training data.
The method of claim 2, wherein the real-time multivariate time series prediction system generates a plurality of dynamic transition models through training data,
And generating the dynamic transition models by linearly combining the autoregressive influence of the dependent variable Y and the influence of the independent variable x of the current time point on the dependent variable Y. The multivariate time series prediction method using the dynamic transition ensemble model.
4. The method of claim 3, wherein the dynamic transition models are generated by the following equation.
[Equation]

Where Y is the dependent variable, B is the backshift operator, d is the difference factor, p is the autoregressive order, I is the display function,

Is the autoregressive coefficient,

Is the regression model constant,

Is the first-order regression coefficient for the independent variable x,

Is the error term and m is the number of independent variables.
The method of claim 4, wherein the real-time multivariate time series prediction system generates the plurality of dynamic transition models through training data,

A method for predicting real-time multivariate time series using a dynamic transition ensemble model, wherein the dynamic transition models are generated, wherein q is an independent variable selection number.
The method of claim 1, wherein the real-time multivariate time series prediction system selects a predetermined number of the generated dynamic transition models to define a dynamic transition ensemble model based on the selected selection models.
Real-time multivariate time series prediction method using the dynamic transition ensemble model, characterized in that the K of the predictive performance of the plurality of dynamic transition models are selected as the selection models.
The method of claim 1, wherein the dynamic transition ensemble model is defined by the following equation.

Where w is the weight of each selection model.
The method of claim 1, wherein the real-time multivariate time series prediction system performs prediction in real time while updating weights of each of the selection models included in the dynamic transition ensemble model or model coefficients of each of the selection models.
Real-time multivariate time series prediction method using a dynamic transition ensemble model, characterized in that for updating the weight using the following equation.

The method of claim 1, wherein the real-time multivariate time series prediction system performs prediction in real time while updating weights of each of the selection models included in the dynamic transition ensemble model or model coefficients of each of the selection models.
A real-time multivariate time series prediction method using a dynamic transition ensemble model, wherein the model coefficients of each of the selected models are updated through an RLS method while performing a forward update using a predetermined window size.
The method of claim 1, wherein the real-time multivariate time series prediction system performs prediction in real time while updating weights of each of the selection models included in the dynamic transition ensemble model or model coefficients of each of the selection models.
A real-time multivariate time series prediction method using a dynamic transition ensemble model, wherein a weight is obtained using a predicted error at a current time t at a current time t.
Generating, by the real-time multivariate time series prediction system, a plurality of dynamic transition models through the training data;
The real-time multivariate time series prediction system receives real-time data using a dynamic transition ensemble model based on selection models selected from a plurality of dynamic transition models generated by the real-time multivariate time series prediction system. A method for predicting real-time multivariate time series using a dynamic transition ensemble model, comprising performing prediction while updating model coefficients of each of the selected models.
A computer program installed in a data processing apparatus and recorded on a recording medium for performing the method according to any one of claims 1 to 11.
A DB for storing information about each of the plurality of dynamic transition models;
An ensemble model module for selecting a predetermined number of the plurality of dynamic transition models to define a dynamic transition ensemble model based on selected models;
Real-time multivariate through the dynamic transition ensemble model, which receives a real-time data and includes a control module for performing prediction while updating weights of each of the selection models included in the dynamic transition ensemble model or model coefficients of each of the selection models. Time series prediction system.
The system of claim 13, wherein the real-time multivariate time series prediction system using the dynamic transition ensemble model comprises:
Real-time multivariate time series prediction system through a dynamic transition ensemble model further comprising a dynamic transition model module for performing the learning through the training data to generate the plurality of dynamic transition models.
A processor; And
A memory storing a computer program executed by the processor,
The computer program, when executed by the processor, to perform the method according to any one of claims 1 to 10.

Images