JP5816148B2 - Output value prediction apparatus, output value prediction method, and program thereof - Google Patents

Description

  The present invention relates to an output value prediction technique for predicting an output value from given data, and in particular, an output value capable of presenting a reliability index that is an index representing the degree of confidence in a predicted output value (predicted value). It relates to prediction technology.

  In various fields, future predictions are often made when determining future actions. In particular, for products manufactured through various manufacturing processes in a relatively large-scale manufacturing plant, such as steel products and chemical products, for example, the amount of input, the amount of operation input, the passage of time, etc. Accordingly, the output value in each manufacturing process and the output value of the final process directly connected to the product often change every moment. For this reason, in order to control the output value, the prediction of the output value is important.

  In such output value prediction, if a prediction model for predicting an output value is given, and data for predicting an output value is given, the given data is used for the given prediction model. Thus, it is possible to predict the output value from the given data and obtain the predicted value. However, the predicted value obtained in this way is how small the difference from the actual result when performing the action such as operation of the manufacturing plant with the given data, and how reliable is the value, That is, it is unclear how reliable the value is. For this reason, it is difficult to determine whether or not a future action may be determined based on the predicted value, and the future action may be erroneously determined. Therefore, if a reliability index, which is an index representing the degree of confidence in the predicted value, is shown, the reliability index can be used to determine whether or not future actions may be determined based on the predicted value. Useful. Such a technique is disclosed in Patent Document 1, for example.

  The reliability scaled prediction apparatus disclosed in Patent Document 1 extracts a similar case set that is a set of cases similar to a predicted case from the known case set when a known case set and a predicted case are input. A similar case extraction unit, a certainty factor calculation unit for calculating a certainty factor of a certain prediction attribute value from the similar case set, a confidence factor for calculating a reliability measure of the certainty factor from the similar case set and the certainty factor And a reliability measure calculation unit for outputting a certainty factor of a certain predicted attribute value and a reliability measure of the certainty factor. Here, the certainty factor is a value indicating the probability that the prediction attribute of the prediction example is a certain value (for example, “the certainty factor that the prediction attribute value is OO value is 0.95”), for example, The certainty factor of c = (number of similar cases where the predicted attribute value is c) / (number of similar cases). The reliability measure is a value indicating the reliability of confidence.

JP 2003-323601 A

  By the way, when determining whether or not a future action may be determined based on a predicted value, one reliability index may be insufficient. For example, manufacturing conditions (for example, types and components of products and materials, upstream processes (upstream processes) to current processes (current process operating conditions, equipment used, command values, target values, manipulated variables, operation patterns and operations) Time) is determined based on the predicted value, the predicted value is data obtained when the past result data is manufactured recently, and the past result data is a product of the same type of product. Furthermore, it is preferable that the value is obtained based on data obtained after the same time has elapsed from a predetermined reference time such as the production start time. In such a case, the reliability index is the acquisition date of the past performance data. As long as the value is based on the actual data, the reliability of the predicted value can be evaluated from the perspective of the date of acquisition of past performance data. The reliability of the predicted value cannot be evaluated from the viewpoint of whether or not the data is past data, or from the viewpoint of whether the past performance data is data obtained after the lapse of the same time from a predetermined reference time point. When a future action is determined based on the predicted value, it may be erroneously determined.

  Since the reliability index of the present application is the reliability of the predicted value, it is similar to the certainty in Patent Document 1. In Patent Document 1, only one certainty is output, The case cannot be dealt with. In Patent Document 1, there are cases where a plurality of certainty factors are calculated. In this case as well, the plurality of certainty factors are integrated into one certainty factor by the certainty factor integrating unit, and only one is output.

  The present invention is an invention made in view of the above-mentioned circumstances, and its purpose is to obtain a reliability index representing the reliability from a plurality of viewpoints with respect to a predicted value, and to obtain the plurality of reliability indices obtained. To provide an output value prediction apparatus, an output value prediction method, and an output value prediction program that can be presented.

As a result of various studies, the present inventor has found that the above object is achieved by the present invention described below. That is, an output value prediction apparatus according to an aspect of the present invention is an output value prediction apparatus that predicts an output value from a given data by using given data for a given prediction model, and obtains a predicted value. A reliability index calculation unit that obtains a plurality of reliability indices that are indices representing the degree of reliability of the predicted value from a plurality of viewpoints; and a plurality of reliability indices that are obtained by the predicted value and the reliability index calculation unit; And the given prediction model is a model obtained based on past performance data, and the reliability index calculation unit is one of the plurality of reliability indices, It is obtained on the basis of conditional similarity that is the degree of similarity between conditional data that is data of a predetermined data item that affects the output value of data items in past performance data and the given data. At least determined and wherein Rukoto conditionally reliability indicators. The output value prediction method according to another aspect of the present invention includes an output value prediction step of predicting an output value from the given data by using the given data for a given prediction model and obtaining a predicted value; A reliability index calculation step for obtaining a plurality of reliability indexes, which are indexes representing the degree of reliability with respect to the predicted value, from a plurality of viewpoints; and a plurality of reliability indexes obtained in the prediction value and the reliability index calculation step; The given prediction model is a model obtained based on past performance data, and the reliability index calculation step includes the reliability index as one of the plurality of reliability indices, Obtained based on conditional similarity that is the degree of similarity between conditional data that is data of a predetermined data item that affects the output value of data items in past performance data and the given data At least it asked and said Rukoto the matter reliability index. An output value prediction program according to another aspect of the present invention is an output value prediction program for causing a computer to execute the given value data by using the given data for the given prediction model. An output value predicting step for predicting an output value from the output value, a reliability index calculating step for obtaining a plurality of reliability indexes from a plurality of viewpoints, which is an index representing a degree of confidence in the predicted value, and the predicted value A presentation step for presenting a plurality of reliability indexes determined in the reliability index calculation step, and the given prediction model is a model determined based on past performance data, and the reliability index The calculation step is a condition that is data of a predetermined data item that affects the output value among the data items in the past performance data as one of the plurality of reliability indexes. And at least asked characterized Rukoto conditionally reliability index obtained based on conditional similarity is similarity between the data and the given data. Here, the given prediction model means a prediction model given in advance before calculating the reliability index, and even if the prediction model is obtained in advance separately from the process of calculating the reliability index. In addition, a prediction model may be obtained in advance before calculating the reliability index in a series of processes of calculating the reliability index. Further, the given data is data given in advance for which a predicted value is to be obtained.

In the output value predicting device, the output value predicting method, and the output value predicting program having such a configuration, an output value is obtained as a predicted value by using given data for a given model, and a plurality of trusts for the predicted value are obtained. Sex indices are determined, and the determined predicted values and a plurality of reliability indices are presented. As described above, the output value prediction apparatus, the output value prediction method, and the output value prediction program according to the present invention obtain a reliability index that represents the reliability of a prediction value from a plurality of viewpoints, and the plurality of reliability thus obtained. Indicators can be presented. For this reason, the user can evaluate the obtained predicted value from a plurality of viewpoints by referring to the plurality of reliability indices presented, and decide the future action based on the predicted value. It is possible to make a reasonable judgment from these multiple viewpoints. The output value prediction apparatus, the output value prediction method, and the output value prediction program having such a configuration can present a conditional reliability index as one of a plurality of reliability indices. Therefore, the user can determine the conditional similarity between the given data and the past performance data by referring to the conditional reliability index, and evaluate the predicted value from this viewpoint. This makes it possible to reasonably judge the determination of future actions based on the predicted values from these conditional viewpoints .

According to another aspect, in the above-described output value prediction apparatus, the reliability index calculation unit is used when further obtaining the given prediction model as one of the plurality of reliability indices. It is characterized in that a daily reliability index obtained based on a daily similarity that is a similarity between an acquisition date on which past performance data is acquired and a predicted date in the given data is obtained.

Such an output value prediction unit configuration, as one of a plurality of reliability index further may present a number of days reliability index. Therefore, the user, by referring to the number of days reliability index can determine the number of days similarity with the past record data and the given data, evaluating the prediction value from this point of view This makes it possible to rationally determine the future action based on the predicted value from the viewpoint of days.

According to another aspect, in the above-described output value prediction apparatus, the given prediction model is a model that is obtained based on past performance data and includes an elapsed time from a predetermined reference time as an input variable, the reliability index calculation unit, as one of the plurality of the reliability index, further prior Symbol given the given data and acquisition date acquired past record data used when obtaining a predictive model The date reliability obtained based on the daily similarity, which is the similarity to the forecast date, and the acquisition time and the place where the past performance data used when obtaining the given forecast model is obtained It is characterized in that at least one of the predicted time reliability indices calculated based on the predicted time similarity that is the similarity to the predicted time in given data is obtained.

The output value prediction apparatus having such a configuration can present a conditional reliability index and at least one of a daily reliability index and a predicted temporal reliability index as a plurality of reliability indices. For this reason, the user may determine the conditional similarity between the given data and the past performance data by referring to the conditional reliability index when the conditional reliability index is presented. The predicted value can be evaluated from this point of view, and the number of days between the given data and the past performance data by referring to the day reliability index when the day reliability index is presented And predicting the given data and the past performance data by referring to the predicted time reliability index when the predicted time reliability index is presented. It is possible to judge temporal similarity, to evaluate the predicted value from this viewpoint, and to reasonably determine future action decisions based on the predicted value from the multiple viewpoints presented It can become.

Further, in another aspect, the output value predicting device described above, prior Symbol conditionally reliability index, the conditional similarity of the mean function of the mean value of the conditional similarity, the conditional similarity The number of past performance data for which is greater than or equal to a predetermined threshold, the function of the number of past performance data for which the conditional similarity is greater than or equal to a predetermined threshold, and the average value of the product of the conditional similarity and the day-to-day similarity , Any one or more of a function of an average value of products of the conditional similarity and the daily similarity.

  According to this configuration, it is possible to provide an output value prediction apparatus that can present any one or more of the above-described values as the conditional reliability index.

  According to another aspect, in the output value prediction apparatus described above, when the daily reliability index is included as one of the plurality of reliability indices, the daily reliability index is the daily similarity index. The average value of the daily similarity, a function of the average value of the daily similarity, the average value of the daily similarity with respect to past performance data in which the conditional similarity is within the upper fixed amount, and the conditional similarity is within the upper predetermined amount A function of the average value of the daily similarity with respect to the past performance data, the average value of the conditional similarity with respect to the past performance data in which the daily similarity is within the upper certain amount, and the day similarity is within the upper certain amount Any one or more of the functions of the average value of the conditional similarity with respect to the past performance data.

  According to this configuration, it is possible to provide an output value prediction apparatus that can present any one or more of the above-described values as the conditional reliability index.

  According to another aspect, in the output value prediction apparatus described above, when the predicted temporal reliability index is included as one of the plurality of reliability indices, the predicted temporal reliability index is the predicted time Average value of the similarities, a function of the average value of the predicted temporal similarity, a product of the conditional similarity and the predicted temporal similarity, a product of the conditional similarity and the predicted temporal similarity A function of the conditional similarity, the daily similarity and the predicted temporal similarity, a function of the conditional similarity, the daily similarity and the predicted temporal similarity, Function of average value of predicted time similarity for past performance data whose conditional similarity is within upper fixed amount, function of average value of predicted time similarity for past performance data whose conditional similarity is within upper fixed amount , The predicted temporal similarity is less than the upper certain amount Any one or more of: an average value of conditional similarity with respect to past performance data, and a function of an average value of conditional similarity with respect to past performance data in which the predicted temporal similarity is within the upper fixed amount It is characterized by being.

  According to this configuration, it is possible to provide an output value prediction apparatus that can present any one or more of the above-described values as the conditional reliability index.

  According to another aspect, the output value prediction apparatus further includes a scaling unit that scales each of the plurality of reliability indexes obtained by the reliability index calculation unit according to a predetermined reference, and the presentation unit includes: The prediction value and a plurality of reliability indices scaled by the scaling unit are presented.

  Since the output value predicting apparatus having such a configuration scales each of the plurality of reliability indexes, the plurality of reliability indexes can be compared with each other.

  According to another aspect, in the output value prediction apparatus described above, the given prediction model is a ladle or a tundish in a process from a converter steelmaking process to a continuous casting process through a molten steel treatment process. The model is characterized by modeling the temperature of molten steel.

  According to this configuration, the molten steel temperature in the ladle or in the tundish in the process from the converter steelmaking process to the continuous casting process through the molten steel treatment process is predicted, and a plurality of reliability indexes for the predicted value predicted. Can be provided.

  According to another aspect, in the above-described output value prediction apparatus, the given prediction model is a model in which a molten steel component or a molten steel temperature is modeled according to an integrated amount of blown blowing oxygen in a converter process. It is characterized by being.

  According to this configuration, in the converter process, the molten steel component or molten steel temperature corresponding to the integrated amount of blown blown oxygen is predicted, and an output value capable of presenting a plurality of reliability indexes for the predicted value predicted. A prediction device can be provided.

  An output value prediction apparatus, an output value prediction method, and an output value prediction program according to the present invention obtain a reliability index representing the reliability of a predicted value from a plurality of viewpoints, and present the plurality of calculated reliability indices. can do.

It is a block diagram which shows the structure of the output value prediction apparatus in 1st Embodiment. It is a flowchart which shows operation | movement of the output value prediction apparatus in 1st Embodiment. It is a figure which shows the data memorize | stored in the actual measurement data storage part in the output value prediction apparatus of 1st Embodiment. It is a figure for demonstrating the Euclidean distance of prediction object data and each past performance data. It is a figure which shows the data memorize | stored in the intermediate data memory | storage part in the output value prediction apparatus of 1st Embodiment. It is a figure which shows the data memorize | stored in the predicted value memory | storage part in the output value prediction apparatus of 1st Embodiment. It is a figure for demonstrating the calculation procedure of the dispersion | variation in a predicted value in the output value prediction apparatus of 1st Embodiment. It is a figure for demonstrating the method of calculating | requiring the probability density distribution curve shown in FIG.7 (C) from the histogram shown in FIG.7 (B). It is a figure which shows the relationship between the temperature fall amount of an object, and elapsed time. It is a figure for demonstrating the calculation method of the error parameter of the prediction time in 2nd Embodiment. It is a figure which shows the data memorize | stored in the actual measurement data memory | storage part in the output value prediction apparatus of 2nd Embodiment. It is a figure which shows the data memorize | stored in the intermediate data storage part in the output value prediction apparatus of 2nd Embodiment. It is a figure for demonstrating the calculation procedure of the dispersion | variation in the predicted value in the output value prediction apparatus of 2nd Embodiment. It is a figure which shows the data memorize | stored in the predicted value memory | storage part in the output value prediction apparatus of 3rd Embodiment. It is a figure which shows the probability density distribution of each prediction value in the output value prediction apparatus of 3rd Embodiment. It is a block diagram which shows the structure of an output value prediction system. It is a figure which shows one specific example of the reliability parameter | index which performed scaling. It is a figure which shows another specific example of the reliability parameter | index which performed scaling. It is a figure for demonstrating the other method of calculating | requiring a probability density distribution curve from the histogram shown to FIG. 7 (B).

Hereinafter, an embodiment according to the present invention will be described with reference to the drawings. In addition, the structure which attached | subjected the same code | symbol in each figure shows that it is the same structure, The description is abbreviate | omitted suitably. Further, in this specification, when referring generically, it is indicated by a reference symbol without a suffix, and when referring to an individual configuration, it is indicated by a reference symbol with a suffix.
(First embodiment)
FIG. 1 is a block diagram illustrating a configuration of an output value prediction apparatus according to the first embodiment. In FIG. 1, the output value prediction apparatus S includes an arithmetic control unit 1, an input unit 2, a presentation unit 3, and a storage unit 4.

  The input unit 2 outputs various commands such as a command for starting an output value prediction program for predicting an output value by using the method of the present invention from previously given data, and various data necessary for predicting the output value. A device that inputs to the prediction device S, such as a keyboard or a mouse. The presenting unit 3 is a device that presents (outputs) the command and data input from the input unit 2 and the output value (predicted value) predicted by the output value predicting apparatus S. For example, the presenting unit 3 includes a CRT display, LCD, A display device such as an organic EL display and a plasma display, and a printing device such as a printer.

  The storage unit 4 is functionally measured data storage that stores a plurality of past performance data acquired in the past and prediction target data for which an output value is to be predicted, which is functionally composed of a predetermined output and a quantifiable factor related to the output. Unit 41, intermediate data storage unit 42 for storing intermediate data generated in the process of predicting an output value based on past performance data from prediction target data, and prediction based on past performance data from prediction target data A prediction value storage unit 43 that stores (calculated) output values (prediction values), and a variation storage unit 44 that stores variation in prediction values, and various programs such as an output value prediction program; It is a device that stores various data such as data necessary for execution and data generated during the execution. The storage unit 4 is, for example, a volatile storage element such as a RAM (Random Access Memory) serving as a so-called working memory of the arithmetic control unit 1, a ROM (Read Only Memory), or a rewritable EEPROM (Electrically Erasable Programmable Read Only Memory). And the like, and a hard disk for storing various programs and various data.

  The arithmetic control unit 1 includes, for example, a microprocessor and its peripheral circuits, and functionally includes a distance calculation unit 11, a similarity calculation unit 12, a parameter calculation unit 13, and a predicted value calculation unit 14. The variation calculation unit 15 and the reliability index calculation unit 16 are provided, and the input unit 2, the presentation unit 3, and the storage unit 4 are controlled in accordance with the function according to the control program.

  The distance calculation unit 11 calculates a predetermined distance between the prediction target data and the past performance data for each of the plurality of past performance data based on the factors of the prediction target data and the past performance data.

  The similarity calculation unit 12 calculates the similarity between data between the prediction target data and the past performance data for each of the plurality of past performance data based on the plurality of distances calculated by the distance calculation unit 11. Is to be calculated.

  The parameter calculation unit 13 uses the input variable as the output variable y and the input variable when the predetermined output is used as an output variable and some or all of the quantifiable factors related to the predetermined output are used as input variables. A first model representing the first relationship with Z; when y = f (Z, Θ) is generated, it corresponds to the value obtained by giving the input value of the input variable to the first model and the input value of the input variable An error parameter α based on a difference from an output value of the output variable to be calculated is calculated for each of a plurality of past performance data based on the input variable and the output variable of the past performance data. In the present embodiment, the error parameter α is the difference between the value obtained by giving the input value of the input variable to the first model and the output value of the output variable corresponding to the input value of the input variable.

  The predicted value calculation unit 14 generates a second model representing the second relationship between the output variable y and the input variable Z using the input variable Z and the error parameter α; y = f (Z, Θ, α), and performs prediction The value of the factor corresponding to the input variable of the factors of the target data and the value of the error parameter α are given to the second model, and the output value of the prediction target data is used as the predicted value, and a plurality of parameters calculated by the parameter calculation unit 13 It is calculated for each error parameter α.

  The variation calculating unit 15 calculates the variation in the output value of the prediction target data based on the plurality of similarities calculated by the similarity calculating unit 12 and the plurality of predicted values calculated by the predicted value calculating unit 14. is there.

The reliability index calculation unit 16 obtains a plurality of reliability indexes for the prediction value calculated by the prediction value calculation unit 14 from a plurality of viewpoints. The reliability index is an index representing the degree of trust. The reliability index calculation unit 16 functionally includes a conditional reliability index calculation unit 161 that obtains a conditional reliability index based on a predetermined conditional similarity as one of the plurality of reliability indexes, As one of the plurality of reliability indexes, a daily reliability index calculation unit 162 that calculates a daily reliability index based on a predetermined daily similarity is configured. The predetermined conditional similarity is conditional data that is data of a predetermined data item that affects an output value among data items in past performance data, and given data that is given in advance to obtain a predicted value. For example, the similarity between the conditional data and the given data. The predetermined number of days of similarity is obtained by acquiring the past data used to obtain the second model (prediction model) generated by the predicted value calculation unit 14 and the given data (prediction target data). X ₀ ) is an index indicating the degree of similarity to the prediction date, and for example, acquisition of past performance data used when obtaining the second model (prediction model) generated by the prediction value calculation unit 14 The degree of similarity between the date and the predicted date in the given data (predicted data X ₀ ).

  The arithmetic control unit 1, input unit 2, presentation unit 3, and storage unit 4 are connected by a bus 5 so that signals can be exchanged with each other.

  The arithmetic control unit 1, the input unit 2, the presentation unit 3, the storage unit 4, and the bus 5 can be configured by, for example, a computer, more specifically, a personal computer such as a notebook type or a desktop type.

  Note that the output value prediction apparatus S may further include an external storage unit as necessary. The external storage unit is, for example, a flexible disk, a CD-ROM (Compact Disc Read Only Memory), a CD-R (Compact Disc Recordable), a DVD-R (Digital Versatile Disc Recordable), and a Blu-ray Disc (registered trademark). For example, a flexible disk drive, a CD-ROM drive, a CD-R drive, a DVD-R drive, a Blu-ray disk drive, and the like are devices that read data from and / or write data to / from a recording medium. A and / or B means at least one of A and B.

  Here, when the output value prediction program or the like is not stored, the output value prediction program is installed in the storage unit 4 via the external storage unit from the recording medium on which the output value prediction program or the like is recorded. The output value prediction device S may be configured. Alternatively, the output value prediction device S may be configured such that data such as past performance data and data for predicting an output value is recorded on a recording medium via an external storage unit.

Next, the operation of the output value prediction apparatus S of this embodiment will be described. FIG. 2 is a flowchart showing the operation of the output value prediction apparatus in the first embodiment. FIG. 3 is a diagram illustrating data stored in the actual measurement data storage unit in the output value prediction apparatus according to the first embodiment. FIG. 4 is a diagram for explaining the Euclidean distance between the prediction target data and each past performance data. FIG. 5 is a diagram illustrating data stored in the intermediate data storage unit in the output value prediction apparatus of the first embodiment. FIG. 6 is a diagram illustrating data stored in a predicted value storage unit in the output value predicting apparatus according to the first embodiment. FIG. 7 is a diagram for explaining a procedure for calculating a variation in predicted values in the output value predicting apparatus according to the first embodiment. FIG. 7A shows the relationship between the similarity w and the output predicted value y ₀ , the horizontal axis is the similarity w, and the vertical axis is the predicted value y ₀ . FIG. 7B shows a histogram of the predicted value y ₀ , the horizontal axis is the weighted frequency Fw, and the vertical axis is the predicted value y ₀ . FIG. 7C shows the probability density distribution of the predicted value y ₀ , the horizontal axis is the probability density P (y ₀ ), and the vertical axis is the predicted value y ₀ .

  For example, when an output command is received from the input unit 2 by a user operation, the output value prediction device S executes an output value prediction program. By executing the output prediction program, the distance calculation unit 11, the similarity calculation unit 12, the parameter calculation unit 13, the predicted value calculation unit 14, the variation calculation unit 15, and the reliability index calculation unit 16 are functionally added to the calculation control unit 1. Composed. And the output value prediction apparatus S estimates an output value (predicted value) from prediction object data based on past performance data by the following operations.

In the prediction of the output value, the actual measurement data storage unit 41 in the storage unit 4 of the output value prediction apparatus S stores, for example, past performance data (X, y) and prediction target data in the table format (table format) shown in FIG. _x0n is stored in advance. Here, the predetermined output y is a set of M elements (output elements) y _j , that is, y = {y _j | j = 1, 2,..., M}, The factor X related to the output y is a set of N elements (factor elements) x _ji , that is, X = {x _ji | i = 1, 2,..., N}.

The actual measurement data table 51 shown in FIG. 3 includes an output field 511 for registering the actually measured output value y _j and each field of the data field 512 for registering the data x _ji of the factor X related to the predetermined output y. is configured with, includes a record for each historical track record data (X, y), further comprises a record of the predicted target data X _0. Incidentally, the predicted target data X _0, in order to distinguishably distinguished historical performance data (X, y) and, 0 is assigned as the first subscript (left subscript), historical performance data (X, y ), 1 to M are assigned as first subscripts to distinguish the M pieces of data in an identifiable manner. Then, the predicted target data X ₀ and historical performance data (X, y), said predetermined first or for the second N data items distinguishably distinguished from each other in the factor X associated with the output y is the N elements 1 to N are added as second subscripts (the right side of the subscripts). X ₀ = [x ₀₁ , x ₀₂ ,..., X _0N ], and X _ji = [x _j1 , x _j2 ,..., X _jN ]. For example, y ₃ represents the third output value in the past performance data (X, y), and for example, x ₂₃ represents the second third data item in the past performance data (X, y). represents the value, also for example, x ₀₄ represents the value of the fourth data item in the prediction target data X _0.

As described above, the factor X related to the predetermined output y is configured to include a plurality of N elements (data items, factor elements). For this reason, the data field 512 includes data corresponding to the number N of elements. It has an item subfield. In the example shown in FIG. 3, the predetermined output y involves at least N elements (first to Nth data items). Therefore, the data item subfield has a first or data item subfield 5121~512N for registering each respective data _x j1 _{~x jN} of the N data items correspond to each element _{x ji} factor X . The past performance data (X, y) is data obtained by actual measurement or the like at different times (time points) t _j in the past under different conditions in the past. In the example shown in FIG. It consists of data. The first to Nth data items include data items on the acquisition date when the data is acquired. In the second embodiment to be described later, each of the first to Nth data items includes each data item of an acquisition date and an acquisition time when the data is acquired. Predicted target data _{X 0,} the a predetermined data _{x 0i} for which you want to predict the value _{y 0} in the output y, for example, and the measured data _{x 0i} until prediction time _{t 0,} Ya value _{x 0i} operation input The operation date / time _x0i , the data _x0i prepared for the simulation, and the like.

Here, the output value prediction apparatus S uses the entire Z or a part Z of the data values x _0i of the prediction target data X ₀ and outputs the output value (predicted value) y ₀ based on the past performance data (X, y). It predicted, and requests a variation of the predicted value y _0.

In the case where such a historical record data (X, y) and the predicted target data X ₀ is stored in the measured data storage unit 41, an output value prediction device S, as shown in FIG. 2, initially, historical performance data In order to evaluate the relationship between (X, y) and the prediction target data X ₀ , the distance between the two data is calculated by the distance calculation unit 11 of the calculation control unit 1, and the calculated distance is stored in the middle of the storage unit 4. The data is stored in the data storage unit 42 (S11). The distance is defined so as to represent the relationship between both data, and for example, Euclidean distance, weighted Euclidean distance, normalized Euclidean distance, or the like is used. The data x used to calculate this distance is all or part x of the factor X related to the predetermined output y, and all or part of the factor X related to the predetermined output y. It may coincide with a certain Z, may not coincide, or may partially coincide (partly disagree).

More specifically, in the present embodiment, the distance calculation unit 11 calculates the Euclidean distance d between the prediction target data X ₀ (♦) and the past performance data X (◯) in the data item space, as shown in FIG. It calculates about each past performance data (X, y). For example, in the present embodiment, the data x used to obtain the Euclidean distance d is all the factors X related to the predetermined output y, and the data item space has N data items. It becomes an N-dimensional space. In the present embodiment, the Euclidean distance d is a weighted distance, and is obtained by, for example, Expression 1-1. In Formula 1-1, the weighted Euclidean distance d _j between the prediction target data X ₀ and the j-th past performance data X is the j-th past performance data in the data x used to obtain the Euclidean distance d. The square of the difference between the i-th data item x _ji in (X, y) and the i-th data item x _0i in the prediction target data X ₀ is the weight a _i of the i-th data item (weights a _i and a _i related to distance _). A value obtained by multiplying the square of ≧ 0) is calculated by taking the sum from the first data item to the Nth data item and obtaining the square root of the result.

Further, for example, the Euclidean distance d may be obtained by Expression 1-2. In Expression 1-2, the weighted Euclidean distance d _j between the prediction target data X ₀ and the j-th past performance data X is the j-th past performance data in the data x used to obtain the Euclidean distance d. The difference between the i-th data item x _ji in (X, y) and the i-th data item x _0i in the prediction target data X ₀ multiplied by the weight a _i of the i-th data item is used as the absolute value of the first data item. To the Nth data item.

Further, for example, the Euclidean distance d may be obtained by Expression 1-3. In Formula 1-3, the weighted Euclidean distance d _j between the prediction target data X ₀ and the j-th past performance data X is the j-th past performance data in the data x used to obtain the Euclidean distance d. The difference between the i-th data item x _ji in (X, y) and the i-th data item x _0i in the prediction target data X ₀ is multiplied by the weight a _i of the i-th data item. It is calculated by selecting the maximum value from the set up to.

ここで、重みａ_ｉは、ユークリッド距離ｄを求めるに当たって、第１ないし第Ｎデータ項目の中で第ｉデータ項目の重要度（重要性の度合い）を表すパラメータであり、要因要素（第ｉデータ項目ｘ_ｉ）が所定の出力ｙにおけるばらつきの大きさに寄与する程度を表すものである。例えば、要因要素（第ｉデータ項目ｘ_ｉ）がそれぞれ標準化もしくは正規化されている場合には、この重みａ_ｉは、所定の出力ｙに影響を与える程度が大きいデータ項目ほど大きくなり、所定の出力ｙに影響を与える程度が小さいデータ項目ほど小さくなるように、設定される。この重みａ_ｉは、公知の手法を用いて決定することができ、例えば、特許第３９４３８４１号明細書に開示されているように重回帰分析によって求めることができ、また例えば、特許第３９１２２１５号明細書に開示されているように、各データ項目をその統計値（例えば平均値や標準偏差）によって正規化しておくことによって求めることができる。

Here, the weight a _i is a parameter representing the importance (degree of importance) of the i-th data item among the first to N-th data items in obtaining the Euclidean distance d, and is a factor element (i. This represents the degree to which the item x _i ) contributes to the magnitude of variation in the predetermined output y. For example, when the factor element (i-th data item x _i ) is standardized or normalized, the weight a _i increases as the data item that has a large degree of influence on the predetermined output y increases. It is set so that the smaller the data item that affects the output y, the smaller the data item. This weight a _i can be determined by using a known method, for example, can be obtained by multiple regression analysis as disclosed in Japanese Patent No. 393441, and for example, Japanese Patent No. 3912215. As disclosed in the document, each data item can be obtained by normalizing it with its statistical value (for example, average value or standard deviation).

Then, the distance calculation unit 11 stores the distances d _j which is the calculated intermediate data storage unit 42 of the storage unit 4.

Next, in order to evaluate how much the output value prediction device S is similar to the prediction target data X ₀ , the output value prediction device S uses the first to M-th past results as the similarity (degree of similarity) w between the two data x. Each of the data (X, y) is calculated by the similarity calculation unit 12 of the calculation control unit 1, and each calculated similarity w is stored in the intermediate data storage unit 42 of the storage unit 4 (S12). The similarity w _j is defined, for example, such that the smaller the weighted Euclidean distance d _j is, the larger the similarity is and a positive value is taken.

More specifically, the similarity calculation unit 12 calculates, for example, the similarity w _j using Equation 2-1. Further, for example, the similarity calculation unit 12 calculates the similarity w _j by using the expression 2-2. Further, for example, the similarity calculation unit 12 calculates the similarity w _j by using the expression 2-3, for example.

ここで、類似度ｗ_ｊは、予測対象データＸ_０に対する第ｊ番目の過去実績データＸの類似度であり、ｃ_１は、０以上の実数の調整パラメータであり、ｃ_２は、実数（負の値でもよい）の調整パラメータであり、ｇ、ｒは、正の実数の調整パラメータである。

Here, the similarity w _j is the similarity of the j-th past performance data X to the prediction target data X ₀ , c ₁ is a real number adjustment parameter of 0 or more, and c ₂ is a real number (negative) And g and r are positive real adjustment parameters.

Then, the similarity calculation unit 12 stores each calculated similarity w _j in the intermediate data storage unit 42 of the storage unit 4.

In addition, when the upper limit value of the similarity score w and / or the lower limit value thereof is provided and the similarity score w _j calculated by any one of the formulas 2-1 to 2-3 exceeds the upper limit value, , When the similarity w _j is replaced with the upper limit and / or the similarity w _j calculated by any one of the formulas 2-1 to 2-3 exceeds the lower limit, The similarity calculation unit 12 may be configured such that the similarity w _j is replaced with the lower limit value. By being configured in this way, it is possible to prevent only the specific past performance data X from excessively increasing in similarity or conversely decreasing. If the degree of similarity of only specific past performance data X becomes excessive, if there is an error in the data measurement value, it will be pulled by the error and the wrong variation will be predicted. Will end up. For this reason, as described above, setting the upper limit value has an effect of becoming stronger in error.

Further, for example, when a predetermined threshold is provided in advance and the similarity w _j calculated by any one of the expressions 2-1 to 2-3 is equal to or less than the threshold, the similarity w _j is 0. The similarity calculation unit 12 may be configured to be replaced with Alternatively, the similarities w up to the predetermined number (or predetermined ratio) set in advance from the smallest are arranged in order of increasing similarity w _j calculated by any one of Expressions 2-1 to 2-3. _The similarity calculation unit 12 may be configured such that _j is replaced with 0. By being constructed as described above, when obtaining the prediction values y _0, it is possible to prevent to consider more than necessary historical performance data X which is not much similar to the predicted target data X _0. Also, it excluded historical performance data X which is not much similar to the predicted target data X ₀ is, the arithmetic processing is not required to be described below, as a result, the arithmetic processing amount reduction (shortening of processing time) becomes possible.

Further, in the case of calculating the similarity w _j, the predicted target data X ₀ and temporally past as actual data X near causes increase its similarity w _j (raised to) evaluation items in the calculation of the similarity w _j It is preferable to enter. For example, one or more evaluation items for increasing the similarity w _j may be included in the data x used for obtaining the Euclidean distance d, and, for example, the equation 2 for calculating the similarity w -1 to Formula 2-3. More specifically, the correction term exp (− (wday × Δday _j ) ² ) is multiplied by the expressions 2-1 to 2-3. Here, wday is the weight of the adjustment parameter, △ day _j is an operational days difference between the j historical performance data _{X j} obtained at time _{t j} and the predicted target data _{X 0.} By configuring in this way, for example, when the characteristics of a system (for example, a manufacturing plant or a manufacturing process) in which variations in output values are predicted are changed due to a change in equipment or operation or a change over time, it is robust. Prediction can be performed.

  Next, the output value prediction device S calculates an error parameter α representing an uncertain element by the parameter calculation unit 13 of the arithmetic control unit 1 for each of the first to M-th past performance data (X, y), The calculated error parameters α are stored in the intermediate data storage unit 42 of the storage unit 4 (S13).

  More specifically, the parameter calculation unit 13 obtains a prediction model (second model) for predicting the output value y based on the M past performance data (X, y), and uses the obtained model. Each error parameter α is obtained for each of the first to Mth past performance data (X, y).

  This prediction model is expressed by, for example, the function f in Expression 3. In this case, the parameter calculation unit 13 obtains the coefficient Θ of the function equation 3 based on the M past performance data (X, y), and uses the obtained function equation 3 to obtain the first to M-th past Each error parameter α is obtained for each result data (X, y).

ここで、Ｚは、例えば操業条件（製造条件）の各条件や製造工程の各工程における各測定項目等の、前記所定の出力ｙに関わる数値化可能な要因Ｘのうちの出力値ｙ_０を予測するために用いられる要因であり、複数Ｌ個の要素ｚを備えて構成される（Ｚ_ｊ＝［ｚ_ｊ１、ｚ_ｊ２、・・・、ｚ_ｊＬ］）。Ｚは、例えば、前記所定の出力ｙに関与する要因Ｘにおける複数Ｎの要因要素ｘ_ｊｉであるデータ項目（Ｘ_ｊ＝［ｘ_ｊ１、ｘ_ｊ２、・・・、ｘ_ｊＮ］）の一部または全部によって構成される。なお、このＺは、さらに、前記データ項目Ｘ以外の要素を含んでいてもよい。また、上述したように、この出力値ｙ_０を予測するために用いられる要因Ｚは、ユークリッド距離ｄを求めるために用いられるデータｘと、一致していてもよく、また不一致であってもよく、また一部一致（一部不一致）であってもよい。このＺは、関数ｆの式３を決定する際に予め設定される。Θは、関数式３の係数等の所定の調整パラメータであり、Ｍ個の過去実績データ（Ｘ、ｙ）に基づいて同定計算によって求められる。この同定には、最小二乗法、最尤推定法、部分最小二乗法、二次計画法およびＰＳＯ（Particle Swarm Optimization）等の、前記所定の出力ｙの実績値ｙ_ｊとその予測値ｙ_０との誤差が所定の評価関数の下（所定の制約条件の範囲内）で最小（または最大）となるように決定する方法が用いられる。あるいは、Θは、所定の物理法則を用いることによって求められてもよい。α_ｊは、不確定要素を表す誤差パラメータであり、ΘおよびＺだけでは出力ｙを表現しきれない要因（ばらつきの要因）を表すものであり、ΘおよびＺを用いて出力ｙを予測した場合における予測値ｙ_０と実績値ｙ_ｊとの誤差に相当する。

Here, Z is an output value y ₀ of a factor X that can be quantified related to the predetermined output y, such as each condition of operation conditions (manufacturing conditions) and each measurement item in each process of the manufacturing process. This is a factor used for prediction, and is configured to include a plurality of L elements z (Z _j = [z _j1 , z _j2 ,..., Z _jL ]). Z is, for example, a part of a data item (X _j = [x _j1 , x _j2 ,..., X _jN ]) that is a plurality of N factor elements x _ji in the factor X involved in the predetermined output y. Consists of everything. The Z may further include elements other than the data item X. As described above, factors Z used to predict the output value y ₀ is a data x that is used to determine the Euclidean distance d, may be coincident, also be a mismatch Also, it may be partially coincident (partly inconsistent). This Z is set in advance when the expression 3 of the function f is determined. Θ is a predetermined adjustment parameter such as a coefficient of the function formula 3, and is obtained by identification calculation based on M pieces of past performance data (X, y). For this identification, the actual value y _j of the predetermined output y and its predicted value y ₀ such as least square method, maximum likelihood estimation method, partial least square method, quadratic programming method and PSO (Particle Swarm Optimization) Is used such that the error is minimized (or maximum) under a predetermined evaluation function (within a predetermined constraint). Alternatively, Θ may be determined by using a predetermined physical law. α _j is an error parameter that represents an uncertain element, and represents a factor (variation factor) in which the output y cannot be expressed only by Θ and Z. When the output y is predicted using Θ and Z corresponding to the error between the predicted value y ₀ and actual y _j in.

When the predicted value y ₀ is predicted by a multiple regression equation, for example, Equation 4-1 can be used as the function equation 3, and the error parameter α in the _jth past performance data (X _j , y _j ) is used. _j is given by Equation 4-2. The model expressed by Expression 4-1 is effective when there are additive elements (factors of variation) additively.

Further, for example, when the predicted value y ₀ is predicted by a multiple regression equation, the equation 3 can be used as the function equation 3, for example, in the jth past performance data (X _j , y _j ). The error parameter α _j is given by equation 5-2. The model expressed by Equation 5-1 is effective when the uncertain element exists in a multiplicative manner.

Further, for example, when the predicted value y ₀ is predicted by a multiple regression equation, for example, Equation 6-1 can be used as the function equation 3, and the j-th past performance data (X _j , y _j ) The error parameter α _j is given by equation 6-2. The model expressed by Expression 6-1 is effective when an uncertain element exists in the influence coefficient of z _j1 .

Further, for example, when predicting the predicted value y ₀ by a multiple regression equation, for example, Equation 7-1 can be used as the function equation 3, and the j-th past performance data (X _j , y _j ) The error parameter α _j is given by Expression 7-2 (Expression 7-2-1 and Expression 7-2-2).

In the above description, a mathematical expression representing the function f is used. However, an arithmetic program including a table representing the function f, a convergence calculation algorithm, an if-then rule, a fuzzy reasoning, a neural network, a JIT (Just in Time) model, and the like. May be used. Here, when the error parameter α _j cannot be obtained from Z _j , Θ, and y _j by back calculation, the value of the error parameter α _j is obtained by, for example, a binary search method, a carpet bombing method, or PSO (Particle Swarm Method). And an error parameter α _j such that the output value matches y _j .

  Then, the parameter calculation unit 13 stores the calculated error parameters α in the intermediate data storage unit 42 of the storage unit 4.

The weighted Euclidean distance d _j , similarity w _j and error parameter α _j calculated by each of the processes S11 to S13 are, for example, the intermediate data storage unit 42 in a table format (table format) as shown in FIG. Is remembered. The intermediate data table 52 shown in FIG. 5 includes an output field 521 for registering the actually measured output value y _j, and a similarity calculation data field 522 for registering data x of the data item used for calculating the similarity w _j. , the predicted value output calculation data field 523 for registering the data z data items used for calculation of y _0, the weighted distance field 524 for registering a weighted Euclidean distance d _j of the historical performance data X, the historical record The first to Mth past achievements are configured to include a similarity field 525 for registering the similarity w _j of the data X and an error parameter field 526 for registering the error parameter α _j of the past achievement data X. data (X, y) includes a record for each further comprises a record of the predicted target data X ₀ The similarity calculation data field 522 includes data item subfields 5221 to 522N corresponding to each data item x used to calculate the similarity w _j . Similarly, the output calculation data field 523, a data item subfield 5231~523L corresponding to each data item z, which is used to calculate the predicted value _{y 0.}

Then, the output value prediction device S, by the predictive value calculating unit 14 of the arithmetic and control unit 1, using the prediction model obtained in process S13, the first to the data value of the N data items in the prediction target data X ₀ Based on x _{01 to} x _0N , the predicted value y ₀ is calculated for each error parameter α _j obtained in step S 13, and the calculated predicted value y ₀ is stored in the predicted value storage unit 43 of the storage unit 4. (S14). Here, in this process S14, the prediction model is changed to each of the error parameters alpha _j the error parameter alpha _j is determined in process S13. For example, in the case where the prediction model is expressed by Equation 3 of the function f described above, the coefficient f calculated in the process S13 and the function f changed to each of the error parameters α _j obtained in the process S13. In Equation 3, each data value x _{01 to} x _0L of the first to Lth data items used for calculating the predicted value y ₀ among the first to Nth data items in the prediction target data X ₀ is used as Z. Accordingly, the predicted value calculation unit 14 calculates the predicted value y ₀ for each error parameter α _j obtained in step S13. The predicted value _{y 0,} each error parameter alpha _j is because it is the M, the M number of predicted value _y 01 _{~y 0M.}

  In the process S14, as in the process S13, an arithmetic program including a table representing the function f, a convergence calculation algorithm, an if-then rule, a fuzzy inference, a neural network, a JIT model, and the like may be used.

Each such prediction value y ₀₁ ~y _0M calculated by step S14 is stored in, for example, the predicted value storing unit 43 by the table format table format as shown in FIG. The predicted value data table 53 shown in FIG. 6 includes a predicted value field 531 for registering the predicted value _y0j calculated in step S14, and each data of the first to Lth data items used for calculating the predicted value _y0j. The output prediction data field 532 for registering the values x _{01 to} x _0L , the error parameter field 533 for registering the error parameter α calculated in step S13, and the past results used for calculating the error parameter of the parameter field 533 Each field includes a similarity field 534 for registering the similarity w _j in the data (X, y), and a record is provided for each of the first to Mth error parameters α _j . Here, each prediction value y ₀₁ ~y _0M calculated by the step S14, as shown in FIG. 6, also the similarity w _j corresponding to the predicted value y _0, the predicted value storing section to correspond to each other 43.

Then, the output value prediction device S, the variation calculation unit 15 of the arithmetic and control unit 1, with each of the prediction value _y 01 _{~y 0M} obtained in processing S14, calculates the variance of the predicted value _{y 0j,} the calculated The variation of the predicted value _y0j is stored in the variation storage unit 44 of the storage unit 4 (S15).

In the present embodiment, based on the rule of thumb that results in similar in similar conditions, the higher the similarity between the data z ₀ of the prediction target data X ₀ and data x _j of the j-th historical performance data X ( If the degree of similarity w _j is large), the error parameter α ₀ of the prediction target data X ₀ is considered to have high similarity. For this reason, the predicted value y ₀ of the prediction target data X ₀ is considered to be the predicted value y _0j predicted using the error parameter α _j with the similarity w _j .

Based on this concept, and more specifically, the variation calculation unit 15, as shown in FIG. 7 (A), taking the similarity w on the horizontal axis with taking the prediction value y ₀ on the vertical axis, First, for each of M prediction values y _0j corresponding to each of the M error parameters α _j (j = 1 to M) calculated from the M past performance data (X, y), The similarity w _{j is made} to correspond. Next, the variation calculation unit 15 _divides a range y _0j including at least each predicted value on the vertical axis y ₀ in FIG. _7A into a finite number of sections (classes, grades), and is included in each section. A weighted frequency F _w is generated by adding all the similarities w _j of the predicted value y _0j , and a histogram representing the variation of the predicted value y ₀ is generated as shown in FIG.

That is, as shown in Equation 8, if the set of j the predicted value _{y 0j} is included in the k-th interval _{(Y k} or _{Y k + 1} less than the section) and _{S k (S k = {j} | Y k ≦ y _0j <Y _{k + 1} }) and the sum of the similarities w _j for j included in the set S _k is the weighted frequency F _w in the k-th interval.

This variation of the predicted value y ₀ as is indicated by the histogram, it is possible to easily know the frequency of occurrence of the predicted value y _0.

As described above, the histogram illustrated in FIG. 7B may be the variation of the predicted value y ₀ , but in the present embodiment, the variation calculation unit 15 further has an area of the histogram illustrated in FIG. Normalize so that The normalized histogram is a probability density distribution of the predicted value y ₀ in the prediction target data X ₀ (empirical probability density distribution), are variations of the predicted value y _0. Furthermore, the variation calculation unit 15 may represent the histogram shown in FIG. 7B with a curve as shown in FIG. 7C while maintaining the area at 1. This curve is a probability density distribution of the predicted value y ₀ in the prediction target data X _0, are variations of the predicted value y _0.

Note that the normalization, for example, when dividing the longitudinal axis y ₀ shown in FIG. 7 (A) into a finite number of intervals, uniform width h = | when a is divided into _| Y k + _{1 -Y} k , According to Equation 9.

  Further, when obtaining the curve shown in FIG. 7C from the histogram shown in FIG. 7B, a known probability density distribution such as a lognormal distribution or a Weibull distribution may be used.

  FIG. 8 is a diagram for explaining a method for obtaining the probability density distribution curve shown in FIG. 7C from the histogram shown in FIG. 7B. 8A shows a state in which the center points of the histogram are connected by a broken line, FIG. 8B shows the cumulative probability density distribution of FIG. 8A, and FIG. 8C shows FIG. FIG. 8B shows a state where the cumulative probability density distribution shown in (B) is smoothed, and FIG. 8D shows a smoothed probability density distribution (probability density distribution curve).

First, as shown in FIG. 8 (A), in the histogram normalized shown in FIG. 7 (B), connecting the center positions of the respective frequencies of (y ₀ direction of the center) in a line. At both ends, a point separated from the end by a half (h / 2) of the section width h is also set to 0 and connected to the broken line. The area surrounded by the broken line is also 1.

Next, as shown in FIG. 8B, a cumulative probability density distribution _SwN is obtained from FIG.

Next, as shown in FIG. 8C, the cumulative probability density distribution _SwN of the broken line in FIG. 8B is smoothed by using, for example, Expression 10-2.

  Then, as shown in FIG. 8D, the smoothed probability density distribution (probability density distribution curve) is obtained by using, for example, Equation 10-3 from the smoothed cumulative probability density distribution shown in FIG. ) Is required.

Thus, the variation of the predicted value y ₀ is indicated by the probability density distribution, and the appearance probability of the predicted value y ₀ can be easily known.

Further, when calculating the weighted frequency F _w , the similarity w _{j of} M pieces of past performance data (X, y) is arranged in descending order (in descending order), and preset from the larger one. The past performance data (X, y) up to a predetermined number (predetermined ratio) may be extracted, and the weighted frequency F _w may be obtained by using only the extracted past performance data. By removing the past performance data (X, y) having a low similarity w _j in advance, it is possible to reduce the amount of computation processing for computing the weighted frequency F _w (shortening the computation processing time). In the case where the similarity w _j is calculated by the above-described Expression 2 (Expression 2-1 to Expression 2-3), the past performance data (X, y) having a low similarity w _j with the prediction target data X ₀ is calculated. However, since the similarity w _j does not become 0, the weighted frequency F _w is affected. For this reason, the width of the weighted frequency F _w shown in FIG. 7B matches the width of the M predicted values y _0j , and when the function f is Equation 4, the width is the prediction target data always constant irrespective of the conditions of X _0. As a result, the base of the probability density distribution shown in FIG. 7C may spread more than necessary. However, as described above, by excluding the past performance data (X, y) having a small similarity w _j , the base of the probability density distribution is prevented from being excessively expanded, and the predicted value y in the prediction target data X ₀ is determined. The feature of ₀ distribution shape is remarkably expressed.

Next, the output value predicting apparatus S calculates a plurality of reliability indices R, which are indices indicating the degree of reliability of the predicted value y _0, from a plurality of viewpoints by the reliability index calculating section 16 of the arithmetic control section 1. The calculated reliability index R is stored in the storage unit 4 (S16). For example, as one of the plurality of reliability indexes R, the reliability index calculation unit 16 calculates a conditional reliability index by the conditional reliability index calculation unit 161, and one of the plurality of reliability indexes R The daily reliability index calculation unit 162 calculates the daily reliability index.

More specifically, for example, the conditional reliability index calculation unit 161 calculates the average value of the conditional similarities w as the reliability index R (conditional reliability index _RC ). This is the fact that the average value of the conditional similarity is large, it just means that in the past actual data (X, y), which contains many conditions similar data to the prediction target data X _0, reliability of the predicted value y ₀ obtained on the basis of such historical performance data (X, y) also is considered to high. The conditional similarity w is the similarity between the conditional data and the prediction target data X _0, and the average value (conditional reliability index R _C ) is defined by, for example, Expression 11. The conditional data is data of a predetermined data item that affects the output value y (predicted value y ₀ ) among the data items in the past performance data (X, y). The conditional data is, for example, It may be all or part of the data of the data item used to generate the prediction model, and for example, there are some data items that are difficult to incorporate into the prediction model because they are non-linear. Thus, in this conditional data, data items other than the data items used for generating the prediction model may be added to all or part of the data items used for generating the prediction model.

Further, for example, the conditional reliability index calculation unit 161 may use a function of the average value of the conditional similarity w as the conditional reliability index _RC . The conditional reliability index _RC may be defined by the equation 11, but the value according to the equation 11 is 0 ≦ R _c ≦ 1 when 0 ≦ w _j ≦ 1, but the conditional similarity If the degree w _j has no such range limitation, it is generally 0 or more and there is no upper limit. Although the user can relatively judge the degree of reliability based on the past experience based on the average value of the conditional similarity w, since it is such an empirical judgment, the conditional reliability index R _C is a function of the average value of the conditional similarity w such that 0 ≦ R _C ≦ 1. Function of the mean value of such conditional similarity (conditionally reliability index _{R C (R Ca ~R Cc)} ) may, for example, is defined by any one of formula 12-1 formula 12-3 .

ここで、ｃ_１は、正の実数の調整パラメータであり、ｃ_２は、実数（負の値でもよい）の調整パラメータである。

Here, c ₁ is a positive real number adjustment parameter, and c ₂ is a real number (which may be a negative value) adjustment parameter.

As described above, the reliability index R, here, the conditional reliability index _RC is converted so as to be within a predetermined numerical range, so that it can be appropriately presented to the presentation unit 3.

Further, for example, the daily reliability index calculation unit 162 calculates the average value of the daily similarity w _D as the reliability index R (the daily reliability index R _D ). This means that the average value of the day-to-day similarity w _D is large, and accordingly, the past performance data (X, y) includes data (latest data) that is day-similar to the prediction target data X _0. means that included many, because such a historical record data (X, y) is considered to be high reliability of the predicted value y ₀ obtained based on. The day-to-day similarity w _D is calculated based on the acquisition date δ _j (j = 1 to M) from which the past performance data (X, y) used when obtaining the prediction model and the prediction date δ in the prediction target data y ₀ . _The degree of similarity with ₀ is defined by, for example, Expression 13-2, and the average value (daily reliability index R _D ) is defined by Expression 13-1.

ここで、ａは、正の実数の調整パラメータである。

Here, a is a positive real adjustment parameter.

For example, the daily reliability index calculation unit 162 may use a function of the average value of the daily similarity w _{D as} the daily reliability index R _D. The daily reliability index R _D may be defined by the equation 13-1, but the value according to the equation 13-1 is 0 ≦ R _D ≦ 1 when 0 ≦ w _Dj ≦ 1. In the case where there is no such range limitation on the daily similarity w _Dj , it is generally 0 or more and there is no upper limit. The user can also determine the degree of reliability relative to the past experience based on the magnitude of the average value of the day-to-day similarity w _D , but since this is an empirical judgment, the day-to-day reliability index R _D is a function of the average value of the day-to-day similarity w _D such that 0 ≦ R _D ≦ 1. Such a function of the average value of the daily similarity degree w _D (the daily reliability index R _D (R _{Da to} R _Dc )) is defined by, for example, any one of Expressions 14-1 to 14-3. Is done.

ここで、ｃ_１は、正の実数の調整パラメータであり、ｃ_２は、実数（負の値でもよい）の調整パラメータである。

Here, c ₁ is a positive real number adjustment parameter, and c ₂ is a real number (which may be a negative value) adjustment parameter.

In this way, the reliability index R, here the daily reliability index _RD , is scaled so as to be within a predetermined numerical range, so that it can be appropriately presented to the presentation unit 3. In this way, the daily reliability index calculation unit 162 also functions as an example of a scaling unit that scales the reliability index R so as to be within a predetermined numerical range. The same applies to the following examples.

Further, for example, the conditional reliability index calculation unit 161 (or the daily reliability index calculation unit 162) uses the average value of the products of the conditional similarity w and the daily similarity w _D as the reliability index R ( Calculated as a conditional days reliability index R _CD ). This is the fact that the average value of the product of conditional similarity w and days Similarity w _D is large, only that, in the past actual data (X, y), conditional also on the predicted target data X ₀ also mean to include many similar data to the number of days, the because such historical performance data (X, y) is considered to be high reliability of the predicted value y ₀ obtained based on. The average value (conditional daily reliability index R _CD ) of the product of the conditional similarity w and the daily similarity w _D is defined by Equation 15.

In addition, for example, the conditional reliability index calculation unit 161 (or the daily reliability index calculation unit 162) can be used for Expression 12 (Expression 12-1 to Expression 12-3) with respect to the above Expression 11 and with respect to the above Expression 13-1. Due to the same situation as Expression 14 (Expression 14-1 to Expression 14-3), the function of the average value of the product of the conditional similarity w and the daily similarity w _D is expressed as the conditional daily reliability index R. _It is good also as _CD . Such a conditional number of days reliability index R _CD is defined by, for example, expressing Expression 15 as Expression 12 or Expression 14.

Also, for example, the conditional reliability index calculation unit 161 (or the daily reliability index calculation unit 162) extracts past performance data (X, y) in which the conditional similarity w is within a certain upper level, The average value of the daily similarity w _D with respect to the extracted past performance data (X, y) is calculated as the reliability index R (upper condition daily reliability index R _D1 ). The conditional similarity w within the upper fixed amount is, for example, the conditional similarity w from the top to a predetermined number (for example, the conditional similarity w within the top 100) or the like from the top to a predetermined ratio. Conditional similarity w (for example, conditional similarity w within the top 30%). This is the fact that a large average value, it only conditionally similar historical performance data (X, y) in the predicted target data X ₀ in the, number of days to similar data to the predicted target data X ₀ means that included many, because such a historical record data (X, y) is considered to be high reliability of the predicted value y ₀ obtained based on. Such an upper condition days reliability index R _D1 is defined by, for example, Expression 16.

ここで、Ω_ｃは、条件的類似度ｗが上位一定量以内となる過去実績データ（Ｘ、ｙ）の集合であり、条件的類似度ｗが大きい方から抽出する過去実績データ（Ｘ、ｙ）の個数をＮｃとし、条件的類似度ｗが大きい方からＮｃ個目の条件的類似度の値をωとすると、Ω_ｃは、｛ｊ｜ｗ_ｊ≧ω｝である（Ω_ｃ＝｛ｊ｜ｗ_ｊ≧ω｝）。

Here, Ω _c is a set of past performance data (X, y) in which the conditional similarity w is within the upper fixed amount, and past performance data (X, y) extracted from the one having the higher conditional similarity w ) Is Nc, and the value of the Nc-th conditional similarity from the larger conditional similarity w is ω, Ω _c is {j | w _j ≧ ω} (Ω _c = { j | w _j ≧ ω}).

In addition, for example, the conditional reliability index calculation unit 161 (or the daily reliability index calculation unit 162) can be used for Expression 12 (Expression 12-1 to Expression 12-3) with respect to the above Expression 11 and with respect to the above Expression 13-1. Due to the same situation as Expression 14 (Expression 14-1 to Expression 14-3), the average value of the day-to-day similarity w _D with respect to the past performance data (X, y) in which the conditional similarity w is within the upper fixed amount May be used as the upper condition days reliability index R _D1 . Such a high-order condition days reliability index R _D1 is defined, for example, by expressing Expression 16 as Expression 12 or Expression 14.

Further, for example, the daily reliability index calculation unit 162 (or the conditional reliability index calculation unit 161) extracts the past performance data (X, y) in which the daily similarity w _D is within the upper fixed amount, The average value of the conditional similarity w with respect to the extracted past performance data (X, y) is calculated as the reliability index R (upper day condition conditional reliability index _RC1 ). The number of days Similarity w _D within the upper fixed amount, for example, the number of days similarity degree from an upper to a predetermined number w _{D (e.g.} days Similarity within 100 upper w _D, etc.) and the proportion from the upper predetermined The above-mentioned day-to-day similarity w _D (for example, the day-to-day similarity w _D within the top 30%). This is the fact that a large average value, it only dates manner similar historical performance data (X, y) in the predicted target data X ₀ in the, condition similar data to the prediction target data X ₀ means that included many, because such a historical record data (X, y) is considered to be high reliability of the predicted value y ₀ obtained based on. Such upper days conditional reliability index _RC1 is defined by, for example, Expression 17.

ここで、Ω_Ｄは、日数的類似度ｗ_Ｄが上位一定量以内となる過去実績データ（Ｘ、ｙ）の集合であり、日数的類似度ｗ_Ｄが大きい方から抽出する過去実績データ（Ｘ、ｙ）の個数をＮ_Ｄとし、日数的類似度ｗ_Ｄが大きい方からＮ_Ｄ個目の条件的類似度の値をωとすると、Ω_Ｄは、｛ｊ｜ｗ_Ｄｊ≧ω｝である（Ω_Ｄ＝｛ｊ｜ｗ_Ｄｊ≧ω｝）。

Here, Ω _D is a set of past performance data that the number of days Similarity w _D is equal to or less than the upper fixed amount (X, y), past performance data to be extracted from a larger number of days Similarity w _D (X , the number of y), and N _D, if the value of N _D th conditional similarity and omega from the side days similarity w _D is large, omega _D is, | is {j w _Dj ≧ ω} (Ω _D = {j | w _Dj ≧ ω}).

Further, for example, the daily reliability index calculation unit 162 (or the conditional reliability index calculation unit 161) performs the expression 12 (expression 12-1 to expression 12-3) with respect to the above expression 11 and the expression 13-1. The average value of the conditional similarity w with respect to past performance data (X, y) in which the daily similarity w _D is within the upper fixed amount due to the same situation as Expression 14 (Expression 14-1 to Expression 14-3) May be used as the upper-day conditional reliability index _RC1 . Such upper days conditional reliability index _RC1 is defined, for example, by expressing Expression 17 as Expression 12 or Expression 14.

Further, for example, the conditional reliability index calculation unit 161 (or the daily reliability index calculation unit 162) is configured such that the conditional similarity w or a product of the conditional similarity w and the day similarity w _D is a predetermined threshold value. The number of past performance data (X, y) as described above is calculated as the reliability index R (reliability index R _N above the threshold). This is the fact that this number is large, it just past actual data (X, y) in the similar data is on conditions similar data or conditional on days specifically to prediction target data X ₀ This is because the reliability of the predicted value y ₀ obtained based on such past performance data (X, y) is considered high. Such threshold above the reliability index R _N is, for example, is defined by equation 18.

ここで、Ｍは、過去実績データ（Ｘ、ｙ）の総数である。

Here, M is the total number of past performance data (X, y).

In addition, for example, the conditional reliability index calculation unit 161 (or the daily reliability index calculation unit 162) can be used for Expression 12 (Expression 12-1 to Expression 12-3) with respect to the above Expression 11 and with respect to the above Expression 13-1. Past results in which the conditional similarity w or the product of the conditional similarity w and the day-to-day similarity w _D is equal to or greater than a predetermined threshold due to the same situation as that of Expression 14 (Expression 14-1 to Expression 14-3) data (X, y) may be a reliability index R _N threshold or a function of the number of. Such a reliability index R _N (R _{Na to} R _Nc ) above the threshold is defined by, for example, Expression 19-1 to Expression 19-3.

ここで、ｃ_１は、正の実数の調整パラメータであり、ｃ_２は、実数（負の値でもよい）の調整パラメータである。

Here, c ₁ is a positive real number adjustment parameter, and c ₂ is a real number (which may be a negative value) adjustment parameter.

The output value prediction device S, the arithmetic control unit 1, the variation of the predicted value y ₀ calculated by the variation calculation unit 15 in the processing S15, and a plurality of calculated by the reliability index calculating unit 16 in the processing S16 The reliability index R is presented to the presentation unit 3 (S17), and the process ends. Here, as seen in FIG. 7 (B) or FIG. 7 (C), the the variation of the predicted value y _0, the predicted value y ₀ is also included in the presentation unit 3, also presented predicted value y ₀ Has been. The arithmetic control unit 1 not only displays the variation of the predicted value y ₀ as described above, but also a predetermined predicted value y _{0 in} the variation of the predicted value y ₀ , for example, the probability density distribution P (y ₀ the largest value of the predicted value y ₀ in) may be presented to the presentation unit 3.

In the present embodiment, when the output value prediction apparatus S of the first embodiment operates as described above, M error parameters α _j (j = 1) calculated from M past performance data (X, y). ~M) by using the predicted prediction value _{y 0j} of the target data _{X 0} is calculated as M, and weighted frequency _{F w} is calculated for the predicted value _{y 0j} according similarity _{w j} from the predicted target data _{X 0} Is done. Further, a probability density distribution P (y ₀ ) is calculated from the weighted frequency F _w . Therefore, regardless of the manner of distribution of the variations of the predicted value y _0, historical performance data (X, y) and the predicted value y ₀ in the prediction target data X ₀ of similarity is considered the predicted target data X ₀ Variation is required with high accuracy. Accordingly, the output value prediction unit S can present variations of the predicted value y _0. Since the variation of the predicted value y ₀ is presented to the presentation unit 3, the user is able to know the variation of the predicted value y _0, the predicted value when performing operations and judgments, etc. on the basis of the predicted value y ₀ variation of y ₀ also becomes possible to consider.

As described above, in the output value predicting apparatus S in the first embodiment, the variation of the predicted value y ₀ is obtained by using the prediction target data X ₀ for the prediction model, and the variation of the predicted value y ₀ . a plurality of the reliability index R sought for, and variations and a plurality of reliability index R of the predicted value y ₀ these calculated was is presented. The output value prediction apparatus S in this embodiment as the relative prediction value y _0, obtains a plurality of reliability index R from a plurality of viewpoints, it is possible to present a plurality of reliability index R which this calculated. For this reason, the user can evaluate the obtained variation of the predicted value y ₀ from a plurality of viewpoints by referring to the plurality of reliability indexes R presented, and the predicted value y ₀ It is possible to reasonably judge the determination of future actions based on these multiple viewpoints.

In the output value prediction apparatus S of the first embodiment, the conditional reliability index _RC and the daily reliability index _RD can be presented as the plurality of reliability indices R. Therefore, the user can determine the condition similarity between the prediction target data X ₀ and historical performance data (X, y) by reference to conditional reliability index R _C, the prediction value y ₀ , And the number of days of similarity between the prediction target data X ₀ and the past performance data (X, y) is determined by referring to the day reliability index R _D. And the variation of the predicted value y ₀ can be evaluated from this point of view. In this way the user, prediction and whether or not the point whether the target data X ₀ is a well there was the case in the past (cases), recently also frequently asked whether or not the point is the case (previously recently were often is Can be interpreted individually (separately). Therefore, the determination of future actions based on the predicted value y ₀ can be rationally determined from these conditions perspective and days standpoint. The determination can be made in consideration of, for example, a risk due to a change in conditions or a risk due to, for example, changes over time in days.

In the present embodiment, the conditionally reliability index R _C, conditionally reliability index R _C of formula 11 or formula 12 described _above, conditional days reliability indicator, such as Equation 15 described above R _CD and threshold above reliability index R _N of the formula 18 or formula 19 above can be cited. As the number of days reliability index R _D, days reliability index R _D of formula 13 or formula 14 described _above, the upper conditions such as Equation 16 described above days reliability index R _D1, and the above equation 17 The upper days conditional reliability index _RC1 . In particular, the only presents the conditional reliability index R _C and conditionally days reliability index R _CD to the presentation unit 3, conditional standpoint and number of days specifically the determination of future actions based on the predicted value y ₀ It can be reasonably judged from the viewpoint. Alternatively, the only higher than the threshold confidence index R _N and conditionally days reliability index R _CD presents the presentation unit 3, conditional standpoint and number of days specifically the determination of future actions based on the predicted value y ₀ It can be reasonably judged from the viewpoint. This is because when the value of the conditional confidence index R _C and the threshold value or higher confidence index R _N is large, if the conditional days reliability index R _CD is small, the number of days specifically the prediction target data X ₀ This means that there are few conditions similar to

In the first embodiment described above, to determine the predicted value y ₀ from seeking error parameter alpha, if similar conditions, i.e., if the data item is similar, similar such output If the rule of thumb that y is obtained holds, the predicted value y ₀ of the prediction target data X ₀ may be obtained by a weighted average according to the following equation 20.

In addition, when the predicted value y ₀ of the prediction target data X ₀ is not predicted at one point as in the above formula 20, but is predicted as a distribution (probability density distribution), the following formula 3 is used instead of the above formula 3. The distribution (probability density distribution) may be obtained by executing the above-described calculation steps described above using Equation 21 and _obtaining a weighted frequency distribution of _y0j .

  Next, another embodiment will be described.

(Second Embodiment)
The output value prediction apparatus S of the present embodiment can be applied to various cases, and can be applied to the following cases, for example. That is, for example, in the case of products manufactured through various manufacturing processes in a relatively large-scale manufacturing plant, such as manufacturing of steel products and chemical products, depending on, for example, input amount, operation input amount, and time passage In many cases, the output value in each manufacturing process or the output value of the final process directly connected to the product changes every moment. For example, in the manufacturing process of steel products, the relationship between the hot metal temperature in the topped car and the elapsed time, the relationship between the molten steel temperature in the ladle and the elapsed time, and the carbon concentration in the molten steel and the integrated oxygen injection value in the converter blowing. The relationship and the relationship between the molten steel temperature and the blown oxygen integrated value in converter blowing are mentioned.

FIG. 9 is a diagram illustrating the relationship between the temperature drop amount of the object and the elapsed time. FIG. 9 shows the relationship between the temperature drop amount (deviation from the initial temperature) y of the object left in the atmosphere and the elapsed time (temperature measurement time) t plotted with ○ for each past performance data. Assume that it was a result. Here, when predicting the temperature drop amount y (t ₀ ) at a predetermined time t ₀ , when the probability density distribution is obtained by using past performance data (X, y) near the time t ₀ , Problems can arise. That is, first, in the time domain where the past performance data (X, y) is small (or does not exist), there is very little data that can be used, and the data used is a part of the past performance data (X, y). Only it is. For this reason, it is difficult to predict the distribution of the temperature drop amount y (t ₀ ) of the prediction target data with high accuracy. Then, the second, larger historical performance data (X, y) of the similarity between the predicted target data is not necessarily to be near the predetermined time t _0, prediction processing away from the time t ₀ target data X ₀ If there is past performance data (X, y) having a large similarity, the past performance data (X, y) is not utilized.

Therefore, for such a problem, as shown for a part of the past performance data (X, y) by a thin broken line in FIG. 9, the elapsed temperature t of the temperature drop amount y in each past performance data (X, y) and Is constructed, and each past performance data (X, y) is projected at a predetermined time t ₀ (ie, the constructed prediction model; y _j (t) = f (Z _j , Θ) , alpha _j, by obtaining the amount of temperature drop y _{(t 0)} at time _{t 0} of t)), in the past record data away from a predetermined time _{t 0} (X, y) be the predicted value y _{(t 0)} It can be utilized for estimation of probability density distribution, and it becomes possible to obtain the variation of the predicted value y (t ₀ ) of the prediction target data with higher accuracy. That is, by using the output value prediction apparatus S of the first embodiment in this case, it is possible to obtain the variation in the predicted value y (t ₀ ) of the prediction target data with higher accuracy. That is, in the output value prediction device S of the first embodiment, the prediction model is an arbitrary model obtained based on the past performance data (X, y), but in the output value prediction device Sa of the first embodiment. The prediction model is a model that is obtained based on past performance data (X, y) and includes an elapsed time from a predetermined reference time as an input variable. In FIG. 9, the prediction model; y _j (t) = f (Z _j , Θ, α _j , t) is indicated by a thick broken line.

Here, when predicting the predicted value y ₀ , for example, a physical law, an empirical rule, or the like has an influence on the predicted value y ₀ (how the output changes with respect to an input change, the output tendency with respect to the input). Although it is considered that there is a known factor z and a factor z whose influence on the predicted value y ₀ is unknown, the error parameter in the first embodiment has a way of affecting the predicted value y ₀ . It is treated as being caused by the factor Z including a part of the known factor Z. Therefore, in the first embodiment, not out the tendency of the output y in the past record data X to be used for the predicted value y ₀ are also excluded as contained different historical performance data X, unfavorable this historical performance data Variations including X are obtained. Therefore, in the output prediction device Sa of the second embodiment, it should be noted that the first model f is the first relationship between the output variable y and the input variable Z, which is derived using a predetermined predetermined relationship in advance. , the input value z of the input variables Z, the prediction value y ₀ influence extracting factors z with known how influence from factor z to the prediction value y ₀ which gives a, the extracted factors z first model f The value obtained by giving to is a criterion for the error parameter α. The previously known predetermined relationship is, for example, at least one of a previously known physical law, a previously known empirical rule, a previously statistically determined relationship, a previously obtained empirical formula for explaining experimental data, and the like. One can be mentioned. “Known in advance” may be obtained during the calculation process for obtaining the predicted value y ₀ , and in short, it may be obtained before using the predetermined relationship. With this configuration, the error parameter α becomes a difference from a predetermined reference (a reference point or a reference line) set in advance, and the predetermined reference is extracted from a factor z that affects the predicted value y _0. Is generated from the predetermined predictable factor z and is obtained with higher accuracy.

  Hereinafter, the second embodiment will be described in more detail. The output value prediction apparatus Sa of the present embodiment further includes the above-described temperature drop amount y and elapsed time t of an object left in the atmosphere. A case where the relationship is applied will be described.

Output value prediction apparatus Sa of the present embodiment, the output value prediction apparatus S of the basic embodiment, the parameter calculation process (S13) for calculating a parameter, prediction value calculation processing for calculating the predicted value y ₀ (S14) and a plurality of Since the calculation process (S16) of the reliability index R is the same as that of the output value prediction apparatus S in the first embodiment described above except that the process is executed as follows, mainly the different points are described in more detail. Explained.

In the present embodiment, the reliability index calculation unit 16 includes the conditional reliability index R _C described in the first embodiment, the daily reliability index R _D described in the first embodiment, and the predicted time reliability. At least two of the indices _RT are calculated, and the calculated at least two reliability indices R are presented to the presentation unit 3. For this reason, in addition to the conditional reliability index calculation unit 161 and the daily reliability index calculation unit 162, the reliability index calculation unit 16 of the arithmetic control unit 1 further includes a plurality of reliability indexes as shown by a broken line in FIG. As one reliability index R, a predicted time reliability index calculation unit 163 that calculates a predicted time reliability index _RT based on a predetermined predicted time similarity is provided. The predetermined prediction time similarity is obtained by obtaining the past time data used when obtaining the second model (prediction model) generated by the prediction value calculation unit 14 and the given data (prediction target). Data X ₀ ) is an index indicating the degree of similarity to the predicted time, for example, past performance data used when obtaining the second model (prediction model) generated by the predicted value calculation unit 14, for example Is the degree of similarity between the acquisition time at which the data is acquired and the prediction time in the given data (prediction target data X ₀ ).

FIG. 10 is a diagram for explaining a method for calculating an error parameter of a predicted time in the second embodiment. FIG. 10A shows an error parameter α _j (t _j ) at a reference line of past performance data, and FIG. 10B shows an error parameter α (t ₀ ) at a reference line of prediction target data. . FIG. 11 is a diagram illustrating data stored in an actual measurement data storage unit in the output value prediction apparatus of the second embodiment. FIG. 12 is a diagram illustrating data stored in the intermediate data storage unit in the output value prediction apparatus of the second embodiment. FIG. 13 is a diagram for explaining a calculation procedure of predicted value variation in the output value predicting apparatus according to the second embodiment. FIG. 13A shows the predicted value y (t ₀ ) at a predetermined time t ₀ , and FIG. 13B shows the relationship between the similarity w and the output predicted value y (t ₀ ). The horizontal axis is the similarity w, and the vertical axis is the predicted value y (t ₀ ).

In the output value prediction apparatus Sa of the second embodiment, the actual data storage unit 41 of the storage unit 4 stores the past performance data X and the prediction target data X ₀ in a table format (table format), as in the first embodiment. Is stored in advance. In the second embodiment, the past performance data X and the prediction target data X ₀ include the temperature drop amount y, the actual measurement time t when the temperature drop amount y is measured, and the factor data x related to the temperature drop amount y. It is prepared for. The temperature drop amount y corresponds to the predetermined output y, and the actual measurement time t can be regarded as one of the elements (factor elements) x in the factor X related to the predetermined output y. That is, the factor X related to the predetermined output y includes at least the time t as an element. The origin of the actual measurement time t is the time when the temperature drop amount y = 0, that is, the time of the initial temperature of the object (the time when the measurement of the temperature of the object is started).

The historical performance data ((X, t), y) the predicted target data _(X 0, _{t 0)} the output value from the (predicted value) prediction of _{y 0} is started on the basis of, in the processing S11, first Similar to the embodiment, in the present embodiment, the distance calculation unit 11 between the past performance data (X, t) and the prediction target data (X ₀ , t ₀ ) in the first to Nth data item spaces. The distance d _j is calculated, and the calculated distance d _j is stored in the intermediate data storage unit 42 of the storage unit 4.

Next, in the process S12, as in the basic mode, the similarity calculation unit 12 calculates the similarity w _j between the prediction target data (X ₀ , t ₀ ) and the past performance data (X, t), Calculation is performed for each of the first to M-th past performance data (X, t), and the calculated similarity w _j is stored in the intermediate data storage unit 42 of the storage unit 4.

Next, in the process S13, in this embodiment, the parameter calculator 13 uses the equation 22-1 instead of the equation 3 in the first embodiment, so that the parameter calculation unit 13 sets the error parameter α _j representing the error from the reference. Each of the first to M-th past performance data ((X, t), y) is calculated, and the calculated error parameters α _j are stored in the intermediate data storage unit 42 of the storage unit 4. Note that Formula 22-1 may be Formula 21-2.

ここで、ｙ_ｊ（ｔ）は、実測時刻ｔによける温度降下量であり、ＺおよびΘは、前記基本態様の式３と同様である。

Here, y _j (t) is the amount of temperature drop at the actual measurement time t, and Z and Θ are the same as in Equation 3 of the basic mode.

In the present embodiment, the function f is a function for calculating a predetermined reference (reference value), and the error parameter α _j (t) is given (calculated) by the function f. The difference from the reference point or reference line). The function f includes, as an input variable Z, a factor z that knows how to affect the predicted value y (t ₀ ) (how the output changes relative to the input change, and how the output tends to change). For example, it is given by a physical law, an empirical rule or a statistical technique. The predetermined reference value given by the function f is, for example, an average value, a median value, an upper limit value, a lower limit value, or the like. Further, the error parameter α _j (t) is treated as a result of a factor whose influence on the predicted value y (t ₀ ) is unknown. Note that the remaining factor obtained by extracting the predetermined predictable factor z from the factor z that affects the predicted value y ₀ is only the factor whose influence on the predicted value y (t ₀ ) is unknown. However, since the designer cannot determine whether or not the factor z is a predictable factor, the predictable factor z may be included. That is, quantifiable factors relating to the predicted value y (t _0), for example physical laws and heuristics or statistical output by a method such as (predicted value y (t ₀₎₎ manner, for example, subjective effects given to The above-mentioned physical laws, empirical rules, statistical methods, etc. are divided into the first factor and the remaining second factor that are known in the above, and the influence of the first factor on the predicted value y (t ₀ ) is used as the reference. By using this, it is possible to calculate the accuracy of the prediction value y (t ₀ ) with higher accuracy, and to capture the variation in the predicted value y (t ₀ ) as the variation from the reference, thereby calculating the variation in the predicted value y (t ₀ ) with higher accuracy. . In other words, the error parameter α _j (t) is a parameter caused by the second factor that is not considered in the modeling in the model relating to the first factor.

Θ of the function f is a known model parameter that can be obtained as described in the first embodiment. Each past performance data ((X, t), y) to y _j (t _j ), Z _{Since j} and t _j are also known, an error parameter α _j (t) is obtained in this process S13. More specifically, in the case of Expression 22-1, as shown in FIG. 10A, the j-th past performance data ((Z _j , t _j ), y _j (t _j )) (point B _j ) and the difference between the reference value (point A _j ) at time t _j on the reference line j; f (Z _j , Θ, t), as represented by Equation 23, the error parameter α _j (t ) Is required.

As can be seen from the difference between the present embodiment and the first embodiment described above, the expressions 22-1 and 22-2 are combined with the function f and the error parameter α _j (t) to regenerate one function f. By considering as “y”, y _j (t) = f ′ (Z _j , Θ, α _j , t), which is an expression corresponding to Expression 3 of the first embodiment.

The weighted Euclidean distance d _j , the similarity w _j and the error parameter α _j calculated by each of the processes S11 to S13 are, for example, shown in FIG. Is remembered. The intermediate data table 52A shown in FIG. 11 corresponds to the intermediate data table 52 shown in FIG. 5 in the first embodiment, and the output field 521 and the similarity calculation data are similar to the intermediate data table 52 shown in FIG. A field 522, an output prediction data field 523, a weighted distance field 524, a similarity field 525, and an error parameter field 526, which are given by a function f serving as a reference for the error parameter α _j (t). A reference line field 527 for registering f (Z _j , Θ, t), and a time field 528 for registering the time (actual measurement time) t of the past performance data ((X, t), y). With a record for each past performance data ((X, t), y). It has a record of the target data X _0.

Next, the process S14, using the prediction time t _0, the other is by treating in the same manner as the first embodiment, the prediction value calculating unit 14 uses the prediction model which has been determined by the processing S13, the prediction target Each predicted value y ₀ (t ₀ ) determined in the above-described step S13 based on the predicted time t ₀ in the data (X ₀ , t ₀ ) and the data values z _{01 to} z _0L of the first to Lth data items calculated for each of the error parameters α _j (t), the predicted value _y 01 _{(t 0)} which is the calculated _~y 0M _{(t 0)} the similarity _w 1 to w _M and association with the prediction of the storage unit 4 Store in the value storage unit 43.

Here, it should be noted that in this embodiment, a value and an input obtained by giving the first model f an input value x corresponding to the input variable X in the past performance data ((X, t), y). The difference between the output value y of the output variable corresponding to the input value x of the variable X is used as the error parameter α (α _j (t _j )) at time t when the past performance data ((X, t), y) is obtained. The obtained difference α _j (t _j ) is converted into the value (α _j (t ₀ ) of the prediction time t ₀ for which the prediction target data (x ₀ , y ₀ ) is to be predicted using a predetermined conversion relationship. 10A, that is, as shown in FIG. 10A, the error parameter α _j (t _j ) (; line segment A _j B at the actual measurement time t _j obtained in the process S13. based on the _j), error path in the prediction time _{t 0} Meter α _{j (t} 0) (; line _C j _{D j)} is obtained, as shown in FIG. 10 (B), error parameters α _{j (t} 0) at the predicted time _{t 0} (; segment _{C j} D _j = line _E j _{F j)} is used for the calculation of the predicted value _{y 01 (t 0) ~y 0M} (t 0). Therefore, the predicted value _y 0j _{(t 0)} of the formula 25-1 In addition, when Expression 22-2 that expresses the difference from the reference in an additive manner is used instead of Expression 22-1 that expresses the difference from the reference value additively, Expression 25-2 is used. Instead of -1, Equation 25-2 is obtained.

The calculation of the error parameter α _j (t ₀ ) at the predicted time t ₀ based on the error parameter α _j (t _j ) at the actual measurement time t _j is the function formula f used for the calculation of the reference, its program, or qualitative Predetermined predetermined conversion relationships such as general knowledge and simple models are used. For example, Formula 24-1 and Formula 24-2 can be used.

Here, in Expression 24-1, the error parameter α _j (t) is constant without depending on time, and the error parameter α _j (t _j ) at the actual measurement time t _j obtained in the processing S13 is the total time t. The error parameter α _j (t) in FIG. Expression 24-2 shows that the error parameter α _j (t) is proportional to the output value of the function f.

Further, in Expression 25-1, the difference (variation) from the reference line in each past performance data ((X, t), y) is mapped at the prediction time t ₀ , and this is the difference from the reference line at the prediction time t ₀ . It shows that it is mapped to. That is, as shown in FIG. 10A, the line segment A _{j at the} actual measurement time t _j in the reference line j; f (Z _j , Θ, t) in the past performance data data ((X, t), y). B _j is mapped to the predicted time t ₀ as a line segment C _j D _j , and this line segment C _j D _j for the reference line in the past performance data ((X, t), y) is shown in FIG. As shown, it is mapped as a line segment E _j F _j for the reference line in the prediction data (X ₀ , t ₀ ).

Incidentally, in FIG. 10, if the measured time _{t i} is also shown, the error parameters in the measured time _{t i α i (t i)} (; segment _A i _{B i)} predicted time obtained based on the t ₀ in the error parameter α _{j (t} 0) (; line _C j _{D j)} is shown, also, the error parameter α _{j (t} 0) (; line _C i _{D i} = the line segment _E i _F i) The state in which is an error parameter α from the reference value at the predicted time t ₀ is also illustrated.

  In the processing S14, as in the processing S13, an arithmetic program including a table representing the expression 25-1 or 25-2, a convergence calculation algorithm, an if-then rule, a fuzzy inference, a neural network, a JIT model, and the like is used. May be.

Such step S14 the predicted value _y 01 calculated by _{(t 0) ~y 0M (t} 0) , for example, stored in the prediction value storing unit 43 by the table format table format as shown in FIG. 12 The The predicted value data table 53A shown in FIG. 12 corresponds to the predicted value data table 53 shown in FIG. 6 in the first embodiment, and, similarly to the predicted value data table 53 shown in FIG. Each field includes a prediction data field 532, an error parameter field 533, and a similarity field 534, and a record is provided for each error parameter α _j . Furthermore, the predicted value data table 53 shown in FIG. 12 includes a predicted time field 535 for registering the predicted time t0. The error parameter field 533 is divided into a subfield 5331 for registering an error parameter α (t _j ) at an actual measurement time t _j and a subfield 5332 for registering an error parameter α (t ₀ ) at an estimated time t ₀ .

Next, the process S15, as in the first embodiment, the variation calculation unit 15 uses the respective predicted value _y 01 obtained by the processing _{S14 (t 0) ~y 0M (} t 0), the predicted value y The variation of ₀ (t ₀ ) is calculated, and the calculated variation of the predicted value y ₀ (t ₀ ) is stored in the variation storage unit 44 of the storage unit 4. More specifically, as shown in FIG. 13 (B), the variation calculation unit 15 takes the predicted value y (t ₀ ) on the vertical axis and the similarity w on the horizontal axis. M prediction values y _0j (t ₀ ) respectively corresponding to M error parameters α _j (j = 1 to M) calculated from each past performance data ((X, t), y). On the other hand, the similarity w _{j is made} to correspond. Next, the variation calculation unit 15 _divides a range including at least each predicted value y _0j (t ₀ ) on the vertical axis y (t ₀ ) in FIG. _13B into a finite number of sections (classes, grades). Then, the weighted frequency F _w is generated by adding all the similarities w _j of the predicted values y _0j (t ₀ ) included in each section, and a histogram representing the variation of the predicted values y ₀ (t ₀ ) is generated. To do. The histogram may be a variation of the predicted value y ₀ (t ₀ ), but in the present embodiment, the variation calculation unit 15 further normalizes the area of the histogram to be 1. This normalized histogram is used as the probability density distribution (empirical probability density distribution) of the predicted value y ₀ (t ₀ ) in the prediction target data (x ₀ , t ₀ ), and the predicted value y ₀ (t ₀ ) It is regarded as variation. Or the variation calculation part 15 calculates | requires the said curve similarly to the above-mentioned from this histogram, further maintaining the area at 1. This curve is a probability density distribution of the predicted value y ₀ (t ₀ ) in the prediction target data (x ₀ , t ₀ ), and is a variation of the predicted value y ₀ (t ₀ ).

Next, the process S16, as in the first embodiment, the reliability index calculating unit 16, a reliability index R for the predicted value y ₀ plurality calculated from a plurality of viewpoints, stores the calculated reliability index R Store in part 4. More specifically, the reliability index calculation unit 16 uses the conditional reliability index calculation unit 161 to include the above-described conditional reliability index R _C (including a function), conditional days, as in the first embodiment. Calculating one or more of the statistical reliability index R _CD (including the function), the upper conditional days reliability index R _D1 (including the function) and the above-threshold reliability index R _N (including the function); and As in the first embodiment, the daily reliability index calculation unit 162 causes the above-described daily reliability index R _D (including a function) and upper days conditional reliability index R _C1 (including a function) to be included. One or more of are calculated. In the present embodiment, the reliability index calculation unit 16 further calculates the prediction time reliability index _RT by the prediction time reliability index calculation unit 163 indicated by a broken line in FIG.

For example, the predicted time reliability index calculation unit 163 calculates the average value of the predicted time similarity w _T as the reliability index R (predicted time reliability index R _T ). This is, that there is a large average value of the prediction time Similarity w _T, it only, in the past performance data (X, y), data (elapsed time is similar to the predicted time to the prediction target data X ₀ similar and has data) means that include many, because such a historical record data (X, considered highly reliable predictions y ₀ obtained based on y). The prediction time similarity w _T is the acquisition time t _j (j = 1 to M) at which the past performance data (X, y) used when obtaining the prediction model and the prediction time in the prediction target data y ₀ . The degree of similarity with t ₀ is defined by, for example, Expression 26-2, and the average value (predicted time reliability finger R _T ) is defined by Expression 26-1.

ここで、ａは、正の実数の調整パラメータである。

Here, a is a positive real adjustment parameter.

Further, for example, the predicted time reliability index calculation unit 163 may use a function of the average value of the predicted time similarity w _T as the predicted time reliability index _RT . The predicted time reliability index _RT may be defined by the equation 26-1, but the value according to the equation 26-1 is 0 ≦ R _T ≦ 1 when 0 ≦ w _Tj ≦ 1. However, in the case where the predicted temporal similarity w _Tj does not have such a range limitation, it is generally 0 or more and there is no upper limit. The user, although the degree of reliability in the magnitude of the average value of the predicted time Similarity w _T can be relatively determined from past experience, since such a empirical judgment, predicted time trust The sex index R _T is a function of the average value of the predicted temporal similarity w _T such that 0 ≦ R _T ≦ 1. Such a function of the average value of the predicted temporal similarity w _T (predicted temporal reliability index R _T (R _{Ta to} R _Tc )) is, for example, any one of Expressions 27-1 to 27-3. Defined by

ここで、ｃ_１は、正の実数の調整パラメータであり、ｃ_２は、実数（負の値でもよい）の調整パラメータである。

Here, c ₁ is a positive real number adjustment parameter, and c ₂ is a real number (which may be a negative value) adjustment parameter.

In this way, the reliability index R, here the predicted temporal reliability index _RT , is scaled so as to be within a predetermined numerical range, so that it can be appropriately presented to the presentation unit 3. As described above, the predicted temporal reliability index calculation unit 163 also functions as an example of a scaling unit that scales the reliability index R so as to be within a predetermined numerical range. The same applies to the following examples.

Further, for example, the predicted time reliability index calculation unit 163 uses the average value of the products of the conditional similarity w and the predicted time similarity w _T as the reliability index R (conditional predicted time reliability index R). _CT ). This means that the average value of the product of the conditional similarity w and the predicted temporal similarity w _T is large, so that the prediction target data X ₀ is conditionally included in the past performance data (X, y). This is because both mean that contain many similar data to predicted time, such a historical record data (X, y) is considered to be high reliability of the predicted value y ₀ obtained based on. The average value of the product of the conditional similarity w and the predicted temporal similarity w _T (conditional predicted temporal reliability index R _CT ) is defined by Equation 28.

In addition, for example, the predicted time reliability index calculation unit 163 includes the expression 12 (expression 12-1 to expression 12-3) for the above expression 11 and the expression 14 (expression 14-1 to expression 14) for the expression 13-1. -3), the function of the average value of the products of the conditional similarity w and the predicted temporal similarity w _T described above may be used as the conditional predicted temporal reliability index R _CT . Such a conditional prediction temporal reliability index R _CT is defined by, for example, expressing Expression 28 as Expression 12 or Expression 14.

Further, for example, the predicted temporal reliability index calculation unit 163 calculates the average value of the products of the conditional similarity w, the daily similarity w _D and the predicted temporal similarity w _T as the reliability index R (conditional Calculated as a daily predicted time reliability index R _CDT ). This is because the average value of the product of the conditional similarity w, the daily similarity w _D, and the predicted temporal similarity w _T is large in the past performance data (X, y). This means that the data X ₀ includes a lot of data that is similar in terms of condition, number of days, and predicted time, and the predicted value y ₀ obtained based on such past performance data (X, y). This is because the reliability of this is considered high. The average value (conditional predicted temporal reliability index R _CT ) of the product of the conditional similarity w, the daily similarity w _D and the predicted temporal similarity w _T is defined by Equation 29.

In addition, for example, the predicted time reliability index calculation unit 163 includes the expression 12 (expression 12-1 to expression 12-3) for the above expression 11 and the expression 14 (expression 14-1 to expression 14) for the expression 13-1. -3), the function of the average value of the product of the conditional similarity w, the daily similarity w _D and the predicted time similarity w _T is used as a conditional daily predicted time reliability index. R _CDT may be used. Such a conditional number of days prediction predicted time reliability index R _CDT is defined by, for example, expressing Expression 29 as Expression 12 or Expression 14.

Further, for example, the predicted temporal reliability index calculation unit 163 extracts the past performance data (X, y) in which the conditional similarity w is within the upper fixed amount, and the extracted past performance data (X, y) the average value of the predicted time similarity w _T against) is calculated as a reliability index R (upper condition predicted time reliability index R _T1). The conditional similarity w within the upper fixed amount is, for example, the conditional similarity w from the top to a predetermined number (for example, the conditional similarity w within the top 100) or the like from the top to a predetermined ratio. Conditional similarity w (for example, conditional similarity w within the top 30%). Data This is the fact that a large average value, it just, in the prediction target data X ₀ conditionally similar historical performance data (X, y), which predicted time similar to the predicted target data X ₀ This is because the predicted value y ₀ obtained based on such past performance data (X, y) is considered to be highly reliable. Such an upper condition predicted temporal reliability index _RT1 is defined by, for example, Equation 30.

ここで、Ω_ｃは、条件的類似度ｗが上位一定量以内となる過去実績データ（Ｘ、ｙ）の集合であり、条件的類似度ｗが大きい方から抽出する過去実績データ（Ｘ、ｙ）の個数をＮｃとし、条件的類似度ｗが大きい方からＮｃ個目の条件的類似度の値をωとすると、Ω_ｃは、｛ｊ｜ｗ_ｊ≧ω｝である（Ω_ｃ＝｛ｊ｜ｗ_ｊ≧ω｝）。

Here, Ω _c is a set of past performance data (X, y) in which the conditional similarity w is within the upper fixed amount, and past performance data (X, y) extracted from the one having the higher conditional similarity w ) Is Nc, and the value of the Nc-th conditional similarity from the larger conditional similarity w is ω, Ω _c is {j | w _j ≧ ω} (Ω _c = { j | w _j ≧ ω}).

In addition, for example, the predicted time reliability index calculation unit 163 includes the expression 12 (expression 12-1 to expression 12-3) for the above expression 11 and the expression 14 (expression 14-1 to expression 14) for the expression 13-1. the same circumstances -3), the conditional similarity w past actual data (X to be within the top fixed amount, estimated time similarity w average function higher conditions predicted time confidence of _T with respect to y) It is good also as sex index _RT1 . Such an upper condition predicted temporal reliability index R _T1 is defined, for example, by expressing Expression 30 as Expression 12 or Expression 14.

Also, for example, the predicted temporal reliability index calculation unit 163 extracts past performance data (X, y) in which the predicted temporal similarity w _T is within a certain upper level, and the extracted past performance data (X , Y), the average value of the conditional similarity w is calculated as the reliability index R (upper predicted time conditional reliability index R _C2 ). The predicted temporal similarity w _T within the upper fixed amount is, for example, the predicted temporal similarity w _T (for example, the predicted temporal similarity w _T within the upper 100) or the like from the upper level to a predetermined number The predicted time similarity w _T up to a predetermined ratio (eg, predicted time similarity w _T within the upper 30%). Data This is the fact that a large average value, it only in the predicted time similar to the historical performance data to the predicted target data X ₀ (X, y), that condition similar to the predicted target data X ₀ This is because the predicted value y ₀ obtained based on such past performance data (X, y) is considered to be highly reliable. Such an upper predicted time conditional reliability index _RC2 is defined by, for example, Equation 31.

ここで、Ω_Ｔは、予測時刻的類似度ｗ_Ｔが上位一定量以内となる過去実績データ（Ｘ、ｙ）の集合であり、予測時刻的類似度ｗ_Ｔが大きい方から抽出する過去実績データ（Ｘ、ｙ）の個数をＮ_Ｔとし、予測時刻的類似度ｗ_Ｔが大きい方からＮ_Ｔ個目の条件的類似度の値をωとすると、Ω_Ｔは、｛ｊ｜ｗ_Ｔｊ≧ω｝である（Ω_Ｔ＝｛ｊ｜ｗ_Ｔｊ≧ω｝）。

Here, Ω _T is a set of past performance data to predict time Similarity w _T is within the top fixed amount (X, y), past performance data to be extracted from the larger prediction time Similarity w _T (X, y) when the number of the _{N T,} the value of _{N T} th conditional similarity from the direction predicted time similarity _{w T} is large and omega, omega _T is, {j _{| w Tj} ≧ ω } (Ω _T = {j | w _Tj ≧ ω}).

In addition, for example, the predicted time reliability index calculation unit 163 includes the expression 12 (expression 12-1 to expression 12-3) for the above expression 11 and the expression 14 (expression 14-1 to expression 14) for the expression 13-1. -3), the function of the average value of the conditional similarity w with respect to the past performance data (X, y) in which the predicted time similarity w _T is within the upper fixed amount is used as the upper predicted time conditional reliability. It may be the sex index _RC2 . Such an upper predicted time conditional reliability index _RC2 is defined, for example, by expressing Expression 31 as Expression 12 or Expression 14.

The output value prediction unit Sa is the arithmetic control unit 1, the variation of the predicted value y ₀ calculated by the variation calculation unit 15 in the processing S15, and a plurality of calculated by the reliability index calculating unit 16 in the processing S16 The reliability index R is presented to the presentation unit 3 (S17), and the process ends.

In the present embodiment, by this way the output value prediction apparatus Sa of the second embodiment operates, the predetermined reference by extracting a predetermined predictable factors from factors affecting the expected value y ₀ product by being, regardless of the manner of distribution of the variations of the predicted value y _0, it is possible to obtain the prediction value y ₀ of the output in the process output changes every moment with the lapse of time, and this predicted value y It is possible to obtain a variation of _zero . Further, in the output value predicting device Sa of the second embodiment, it is possible to obtain the predicted value y ₀ even in a time region where the past performance data ((X, t), y) is small (or does not exist). variation of y ₀ also becomes possible to obtain. Further, in the output value prediction device Sa of the second embodiment, past performance data ((X, t), y) separated from a predetermined time t ₀ is also used for estimation of variation in the predicted value y ₀ (t ₀ ). Therefore, it is possible to obtain the variation of the predicted value y ₀ (t ₀ ) of the prediction target data with higher accuracy. In the output value prediction apparatus Sa of the second embodiment, the error parameter α _j is a predetermined error parameter α (t _j ) at time t _j when the past performance data ((X, t), y) is obtained. Since the prediction target data (x ₀ , y ₀ ) is converted into a value at the prediction time t ₀ to be predicted by the conversion relationship and is set as the error parameter α _j , the error parameter is obtained more appropriately.

As described above, in the output value prediction device Sa in the second embodiment, the variation of the prediction value y ₀ is obtained by using the prediction target data X ₀ for the prediction model, and the variation of the prediction value y ₀ A plurality of reliability indexes R are obtained, and the obtained variation of the predicted value y ₀ and a plurality of reliability indexes R are presented. Thus the output values predicting apparatus Sa of the present embodiment, with respect to the predicted value y _0, obtains a plurality of reliability index R from a plurality of viewpoints, it is possible to present a plurality of reliability index R which this calculated. For this reason, the user can evaluate the obtained variation of the predicted value y ₀ from a plurality of viewpoints by referring to the plurality of reliability indexes R presented, and the predicted value y ₀ It is possible to reasonably judge the determination of future actions based on these multiple viewpoints.

In the output value prediction apparatus Sa of the second embodiment, as the plurality of reliability indexes R, at least one of the conditional reliability index R _C , the daily reliability index R _D, and the predicted temporal reliability index R _T. Two can be presented. Therefore, the user, if the condition reliability index R _C are presented in the past record data and the predicted target data X ₀ by referring to the condition reliability index R _C (X, y) and the Conditional similarity can be determined, and variations in the predicted value y ₀ can be evaluated from this point of view, and if a daily reliability index R _D is presented, this daily reliability index R historical performance data and the prediction target data X ₀ by referring to _D (X, y) can be determined the number of days similarity with, it is possible to evaluate the variations of the predicted value y ₀ in this respect, Then, the predicted time of the case, the prediction target data X ₀ by referring to the predicted time reliability index R _T with past actual data (X, y) of the predicted time reliability index R _T are presented To judge similar similarity Can, variations of the predicted value y ₀ can be evaluated from this viewpoint. In this way the user, prediction and whether or not the point whether the target data X ₀ is a well there was the case in the past (cases), recently also frequently asked whether or not the point is the case (previously recently were often is It is possible to interpret separately (separately) whether there are few cases or not, and whether or not the case is a common predicted time. For this reason, the determination of the future action based on the predicted value y ₀ is rationally performed from the viewpoint of the condition according to the plurality of reliability indexes R presented to the presentation unit 3, the viewpoint of days, and the viewpoint of the predicted time. It becomes possible to judge. The determination can be made in consideration of, for example, a risk due to a change in conditions, a risk due to changes over time such as days, and a risk due to changes in the elapsed time of the process.

In particular, the output changes from moment to moment according to the elapsed time of the process and the integrated value of the operation input, for example, hot metal temperature in the topped car, molten steel temperature in the ladle, Predicts the value for the accumulated value of transport time, processing time and operation input, which is not so much in the past, such as carbon concentration in molten steel, molten steel temperature in converter blowing, phosphorus concentration in molten iron, and phosphorus concentration in molten steel The user's (operator) who considers the risk depends on the degree of reliability that the predicted value has in the case of predicting the value for the accumulated value of the transfer time, processing time, and operation input amount that was often in the past. The operation is usually different. For this reason, it is effective to present the predicted time reliability index _RT from the viewpoint of similarity in predicted time.

In the present embodiment, the conditionally reliability index R _C, conditionally reliability index R _C of formula 11 or formula 12 described _above, such as formula 15 above conditional days reliability index R _CD, The above-mentioned threshold value reliability index R _{N of} the above-described equation 18 or 19, the conditional prediction time-of-day reliability index R _CT such as the above-described equation 28, and the conditional number-of-days prediction-time reliability such as the above-described equation 29 An index R _CDT is mentioned. As the number of days reliability index R _D, days reliability index R _D of formula 13 or formula 14 described _above, the upper conditions such as Equation 16 described above days reliability index R _D1, and the above equation 17 The upper days conditional reliability index _RC1 . In particular, the only presents the conditional reliability index R _C and conditionally days reliability index R _CD to the presentation unit 3, conditional standpoint and number of days specifically the determination of future actions based on the predicted value y ₀ It can be reasonably judged from the viewpoint. Alternatively, the only higher than the threshold confidence index R _N and conditionally days reliability index R _CD presents the presentation unit 3, conditional standpoint and number of days specifically the determination of future actions based on the predicted value y ₀ It can be reasonably judged from the viewpoint. This is because when the value of the conditional confidence index R _C and the threshold value or higher confidence index R _N is large, if the conditional days reliability index R _CD is small, the number of days specifically the prediction target data X ₀ This means that there are few conditions similar to As the predicted time reliability index R _T, predicted time reliability index R _T of Formula 26 or Formula 27 described _above, the upper condition predicted time reliability index R _T1 and above of formula 30 such as described above An upper predicted time conditional reliability index _RC2 such as 31 is cited. In particular, the only presents the conditional reliability index R _C and conditionally predicted time reliability index R _CT to the presentation unit 3, conditional aspects and estimated time to determine the future actions based on the predicted value y ₀ Can be reasonably judged from a general viewpoint. Alternatively, the only higher than the threshold confidence index R _N and conditionally predicted time reliability index R _CT are presented to the presentation unit 3, conditional aspects and estimated time to determine the future actions based on the predicted value y ₀ Can be reasonably judged from a general viewpoint. This is because when the value of the conditional confidence index R _C and the threshold value or higher confidence index R _N is large, if the condition prediction time reliability index R _CT is small, the prediction target data X ₀ prediction This is because there are few conditions that are similar in time.

Further, as can be seen from FIG. 10, the output value prediction apparatus Sa of the second embodiment can obtain the predicted values y ₀ (t) at a plurality of different times t, and the variation in the predicted values y ₀ (t). Can also be sought. Therefore, it is possible to determine the processing end timing with the least risk by comparing the variation of the predicted value y ₀ (t) at each time t. For example, in a heating furnace of a steel product manufacturing process, it is possible to terminate the heat treatment at a low risk timing, considering not only whether the steel was heated as intended but also considering the probability of temperature deviation. It becomes. Further, for example, in converter blowing, it is possible to terminate the blowing at a timing with less risk in consideration of the molten steel temperature and the probability that the components in the molten steel deviate from the target. In the case of this converter blowing, the horizontal axis of FIG. 14 is the integrated value of the blown blowing oxygen amount.

Further, in order to obtain with high accuracy, the amount of calculation processing is usually relatively large. However, in the output value prediction device Sa of the second embodiment, a quantifiable factor Z related to the output (predicted value y (t ₀ )). Are separated into a first factor z and a residual second factor z that subjectively know how to affect the output (predicted value y (t ₀ )), so that the first factor used as a reference It is only necessary to perform calculation with higher accuracy than only the output related to z, and by storing this result, it is not necessary to execute the calculation related to the first factor z thereafter. Therefore, the variation of the predicted value y (t ₀ ) can be obtained with a relatively small amount of calculation processing, and can be calculated by a simpler computer. Furthermore, by calculating and storing the error parameter α _j (t) from the reference line in advance at predetermined time intervals, the predicted value y (t) can be obtained with a smaller amount of calculation processing and by a simpler computer. ₀ ) can be obtained.

  Next, another embodiment will be described.

(Third embodiment)
In the steel product manufacturing process, after the converter is blown, the molten steel is transferred from the converter to the ladle, and after the molten steel is processed, the molten steel is transported to the continuous casting equipment. Therefore, it is preferable that the molten steel temperature is slightly higher than the solidification temperature when the ladle arrives at the continuous casting facility. If the molten steel temperature is too low, the molten steel solidifies, which is not preferable. If the molten steel temperature remains high, the casting speed must be reduced. Depending on the charge, the molten steel composition, the amount of molten steel, the type of ladle, the initial state of the ladle (the refractory melting condition, the temperature distribution inside the ladle (cooling condition)), and the ladle when receiving steel from the converter The amount of temperature drop varies depending on the amount of alloy and type of alloy placed in advance. For this reason, it is difficult to predict the molten steel temperature that changes from moment to moment at a single point. Therefore, it is important to accurately estimate the variation in molten steel temperature in the ladle for the charge.

  In the third embodiment, the predetermined output is the molten steel temperature in the ladle in the process from the converter steelmaking process through the molten steel treatment process to the continuous casting process, and the output value prediction device Sa of the second embodiment is applied. The output value prediction device Sb in the third embodiment estimates the probability density distribution for the temperature drop of the molten steel until the molten steel transferred from the converter to the ladle is transported to the molten steel processing facility. Is. Therefore, the output value prediction device Sb in the third embodiment is the same as the output value prediction device Sa in the second embodiment, in the distance calculation process (S11) for calculating the distance, the parameter calculation process (S13) for calculating the parameter, and the predicted value. The predicted value calculation process (S14) for calculating is the same as the output value prediction apparatus S in the second embodiment except that the process is executed as follows, and the description of the same points is omitted.

  FIG. 14 is a diagram illustrating data stored in a predicted value storage unit in the output value predicting apparatus according to the third embodiment. FIG. 15 is a diagram illustrating a probability density distribution of each predicted value in the output value predicting apparatus according to the third embodiment. The horizontal axis in FIG. 15 is the elapsed time t expressed in units of minutes (min), and the vertical axis thereof is the temperature drop amount y (t) expressed in degrees (° C.).

In the output value prediction device Sb of the third embodiment, the actual data storage unit 41 of the storage unit 4 stores the past performance data ((X, t), y in a table format (table format) as in the first embodiment. ) And prediction target data (X ₀ , y ₀ ) are stored in advance. In the third embodiment, the past performance data ((X, t), y) and the prediction target data (X ₀ , y ₀ ) are the temperature drop amount y, the actual measurement time t when the temperature drop amount y is measured, And the factor data x related to the temperature drop amount y is provided. The temperature drop amount y corresponds to the predetermined output y, and the actual measurement time t can be regarded as one of the elements (factor elements) x in the factor X related to the predetermined output y. The origin of the actual measurement time t is the time when the temperature drop amount y = 0, that is, the time of the initial temperature of the object (the time when the measurement of the temperature of the object is started). Each element (data item) x _ji in the factor used to obtain the similarity among the factors X related to the temperature drop y is the number of times the ladle is received, the type of deoxidizer, the molten carbon concentration, These are items such as the state of the pan in the pan, the output temperature, the solidification temperature, and the operation team. Here, in the present embodiment, the number of times the steel is received by the ladle is converted by a non-linear function so as to be the square root of the number of times the steel is received, for example. The type of deoxidizer is quantified according to the strength of deoxidation. When the ladle is in an empty pan state (the state in which no molten steel is contained), the standing time, the heat retaining time, the standing time after the heat retaining, and the like are quantified by a nonlinear function. The operation group is given an identifier for each group.

When the prediction of the output value (predicted value) y ₀ is started from the prediction target data (X ₀ , t ₀ ) based on the past performance data ((X, t), y), in the process S11, Expression 32 The distance calculation unit 11 uses the distance d _j defined in the above to perform the same processing as in the second embodiment, so that the distance calculation unit 11 in the present embodiment, in the first to Nth data item spaces, X, t) and the distance d _j between the prediction target data (X ₀ , t ₀ ) are calculated, and the calculated distance d _j is stored in the intermediate data storage unit 42 of the storage unit 4.

ここで、ｆ_ｄ（ｘ_ｊｉ，ｘ_０ｉ）は、ｘ_ｊｉとｘ_０ｉとが同じ場合に０をとり、ｘ_ｊｉとｘ_０ｉとが異なる場合に１をとる関数である。そして、本実施形態では、ａ_ｉ（ｉ＝１〜Ｎ）＝１とされる。Ｎは、データ項目数である。また、ｋ＜Ｎである。

Here, f _d (x _ji , x _0i ) is a function that takes 0 when x _ji and x _0i are the same, and takes 1 when x _ji and x _0i are different. In this embodiment, a _i (i = 1 to N) = 1. N is the number of data items. Further, k <N.

When comparing the operating conditions of the relevant charge with the operating conditions of each past charge, it may be subtracted, such as the operating team or equipment number (number of equipment used for processing among multiple equipment). can not data items or, some data items insignificant to the subtraction itself, a distance d _j defined by equation 32, valid for data items such data item is significant whether the same It is.

Further, date and time or date may be added as a data item when calculating the similarity w _j . Some processes have characteristics that change over time, such as aging and seasonal variation. In such a case, even if the operating conditions are the same, the result may be different if the date is different. By adding the month and day as a data item, it is possible to reduce the similarity w _j of the old data and make a prediction considering the secular change. Note that the date may be expressed as the number of days elapsed from a reference date (for example, January 1, 1900).

Next, in the process S12, as in the second embodiment, the similarity calculation unit 12 calculates the similarity w _j between the prediction target data (X ₀ , t ₀ ) and the past performance data (X, t). The first through M-th past performance data ((X, t), y) are calculated, and the calculated similarities w _j are stored in the intermediate data storage unit 42 of the storage unit 4.

Here, the similarity defined by Expression 33 is used instead of Expressions 2-1 to 2-3 as the similarity w _j .

ここで、μは、距離ｄ_ｊ（ｊ＝１〜Ｍ）の平均値であり、σは、距離ｄ_ｊ（ｊ＝１〜Ｍ）の標準偏差である。また、本実施形態では、ｇ＝１とされる。そして、本実施形態では、予め設定された所定の閾値よりも小さい類似度ｗ_ｊは、０とされる。類似度ｗ_ｊの低い過去実績データ（（Ｘ、ｔ）、ｙ）を予め除去することによって、例えば重み付き度数Ｆ_ｗを算出するための演算処理量等の以下の演算処理量の軽減（演算処理時間の短縮）が可能となる。

Here, μ is an average value of the distance d _j (j = 1 to M), and σ is a standard deviation of the distance d _j (j = 1 to M). In the present embodiment, g = 1 is set. In the present embodiment, the similarity w _j smaller than a predetermined threshold value set in advance is set to zero. By removing the past performance data ((X, t), y) having a low similarity w _j in advance, for example, the following calculation processing amount such as the calculation processing amount for calculating the weighted frequency F _w is reduced (calculation Processing time).

Next, in the process S13, by using the formula 22-1 instead of the formula 3 of the basic mode, the parameter calculation unit 13 sets the error parameter α _j representing the error from the reference to the first to M-th past. Each result data ((X, t), y) is calculated, and each calculated error parameter α _j is stored in the intermediate data storage unit 42 of the storage unit 4. For the reference line in Formula 22-1; f (Z _j , Θ, t _j ), for example, the method disclosed in Japanese Patent Application Laid-Open No. 2007-167858 or the method disclosed in Japanese Patent Application Laid-Open No. 2007-186762 can be used. it can. Note that Formula 22-2 may be used instead of Formula 22-1.

Next, the process S14, using the prediction time _{t 0,} the predicted value calculation unit 14 uses the prediction model which has been determined by the processing S13, the prediction time _{t 0} and the in the prediction target data _(X 0, _{t 0)} A predicted value y ₀ (t ₀ ) is calculated for each error parameter α _j (t) obtained in step S13 based on the data values z _{01 to} z _0L of the first to Lth data items, and this calculation is performed. stored in the prediction value _{y 01 (t 0) ~y 0M} (t 0) the predictive value storage unit 43 of the similarity _w 1 to w _M and association with the storage unit 4 described.

Here, in the present embodiment, the calculation of the error parameter α _j (t ₀ ) at the predicted time t ₀ based on the error parameter α _j (t _j ) at the actual measurement time t _j is performed as follows.

In the first calculation mode, when the reference value (reference line) is the upper limit value (upper limit line) and the variation from the reference value is considered to be dependent (caused) by the variation in the initial temperature of the ladle refractory, , The error parameter α _j (t _j ) is mapped to the predicted time t ₀ by Equation 34.

ここで、Ｔ_ｍｊ（ｔ）は、式２２における基準；Ｔ_ｍｊ（ｔ）＝ｆ（Ｚ_ｊ、Θ、ｔ）であり、第ｊ番目の過去実績データの時刻ｔにおける溶鋼温度の基準値（上限値）であり、Ｔ_ｒｊ（０）は、第ｊ番目の過去実績データにおける取鍋耐火物の初期温度（時刻ｔ＝０における温度）である。

Here, T _mj (t) is the reference in Equation 22; T _mj (t) = f (Z _j , Θ, t), and the reference value of the molten steel temperature at time t of the j-th past performance data ( T _rj (0) is the initial temperature (temperature at time t = 0) of the ladle refractory in the j-th past performance data.

In the second calculation mode, the reference value (reference line) is set to the upper limit value (upper limit line), and the error parameter α is used when it is assumed that the variation from the reference value depends on (because) the variation in the molten steel initial temperature. _j (t _j ) is mapped to the predicted time t ₀ by Equation 35.

In the third calculation mode, the reference value (reference line) is set to the upper limit value (upper limit line), and when it is assumed that the variation from the reference value depends on (because) the variation in the heat transfer coefficient, the error parameter α _j (t _j ) is mapped to the predicted time t ₀ by Equation 36.

In the fourth calculation mode, when the reference value (reference line) is set to the upper limit value (upper limit line) and the variation from the reference value is considered to be dependent (caused) by a plurality of events, the error parameter α _j ( t _j ) is mapped to the predicted time t _{0 according} to Equation 37-1 or Equation 37-2 with the degree of influence on the variation in each event as the weight ν _k . Note that the sum of the weights ν _k satisfies a predetermined constant value, and is, for example, 1 as represented by Expression 38.

Here, for example, the equation 37-1 can be expressed by using each error parameter α _j (t ₀ ) at the prediction time t ₀ obtained in each of the first to third calculation modes as α _jk (t ₀ ). It is a weighted average of error parameters α _j (t ₀ ).

For example, Expression 37-2 is _obtained by weighting the error parameter α _j (t _j ) before mapping to the predicted time t ₀ with a weight ν _k in each of the first to third calculation modes. This is the sum of α _jk (t ₀ ) mapped to the predicted time t ₀ and mapped.

That is, the predetermined conversion relation is a plurality of relations, and the error parameter α _j (t ₀ ) is calculated by weighted averaging the conversion results respectively obtained from the plurality of relations. Alternatively, the predetermined conversion relationship is the prediction that the error parameter α _j (t _j ) at the time t _j when the past performance data is acquired based on a plurality of conversion factors is to be predicted for the prediction target data (x ₀ , y ₀ ). a relationship that converts the value of the time t _0, the error parameters α _{j (t 0)} is calculated by performing weighted averaging of the respective conversion results obtained for each of the plurality of conversion factors.

Such a weight ν _k is set in advance in accordance with, for example, the degree of influence on the variation. For example, (as equivalent to dominant variation factors) the degree of influence on the variation according to increases, the greater the weight [nu _k corresponding to the error parameter elements alpha _jk, among error parameters alpha _jk, the variation The weight ν _k corresponding to the most influential parameter element α _jk is _maximized . For example, the weights ν _k may be equal, and the weights ν _k corresponding to the error parameters α _jk may be the same. That is, it is simply averaged. The weighted average includes a simple average when each weight is equal.

Such step S14 the predicted value _y 01 calculated by _{(t 0) ~y 0M (t} 0) , for example, stored in the prediction value storing unit 43 by the table format table format as shown in FIG. 14 The The predicted value data table 53B shown in FIG. 14 corresponds to the predicted value data table 53A shown in FIG. 12 in the second embodiment, and similarly to the predicted value data table 53 shown in FIG. Each field includes a prediction data field 532, an error parameter field 533, and a similarity field 534, and a record is provided for each error parameter α _j . Furthermore, the expected value data table 53B shown in FIG. 14, a prediction time field 535 for registering the predicted time _{t 0.} The error parameter field 533 is provided for each event and a subfield 5331 for registering the error parameter α _jk (t) for each event. The subfield 5331 is for each event and registers the error parameter α (t _j ) at the actual measurement time t _j . (53321 to 5332k).

Next, the process S15, similarly to the basic embodiment, the variation calculation unit 15 uses the respective predicted value _y 01 obtained by the processing _{S14 (t 0) ~y 0M (} t 0), the predicted value _y 0 ( t ₀ ) variation (for example, histogram, probability density distribution, etc.) is calculated, and the calculated predicted value y ₀ (t ₀ ) variation is stored in the variation storage unit 44 of the storage unit 4.

FIG. 15 shows the probability density distribution of the predicted value y ₀ (t ₀ ) of the temperature drop amount every 10 minutes. In other words, the predicted time point is every 10 minutes. FIG. 15 also shows a reference line (upper limit line) f (Z ₀ , Θ, t ₀ ). In FIG. 15, the scale of the horizontal axis of the probability density distribution (corresponding to the horizontal axis of FIG. 7C) is enlarged for easy viewing.

Then, then the processing S16, similarly to the first embodiment and the second embodiment, the reliability index calculating section 16 calculates a plurality of reliability index R for the predicted value y ₀ from a plurality of viewpoints, the calculated The reliability index R is stored in the storage unit 4.

Then, the processing S17, the output value prediction unit Sb, the operation controlling unit 1, the variation of the predicted value _{y 0} calculated by the variation calculation unit 15 in the processing S15, and by reliability index calculating unit 16 in the processing S16 The plurality of calculated reliability indices R are presented to the presentation unit 3, and the process is terminated.

  By operating in this way, the output value prediction device Sb of the third embodiment predicts the molten steel temperature in the ladle of the charge in the process from the converter steelmaking process to the continuous casting process through the molten steel treatment process, It is possible to obtain the predicted dispersion of the molten steel temperature in the ladle with higher accuracy.

More The output value prediction apparatus Sb of this embodiment, like in the first embodiment and the second embodiment, with respect to the predicted value y _0, obtains a plurality of reliability index R from a plurality of viewpoints, which this calculated Reliability index R can be presented. For this reason, the user can evaluate the obtained variation of the predicted value y ₀ from a plurality of viewpoints by referring to the plurality of reliability indexes R presented, and the predicted value y ₀ It is possible to reasonably judge the determination of future actions based on these multiple viewpoints.

  In the third embodiment described above, the predetermined output is the molten steel temperature in the ladle in the process from the converter steelmaking process through the molten steel treatment process to the continuous casting process, but the predetermined output is It may be the molten steel temperature in the tundish in the process from the converter steelmaking process through the molten steel treatment process to the continuous casting process. By configuring in this way, it is possible to predict the molten steel temperature in the tundish in the process from the converter steelmaking process to the continuous casting process through the molten steel treatment process, and obtain the variation of this predicted value It becomes.

  In the third embodiment described above, the predetermined output may be a molten steel component or a molten steel temperature corresponding to the integrated amount of blown blown oxygen in the converter process. By comprising in this way, it becomes possible to predict the molten steel component or molten steel temperature according to the integration amount of blown blowing oxygen in a converter process, and to obtain | require the dispersion | variation in this estimated predicted value.

  In the third embodiment described above, the predetermined output may be the steel material temperature of the steel material in accordance with the heating time or the integrated amount of heat in the steel material heating furnace step. By comprising in this way, it becomes possible to estimate the steel material temperature of the steel material according to the heating time or the integrated amount of the heating heat amount in the heating furnace process of the steel material, and to obtain the variation of the predicted value.

  In the first to third embodiments described above, the past performance data X may be all the actual measurement data generated within a predetermined period, or the actual measurement data may be classified by layer, that is, for example, a product. It may be classified from a predetermined viewpoint such as the component composition of the past, and may be past performance data belonging to the same classification as the prediction target data.

  In the first to third embodiments described above, the distance calculation unit 11 functionally includes a weight calculation unit, and uses the A weight calculated by the weight calculation unit to predict data and past results. The predetermined distance from the data may be calculated for each of the plurality of past performance data based on the factors of the prediction target data and the past performance data.

  The weight calculation unit is configured to calculate the factor as the Ath weight to which the factor contributes to the magnitude of variation in the predetermined output. More specifically, this weight calculation unit is based on a plurality of past performance data when the magnitude of variation in the predetermined output is an A output variable and the variable related to the factor is an A input variable. An A model representing the relationship between the A input variable and the A output variable is generated, and the A weight is calculated based on the A model. The A-th model is represented by, for example, a function equation having two variables or three or more variables (a regression equation such as a single regression equation or a multiple regression equation), and is regressed from the A input variable and the A output variable. Calculated by calculation. Examples of the regression calculation include a least square method and a partial least square method (PLS, Partial Least Square).

  Here, the degree to which the factor contributes to the magnitude of the variation in the predetermined output is large when the factor greatly contributes to the magnitude of the variation in the predetermined output, and the factor is the magnitude of the variation in the predetermined output. If it does not contribute much to, it will be a small value. That is, when the range of variation in the predetermined output (range of the range) is relatively dependent on the value of the factor, the factor greatly contributes to the magnitude of variation in the predetermined output, and the factor is the predetermined output. The magnitude of the influence on the magnitude of variation in the output is large, and if the range of variation in the predetermined output (range) is relatively independent of the factor value, the factor is the magnitude of variation in the predetermined output. In this case, the influence of the factor on the magnitude of the variation in the predetermined output is a small value. In other words, the magnitude of the A-th weight is determined according to the magnitude of the influence exerted on the predetermined output variation by the factor.

  Further, the weight calculation unit may calculate a B-th weight according to the number of past past record data, and generate the A-th model using the B-th weight. With this configuration, it is possible to reduce the influence of the error included in the factor on the Ath weight, and to generate the Ath model with higher accuracy. As a result, the Ath weight with higher accuracy. Can be calculated.

In addition, the weight calculation unit uses the degree to which the factor contributes to the magnitude of variation in the predetermined output and the degree to which the factor contributes to the magnitude of the absolute value in the predetermined output as the A weight. It may be calculated. With this configuration, the Ath weight a _i can be obtained in consideration of the influence of the factor on the variation in the predetermined output and the influence of the factor on the absolute value of the predetermined output. , The accuracy of the variation of the predicted value is improved.

  In the first to third embodiments, the error parameter α is obtained by weighted averaging the respective conversion results obtained from the plurality of relationships, or the respective conversion results obtained for each of the plurality of conversion factors. The error parameter α may be calculated by dividing the obtained difference into a plurality of parts and then mapping them, and then obtaining the sum thereof. Even with such a configuration, even when the predetermined conversion relationship is caused by a plurality of events, the error parameter can be obtained more appropriately.

That is, the first to third weights indicating the degree of influence of the first to third variation factors (factors of the error parameter α) on the error parameter α are represented by ν ₁ (t), ν ₂ (t), and ν ₃ (t). These first to third weights ν ₁ (t), ν ₂ (t), and ν ₃ (t) satisfy Expression 39 for an arbitrary time t.

Here, when the variation α _j (t _j ) at time t _j is separated into variations due to the variation factors ν ₁ (t), ν ₂ (t), and ν ₃ (t), respectively, ν ₁ (t _j ) × α _j (t _j ), ν ₂ (t _j ) × α _j (t _j ), and ν ₃ (t _j ) × α _j (t _j ). In other words, at time t _j , variation occurs at a ratio of ν ₁ (t _j ): ν ₂ (t _j ): ν ₃ (t _j ), and the sum of these _results in variation α _j (t _j ) at time t _j . I think. This is represented by Equation 40.

Here, as in the third embodiment, the molten steel is transferred from the converter to the ladle after the converter blowing in the steel product manufacturing process, and the molten steel is transported to the continuous casting equipment through the molten steel treatment. In the case of the process, first, since the variation in initial temperature of the ladle refractory is ν ₁ (t _j ) × α _j (t _j ) at time t _j , the error parameter α _j1 (t ₀ ) Is mapped to the predicted time t ₀ by Equation 41.

Second, since the variation in molten steel initial temperature (measurement error of molten steel temperature) is ν ₂ (t _j ) × α _j (t _j ) at time t _j , the error parameter α _j2 (t ₀ ) is This is mapped to the predicted time t ₀ by Equation 42.

Third, since the variation in the heat transfer coefficient (the heat transfer coefficient between the molten steel and the ladle refractory) is ν ₃ (t _j ) × α _j (t _j ) at time t _j , the error parameter α _j3 (t ₀ ) is mapped to the predicted time t ₀ by Equation 43.

Here, the equation 43, the time _{t 0} to be mapped ^{_{(~ t y / t j)}} (~ t y is on ^~ in Formula 43 is _{t y)} multiplied by the reference line at the time of the (upper line) It shows that the difference between the value and the value of the reference line (upper limit line) at the time t ₀ to be mapped is the error parameter α _j3 (t ₀ ).

Then, as shown in Equation 44, by summing these equations 41 to Formula 43, the values alpha _{j (t 0)} which is mapped to the time _{t j} variation alpha _{j (t j)} the time _{t 0} is obtained.

  In addition, although illustrated also above, in the above-described embodiment, specific examples of x, X, and Z are as follows. In this example, X is a combination of x and Z.

  In the case of the temperature of a predetermined object accommodated in a predetermined container, x can include the volume of the object (for example, a liquid state), the number of times the container is used, the solidification temperature of the object, the temperature measurement time, and the like. Z is, for example, the initial temperature of the object, the initial temperature of the container, the specific heat / density / volume of the object, the specific heat / density / volume of the container, the contact area between the object and the container, and a heat transfer coefficient calculation value (for example, physical property value). For example).

  In the case of the temperature of the molten steel in the ladle, x is, for example, the number of times the ladle is received, the type of deoxidizer, the concentration of molten steel, the state of the ladle in the empty ladle, the ladle transport time, the outgoing steel temperature, the solidification temperature , Various alloy amounts, steel types, operation groups, treatment days (for example, the number of days elapsed since January 1, 1900), and Z is, for example, ladle type, molten steel carbon concentration, deoxidation Type of agent, ladle time, heat retention time, temperature of outgoing steel, number of received steel in ladle, amount of various alloys, amount of molten steel, molten steel specific heat, molten steel density, molten steel thermal conductivity, ladle refractory temperature, ladle fire resistance Specific heat, ladle refractory density, ladle refractory thermal conductivity, ladle shape and the like can be mentioned.

  In the case of the molten steel temperature in the tundish, x is, for example, the number of times the ladle is received, the type of deoxidizer, the molten steel carbon concentration, the state of the ladle empty, the number of times the tundish is used, and the ladle transport time Steel temperature, solidification temperature, various alloy amounts, steel type, operation team, molten steel treatment type, treatment date (e.g., number of days since January 1, 1900), molten steel treatment time, temperature rise in molten steel treatment, Casting time etc. can be mentioned. Z is, for example, ladle type, molten steel carbon concentration, deoxidizer type, empty ladle time, heat retention time, steel output temperature, number of times the ladle is received, various alloys Amount, molten steel temperature after molten steel treatment, various manipulated variables in molten steel treatment, molten steel amount, molten steel specific heat, molten steel density, molten steel thermal conductivity, ladle refractory temperature, ladle refractory specific heat, ladle refractory density, ladle refractory Material thermal conductivity, ladle shape, tundish refractory temperature, tundish refractory Heat, tundish refractories density, tundish refractory Mononetsu conductivity, tundish shape, ladle transfer time, can be mentioned casting time, and the like.

  Moreover, in the case of the molten steel temperature and molten steel component in converter blowing, x is, for example, the amount of steel output, molten iron temperature, molten iron component (C, Si, Mn, P, S, etc.), blowing target temperature, blowing Target components (target molten steel components, C, Mn, P, S, etc.), hot metal compounding ratio, various auxiliary raw material inputs, alloy inputs, slag basicity, slag amount, acid feed rate, number of furnaces, number of lances, lance height In addition, furnace shutdown time, sublance measurement temperature, sublance measurement component, treatment date (for example, the number of days elapsed since January 1, 1900), number of converters, operation group, pre-charge information, oxygenated amount integrated value, etc. Z is, for example, main raw material (hot metal, cold metal, scrap) input amount, hot metal temperature, hot metal component, various auxiliary material input amounts, various auxiliary material compositions, alloy input amount, acid feed amount, blowing rate Stop target component, sublance measurement temperature, sublance measurement component, pre-charge information A oxygen-flow amount integrated values or the like.

  As described in the first to third embodiments, the output value prediction devices S, Sa, and Sb combine the output value in each process of the operation process and the manufacturing process and the output value of the final process directly connected to the product together with variations. Here, a configuration example of an output value prediction system in which the output value prediction devices S, Sa, and Sb are applied to an operation process and a manufacturing process will be described.

  FIG. 16 is a block diagram showing the configuration of the output value prediction system. In FIG. 16, the output value prediction system includes a performance data collection device 101 that collects performance data from the operation process / manufacturing process 201, and past operation data storage that stores performance data collected by the performance data collection device 101 as past performance data. Device 105, parameter fitting arithmetic device 102 for calculating error parameter α based on past actual data collected by actual data collecting device 101, and parameter estimated value storage device for storing error parameter α calculated by parameter fitting arithmetic device 102 106, a prediction target data operation condition collection device 104 that collects prediction target data from the operation process / manufacturing process 201, a similarity calculation device 108 that calculates a similarity between the prediction target data and each past performance data, and a prediction target Data operating condition collection device The prediction target data output prediction calculation device 103 that predicts the output value (prediction value) of the prediction target data from the prediction target data collected in 104 based on the calculated error parameter α, and the prediction target data output prediction calculation device 103 performs prediction. An output predicted value probability density distribution estimating device 107 that calculates a probability density distribution of the predicted value of the prediction target data based on the output value (predicted value) of the predicted target data and the similarity calculated by the similarity calculating device 108; A reliability index calculation device 109 that calculates a plurality of reliability indexes for the predicted value of the target data, and a probability density distribution and reliability index calculation device of the predicted value of the prediction target data calculated by the output predicted value probability density distribution estimation device 107 And a probability density distribution display device 110 that displays a plurality of reliability indexes calculated in 109.

  When the output value prediction system shown in FIG. 16 is compared with the output value prediction device S shown in FIG. 1, the similarity calculation device 108 is substantially the same as the distance calculation unit 11, the similarity calculation unit 12, and the intermediate data storage unit 42. The parameter fitting calculation device 102 has substantially the same function as the parameter calculation unit 13, and the prediction target data output prediction calculation device 103 is substantially the same as the prediction value calculation unit 14 and the prediction value storage unit 43. The output predicted value probability density distribution estimation device 107 has substantially the same functions as the variation calculation unit 15 and the variation storage unit 44, and the reliability index calculation device 109 is the same as the reliability index calculation unit 16. The past operation data storage device 105 has substantially the same function as the actual data storage unit 41, and the parameter estimated value storage device 106 is an intermediate data storage device. Substantially have the same function as 42.

  When the prediction target data is collected during the execution of the process, the output value prediction system having such a configuration can obtain a prediction value and its probability density distribution in the prediction target data, and can display them. it can. For this reason, a user such as an operator can appropriately adjust the process based on the predicted value in the prediction target data and its probability density distribution, and perform the execution.

  In the first to third embodiments described above, various reliability indexes are calculated in the example, but the prediction model may be any model. For example, the prediction model is a model disclosed in Japanese Patent Laid-Open No. 9-256021, a model disclosed in Japanese Patent Laid-Open No. 2003-231908, a model disclosed in Japanese Patent Laid-Open No. 2007-167858, It may be a model such as a model disclosed in Japanese Patent Application Publication No. 2009-241139.

FIG. 17 is a diagram illustrating a specific example of the reliability index that has been scaled. FIG. 18 is a diagram illustrating another specific example of the scaled reliability index. The horizontal axis in FIGS. 17 and 18 is a charge number as an identifier assigned to a so-called charge, and the vertical axis is a reliability index R. The broken line in (A) shown in FIG. 17 is the conditional reliability index _RC according to Equation 11, the broken line in (B) shown in FIG. 17 is the daily reliability index _RD according to Equation 13 above, The broken line in (C) shown in FIG. 17 is the predicted temporal reliability index _RT according to the above-described equation 26. The broken line in (A) shown in FIG. 18 is the conditional reliability index _RC according to the equation 11, and the broken line in (B) shown in FIG. 18 is the upper conditional daytime reliability index R _{D1 in the} above-described equation 16 and the like. The broken line in (C) shown in FIG. 18 is the upper condition predicted temporal reliability index R _T1 such as Equation 30 described above. The upper fixed amount is a ratio of the upper 10%.

In the first to third embodiments described above, the arithmetic control unit 1 functionally compares a plurality of reliability indices R with respect to the plurality of predicted values y ₀ in order to relatively compare each other. An unillustrated scaling unit (standardization unit) that scales (standardizes) each of the indices R with a predetermined reference is provided, and the presentation unit 3 includes the predicted value y ₀ (the variation in the predicted value y _{0 in} the above example) and the scaling And a plurality of reliability indexes R scaled by the unit may be presented. With such a configuration, each of the plurality of reliability indexes R is scaled (standardized) based on a predetermined reference, so that the plurality of reliability indexes R can be compared with each other. In addition, the scaled reliability index R can be kept within a predetermined range by scaling.

For example, as shown in FIG. 17 and FIG. 18, as the predetermined reference, each reliability is set such that the average value of the reliability index R of the predicted value y ₀ for each charge is 0 and its standard deviation is 1. The sex index R is scaled (standardized). In the example shown in FIGS. 17 and 18, each reliability index R is scaled so that the average value of the reliability index R for each of the 12 charges is 0 and the standard deviation is 1. . That is, for each of the twelve charges, scaling is performed by subtracting the average value from the reliability index R and dividing the result by the standard deviation (scaled reliability index R = (before scaling) Reliability index R-average value) / standard deviation). The predetermined criterion is not limited to this, and each reliability index R may be scaled so that the difference from the minimum value is normalized by the difference between the maximum value and the minimum value. (Reliability index R after scaling = (Reliability index R before scaling−Minimum value) / (Maximum value−Minimum value). By scaling in this way, the reliability index R after scaling is changed from 0 to 1. It can also be within the range.

Then, the numerical value of the reliability index R after scaling may be presented to the presentation unit 3, but by presenting the presentation unit 3 such a graph as in FIG. 17 and FIG. 18, with respect to past predicted value y ₀ this reliability of the predicted value y ₀ Te becomes possible to grasp at a relatively glance. For example, the off track record results for the predicted value y ₀ of the previous, the reliability index R was also low in this case. In this case, if even lower total reliability index R against current predicted value y _0, a user can determine that there is a high possibility that deviates even predicted value y ₀ of the time. Conversely, by applying for example, in the case, the higher the total reliability index R against current predicted value y _0, the user, the predicted value y ₀ of the current can be determined as reliable.

  Alternatively, only the numerical value of the reliability index R of scaling may be presented to the presentation unit 3. Thus, for example, when the numerical value of the reliability index R is 0, the user may determine that the reliability is average among the past limited number of past performance data in the scaling based on the predetermined standard. For example, if the numerical value of the reliability index R is a positive large value, it can be determined that the reliability is relatively high among the past limited number of past performance data. If the numerical value is a large negative value, it can be determined that the reliability is relatively low among the past limited number of past performance data.

For example, in FIG. 17, in each reliability index R in charge number 3, the conditional reliability index R _C (line (A)) according to Expression 11 is relatively large, and the daily reliability index R according to Expression 13 above. _D (broken line (B)) and the predicted temporal reliability index R _T (broken line (C)) according to the above-described equation 26 are relatively small. In other words, this allows the user to historical performance data condition similar to the predicted target data X ₀ (X, y) is relatively large, but less recently and can see that the predicted time is also specific, It can be determined that the overall reliability is low.

Further, for example, in FIG. 17, in each reliability index R at the charge number 12, the conditional reliability index R _C (line (A)) according to Equation 11 and the daytime reliability index R _D (line ( B)) and the predicted temporal reliability index R _T (polygonal line (C)) according to the above-described equation 26 are relatively large. In other words, this allows the user to determine that the overall reliability is high.

Further, for example, in FIG. 18, in each reliability index R at charge number 3, the conditional reliability index R _C (line (A)) according to Expression 11 is relatively large, and the upper condition condition days such as Expression 16 described above are obtained. The reliability index R _D1 (broken line (B)) and the upper condition predicted temporal reliability index R _T1 (broken line (C)) such as Equation 30 described above are relatively small. In other words, the user has a relatively large amount of past performance data (X, y) that is conditionally similar to the prediction target data X ₀ , but the top 10 percent of past performance data (X, y) It can be seen that the estimated time is different from this time, not the data, and the overall reliability can be judged to be low. If the user has the knowledge that “the output (for example, temperature) tends to shift downward in the last month, even under the same manufacturing conditions as before”, the conditional confidence according to Equation 11 even sex index R _{C (broken} line (a)) is relatively large, if a relatively small upper condition days reliability index R _D1 such expression 16 described above, the actual output value with respect to the predicted value y ₀ is lower There is a risk of shifting to the side, and considering this risk larger, it is possible to make correction judgments on manufacturing conditions.

Further, for example, in FIG. 18, in each reliability index R at the charge number 12, the conditional reliability index R _C (line (A)) according to Equation 11 and the upper condition days reliability index R _D1 such as Equation 16 described above. The upper condition predicted time reliability index R _T1 (the broken line (C)) such as (the broken line (B)) and Equation 30 described above is relatively large. In other words, this allows the user to determine that the overall reliability is high.

  In the above-described embodiment, the probability density distribution curve is obtained from the histogram shown in FIG. 7B by the above-described method shown in FIG. 8, but the probability is calculated from the histogram shown in FIG. 7B by the following method. A density distribution curve may be determined.

  FIG. 19 is a diagram for explaining another method for obtaining a probability density distribution curve from the histogram shown in FIG.

In this method, in the histogram shown in FIG. 7B, that is, in the histogram shown in FIG. 19A, first, as shown in FIGS. 19B to _19D, the probability density F _{wN of} each section is _set. For (k), a normal distribution is created in which the sum of the probability densities is the probability density F _wN (k) of the section and the probability density of the section Y _{k to} Y _{k + 1} is the peak.

More specifically, the probability density F _wN (k) of the section shown by hatching in FIG. 19B is the same number of sections before and after the section (FIG. 19 (C)). In the example shown in C), it is assumed that the area has been expanded into two sections. Then, the probability density _~ F _wN (k) of the section after smoothing and the probability density _~ F _wN (k-2), _~ F _wN (k-1), _~ F _wN ( As shown in FIG. _19D , k + 1) and _{˜F wN} (k + 2) are respectively obtained by multiplying the probability density F _wN (k) of the section by a predetermined coefficient λ for smoothing. . In addition, said _-FwN is the same as _FwN which attached _| subjected-in FIG. Also, in FIG. _{19D, “} F _wN (k + i, k + j)” indicates the probability density after smoothing in the k + i-th section with respect to the section k + j.

The predetermined coefficient λ may be arbitrarily determined as long as the sum is 1. Here, the predetermined coefficient λ is set to be a normal distribution and the sum thereof is 1. Here, the normal distribution is used, but another distribution such as a binomial distribution may be used. By setting the sum of the coefficients λ to 1, the sum of the probability density after smoothing becomes equal to the probability density F _wN (k) of the section. For example, in order to obtain the probability density to F _wN (k−2) after smoothing (after normal distribution in this case), the coefficient λ ₋₂ to be multiplied by the probability density F _wN (k) in the section is 0. 02 (λ ₋₂ = 0.02, probability density after normal distribution to F _wN (k−2) = λ ₋₂ × F _wN (k)), and probability density after normal distribution to F _wN ( In order to obtain k−1), the coefficient λ ₋₁ multiplied by the probability density F _wN (k) of the section is set to 0.22 (λ ₋₁ = 0.22, the probability density after normal distribution) _{F wN (k-1) =} λ -1 × F wN (k)), in order to obtain the probability density _{to F} wN after normal distribution of (k), multiplying the probability density _F wN (k) of the section coefficient lambda ₀ is 0.52 and is (lambda ₀ = 0.52, the probability density after the normal distribution of _{~F wN (k) = λ 0} × F wN (k)), normal To determine the probability density _~F wN (k + 1) after Nunoka, coefficients lambda ₊₁ to be multiplied by the probability density _F wN (k) of the section is a 0.22 (λ ₊₁ = 0.22, normal distribution Probability density _FwN (k + 1) = λ _{+ 1} × _FwN (k)) and normal distribution probability density _FwN (k + 2) are obtained to obtain the probability density F _wN ( The coefficient λ ₊₂ to be multiplied by k) is 0.02 (λ ₊₂ = 0.02, probability density after normal distribution to F _wN (k + 2) = λ ₊₂ × F _wN (k)). By determining in this way, the sum of the coefficients λ becomes λ ₋₂ + λ ₋₁ + λ ₀ + λ ₊₁ + λ ₊₂ = 1. In this way, the probability density F _wN (k) of the section is expanded in a normal distribution form.

When the probability density F _wN (k) of each section is expanded to a normal distribution, each probability density after normal distribution is added for each section (Equation 45), and the smoothing shown in FIG. A probability density distribution is obtained. A smoothed probability density distribution curve is obtained by connecting the center positions of the respective sections with a curve. In this example, since the probability density after smoothing is expanded by two in the preceding and following sections, both ends of the entire distribution are expanded by two sections. For this reason, at each both ends, a point separated from the end by a half (h / 2) of the width h of the section is also connected to the curve as 0.

ここで、式４５の右辺の〜Ｆ_ｗＮ（ｋ＋ｉ、ｋ＋ｊ）において、定義されていない値〜Ｆ_ｗＮ（ｋ＋ｉ、ｋ＋ｊ）は、０とする。

Here, in _~ F _wN (k + i, k + j) on the right side of Expression 45, an _undefined value _~ F _wN (k + i, k + j) is 0.

  With such a technique, a probability density distribution curve may be obtained from the histogram shown in FIG.

  In order to express the present invention, the present invention has been properly and fully described through the embodiments with reference to the drawings. However, those skilled in the art can easily change and / or improve the above-described embodiments. It should be recognized that this is possible. Therefore, unless the modifications or improvements implemented by those skilled in the art are at a level that departs from the scope of the claims recited in the claims, the modifications or improvements are not covered by the claims. To be construed as inclusive.

S, Sa Output value prediction device 1 Calculation control unit 4 Storage unit 11 Distance calculation unit 12 Similarity calculation unit 13 Parameter calculation unit 14 Predicted value calculation unit 15 Variation calculation unit 16 Reliability index calculation unit 41 Actual measurement data storage unit 42 Intermediate data Storage unit 43 Predicted value storage unit 44 Variation storage unit 102 Parameter fitting calculation unit 103 Prediction target data output prediction calculation unit 105 Past operation data storage unit 107 Output predicted value probability density distribution estimation unit 108 Similarity calculation unit 161 Conditional reliability index Calculation unit 162 Daily reliability index calculation unit 163 Predictive time reliability index calculation unit

Claims (11)

In an output value predicting device that predicts an output value from the given data by using the given data for a given prediction model and obtains a predicted value,
A reliability index calculation unit for obtaining a plurality of reliability indices from a plurality of viewpoints, which is an index representing the degree of confidence in the predicted value;
A presentation unit that presents the predicted value and a plurality of reliability indexes obtained by the reliability index calculation unit;
The given prediction model is a model obtained based on past performance data,
The reliability index calculation unit, as one of the plurality of reliability indexes, includes conditional data that is data of a predetermined data item that affects the output value among data items in the past performance data, and the predetermined data item. An output value prediction apparatus characterized in that it obtains at least a conditional reliability index obtained based on a conditional similarity that is a similarity to given data. The reliability index calculation unit further includes, as one of the plurality of reliability indices, an acquisition date when acquiring past performance data used when obtaining the given prediction model, and a prediction in the given data The output value prediction apparatus according to claim 1, wherein a day reliability index obtained based on a day degree similarity that is a degree of similarity with a day is obtained. The given prediction model is a model that is obtained based on past performance data and includes an elapsed time from a predetermined reference time as an input variable;
The reliability index calculation unit further includes, as one of the plurality of reliability indices, an acquisition date when acquiring past performance data used when obtaining the given prediction model, and a prediction in the given data The daily reliability index obtained based on the daily similarity that is the degree of similarity to the day, the acquisition time when the past performance data used when obtaining the given prediction model, and the given 2. The output value prediction according to claim 1, wherein at least one of the predicted temporal reliability indexes obtained based on the predicted temporal similarity that is the similarity to the predicted time in the data of apparatus. The conditional reliability index includes an average value of the conditional similarity, a function of the average value of the conditional similarity, the number of past performance data in which the conditional similarity is a predetermined threshold or more, and the conditional similarity A function of the number of past performance data whose degree is equal to or greater than a predetermined threshold, an average value of a product of the conditional similarity and the daily similarity, and an average of a product of the conditional similarity and the daily similarity The output value prediction apparatus according to claim 2 , wherein the output value prediction apparatus is any one or a plurality of value functions. In the case where the daily reliability index is included as one of the plurality of reliability indices, the daily reliability index includes an average value of the daily similarity, a function of the average value of the daily similarity, The average value of the day-to-day similarity with respect to the past performance data whose conditional similarity is within the upper fixed amount, the function of the average value of the day-to-day similarity with respect to the past performance data whose conditional similarity is within the upper certain amount, An average value of conditional similarity for past performance data whose daily similarity is within the upper certain amount, a function of an average value of conditional similarity for past performance data whose daily similarity is within the upper certain amount, The output value prediction apparatus according to claim 2, wherein the output value prediction apparatus is any one or more of them. In the case where the predicted temporal reliability index is included as one of the plurality of reliability indices, the predicted temporal reliability index includes an average value of the predicted temporal similarity and an average value of the predicted temporal similarity A function of the conditional similarity and the predicted temporal similarity, a function of a product of the conditional similarity and the predicted temporal similarity, the conditional similarity and the daily similarity and the Prediction with respect to past performance data in which the conditional similarity is a product of the conditional similarity, a function of the product of the conditional similarity, the daily similarity, and the predicted temporal similarity, and the conditional similarity is within the upper fixed amount Average value of temporal similarity, function of average value of predicted temporal similarity with respect to past performance data in which the conditional similarity is within the upper certain amount, past performance in which the predicted temporal similarity is within the upper certain amount The average conditional similarity to the data, The output according to claim 3, wherein the output is a function of an average value of conditional similarity with respect to past performance data whose predicted temporal similarity is within the upper fixed amount. Value prediction device. A scaling unit that scales each of the plurality of reliability indexes obtained by the reliability index calculation unit according to a predetermined reference;
The output value prediction apparatus according to any one of claims 1 to 6, wherein the presenting unit presents the predicted value and a plurality of reliability indexes scaled by the scaling unit. The given prediction model is a model in which a molten steel temperature in a ladle or a tundish is modeled in a process from a converter steelmaking process to a continuous casting process through a molten steel treatment process. The output value prediction apparatus according to any one of claims 1 to 7. The given prediction model is a model in which a molten steel component or a molten steel temperature is modeled according to an integrated amount of blown blown oxygen in a converter process. The output value prediction apparatus according to claim 1. An output value prediction step of predicting an output value from the given data by using the given data for a given prediction model, and obtaining a predicted value;
A reliability index calculation step for obtaining a plurality of reliability indices from a plurality of viewpoints, which is an index representing a degree of confidence in the predicted value;
A presentation step of presenting the predicted value and a plurality of reliability indexes obtained in the reliability index calculation step,
The given prediction model is a model obtained based on past performance data,
The reliability index calculating step includes conditional data that is data of a predetermined data item that affects the output value of data items in the past performance data as one of the plurality of reliability indexes, and the predetermined data item. An output value prediction method characterized in that at least a conditional reliability index obtained based on a conditional similarity that is a similarity with given data is obtained. An output value prediction program for causing a computer to execute,
An output value prediction step of predicting an output value from the given data by using the given data for a given prediction model, and obtaining a predicted value;
A reliability index calculation step for obtaining a plurality of reliability indices from a plurality of viewpoints, which is an index representing a degree of confidence in the predicted value;
A presentation step of presenting the predicted value and a plurality of reliability indexes obtained in the reliability index calculation step,
The given prediction model is a model obtained based on past performance data,
The reliability index calculating step includes conditional data that is data of a predetermined data item that affects the output value of data items in the past performance data as one of the plurality of reliability indexes, and the predetermined data item. An output value prediction program characterized by at least obtaining a conditional reliability index obtained based on a conditional similarity that is a similarity to given data.

Images