JP5592813B2 - Lifetime demand forecasting method, program, and lifetime demand forecasting device - Google Patents

Description

  The present invention relates to a technique for a lifetime demand prediction method, a program, and a lifetime demand prediction apparatus.

  In durable consumer goods such as home appliances and automobiles, the demand for maintenance parts continues for a long time even after the production of the product parts ends. For this reason, the manufacturer procures the relevant maintenance parts together at the end of the production of the parts and holds them as inventory. In order to determine the amount of procurement, the demand for maintenance parts until the end of the maintenance period (hereinafter referred to as lifetime demand) is predicted.

  Conventionally, a manufacturer predicts the lifetime demand for maintenance parts by assuming that the demand is attenuated at a certain rate of decrease or at a certain rate. These methods are premised on application to maintenance parts in a period of continuous decline in demand.

  For example, Patent Document 1 discloses a method for estimating a demand trend in the following year by performing regression analysis using the actual demand in each month of the previous year as an explanatory variable in order to obtain the total annual demand forecast during the demand increase period. Has been.

FIG. 22 is a graph showing an example of regression analysis disclosed in Patent Document 1.
In the example shown in FIG. 22, the horizontal axis is the number of elapsed months, the vertical axis is the component usage record of the maintenance part, and the regression line of the component use record in each elapsed month is calculated.

  Non-Patent Document 1 includes a moving average method, an exponential smoothing method, and an ARIMA (Auto Regressive Integrated Moving Average) model as a method used for demand prediction.

JP 2009-230555 A

R. L. Goodrich, “Applied Statistical Forecasting,” Business Forecast Systems, Inc. (1992), p.45-62,90-92

  However, in recent years, new products have been introduced and new technologies have been introduced in a short period of time, and the production end time of parts has been accelerated. For this reason, there is an increasing need for procurement by predicting the lifetime demand for maintenance parts at the beginning of the maintenance period. That is, there is an increasing need for procurement by predicting lifetime demand even in a situation where there is not enough data to predict lifetime demand by the techniques described in Patent Literature 1 and Non-Patent Literature 1.

Here, the techniques described in Patent Document 1 and Non-Patent Document 1 are time-series prediction methods that perform prediction on the assumption that the demand trend of the parts that have been predicted will continue in the future.
However, the trend of demand for maintenance parts is not uniform, increasing in the early stages, stabilizing in the middle, and decreasing in the final stages.
Therefore, in the time series prediction methods described in Patent Document 1 and Non-Patent Document 1 that perform prediction on the assumption that the demand trend will continue, only the data at the beginning of the maintenance period in which the demand is increasing is available. In the situation where there is not enough data, it is impossible to predict the lifetime demand that changes the trend.

  The present invention has been made in view of such a background, and provides a lifetime demand prediction method, a program, and a lifetime demand prediction device for predicting the lifetime demand of maintenance parts even when sufficient performance values are not available. Objective.

In order to solve the above-mentioned problems, the present invention provides a conversion function for each part and for each number of elapsed periods from the start of maintenance of the parts, and the cumulative number of product shipments up to the elapsed period and the parts demand per product. Normalized part demand amount calculation unit 101 that calculates the normalized part demand amount converted by the above, a component grouping unit 103 that classifies parts into groups, and every group, every number of elapsed periods after the elapsed (k + 1) period The linear regression formula construction unit 104 that constructs a linear regression formula for predicting the normalized part demand for the number of elapsed periods, and the group to which the forecast target part input via the input device 40 belongs are identified and linear It has the demand amount predicted value calculation part 106 which calculates the predicted value of the demand amount of the components in the prediction object period of the prediction object part using the linear regression equation which the regression equation construction part 104 calculated.
Other embodiments will be described later in the embodiments.

  According to the present invention, it is possible to provide a lifetime demand prediction method, a program, and a lifetime demand prediction apparatus for predicting the lifetime demand of maintenance parts even when sufficient performance values are not available.

It is a block diagram which shows the structural example of the lifetime demand prediction apparatus which concerns on this embodiment. It is a functional block diagram which shows the structural example of the process part which concerns on this embodiment. It is a figure which shows the example of the product shipment quantity table which concerns on this embodiment. It is a figure which shows the example of the components shipment quantity table which concerns on this embodiment. It is a figure which shows the example of the product component corresponding | compatible table which concerns on this embodiment. It is a figure which shows the example of the product attribute table which concerns on this embodiment. It is a figure which shows the example of the components attribute table which concerns on this embodiment. It is a figure which shows the example of the maintenance start year table which concerns on this embodiment. It is a figure which shows the example of the cumulative product shipment number table which concerns on this embodiment. It is a figure which shows the example of the normalized component demand amount table which concerns on this embodiment. It is a figure which shows the example of the group table which concerns on this embodiment. It is a figure which shows the example of the prediction information table which concerns on this embodiment. It is a figure which shows the example of the linear regression type table which concerns on this embodiment. It is a figure which shows the example of the demand amount estimated value table which concerns on this embodiment. It is a flowchart which shows the whole process of the lifetime demand prediction method which concerns on this embodiment. It is a figure which shows the example of a manual grouping screen which concerns on this embodiment. It is a flowchart which shows the detailed procedure of the normalized components demand amount calculation process which concerns on this embodiment. It is a flowchart which shows the detailed procedure of the linear regression type construction process which concerns on this embodiment. It is an example of the graph of a linear regression formula in the case of k = 1. It is a flowchart which shows the detailed procedure of the demand amount predicted value calculation process which concerns on this embodiment. It is a figure which shows an example of the lifetime demand prediction screen which concerns on this embodiment. 5 is a graph showing an example of regression analysis disclosed in Patent Document 1.

  Next, a mode for carrying out the present invention (referred to as “the present embodiment”) will be described in detail with reference to the drawings as appropriate. In this embodiment, “maintenance part” is appropriately described as “part”. Further, in the present embodiment, the demand prediction is performed in an annual manner, and the number of elapsed periods is described as “elapsed year”, the prediction target period is described as “prediction target year”, etc., but the present invention is limited to the year. It is a technology that can be applied to demand forecasts in any period such as monthly or quarterly.

"Device configuration"
FIG. 1 is a block diagram illustrating a configuration example of a lifetime demand prediction apparatus according to the present embodiment.
The lifetime demand prediction device 1 includes a memory 10 such as a RAM (Random Access Memory), a storage device 20 such as an HDD (Hard Disk Drive), a CPU (Central Processing Unit) 30, and an input device (input unit) 40 such as a keyboard and a mouse. And a display device (display unit) 50 such as a display or a printer.
The storage device 20 includes a product shipment number table (product shipment number information) 201, a part shipment number table (part shipment number information) 202, a product part correspondence table (product part correspondence information) 203, and a product attribute table, which will be described later. 204, a parts attribute table 205, a maintenance start year table 206, a cumulative product shipment quantity table 207, a normalized parts demand quantity table 208, a group table 209, a prediction information table 210, a linear regression equation table 211, and a demand quantity forecast value table 212. Stored.
The memory 10 includes a processing unit 100 that is implemented by loading a program stored in the storage device 20 and executing the program by the CPU 30. Details of the processing unit 100 will be described later with reference to FIG.

(Processing unit configuration)
FIG. 2 is a functional block diagram illustrating a configuration example of the processing unit according to the present embodiment.
The processing unit 100 includes a normalized part demand amount calculation unit 101, a prediction information setting unit 102, a part grouping unit 103, a linear regression formula construction unit 104, a prediction target part setting unit 105, a demand amount predicted value calculation unit 106, and a prediction graph. A display unit 107 and a predicted value display unit 108 are provided.
The normalized part demand calculation unit 101 converts the cumulative number of product shipments up to the elapsed year and the parts demand per product using a conversion function for each part and every year since the maintenance of the part started. Calculate the normalized parts demand.
The prediction information setting unit 102 sets information regarding prediction input via the input device 40.
The component grouping unit 103 classifies components into groups.
The linear regression formula construction unit 104 constructs a linear regression formula for predicting the normalized part demand for the elapsed year for each group and every elapsed year after the (k + 1) th year.
The prediction target component setting unit 105 sets information on the prediction target component input via the input device 40.
The demand amount predicted value calculation unit 106 identifies a group to which the prediction target part belongs, and uses the linear regression formula constructed by the linear regression formula construction unit 104 to predict the demand amount of the part in the forecast target year of the prediction target part. Calculate the value.
The prediction graph display unit 107 displays the prediction value on the display device 50 as a graph.
The predicted value display unit 108 displays a list of predicted values on the display device 50.

  In the present embodiment, the lifetime demand prediction device 1 is a single device, but the storage device 20 is a database and another device, or any function of the processing unit 100 is given to another computer. Also good.

《Various tables》
Next, various tables used in this embodiment will be described with reference to FIGS. 3 to 14 while referring to FIG.

(Product shipment table)
FIG. 3 is a diagram illustrating an example of a product shipment quantity table according to the present embodiment.
The product shipment quantity table 201 is a table in which the product shipment quantity of the product in each year after the product shipment start year (product shipment start period) is registered for each product, and the product ID (Identification) and year (period) ) And the number of product shipments of the product in that year.

(Parts shipment quantity table)
FIG. 4 is a diagram illustrating an example of a component shipment quantity table according to the present embodiment.
The part shipment quantity table 202 is a table in which the part shipment quantity of the part in each year is registered for each part, and the part ID, the year, and the part shipment quantity of the part in that year are set information. Is stored as The “year” here is the calendar year.

(Product parts correspondence table)
FIG. 5 is a diagram showing an example of the product part correspondence table according to the present embodiment.
The product part correspondence table 203 is a table in which which part is used for which product, and a product ID and a part ID are stored as a pair of information.

(Product attribute table)
FIG. 6 is a diagram illustrating an example of a product attribute table according to the present embodiment.
The product attribute table 204 is a table in which product types used for component classification and product shipment information are registered, and a product ID uniquely assigned to a product, a product type, and a product shipment. A start year, a product shipment period indicating how many years the product has been shipped from the product shipment start year, and the total number of products shipped in the product shipment period are stored as a set of information.

(Part attribute table)
FIG. 7 is a diagram illustrating an example of a component attribute table according to the present embodiment.
The component attribute table 205 is a table in which component types used for component classification are registered. A component ID uniquely assigned to a component and a component type are stored as a pair of information. Yes.

(Maintenance start year table)
FIG. 8 is a diagram illustrating an example of a maintenance start year table according to the present embodiment.
The maintenance start year table 206 is a table for storing the maintenance start year calculated by the normalized part demand amount calculation unit 101 for each part. The part ID and the maintenance start year (maintenance start period) are a pair. Stored as information.

(Cumulative product shipment quantity table)
FIG. 9 is a diagram showing an example of the cumulative product shipment quantity table according to the present embodiment.
The cumulative product shipment quantity table 207 is a table that stores the cumulative product shipment quantity of parts for each part and every elapsed year (elapsed period). The part ID, the elapsed year, and the cumulative product shipment quantity are a set. Stored as information. Here, the elapsed year is the number of years counted from the maintenance start year as the first year.

(Normalized parts demand table)
FIG. 10 is a diagram illustrating an example of a normalized parts demand amount table according to the present embodiment.
The normalized part demand amount table 208 is a table that stores the normalized part demand amount calculated by the normalized part demand amount calculation unit 101 for each part and every elapsed year. Quantized parts demand amount is stored as a set of information.

(Group table)
FIG. 11 is a diagram illustrating an example of a group table according to the present embodiment.
The group table 209 is a table for storing information related to a group of components created by the component grouping unit 103, and stores a component ID and a group ID uniquely assigned to the group as a pair of information. Yes.

(Prediction information table)
FIG. 12 is a diagram illustrating an example of a prediction information table according to the present embodiment.
The prediction information table 210 is a table that stores prediction setting information set by the prediction information setting unit 102 and information on the prediction target component set by the prediction target component setting unit 105. The order k of the formula, the prediction start year (prediction start period), and the prediction end year (prediction end period) are stored.

(Linear regression table)
FIG. 13 is a diagram illustrating an example of a linear regression equation table according to the present embodiment.
The linear regression equation table 211 is a table that stores the linear regression coefficient of the linear regression equation constructed by the linear regression equation construction unit 104 for each group and every elapsed year.
The elapsed year, the linear regression coefficient index, and the linear regression coefficient are stored as a set of information. Here, the index is a number assigned to the explanatory variable.

(Demand forecast table)
FIG. 14 is a diagram illustrating an example of a demand amount prediction value table according to the present embodiment.
The demand amount predicted value table 212 is a table that stores the predicted value of the demand amount of the part calculated by the demand amount predicted value calculation unit 106 for each part and for each prediction target year. The part ID, the prediction target year, The predicted value of the demand amount is stored as a set of information.

<Description of processing>
Next, the lifetime demand amount prediction method according to the present embodiment will be described with reference to FIGS. 15 to 20 with reference to FIGS.

(Overall processing)
FIG. 15 is a flowchart showing the entire process of the lifetime demand prediction method according to the present embodiment.
First, the normalized part demand calculation unit 101 uses a conversion function to calculate the cumulative number of products shipped up to the elapsed year and the part demand per product for each part and every year since the maintenance of the part started. The converted normalized part demand is calculated (S101). Step S101 will be described later with reference to FIG.
Next, prediction information is input via the input device 40 (S102). Here, the input prediction information includes the order k of the linear regression equation, the prediction start year (prediction start period), the prediction end year (prediction end period), and the prediction information setting unit 102 receives the input prediction information. Is stored in the prediction information table 210 (FIG. 12) of the storage device 20.

Then, the component grouping unit 103 performs component grouping to classify components into groups (S103).
The parts grouping is performed so that parts with similar parts shipping tendency (for example, parts of the same type used in refrigerators of the same system) belong to the same group. In the component grouping, the component grouping unit 103 issues a group ID to the component group, associates the component ID of each component with the group ID, and stores them in the group table 209 (FIG. 11).

The following two methods are conceivable as the component grouping process.
(1) The user manually performs component grouping via the input device 40. At this time, the user inputs product attribute and part attribute information to the lifetime demand prediction apparatus 1 via the input device 40, and the part grouping unit 103 selects parts corresponding to the input product attribute or part attribute. Belong to the same group.
Manual grouping will be described later with reference to FIG.

(2) The part grouping unit 103 automatically classifies parts from product attributes or part attributes by using an existing classification method such as multivariate regression tree classification.
When multivariate regression tree classification is used as a means of part grouping processing, product attributes such as product shipment start year, product shipment period, total product shipment quantity, and part attributes such as part type are used for regression tree classification. Variable. The component grouping unit 103 constructs a decision tree that branches under conditions related to these variables, and groups components. As an example, when the period from the prediction start year to the prediction end year is L and the coefficient of determination of the linear regression equation in the elapsed t-year constructed in step S104 described later is Rt, R1 + R2 +. The decision tree is constructed so that
Since regression tree classification is an existing technique, detailed description is omitted.

Next, the linear regression formula construction unit 104 predicts the normalized parts demand in the elapsed year for each group and every elapsed year after the elapsed (k + 1) year (elapsed (k + 1) period). Is constructed (S104).
Step S104 will be described later with reference to FIG.

Then, information on the prediction target part is input via the input device 40 (S105). The prediction target component setting unit 105 stores the input information of the prediction target component in the prediction information table 210 (FIG. 12).
Next, based on the linear regression equation (constructed in step S104) of the group to which the prediction target part belongs, the demand amount prediction value calculation unit 106 calculates the demand amount of the part in the prediction target year (prediction target period) of the prediction target part. A predicted value is calculated (S106).
Step S106 will be described later with reference to FIG.

After calculating the predicted value of the demand amount in step S106, the predicted value of the demand amount calculated by the prediction graph display unit 107 is displayed on the display device 50 as a prediction graph (S107).
Further, the predicted value display unit 108 displays the contents (predicted values) of the demand amount predicted value table 212 as a list on the display device 50 (S108). Note that step S108 can be omitted.

Here, manual grouping in step S103 will be described.
FIG. 16 is a diagram illustrating an example of a manual grouping screen according to the present embodiment.
The manual grouping screen 1600 has an unclassified component display area 1601 and a group information display area 1602.
The unclassified component display area 1601 includes component information displayed in the unclassified component list display area 1611 and a filter list display area 1612.
In the unclassified parts list display area 1611, product IDs, product ID values, product types, part IDs 1 and the like of parts not grouped (to be grouped from now on) are displayed. Here, in addition to the component ID 1 shown in FIG. 16, the component ID is managed by one component with a plurality of component IDs, such as a component ID 2, component ID 3,. To do. The numerical value of the product ID is a numerical value that appears first when viewed from the left side of the product ID. For example, if the product ID is “M00001”, the numerical value of the product ID is “1”. The numerical value of the product ID represents, for example, the product model number or specification.

The filter list display area 1612 is a display area for filtering the unclassified component list display area 1611. When the user inputs filtering information in the filter list display area 1612 and then selects and inputs a filter execution button, the result filtered by the input filtering information is reflected / displayed in the unclassified component list display area 1611. .
In the example of FIG. 16, the information regarding the parts used in the product whose product type is “Reizoko” with a forward match and the total number of products shipped so far is 100,000 or more is an unclassified parts list display area 1611. Is displayed.
It should be noted that by selecting and inputting the row addition button in the filter list display area 1612, a row having the item input in the item window 1614 is added.
Further, when the user selects and inputs a row deletion button in the filter list display area 1612, the row selected in the filter list display area 1612 is deleted.

When a check is selected and input in the check field 1613 displayed in the mark item of the uncategorized parts list display area 1611 and the group addition button 1631 is pressed, the checked parts are displayed in the group information display area 1602. Added to the group that has been added.
The user can input a check in all check fields 1613 in the unclassified parts list display area 1611 by selecting and inputting the “mark all parts” button in the unclassified parts list display area 1611. By selecting and clearing the “clear mark” button, all the check boxes 1613 in the unclassified component list display area 1611 can be unchecked.

The group information display area 1602 has a group list display area 1621 and an in-group part information display area 1622 displaying a list of parts already grouped.
The group list display area 1621 is an area in which a list of parts groups that have been created or are being created is displayed. In the example of FIG. 16, information on parts that belong to the group “refrigerator door 2” and whose total product shipment is classified as 500,000 or more is displayed in the in-group part information display area 1622.
The items displayed in the in-group component information display area 1622 are the same as those in the unclassified component list display area 1611, and thus the description thereof is omitted.

When the user inputs a group name in the group name input window 1624 and selects and inputs a group addition button, a new line is displayed in the group list display area 1621 and a new group can be edited.
When the user selects and inputs a group delete button, the row selected in the group list display area 1621 is deleted and the corresponding group is deleted.

When the user selects and inputs a check in the check field 1623 displayed in the mark item of the in-group part information display area 1622 and presses the group exclusion button 1632, the checked part is replaced with the in-group part. It is deleted from the information display area 1622 and excluded from the group.
When the user selects and inputs the “mark all components” button in the in-group component information display area 1622, a check can be input to all check fields 1623, and the “clear all marks” button is clicked. By selecting and inputting, all the check boxes 1623 in the in-group component information display area 1622 can be unchecked.

  Furthermore, it is possible to input a prediction period in step S102 of FIG.

(Normalized parts demand calculation processing)
FIG. 17 is a flowchart showing a detailed procedure of normalized part demand amount calculation processing (step S101 in FIG. 15) according to the present embodiment.
First, the normalized component demand amount calculation unit 101 acquires one component ID registered in the component attribute table 205 (S201). Hereinafter, in the description of the processing in FIG. 17, a component corresponding to the acquired component ID is referred to as a component p.
Next, the normalized component demand amount calculation unit 101 acquires product IDs of all products using the component p from the product component correspondence table 203 using the acquired component ID as a key (S202). Since the part p may be used in a plurality of products, the ID acquired in step S202 is not necessarily one. Hereinafter, in the description of the processing in FIG. 17, a set of all products registered in the product part correspondence table 203 and using the part p is denoted by Mp.

Then, the normalized part demand amount calculation unit 101 refers to the product attribute table 204 and identifies the product with the earliest product shipment start year among the products belonging to the product Mp (S203).
Next, the normalized part demand calculation unit 101 sets the product shipment start year (product shipment start period) of the product identified in step S203 as the maintenance start year (maintenance start period) of the part p (S204), and It is stored in the maintenance start year table 206 together with the component ID. And later, a maintenance start year of parts p and _{Y p,} and elapsed t th year of parts p the _(Y p + t-1) year.

Subsequently, the normalized part demand amount calculation unit 101 substitutes 1 for a variable t representing the elapsed year (S205).
Then, the normalized parts demand calculation unit 101 calculates the cumulative product shipment number _Sp (t) of the parts p in the elapsed t years (S206). When the product shipment start year of the product m is y _m and the product shipment number of the product m in year y is s _m (y), the cumulative product shipment number Sp (t) is calculated by the following equation (1). That is, the cumulative product shipment quantity S _p (t) of the part p is the cumulative shipment quantity from the shipment start year of the product to the elapsed t year for all products in which the part p is used. That is, the normalized part demand amount calculation unit 101 uses the part among the products registered in the product part correspondence table 203 for each elapsed year (elapsed period) after the maintenance start year (maintenance start period). The total number of products shipped from the maintenance start year (maintenance start period) to the elapsed years (elapsed period) of all the products is calculated.

The normalized part demand amount calculation unit 101 stores the calculated cumulative product shipment quantity in the cumulative product shipment quantity table 207 together with the part ID and elapsed year of the part p to be calculated.
Subsequently, the normalized part demand amount calculation unit 101 acquires the part shipment quantity d _p (t) in (Y _p + t−1) years from the part shipment quantity table 202 using the part ID of the part p as a key (S207). .
Next, the normalized part demand amount calculation unit 101 calculates a normalized part demand amount n _p (t) in the elapsed t years of the part p (S208). That is, the normalized part demand amount calculation unit 101 calculates a normalized part demand amount obtained by dividing the number of parts shipped for the maintenance part in the elapsed years (elapsed period) by the cumulative number of products shipped. Here, in order to reduce the variance non-uniformity of the linear regression equation calculated later, the normalized part demand amount calculation unit 101 converts the number of parts shipped per product shipment by Box-Cox conversion or the like. Calculate the normalized parts demand. Variance non-uniformity means that the variance of errors varies from sample to sample and is not constant.
When the Box-Cox transformation is used, the transformation function f _λ is given by the following equation (2).

Here, λ is a Box-Cox conversion parameter, and is specified by the user via the input device 40.
When the conversion function is f _λ , n _p (t) is calculated by the following formula (3) using the cumulative product shipment quantity S _p (t) and the part shipment quantity d _p (t) in the elapsed t years. Is done.

Although Box-Cox conversion is used here, other conversions such as logit conversion may be used without being limited to Box-Cox conversion.
Further, if it is expected that the variance non-uniformity of the error in the linear regression equation will be reduced, the conversion in step S208 may not be performed, and in this case, all the conversion functions in the following description can be read as identity functions. That's fine.
The cumulative product shipment quantity S _p (t) cannot be calculated, or the part shipment quantity d _p (t) in the elapsed t years cannot be obtained, or d _p (t) / S _p (t) is When not belonging to the definition area of f, the normalized part demand amount calculation unit 101 does not calculate the normalized part demand amount n _p (t).
The normalized part demand amount calculation unit 101 calculates the normalized part demand amount n _p (t) together with the elapsed year for which the part ID and the normalized part demand amount n _p (t) are calculated. Store in the quantity table 208.

Next, the normalized part demand amount calculation unit 101 uses the part ID of the part p to be processed as a key, and the data on the number of parts shipped after the elapsed (t + 1) years of the part p in the parts shipment quantity table 202. It is determined whether there is (parts shipment data) (S209).
If there is part shipment data as a result of step S209 (step S209 → Yes), the normalized part demand amount calculation unit 101 adds 1 to the variable t representing the elapsed year (S210), and then returns the process to step S206.

If there is no part shipment data as a result of step S209 (step S209 → No), the normalized part demand amount calculation unit 101 completes the processes from step S201 to step S210 for all parts registered in the part attribute table 205. It is determined whether or not (S211).
As a result of step S211, when there is a part that has not been processed (step S211 → No), the normalized part demand amount calculation unit 101 returns the process to step S201.
As a result of step S211, when all the parts have been processed (step S211 → Yes), the processing unit 100 ends the process of step S101 in FIG. 15, and thus proceeds to step S102.

(Linear regression formula construction process)
FIG. 18 is a flowchart showing a detailed procedure of the linear regression equation construction process (step S104 in FIG. 15) according to the present embodiment.
First, the linear regression formula construction unit 104 acquires one group ID registered in the group table 209 (S301). Hereinafter, in the description of the processing in FIG. 18, a group corresponding to the acquired group ID is referred to as a group G, and is referred to as a set of components.
Next, the linear regression formula construction unit 104 substitutes (k + 1) for a variable t representing the elapsed year (S302). Here, k is the order k input in step S102 of FIG.
Subsequently, the linear regression formula construction unit 104 constructs a data set for constructing a linear regression formula in the elapsed t years of group G (S303). A data set is a set of normalized part demands that satisfy equation (4).

Here are the points to note when building a dataset.
Since the demand for parts decreases with time, there are cases where the number of parts shipped becomes “0” in many years. If data for the year in which the number of parts shipped is “0” is included in the data set, the degree of fit of the linear regression equation is low. Therefore, the predicted value of the normalized component demand calculated by the linear regression equation is calculated as a conditional expected value when the demand is positive. And it correct | amends to the predicted value of the demand amount also considered when the number of parts shipped becomes "0" by step S409, S410 of FIG. 20 mentioned later (details are demonstrated by step S409, S410 of FIG. 12). The data set is a set of (k + 1) -dimensional vectors represented by the following equation (4).

In equation (4), n _p (t−j) is the normalized component demand, and d _q (t−j) is the number of components shipped in the elapsed (t−j) year of component q.
Further, the condition d _q (t−j)> 0 in the expression (4) is a condition for excluding the case where the normalized component demand amount in the elapsed (t−j) year of the component q is “0”. is there.

Next, the linear regression formula construction unit 104 determines whether or not the number of data in the data set constructed by Step S303 (Formula (4)) is greater than or equal to a threshold value (S304).
Here, since the number of data must be (k + 1) or more in order to construct a linear regression equation, the threshold value is at least (k + 1).
If the number of data is less than the threshold value as a result of step S304 (S304 → No), the linear regression equation constructing unit 104 advances the process to step S311.

If the number of data is greater than or equal to the threshold value as a result of step S304 (step S304 → Yes), the linear regression formula construction unit 104 uses the data set constructed in step S303 to determine the linear regression coefficient of the linear regression formula shown in formula (5). A linear regression equation is constructed by calculating β _j (t) (j = 0,..., k) by the least square method (S305).

Here, n (t−j) (j = 0,..., K) is a variable representing the normalized component demand in the elapsed (t−j) year of group G.
In the case of k = 1, the linear regression equation (formula (5)) is a primary using the normalized part demand in the elapsed (t-1) year as an explanatory variable and the normalized part demand in the elapsed t year as an objective variable. It becomes an expression. An example of the graph of the linear regression equation in this case is shown in FIG.

FIG. 19 is an example of a graph of a linear regression equation when k = 1.
In FIG. 19, the horizontal axis represents the normalized part demand in the elapsed (t−1) year, and the vertical axis represents the normalized part demand in the elapsed t year. That is, the normalized part demand in the elapsed (t-1) year is used as an explanatory variable, and the normalized part demand in the elapsed t year is used as an objective variable.
A point in FIG. 19 is a normalized part demand amount of each part in the group G, and a line 1901 is a graph representing a linear regression equation of the normalized part demand amount. The slope of this straight line is the coefficient β ₁ (t).

Returning to the explanation of FIG. 18, after step S304, the linear regression formula construction unit 104 determines whether or not there is an outlier in the data set constructed in step S303, using the linear regression formula constructed in step S305. (S306). This is because if the outlier exists in the data set used to construct the linear regression equation, the degree of fit of the linear regression equation is reduced.
Here, in order to determine whether or not the data of the component qεG is an outlier, the Cook distance D _q (t) defined by the following equation (6) is used.

Here, c _r (t) is a predicted value when the normalized part demand in the elapsed t years of the part r is predicted in the linear regression equation (formula (5)), and c _r (q; t) Is a predicted value when a normalized regression demand for the elapsed t years of the component r is predicted after the linear regression equation (Equation (5)) is constructed excluding the component q.
When D _q (t) in Expression (6) exceeds the threshold value, the linear regression formula construction unit 104 determines that the data of the part q is an outlier. For example, “1.0” is used as the threshold value.

  In this embodiment, the Cook distance is used as the outlier determination method. However, the present invention is not limited to this, and other determination methods such as determination by lever ratio may be used. In addition, since the Cook distance is used in the present embodiment, the outlier is determined after the linear regression coefficient is calculated. However, if possible, the outlier may be determined when the data set is constructed in step S303, for example.

As a result of step S306, if an outlier exists (S306 → Yes), that is, if the distance of Cook is used for the determination, there is data of the normalized parts demand amount in which Cook distance D _q (t) is larger than the threshold. If so, the linear regression formula construction unit 104 removes the normalized part demand data of the part q determined to be an outlier from the data set (S307), returns the process to step S304, and removes the outlier. The linear regression coefficient is calculated again based on the data set.

As a result of step S306, when there is no outlier (S306 → No), that is, when the Cook distance is used to determine the outlier, the Cook distance D _q (t) in the data for all the parts q is equal to or less than the threshold value. In the case of the linear regression equation construction unit 104, for j = 0,..., K in Equation (5), the group ID of the group G, the elapsed year t, the index j, and the linear regression calculated in step S305. The coefficient β _j (t) is stored as a set of data in the linear regression equation table 211 (S308). Here, “j” of the linear regression coefficient β _j (t) is an index of the linear regression equation table 211.

Next, the linear regression formula construction unit 104 determines whether or not the elapsed year t is the final elapsed year with reference to the normalized parts demand amount table 208 (S309). Here, the last elapsed year is an elapsed year with no data after the corresponding elapsed year in the normalized component demand table 208. .
As a result of step S309, when the elapsed year t is the last elapsed year (S309 → Yes), the linear regression formula construction unit 104 advances the processing to step S311.
As a result of step S309, when the elapsed year t is not the final elapsed year (S309 → No), the linear regression formula construction unit 104 adds “1” to the variable t representing the elapsed year (S310), and the process proceeds to step S303. Go back and build a linear regression equation when the elapsed year increases by one year.

If the number of data is less than the threshold value as a result of step S304 (S304 → No), or if the elapsed year t is the last elapsed year as a result of step S309 (S309 → Yes), the linear regression formula construction unit 104 displays the group table. It is determined whether or not the processing from steps S301 to S310 has been completed for all groups registered in 209 (S311).
As a result of step S311, when there is a group that has not been processed (S311 → No), the linear regression formula construction unit 104 returns the process to step S301.
As a result of step S311, when the processing has been completed for all the groups (S311 → Yes), the processing unit 100 ends the processing of step S104 in FIG. 15 and advances the processing to step S105.

(Demand forecast calculation processing)
FIG. 20 is a flowchart showing a detailed procedure of the demand amount predicted value calculation process (step S106 in FIG. 15) according to the present embodiment.
First, the demand amount predicted value calculation unit 106 acquires one group ID from the group table 209 using the part ID of the part to be predicted set in step S105 in FIG. 15 as a key, so that the group to which the part to be predicted belongs belongs. Is identified (S401). As a result, the group to which the prediction target part belongs is specified. Hereinafter, in the description of the processing in FIG. 18, the prediction target component is referred to as a prediction target component p, and a group corresponding to the acquired group ID is referred to as a group G, and is referred to as a set of components.
Next, the demand amount predicted value calculation unit 106 refers to the normalized part demand amount table 208 using the part ID of the prediction target part p as a key, and normalizes parts from k years before the start of prediction to 1 year before the start of prediction. A demand amount is acquired (S402). Here, k is the order k input in step S102. Also, k years before the start of prediction is, for example, k = 2 and if the prediction start year is 2011, the year before the start of prediction is 2009.
From the demand amount prediction value calculation unit 106 prediction information table 210, obtains the predicted start year, to the prediction starting year and Y _0. In step S402, for j = 1,..., K, the normalized component demand n _p (Y ₀ −j−Y _p +1) of the prediction target component p in (Y ₀ −j) year is obtained. It is. If there is a normalized part demand amount that cannot be acquired, it is assumed that the demand amount cannot be predicted, and the demand amount prediction value calculation unit 106 causes the display device 50 to display an error and ends the process. Incidentally, Y _p represents the maintenance start year of the prediction target component p as described above.

Subsequently, the demand amount prediction value calculation unit 106 substitutes the predicted start year Y ₀ into a variable y representing the prediction target year (S403), converts the prediction target year y in year elapsed t by the following equation (7) ( S404).

t = y−Y _p +1 (7)

Then, the predicted demand amount calculation unit 106 uses the group ID acquired in step S401 as a key, and the linear regression coefficient of the linear regression equation (formula (5)) in the elapsed t years of the corresponding group G from the linear regression equation table 211. Are all acquired (S405).
Next, the demand amount predicted value calculation unit 106 uses the linear regression equation (Formula (5)) using the linear regression coefficient acquired in step S405 to normalize the part demand for the prediction target part p in the elapsed t years. A predicted value c _p (t) of the quantity is calculated (S406).
As the explanatory variable, the normalized component demand amount of the prediction target component p in the elapsed (t−j) year (j = 1,..., K) is used. When the elapsed (t−j) year falls before the prediction start year Y ₀ , that is, when t−j <Y ₀ −Y _p +1, the actual value n _p (t−j of the normalized parts demand amount) ) Is used. When t−j ≧ Y ₀ −Y _p +1, the predicted value c _p (t−j) of the normalized component demand is used.

Subsequently, the demand amount predicted value calculation unit 106 converts the normalized component demand amount predicted value c _p (t) calculated in step S406 with the inverse conversion function g of the conversion function f _λ used in step S208 of FIG. (S407) Thus, g (c _p (t)) is calculated. For example, in the Box-Cox conversion when λ = 0, since f _λ (x) = log (x), g (x) = exp (x). If no conversion is performed in step S208 in FIG. 17, step S407 is omitted.

Then, the demand amount predicted value calculation unit 106 uses the part ID of the prediction target part p as a key to obtain the cumulative product shipment quantity _Sp (t) in the elapsed t years of the prediction target part p from the cumulative product shipment quantity table 207. (S408).
Next, the predicted demand amount calculation unit 106 calculates the probability that the demand amount will be a “positive” value (S409).
As described above in step S303 in FIG. 18, the predicted value c _p (t) of the normalized component demand amount calculated in step S406 is a condition when the demand amount of the prediction target component p in the elapsed t years becomes positive. Expected value. That is, in the process so far, the case where the demand amount of the prediction target component p is “0” is intentionally excluded. Actually, there is a year when the demand amount becomes “0”, so it is conceivable that the predicted value calculated in a state where the demand amount is “0” intentionally removed is not accurate. In order to complement this, the processing of step S409 and step S410 is performed.

  Specifically, in step S409, the demand amount predicted value calculation unit 106 calculates the probability Q that the demand amount of the component is positive by the following equation (8).

Here, | · | is a symbol representing the number of elements in the set.
After step S409, the demand amount predicted value calculation unit 106 uses the probability Q that the demand amount calculated in step S409 is “positive”, and the demand amount predicted value x _p in the elapsed t years of the prediction target component _p. (T) is calculated by the following equation (9) (S410).

x _p (t) = Q × g (c _p (t)) × S _p (t) (9)

Here, g (c _p (t)) × S _p (t) is a predicted value of the demand amount when the demand amount of parts becomes positive in the elapsed t years of group G. To be precise, g (c _p (t)) is a value obtained by inversely converting the predicted value of the normalized component demand calculated in step S406 in step S407.

Next, the demand amount predicted value calculation unit 106 determines whether the prediction target year y is the prediction end year stored in the prediction information table 210 (S411).
As a result of step S411, when the prediction target year y is not the prediction end year (step S411 → No), the demand amount prediction value calculation unit 106 adds 1 to the variable y indicating the prediction target year (S412), and goes to step S404. Return and forecast the demand for the next forecast year.
As a result of step S411, when the prediction target year y is the prediction end year (step S411 → Yes), the processing unit 100 ends the process of step S106 of FIG. 15, and thus proceeds to step S107.

FIG. 21 is a diagram illustrating an example of a lifetime demand prediction screen displayed in the process of step S107 of FIG.
The lifetime demand prediction screen 2100 includes a prediction setting area 2110, a group display area 2120, a parts display area 2130, a group information display area 2140, a product information display area 2150, and a prediction result display area 2160.
First, the user selects and sets the prediction method and the order of the linear regression equation (if the selected prediction method is a linear regression equation) in the prediction method selection window 2111 in the prediction setting area 2110 (in the example of FIG. 21, “Linear regression prediction” is selected as the prediction method, and “order: 1” is selected).

  When the user selects a group displayed in the group display area 2120, the processing unit 100 refers to the group table 209 and displays a list of parts included in the group in the part display area 2130. At this time, information regarding the selected group is displayed in the group information display area 2140. At this time, the table referred to by the processing unit 100 is a group attribute table (not shown).

When the user selects a component displayed in the component display area 2130 ("PY900-017_2 electronic component B type c" is selected in the example of FIG. 21), information on the product in which the selected component is used. Is displayed in the product information display area 2150. Information displayed in the product information display area 2150 is based on the product part correspondence table 203 and the like.
The work so far is the process of inputting the prediction target part (S105) in FIG.

Then, when the user selects and inputs the prediction execution button 2112 in the prediction setting area 2110, the demand amount prediction value calculation unit 106 calculates the prediction value of the demand amount (S106) in FIG. 15, and the prediction graph display unit 107 displays the prediction result. The prediction result is displayed in the area 2160 (corresponding to the prediction graph display (S107) in FIG. 15).
In the prediction result display area 2160, the actual product shipment value is indicated by a bar graph, the actual part demand value (number of parts shipped) is indicated by a line graph 2161, and the predicted value of the demand amount of the part is indicated by a line graph 2162. It is shown in
Further, the processing unit 100 displays the contents of the demand amount prediction value table 212 as a list on the display device 50 separately from the lifetime demand prediction screen 2100 (corresponding to the prediction value display (S108) in FIG. 15).

<Summary>
According to this embodiment, in each elapsed year (elapsed period), by obtaining a linear regression coefficient using the normalized parts demand for the past k years (past k period) from the previous year as an explanatory variable, Alternatively, for parts (maintenance parts) that have only data up to the middle of the early stage, it is possible to predict lifetime demand for parts (maintenance parts) corresponding to the early stage, middle stage, and final stage. That is, according to the present embodiment, even if there are not enough actual values for predicting lifetime demand by the time series prediction method as described in Patent Document 1 and Non-Patent Document 1, according to the present embodiment, the component (maintenance) Lifetime demand for parts) can be predicted.

1 Lifetime demand forecasting device 10 Memory 20 Storage device 30 CPU
40 Input device (input unit)
50 Display device (display unit)
DESCRIPTION OF SYMBOLS 100 Processing part 101 Normalization parts demand amount calculation part 102 Prediction information setting part 103 Part grouping part 104 Linear regression formula construction part 105 Prediction object part setting part 106 Demand amount predicted value calculation part 107 Prediction graph display part 108 Prediction value display part 201 Product shipment quantity table (Product shipment quantity information)
202 Parts shipment quantity table (part shipment quantity information)
203 Product parts correspondence table (Product parts correspondence information)
204 Product Attribute Table 205 Parts Attribute Table 206 Maintenance Start Year Table 207 Cumulative Product Shipment Number Table 208 Normalized Parts Demand Amount Table 209 Group Table 210 Prediction Information Table 211 Linear Regression Equation Table 212 Demand Amount Prediction Value Table

Claims (5)

A lifetime demand forecasting method by a lifetime demand forecasting device that forecasts the demand for maintenance parts used in products,
The lifetime demand prediction device is at least in the storage unit,
Product shipment information on the product shipment start period for each product and the product shipment quantity of the product in each period after the product shipment start period,
Parts shipment unit information in which the number of parts shipped for maintenance parts in each period is registered,
Product part correspondence information in which maintenance parts used for the product are registered for each product;
And store
The lifetime demand forecasting device is
For each maintenance part registered in the part shipment quantity information, the product registered in the product part correspondence information is identified as the product that uses the maintenance part and has the earliest product shipment start period. And the product shipment start period of the identified product as the maintenance start period of the maintenance parts,
In the product shipment quantity information, for each elapsed period after the maintenance start period, out of the products registered in the product part correspondence information, all the products using the maintenance part from the maintenance start period. Calculate the total number of products shipped up to the number of elapsed years,
Calculate the normalized parts demand by dividing the number of parts shipped of the maintenance parts in the elapsed period by the cumulative number of products shipped;
When the order k of the linear regression equation, the prediction start period, and the prediction end period are input via the input unit,
Classify the maintenance parts registered in the part shipment quantity information into groups,
For each group, for each elapsed period after the elapsed (k + 1) period, the normalized part demand from the k period before the elapsed period to one period before the elapsed period is used as an explanatory variable. By calculating the coefficient of the variable, a linear regression equation for the explanatory variable is constructed.
When information on the prediction target part is input via the input unit,
Identify the group to which the prediction target part belongs,
For each prediction target period from the prediction start period to the prediction end period, convert the prediction target period to the number of elapsed periods of the prediction target part,
Based on the calculated linear regression equation, calculate a predicted value of the normalized part demand amount of the prediction target part in the converted number of elapsed times,
By multiplying the predicted value of the normalized part demand amount by the cumulative product shipment quantity in the converted elapsed period number, the demand amount of the maintenance part in the prediction target period of the prediction target part is predicted. Calculate the value
Lifetime demand forecasting method characterized by that. The lifetime demand prediction device is:
The normalized part demand is converted by a conversion function,
The lifetime demand prediction method according to claim 1, wherein the predicted value of the normalized part demand is converted by an inverse conversion function of the conversion function. The lifetime demand prediction device is:
When calculating the linear regression equation, use data in which the number of parts shipped is not 0,
After excluding the normalized parts demand data that will be outliers, calculate the linear regression equation,
Calculating the probability that the demand for maintenance parts will not be zero;
The lifetime demand prediction method according to claim 1, wherein the predicted value of the calculated demand amount of maintenance parts is multiplied by the probability.   The program which makes a computer perform the lifetime demand prediction method as described in any one of Claims 1-3. A lifetime demand prediction device that predicts the demand for maintenance parts used in products,
Product shipment information on the product shipment start period for each product and the product shipment quantity of the product in each period after the product shipment start period,
Parts shipment unit information in which the number of parts shipped for maintenance parts in each period is registered,
Product part correspondence information in which maintenance parts used for the product are registered for each product;
A storage unit storing
For each maintenance part registered in the part shipment quantity information, among the products registered in the product part correspondence information, the product with the earliest product shipment start period among the products using the maintenance part is selected. The product shipment start period of the identified product is defined as the maintenance start period of the maintenance part,
Shipment from the maintenance start period to the elapsed period of all products that use the maintenance part among the products registered in the product part correspondence information for each elapsed period after the maintenance start period Calculate the total number of products shipped,
A normalized part demand amount calculation unit that calculates a normalized part demand amount obtained by dividing the number of parts shipped of the maintenance part in the elapsed period by the cumulative number of product shipments;
When the order k of the linear regression equation, the prediction start period, and the prediction end period are input via the input unit,
A parts grouping unit for classifying maintenance parts registered in the parts shipment unit information into groups;
For each group, for each elapsed period after the elapsed (k + 1) period, the normalized part demand from the k period before the elapsed period to one period before the elapsed period is used as an explanatory variable. By calculating the coefficient of the variable, when the linear regression formula construction unit that constructs the linear regression formula for the explanatory variable and the information of the prediction target part are input via the input unit,
Identify the group to which the prediction target part belongs,
For each prediction target period from the prediction start period to the prediction end period, convert the prediction target period to the number of elapsed periods of the prediction target part,
Based on the calculated linear regression equation, calculate a predicted value of the normalized part demand amount of the prediction target part in the converted number of elapsed times,
By multiplying the predicted value of the normalized part demand amount by the cumulative product shipment quantity in the converted elapsed period number, the demand amount of the maintenance part in the prediction target period of the prediction target part is predicted. A demand amount predicted value calculation unit for calculating a value;
A lifetime demand prediction device characterized by comprising:

Images