JP2005122438A - Predicting method, predicting device, predicting program and recording medium - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a predicting method, a predicting device, a predicting program and a recording medium for working a plurality of prediction data as highly precise prediction data so as to increase predicting precision. <P>SOLUTION: This predicting device which calculates prediction data related to future prediction is provided with a predicting means 50 which calculates n types of prediction data by carrying out prediction by n types of different prediction methods and a prediction value working means 60 which calculates the mean value of the n types of prediction data as prediction data. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to a prediction method, a prediction device, a prediction program, and a recording medium for performing various predictions (energy demand prediction, sales volume prediction, price prediction, economic index prediction) using a computer.

Conventionally, predictions have been made in various fields. For example, an electric power company predicts the next day's power demand and plans the operation of the generator based on the result. Stores and vending machines predict sales volume and replenish products. In the financial field, the company forecasts commodity prices, stock prices, and exchange rates, and makes various trades.
Thus, various predictions are made in many fields. As the prediction method, there are a method that relies on human intuition, a method of prediction by a computer, etc., but there is no method that is determined by the prediction target or field, various methods have been studied, and various methods have been put to practical use. is the current situation.

  The problem with building a prediction system is that there is no method determined by the target or field of prediction, and therefore long-term data analysis and simulation verification are required to establish the prediction method. Therefore, it takes a long time to build a prediction system, resulting in an increase in cost, or a system with high prediction accuracy unless sufficient data analysis and simulation verification are performed to reduce costs. I could not.

As a solution to such a problem, there is a method of incorporating a plurality of prediction methods into the system. That is, by incorporating a plurality of candidate prediction methods, the examination / verification time for narrowing down to one best prediction method is reduced, and the cost is reduced. If a prediction method with high prediction accuracy can be appropriately selected, a high prediction result can be obtained as a result.
As such a conventional technique, Patent Document 1 discloses an invention (a data prediction method, a data prediction apparatus, and a recording medium) that selects one of a plurality of prediction methods.

  Further, although not a prediction technique, there is a technique that obtains a final prediction value by combining a plurality of prediction values. For example, there is a stock investment allocation (portfolio theory) technique in the financial field. As long as the risks and returns of individual stock investments can be assumed, this is a method for determining the optimal investment allocation that gives the minimum risk for any return.

  For example, consider the case of investing in stocks of an umbrella store and an ice cream store. If summer is sunny and hot, the ice cream shop will make money, but the umbrella shop won't make money. Conversely, if it continues to rain and cold days continue, the umbrella shop will make money, but the ice cream shop will not make money. By investing in both ice cream and umbrella stocks, it is possible to obtain stable returns regardless of weather risks.

  This portfolio theory will be described with reference to the drawings. FIG. 8 is an explanatory diagram for explaining the portfolio theory. As shown in FIG. 8, it is assumed that the risks and returns of stocks A to C can be assumed. Here, the risk is the variance (variation) in stock price movements, and the return is the rate of price increase (the future state is unknown, so the stock price variance and average price increase rate over the past few years are usually used. ).

  The risk and return when stock A and stock B are held in the ratio of xa: xb is always one point on arc AB, and the risk is lower than when stock A and stock B are individually owned. . The risk and return when the stock A, the stock B, and the stock C are held at a ratio of xa: xb: xc are one point on the plane surrounded by the arc AB, the arc BC, the arc CD, and the arc DA. The optimal solution with small risk (variance) and large return is D point.

  Generally, stock investment allocation is a problem of minimizing the objective function (risk) under the following constraints (more than an arbitrary return). The optimal solution (investment allocation) is obtained using quadratic programming. It is done.

(Equation 1)
Objective function: f (x) = 1 / 2ρ ²
Constraint formula: r ₁ x ₁ + r ₂ x ₂ +... + R _n x _n ≧ E

Here, x: investment ratio, r: individual rate of return, E: arbitrary rate of return, and ρ: portfolio variance.
Such detailed explanation of portfolio theory is also described in the following books, for example.
"Financial engineering and optimization, written by Kiki, Asakura Shoten"

JP 2001-14295 A (paragraph numbers 0020 to 0035)

  In the above-described invention of Patent Document 1 (data prediction method, data prediction device, and recording medium), a method for selecting one prediction method from a plurality of prediction methods is used, but the prediction result is one of the prediction methods prepared in advance. Limited. That is, when n prediction models are prepared, the best one can be selected from the n prediction models, but since the number of selection candidates is limited to n, a highly accurate prediction result is obtained. If you want, you need to increase the number of n. When many prediction methods are prepared, there is a problem that the cost of systemization increases accordingly.

  In addition, the portfolio theory set forth as the prior art is a method for stock investment. Even if it is directly applied, it seeks a ratio with high return (average value) and small variation (standard deviation), so it can be applied directly to the prediction problem. Can not do it. In other words, there is no current application method, and constraint equations and objective functions cannot be applied as they are.

  The present invention has been made in order to solve the above-described problems, and an object of the present invention is to provide a prediction method and a prediction apparatus that improve prediction accuracy by processing a plurality of prediction data into high-precision prediction data. Another object is to provide a prediction program and a recording medium.

  The present invention solves the above-described problems. In predictions such as energy demand forecasting, sales volume demand forecasting, price forecasting, and economic index forecasting, forecast data is calculated by a plurality of forecasting methods, and each forecast data is processed. As a result, more accurate prediction data is created. The prediction method uses statistical methods and financial engineering. Since it has a function of processing the prediction data, even if the number of prediction methods to be implemented is relatively small, it can have the same performance as a large number of prediction methods. This has the following advantages.

(1) It is possible to save time and effort for developing and improving the prediction method, and to reduce costs.
(2) Even if a plurality of prediction methods are implemented, highly accurate prediction results can be automatically obtained without depending on human experience and intuition.
(3) Even if only a small number of prediction methods are implemented, performance sufficient to implement a large number of prediction methods can be obtained. That is, a highly accurate prediction result can be obtained. (4) Due to the feature (3) described above, it is not necessary to implement a large number of prediction methods, and the cost of system construction can be further reduced.

The prediction method according to claim 1 of the present invention includes:
A prediction method for calculating prediction data related to future prediction,
a prediction procedure for performing prediction using n different prediction methods and calculating n types of prediction data;
a prediction data processing procedure for calculating an average of n types of prediction data as prediction data;
It is characterized by having.

Moreover, the prediction method according to claim 2 of the present invention includes:
A prediction method for calculating prediction data related to future prediction,
a prediction procedure for performing prediction using n different prediction methods and calculating n types of prediction data;
An index calculation procedure for calculating an index based on a prediction error,
A prediction data processing procedure for calculating prediction data by weighted averaging of n types of prediction data based on an index;
It is characterized by having.

Moreover, the prediction method according to claim 3 of the present invention includes:
The prediction method according to claim 2,
The indicator calculation procedure is a procedure for calculating n types of error indicators and error indicator sums calculated for each of the n types of predictions.
The prediction data processing procedure uses n types of prediction data, n types of error indices, and error index summation,
Prediction data = Σ {prediction data i × (error index total−error index i) / error index total}
It is the procedure which processes prediction data so that it may become.

A prediction method according to claim 4 of the present invention is:
The prediction method according to claim 2,
The index calculation procedure is a procedure for calculating the weighted ratio so that the weighted ratio constrained by the constraint equation and the objective function including the index are minimized or maximized.
The prediction data processing procedure is a procedure for processing into prediction data based on a weighted average of n types of prediction data using this weight ratio.

A prediction device according to claim 5 of the present invention is
A prediction device for calculating prediction data related to future prediction,
a prediction means for performing prediction by n different prediction methods and calculating n types of prediction data;
prediction data processing means for calculating an average of n types of prediction data as prediction data;
It is characterized by having.

Moreover, the prediction device according to claim 6 of the present invention provides:
A prediction device for calculating prediction data related to future prediction,
a prediction means for performing prediction by n different prediction methods and calculating n types of prediction data;
Index calculation means for calculating an index based on the prediction error;
Prediction data processing means for calculating prediction data by weighted averaging of n types of prediction data based on an index;
It is characterized by having.

Moreover, the prediction device according to claim 7 of the present invention provides:
The prediction device according to claim 6,
The index calculation means is a means for calculating n types of error indexes calculated for each of the n types of predictions and a total of error indexes.
The prediction data processing means uses n types of prediction data, n types of error indicators, and error indicator summation,
Prediction data = Σ {prediction data i × (error index total−error index i) / error index total}
It is a means which processes prediction data so that it may become.

Moreover, the prediction device according to claim 8 of the present invention provides:
The prediction device according to claim 6,
The index calculation means is a means for calculating the weight ratio so that the objective function including the weight ratio and the index constrained by the constraint equation is minimized or maximized.
The prediction data processing means is a means for processing into prediction data based on a weighted average of n types of prediction data using this weight ratio.

The prediction program according to claim 9 of the present invention is
A prediction program for calculating prediction data related to future prediction,
a prediction means for performing prediction by n different prediction methods and calculating n types of prediction data;
prediction data processing means for calculating an average of n types of prediction data as prediction data;
It is a program which makes a computer function as.

Moreover, the prediction program according to claim 10 of the present invention includes:
A prediction program for calculating prediction data related to future prediction,
a prediction means for performing prediction by n different prediction methods and calculating n types of prediction data;
Index calculation means for calculating an index based on the prediction error;
Prediction data processing means for calculating prediction data by weighted averaging of n types of prediction data based on an index;
It is a program which makes a computer function as.

A prediction program according to claim 11 of the present invention is
The prediction program according to claim 10,
The index calculation means is a means for calculating n types of error indexes calculated for each of the n types of predictions and a total of error indexes.
The prediction data processing means uses n types of prediction data, n types of error indicators, and error indicator summation,
Prediction data = Σ {prediction data i × (error index total−error index i) / error index total}
It is a means which processes prediction data so that it may become.

A prediction program according to claim 12 of the present invention includes:
The prediction program according to claim 10,
The index calculation means is a means for calculating the weight ratio so that the objective function including the weight ratio and the index constrained by the constraint equation is minimized or maximized.
The prediction data processing means is a means for processing into prediction data based on a weighted average of n types of prediction data using this weight ratio.

A recording medium according to claim 13 of the present invention provides:
A recording medium in which a prediction program for calculating prediction data related to future prediction is recorded in a computer-readable manner,
a prediction means for performing prediction by n different prediction methods and calculating n types of prediction data;
prediction data processing means for calculating an average of n types of prediction data as prediction data;
As described above, a program for causing a computer to function is recorded.

A recording medium according to claim 14 of the present invention is
A recording medium in which a prediction program for calculating prediction data related to future prediction is recorded in a computer-readable manner,
a prediction means for performing prediction by n different prediction methods and calculating n types of prediction data;
Index calculation means for calculating an index based on the prediction error;
Prediction data processing means for calculating prediction data by weighted averaging of n types of prediction data based on an index;
As described above, a program for causing a computer to function is recorded.

A recording medium according to claim 15 of the present invention is
The recording medium according to claim 14,
The index calculation means is a means for calculating n types of error indexes calculated for each of the n types of predictions and a total of error indexes.
The prediction data processing means uses n types of prediction data, n types of error indicators, and error indicator summation,
Prediction data = Σ {prediction data i × (error index total−error index i) / error index total}
It is a means which processes prediction data so that it may become.

A recording medium according to claim 16 of the present invention provides:
The recording medium according to claim 14,
The index calculation means is a means for calculating the weight ratio so that the objective function including the weight ratio and the index constrained by the constraint equation is minimized or maximized.
The prediction data processing means is a means for processing into prediction data based on a weighted average of n types of prediction data using this weight ratio.

  According to the present invention as described above, it is possible to provide a prediction method, a prediction device, a prediction program, and a recording medium that improve prediction accuracy by processing a plurality of prediction data into high-precision prediction data. Can do.

Hereinafter, a prediction method, a prediction apparatus, a prediction program, and a recording medium according to the best mode of the first invention will be described with reference to the drawings.
First, the prediction method will be described with reference to the drawings. FIG. 1 is a flowchart for explaining the prediction method of this embodiment.

  Step S10 is a procedure (prediction procedure) in which n types of prediction data are calculated by performing prediction using n different types of prediction methods. Here, the details of the n different prediction methods are not the gist of the present invention and will not be described in detail. However, typical prediction methods include multiple regression equations, neural networks, fuzzy inference, and autoregressive models. and so on. In particular, the forecast method for the financial sector is slightly different from the forecast, but the moving average value, RSI (Relative Strength Index), RCI (Rank Correlation Index) used in the financial sector are slightly different from the forecast. ), Various statistical methods such as candlestick and Bollingger Band and chart display methods are also included in the prediction method. Furthermore, a prediction method for calculating processed prediction data so that the objective function is maximized or minimized may be included. In short, the prediction procedure includes various methods for assisting prediction as prediction methods in addition to the prediction method for calculating prediction data.

  Step S11 is a procedure (predicted data processing procedure) for calculating an average value of n types of predicted data as predicted data. An average obtained by dividing the sum of all n types of predicted data 1, predicted data 2,..., predicted data n by n is used as predicted data, and is expressed by the following equation.

(Equation 2)
Prediction data = (Σ prediction data i) / n

  The prediction data calculated in this way has an advantage that high-precision prediction data can be calculated by simple and high-speed processing.

  Next, the prediction device will be described with reference to the drawings. FIG. 2 shows a prediction apparatus according to this embodiment. The prediction apparatus includes a data input unit 10, a data storage unit 20, a central processing unit 30, and a data output unit 40. The central processing unit 30 includes a prediction unit 50 and a prediction data processing unit 60.

The data input means 10 is composed of a keyboard or a communication device such as a LAN / Internet and is a means for inputting performance data, calendar data, and parameters necessary for prediction.
The data storage means 20 is composed of a hard disk, a magneto-optical disk, a flexible disk, and a memory, and is a means for storing input performance data, calendar data, and parameters. In addition, various calculation results calculated during the prediction are stored.

The central processing unit 30 is equipped with a prediction program including the prediction procedure and the prediction data processing procedure as described above, reads actual data, calendar data, and parameters from the data storage means 20, and predicts the prediction means 50 and the prediction data processing. It is a means that functions as the means 60 and performs arithmetic processing, and writes predicted data into the data storage means 20.
The data output means 40 includes a display, a transmission device, and a printer, and displays and transmits prediction results based on final prediction data, prediction results of each prediction method, and various calculation results calculated during prediction. Is a means for displaying or transmitting. For example, in a prediction apparatus related to the financial field, charts such as candlestick, moving average, RCI, and RSI are displayed.

  The prediction unit 50 is a unit that calculates n types of prediction data by performing predictions using n types of prediction units 1 to n, respectively. Here, typical prediction methods of the respective prediction means 1 to n include a multiple regression equation, a neural network, a fuzzy reasoning, an autoregressive model, and the like. Further, in the prediction apparatus used in the financial field, various statistical methods such as moving average values, RSI, RCI, candlesticks, Bollinger bands, and chart display methods used in the financial field are included in the prediction method. Furthermore, a prediction method for calculating processed prediction data so that the objective function is maximized or minimized may be included.

The predicted data processing means 60 is a means for calculating an average value of n types of predicted data as predicted data. The average data obtained by dividing the sum of all n types of prediction data 1, prediction data 2,..., prediction data n by n is used as prediction data.
The algorithm for calculating the prediction data in this way has advantages such as fast processing and high prediction accuracy.

  The operation of such a prediction device will be described. In the prediction device, actual data, calendar data, and parameters necessary for prediction are registered in the data storage unit 20 through the data input unit 10 in advance. When the prediction apparatus predicts, first, the central processing unit 30 functions as a reading unit that reads out the actual data, calendar data, and parameters and temporarily stores them in a storage unit (not shown). Subsequently, the central processing unit 30 functions as a prediction unit that calculates n types of prediction data as described above. Subsequently, the central processing unit 30 functions as a prediction data processing unit that calculates an average of n types of prediction data as prediction data. Subsequently, the central processing unit 30 functions as a writing unit that writes the calculated prediction data in the data storage unit 20. Finally, the central processing unit 30 functions as an output unit that outputs prediction data from the data storage unit 20 via the data output unit 40.

  In this way, the prediction apparatus of the present embodiment predicts. The present invention is not limited to the above-described apparatus configuration. For example, one computer including the data input means 10, the data storage means 20, the central processing unit 30, and the data output means 40. It is also good. If the central processing unit 30 functions as the prediction unit 50 and the prediction data processing unit 60, the computer can perform the prediction.

  The prediction program is a program that causes the central processing unit 30 of the prediction apparatus to function as the prediction unit 50 and the prediction data processing unit 60, and is stored in a storage unit (not shown) connected to the central processing unit 30. ing. A prediction program is installed in the prediction device using a recording medium on which such a prediction program is recorded (for example, a medium such as a CD-ROM, MO, FD, and HD that is recorded / reproduced by magnetism, light, and magnetomagnetism). Alternatively, the prediction program may be installed in a storage unit (not shown) of the prediction apparatus via a network / LAN (not shown) such as the Internet.

Next, a prediction method, a prediction apparatus, a prediction program, and a recording medium according to another best mode of the second invention will be described with reference to the drawings.
First, the prediction method will be described with reference to the drawings. FIG. 3 is a flowchart for explaining the prediction method of this embodiment.

  Step S20 is a procedure (prediction procedure) in which n types of prediction data are calculated by performing prediction using n different types of prediction procedures. Here, the prediction method used in each prediction procedure includes a multiple regression equation, a neural network, fuzzy reasoning, an autoregressive model, and the like. In addition, various statistical methods such as moving average values, RSI, RCI, candlesticks, and Bollinger bands used in the financial field, chart display methods, and the like are also included in the prediction method.

  Step S21 is a procedure (index calculation procedure) for calculating an index based on the prediction error. The prediction error is calculated by calculating the prediction data 1 to n of a certain day by the prediction means 1 to n using the calendar data, the actual data, and the parameters, and predicts the error between the prediction data and the actual actual value on that day. An error is calculated for each of n, and the errors accumulated for such predictions 1 to n over many years are prediction errors 1 to n. The prediction error is obtained for each of the prediction data 1 to n. The indices 1 to n of the prediction data 1 to n are calculated based on the prediction errors 1 to n such as the past n hours, the past n days, and the past n months. For example, the index 1 for the prediction data 1 of the prediction 1 is calculated using the past prediction error 1, and similarly, the index n for the prediction data n of the prediction n is calculated using the past prediction error n.

  Normally, the average value, standard deviation, variance, maximum / minimum value, absolute value average value, and relative coefficient of the prediction errors 1 to n are the indices 1 to n, but are not limited to these statistical indices. For example, an index based on a prediction error between predicted data obtained by statistically calculating a value representing a stock price peak or a value representing a stock price bottom in an RSI or RCI index often used in the financial field and the like is included. These indexes and methods largely depend on the prediction target and the prediction method, but the present invention is not limited to one index and method, and various types of adoption are possible.

  Step S22 is a procedure (prediction data processing procedure) for calculating one prediction data by performing weighted averaging of n types of prediction data based on the indices 1 to n. At this time, indices 1 to n are used. For example, an average value of prediction errors for the past k days for a certain prediction i is set as an error index i, and the error index i is calculated by adding the error index i for each of predictions 1 to n as an error index sum, and these n types of predictions Predictive data is calculated using the data, n types of error indices, and error index summation as in the following equation.

(Equation 3)
Prediction data = Σ {prediction data i × (error index total−error index i) / error index total}

Prediction data calculated in this way has the advantage that the specific gravity of prediction data with a small error increases and prediction accuracy increases.
Furthermore, the average error of the processed prediction data may be reduced so that the variance of the error is small so that the objective function is maximized or minimized.

  Next, the prediction device will be described with reference to the drawings. FIG. 4 is a configuration diagram of the prediction apparatus of the present embodiment. The prediction apparatus includes a data input unit 10, a data storage unit 20, a central processing unit 30, and a data output unit 40. The central processing unit 30 includes a prediction unit 50, a prediction data processing unit 60, and an index calculation unit 70.

The data input means 10 is composed of a keyboard or a communication device such as a LAN / Internet and is a means for inputting performance data, calendar data, and parameters necessary for prediction.
The data storage means 20 is composed of a hard disk, a magneto-optical disk, a flexible disk, and a memory, and is a means for storing input performance data / calendar data / parameters. It also stores various results calculated in the system.

  The central processing unit 30 is equipped with a prediction program including a prediction procedure, an index calculation procedure, and a prediction data processing procedure as described above, and functions as a means for reading actual data, calendar data, and parameters from the data storage means 20; It functions as a prediction unit 50, an index calculation unit 70, and a prediction data processing unit 60 to obtain prediction data and write the prediction data in the data storage unit 20.

  The data output means 40 includes a display, a transmission device, and a printer, and displays and transmits prediction results based on final prediction data, prediction results of each prediction method, and various calculation results calculated during prediction. Is a means for displaying and transmitting. For example, in a prediction apparatus related to the financial field, charts such as candlestick, moving average, RCI, and RSI are displayed.

  The prediction unit 50 is a unit that calculates n types of prediction data by performing prediction using n different types of prediction methods. Here, typical prediction methods include multiple regression equations, neural networks, fuzzy reasoning, and autoregressive models. Further, in the prediction apparatus in the financial field, various statistical methods such as moving average value, RSI, RCI, candlestick, Bollinger band, chart display method, and the like are also included in the prediction method.

The index calculation means 70 is a means for calculating an index based on the prediction error.
The indices 1 to n of the prediction data 1 to n are calculated based on the prediction errors 1 to n such as the past n hours, the past n days, and the past n months. For example, the index 1 for the prediction data 1 related to the prediction 1 is calculated using the past prediction error 1, and similarly, the index n for the prediction data n related to the prediction n is calculated using the past prediction error n.
Normally, the average value, standard deviation, variance, maximum / minimum value, absolute value average value, and relative coefficient of the prediction errors 1 to n are the indices 1 to n, but are not limited to these statistical indices. For example, an index based on a prediction error between the prediction data obtained by statistically calculating the value representing the stock price peak or the value representing the stock price bottom in the RSI or RCI index often used in the financial field and the like is included. These indexes and methods largely depend on the prediction target and the prediction method, but the present invention is not limited to one index and method, and various types of adoption are possible.

The prediction data processing unit 60 is a unit that calculates one prediction data by performing weighted averaging of n types of prediction data based on the indices 1 to n. At this time, indices 1 to n are used. For example, an average value of prediction errors for the past k days for a certain prediction i is set as an error index i, and the error index i is calculated by adding the error index i for each of predictions 1 to n as an error index sum, and these n types of predictions Using the data, n types of error indices, and error index summation, for example, the prediction data is calculated as shown in Equation 3 above. In such prediction data, there is an advantage that the specific gravity of the prediction data i having a small prediction error is increased and the prediction accuracy is increased.
Furthermore, the average error of the processed prediction data may be reduced so that the variance of the error is small so that the objective function is maximized or minimized.

  The operation of such a prediction device will be described. In the prediction device, actual data, calendar data, and parameters necessary for prediction are registered in the data storage unit 20 through the data input unit 10 in advance. When the prediction apparatus predicts, first, the central processing unit 30 functions as a reading unit that reads out the result data, calendar data, and parameters and temporarily stores them in a storage unit (not shown). The central processing unit 30 functions as a prediction unit that calculates n types of prediction data as described above, functions as an index calculation unit that calculates an index based on a prediction error, and weights n types of prediction data based on the index. Functions as prediction data processing means for calculating prediction data on average. The central processing unit 30 functions as a writing unit that writes prediction data into the data storage unit 20. Finally, the central processing unit 30 outputs prediction data from the data storage unit 20 via the data output unit 40.

  The prediction device predicts in this way. The present invention is not limited to the above-described apparatus configuration. For example, one computer including the data input means 10, the data storage means 20, the central processing unit 30, and the data output means 40. If the central processing unit 30 functions as the prediction unit 50, the index calculation unit 70, and the prediction data processing unit 60, the computer can perform the prediction.

  The prediction program is a program that causes the central processing unit 30 of the prediction apparatus to function as the prediction unit 50, the index calculation unit 70, and the prediction data processing unit 60, and a storage unit (see FIG. (Not shown). A prediction program is installed in the prediction device using a recording medium on which such a prediction program is recorded (for example, a medium such as a CD-ROM, MO, FD, and HD that is recorded / reproduced by magnetism, light, and magnetomagnetism). Alternatively, the prediction program may be installed in a storage unit (not shown) of the prediction apparatus via a network / LAN (not shown) such as the Internet.

Next, Example 1 of the first invention will be described with reference to the drawings. 5 and 6 are explanatory diagrams for explaining a specific example of the prediction means, and FIG. 7 is a power-day characteristic diagram for comparing the actual value with the prediction data.
In the first embodiment, the average of the prediction data 1 and 2 of the prediction means 1 and 2 is used as the prediction data, and the prediction device shown in FIG. 2 is used. The prediction means 50 and the prediction data processing means 60 are used. The operation of will be specifically described. As an example of prediction, the maximum power demand prediction will be described.

  The maximum power demand prediction is a problem of predicting the maximum power of the next day based on the weather and past actual power values. In the present embodiment, in order to simplify the explanation, a means having two different prediction means 1 and 2 is assumed. An example of a specific prediction method is shown. The prediction model of the prediction means 1 shown in FIG. 5 is the neural network 1 that outputs the maximum power prediction data of the target day from the highest and lowest temperatures of the nearest few days, and the prediction model of the prediction means 2 shown in FIG. This is a neural network 2 that outputs daily hourly power prediction data. The evaluation target in this example was the maximum power, and in the prediction model 2, only the maximum of 24 points was evaluated as the maximum power value.

  Predicted data 1 and 2 are calculated by these predicting means 1 and 2. There is a slight prediction error between the prediction data and the actual data as shown in FIG. A prediction error is calculated from such prediction data and actual data. Prediction error is calculated by calculating the forecast data 1 and 2 for a certain day using the forecast means 1 and 2 using the meteorological result data about the weather, the calendar data about the month and day, and the past power data. The error between the prediction data on that day and the actual actual value is defined as the errors 1 and 2 of the day, and the errors 1 and 2 accumulated for many years are the prediction errors 1 and 2. Statistical indexes 1 and 2 are calculated using such prediction errors 1 and 2. Statistical indices 1 and 2 for one-year prediction errors 1 and 2 by the prediction means 1 and 2 are shown in the following table.

  In the two prediction means 1 and 2, the standard deviation and the absolute value average error are approximately the same, but the average error is slightly different. Although the prediction means 1 and 2 output different prediction data 1 and 2, it can be seen that there is no significant difference.

The prediction unit 50 performs the maximum power prediction for i days. The predicted maximum power of i day by the prediction unit 1 is represented by prediction data 1 (i), and the predicted maximum power value 2 (i) of i day by the prediction unit 2.
The predicted data processing means 60 calculates the average value of predicted data using the following formula.

(Equation 4)
Prediction data = avg (prediction data 1 (i), prediction data 2 (i))

  Table 2 shows statistical indexes of prediction errors due to prediction by the prediction apparatus of the first embodiment. When compared with the statistical index of Table 1 that does not take an average, the value is reduced compared with the statistical index of Table 1, indicating the effectiveness of the present invention.

Next, Example 2 of the second invention will be described with reference to the drawings.
In the first embodiment, the prediction device shown in FIG. 4 is used, and the operations of the prediction unit 50, the prediction data processing unit 60, and the index calculation unit 70 of the prediction device of the second invention will be specifically described. As an example of prediction, the above-described maximum power demand prediction will be described. Also in the second embodiment, in order to simplify the explanation, the prediction means 30 is assumed to be two prediction means 1 and 2, and the prediction model of the prediction means 1 is the target date from the highest and lowest temperatures of the nearest few days described in FIG. The prediction model of the prediction means 2 is the neural network 2 that predicts the hourly power of the target day described with reference to FIG.

  The prediction unit 50 performs the maximum power prediction for i days. The predicted maximum power of i day by the prediction unit 1 is represented by the prediction data 1 (i), and the predicted maximum power of day i by the prediction unit 2 is represented by the prediction data 2 (i).

Subsequently, error indexes 1 and 2 for the prediction data 1 and 2 are calculated by the index calculation means 70.
Here, the error indexes 1 and 2 are calculated using a prediction error. In the second embodiment, an error index based on the nearest absolute value average error for the past two days is used. The reason for using the nearest absolute value average error is that it is impractical to use the above statistical indicators over a long period of time because the processing becomes longer or the situation in the immediate vicinity cannot be reflected. is there.

   First, the prediction errors 1 and 2 of the previous day (i-1 day) by the respective prediction means 1 and 2 are represented by the prediction error 1 (i-1) and the prediction error 2 (i-1). Similarly, when the prediction error two days ago (assumed to be i-2 days) is represented by prediction error 1 (i-2) and prediction error 2 (i-2), each error index 1, 2 and error index sum total Can be calculated by the following formula.

(Equation 5)
Error index 1 = {| prediction error 1 (i-1) | + | prediction error 1 (i-2) |} / 2
Error index 2 = {| prediction error 1 (i-2) | + | prediction error 2 (i-2) |} / 2
Error index sum = error index 1 + error index 2

  Subsequently, the predicted data processing means 60 calculates predicted data for the maximum power demand by a weighted average based on the error indexes 1 and 2 described above. In this embodiment, the general formula for weighted average of the prediction data is as follows.

(Equation 6)
Prediction data = Σ {prediction data k × (total error index−error index k) / total error index}

here,
Prediction data k: Prediction data error index by prediction means k k: Statistical index error index calculated using past prediction errors (variance, absolute value average error, standard deviation, maximum, minimum error, average error) by prediction means k Of Σ: Σ error index k

  Here, assuming that error index 1 = 1%, error index 2 = 3%, and total error index = 4%, final prediction data is processed as follows using these error indexes. In this example, the prediction data 1 and the prediction data 2 are weighted at a ratio of 3 to 1 and weighted average is the final prediction data.

(Equation 7)
Prediction data = Σ {prediction data k × (total error index−error index k) / total error index}
= {Prediction data 1 × (4-1) + Prediction data 2 × (4-3) / 4
= {3 × predicted data 1 + 1 × predicted data 2} / 4

  The statistical index calculated using the prediction error between the prediction data and the actual value is shown in the following table. When compared with the statistical index of Table 3 and the statistical index of Table 1 that does not take a weighted average, the value is reduced compared to the statistical index of Table 1, indicating the effectiveness of the present invention.

In the above, the prediction apparatus of Example 1, 2 was demonstrated. In the prediction apparatus of the present embodiment, the absolute value average error of the past two days has been described as an example. However, the statistical index may not be the absolute value average error, and may be an index that can determine whether the prediction result is good, For example, the maximum error or the average error may be used.
Further, the target number of past days may not be two days, and may be one day or seven days. In this embodiment, two examples of prediction means have been described. However, when there are a large number of prediction means, the top n weighted averages with a small error, or the weighted average values of only those below a certain error index. It may be.

  Subsequently, Example 3 according to the second invention will be described. The present embodiment is an invention for processing forecast data by a method improved from the concept of a so-called financial engineering portfolio theory. In portfolio theory in financial engineering, a ratio with high return (average) and small variation (standard deviation) is calculated. However, if applied directly to the prediction problem, the standard deviation is small, but the one with a large error is selected, so the portfolio theory of financial engineering cannot be applied directly. In the third embodiment, an improvement is added.

Hereinafter, this embodiment will be specifically described.
The prediction means 50 and the index calculation means 70 are the same as those in the second embodiment described above. Two types of prediction data 1 and 2 are calculated from the two prediction means 1 and 2 of the prediction means 50, and the index calculation means 70 performs the prediction. Error indexes 1 and 2 for data 1 and 2 are calculated. Here, the error index is not limited to the absolute value average error, and includes standard deviation, variance, average error, maximum error, minimum error, and the like.

  The prediction data processing means 60 constructs a portfolio of prediction data 1 and 2 and processes the prediction data. That is, when there are two prediction means, two pieces of prediction data are obtained. The two pieces of prediction data are weighted and averaged at a certain ratio to calculate the prediction data. Of course, when there are n prediction means, the final prediction data is calculated by performing a weighted average using n prediction data.

  The weighted average can be calculated by obtaining an objective function under a certain constraint equation. The constraint equation and the objective function can be arbitrarily set. In the present embodiment, the following constraint equation and objective function will be described as an example.

(Equation 8)
[Constraint expression]
a + b = 10
a and b are integers of 0 to 10, a and b are weighted ratios [objective function]
1Minimizing absolute average error over the past week

  The above constraint equation is a condition that must be satisfied. That is, the weighting ratio of the prediction data means any one of 11 types of 0:10, 1: 9,..., 10: 0. The objective function is the minimization of the absolute average error over the past week. That is, the absolute value average errors 1 and 2 from the i-7 day to the i-1 day are obtained, and the error index = a × the absolute value average error 1 + b × the absolute value average error 2 is calculated. Assuming that the weight ratio with the smallest error index is 6: 4, the prediction data for i day is calculated by the following equation.

(Equation 9)
Prediction data = (6 × prediction data 1 + 4 × prediction data 2) / 10

  The optimal weight ratio calculation method varies depending on the constraint equation and the objective function, but can be easily calculated by a general method such as quadratic programming, GA (genetic algorithm), or enumeration. In this example, the calculation was performed using an enumeration method (all 11 types of prediction data errors were calculated, and the weighted ratios (values a and b) indicating the minimum values were used as the optimal weighted ratio).

  For example, in the above constraint equation, if the condition that a and b are integers is removed, the weighting ratio becomes an infinite combination, so the enumeration method cannot be used. The ratio can be calculated. (Secondary programming itself can use common commercial tools).

  Note that the present invention is not limited to the above-described constraint equations and objective functions described in the present embodiment, and can be set with any constraint equation or objective function. In the building part). In the following, three types of examples of objective functions are added, and Table 4 shows error indicators for a total of four types of objective functions.

(Equation 10)
[Constraint expression]
a + b = 10
a and b are integers of 0 to 1, a and b are weighted ratios [objective function]
1Minimization of absolute average error over the past week
2Minimizing the average error over the past week
3Minimization of standard deviation over the past week
4Minimization of average error + standard deviation over the past week

  Table 4 shows the statistical indicators. Although four types of formulation methods have been implemented, the values are smaller than those in Table 1 that do not take a weighted average, indicating the effectiveness of the present invention.

  In the third embodiment, eleven pieces of prediction data can be generated despite only two types of prediction means. That is, it can be said that the same effects as those obtained by mounting 11 prediction means are obtained. Of course, if the number of prediction means to be implemented increases, it is possible to generate a large amount of prediction data exponentially, and it goes without saying that the effect of the present invention is increased.

The best mode and embodiment of the present invention have been described above.
By using the present invention, it is possible to shorten the construction time of the prediction system and realize cost reduction. In addition, high prediction data can be obtained.
Conventionally, it is common to implement one prediction method for one prediction target. In order to implement a prediction method with a high degree of accuracy as much as possible, it takes a long time to analyze a very long data analysis, to verify one prediction method from multiple prediction methods, or to develop a new prediction method, These are the causes of increased costs during system development. As a conventional method for solving this, there is a method of mounting a plurality of prediction methods without being limited to one prediction method. This method can shorten the data analysis period and the method selection period, and can reduce the cost by shortening the development period. However, even in this method, there is no standard for selecting which prediction data should be adopted from a plurality of prediction data.

In the present invention, since one prediction data can be obtained by selecting and processing a plurality of prediction data, it is possible to simultaneously realize cost reduction and high prediction accuracy by shortening the development period. In particular, in the present invention, even if n prediction methods are implemented, the prediction data can be processed by a combination of n prediction methods or an average process. It has the same effect as the system that selects That is, highly accurate prediction is possible even if there are few prediction methods to be mounted, and the cost of the system can be reduced by reducing the number of prediction methods to be mounted.
In addition, although prediction of the maximum power demand has been described in this embodiment, the present invention can be applied to various predictions such as energy demand prediction, sales volume prediction, price prediction, and economic index prediction.

  Further, since the invention of Patent Document 1 is a method of selecting one optimal prediction data from a plurality of prediction means, the prediction data is limited to the prediction data of the prediction means prepared in advance. That is, when n prediction means are prepared, one prediction data is selected from the n prediction data. Therefore, when high-precision prediction is expected, a number of prediction methods are implemented, and there is a drawback that the cost of system construction increases.

  On the other hand, in the present invention, it is possible to process a plurality of prediction data and obtain prediction data with a smaller error. Therefore, when n prediction means are prepared, the combination can provide the same effect as that provided by far more than n prediction means. In the case of the embodiment of claim 3 described above, only two prediction means are implemented, but 11 prediction data are constructed by the combination thereof. Compared with the case of constructing a system having 11 prediction means in the invention of Patent Document 1, the cost can be reduced to 2/11. Of course, in the present invention, when the number of prediction methods to be implemented is increased to 3, 4, it goes without saying that the combination increases exponentially and the effect of the present invention is further increased.

It is a flowchart explaining the prediction method of the best form for implementing 1st invention. It is a block diagram of the prediction apparatus of the best form for implementing 1st invention. It is a flowchart explaining the prediction method of the best form for implementing 2nd invention. It is a block diagram of the prediction apparatus of the best form for implementing 2nd invention. It is explanatory drawing explaining the specific example of a prediction means. It is explanatory drawing explaining the specific example of a prediction means. It is an electric power-day characteristic figure which compares a track record value and prediction data. It is explanatory drawing explaining portfolio theory.

Explanation of symbols

10: data input means 20: data storage means 30: central processing unit 40: data output means 50: prediction means 60: prediction data processing means 70: index calculation means

Claims (16)

A prediction method for calculating prediction data related to future prediction,
a prediction procedure for performing prediction using n different prediction methods and calculating n types of prediction data;
a prediction data processing procedure for calculating an average of n types of prediction data as prediction data;
The prediction method characterized by having. A prediction method for calculating prediction data related to future prediction,
a prediction procedure for performing prediction using n different prediction methods and calculating n types of prediction data;
An index calculation procedure for calculating an index based on a prediction error,
A prediction data processing procedure for calculating prediction data by weighted averaging of n types of prediction data based on an index;
The prediction method characterized by having. The prediction method according to claim 2,
The indicator calculation procedure is a procedure for calculating n types of error indicators and error indicator sums calculated for each of the n types of predictions.
The prediction data processing procedure uses n types of prediction data, n types of error indices, and error index summation,
Prediction data = Σ {prediction data i × (error index total−error index i) / error index total}
The prediction method characterized by being the procedure which processes prediction data so that it may become. The prediction method according to claim 2,
The index calculation procedure is a procedure for calculating the weighted ratio so that the weighted ratio constrained by the constraint equation and the objective function including the index are minimized or maximized.
The prediction data processing procedure is a procedure for processing into prediction data based on a weighted average of n types of prediction data using this weight ratio. A prediction device for calculating prediction data related to future prediction,
a prediction means for performing prediction by n different prediction methods and calculating n types of prediction data;
prediction data processing means for calculating an average of n types of prediction data as prediction data;
The prediction apparatus characterized by having. A prediction device for calculating prediction data related to future prediction,
a prediction means for performing prediction by n different prediction methods and calculating n types of prediction data;
Index calculation means for calculating an index based on the prediction error;
Prediction data processing means for calculating prediction data by weighted averaging of n types of prediction data based on an index;
The prediction apparatus characterized by having. The prediction device according to claim 6,
The index calculation means is a means for calculating n types of error indexes calculated for each of the n types of predictions and a total of error indexes.
The prediction data processing means uses n types of prediction data, n types of error indicators, and error indicator summation,
Prediction data = Σ {prediction data i × (error index total−error index i) / error index total}
A prediction device characterized by being means for processing prediction data so that The prediction device according to claim 6,
The index calculation means is a means for calculating the weight ratio so that the objective function including the weight ratio and the index constrained by the constraint equation is minimized or maximized.
The prediction data processing means is means for processing into prediction data based on a weighted average of n types of prediction data based on the weighted ratio. A prediction program for calculating prediction data related to future prediction,
a prediction means for performing prediction by n different prediction methods and calculating n types of prediction data;
prediction data processing means for calculating an average of n types of prediction data as prediction data;
A prediction program characterized by being a program that causes a computer to function as. A prediction program for calculating prediction data related to future prediction,
a prediction means for performing prediction by n different prediction methods and calculating n types of prediction data;
Index calculation means for calculating an index based on the prediction error;
Prediction data processing means for calculating prediction data by weighted averaging of n types of prediction data based on an index;
A prediction program characterized by being a program that causes a computer to function as. The prediction program according to claim 10,
The index calculation means is a means for calculating n types of error indexes calculated for each of the n types of predictions and a total of error indexes.
The prediction data processing means uses n types of prediction data, n types of error indicators, and error indicator summation,
Prediction data = Σ {prediction data i × (error index total−error index i) / error index total}
A prediction program characterized by being means for processing the prediction data so that The prediction program according to claim 10,
The index calculation means is a means for calculating the weight ratio so that the objective function including the weight ratio and the index constrained by the constraint equation is minimized or maximized.
The prediction data processing means is means for processing into prediction data based on a weighted average of n types of prediction data based on the weighted ratio. A recording medium in which a prediction program for calculating prediction data related to future prediction is recorded in a computer-readable manner,
a prediction means for performing prediction by n different prediction methods and calculating n types of prediction data;
prediction data processing means for calculating an average of n types of prediction data as prediction data;
As a recording medium, a program for causing a computer to function as a recording medium is recorded. A recording medium in which a prediction program for calculating prediction data related to future prediction is recorded in a computer-readable manner,
a prediction means for performing prediction by n different prediction methods and calculating n types of prediction data;
Index calculation means for calculating an index based on the prediction error;
Prediction data processing means for calculating prediction data by weighted averaging of n types of prediction data based on an index;
As a recording medium, a program for causing a computer to function as a recording medium is recorded. The recording medium according to claim 14,
The index calculation means is a means for calculating n types of error indexes calculated for each of the n types of predictions and a total of error indexes.
The prediction data processing means uses n types of prediction data, n types of error indicators, and error indicator summation,
Prediction data = Σ {prediction data i × (error index total−error index i) / error index total}
A recording medium, which is means for processing predicted data so that The recording medium according to claim 14,
The index calculation means is a means for calculating the weight ratio so that the objective function including the weight ratio and the index constrained by the constraint equation is minimized or maximized.
The prediction data processing means is means for processing into prediction data based on a weighted average of n types of prediction data based on the weighted ratio.

Images