JPH09146915A - Chaos time-series short-period predicting device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve the prediction precision. SOLUTION: An input part 11 fetches time-series data from an object system at an equal sampling interval and stores those data in a data storage part 12. The data stored in the storage part 12 are inputted to a preprocessing execution part 18 before prediction, and the data are preprocessed there and supplied to a prediction part 14. The preprocessing part 18 obtains new time-series data by calculating the ratio and difference of last time-series data to be predicted from the storage part 12. A parameter determination part 13, on the other hand, moves parameters within a certain range and uses several past observed data in the data storage part 12 to predict remaining data for a short period. Such parameters that the prediction error rate is minimum and the correlation coefficients of a predicted value and a measured value are maximum are found. The prediction part 14 predicts the value of data by using the parameters determined by the parameter determination part 13 and inputs the value to a postprocessing execution part 19, which postprocesses the data. As the postprocessing, conversion reverse to the preprocessing is performed.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a chaotic non-linear time-series data short-term predictor, and more particularly to a chaotic time-series short-term predictor characterized in how to handle time-series data to be predicted.

[0002]

2. Description of the Related Art Conventionally, an ARIMA model mainly based on autoregressive analysis has been mainly used for prediction of time series data, but in recent years, a deterministic chaos theory has been used for the purpose of elucidating chaos phenomena. Prediction of non-linear time series data with chaotic property by the conventional local reconstruction method has been studied.
As the local reconstruction method, Gram-Schmidt orthogonal method,
The tessellation method, the local fuzzy reconstruction method, etc. are mentioned. The outline of deterministic chaos is shown below.

Dominated by determinism (if the initial value is determined,
A seemingly irregular, unstable, and complex behavior can often be generated from a differential equation or a difference equation whose subsequent states are all determined in principle. This is the deterministic chaos of dynamical systems.

The concept of deterministic chaos has revealed that the irregular phenomena in the world are not necessarily dominated by non-determinism. In other words, among the irregular phenomena in the world, there are not a few phenomena that produce their behavior according to determinism. Common to dynamical systems that produce chaotic behavior is that their equations are non-linear.

By the way, if the behavior of certain time series data is chaotic, it can be considered that the behavior complies with the deterministic law. Then, if the nonlinear deterministic law can be estimated, it will be the near future until the deterministic causality is lost due to the "sensitive dependence on the initial value" of chaos from the measured data at a certain point. It is possible to predict the data of.

Prediction from the standpoint of such a deterministic dynamical system theory is based on the theory of embedding time-series data, "reconstructing the attractor of the original dynamical system from one observation time-series data". ing. The outline of this theory is as follows.

From some observed time series data y (t), vectors (y (t), y (t-τ), y (t-2τ), ..., Y (t)
-(N-1) τ)) (τ is a time delay).

This vector is the n-dimensional reconstructed state space R ⁿ
Will indicate one point. Therefore, if t is changed,
Orbits can be drawn in this n-dimensional reconstructed state space.

If the target system is a deterministic dynamical system, and observation time series data is obtained through an observational system corresponding to the C ¹ continuous mapping from the state space of this dynamical system to the one-dimensional Euclidean space R. This reconstructed orbit is an embedding of the original deterministic system, assuming that n is large enough.

That is, if the observed time series data is derived from the original attractor of the dynamical system, the attractor which preserves the phase structure of this attractor is reproduced in the reconstructed state space. Although n is usually called “embedding dimension”, since the reconstruction operation is “embedding”, this dimension n is expressed by the following equation (1) when the dimension of the state space of the original dynamical system is m. It has been proved to be sufficient if it holds.

[0011]

## EQU1 ## n ≧ 2m + 1 (1) However, this is a sufficient condition, and depending on the data, 2
Even if it is less than m + 1, it may be embedded. Furthermore, n
It is also shown that if> 2d (where d is the box count dimension of the original attractor of the dynamical system), the reconstruction operation is a one-to-one mapping.

As described above, if the deterministic chaos is,
The structure of the system is clarified even for seemingly chaotic data, and it becomes possible to make highly accurate short-term predictions depending on the data.

On the other hand, in the present technology for short-term prediction of chaotic time series, the structure of the target system is modeled by a mathematical formula and the prediction is performed using the mathematical formula. This equation contains many parameters, and its identification requires a large amount of calculation.

[0014]

However, there is always data that does not fit the mathematical model in the prediction target system,
Such data is noise and measurement error. When the dynamics of the target system changes, it is necessary to change the model formula itself, but it is difficult to make a prediction by changing the model formula itself with the current technology. In particular, it is very difficult to prepare in advance an algorithm that automatically associates changes in the target system with changes in the model. Moreover, since the accuracy of prediction depends on the structure of the time series data to be predicted, there is a problem that the accuracy of prediction decreases if the chaotic nature of the time series data is not clear.

The present invention has been made in view of the above circumstances, and it is an object of the present invention to provide a chaotic time series short-term prediction device capable of responding to a change in the dynamism of time series data in real time and improving the accuracy of prediction. It is an issue.

[0016]

In order to solve the above-mentioned problems, the present invention first stores the time-series data actual measurement value in the data storage means, and the time-series data stored in this data storage means is used as the immediately preceding data. Preprocessing is performed by the preprocessing execution unit by taking the ratio and difference with On the other hand, the parameter determining means selects the parameters that are the components of the data vector,
The time-series data pre-processed by the pre-processing execution unit is supplied to the prediction unit, and a data vector is generated from the time-series data according to the parameter determined by the parameter determination unit, and the attractor is reconfigured to perform the prediction. The predicted value is obtained by means. The post-processing execution unit performs the inverse conversion of the pre-processing on the prediction value predicted by the prediction means. The predicted value post-processed by the post-processing execution unit is stored in the prediction result storage means.

With respect to the actual measurement value of the time series data, a prediction value corresponding to the actual measurement value is detected from the post-processing execution unit, and the actual measurement value and the predicted value are compared to evaluate the prediction accuracy by the prediction result evaluation means. However, the parameter determination means, while extracting the sample data from the data storage means, to extract a predetermined number of past data required to predict the sample data, for each combination of the parameters,
Each of the predicting means calculates a predicted value of the sample data based on the past data, compares the calculated value with the value of the actual sample data, and selects a combination of parameters that gives the highest prediction accuracy of the sample data. While selecting and outputting to the prediction means, when the prediction accuracy becomes smaller than the threshold value by the prediction result evaluation means, the combination of parameters is reselected.

Sample data is extracted from the data storage means, a predictive value is obtained for each combination of the parameters by the predicting means based on the sample data, and the predicted value is stored as a corresponding actual measured value. By comparing and selecting the combination with the highest prediction accuracy, the combination of parameters corresponding to the dynamics of the time-series data to be predicted is generated. In particular, when the prediction accuracy in the prediction result evaluation means becomes smaller than the threshold, by reselecting the combination of dynamics parameters, even if the dynamics are changed, the parameters are changed in response to the change. It

Further, the predicting means of the present invention uses the reconstructed attractor to detect a plurality of neighboring vectors located in the vicinity of the data vector to be processed by the neighboring vector detecting means, and to detect the axis component to be processed. The axial component value of each of the detected neighboring vectors and the axial component value of the processing target data vector is detected by an axial component detecting means, and the axial component value of each of the neighboring vectors after a predetermined step. And a component value after a predetermined step of the axis component of the processing target data vector is determined by a fuzzy inference unit so that the difference from the one in which the difference in the original data vector is small becomes small. There is a local fuzzy reconstructing unit characterized in that the processing result in the axis component determination unit is converted into time series data and a predicted value of the data is generated by the data predicted value generation unit. That. Also, in local fuzzy reconstruction that performs prediction that reflects local vector motion,
Although the calculation time is proportional to the number of dimensions, the number of program steps in the prediction process is small, so that the prediction can be performed in a short time. Therefore, the predicted value can be calculated in real time, and the above parameters are also redetermined in real time.

[0020]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a functional block diagram showing a first form of a chaotic time series short-term prediction device. In FIG. 1, 11 is a time series data input unit, 12 is a data storage unit, 13 is a parameter determination unit, and 14 is a prediction unit. , 15 is an output unit, 16 is a prediction result storage unit, 17 is a prediction result evaluation unit,
Reference numeral 18 is provided between the data storage unit 12 and the prediction unit 14, and a pre-processing execution unit that pre-processes data from the data storage unit 12 and supplies the data to the prediction unit 14 as described later. The post-processing execution unit is provided between the output unit 15 and the prediction result storage unit 16 and post-processes the output from the prediction unit 14 and supplies the output to the output unit 15 and the prediction result storage unit 16.

The input unit 11 outputs time-series data y (t) from the target system as shown in FIG. 7 at equal sampling intervals.
Is stored in the data storage unit 12. The data stored in the data storage unit 12 is input to the pre-processing execution unit 18 before performing the prediction, where the following processing is performed, and then the data is supplied to the prediction unit 14.

In the preprocessing execution unit 18, the data storage unit 12
The new time series data is obtained by taking the ratio of the time series data to be predicted immediately before. For example, when the time series data to be predicted is a ₁ , a ₂ , ... A _n , ... And the processed new time series data is b ₁ , b ₂ , ... b _n ,. The newly created time series data is as follows.

[0023]

(Equation 2)

On the other hand, the parameter determination unit 13 moves the parameters within a certain range, and uses some of the observed past data in the data storage unit 12 to make a short-term prediction of the remaining data. Then, the prediction error rate is minimized,
Find the parameter that maximizes the correlation coefficient between the predicted value and the measured value (automatic parameter identification).

In this example, three of the embedding dimension, the time delay, and the number of neighboring data are selected, and the optimum combination is obtained from the combinations of the respective values. These parameters are necessary when the prediction unit 14 generates a data vector.

Since the data processed by the preprocessing execution unit 18 is supplied to the prediction unit 14, the prediction unit 14 uses the parameters determined by the parameter determination unit 13 after a predetermined time elapses by the local fuzzy reconstruction method. Predict the value of the data in. At the time of this prediction, among the new time series data processed by the pre-processing execution unit 18, b ₁ to b _n− 1th data are embedded and predicted from the b _n th value. ∧b _n , ∧b _{n + 1} ,… ∧b _m ,…
The post-processing execution unit 1 (hereinafter, the predicted value is indicated by adding a ∧ code)
Input to 9 and post-process data. As the post-processing, the reverse conversion of the pre-processing is performed. The time-series data when this inverse transformation is applied are ∧a _n , ∧a _{n + 1} , ... ∧a _m ,. Since the predicted time series data has a relationship of b _n → ∧b _n = a _n / ∧a _n , the inversely transformed time series data (time series data to be obtained) is as follows. .

[0027]

(Equation 3)

The result obtained by the post-processing execution unit 19 as described above is output to the output unit 15 and the prediction result storage unit 16. The output unit 15 outputs the prediction result, and the prediction result storage unit 16 stores the prediction result. The prediction result evaluation unit 17 compares the prediction result obtained from the post-processing execution unit 19 with the actual data after the predetermined time has elapsed. This actual data is obtained from the data storage unit 12.

When the prediction error rate or the correlation coefficient is deteriorated by constantly comparing the above-mentioned predicted data with the actual data, it is judged that the dynamics of the target system has changed, and the parameter determination unit 13 is determined. A signal for redetermining the value of each parameter is output.

The operation of the parameter determining unit 13 will be described in detail below with reference to the flowcharts of FIGS. 2 and 3 by taking predetermined time series data as an example. In this example, data for 130 steps was obtained from the time when the input was started.

First, the parameter determining unit obtains all combinations of the values of the embedding dimension n, the delay time τ, and the number N of data vectors included in the neighborhood (S21). At this time, an upper limit value and a lower limit value for n, τ, and N are determined in advance.

In this example, n is 2 to 6, τ is 3 to 7, and N is 2 to 8. Since n can take 5 kinds of values of 2 to 6, τ can take 5 kinds of values, and N can take 7 kinds of values, there are 5 × 5 × 7 = 175 combinations. .

For each of these combinations, the degree of agreement between the predicted value and the actually measured value is examined using past data.
When the data from the start of prediction to the 130th step is stored in the data storage unit 12, the prediction reference step is set to the α step, and the β step data before that is used to convert the data of the (α + 1) th step. The ratio is calculated and the processed data is predicted using the prediction unit 14, and
The post-processing execution unit 19 performs inverse transformation to obtain a predicted value (S2
2).

If α = 121 and β = 20, the 102nd
The data (test data) in the 122nd step is predicted using the data (past data) from the step to the 121st step.

Similarly, when α = 122, the data from the 103rd step to the 122nd step becomes the past data, and the data of the 123rd step becomes the test data. In this way, the prediction is performed up to α = 122 to (122 + γ) steps. In this example, γ = 9 and the 122nd to
The prediction was performed up to 130 steps. Thus, by using the latest data group obtained at that time as test data, an attractor reflecting the latest state of the system to be measured can be obtained. The values of β and γ may be set freely as appropriate.

The predicted values of the 122nd to 130th steps are compared with the actual values of the 122nd to 130th steps stored in the data storage unit 12, and the correlation coefficient, average error, average error rate, etc. An evaluation value is calculated by using the evaluation function that is set arbitrarily (S23). This evaluation value is calculated for each combination of all parameter values (S24, 25), the combination with the highest evaluation value is detected (S26), and the combination is output to the prediction unit 14 as the optimum combination at that time (S27).

In this example, values of n = 3, τ = 4 and N = 3 were obtained.

In the predicting unit 14, the optimum combination (n
= 3, τ = 4, N = 3), the prediction is performed using the latest observation data. In this example, since the optimum combination was calculated before the data of the 131st step was obtained, the prediction of the 131st step and thereafter was performed using this combination.

The prediction results after the 131st step are output to the output unit 15 via the post-processing execution unit 19, and
It is stored in the prediction result storage unit 16.

The prediction result evaluation section 17 compares the latest δ actual measurement values which are always obtained from the latest input section 11 with the predicted values respectively corresponding to the actual measurement values, and calculates the correlation coefficient, error,
The evaluation value and the predicted value are evaluated using an evaluation function such as an error rate (S31). When the evaluation value is lower than a predetermined threshold value, a signal for operating the parameter determination unit 13 is output (S32, 33).

The parameter determining unit 13 receives this output, finds the optimum parameter combination again, and outputs it to the predicting unit 14.

As described above, the optimum combination of parameters is determined.

In the above example, the evaluation function such as the correlation coefficient, the error, the error average rate, etc. is used as the evaluation criterion of the parameter, but this evaluation function is arbitrarily set in accordance with the prediction purpose of the target system. Can be changed. For example, when predicting power demand, prediction accuracy at the peak value of power demand is required.

Thus, if special precision is required,
This evaluation is also a parameter determination standard.

The detailed configuration of the prediction unit 14 using the local fuzzy reconstruction will be described below with reference to FIG.

The attractor reconstructing unit 42 reconstructs the attractor of the data vector in the n-dimensional state space from the data obtained from the data storage unit 12 using the Takens embedding theory to create the n-dimensional state space database.

Reference numeral 43 is a processing object data vector selection unit, 4
4 is a neighborhood vector detection unit, 45 is an s-step vector detection unit, 46 is an axial component detection unit, 47 is a fuzzy inference unit,
Reference numeral 48 is a data prediction value data prediction value generation unit, and these units form a local fuzzy reconstruction unit.

The function of each unit will be described in detail below.

The attractor reconstructing section 42 generates a vector x (t) represented by the following equation.

[0050]

X (t) = (y (t), y (t-τ), y (t-2)
τ), ... y (t− (n−1) τ)) t: (n−1) τ + 1 ≦ t ≦ N The values determined by the parameter determination unit 13 as the values of the embedding dimension n and the delay time τ are It is used to reconstruct the state space and attractor by embedding operations.

As described above, this vector indicates one point in the n-dimensional reconstructed state space R ⁿ , and when t is changed, a trajectory can be drawn in this n-dimensional reconstructed state space. An example of this is shown in FIG.
Shown in

When the data to be predicted follows deterministic chaos, the attractor becomes a smooth manifold as shown in FIG. 6, and it is possible to predict the data using this attractor.

Next, the data vector to be processed (usually the data vector obtained by the latest observation) is z.
As (T), the data vector z (T
The predicted data vector z ″ (T + s) of + s) is obtained as follows.

In the processing object data vector selection unit 43,
A data vector to be processed [normally the latest data vector z (T)] is selected and output to the neighborhood vector detection unit 44. In the neighborhood vector detection unit 44, the attractor reconstruction unit 42
Detects a plurality of neighborhood vectors x (i) in the neighborhood of z (T) from the data vector in [iεN (z (T)),
N (z (T)): set of indices i of neighborhood x (i) of z (T)].

In the axis component detecting section 46, z (T) and x
The axis component with (i) is detected and output to the fuzzy inference unit 47.

After the s-step, the vector detection unit 45 detects each x
For (i), the vector x (i +
s) is detected and output to the fuzzy inference unit 47.

In the fuzzy inference unit 47, z is set for each axis.
The component values of (T) and x (i) are compared, and each component value of the prediction vector z ″ (T + s) is determined from the comparison result and the component value at each x (i + s). The prediction vector z ″ (T + s) is output to the data prediction value generation unit 48.

Since the trajectory of the attractor composed of actual data is not usually an ideal manifold, it is possible to determine the axial component nonlinearly using fuzzy inference rather than linearly determining the axial component. preferable.

The process of determining the axial component using the fuzzy theory is shown below.

If the data were generated according to the deterministic law and the value of s is small enough,
It is considered that the process in which x (i) reaches x (i + s), which is s-step ahead data, is based on deterministic dynamism.

This dynamism can be expressed by the following linguistic expressions.

[0062]

## EQU00005 ## IF x (T) is x (i) THEN x (T + s) is x (i + s) where x (T): z in the n-dimensional reconfigurable state space
A set x (T + s) representing a data vector in the vicinity of (T): a set representing a data vector after s steps of x (T) iεN (z (T)), N (z (T)): z ( Since the set of indices i of the neighborhood x (i) of T) x (i) is the data vector of the neighborhood of z (T), the step s is deterministic due to the "sensitive dependence on the initial value" of chaos. Before loss of causality, state z (T)
It can be assumed that the dynamics from state z (T + s) to is approximately equivalent to the dynamics from state x (i) to state (i + s).

When the attractor embedded in the n-dimensional reconstruction state space is a smooth manifold, z (T) to z
The vector distance to (T + s) is affected by the vector distance from z (T) to x (i). That is, it can be considered that an orbit of x (i) closer to z (T) has a larger influence on the orbit from z (T) to z (T + s), and a farther the influence is smaller.

By the way,

[0065]

X (i) = (y (i), y (i−τ), ..., Y (i− (n−1) τ)) x (i + s) = (y (i + s), y (i + s) −τ), ..., y (i + s− (n−1τ)) (1) Therefore, focusing on the j axis in the n-dimensional reconstructed state space, the equation (1) becomes

[0066]

## EQU00007 ## IF aj (T) is yj (i) THEN aj (T + s) is y (i + s) (j = 1 to n) (2) where aj (T): neighborhood value of z (T) The j-axis component of x (i) in the n-dimensional reconstructed state space aj (T + s): the j-axis component of x (j + s) in the n-dimensional reconstructed state space n: embedded dimension number

The orbit from z (T) to z (T + s) is
Although affected by the vector distance from z (T) to x (i), this effect is expressed in a non-linear form because the attractor, which is the trajectory of this vector, is a smooth manifold. Therefore, in order to make the effect non-linear, if equation (2) is expressed by a fuzzy function,

[0068]

[Equation 8] IF aj (T) is y'j (i) THEN aj (T + s) is y'j (i + s) where (j = 1 to n) (3) Note that normally the function y (i) is The symbol "-" is used for fuzzy conversion, but the symbol "'" is used here.

In addition,

[0070]

## EQU00009 ## z (T) = (y (T), y (T-.tau)), ..., y (T-
(n−1) τ)), so j in the n-dimensional reconstruction state space of z (T)
The axis component is yj (T). Therefore, if the predicted value of the data vector z (T + s) after s steps of the data vector z (T) is z ″ (T + s), its j-axis component is yj (T) in aj (T) of the equation (3). ), And fuzzy inference is performed, it is possible to obtain as aj (T + s).

As described above, the value of each parameter obtained by the parameter determining unit 13 and the values obtained by the pre-processing executing unit 18 and the post-processing executing unit 19 are used to accurately predict the time series data. be able to.

In the above-mentioned first embodiment, the case where the ratio with the immediately preceding data is taken has been described, but the following second to fifth embodiments may be used.

In the second mode, the difference from the immediately preceding data is taken, and the time series data newly created by the processing is as follows.

[0074]

B ₁ = a ₁ -a ₂ , b ₂ = a ₂ -a ₃ , ..., b _n-1
= A _n-1 −a _n , ... The time series data obtained by subjecting the predicted value to the inverse transformation is as follows.

[0075]

∧a _n = a _n −∧b _n , ∧a _{n + 1} = a _{n + 1} −∧b
_{n + 1, ..., ∧a m} = a n-1 + m -∧b m, ... third embodiment obtained by to take common logarithm of the data,
The newly created time series data after processing is as follows.

[0076]

## EQU12 ## b ₁ = loga ₁ , b ₂ = loga ₂ , ..., b _n
= Loga _n, ... time-series data that has been subjected to inverse transformation to the prediction value is as follows.

[0077]

(Equation 13)

The fourth form is adapted to take the natural logarithm of the data, and the time series data newly created by the processing are as follows.

[0079]

[Mathematical formula-see original document] b ₁ = lna ₁ , b ₂ = lna ₂ , ..., _Bn = lna _n , .. Time series data obtained by subjecting the predicted value to inverse transformation is as follows.

[0080]

(Equation 15)

In the fifth embodiment, an arbitrary function is used. An arbitrary function is used for preprocessing and its inverse function is used for postprocessing.

As the prediction method, the above-described Gram-Schmidt orthogonal system method, tessellation method, local fuzzy reconstruction method, and the like can be mentioned.

[0083]

As described above, according to the present invention,
For the time-series data to be predicted, pre-processing such as taking the ratio or difference from the immediately preceding data is performed, and the time-series data that has been pre-processed is predicted as a new prediction target Since the predicted value is obtained by performing the inverse conversion, it is possible to improve the prediction accuracy.

[Brief description of the drawings]

FIG. 1 is a functional block diagram of a chaotic time series short-term prediction device.

FIG. 2 is a flowchart showing how to find an optimum combination of parameters.

FIG. 3 is a flowchart showing the operation of a prediction result evaluation unit.

FIG. 4 is a functional block diagram of a prediction unit.

FIG. 5 is an explanatory diagram of a trajectory of a data vector.

FIG. 6 is an explanatory diagram of an attractor.

FIG. 7 is an explanatory diagram of time series data.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 11 ... Input part 12 ... Data storage part 13 ... Parameter determination part 14 ... Prediction part 15 ... Output part 16 ... Prediction result storage part 17 ... Prediction result evaluation part 18 ... Pre-processing execution part 19 ... Post-processing execution part

Claims (5)

[Claims] 1. A method for embedding time series data to perform n
A chaotic time-series short-term predictor for reconstructing the attractor of the time-series data in a dimensional state space to perform data prediction, the data storage means storing the time-series data measured value, and the data storage means stored in the data storage means. Pre-processing execution unit for pre-processing the time-series data, parameter determining means for selecting a parameter to be a component of the data vector, and time-series data pre-processed by the pre-processing execution unit are supplied to determine the parameters. Predicting means for generating a data vector from the time-series data according to the parameter determined by the means, and for reconstructing the attractor to obtain a predicted value; and an inverse conversion of the preprocessing to the predicted value predicted by the predicting means. A post-processing execution unit that performs the above, a prediction result storage unit that stores the prediction value post-processed by the post-processing execution unit, and time-series data measurement For the value, the predicted value corresponding to the actual measurement value is detected from the post-processing execution unit, and a prediction result evaluation unit that evaluates the prediction accuracy by comparing the actual measurement value and the predicted value is included, and the parameter Determining means, for each combination of the parameters, to obtain a predicted value of the sample data based on the past data in the prediction means, respectively, by comparing the obtained value and the value of the actual sample data, A combination of parameters that maximizes the prediction accuracy of the sample data is selected and output to the prediction means, and when the prediction accuracy of the prediction result evaluation means becomes smaller than a threshold value, the combination of parameters is reselected. A chaotic time series short-term predictor characterized by the above. 2. The predicting means uses the reconstructed attractor to detect a plurality of neighboring vectors located in the vicinity of a data vector to be processed, and a detecting means for the axis component to be processed. Axis component detecting means for detecting a difference between the axis component value of each neighborhood vector and the axis component value of the processing target data vector, and detecting the axis component value after a predetermined step of each neighborhood vector, A fuzzy inference unit that determines a component value of the axis component of the processing target data vector after a predetermined step so that a difference from the original data vector having a small difference becomes small; and the axis component determination unit. A local fuzzy reconstructing unit having: a data prediction value generation unit that converts the processing result in step S1 into time series data to generate a data prediction value. 1 chaotic time series short-term prediction device as claimed. 3. The chaotic time series short-term prediction device according to claim 1, wherein the preprocessing execution unit takes a ratio or a difference from immediately preceding time series data. 4. The preprocessing execution unit takes a common logarithm or a natural logarithm of the time series data.
The described chaotic time series short-term prediction device. 5. The chaotic time series short-term prediction device according to claim 1, wherein an arbitrary function is used in the preprocessing execution unit.