JP5416811B2 - Prediction device, prediction method, and computer program - Google Patents

Description

  The present invention relates to a prediction object that changes in time series and a method for analyzing a plurality of items related to the prediction object, and in particular, a prediction device, a prediction method, and a computer that can improve the accuracy of prediction results. The present invention relates to a computer program that functions as a computer program.

  Economic forecasts such as product demand forecasts and sales forecasts are extremely important in examining the direction and strategy in company management. Then, how to connect the predicted demand to the plans of each department such as sales, inventory, production, distribution, and development is a management issue. Furthermore, not only forecasts for economic events such as demand forecasts and sales forecasts, but forecasting future events using information that has changed over time is an important issue in various fields. .

  Various time-series analysis methods have been proposed as methods for predicting subsequent changes in time-series events such as the number of products demanded or the number of products sold. Examples of such an analysis method include multivariate analysis such as multiple regression analysis and T method (for example, Patent Document 1 and Non-Patent Document 1). Furthermore, various application proposals have been proposed for these analysis methods (Non-Patent Document 2, etc.).

Japanese Patent No. 3141164

Edited by Kazuo Tatebayashi, Shoichi Teshima, Ryoko Hasegawa, “Introductory MT System”, Nikka Giren Shuppansha, December 2008 Yukiya Masuda, “Research of T Method Considering Nonlinear Components”, Proceedings of 17th Quality Engineering Research Conference, p. 422-425 2009

  Among various time series analysis methods, when the analysis is performed using the T method, how to select items used for the analysis is a problem. Regarding the selection of items, a method called a two-sided T method disclosed in Non-Patent Document 1 has been proposed. However, the inventor considered that the item selection by the two-sided T method is insufficient in prediction accuracy, and studied a method for selecting an item in which the total estimated SN ratio of Non-Patent Document 2 is maximized. As a result, we obtained the knowledge that the target period for item selection should be selected in consideration of the trend of the past events over time.

  The present invention has been made based on such knowledge, and an object of the present invention is to provide a prediction device, a prediction method, and a computer program that further improve the prediction accuracy by reflecting the tendency of change over time.

  The prediction device according to the present invention records a value related to an event related to a plurality of factors, and records the plurality of factors related to the value in time series in association with the values as data for a plurality of items. And means for reading out the value and the data for each of the plurality of items corresponding to the value from the recording means, the calculating means correlating the change of the values of the plurality of items. In the prediction device that selects a plurality of items from the plurality of items based on the factor effect value indicating the strength of the item, and predicts the value related to the event in the prediction target period based on the data of the selected item, the calculation A means for identifying a time-dependent change tendency during a recording period of the value recorded in the recording means, and a means similar to the time-dependent change tendency of the value in a corresponding period before the prediction target period. A period selection means for selecting the period from the recording period, data for each of the plurality of items in the selection period selected by the period selection means, and the factor obtained by a value after a predetermined period from the selection period An item selection unit that selects a plurality of items based on an effect value; and a prediction unit that calculates a predicted value in the prediction target period from data of the item selected by the item selection unit in the corresponding period. To do.

  In the prediction apparatus according to the present invention, the period selection unit includes a determination unit that determines a gradient of a change with time in the corresponding period, and a first period in which the value changes with the gradient determined by the determination unit. Means for selecting from a period; first calculation means for calculating a numerical value representing similarity between the data for each of the plurality of items in the corresponding period; and the data for each of the plurality of items in the first period; Second calculation means for calculating a numerical value representing a change tendency of the calculated numerical value, and a second period having a high similarity in the first period based on a comparison between the numerical value calculated by the second calculation means and a threshold value And selecting means for selecting.

  In the prediction apparatus according to the present invention, the first calculating means calculates the Mahalanobis generalized distance, and the second calculating means uses the Mahalanobis generalized distance trend calculated by the first calculating means based on the trend model. The selection means selects the period in which the trend calculated by the second calculation means is equal to or less than a predetermined value as the second period.

  In the prediction apparatus according to the present invention, the selection unit selects the second period by changing the length, and the value after a predetermined time of the data for each item in the second period is selected for each length. Further comprising a determining means for determining a length that maximizes a value related to the prediction accuracy for, wherein the period selecting means selects a second period of the length determined by the determining means. To do.

  The prediction method according to the present invention records a value related to an event related to a plurality of factors, and records the plurality of factors related to the value in time series in association with the value as data for a plurality of items. Means for reading the value and the data for each of the plurality of items corresponding to the value from the recording means, and for calculating the value of the plurality of items. In a prediction method for selecting a plurality of items from the plurality of items based on a factor effect value indicating the strength of correlation to change, and predicting a value related to the event in a prediction target period based on data of the selected items The calculation means identifies a time-dependent change tendency of the value recorded in the recording means during the recording period, and is similar to the time-dependent change tendency of the value in a corresponding period before the prediction target period. A period is selected from the recording period, and a plurality of items are selected based on the data for each of the plurality of items in the selected selection period and the factor effect value obtained by a value after a predetermined period from the selection period. And the predicted value in the said prediction object period is calculated from the data of the some item selected in the said corresponding | compatible period, It is characterized by the above-mentioned.

  A computer program according to the present invention includes a computer having recording means, a plurality of values related to an event related to a plurality of factors recorded in time series in the recording means, and a plurality of the factors related to the values. Based on the factor effect value indicating the strength of correlation with respect to the change in the value of the plurality of items, and selecting a plurality of items from the plurality of items, and based on the data of the selected item In the computer program for predicting the value related to the event in the prediction target period, the computer has a specifying means for specifying a tendency to change over time during the recording period of the value recorded in the recording means, than the prediction target period. A period selecting means for selecting a latest period similar to the time-dependent change trend of the value in the corresponding period before the predetermined period from the recording period; Item selection means for selecting a plurality of items based on the data for each of the plurality of items in the selection period and the factor effect value obtained by a value after a predetermined period from the selection period, and selection in the corresponding period The function of the prediction means which calculates the predicted value in the said prediction object period from the data of the some item made | formed is characterized.

  In the present invention, the value and the data for each item are read out from the recording means in which the value relating to the event that is time-series data and the data for each of the plurality of items that are a plurality of factors related to the value are recorded. The time-dependent change tendency of the time series value is specified. Based on the time-dependent change tendency of the time series value, a corresponding period before the prediction target period, that is, a period similar to the time change tendency immediately before the prediction time point and close to the prediction target period is selected. A relational expression between the data for each of the plurality of items in the selected period and the value after the predetermined period is specified, and a plurality of items having a strong correlation with the change in the value are selected, and among the data for each item in the corresponding period Based on the data of the selected item, the value in the prediction target period is predicted.

  In the present invention, when selecting a period, a period similar to the change tendency of the signal value before the prediction target period is specified based on the slope of the change tendency of the value, and among the periods having the same slope, the item A period in which the time-series tendency of each data is similar is selected. In particular, since a similar period that is as close as possible to the prediction target period is selected, it is possible to perform a prediction that reflects the tendency of data for each item that is a sign.

  In the present invention, Mahalanobis's general distance is used as a numerical value representing the similarity of data for each item. Further, the Mahalanobis pan-distance trend is used as a numerical value representing a time-series transition of data for each item. This makes it possible to select objectively similar periods using a relatively easy method, and to improve prediction accuracy.

  In the present invention, the length with the highest prediction accuracy is selected by changing the length of the selection period. Thereby, it is possible to improve the prediction accuracy.

  In the present invention, when selecting an item to be used for analysis, a signal value at a prediction time point and a signal value after a predetermined time are selected from the period in which the change tendency of the item data is determined to be similar in the signal period. Is specified, and prediction is performed from the relational expression. That is, the prediction is performed based on the data at the time when the temporal change with high similarity is made from the past event. As a result, it is possible to eliminate the influence due to the time-series data having a temporal change different from the prediction time point and improve the prediction accuracy.

It is explanatory drawing which shows the content example of the signal value and item data which are used by T method. It is explanatory drawing which shows an example of the proportionality constant (beta) and SN ratio (eta) (square ratio) of each item used by T method. It is the figure which showed the actual value of each member's actual signal value or the actual value of the standardized signal value, and the comprehensive estimated value calculated | required about each member in list format. It is a factor effect figure which shows the S / N ratio of the comprehensive estimated value of each item in T method. It is a graph which shows the influence with respect to the factor effect of each item in T method. It is explanatory drawing which shows a time difference model notionally. It is a graph which shows the relationship between the signal value before and behind conversion processing, and the value of the data for every item. It is a graph which shows notionally the processing for determining the optimal number of item selection. It is a graph which shows a response | compatibility with the number of item selections, and SN ratio of a comprehensive estimated value. It is a block diagram which shows the structure of the prediction apparatus in this Embodiment. It is a flowchart which shows an example of the prediction process procedure by the prediction apparatus in this Embodiment. It is a flowchart which shows an example of the signal period selection processing procedure by the prediction apparatus in this Embodiment. It is explanatory drawing which shows the content of the item in a present Example. It is a graph which shows the example of the content of the signal value used for prediction in a present Example. It is a graph which shows the trend of the signal value calculated | required by the control part of the prediction apparatus. It is a graph which shows MD during the period selected by the control part of a prediction device, and the period concerned. It is a graph which shows the prediction result by the control part of the prediction apparatus in a present Example. It is a graph which shows the predicted value at the time of making a signal period into the latest 2 years. It is a graph which shows the predicted value at the time of making a signal period into the latest one year. It is a graph which shows the comparison of the SN ratio of the comprehensive estimated value by the difference in the selection method of a signal period.

  Embodiments of the present invention will be specifically described below with reference to the drawings.

The prediction method disclosed below is an invention called a “time difference model” for predicting changes over time of a signal (value related to an event) based on an MT system that has been used in the past, in particular, a method using the T method. Is added to improve the prediction accuracy, and from the past signals and items (factors) corresponding to the signals, a “method for appropriately selecting signals and items” used for prediction is added to improve the prediction accuracy. is there. In addition, the following prediction methods are implemented by calculation means, such as a computer (prediction apparatus 1 mentioned later) which can read each data from the recording means which has recorded the data used for prediction.
In particular, the following discloses a “signal and item selection method”. However, the techniques already disclosed above, such as the T method and the time difference model, are the techniques and inventions premised on the selection method, and are intended to assist in understanding the signal and item selection method of the present invention. Since it is necessary, it will be described below.

First, an outline of the T method will be described. FIG. 1 is an explanatory diagram showing an example of the contents of signal values and item data used in the T method. In FIG. 1, a member is an index indicating data for each unit period, and is indicated as 1, 2,. The data value itself is recorded as a signal value M. "Item 1", "item 2", ..., "item k" is an item which is a factor associated with the signal value _{M, X 11, X 12,} ..., X 1k is, the signal value M ₁ of the member 1 The data of each item related to is shown. For example, if the member has monthly electricity usage, the monthly electricity usage (W) is the signal value M, and the items are “month”, “temperature”, “wind speed”, “precipitation (monthly average) ) ”,“ Sunshine hours (monthly average) ”,“ maximum temperature (monthly average) ”,“ minimum temperature (monthly average) ”, and so on. The item data in this case is the value of each item, for example, data indicating what month it is, the monthly average value of temperature, the monthly average value of wind speed, and the like.

  Actual values may be used as the signal value M and the data X of each item, but it is preferable to use them after normalization using a calculation means. In the normalization method, for example, the calculation means subtracts the average value of the data of the item (the average value of the monthly average temperature of all members to be used) from the data X of each item (for example, the monthly average temperature). There are methods. Thus, when plotting the data value of the item against the signal value (referred to as a unit space), it is possible to express the prediction formula as a straight line passing through the origin.

  In the prediction method, a calculation formula is used to obtain a specific expression for specifying the relationship between the signal value M of each member and the related multivariable item data, and the data of the next unit period item to be predicted is predicted. Then, the predicted item data is applied to a specific formula for prediction. However, the item data is multivariate. For example, there may be 12 members and there are 20 items (k = 20). In such a case, it is difficult to obtain a specific expression for predicting the next signal value. In the T method, for each of a large number of items, a value indicating the strength of the factor effect with respect to signal value fluctuation is evaluated, and a value indicating the factor effect is weighted. Thereby, by selecting effective items, it is possible to perform prediction with sufficiently high accuracy by using the selected items without using data of all items.

  Therefore, in the T method, the calculation means calculates the proportionality constant β and the SN ratio η (square ratio) for each item by applying the following formulas 1 and 2 using the member having the signal value M. . The S / N ratio is a value indicated by using the reciprocal of dispersion as shown in the following formula 2, is the sensitivity of the signal value for each item, and indicates the strength of correlation between each item and the signal value.

  In addition, the above-mentioned Formula 1 and Formula 2 are formulas for obtaining the proportionality constant β and the SN ratio η (square ratio) for Item 1. The calculation means performs the same calculation as for item 1 for item 2 to item k. FIG. 2 is an explanatory diagram showing an example of the proportionality constant β and the SN ratio η (square ratio) of each item used in the T method. In FIG. 2, the proportionality constant β and SN ratio η (square ratio) for each item calculated by applying Equation 1 and Equation 2 described above to each item are shown in a table format.

  Next, using the proportionality constant β and the SN ratio η (square ratio) for each item, an estimated value of the output for each item is obtained for each member by the computing means. For the i-th member, the estimated output value of item 1 can be expressed by the following equation 3. Similarly, the estimated value of the output from item 2 to item k is obtained by the calculation means.

  Thereby, the calculating means can derive the comprehensive estimation formula (Formula 4) as a prediction formula indicating the relationship between the item data and the total estimated value of the signal value. However, the comprehensive estimation formula using all items (1 to k) with respect to the signal value to be predicted does not have the highest prediction accuracy. Therefore, in order to increase the contribution to the influence on the prediction target and increase the prediction accuracy, an appropriate combination of items is selected from all items by the calculation means.

Therefore, the calculation means calculates an overall estimated value using the SN ratios η ₁ , η ₂ ,..., Η _k (square ratio), which is the estimation accuracy for the estimated value of each item, as a weighting coefficient. Therefore, the total estimated value of the i-th member can be expressed by the following Equation 4.

  FIG. 3 is a table showing the actual signal value of each member or the actual value of the standardized signal value and the total estimated value obtained for each member in a list format. Then, using the actual value and the total estimated value of each member obtained as shown in FIG. 3, the computing means calculates the SN ratio η (db) of the total estimated value for each item by the following equation 5. .

  In the T method, a method of obtaining a value called a factor effect value by using the total estimated value obtained for each item as described above, and selecting an item based on the factor effect value is used. FIG. 4 is a factor effect diagram showing the SN ratio of the overall estimated value of each item in the T method, and FIG. 5 is a graph showing the influence of each item on the factor effect in the T method.

In FIG. 4, the horizontal axis indicates items to be selected, and the vertical axis indicates the SN ratio of each item by taking the SN ratio of the comprehensive estimated value. In FIG. 4, for each item, the SN ratio of the total estimated value for each data including the data of the item, that is, the strength of the correlation with the signal value is shown on the left side, and each data excluding the data of the item is shown on the right side. The SN ratio of the comprehensive estimated value with respect to is shown. In the example shown in FIG. 4, there are 36 items, and there are 2 ³⁶ −1 combinations of selections. The calculation means derives the SN ratio of the prediction target for one or a plurality of items for each of these combinations. And a calculating means calculates the average value of SN ratio of the combination containing the said item used as object, and the average value of S / N ratio of the combination which does not contain the said item. In FIG. 4, the average value of the SN ratio including the data of the item calculated in this way is shown on the left side, and the average value of the SN ratio not including the item is shown on the right side for each item.

  In FIG. 5, the horizontal axis indicates items to be selected, and the vertical axis indicates the factor effect value, and the degree of the factor effect value for each item is indicated. The factor effect value on the vertical axis in FIG. 5 is the left side of the SN ratio on the right side (excluding items) of the SN ratio of the overall estimated value of each item indicating the strength of the correlation of the prediction target in FIG. Of the SN ratio (including items), that is, a value obtained by subtracting the right SN ratio from the left SN ratio. That is, for each item, the degree of effect on the S / N ratio when the item is included is not shown. Therefore, an item for which the calculated factor effect value is positive indicates that the SN ratio of the comprehensive estimated value is increased by using the item. In a method called a two-sided T method, only items having such positive factor effect values are selected, and an analysis by the T method as described above is performed.

  When prediction is performed using the T method (two-sided T method) as described above, the actual prediction accuracy still does not increase. Therefore, the inventor first applies the concept of a time difference model to the above-described T method, performs conversion processing in consideration of the nonlinearity between the signal value and the value of the item data, and further, as the item selection method, the factor effect value Instead of the method of selecting by positive / negative of the method, the method of maximizing the SN ratio of the comprehensive estimated value is used.

First, the time difference model will be described.
The T method is not originally a method for prediction of time series data. There is no particular concept of the time axis, and it is a method for selecting which factor has the most influence on the value of an event related to a large number of factors. When the T method is used as it is for the prediction method, as described above, the relationship between the data for each item of each member and the signal value is specified, and the signal value of the next member is predicted. Therefore, in order to predict the signal value of the next member, data for each item of the next member must be predicted. It is easily assumed that the estimation error of the data of each item at this time overlaps and the prediction accuracy of the signal value is deteriorated. In addition, when predicting the next signal value based on time-series data, especially in fields such as economic forecasting or sales forecasting, there should be signs of future events. is there. Therefore, the inventor decided to associate the association between the signal value and the data for each item as described above not with the members of the same period but with the ones shifted by a predetermined time. Specifically, the computing means that implements the prediction method of the present embodiment associates the signal value after a predetermined period with the data for each item in a certain period, and calculates Equations 1 to 5 between them. Then, the signal value after a predetermined period is estimated (predicted) based on the data for each item in the current (most recent) period using the comprehensive estimation formula obtained as a result.

FIG. 6 is an explanatory diagram conceptually showing the time difference model. FIG. 6 shows the passage of time from right to left. Each rectangle at the bottom of FIG. 6 represents a signal for each unit period, and each rectangle at the top represents data of each item for each unit time at the same time as the signal. For example, a signal for each month as a unit period and data of items for each month are shown. There are a plurality of items for each signal. The time difference model, rather than associating the data signal and the item for the same period, with respect to signal values M _i, the data X _{it 1} of a predetermined period before _{entry, X it 2, ..., X} it k ( e.g. t = 4 months). In the prediction method of the present embodiment, the calculation means applies the signal value and item data to which the time difference model is applied to M ₁ in the above (Equation 1 to Equation 5) and the data X ₁₁ , X ₁₂ ,. The T method is applied by applying the relationship of X _1k . The signal value of the prediction target period is applied to Equation 4 using the value of the data of the item before the predetermined period, that is, the value of the data of each item in the corresponding period (here, four months) immediately before the prediction target period. Is required. In other words, in the time difference model in the prediction method of the present embodiment, the calculation means specifies the relationship between the data of a plurality of items and the signal value after a predetermined period, so that the future signal can be obtained from the past or current item data. It is possible to predict the value.

In the prediction method according to the present embodiment, the signal value X is subjected to conversion processing in consideration of nonlinearity. FIG. 7 is a graph showing the relationship between signal values before and after the conversion process and data values for each item. The left side of FIG. 7 shows the relationship of signal values to data of a certain item before conversion processing, and the right side shows the relationship of signal values to data of a certain item after conversion processing. In the T method, for each item, a straight line passing through the zero point is set for the relationship between the item data and the signal value, and weighting based on the deviation from this straight line (quantified as an SN ratio) is performed. That is, the relationship on the right side of FIG. 7 is assumed. Items with less variation with respect to a straight line are items that contribute to changes in the signal value.There are more variations, that is, items with no correlation between the change in the data value of the item and the change in the signal value, the change in the signal value. This item does not contribute. However, it is natural that the data of all items does not necessarily have a straight line, that is, a linear relationship in relation to the signal value. Among them, as shown on the left side of FIG. 7, there is an item indicating a strong relationship indicated by a quadratic function that does not pass through the origin. Therefore, in the prediction method according to the present embodiment, the calculation means uses the item data X _ij (i: number of members (1 to 1), j: number of items (1 to k)) as a non-linear relationship (quadratic function). ) _Is linearly converted into x _ij indicating a linear relationship.

Specifically, the arithmetic means, as the unit space data as a reference between the items of data X _ij (1 to L) and the signal value M _ij (1 to L), the data X _ij and the signal values M _ij Find the average value. The calculation means performs a normalization process for subtracting the average value in the unit space from each value of the signal value M _ij and the item data X _ij . Then, the calculation means approximates with a polynomial such as a quadratic expression using the value of the item data X _ij that changes with respect to the signal value M _ij . Next, the computing means converts the value of X _ij (1 to 1) into x _ij using this approximated value. If the relationship between the signal value and the item data is linear, although data X _ij where the item data value X _ij (1 to L) is directly or normalized (1 to L) is applied to Equation 4 In the case of non-linearity, the value of x _ij (1 to 1) after conversion is applied to Equation 4.

  In addition, in the prediction method according to the present embodiment, as a method of item selection, a response after applying the above-described time difference model to the signal value and the data for each item, and further performing a conversion process in consideration of nonlinearity Based on the data and signal value for each item to be calculated, the calculation means calculates the SN ratio of the comprehensive estimated value and selects an item that maximizes the SN ratio of the comprehensive estimated value. FIG. 8 is a graph conceptually showing a process for determining the optimum number of item selections. First, the S / N ratio of the comprehensive estimated value of each item is calculated by applying Equations 1 to 5 for the signal value and the data for each item that have been subjected to the linear transformation process using the time difference model. Then, as described with reference to FIGS. 4 and 5, the calculation means calculates a factor effect value for each item from the SN ratio of the comprehensive estimated value. As shown in FIG. 8, the computing means first sets the minimum value of the factor effect value for each item as an initial threshold value. The computing means selects an item whose factor effect value is equal to or greater than a threshold value, and calculates the SN ratio of the comprehensive estimated value with respect to the signal value (for example, all 1 to 1). If the threshold is the initial value, all items are selected. Next, the calculation means sets a value obtained by adding a predetermined value to the threshold value as the next threshold value, similarly selects an item whose factor effect value is equal to or greater than the threshold value, and calculates the SN ratio of the comprehensive estimated value with respect to the signal value. . The calculation means repeats such processing until the threshold reaches the maximum factor effect value, that is, until the horizontal line shown as MAX in FIG. 8 is reached, whereby the SN ratio of the overall estimated value for the data of the selected items is selected. Can be calculated for each number of items. The calculation means sets the initial value of the threshold value to the maximum value MAX or more, decreases the threshold value by a predetermined value, selects an item that is a factor effect value above each threshold value, and calculates the SN ratio of the comprehensive estimated value. You may make it go.

  FIG. 9 is a graph showing the correspondence between the number of item selections and the SN ratio of the overall estimated value. The SN ratio of the comprehensive estimated value with respect to the number of items when an item is selected by the method shown in FIG. In FIG. 9, the transition indicated by the black circle is the SN ratio of the comprehensive estimated value when linear conversion is performed, and the transition indicated by the black square is the SN ratio of the comprehensive estimated value when linear conversion is not performed. As shown in FIG. 9, the SN ratio of the overall estimated value as a whole is higher when linear conversion is performed. That is, the accuracy of prediction is high. In addition, among the S / N ratio values, the white square is the S / N ratio of the overall estimated value when an item is selected using the two-sided T method in which an item having a positive factorial effect value is selected. Open diamonds indicate the SN ratio of the overall estimated value when all items are selected. White circles indicate the number of items with the maximum total SN ratio. As shown in FIG. 9, in the method of selecting an item having a positive factor effect value, the SN ratio of the comprehensive estimated value is not maximized when nonlinear conversion is performed. As a matter of course, as shown in FIG. 9, when the calculation means performs the method shown in FIG. 8, the item having the highest SN ratio of the total estimated value can be optimally selected.

  As described above, the inventor of the present application applies a time difference model to the T method, calculates the SN ratio of the overall estimated value after performing a conversion process considering the nonlinearity between the signal value and the item data, The present inventors have obtained the knowledge that the prediction accuracy can be improved by applying a method for selecting an item that maximizes the S / N ratio. Furthermore, in addition to the method such as the time difference model described above, the inventor of the present application is suitable for prediction from the recorded signal period (members 1 to 1) when selecting an item that maximizes the SN ratio of the overall estimated value. The knowledge that the prediction accuracy can be improved by selecting the signal period was obtained. The signal period suitable for the prediction is selected by the calculation means based on the signal trend and the item similarity trend. Hereinafter, a method of selecting a signal period that can improve the prediction accuracy will be specifically disclosed.

  First, the structure of the prediction apparatus which implements the whole prediction method including the method demonstrated above is demonstrated. FIG. 10 is a block diagram illustrating a configuration of the prediction device 1 according to the present embodiment. The prediction device 1 uses a computer such as a personal computer or a server computer. The prediction device 1 includes a control unit 10, a recording unit 11, a temporary storage unit 12, an input unit 13, and an output unit 14.

  The control unit 10 uses a CPU (Central Processing Unit). The control part 10 controls a personal computer or a server computer based on the prediction program 2 demonstrated below, and exhibits the function as the prediction apparatus 1 in this Embodiment. In addition, the control part 10 exhibits the function of the calculating means which implements the prediction method mentioned above.

  The recording unit 11 uses a nonvolatile memory such as a ROM (Read Only Memory) or a hard disk drive. The recording unit 11 may be an external hard disk drive, an optical disk drive, or another recording device connected via a communication network. That is, the recording unit 11 is a general term for one or a plurality of information recording media accessible from the control unit 10.

  The recording unit 11 records a prediction program 2 including various procedures for realizing the prediction method of the present embodiment. A part of the recording area of the recording unit 11 is used as a database (DB) 110 that records signal values and data of a plurality of items corresponding to the values. The control unit 10 can read and write signal values and data of a plurality of items with respect to the database 110. The database 110 records the signal value of each member and data for each item in time series, for example, in the format shown in FIG.

  The temporary storage unit 12 uses a nonvolatile memory such as DRAM (Dynamic Random Access Memory) or SRAM (Static Random Access Memory). The temporary storage unit 12 temporarily stores information generated by the processing of the control unit 10.

  The input unit 13 receives a user operation input using a keyboard, a mouse, or the like.

  The output unit 14 uses a display unit such as a liquid crystal monitor or a printing unit such as a printer, and outputs information processing results by the control unit 10.

  In the prediction apparatus 1 configured as described above, the control unit 10 executes a process based on the prediction program 2 to predict a signal value related to a future event.

  FIG. 11 is a flowchart illustrating an example of a prediction processing procedure performed by the prediction device 1 according to the present embodiment.

  The control unit 10 receives an input from the input unit 13 regarding the signal value and data of a plurality of items related thereto, and records the received signal value and data for each item in the database 110 of the recording unit 11 (step S101). . The signal values and data for each item to be recorded in the database 110 may be input not only from the input unit 13 but also from other devices via a communication network, or from other information recording media. It may be entered.

  The control unit 10 generates a time difference model based on the signal value recorded in the database 110 of the recording unit 11 and the data for each item (step S102). As described with reference to FIG. 6, the time difference model generated in step S102 is obtained by associating data for each item with a signal value after a predetermined period of the data of the item. That is, in step S102, the control unit 10 associates the signal values stored in time series with the data for each item while shifting by a predetermined period.

  The control unit 10 linearly transforms the data for each item based on the relationship between the corresponding signal value and the data for each item in the generated time difference model (step S103). The control part 10 performs the process which selects the signal period used for prediction using the data after linear transformation (step S104). Details of the process of selecting the signal period will be described with reference to the flowchart of FIG.

  Based on the signal value in the signal period selected in step S104 and the corresponding item data in the time difference model, the control unit 10 uses the above-described Equation 1 and Equation 2 to calculate the proportionality constant β and SN ratio for each item. η (square ratio) is calculated (step S105).

  The control unit 10 calculates the estimated value of the output by Expression 3 for each member in the selected signal period using the proportional constant β and the SN ratio η (square ratio) for each item using Expression 3. Step S106).

  The control unit 10 calculates an overall estimated value using the SN ratio (square ratio), which is the estimation accuracy of the estimated value, as a weighting coefficient by Equation 4 (step S107).

  Next, based on Equation 5, the control unit 10 calculates the SN ratio (db) of the overall estimated value of each item based on the signal value and the overall estimated value (step S108).

  The control unit 10 derives a factor effect value for each item (step S109). As described above, the factor effect value is the difference between the SN ratio of the overall estimated value for the data of each item including the item with respect to the SN ratio of the comprehensive estimated value for the data of each item excluding the item, as described above. It is calculated by obtaining. The SN ratio of the comprehensive estimated value is the strength of the correlation of the data of the item with respect to the signal value, and is a value expressed as a logarithm of a value proportional to the reciprocal of variance.

  Next, the control unit 10 calculates the SN ratio of the comprehensive estimated value for the data of a plurality of items selected in descending order of the factor effect value for each number of items (step S110). Details of the processing in step S110 have been described above with reference to FIG.

  The control unit 10 determines the number of items that maximizes the SN ratio based on the SN ratio of the comprehensive estimated value for each number of items (step S111).

  The control unit 10 selects as many items as the number of items determined in step S111 (step S112). And the control part 10 is the data of the selected item among the data for every item before the predetermined period matched with the prediction object period in a time difference model, ie, the data for every item in the corresponding period nearest to a prediction object period. Is applied to Equation 4 to calculate a predicted value (step S113). Β and η in Equation 4 are those calculated in step S105 (calculated from data after time difference model, linear conversion, and signal period selection). In step S113, the predicted value is output from the output unit 14 or recorded in the recording unit 11. If the signal value is normalized, it can be obtained by performing inverse transformation.

  FIG. 12 is a flowchart illustrating an example of a signal period selection processing procedure by the prediction device 1 according to the present embodiment. The processing procedure shown below corresponds to the details of the signal period selection processing in step S104 of FIG.

  In order to reduce the fluctuation rate, the control unit 10 normalizes the signal value (subtracts the average value) and performs conversion that takes a logarithmic value (step S41).

  The control unit 10 calculates a trend for all members or a predetermined period (for example, 10 years) with respect to the converted signal value (step S42). For the calculation of the signal trend, it is possible to use a known trend calculation method for various time series data, such as applying a trend component model based on a state space model.

  The control part 10 specifies the trend of the signal value of the item data matched with the signal value of a prediction object period, and the signal period of the same period (corresponding period) before application of a time difference model among the trends calculated in step S42 ( Step S43). Specifically, in step S43, the control unit 10 determines whether the trend in the corresponding period is a positive gradient or a negative gradient.

  The control unit 10 selects a signal value and data for each item (after applying the time difference model) in the section having the same trend characteristic as the trend (positive slope / negative slope) identified in step S43 (step S44). ).

  The control unit 10 calculates the Mahalanobis general distance (hereinafter referred to as MD (Mahalanobis Distance)) of the item data (after linear conversion) in the selected section (step S45).

  The control unit 10 calculates the MD trend in the selected section (step S46).

  The control unit 10 sequentially selects a section close to the prediction target period from a plurality of discontinuous sections (sections having the same signal trend) constituting the selected section (step S47). The control unit 10 determines whether or not the MD trend is larger than a predetermined value (for example, 1.0) in the entire selected section (step S48). If the control unit 10 determines that the value is larger than the predetermined value in step S48 (S48: YES), the control unit 10 returns the process to step S47, and then selects another section close to the prediction target period.

  When the control unit 10 determines that the MD trend is not larger than the predetermined value in the entire selected section, that is, a part of the MD trend includes a section having a predetermined value or less (S48: NO), the control section 10 Is selected as a signal period (step S49), and the process ends.

  When the control unit 10 selects the signal period as shown in the flowchart of FIG. 12, the aspect close to the characteristic of the trend of the data value of the item corresponding to the signal value (prediction target) of the prediction target period and the signal value of the same period. Prediction is performed from the relationship between the item data in the period having the characteristic and the signal value after the predetermined period. Thereby, it is possible to reduce the influence of the relationship between the item data and the signal value in a period showing a sign that is not similar to (previously) a predetermined period before the prediction target period, and to improve the accuracy of the prediction. Can be guessed.

  Next, a specific embodiment to which the prediction method by the prediction device 1 is applied will be described.

  As an example, an example in which the prediction method of the present embodiment is applied to the demand prediction of a construction machine will be described. FIG. 13 is an explanatory diagram showing the contents of items in the present embodiment. In this example, the prediction target (signal value) was “the number of construction machines shipped”, and items related to the prediction target were various economic indicators. As an economic index item, as shown in FIG. 13, “month”, “Japan unemployment rate (%)”, “domestic bank lending interest rate (%)”, etc. are set. In each of these items, data for each month is associated with the member. In addition, items indicating months such as “January” and “February” are items indicating how many degrees each item associated as a member corresponds to. , “1” is recorded, and “0” is recorded when it is not the month. For example, for a member in January 2010, the data for the item “January” is “1”, and the data for the items for other months “February” to “December” is “0”. . In the embodiment, the prediction device 1 predicts “the number of construction machines shipped” based on a total of 36 items including 24 items related to economic indicators (partially omitted or changed in detail in FIG. 13) and 12 items related to monthly degrees.

  FIG. 14 is a graph showing an example of the contents of signal values used for prediction in this embodiment. The horizontal axis in FIG. 14 indicates the year and month in time series, and the vertical axis indicates the number of units shipped. In FIG. 14, the number of units shipped is standardized by the monthly average number in 2005. In the present embodiment, the prediction unit 1 is used to predict the shipment quantity for 12 months from January to December 2010 based on the data for each item shown in FIG. 13 with respect to the time series data of the shipment quantity shown in FIG. To do. In addition, evaluation with respect to the predicted value by the prediction apparatus 1 is performed based on the actual value of the number of units shipped for 12 months from January to December 2010, which will be described later.

  First, the control unit 10 of the prediction device 1 generates time-series data to which a time difference model is applied for the signal values of the time-series data and the data for each item for each signal value as shown in FIG. 14 (S102). In this embodiment, since the prediction period is 12 months, a time difference model shifted by 12 months is applied.

  Then, the control unit 10 performs linear conversion on the data of the item to which the time difference model is applied (S103), and performs conversion processing for normalizing and logarithmically converting the signal value as described above (S41). . And the control part 10 calculates | requires the trend of the signal value after conversion (S42).

  FIG. 15 is a graph showing a trend of signal values obtained by the control unit 10 of the prediction device 1. The graph shown in FIG. 15 is displayed so as to overlap the shipment number graph of FIG. In FIG. 15, the signal trend is indicated by black circles and white circles. A white circle indicates a section where the slope of the signal trend is plus (+), and a black circle indicates a section where the slope of the signal trend is minus (−). The abscissa in FIG. 15 shows a plus section and a minus section.

  Based on the signal trend as shown in FIG. 15, the control unit 10 sets the data corresponding to the prediction target period, that is, 2009 before a predetermined period (12 months) from 12 months from January to December 2010. The slope of the signal trend in the latest period from January to December is determined (S43). The control unit 10 determines that the gradient is positive by the process of S43.

  The control unit 10 identifies a section in which the slope of the signal trend is positive from the entire period or a predetermined period (such as 10 years) (S44). FIG. 16 is a graph showing the period selected by the control unit 10 of the prediction apparatus 1 and the MD in the section. In FIG. 16, the horizontal axis indicates the passage of time, and the vertical axis indicates the MD value. The range of the horizontal axis is partly different from FIG. The control unit 10 selects item data in a section in which the identified signal trend is positive (S44), and calculates MD (S45). The calculated MD is indicated by a circle in FIG. Further, the control unit 10 calculates the MD trend (S46). The calculated MD trend is indicated by a thick line in FIG.

  As shown in FIG. 16, the section with a positive signal trend is composed of three discontinuous sections. The control unit 10 first selects the latest section (July to December 2009) among the three sections (S47). The control unit 10 refers to the MD trend calculated for the section, determines that the MD trend is greater than the predetermined value (1.0) (S48: YES), and excludes the section. Next, the control unit 10 selects the next closest section (September 2002 to February 2007) among the three sections (S47). Although the MD trend calculated for the section is partially larger than 1.0, most of the MD trend is 1.0 or less (S48: NO). (March 2006 to February 2007) is selected as the signal period (S49).

  Based on the signal value and the data for each item in the selected signal period (March 2006 to February 2007), the control unit 10 obtains the SN ratio of the comprehensive estimated value for each item using the two-sided T method, Processing for selecting the number of items with the maximum S / N ratio is performed (S105 to S111). And the control part 10 selects the item of the number of selected items (S112), and calculates a predicted value based on the data of the selected item in a corresponding period (January to December, 2009) (S113).

  FIG. 17 is a graph showing a prediction result by the control unit 10 of the prediction device 1 in the present embodiment. The horizontal axis in FIG. 17 indicates the passage of time, and the vertical axis indicates the number of shipments. Values indicated by black circles from January 2005 to January 2010 are actual values (after normalization) and are the same as those shown in FIG. A waveform with a solid line in FIG. 17 indicates a signal trend. The section shown by the arrow in FIG. 14 is the selected period from March 2006 to February 2007. And the value shown in the thick line between January 2010 and December 2010 in FIG. 14 is a value predicted by the control unit 10 based on the data in the selected period. On the other hand, the values indicated by white squares are actual values (after normalization) from January 2010 to December 2010. As shown in FIG. 17, it can be seen that each predicted value can be predicted with sufficient accuracy as a predicted value for 12 months, although it differs from the actual value when viewed monthly.

  For comparison, based on the signal trend and the MD trend as shown in the flowchart of FIG. 12, the prediction accuracy when the target signal period is selected for the last two years or one year is shown. FIG. 18 is a graph showing a predicted value when the signal period is the latest two years, and FIG. 19 is a graph showing a predicted value when the signal period is the latest one year. In FIG. 18, as indicated by arrows, the period from January 2008 to December 2009 is the signal period, and in FIG. 19, the period from January 2009 to December 2009 is also the signal period. In FIG. 18 and FIG. 19, the predicted value is indicated by a white diamond. Comparing FIG. 17 with FIG. 18 and FIG. 19, it can be seen that the predicted value by the prediction device 1 in the present example has the highest accuracy.

  FIG. 20 is a graph showing a comparison of the S / N ratio of the comprehensive estimated value due to the difference in the signal period selection method. FIG. 20 shows the length of the signal period to be selected on the horizontal axis, and the SN ratio of the comprehensive estimated value on the vertical axis. In FIG. 20, the black circle indicates the SN ratio of the total estimated value when the signal period is selected by the prediction apparatus 1 of the present embodiment, and the black square indicates the SN of the total estimated value when the latest period is used. Indicates the ratio. As shown in FIG. 20, when the signal period is predicted based on 12 months, the SN ratio of the total estimated value is the highest, and the accuracy of the comprehensive estimation formula is improved.

  In this way, by applying a method that takes into account the signal trend, the prediction device 1 can realize highly accurate shipment number prediction.

  It should be understood that the embodiment disclosed above is illustrative in all respects and not restrictive. The scope of the present invention is defined by the terms of the claims, rather than the meanings described above, and is intended to include any modifications within the scope and meaning equivalent to the terms of the claims.

DESCRIPTION OF SYMBOLS 1 Prediction apparatus 10 Control part (calculation means)
11 Recording section (recording means)
13 Input unit 14 Output unit 110 Database 2 Prediction program (computer program)

Claims (6)

A value relating to an event related to a plurality of factors, and a recording means for recording the plurality of factors related to the value in association with the value as data for a plurality of items in time series, from the recording means to the A factor and an effect value indicating a strength of correlation with respect to a change in the value of the plurality of items, the calculation unit including a value and data for each of the plurality of items corresponding to the value. In the prediction device that selects a plurality of items from the plurality of items, and predicts the value related to the event in the prediction target period based on the data of the selected items.
The computing means is
A specifying means for specifying a tendency to change over time during a recording period of the value recorded in the recording means;
A period selection means for selecting the most recent period similar to the temporal change tendency of the value in the corresponding period before the prediction target period from the recording period;
Item selection means for selecting a plurality of items based on the data for each of the plurality of items in the selection period selected by the period selection means and the factor effect value obtained by a value after a predetermined period from the selection period; ,
A prediction device comprising: prediction means for calculating a prediction value in the prediction target period from data of items selected by the item selection means in the corresponding period. The period selection means includes
A discriminating means for discriminating a gradient of change over time of the value in the corresponding period;
Means for selecting, from the recording period, a first period whose value changes with the gradient determined by the determining means;
First calculation means for calculating a numerical value representing the similarity between the data for each of the plurality of items in the corresponding period and the data for each of the plurality of items in the first period;
Second calculating means for calculating a numerical value representing a change tendency of the numerical value calculated by the first calculating means;
The selection means which selects the 2nd period with the most recent similarity in the 1st period based on the comparison with the numerical value and the threshold which the 2nd calculation means computed. Prediction device. The first calculating means calculates the Mahalanobis general distance,
The second calculation means calculates the Mahalanobis pan-distance trend calculated by the first calculation means based on the trend model,
The prediction device according to claim 2, wherein the selection unit selects a period in which the trend calculated by the second calculation unit is equal to or less than a predetermined value as the second period. The selection means is adapted to select by changing the length of the second period,
For each length, further comprising a determining means for determining a length that maximizes a value related to prediction accuracy with respect to a value after a predetermined time of data for each item in the second period;
The prediction apparatus according to claim 2, wherein the period selection unit is configured to select a second period having a length determined by the determination unit. A value relating to an event related to a plurality of factors, and a recording means for recording the plurality of factors related to the value in association with the value as data for a plurality of items in time series, from the recording means to the A calculation means that reads out data for each of the plurality of items corresponding to the value and performs calculation, and the calculation means indicates the strength of correlation with respect to the change in the value of the plurality of items. In the prediction method of selecting a plurality of items from the plurality of items based on the factor effect value, and predicting the value related to the event in the prediction target period based on the data of the selected item,
The computing means is
Identify the trend of change over time during the recording period of the value recorded in the recording means,
The most recent period similar to the time-dependent change trend of the value in the corresponding period before the prediction target period is selected from the recording period,
Based on the data for each of the plurality of items in the selected selection period, and the factor effect value obtained by a value after a predetermined period from the selection period, select a plurality of items,
A prediction method for calculating a predicted value in the prediction target period from data of a plurality of selected items in the corresponding period. A computer having a recording unit, reading a value related to an event related to a plurality of factors recorded in time series in the recording unit, and reading data for each of the plurality of factors related to the value; Based on a factor effect value indicating the strength of correlation with respect to a change in the value of the plurality of items, the plurality of items are selected from the plurality of items, and the event in the prediction target period based on data of the selected items In a computer program for predicting a value related to
In the computer,
A specifying means for specifying a tendency to change over time during a recording period of a value recorded in the recording means;
A period selection means for selecting, from the recording period, the most recent period similar to the time-dependent change trend of the value in the corresponding period before the prediction target period;
Item selection means for selecting a plurality of items based on the data for each of the plurality of items in the selected selection period, and the factor effect value obtained by a value after a predetermined period from the selection period, and
A computer program for executing a function of a prediction means for calculating a predicted value in the prediction target period from data of a plurality of selected items in the corresponding period.

Images