JP5139967B2 - Demand forecasting device - Google Patents

Description

The present invention relates to a demand prediction apparatus that performs demand prediction of new products with different sales for each attribute of a due date.
Note that the attribute of the due date includes whether it is a day of the week, a holiday, a holiday, a vacation, an event period, or the like. In addition, new products include those in which new functions are added to the old products and those that are greatly improved.

  For companies that manufacture consumer products, maintaining an appropriate inventory level of the product is an important management issue. For that purpose, it is necessary to predict the future demand for the product as accurately as possible and to make a production plan and an inventory plan based on it. Standard products that already have long-term stable demand fluctuations are easy to predict by statistical methods.

  On the other hand, new products that have just been put on the market are difficult to predict by statistical methods because of the short sales period. In particular, for products in the field where new products are released one after another, it is necessary to predict the future sales based on the few sales results immediately after sales, and to adjust the production volume and inventory volume. In such a case, it is considered that prediction using a prediction model is effective for demand prediction of new products. Thus, conventionally, demand prediction has been performed using a prediction model to which certain parameters are applied.

  Conventionally, even if alternative products exist, in order to forecast the demand of the products, the new demand and replacement demand of the alternative products are calculated in consideration of the effect that a part of the new demand and replacement demand is replaced by the alternative products. Calculate the number of replacements by multiplying the demand for the product for each year, and subtract this from the demand for the product for each year to obtain the demand for the product when there is an alternative product. "Apparatus" is known (for example, see Patent Document 1).

  In addition, a life cycle pattern is obtained and accumulated based on the life cycle parameters calculated from the sales performance data, and the most similar to the life cycle pattern of the demand forecast target product is searched from here, and the searched life cycle pattern and sales performance Perform multiple demand forecast calculations using the data, set the evaluation standard limit values for these evaluation indexes, select the demand prediction calculation results that each evaluation index does not exceed the evaluation standard limit value, and select the optimal prediction data And “sales plan creation support system” that creates sales plan data from the created optimum prediction data is known (see, for example, Patent Document 2).

JP 2005-258799 A JP 2008 21020 A

However, many consumer products such as general households tend to sell more on Saturdays and Sundays than on weekdays (days except Saturdays and Sundays). Tend to sell more on weekdays than on holidays.
In the prior art, since the parameters of the prediction model are always the same value, the difference in sales by product day of the week is not reflected in the prediction result, so that there is a problem that the daily demand cannot be predicted sufficiently accurately.
Similar to the difference in the day of the week, there were similar problems in whether it was a holiday, holiday, vacation, or event period.

  A first demand prediction device according to the present invention is a demand prediction device that predicts a product demand prediction value of a new product by applying it to a prediction model, and predicts a prediction parameter for each attribute of a due date applied to the prediction model. Predicting the daily forecast value calculation means and the daily forecast value calculation means for calculating the product demand forecast value of the day to be forecasted by the forecast model using the estimation means and the forecast parameter corresponding to the day to be forecasted Product demand forecast value calculation means for calling each day of the period to be calculated and calculating a product demand forecast value for each day in the period is provided.

  A second demand prediction apparatus according to the present invention is a demand prediction apparatus that predicts a product demand prediction value of a new product by applying it to a prediction model, and predicts a prediction parameter for each day of the week to be applied to the prediction model. And the forecast parameter corresponding to the day of the week to be forecasted, the forecasted product demand forecast value of the day to be forecasted by the forecast model, and the forecasted daily forecast value calculator Product demand forecast value calculation means for calling each day of the period to be calculated and calculating a product demand forecast value for each day in the period is provided.

ADVANTAGE OF THE INVENTION According to this invention, even if it is a product from which sales differ according to the attribute of a period, the demand prediction apparatus which can estimate a demand correctly can be provided.
The attribute of the period includes, in addition to the day of the week, whether it is a holiday, a vacation, an event period, or the like.

  Hereinafter, an example of the best mode for carrying out the present invention (hereinafter referred to as “embodiment”) will be described in detail with reference to the accompanying drawings.

<First Embodiment>
FIG. 1 is a block diagram showing the overall configuration of a demand prediction apparatus 1 according to the first embodiment of the present invention.
This demand forecasting device 1 outputs product demand forecast value information 36 when product sales record information 11 is inputted, product sales record input means 12, product sales start date input means 13, and product sales record storage. Means 14, product sales start date storage means 15, prediction parameter storage means 16, prediction parameter estimation means 21, daily prediction value calculation means 25, prediction model selection information storage means 30, and prediction model selection means 31 A forecast period specifying means 32, a product demand forecast value calculating means 33, a product demand forecast value storage means 34, a product demand forecast value output means 35, and a product sales performance / predicted value display means 37. ing.

  The demand forecasting apparatus 1 typically includes a computer (hardware) having a calculation function, a control function, a storage function, an input / output function (including a display function), and the like, and a necessary program and data (software). By installing, each component described above is realized. Some or all of these components may be realized in hardware by, for example, burning data and programs into an EPROM.

  The “product” in the present embodiment is a product sold to a consumer by a supplier (particularly, a product sold to a consumer by a seller), and includes the concept of “product”. . The “product” is an industrial product that is mass-produced and preferably has a relatively short product life (for example, about 1 to 6 months). Of course, the “product” may be a finished product, a semi-finished product, a part, a raw material, or an agricultural, forestry and fishery product. Further, in addition to tangible objects, intangible objects such as music and computer programs may be used.

  In addition, “product” in this embodiment means “new product” in principle. “New products” include products that have been completely designed and manufactured, new products added to old products, and products that are partly or entirely improved (modified) from old products.

FIG. 2 is a diagram showing a data list of product sales performance information 11 input to the demand prediction apparatus 1.
As described above, the product sales record information 11 has a date field and a cumulative sales volume field, and is information that associates the date and the cumulative sales volume for each date up to this date.

  The product sales record input means 12 reads the product sales record information 11 and stores it in the product sales record storage means 14.

The product sales record storage means 14 adds “day of the week” information to the product sales record information 11 and stores it.
FIG. 3 is a diagram showing a data list of the product sales record information 200 stored in the product sales record storage unit 14.
In the product sales performance information 200, the first column is a date field, the second column is a day of the week field, and the third column is a cumulative sales volume field up to the date. It has a corresponding record (information).

  Returning to FIG. 1, the product sales start date input means 13 accepts the input of product sales start date information 300 (described later) and stores it in the product sales start date storage means 15.

The product sales start date storage means 15 stores the product sales start date information 300.
FIG. 4 is a diagram showing a data list of the product sales start date information 300 stored in the product sales start date storage means 15.

  Returning to FIG. 1, the prediction model selection means 31 receives an input for selecting one prediction model from a plurality of prediction models (in this embodiment, four from the first prediction model to the fourth prediction model), and the prediction model It is stored in the selection information storage means 30.

  The prediction model selection information storage unit 30 stores the prediction model selection information received by the prediction model selection unit 31 (any one of the first to fourth in this embodiment).

  The prediction parameter estimation means 21 is the product sales performance information 200 read from the product sales performance storage means 14, the product sales start date information 300 read from the product sales start date storage means 15, and the calculation result of the daily predicted value calculation means 25. Based on the above, a prediction parameter is estimated and stored in the prediction parameter storage means 16.

FIG. 5 is a block diagram showing the prediction parameter estimation means 21 in detail.
The prediction parameter estimation unit 21 includes a prediction parameter initialization unit 22 and a prediction parameter update unit 23.
The prediction parameter initialization unit 22 has a function of initializing the prediction parameters stored in the prediction parameter storage unit 16.
The prediction parameter update unit 23 has a function of updating the prediction parameter stored in the prediction parameter storage unit 16 with the prediction parameter estimated by the prediction parameter estimation unit 21.

  Returning to FIG. 1, the prediction parameter storage unit 16 stores the prediction parameters estimated by the prediction parameter estimation unit 21 separately according to the prediction model selection information read from the prediction model selection information storage unit 30.

FIG. 6 is a block diagram showing the prediction parameter storage unit 16 in detail.
The prediction parameter storage unit 16 includes a first prediction model parameter storage unit 17 that stores a prediction parameter corresponding to the first prediction model, and a second prediction model parameter storage unit 18 that stores a prediction parameter corresponding to the second prediction model. A third prediction model parameter storage unit 19 for storing a prediction parameter corresponding to the third prediction model, and a fourth prediction model parameter storage unit 20 for storing a prediction parameter corresponding to the fourth prediction model. . Although the case where there are four prediction models has been described, the prediction parameter storage means 16 has prediction model parameter storage means according to the number of prediction models that can be selected.

  Returning to FIG. 1, the daily forecast value calculation means 25 applies the forecast parameter for each estimated day of the week using the selected forecast model, and calculates the product demand forecast value for that day.

FIG. 7 is a block diagram showing the daily forecast value calculation means 25 in detail.
The daily prediction value calculation means 25 includes a first prediction model daily prediction value calculation means 26, a second prediction model daily prediction value calculation means 27, a third prediction model daily prediction value calculation means 28, a fourth prediction model daily prediction value calculation means 28, A prediction model daily prediction value calculation means 29. Although the case where the number of prediction models is four is shown, the daily prediction value calculation unit 25 has a prediction model daily prediction value calculation unit according to the number of prediction models that can be selected.

  The first forecast model daily forecast value calculation means 26 calculates the product demand forecast value f of the day using the first forecast model expressed by the following formula. The first prediction model is an improved PNE model. m is the value of the product sales performance on the day before the day to be predicted or the product demand prediction value. λ, γ, N, and b are prediction parameters.

  The second forecast model daily forecast value calculation means 27 calculates the product demand forecast value f of the day using the second forecast model expressed by the following equation. The second prediction model is an improved Bass model. m is the value of the product sales performance on the day before the day to be predicted or the product demand prediction value. λ, μ, N, and b are prediction parameters.

  The third forecast model daily forecast value calculation means 28 calculates the product demand forecast value f of the day using the third forecast model expressed by the following equation. The third prediction model is an improved inverse exponential model. m is the value of the product sales performance on the day before the day to be predicted or the product demand prediction value. λ, N, and b are prediction parameters.

  The fourth forecast model daily forecast value calculation means 29 calculates the product demand forecast value f of the day using the fourth forecast model expressed by the following equation. The fourth prediction model is an improved logistic model. m is the value of the product sales performance on the day before the day to be predicted or the product demand prediction value. μ, N, and b are prediction parameters.

  These prediction models are generally suitable examples, and depending on the type of product and sales method, these prediction models may be modified or used. Good.

  Of the prediction parameters, one or both of N and b may use a predetermined value (that is, a constant value that does not change within this period) on each day within the period to be predicted. . In these prediction models, the value of N represents the expected size of the market (the value of the quantity of products sold after sufficient time (days) has passed in the process of diffusion). The value of b represents the quantity of products sold by prior sales, reservation sales, or the like before the official sale of the product related to the prediction (before the period to be predicted). Therefore, the value of N and the value of b can be handled as those that do not substantially change within the period to be predicted.

  Returning to FIG. 1, the product demand forecast value calculation means 33 reads the product sales start date from the product sales start date storage means 15 and reads the product sales record from the product sales record storage means 14. And the daily forecast value calculation means 25 is called, the product demand forecast value of each day is calculated, and the value is obtained.

  The prediction period designating unit 32 receives designation of a prediction period (period to be predicted).

  The product demand forecast value storage means 34 stores the product demand forecast value for the designated forecast period.

  The product demand predicted value output means 35 reads the product demand predicted value from the product demand predicted value storage means 34 and outputs the information as product demand predicted value information 36 to the outside.

  The product sales performance / predicted value display means 37 reads the product demand predicted value from the product demand predicted value storage means 34, reads the product sales start date from the product sales start date storage means 15, and The product sales record is read and this information is displayed.

Next, the input procedure by the product sales record input means 12 will be described in detail.
FIG. 8 is a flowchart showing in detail a procedure for inputting product sales results.

  First, product sales record information 11 is opened (step S101).

  Subsequently, the line number i is set to 1. That is, i indicating the line number being read is set to 1, that is, the first line (step S102).

It is determined whether or not the end of the product sales information has been reached (step S103). That is, the following processing is repeated until all the rows of the product sales record information 11 are read.
If the end of the product sales information has been reached (Yes in step S103), the process proceeds to step S111 and subsequent steps.
If the end of the product sales information has not been reached (No in step S103), the process proceeds to the next step S104 and subsequent steps.

The date is read from the i-th line of the product sales record information 11 (step S104), and this is written to the i-th line of the product sales record storing means 14 (step S105).
Subsequently, the day of the week is calculated (step S106). Then, the calculated day of the week is written in the i-th row of the product sales record storage means 14 (step S107).

  The accumulated sales amount is read from the i-th line of the product sales record information 11 (step S108), and written in the i-th line of the product sales record storing means 14 (step S109).

When all the rows have been read, the product sales record information 11 is closed (step S111).
Subsequently, the data stored in the product sales record storage means 14 is rearranged so that the data is arranged in the order of date (ascending order), that is, the old data comes to the first line (step S112).

  When the input of the product sales record information 11 is completed, the main screen 600 shown in FIG. 9 is displayed on the product sales record / predicted value display means 37 (see FIG. 1). The upper portion of the main screen 600 is a graph and the lower portion is a table. The upper portion 601 of the graph and the table is the product sales performance / predicted value display means 37 (see FIG. 1), and the lower portion 602 of the table is the product sales start date input. Means 13 (see FIG. 1).

  Product sales performance information stored in the product sales performance storage means 14 is displayed in the graph and the table. There is a prediction execution button 605 on the upper right of the screen.

  When the product sales start date input means 13 in the main screen 600 of FIG. 9 is clicked with the mouse, that is, an arbitrary frame of the portion 602, the date corresponding to the frame is stored as the product sales start date storage means. 15 (see FIG. 1).

  The product sales start date storage means 15 stores the product sales start date in the form as shown in FIG.

  When the product sales start date is stored in the product sales start date storage means 15, as shown in the main screen (sales start date input) shown in FIG. A black circle 603) is displayed, and a vertical dotted line is displayed at a position corresponding to the date in the graph portion at the top of the main screen, indicating that this is the sales start date.

  When the prediction execution button 605 at the upper right of the main screen is pressed, a prediction model selection and prediction period designation screen shown in FIG. 11 is displayed. The upper part 701 of this screen is the prediction model selection means 31 (see FIG. 1), and the lower part 702 is the prediction period designation means 32 (see FIG. 1).

  In the demand prediction apparatus 1 of the present embodiment, four types of prediction models are incorporated in advance, and the user can select one type of prediction model from among them. When the user selects one model, a number indicating the selected model is stored in the prediction model selection information storage unit 30 (see FIG. 1).

  As shown in FIG. 12, the prediction model selection information storage means 30 stores the number of the selected prediction model.

  When the prediction period is specified by the prediction period specifying means 32 (reference numeral 702 in the figure) and the “OK” button is pressed, the prediction period specifying process shown in FIG. 13 is executed. Here, the product demand prediction value storage means 34 (see FIG. 1) performs a process of securing an area that can store all the prediction values during the prediction period. This will be described with reference to FIG.

First, the date designated by the forecast period designating means 32 is set to Q (step S401).
Subsequently, the date of the last line, that is, the line of the final result is acquired from the product sales result storage means 14, and this is set as P (step S402).
Next, the product demand forecast value storage means 34 is cleared (there is no line) (step S403). As shown in FIG. 14, the product demand forecast value storage means 34 at this stage is in a state where there is no data.

  While P is the date before Q (P is added day by day, but during all prediction periods), the following processing is repeated (steps S404 to S407).

First, one day is added to P, and the date of P is advanced by one day (the next day after P is set as a new P) (step S405).
The day of the week indicated by P is calculated, and is set as w (step S406).

  An empty row (record) is added to the product demand forecast value storage means 34, the date indicated by P in the year / month / day column (field) of this new row, and w in the day of week column (field). Write (step S407).

  As a result of the above processing, as shown in FIG. 15, the product demand forecast value storage means 34 has rows for the number of days in the forecast period, stores the date and day of the week, and the column of the cumulative sales volume is blank. It becomes.

  Subsequently, the prediction parameter estimation means 21 in FIG. 1 performs a prediction parameter value estimation process.

  The prediction parameter estimation processing is performed using a known optimization technique generally known to those skilled in the art. This technology includes the BFGS method, which is a quasi-Newton method, and a real-valued genetic algorithm.

FIG. 16 is a first surface of a flowchart showing the prediction parameter estimation processing, and FIG. 17 is a second surface of the flowchart showing the prediction parameter estimation processing.
Referring to FIG. 16, first, the number of the selected prediction model is acquired from the prediction model selection information storage unit 30, and this is set as k (step S201).
Subsequently, the parameter values of the k-th prediction model parameter storage means (17 to 20) are initialized by the prediction parameter initialization means 22 of the prediction parameter estimation means 21 (step S202).

FIG. 18 is a diagram showing a data list of the first prediction model parameter information 801 stored in the first prediction model parameter storage unit 17 (see FIG. 6).
In the first prediction model parameter information 801, each column corresponds to a field corresponding to N, λ, γ, and b that are parameters of the first prediction model, and each row corresponds to a record corresponding to each day of the week from Sunday to Saturday. To do. The first prediction model parameter information 801 has information on values of N, λ, γ, and b for each day of the week.

FIG. 19 is a diagram showing a data list of the second prediction model parameter information 802 stored in the second prediction model parameter storage unit 18 (see FIG. 6).
In the second prediction model parameter information 802, each column corresponds to a field corresponding to N, λ, μ, and b that are parameters of the second prediction model, and each row corresponds to a record corresponding to each day of the week from Sunday to Saturday. To do. The second prediction model parameter information 802 has information on values of N, λ, μ, and b for each day of the week.

FIG. 20 is a diagram showing a data list of the third prediction model parameter information 803 stored in the third prediction model parameter storage unit 19 (see FIG. 6).
In the third prediction model parameter information 803, each column corresponds to a field corresponding to N, λ, b that is a parameter of the third prediction model, and each row corresponds to a record corresponding to each day of the week from Sunday to Saturday. The third prediction model parameter information 803 has information on the values of N, λ, and b for each day of the week.

FIG. 21 is a diagram showing a data list of the fourth prediction model parameter information 804 stored in the fourth prediction model parameter storage unit 20 (see FIG. 6).
In the fourth prediction model parameter information 804, each column corresponds to a field corresponding to N, μ, and b that are parameters of the fourth prediction model, and each row corresponds to a record corresponding to each day of the week from Sunday to Saturday. The fourth prediction model parameter information 804 includes information on values of N, μ, and b for each day of the week.

  In the present embodiment, the prediction model can be selected from four types, and correspondingly, first to fourth prediction model parameter storage means (17 to 20) are provided. Each prediction model has an area for storing prediction model parameters for each day of the week.

  Returning to FIG. 16, the prediction parameter initialization means 22 is initialized to a random real value if it is a real value genetic algorithm, for example, depending on the optimization technique used (step S202).

  The previous error ep is initialized to an infinite value (step S203). This is used to determine the end of the loop, and since the previous value does not exist in the first loop, it is assumed to be infinite.

Subsequently, the current error e is initialized to 0 (step S204).
The product sales start date is acquired from the product sales start date storage means 15, and the same date as the product sales start date is searched for in the date column of the product sales result storage means 14. The row number with this date is set to the row number i (step S205).

  17, the following processing is repeated until the line number i becomes the last line of the product sales record storage means 14 (steps S211 to S217).

  The cumulative sales volume in the i-th row of the product sales record storage means 14 is set to m0 (step S212), the cumulative sales volume in the i + 1-th row is set to m1 (step S213), and the day of the week in the i + 1-th row is set to w (step S214). .

  The m0 and w are passed to the daily predicted value calculation means 25, a predicted value is acquired, and this predicted value is set to f1 (step S215).

The prediction error e1 for the day is obtained by m1-f1.
As shown in FIG. 22, based on the sales performance value m0 on the i-th day, the predicted value f1 on the i + 1-th day is calculated, and the difference between this f1 and the sales performance value m1 on the i + 1-th day is calculated as an error e1. It is said.
Returning to FIG. 17, the square of the error e1 is added to e that was initially initialized to 0 (step S216).

  The line number i is incremented by 1 (step S217), and the square error of the predicted value for each day is sequentially added to e until i becomes the last line of the product sales record storage means 14.

  In this way, the cumulative e of the square error of the predicted value is calculated for the period after the product sales start date input from the product sales start date input means 13 in the period in which the sales record value exists.

  When the squared error accumulated value e is calculated, this value is compared with the previous error ep, and when the absolute value of the difference is smaller than a predetermined threshold value, it is determined that the termination condition is satisfied (in step S221). Yes).

  When the end condition is not satisfied (No in step S221), the parameter value of the k-th prediction model prediction parameter storage unit (17 to 20) is initialized by the prediction parameter update unit 23 of FIG. 1 (step S222). The update method differs depending on the optimization technique used. For example, in the case of a real-valued genetic algorithm, the update is performed by crossover or mutation.

Thereafter, e is substituted for the previous error ep (step S223), and the row number i is incremented by 1 (step S224).
The process of calculating the square error cumulative value e is repeated for the updated prediction model parameters (step S204 and subsequent steps).

  When the termination condition is satisfied (Yes in step S221), the estimation of the prediction parameter is terminated.

  The daily forecast value calculation means 25 in FIG. 1 performs the procedure shown in FIGS. 23 and 24 from the outside on the previous day value (the previous day value is either the previous day sales actual value or the previous day demand forecast value). If yes, the current day of the week w is received and the daily forecast value for that day is calculated.

First, the number of the selected prediction model is acquired from the prediction model selection information storage unit 30, and this is set as k (step S301).
When k is 1 (Yes in step S302), the first predicted model daily predicted value calculation means 26 calculates the daily predicted value (steps S303 and S304).
When k is 2 (Yes in step S305), the daily prediction value is calculated by the second prediction model daily prediction value calculation means 27 (steps S306 and S307).
When k is 3 (Yes in step S308), the third predicted model daily predicted value calculation means 28 calculates the daily predicted value (steps S309 and S310).
If k is none of the above (No in steps S302, S305, and S308), k is 4 in the present embodiment, and thus the fourth prediction model daily prediction value calculation means 29 calculates the daily prediction value. (Steps S311 and S312).

The first prediction model daily prediction value calculation means 26 acquires parameters from the first prediction model parameter storage means 17 and calculates a prediction value by Equation 1.
The second prediction model daily prediction value calculation means 27 acquires parameters from the second prediction model parameter storage means 18 and calculates a prediction value by Expression 2.
The third prediction model daily prediction value calculation means 28 acquires parameters from the third prediction model parameter storage means 19 and calculates the prediction value by Expression 3.
The fourth prediction model daily prediction value calculation means 29 acquires parameters from the fourth prediction model parameter storage means 20 and calculates a prediction value by Expression 4.

  When the estimation of the parameters of the prediction model ends, the product demand prediction value is calculated by the product demand prediction value calculation means 33 in FIG.

This process will be described with reference to FIG.
First, the cumulative sales volume of the last line is acquired from the product sales record storage means 14, and this is set to p (step S501).

The line number i is set to 1 (step S502).
The following processing is repeated until the row number i exceeds the number of rows in the product demand forecast value storage means 34 (steps S503 to S508).

  The day of the i-th row of the product demand forecast value storage means 34 is acquired and set to w (step S504).

P and w are passed to the daily forecast value calculation means 25.
Here, the value of p is input to the sales performance value m0 of the previous day in FIGS. 23 and 24, and the value of w is input to the day of the week w of the day.
Then, since a daily forecast value is calculated, it is set as f (step S505).

The value of f is written in the cumulative predicted value of the i-th line of the product demand predicted value storage means 34 (step S506).
Subsequently, the value of f is substituted for p (step S507), the line number i is incremented by 1 (step S508), and the predicted value of the next day is sequentially calculated.

As shown in FIG. 26 , this is calculated from the final sales result value on the first day of the forecast, and then the forecast value on the second day from the forecast value on the first day, and the third day from the forecast value on the second day. This is a process of increasing the prediction sequentially, such as the predicted value of the eye.

  When the above processing is completed, the product demand prediction value storage means 34 is in a state where a value is set in the column of cumulative prediction values as illustrated in FIG.

  Further, as shown in FIG. 27, the main screen 600 displays the product demand forecast value stored in the product demand forecast value storage means 34 in a graph and a table. In this example, the predicted value is displayed as a white circle on the dotted line.

  Further, the product demand forecast value output means 35 of FIG. 1 outputs the contents of the product demand forecast value storage means 34 to the product demand forecast value information 36 according to the procedure shown in FIG.

  The product demand forecast value information 36 has a format as shown in FIG.

  The product demand forecast value output procedure will be described with reference to FIG.

  First, the product demand predicted value information 36 is opened (step S601), and the product demand predicted value information 36 is emptied (step S602).

  The line number i is set to 1 (step S603).

  The following processing is repeated until all rows of the product demand predicted value storage unit 34 are output (steps S604 to S610).

  The date is read from the i-th line of the product demand forecast value storage means 34 (step S605), and written to the i-th line of the product demand forecast value information 36 (step S606).

The day of the week is read from the i-th row of the product demand forecast value storage means 34 (step S607), and written in the i-th row of the product demand forecast value information 36 (step S608).
The cumulative predicted value is read from the i-th line of the product demand predicted value storage means 34 and written to the i-th line of the product demand predicted value information 36 (step S609).
The line number i is incremented by 1 (step S610), and the next line is sequentially processed.

Second Embodiment
FIG. 31 is a block diagram showing the overall configuration of the demand prediction apparatus 1b according to the second embodiment of the present invention.
In addition to the configuration of the demand prediction apparatus 1 of the first embodiment, the demand prediction apparatus 1b further includes a due date attribute input means 42 for inputting due date attribute information 41, and a due date attribute storage means for storing due date attribute information 41. 43.

  The date attribute information 41 includes, for example, national holidays, holidays such as administrative bodies and companies, vacations that members belonging to organizations, etc. arbitrarily or forcibly acquire, and periods of events related to product purchases. Have been described.

  The prediction parameter estimation means 21, the product demand prediction value calculation means 33, and the product sales performance / prediction value display means 37, in addition to (or in preference to) the day of the week to be predicted, the due date attribute storage means 43 The processing is performed with reference to the due date attribute information 41 stored in.

It is a block diagram which shows the whole structure of the demand prediction apparatus of 1st Embodiment by this invention. It is a figure which shows the data list of the product sales performance information input into a demand prediction apparatus. It is a figure which shows the data list of the product sales performance information stored in the product sales performance storage means. It is a figure which shows the data list of the product sales start date information stored in the product sales start date storage means. It is a block diagram which shows a prediction parameter estimation means in detail. It is a block diagram which shows a prediction parameter storage means in detail. It is a block diagram which shows a daily forecast value calculation means in detail. It is a flowchart which shows the input procedure of a product sales performance in detail. It is a screen figure which shows the main screen displayed on a product sales performance and prediction value display means (refer FIG. 1) (after product sales performance information reading). It is a screen figure which shows the main screen displayed on a product sales performance and prediction value display means (refer FIG. 1) (sales start date input). It is a screen figure which shows a prediction model selection and prediction period designation | designated screen. It is a figure which shows the data list of the number of the selected prediction model stored in the prediction model selection information storage means. It is a flowchart which shows a prediction period designation | designated process. It is a figure which shows the data list of the content of the product demand predicted value storage means of the cleared state. It is a figure which shows the data list of the product demand forecast value information after the forecast period specification stored in the product demand forecast value storage means. It is a 1st surface of the flowchart which shows a prediction parameter estimation process. It is a 2nd surface of the flowchart which shows a prediction parameter estimation process. It is a figure which shows the data list of the 1st prediction model parameter information stored in the 1st prediction model parameter storage means (refer FIG. 6). It is a figure which shows the data list of the 2nd prediction model parameter information stored in the 2nd prediction model parameter storage means (refer FIG. 6). It is a figure which shows the data list of the 3rd prediction model parameter information stored in the 3rd prediction model parameter storage means (refer FIG. 6). It is a figure which shows the data list of the 4th prediction model parameter information stored in the 4th prediction model parameter storage means (refer FIG. 6). It is explanatory drawing which shows the method of prediction of a period with a product sales track record concretely. It is a 1st surface of the flowchart which shows a daily forecast value calculation process. It is a 2nd surface of the flowchart which shows a daily forecast value calculation process. It is a flowchart which shows a product demand predicted value calculation process. It is explanatory drawing which shows the method of the prediction of a period without a product sales track record concretely. It is a screen figure which shows the main screen displayed on a product sales performance and prediction value display means (refer FIG. 1) (after prediction). It is a figure which shows the data list of the product demand forecast value information after the forecast execution stored in the product demand forecast value storage means. It is a flowchart which shows a product demand predicted value output process. It is a figure which shows the data list of the product demand forecast value information output. It is a block diagram which shows the whole structure of the demand prediction apparatus of 2nd Embodiment by this invention.

Explanation of symbols

1 Demand forecasting device (first embodiment)
1b Demand prediction device (second embodiment)
DESCRIPTION OF SYMBOLS 11 Product sales performance information 12 Product sales performance input means 13 Product sales start date input means 14 Product sales performance storage means 15 Product sales start date storage means 16 Prediction parameter storage means 17 First prediction model parameter storage means 18 Second prediction model parameter Storage means 19 Third prediction model parameter storage means 20 Fourth prediction model parameter storage means 21 Prediction parameter estimation means 22 Prediction parameter initialization means 23 Prediction parameter update means 25 Daily prediction value calculation means 26 First prediction model daily prediction value Calculation means 27 Second prediction model daily prediction value calculation means 28 Third prediction model daily prediction value calculation means 29 Fourth prediction model daily prediction value calculation means 30 Prediction model selection information storage means 31 Prediction model selection means 32 Prediction period Designation means 33 Product demand forecast value calculation means 3 Product demand forecast value storage means 35 Product demand forecast value output means 36 Product demand forecast value information 37 Product sales performance / predicted value display means 41 Due date attribute information 42 Due date attribute input means 43 Due date attribute storage means 200 Product sales performance information 300 Product sales Start date information

Claims (5)

A demand forecasting device that predicts a product demand forecast value of a new product by applying it to a forecasting model,
Prediction parameter estimation means for estimating a prediction parameter for each day of the week to be applied to the prediction model;
Daily forecast value calculation means for calculating the product demand forecast value of the day to be forecasted by the forecast model using the forecast parameter corresponding to the day of the week to be forecasted;
Anda product demand forecast value calculating means for calculating the product demand forecast value for each day within the period call for each day of the period to be predicted the daily predicted value calculating means,
The prediction model is
The product demand forecast value on the first day of forecast is calculated from the final sales actual value, and the product demand forecast value on and after the second day of forecast is calculated from the product demand forecast value on the previous day,
Assuming that the final sales performance value or the product demand forecast value on the day before the forecast date is m, and the forecast parameters of the forecast model for the day of the forecast date are λ, γ, μ, N, and b. The demand forecasting device f is characterized in that the product demand forecast value f on the day to be forecasted is represented by one of the following four mathematical formulas .

The prediction parameter estimation means calculates the prediction parameter for each square error between the value of each day of the product sales performance of the new product and the product demand prediction value of each day in the period in which the product sales performance exists. and, demand prediction apparatus according to claim 1, characterized in that the entire time period in which the sales history is present, total of the square error is estimated as a minimum. It has a product sales start date input means that accepts the input of the product sales start date,
The prediction parameter estimation means estimates the prediction parameter for each day of the period after the product sales start date in the period in which the value of the product sales performance exists.
The demand forecasting device according to claim 2 characterized by things. A prediction model selection unit that accepts one selection from the plurality of prediction models;
The daily prediction value calculation means includes a plurality of prediction model-specific daily prediction value calculation means corresponding to the plurality of prediction models, and uses the prediction parameters of the prediction model selected by the prediction model selection means. Calculating the product demand forecast value for the day to be forecasted,
The demand prediction apparatus according to claim 1, wherein: The demand forecasting device according to claim 1 , wherein the daily forecast value calculation means uses a predetermined value for N or b among the forecast parameters within the period to be forecasted.

Images