JP2004355200A - Time series prediction expression correction method and apparatus - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a means for constructing a precise prediction expression even in a small number of observation data. <P>SOLUTION: The time series prediction expression correction method comprises a data input step wherein a data inputting means inputs an observation value after detecting a structure change; a step for creating data for correcting a prediction expression wherein a means for creating data for correcting the prediction expression creates a time series prediction value after detecting a structure change, uses an observation value after detecting a structure change inputted in the data input step, and creates data used for correcting the prediction expression; and a step wherein a means for creating the prediction expression reconstructs the prediction expression with the output data of the step for creating data for correcting the prediction expression as an input. As a result, a conventional problem can be solved that the number of observation data after alienation becomes small when the time-series prediction expression is corrected rapidly after the alienation and a precise prediction expression cannot be constructed in time-series prediction (1), and the number of observation data after alienation needs to be increased and much time is required for reconstructing the prediction expression for constructing a precise prediction expression (2). <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a method and an apparatus for predicting time-series data.
[0002]
[Prior art]
There are various patents and literatures on the method of constructing the time series prediction formula and the method of detecting the deviation from the actual data. No such document was known at the time of filing the patent application. Although patent documents were searched, it was not possible to find a patent document that describes a conventional technical content. The technical scopes searched are as follows.
◇ Search target: Published patent application from 1993 to the day before the application and patent gazette from 1994 ◇ Search method: Full-text search by the following logical formula ・ Search by “time series” + “prediction” + “correction method”・ Search by “Time Series” + “Prediction Formula Correction Method” ・ Search by “Time Series” + “Prediction Formula Correction” As a normal method, the prediction formula is reconstructed using the observation data after the deviation occurs It is considered that a technique is used.
[0003]
[Problems to be solved by the invention]
If an attempt is made to quickly correct the time-series prediction formula after the occurrence of the deviation, the number of observation data after the deviation becomes small, and it becomes difficult to construct a highly accurate prediction expression. In order to construct a highly accurate prediction formula by the conventional method, the number of observation data after the deviation must be increased, and it takes time to reconstruct the prediction formula.
[0004]
The present invention has been made to solve the above problems, and provides means for constructing a highly accurate prediction formula even with a small number of observation data.
[0005]
[Means for Solving the Problems]
According to the present invention, the above-mentioned problem is solved by the means described in the "claims".
[0006]
That is, the invention of claim 1 is a method for correcting a prediction equation after detecting a time-series structural change in a time-series prediction, wherein the data input means inputs an observation value after the structural change is detected. Prediction formula correction data generating means, using the time series prediction value after structural change detection and the observation value after structural change detection input in the data input step, to generate data used for prediction formula correction. A time-series predictive formula correcting method, comprising: a formula correcting data generating step; and a predictive formula generating step of, using the output data of the predictive formula correcting data generating step as input, reconstructing a predictive formula. is there.
[0007]
The invention of claim 2 is the time-series prediction formula correction method according to claim 1, wherein in the prediction formula correction data creating step, the time corresponding to the times 1, 2, 3,. a series prediction value _{P '1, P' 2,} P '3 ,,, P' n, observations using _{P 1, P 2, P 3} ,,, P n " number 12", the "number 13" Calculate,
[0008]
(Equation 12)
Δε _i = P _i −P ′ _i
[0009]
(Equation 13)
^{_{Δ 2 ε i = Δε i -Δε}} i-1
By performing on i = 1, 2,..., n, “Equation 14” is obtained.
[Equation 14]
E ² = {Δ ² ε _i | i = 1,2,2,, n}
From Equation 14, k pieces of Δ ² ε _i extracted by random reconstruction extraction are defined as Equation 15,
[0011]
(Equation 15)
E _k = {δ _i | i = 1,2,2,, k}
"Number 15" to determine the average of _{E k} than, the "number 16" [0012]
(Equation 16)
x _k = Σδ _i / (k−1)
This is repeated m times to obtain “Equation 17”.
[Equation 17]
X = {x _ki | i = 1,2,2, m}
From “Equation 17”, the distribution function of “Equation 17” is given by “Equation 3”.
(Equation 18)
F _x (x) (Equation 3)
Using “Equation 3” and random numbers, generate N (> n) “Equation 19”,
[0015]
[Equation 19]
_{y 1, y 2 ,,, y n} ,,, y N
Time series prediction values for 1, 2, 3,..., N,.
[0016]
(Equation 20)
_{Q '1, Q' 2,} Q '3 ,,, Q' n ,,, Q 'N
From "number 19" and "number 20", using a "type 4" to calculate the _{Q 1, Q 2, Q 3} ,,, Q n ,,, Q N, and "number 22", "number 22" Is output as data used for correcting the prediction formula,
[0017]
(Equation 21)
Q _i = Q ′ _i + (i−1) * y _i (Equation 4)
[0018]
(Equation 22)
_{Q 1, Q 2, Q 3} ,,, Q n ,,, Q N
In the prediction formula generation step, there is provided a time-series prediction formula correction method for constructing a time-series prediction formula based on "Equation 22" output from the prediction formula correction data creating step.
[0019]
The invention according to claim 3 is a time-series prediction formula correcting device that corrects a prediction formula after detecting a time-series structural change in the time-series prediction, wherein the data input unit inputs an observation value after the structural change is detected, Prediction formula correction data creating means for creating data used for predictive equation correction using a time-series predicted value after structural change detection and an observed value after structural change detection input by the data input means; and A time-series prediction formula correcting device having prediction formula generating means for reconstructing a prediction formula by using output data of the correction data generating means as input.
[0020]
BEST MODE FOR CARRYING OUT THE INVENTION
FIG. 1 is a diagram showing a system related to a predictive expression correcting device that embodies the present invention. In the figure, numeral 1 denotes a prediction formula correction device, 2 denotes a data input device, 3 denotes a prediction formula correction data generation device, and 4 denotes a prediction formula generation device.
[0021]
The data input device 2 is a device for inputting observation data after structural change detection.
[0022]
The prediction formula correction data creation device 3 uses the time series prediction value after structural change detection created using the existing time series prediction formula and the observation data after structural change detection input by the data input device 2. This is an apparatus for creating data used for correcting the prediction formula.
[0023]
The prediction formula creation device 4 is a device that reconstructs a prediction formula by using data output from the prediction formula correction data creation device 3 as an input.
[0024]
FIG. 2 is a flowchart showing details of a procedure for correcting a prediction equation after detecting a structural change in the time-series prediction model which is an example of the embodiment of the present invention. In the figure, reference numerals S0 to S4 indicate the respective steps of the processing, and correspond to the descriptions of steps 0 to 4 in the following description.
[0025]
Step 0
The data input device 2 inputs observation values P ₁ , P ₂ , P _3, ... P _n at times 1, ₂ , ₃ ,.
[0026]
Step 1
The prediction formula correction data creating device 3 uses the existing prediction model to predict values P ′ ₁ , P ′ ₂ , P ′ _3, ..., P ′ for times 1, ₂ , ₃ ,. calculating a _n, by using the observed value _{P 1, P 2, P 3} ,,, P n input at step 0 "number 23", calculates the "number 24",
[0027]
[Equation 23]
Δε _i = P _i −P ′ _i
[0028]
(Equation 24)
^{_{Δ 2 ε i = Δε i -Δε}} i-1
“Equation 25” is obtained by performing the processing for i = 1, 2,.
[0029]
(Equation 25)
E ² = {Δε _i | i = 1,2,2, n}
Step 2
The prediction formula correction data creating device 3 randomly extracts and extracts k Δ ² ε _i from “Equation 25”, and sets it as “Equation 26”.
[0030]
(Equation 26)
E _k = {Δε _i , _j | j = 1,2 ,,, k}
The average of _Ek is calculated from “Equation 26” and is set as “Equation 27”.
[Equation 27]
x _k = ΣΔε _i , _j / (k−1)
This is repeated m times to obtain “Equation 28”.
[Equation 28]
X = {x _ki | i = 1,2,2,, m}
From “Equation 28”, the distribution function of “Equation 28” is given by “Equation 5”.
[0033]
(Equation 29)
F _x (x) (Equation 5)
Step 3
The prediction formula correction data generating device 3 generates N (> n) “expression 30” using “expression 5” and random numbers,
[0034]
[Equation 30]
_{y 1, y 2 ,,, y n} ,,, y N
Time series prediction values for 1, 2, 3,..., N,.
[0035]
[Equation 31]
_{Q '1, Q' 2,} Q '3 ,,, Q' n ,,, Q 'N
From "number 30" and "number 31", using a "type 6" calculates the _{Q 1, Q 2, Q 3} ,,, Q n ,,, Q N, and "number 33", "number 33" Is the data used for correcting the prediction formula.
[0036]
(Equation 32)
Q _i = Q ′ _i + (i−1) * y _i (Equation 6)
[0037]
[Equation 33]
_{Q 1, Q 2, Q 3} ,,, Q n ,,, Q N
Step 4
The prediction formula creation device 4 reconstructs a prediction formula by using “Equation 33” which is the output of the prediction formula correction data creation device 3 as an input.
[0038]
As described above, the invention made by the inventor has been specifically described based on the embodiment. However, the present invention is not limited to the embodiment, and can be variously modified without departing from the gist of the invention. Needless to say,
[0039]
【The invention's effect】
As described above, according to the present invention, it is possible to quickly reconstruct a model after a structural change has occurred in a time-series prediction model.
[Brief description of the drawings]
FIG. 1 is a diagram illustrating a system configuration for correcting a time-series prediction expression according to an embodiment of the present invention.
FIG. 2 is a flowchart showing a procedure for correcting a time-series prediction formula according to the embodiment of the present invention.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 1 Prediction formula correction device 2 Data input device 3 Prediction formula correction data creation device 4 Prediction formula creation devices S0 to S4 Processing steps

Claims (3)

In the time series prediction, a method of correcting a prediction equation after detecting a time series structural change,
A data input step of inputting an observation value after the structural change is detected,
The prediction formula correction data generating means uses the time-series predicted value after the structural change detection and the observation value after the structural change detection input in the data input step to generate data used for the prediction formula correction. Data creation step;
A prediction formula creation unit configured to, using the output data of the prediction formula correction data creation step as input, to reconstruct a prediction formula,
A time-series prediction formula correction method having The time-series prediction formula correction method according to claim 1,
In the prediction formula correction data creating step, time series prediction values P ′ ₁ , P ′ ₂ , P ′ _3, ... P ′ _n for times 1, ₂ , ₃ ,. By using the values P ₁ , P ₂ , P _3, ..., _Pn , “Equation 1” and “Equation 2” are calculated,

This is performed for i = 1, 2,... N to obtain “Equation 3”.

From Equation 3, k pieces of Δ ² ε _i extracted by random reconstruction extraction are defined as Equation 4,

_Calculate the average of Ek from “Equation 4” and set it as “Equation 5”.

This is repeated m times to obtain “Equation 6”.

From “Equation 6”, the distribution function of “Equation 6” is given by “Equation 1”.

Using “Expression 1” and random numbers, generate N (> n) “Equation 8”,

Time series prediction values for 1, 2, 3,..., N,.

From "Equation 8" and "Number 9", with "type 2" to calculate the _{Q 1, Q 2, Q 3} ,,, Q n ,,, Q N, and "number 11", "number 11" Is output as data used for correcting the prediction formula,

A time-series prediction formula correction method for constructing a time-series prediction formula based on "Equation 11" output from the prediction formula correction data creating step in the prediction formula generation step. In the time-series prediction, a time-series prediction formula correction device that corrects a prediction formula after detecting a time-series structural change,
Data input means for inputting observation values after structural change detection,
Using a time-series predicted value after structural change detection and an observed value after structural change detection input by the data input means, predictive equation correction data creating means for creating data used for predictive equation correction,
Prediction formula correction means for reconstructing a prediction formula, using output data of the prediction formula correction data generation means as input,
A time-series prediction formula correction device having:

Images