KR20170075312A - System and method for time-series predicting using integrated forward and backward trends, and a recording medium having computer readable program for executing the method - Google Patents

System and method for time-series predicting using integrated forward and backward trends, and a recording medium having computer readable program for executing the method Download PDF

Info

Publication number
KR20170075312A
KR20170075312A KR1020150184798A KR20150184798A KR20170075312A KR 20170075312 A KR20170075312 A KR 20170075312A KR 1020150184798 A KR1020150184798 A KR 1020150184798A KR 20150184798 A KR20150184798 A KR 20150184798A KR 20170075312 A KR20170075312 A KR 20170075312A
Authority
KR
South Korea
Prior art keywords
data
prediction
model
prediction data
time series
Prior art date
Application number
KR1020150184798A
Other languages
Korean (ko)
Other versions
KR101901654B1 (en
Inventor
안도훈
이기천
이운섭
Original Assignee
(주) 우림인포텍
한양대학교 산학협력단
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by (주) 우림인포텍, 한양대학교 산학협력단 filed Critical (주) 우림인포텍
Priority to KR1020150184798A priority Critical patent/KR101901654B1/en
Publication of KR20170075312A publication Critical patent/KR20170075312A/en
Application granted granted Critical
Publication of KR101901654B1 publication Critical patent/KR101901654B1/en

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management

Abstract

A forward predicted data calculating unit, a forward predicted data calculating unit, a forward predicted data calculating unit, a forward predicted data calculating unit, a forward predicted data calculating unit, a time-series data input unit, a forward directional arithmetic analysis model building unit, . And the first predictive data calculating unit calculates a first predictive data for the future using the forward directional arithmetic analysis model based on the time series data, Data is calculated and the reverse arithmetic model is constructed by using the reverse arithmetic model building part in the reverse direction using the reverse arithmetic model building part and the learning data calculating part calculates the fitted data using the reverse arithmetic analysis model, The second predictive data calculation unit calculates second predictive data for a future time point set in advance using the proximity prediction model, and the final predictive data calculation unit calculates The first prediction date And the second predicted data are combined to calculate final predicted data.

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a system and method for predicting forward and backward directional integrated time series, and a recording medium storing a computer readable program for executing the method. BACKGROUND OF THE INVENTION 1. Field of the Invention EXECUTING THE METHOD}

The present invention relates to time series analysis methods and systems, and more particularly, to a method and system for predicting future data using time series analysis.

Time series analysis is a field that analyzes the observed data with time. The most interesting part of the time series analysis is the prediction, and the time series analysis assumes that the pattern of the changes obtained from the past data will be maintained in the future and is a scientific prediction analysis. Currently, time-related data such as demographics, financial markets or weather forecasts are used in many industries where it is used.

The most widely used analytical model for time series prediction is ARIMA (AutoRegressive Integrated Moving Average). The Arima model is generally useful in short-term predictions and is useful when the components of a time series fluctuate rapidly over time, and is used in various industries.

However, in ARIMA model, if there is a sudden change in factors such as abnormal weather or holiday, like the price of agricultural and marine products, even if a new model is built to reflect this effect, there is a phenomenon that the prediction of the model follows the actual change late, There is a problem of lowering.

SUMMARY OF THE INVENTION It is an object of the present invention to provide a method and apparatus for estimating a model predicted by a model that can rapidly follow actual changes and improve predictive power even when sudden changes such as abnormal weather or holidays occur, System, and method.

In order to achieve the above object, a time series prediction system according to the present invention includes a time series data input unit, a forward directional arithmetic analysis model building unit, a first prediction data calculation unit, a reverse directional arithmetic model building unit, a fitting data calculation unit, A second prediction data calculation unit, and a final prediction data calculation unit.

And the first predictive data calculating unit calculates a first predictive data for the future using the forward directional arithmetic analysis model based on the time series data, And the reverse arithmetic model building unit constructs a reverse arithmetic analysis model using the time series data in the reverse direction, the matching data calculating unit calculates fitted data using the reverse arithmetic analysis model, and the proximity prediction model building unit The second prediction data calculation unit calculates second prediction data for a future time point set in advance using the proximity prediction model, and the final prediction data calculation unit calculates the second prediction data using the proximity prediction model, The first prediction date And the second predicted data are combined to calculate final predicted data.

This arrangement allows the forward ARIMA model to be influenced by future forecasts of the ARIMA model in the reverse order of time so that even if there is a sudden change in factors such as weather, The prediction of the model can follow the actual change quickly and increase the predictive power.

At this time, the proximity prediction model using the non-parametric analysis method may be a locally weighted polynomial regression (LOESS) analysis model. According to this configuration, the future prediction can be performed by reflecting the prediction trend of the reverse arithmetic analysis model using the characteristic of the LOESS model which places a lot of weight on the data of the adjacent section.

Also, the final prediction data can be calculated using the weighted average of the first prediction data and the second prediction data.

The time series prediction system further includes a prediction data evaluation unit for evaluating the final prediction data and a smoothing parameter changing unit for changing the smoothing parameter of the local weighted polynomial analysis model when the final prediction data is out of the preset evaluation reference . According to this configuration, the setting of the prediction model can be adjusted in accordance with the prediction result to construct an optimum prediction model.

At this time, the prediction data evaluation unit may compare the final prediction data with the first prediction data, and may compare the final prediction data calculated using different smoothing parameters with each other.

The apparatus may further include a time series data separator for separating the data used for calculating the final predictive data from the time series data and the evaluation data used for evaluating the final predictive data. According to this configuration, it is possible to perform prediction of future prediction as well as prediction using input time series data.

The time series data preprocessing unit may further include a time series data preprocessing unit for detecting and replacing an abnormal value in the input time series data, and the abnormal value detection and replacement in the time series data may be performed by a spline method. According to such a configuration, it is possible to perform more effective prediction by excluding an ideal value that greatly affects the prediction result from the input time series data.

In addition, the invention in which the system is implemented in the form of a method and a computer readable program for executing the method are disclosed together.

According to the present invention, by adding the influence of the future prediction of the ARIMA model to the ARIMA model in the reverse order of time, even when there is a sudden change in factors such as abnormal weather or holidays, Predicts the actual change quickly, thereby increasing the predictive power.

Also, using the features of the LOESS model, which weights a lot of data in the adjacent segments, future predictions can be made by reflecting the trend of the reverse arithmetic analysis model.

In addition, it is possible to construct an optimum prediction model by adjusting the setting of the prediction model according to the prediction result.

In addition, it is possible to evaluate prediction results as well as future predictions using input time series data.

Further, by excluding an ideal value having a large influence on the prediction result from the input time series data, more effective prediction can be performed.

1 is a schematic block diagram of a time-series prediction system in accordance with an embodiment of the present invention.
Figure 2 is a schematic flow diagram of a time series prediction method performed by the system of Figure 1;
Figure 3 is a more specific flow chart of Figure 2;
4 is a graph showing an example of replacing an ideal value due to an input error with a spline value.
FIG. 5 is a graph showing an example of a Forward ARIMA analysis and prediction of 5 days.
FIG. 6 is a graph showing an example in which backward ARIMA analysis is performed in the reverse direction of time.
FIG. 7 is a graph illustrating an example in which fit data learned in the model of FIG.
8 is a graph showing the results of repeatedly performing LOESS analysis using different parameters a (span).

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

1 is a schematic block diagram of a time-series prediction system according to an embodiment of the present invention.

1, the time series prediction system includes a time series data input unit 110, a time series data preprocessing unit 120, a time series data separating unit 130, a forward directional arithmetic analysis model building unit 140, A second predictive data calculation unit 190, a final predictive data calculation unit 200, and a second predictive data calculation unit 140. The first predictive data calculation unit 150, the second predictive data calculation unit 160, A prediction data evaluation unit 210, and a parameter changing unit 220.

In FIG. 1, each component of the time series prediction system 100 may be implemented in hardware, but it may be more general to be implemented in software that operates on hardware.

The time series data input unit 110 receives the time series data, and the time series data preprocessing unit 120 detects and replaces the abnormal value in the input time series data.

There are many outliers in the data input by man, such as agricultural auction price data. According to the time-series data preprocessing unit 120, it is possible to perform more effective prediction by excluding an ideal value that greatly affects the prediction result from the input time series data.

In order to find the outliers, we perform spline analysis similar to the trend of time series, find data whose difference between the actual data value and the spline value is above the threshold value, determine the found data as the outlier value, ) Value.

The time series data separator 130 separates the data used for calculating the final prediction data from the evaluation data used for evaluation of the final prediction data from the input time series data. According to this configuration, it is possible to perform prediction of a future time point as well as estimation of a prediction result using input time series data.

The forward directional arithmetic analysis model building unit 140 constructs a forward directional arithmetic analysis model using the time series data, and the first prediction data calculation unit 150 uses the forward directional arithmetic analysis model to calculate first predicted data for the future .

Forward ARIMA means performing ARIMA analysis in the order of time from the past to the future when performing time series analysis in the same way as in the past. The ARIMA model is largely estimated through four steps: Identification of the model, Estimation of the parameter, Diagnotic, and Forecasting.

Identify the characteristics of the agricultural and fisheries time series data in the identification step and select the order and the number of the differences and the periods p and q of the time series data to determine the form of the ARIMA (p, d, q) model. Before determining the number of differences, the unit root test is performed on the raw series data, and the difference is not stable.

The next step is to estimate the minimum squared estimation method, the maximum likelihood estimation method and so on, and estimate the values of AIC (Akaike Information Ciriterion) and BIC (Bayesian Information Criterion) do.

The stage of diagnosis is to determine how well the estimated time series is explained by the model and to determine the suitability of the model by residual analysis. If it does not, it returns to the identification or estimation stage, repeats the process of model estimation again, and makes the actual prediction if it is judged to be appropriate.

The reverse arithmetic model building unit 160 constructs a reverse arithmetic analysis model using the time series data in the reverse direction, and the matching data calculating unit 170 calculates the fitted data using the reverse arithmetic analysis model.

Backward ARIMA means to perform ARIMA analysis in the reverse order of time in the past in the future. If you create an ARIMA model in the reverse order of time, the past predictions of the model will eventually become the future predictors of the original series data. In other words, the backward ARIMA performs the same ARIMA analysis procedure with the original series data in the reverse direction.

Figure pat00001

However, the backward ARIMA model predicts the past of the original series data, but the future of the original series data is difficult to predict because it is the past forecast in the reverse order of time data.

Accordingly, the proximity prediction model construction unit 180 constructs a proximity prediction model using the nonparametric analysis method using the fitness data in the opposite direction, and the second prediction data calculation unit 190 calculates the proximity prediction model And calculates second prediction data for the time.

At this time, the proximity prediction model using the non-parametric analysis method may be a locally weighted polynomial regression (LOESS) analysis model. According to this configuration, the future prediction can be performed by reflecting the prediction trend of the reverse arithmetic analysis model using the characteristic of the LOESS model which places a lot of weight on the data of the adjacent section.

In other words, we use ARIMA Fitted data (ARIMA Fitted data) in the backward ARIMA model to perform past predictions (ie, future predictions of source series data) Make predictions.

LOESS predicts the backward ARIMA model with the learned data to best fit the trend of the backward model. A non-parametric regression analysis, LOESS, allows us to predict past backward ARIMA models. At this time, the purpose of the backward ARIMA analysis is to influence the backward ARIMA model against the forward ARIMA prediction.

The final prediction data calculation unit 200 calculates the final prediction data by combining the first prediction data and the second prediction data. At this time, the final prediction data can be calculated using the weighted average of the first prediction data and the second prediction data.

This arrangement allows the forward ARIMA model to be influenced by future forecasts of the ARIMA model in the reverse order of time so that even if there is a sudden change in factors such as weather, The prediction of the model can follow the actual change quickly and increase the predictive power. An example of a Forward-Backward Loess algorithm is as follows.

Figure pat00002

The prediction data evaluating unit 210 evaluates final prediction data, and the parameter changing unit 220 changes a smoothing parameter of the local weighted polynomial analysis model when the final prediction data is out of a preset evaluation reference. According to this configuration, the setting of the prediction model can be adjusted in accordance with the prediction result to construct an optimum prediction model.

At this time, the prediction data evaluation unit 220 can compare the final prediction data with the first prediction data, and compare the final prediction data calculated using different smoothing parameters with each other.

For example, the parameter α (span) of the LOESS model is set at (λ + 1) / n <α <1 (λ: polynomial degree, n: whole observations) You can set the value that brings about the predictive power or you can decide by the user's experience. For example, if you are making a daily forecast of agricultural and fishery products, you can set α based on your experience that close-in-the-past data about 1-2 weeks is important.

FIG. 2 is a schematic flowchart of a time series prediction method performed by the system of FIG. 1, and FIG. 3 is a more specific flowchart of FIG.

2 and 3, the time series system first receives time series data (S100), and performs preprocessing on the input time series data (S110). To this end, an abnormal value is detected and substituted in the inputted time series data (S112), and the train data (Train data) used for calculating the final prediction data and the test data (test data) used for evaluation of the final prediction data are separated (114). Spline analysis methods are preferred for outlier detection and missing value processing, but other alternative analytical methods are also possible.

4 is a graph showing an example of replacing an ideal value due to an input error with a spline value. 4, an abnormal value is shown at the upper end of FIG. 4. However, it can be confirmed at the lower end of FIG. 4 that a value greater than a specific value is replaced with a spline value by spline analysis.

In addition, the length of the train data can be preset by a user or the like, and can be determined to be about 2/3 or 3/4 of the entire time series data, for example.

Next, the forward directional arithmetic analysis is performed using time series data to generate predictive data (S120). More specifically, an ARIMA analysis model is constructed with train time series data (S122), predicts n future data (S124), stores forward directional predictive data (S126) The predicted value (MSE) of the forward arithmetic prediction and the test data is stored (S128). FIG. 5 is a graph illustrating an example in which a Forward ARIMA analysis is performed and 5 days are predicted. A plot showing the fit values together with the raw diagram after forward arithmetic analysis is shown.

Next, reverse arithmetic analysis is performed using time series data (S130) to generate backward compatibility prediction data (S140). More specifically, after taking an inverse of the train time series data, the arithmetic analysis model is constructed (Backward ARIMA) (S132), and the fitted data of the Arima model is taken again in the reverse direction (S134) And generates Reversed fitted data (S140).

FIG. 6 is a graph showing an example in which backward ARIMA analysis is performed in the reverse direction of time, and FIG. 7 is a graph illustrating an example in which relevance data learned in the model of FIG. An example of the model calculated by the reverse arithmetic analysis in FIG. 6 is as follows.

Figure pat00003

Next, the prediction of the backward ARIMA model is approximated by a non-parametric analysis method (S150). For non-parametric analysis, it is preferable to use Local Weighted Polynomial Regression (LOESS), but other methods may be used. At this time, the local data is the number of proximate data that determines a smoothing parameter.

More specifically, a 'near' value, which is the number of local data, is determined (S152), a LOESS analysis model is constructed with reversed fitted data, and n future data are predicted in the future (S154) And stores the prediction data (S156). In this case, the parameter α (span) can be determined by repeatedly performing the LOESS analysis on the learned prediction data of the backward ARIMA model.

The parameter a (span) can be expressed as:

α (span) = near (number of local data) / all data (number of total data)

This means that the near value can be expressed as a product of α (span) and total data. Therefore, instead of inputting the parameter α (span), LOESS analysis can be repeated using different near values. FIG. 8 is a graph showing a result of repeatedly performing LOESS analysis using different parameters a (span).

Next, the final predicted value is obtained by applying a weight to each of the predicted value of the forward-directional ARIMA model and the backward-directional ARAMA LOESS predicted value (S160). At this time, the weight value can use a simple average or various other ensemble techniques.

Finally, the final prediction data is evaluated (S170), and the final estimation data that has been evaluated is output (S180). For the evaluation, we can compare the MSE values obtained by predicting 5 days after the FBIOESS analysis for the same interval as the MSE value obtained by the 5-day forecast after performing the conventional simple ARIMA analysis.

In the present invention, a model for short-term prediction of auction prices of agricultural and marine products through time series analysis is proposed. Algorithm that further improves predictive power by integrating Forward ARIMA model and Backward ARIMA model in ARIMA model which is the most widely used analytical model for existing time series forecast and suitable for agricultural and marine auction price data .

If the forecast of agricultural and marine product prices is made by the ARIMA analysis, the predicted value often shows a time difference, and the pattern of the observed change is late. Therefore, the forecasted level is often low. In the present invention, ARIMA Forward-Backward Loess Algorithm for Forward and Backward.

The basic idea of this model is to add the influence of the future prediction of the ARIMA model in the reverse order of time in the existing ARIMA model to compensate for the later predicted problem and expect better prediction results. In other words, we build a reverse ARIMA model and use it to predict future values, rather than simply use it to predict past values. To do this, we predict the future predictions of the ARIMA model in the reverse order of the predicted value and time in the existing ARIMA model by ensemble.

The time series analysis through the present invention solves the problem of the late prediction, which is a problem of the conventional ARIMA time series analysis, and shows a further improved prediction ability from the fluctuating data. The present invention can be applied to any industry in which daily data, such as agricultural and aquatic products auction prices, or daily or hourly data or time series analysis of big data are required. Especially useful for short-term prediction.

Although the present invention has been described in terms of some preferred embodiments, the scope of the present invention should not be limited thereby but should be modified and improved in accordance with the above-described embodiments.

100: Time Series Prediction System
110: Time series data input unit
120: time series data preprocessing section
130: Time series data separator
140: Forward Arima analysis model construction part
150: first prediction data calculation unit
160: Reverse arima model construction part
170: Adaptation data calculation unit
180: Proximity prediction model construction unit
190: second prediction data calculation unit
200: final prediction data calculation unit
210: prediction data evaluation unit
220: Parameter changing section

Claims (17)

A time series data input unit for receiving time series data;
An arithmetic analysis model building unit for constructing a forward directional arithmetic analysis model using the time series data;
A first predictive data calculation unit for calculating first predictive data for a future time point set in advance using the forward-directional arithmetic analysis model;
A reverse arithmetic model building unit for constructing a reverse arithmetic analysis model using the time series data in a reverse direction;
A learning data calculation unit for calculating fitted data using the reverse arithmetic analysis model;
A proximity prediction model building unit for constructing a proximity prediction model using a nonparametric analysis method using the prediction data in a reverse direction;
A second prediction data calculation unit for calculating second prediction data for a time point in the predetermined future by using the proximity prediction model; And
And a final prediction data calculation unit for calculating the final prediction data by combining the first prediction data and the second prediction data.
The method according to claim 1,
Wherein the proximity prediction model using the non-parametric analysis method is a locally weighted polynomial regression (LOESS) analysis model.
3. The method of claim 2,
Wherein the final prediction data is calculated using a weighted average of the first prediction data and the second prediction data.
3. The method of claim 2,
A prediction data evaluation unit for evaluating the final prediction data; And
And a smoothing parameter changing unit for changing a smoothing parameter of the local weighted polynomial analysis model when the final prediction data is out of a predetermined evaluation criterion.
5. The method of claim 4,
Wherein the prediction data evaluating unit comprises:
And compares the final prediction data with the first prediction data.
5. The method of claim 4,
The evaluating unit,
And compares the final prediction data calculated using different smoothing parameters with each other.
5. The method of claim 4,
And a time-series data separator for separating the data used for calculating the final predictive data from the time-series data and the evaluation data used for evaluation of the final predictive data.
3. The method of claim 2,
And a time series data preprocessing unit for detecting and replacing an abnormal value in the input time series data.
The time-
A time series data input step of receiving time series data;
Constructing a forward directional arithmetic analysis model using the time series data;
A first predicted data calculation step of calculating first predicted data for a future time point set in advance using the forward directional arithmetic analysis model;
A reverse arithmetic analysis model building step of constructing a reverse arithmetic analysis model using the time series data in a reverse direction;
A fitting data calculation step of calculating fitted data using the reverse arithmetic analysis model;
A proximity prediction model building step of building a proximity prediction model using a nonparametric analysis method using the adaptation data in a reverse direction;
A second predicted data calculation step of calculating second predicted data for the future future time point using the proximity prediction model; And
And a final prediction data calculation step of calculating final prediction data by combining the first prediction data and the second prediction data.
10. The method of claim 9,
Wherein the proximity prediction model using the non-parametric analysis method is a locally weighted polynomial regression (LOESS) analysis model.
11. The method of claim 10,
Wherein the final prediction data is calculated using a weighted average of the first prediction data and the second prediction data.
11. The method of claim 10,
A prediction data evaluation step of evaluating the final prediction data; And
And a smoothing parameter changing step of changing a smoothing parameter of the local weighted polynomial analysis model when the final prediction data is out of a predetermined evaluation criterion.
13. The method of claim 12,
Wherein the step of evaluating the prediction data comprises:
And comparing the final prediction data with the first prediction data.
13. The method of claim 12,
Wherein the step of evaluating the prediction data comprises:
And comparing the calculated final prediction data using different smoothing parameters.
13. The method of claim 12,
And separating the data used for calculating the final prediction data from the evaluation data used for evaluation of the final prediction data from the time series data.
11. The method of claim 10,
Further comprising a time series data preprocessing step of detecting and replacing an abnormal value in the input time series data.
A recording medium on which a computer-readable program for executing the method according to any one of claims 9 to 16 is recorded.
KR1020150184798A 2015-12-23 2015-12-23 System and method for time-series predicting using integrated forward and backward trends, and a recording medium having computer readable program for executing the method KR101901654B1 (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
KR1020150184798A KR101901654B1 (en) 2015-12-23 2015-12-23 System and method for time-series predicting using integrated forward and backward trends, and a recording medium having computer readable program for executing the method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
KR1020150184798A KR101901654B1 (en) 2015-12-23 2015-12-23 System and method for time-series predicting using integrated forward and backward trends, and a recording medium having computer readable program for executing the method

Publications (2)

Publication Number Publication Date
KR20170075312A true KR20170075312A (en) 2017-07-03
KR101901654B1 KR101901654B1 (en) 2018-09-27

Family

ID=59358082

Family Applications (1)

Application Number Title Priority Date Filing Date
KR1020150184798A KR101901654B1 (en) 2015-12-23 2015-12-23 System and method for time-series predicting using integrated forward and backward trends, and a recording medium having computer readable program for executing the method

Country Status (1)

Country Link
KR (1) KR101901654B1 (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110020374A (en) * 2017-12-28 2019-07-16 中国移动通信集团陕西有限公司 A kind of the transfer amount prediction technique and device of release information
KR20190131845A (en) * 2018-05-17 2019-11-27 정규식 System and method for predicting error of electric rail car
KR20200000660A (en) * 2018-06-25 2020-01-03 (주) 우림인포텍 System and method for generating prediction model for real-time time-series data
KR20220109632A (en) * 2021-01-29 2022-08-05 인하대학교 산학협력단 System for predicting carbon credit prices using trend of carbon credit prices and time series analysis and method thereof
KR20220168056A (en) * 2021-06-15 2022-12-22 한국전력공사 System measurement information status determination method and apparatus thhough double time series analysis

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR20010106166A (en) * 2001-03-24 2001-11-29 조종암 Apparatus for monitoring performance of database and method thereof
KR20080062998A (en) * 2006-12-29 2008-07-03 포스데이타 주식회사 Traffic information guiding system and method
JP4230975B2 (en) * 2004-09-17 2009-02-25 キヤノンＩｔソリューションズ株式会社 Data prediction apparatus, data prediction method, and program
JP2010276848A (en) * 2009-05-28 2010-12-09 Nippon Telegr & Teleph Corp <Ntt> Bidirectional predictive coding device and method, bidirectional predictive decoding device and method, and program and recording medium therefor
JP4673828B2 (en) * 2006-12-13 2011-04-20 日本電信電話株式会社 Speech signal section estimation apparatus, method thereof, program thereof and recording medium
KR20150096103A (en) * 2014-02-14 2015-08-24 한국전자통신연구원 Prediction Apparatus and Method for Yield of Agricultural Products

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR20010106166A (en) * 2001-03-24 2001-11-29 조종암 Apparatus for monitoring performance of database and method thereof
JP4230975B2 (en) * 2004-09-17 2009-02-25 キヤノンＩｔソリューションズ株式会社 Data prediction apparatus, data prediction method, and program
JP4673828B2 (en) * 2006-12-13 2011-04-20 日本電信電話株式会社 Speech signal section estimation apparatus, method thereof, program thereof and recording medium
KR20080062998A (en) * 2006-12-29 2008-07-03 포스데이타 주식회사 Traffic information guiding system and method
JP2010276848A (en) * 2009-05-28 2010-12-09 Nippon Telegr & Teleph Corp <Ntt> Bidirectional predictive coding device and method, bidirectional predictive decoding device and method, and program and recording medium therefor
KR20150096103A (en) * 2014-02-14 2015-08-24 한국전자통신연구원 Prediction Apparatus and Method for Yield of Agricultural Products

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
일본 특허공보 특허 제 4230975호(2009.02.25.) 1부. *
일본 특허공보 특허 제 4673828호(2011.04.20.) 1부. *

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110020374A (en) * 2017-12-28 2019-07-16 中国移动通信集团陕西有限公司 A kind of the transfer amount prediction technique and device of release information
CN110020374B (en) * 2017-12-28 2021-06-08 中国移动通信集团陕西有限公司 Forwarding amount prediction method and device for release information
KR20190131845A (en) * 2018-05-17 2019-11-27 정규식 System and method for predicting error of electric rail car
KR20200000660A (en) * 2018-06-25 2020-01-03 (주) 우림인포텍 System and method for generating prediction model for real-time time-series data
KR20220109632A (en) * 2021-01-29 2022-08-05 인하대학교 산학협력단 System for predicting carbon credit prices using trend of carbon credit prices and time series analysis and method thereof
KR20220168056A (en) * 2021-06-15 2022-12-22 한국전력공사 System measurement information status determination method and apparatus thhough double time series analysis

Also Published As

Publication number Publication date
KR101901654B1 (en) 2018-09-27

Similar Documents

Publication Publication Date Title
KR101901654B1 (en) System and method for time-series predicting using integrated forward and backward trends, and a recording medium having computer readable program for executing the method
CN107765347B (en) Short-term wind speed prediction method based on Gaussian process regression and particle filtering
US20200371491A1 (en) Determining Operating State from Complex Sensor Data
US7778715B2 (en) Methods and systems for a prediction model
US9286573B2 (en) Cost-aware non-stationary online learning
US20190251458A1 (en) System and method for particle swarm optimization and quantile regression based rule mining for regression techniques
KR101876051B1 (en) Machine learning system and method for learning user controlling pattern thereof
Carr et al. An approximate algorithm for prognostic modelling using condition monitoring information
Guthke Defensible model complexity: A call for data‐based and goal‐oriented model choice
JP6066086B2 (en) Data discrimination device, method and program
KR102225370B1 (en) Prediction system based on parameter improvement through learning and method thereof
WO2016047118A1 (en) Model evaluation device, model evaluation method, and program recording medium
JP6896590B2 (en) Predictive model maintenance system, predictive model maintenance method and predictive model maintenance program
EP3899758A1 (en) Methods and systems for automatically selecting a model for time series prediction of a data stream
JP2018092445A5 (en)
Shi et al. A comparison of single and multiple changepoint techniques for time series data
JP6529096B2 (en) Simulation system, simulation method and program for simulation
KR102533460B1 (en) Attention-based stacking method for time series forecasting
CN113111589A (en) Training method of prediction model, method, device and equipment for predicting heat supply temperature
CN113537630A (en) Training method and device of business prediction model
JP2018528511A (en) Optimizing output efficiency in production systems
CN106021080B (en) Using middleware database connection pool resource consumption trend intelligent Forecasting
Park et al. On the long-term predictive capability of data-driven software reliability model: an empirical evaluation
CN108984851B (en) Weighted Gaussian model soft measurement modeling method with time delay estimation
US20160267394A1 (en) Model estimation device, model estimation method, and model estimation program

Legal Events

Date Code Title Description
A201 Request for examination
E902 Notification of reason for refusal
AMND Amendment
E601 Decision to refuse application
AMND Amendment
E902 Notification of reason for refusal
AMND Amendment
X701 Decision to grant (after re-examination)
GRNT Written decision to grant