JP2016012224A - Demand prediction device and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To enhance prediction accuracy of demand prediction of seasonal merchandise and promotional merchandise.SOLUTION: A demand prediction device comprises: storage means storing past purchase quantity of each merchandise and variation factor information obtained by digitizing a variation factor varying the purchase quantity as result data; regression coefficient calculation means calculating a regression coefficient indicating a degree of contribution of the variation factor information to variations of the purchase quantity on the basis of first result data for a predetermined period in proximity to reference date and second result data for a predetermined period from date before a previous year corresponding to the reference date across a future direction out of the result data stored in the storage means; and prediction purchase quantity calculation means calculating prediction purchase quantity for each merchandise purchased on the reference date by using the regression coefficient calculated and the variation factor information on the reference date. The regression coefficient calculation means calculates the regression coefficient on the basis of each mean value of the purchase quantity in divided periods obtained by dividing a period associated with the second result data into a predetermined number of periods for merchandise having a difference between the purchase quantity in the first result data and the second result data equal to a predetermined value or more.

Description

  Embodiments described herein relate generally to a demand prediction apparatus and a program.

  Conventionally, POS (Point of Sales) systems have become widespread, and purchase (sales) data for each product can be easily collected. And the demand forecast according to goods, such as a retail store, based on the purchase data for every goods began to be used. For example, a demand forecast obtained by analyzing purchase data can be used to suppress the loss of unsold goods due to an excessive order quantity, or the occurrence of opportunity loss due to out of order goods due to an insufficient order quantity. Then, a technology is being proposed that combines demand forecasting with the digitization of business-to-business transactions using EDI (Electronic Data Interchange) to develop an automatic ordering system.

  As a demand forecasting technique that can be used for an automatic ordering system, there is a known multiple regression analysis. And a demand prediction can be performed for every goods by using a multiple regression analysis.

  By the way, among the products sold at the retail stores and the like, there are products whose sales numbers fluctuate in relation to a specific time and season (hereinafter referred to as seasonal products). In addition, among the products sold at the retail stores and the like, there are also products (hereinafter referred to as special sale products) that are provided at a low price on a specific date and time for attracting customers.

  However, in the demand forecast by the conventional multiple regression analysis, although the forecast is also made using past purchase data, the number of purchases of seasonal products and special sales products is not constant throughout the year. The prediction accuracy tends to be low.

  The problem to be solved by the present invention is to provide a demand prediction apparatus and program capable of improving the prediction accuracy of demand prediction of seasonal products and special sale products.

  The demand prediction device of the embodiment includes a storage means for storing past purchase numbers of each product and variable factor information obtained by quantifying the variable factors that change the number of purchases as actual data, and a reference for demand prediction The receiving means for receiving the date input, the first result data for a predetermined period closest to the reference date among the record data stored in the storage means, and the future from the date before the previous year corresponding to the reference date Calculated by a regression coefficient calculating means for calculating a regression coefficient indicating a degree that the variable factor information contributes to the fluctuation in the number of purchases based on the second result data for a predetermined period in the direction, and calculated by the regression coefficient calculating means Predicted purchase quantity calculating means for calculating, as the estimated purchase quantity, the number of purchases for each product purchased on the reference date using the regression coefficient and the variable factor information on the reference date. The regression coefficient calculating means divides a period related to the second actual data by a predetermined number for products in which the difference in the number of purchases between the first actual data and the second actual data is a predetermined value or more. The regression coefficient is calculated based on each average value of the number of purchases in the period.

FIG. 1 is a block diagram illustrating a system configuration including a demand prediction apparatus according to an embodiment. FIG. 2 is a block diagram illustrating a hardware configuration of a computer that is a main body of the demand prediction apparatus. FIG. 3 is a diagram for explaining the latest results and the results of the previous year. FIG. 4 is a diagram schematically showing multiple regression analysis. FIG. 5 is a flowchart showing a procedure of processing related to demand prediction in the system.

  FIG. 1 is a block diagram illustrating a system configuration including a demand prediction apparatus according to an embodiment. As shown in FIG. 1, the retailer system includes a demand prediction device 1 installed at a headquarters such as a retail store, a data collection / conversion system 2, a business support terminal 3, and a business system 4.

  The demand prediction apparatus 1 simulates the purchase status of a customer's product in relation to the sales status (sales price), display, sales promotion, etc. of the product on the sales side in a retail store. The demand forecasting device 1 uses the historical data indicating the past sales state and purchase status, and uses a regression coefficient of the number of purchases that can be regarded as not changing in time and a part that changes more quickly than the regression coefficient. Build a purchasing model for buyers divided into certain variable factors. Then, the demand prediction device 1 applies the sales state and the purchase environment on the reference date that is the target of the demand prediction to the fluctuation factor of the constructed purchase model, and the number of purchases for each product at that time Predict.

  The data collection / conversion system 2 collects past information such as purchase data, ordering data, inventory data, sales price / promotion data, and weather data of a plurality of products in the store from the present time to a predetermined past time as actual data. To do. The data collection / conversion system 2 converts the collected performance data into a predetermined data structure and transmits the data to various databases of the demand prediction apparatus 1. The data collection / conversion system 2 includes various data collection systems such as a known POS system, an ordering system, a delivery product inspection system, a weather data reception system, a local event input system, and a data conversion system. ing.

  The business support terminal 3 is a user interface for the demand prediction apparatus 1 and inputs data for controlling a purchase situation simulation. For example, the user inputs the sales state and purchase environment on the reference date that is the target of demand prediction from the business support terminal 3. In addition, when the initially set sales state and the predicted number of purchases with respect to the purchase environment are not the target values for the user of the demand prediction device 1, the user re-predicts from the business support terminal 3 by changing the planned selling price or the like. As described above, control data is input to the demand prediction apparatus 1.

  The business system 4 is a variety of systems that use the predicted purchase data calculated by the demand prediction device 1, and includes an ordering system, a processing instruction system, a price display system, a selling price registration system, and the like. For example, a processing instruction system that is one of the business systems 4 processes the predicted purchase data into recommended order data according to a predetermined ordering unit or order lead time, and converts it into a known ordering system that is one of the business systems 4. send. In the ordering system, ordering data is transmitted to a predetermined fresh processing center, wholesaler, producer, etc. for each product by a predetermined date and time.

  Each of the above demand forecasting device 1, various systems 2, 4 and business support terminal 3 includes a general-purpose computer system and a portable device having the same function as the general-purpose computer system. This is realized by executing a computer program corresponding to the system. In addition, large-scale data handled by the demand prediction device 1, the various systems 2, 4 and the business support terminal 3 is constructed as a database on a storage device of a general-purpose computer system by general-purpose database software. .

  Moreover, the structure of the demand prediction apparatus 1 shown in FIG. 1 and its peripheral system can be implemented with respect to the retail industry which consists of one store or several stores.

  In addition, the demand prediction apparatus 1 of this embodiment may be a service (application) provided in the form of cloud computing, for example, SaaS (Software as a Service).

  Here, the computer which becomes the main body of the demand prediction apparatus 1 is demonstrated. FIG. 2 is a block diagram illustrating a hardware configuration of the computer 100 that is the main body of the demand prediction apparatus 1. As shown in FIG. 2, a computer 100 includes a CPU (Central Processing Unit) 101 that performs information processing, a ROM (Read Only Memory) 102 that is a read-only memory that stores BIOS, and a RAM that stores various data in a rewritable manner. (Random Access Memory) 103 is provided.

  Further, the computer 100 functions as various databases and stores information using an HDD (Hard Disk Drive) 104 and a storage medium 110 which are storage units for storing various programs, distributes information to the outside, and externally. A medium reading device 105 such as a DVD drive for obtaining information, a communication control device 106 for transmitting information to other external devices via each communication line, and an LCD (displaying process progress, results, etc. to the user) A display unit 107 such as a liquid crystal display) and an input unit 108 that is a pointing device such as a keyboard and a mouse for a user to input commands and information to the CPU 101 are provided.

  In the computer 100, data transmitted / received between the above-described units is arbitrated by the bus controller 109.

  In such a computer 100, when the user turns on the power, the CPU 101 activates a program called a loader in the ROM 102, loads a program for managing the hardware and software of the computer called OS (Operating System) from the HDD 104 into the RAM 103, and Start the OS. Such an OS activates a program, reads information, and stores information in accordance with a user operation. As a representative OS, Windows (registered trademark) and the like are known. These operation programs running on the OS are called application programs. The application program is not limited to one that runs on a predetermined OS, and may be one that causes the OS to execute some of the various processes described below, or constitutes predetermined application software, an OS, or the like. It may be included as part of a group of program files.

  In general, application programs installed in the HDD 104 of the computer 100 include various types of optical disks such as CD-ROM and DVD, various magnetic disks such as various magneto-optical disks and flexible disks, and various types of semiconductor memories. The operation program recorded on the storage medium 110 such as a medium and recorded on the storage medium 110 is installed in the HDD 104. Therefore, the portable storage medium 110 such as an optical information recording medium such as a CD-ROM or DVD and a magnetic medium such as an FD can also be a storage medium for storing an application program. Furthermore, the application program may be imported from the outside via the communication control device 106 and installed in the HDD 104, for example.

  In the computer 100, when an application program operating on the OS is started, the CPU 101 executes various arithmetic processes according to the application program, and controls each unit intensively.

  The application program executed by the computer 100 which is the main body of the demand prediction apparatus 1 of the present embodiment is stored on a computer connected to a network such as the Internet and provided by being downloaded via the network. You may do it. Moreover, you may comprise so that the application program run with the computer 100 used as the main body of the demand prediction apparatus 1 of this embodiment may be provided or distributed via networks, such as the internet. Furthermore, an application program that is executed by the computer 100 that is the main body of the demand prediction apparatus 1 according to the present embodiment may be provided by being incorporated in advance in a ROM or the like.

  In the following, demand prediction processing, which is characteristic processing of the present embodiment, will be described among various arithmetic processing executed by the application program by the CPU 101 of the computer 100 that is the main body of the demand prediction device 1.

  As illustrated in FIG. 1, the demand prediction apparatus 1 includes a storage unit 11 that functions as a storage unit, a regression coefficient calculation unit 12 that functions as a regression coefficient calculation unit, and a predicted purchase quantity when the CPU 101 functions according to an application program. A predicted purchase number calculation unit 13 that functions as a calculation unit and an input reception unit 14 that functions as a reception unit are provided.

  The storage unit 11 includes a product master database 111, a purchase quantity database 112, an order database 113, an inventory database 114, a selling price / promotion database 115, a weather database 116, and the like. Further, the storage unit 11 manages not only the demand prediction apparatus 1 but also information used in other terminals and systems. In addition, the storage unit 11 is not only built in the demand prediction apparatus 1 but may be configured to be different from the demand prediction apparatus 1 as an external database server.

  The product master database 111 manages the product code, product name, product classification (category), etc. of each product. As the product classification, for example, a class code for classifying a product group including a plurality of products as a class, a line code for classifying a class group including a plurality of classes as a line, and a line group including a plurality of lines as a dept (department) Examples include hierarchical product classification codes such as dept codes to be classified.

  The purchase number database 112 is a database for managing purchase data. The number-of-purchases database 112 manages a history of the number of purchases from the present time to a predetermined past time for each product identified by the product code. Here, the time unit for managing various data may be determined according to the embodiment. For example, the number of purchases may be managed in units of one day, or the number of purchases may be managed in units of several hours.

  The order database 113 is a database for managing order data. The order database 113 manages, for each product identified by the product code, a history of the number of orders from the current time to a predetermined time in the past. The inventory database 114 is a database for managing inventory data. The inventory database 114 manages, for each product identified by the product code, a history of the number of stocks from the current time to a predetermined past time.

  In addition, the storage unit 11 has a database that stores information that can be a variation factor that fluctuates the number of purchases of the product group. For example, the selling price / sales promotion database 115 stores selling price data, which is a kind of variable factor, and sales promotion data indicating a sales promotion state. Specifically, the selling price / promotion database 115 is based on a standard price or a special sale indicating a price in a normal sales state applied to a product identified by a product code from a current time to a predetermined time in the past. It manages the sale price indicating the actual sale price that fluctuates from the standard price, the sale period indicating the date and period of application of the sale price, and the flyer placement indicating whether or not the flyer is placed. The weather database 116 stores weather data such as temperature, humidity, weather, and precipitation from the present time to a predetermined time in the past.

  As described above, each database in the storage unit 11 stores and manages the performance data collected by the data collection / conversion system 2.

  The regression coefficient calculation unit 12 calculates, for each product, a regression coefficient indicating the degree to which each data that can be a variation factor contributes to fluctuations in the number of purchases of the product based on the actual data stored and managed by the storage unit 11. . Furthermore, the regression coefficient calculation unit 12 calculates, for each product, a purchase number regression coefficient indicating the degree to which each purchase number of other products contributes to fluctuations in the purchase number of the product.

  This embodiment demonstrates the case where the goods which comprise a goods group are N types. N types of products are represented as product 1, product 2,..., Product n,.

Let Y _n be the number of purchases for this product n (n = 1, 2,..., N). That is, items 1, item 2, ..., item n, ... each purchase number of the product N, buying number Y _1, Y _2, ..., Y _n, ..., indicated as Y _N. The past purchase numbers are indicated as Y ~ ₁ , Y ~ ₂ , ..., Y ~ _n , ..., Y ~ _N.

In this embodiment, it is assumed that there are M types of data that can be a variation factor. And the variation factor with respect to the goods n is shown as _Xn1 , _Xn2 , ..., _Xnm , ..., _XnM (m = 1,2, ..., M). The variation factor X _nm (t) is a sales price, a day of the week, a holiday, a temperature, a precipitation amount, a form according to sales, and the like. It is assumed that data that may change the number of purchases is quantified. Here, for example, “standard” for sale at a standard price, “monthly special sale” for sale at a special price and a special sale period of two weeks or more, “special sale” other than monthly special sale, etc. It is done.

Then, in the regression coefficient calculation unit 12, the number of purchases Y to _{n in} the past predetermined period of the product n can be expressed by the following equation (1).
Y ~ _n = _{a n1 Y ~ 1 + a n2} Y ~ 2 + ... + a n (n1) Y ~ n1 + a n (n + 1) Y ~ n + 1 + ... + a nN Y ~ N + b n0 + b n1 X n1 + b _{n2 Xn2} + ... + b _{nM XnM} (1)

In the above formula (1), b _n0 is a regression coefficient of product n which is a constant term. b _n1 ,..., b _nM are regression coefficients for each data that can be M types of variation factors for the product n. a _n1 ,..., a _{n (n-1)} , a _{n (n + 1)} ,..., a _nN are purchase number regressions indicating the degree to which the number of purchases of other products fluctuates the number of purchases of product n It is a coefficient.

Then, the regression coefficient calculating section 12, the above formula (1) is generated for each item N kinds of, Y ~ ₁ of the formula past number of purchasing each product for each predetermined period or more M + N times (1), .., Y to _N , and regression coefficients b _n0 , b _n1 ,..., B _nM (n = 1,..., N) and purchase number regression coefficients a _{n1,. All of nN} (n = 1,..., N) are calculated.

<Predicted products>
The regression coefficient calculation unit 12 according to the present embodiment calculates the above-described regression coefficient and purchase number regression coefficient using a known multiple regression analysis.

  The regression coefficient calculation unit 12 uses the record data stored in the storage unit 11 for calculating these regression coefficients. In the present embodiment, the regression coefficient calculation unit 12 calculates a regression coefficient and a purchase number regression coefficient using actual data that can be the number of purchases and a variation factor in a predetermined period.

  Specifically, the regression coefficient calculation unit 12 is based on a reference date that is a target of demand prediction, actual data for the past predetermined period closest to the reference date (hereinafter referred to as the latest actual result), and before the reference date. A regression coefficient and a purchase number regression coefficient are calculated using actual data for a predetermined period from the date of the fiscal year to the future direction (hereinafter referred to as the previous fiscal year result).

  Here, FIG. 3 is a diagram for explaining the latest results and the results of the previous year. In the example of FIG. 3, the reference date that is the target date for the demand prediction is the current date (the current day), December 1, 2014. The result data R for the past 35 days (October 27, 2014 to November 30, 2014) closest to the reference date is set as the first latest result A1. Further, the record data R for the past 70 days (September 22, 2014 to November 30, 2014) that is closest to December 1, 2014 is the second most recent record A2.

  The actual data R for the 35 days (December 2, 2013 to January 5, 2014) from the date of the previous year corresponding to the reference date (December 2, 2013) to the future direction is The previous year's result is B1. Furthermore, the actual data R for 70 days (December 2, 2013 to February 9, 2014) from the date of the previous year (December 2, 2013) corresponding to the reference date to the future direction is stored in the second The previous year's result is B2. In FIG. 3, the date of the previous year corresponding to the reference date is the day after the reference date. However, the present invention is not limited to this, and may be the same day as the reference date. In addition, the results for the previous year are not limited to the results data for the previous year, and results data for the previous year may be used.

  Then, the regression coefficient calculation unit 12 uses the latest results and the previous year results in the following three patterns according to the presence / absence of the result data related to each product and the tendency of the result data.

<<< Pattern P1 >>
(1) For products for which the previous year results (first previous year results B1, second previous year results B2) do not exist, and (2) products for which the difference between the latest results and the previous year results is less than or equal to the first threshold, the pattern The second most recent record A2 is used as P1. Here, the first threshold value is preferably a value indicating a slight difference, for example, within 20%. In this way, examples of the products for which there is a slight difference between the latest results and the results of the previous year include standard products and daily products that are constantly purchased throughout the year regardless of the season.

<<< Pattern P2 >>
(1) Products whose sales are below the second threshold in the second most recent record A2, or (2) Products where the difference between the most recent record and the previous year is above the third threshold, the pattern P2 is the second previous year record B2. Is used. Here, the second threshold value is preferably a value indicating that the number of sales is small, such as 7 or less (0.1 / day). The third threshold value is preferably a value indicating a large difference, for example, 80% or more. In this way, products that have a large difference between the latest results and the previous year's results are, for example, products that are purchased in relation to a specific time or season (seasonal products) or products that are inexpensively offered at a specific date and time (special sales) Product).

  By the way, in this embodiment, the regression coefficient calculation part 12 does not take the simple average about the number of purchases of seasonal goods and sale goods in the period (for example, 70 days) of 2nd last year results B2. The regression coefficient calculation unit 12 calculates the regression coefficient and the purchase number regression coefficient by using the average value of the number of purchases in a divided period obtained by dividing the period of the second previous year result B2 by a predetermined number (for example, divided into two). In this way, by further dividing the period of the second previous fiscal year result B2, it is possible to moderately capture the months of special sales and trends of seasonal products.

  In the present embodiment, the period of the second previous year result B2 is set to 70 days. However, the period is not limited to this, and can be set to an arbitrary period. The reason for setting the period of the second previous year result B2 to 70 days is that the period is appropriate to include the seasonal factor. Note that, for example, when the consumption tax changes, a long period (for example, three months) may be set.

  Moreover, in this embodiment, although 2nd previous year results B2 was divided into 2, it is not restricted to this, It can set to arbitrary division | segmentation numbers. For example, you may make it divide | segment the period of 2nd last year results B2 for every month.

<<< Pattern P3 >>
(1) For a product that belongs to a predetermined product category or a predetermined product category, the first latest result A1 and the first previous year result B1 are used as the pattern P3. The pattern P3 may also be used for products that do not meet the conditions of the patterns P1 and P2.

  The period referred to in the pattern P3 is not limited to the 70 days of the first most recent record A1 and the first previous year record B1, but may be a specific day or period included in the 70 days. For example, a specific day or a specific period having weather data close to the expected temperature or the weather on the reference day of the first most recent record A1 and the first previous year record B1 may be referred to. In addition, for example, when the past one week's performance data is compared with the previous performance data, and a change in season can be detected from the temperature difference between the two periods, the last one week's performance data and the first previous year You may use track record B1.

  In addition, the regression coefficient calculation unit 12 may obtain the actual data of the entire product category to which the product belongs (the latest result, the previous year) for a product for which neither the previous year result nor the latest result exists or a product with very little past result. The regression coefficient is calculated using the average value of the results) and the result data of the similar products similar to the product (the latest results and the results of the previous year).

  Note that the demand prediction may be performed for each store, or the total demand prediction in all stores may be calculated. When the demand prediction is performed for each store, the product group to be predicted is all the products purchased in the past at the target store. However, products that can be purchased only a few in a month and for which a rapid increase in the number of purchases can not be expected in the future may be excluded from the product group.

  In this way, the regression coefficient calculation unit 12 selects the pattern P1 to P3 according to the latest record of each product and the tendency of the number of purchases in the previous year's record, so that the latest record used to calculate the regression coefficient is selected. And change the period length of the previous year's results. As described above, it is preferable to set each pattern so that the sum of the period lengths related to the latest results and the previous year results is the same number of days (70 days in the example of FIG. 3).

In the present embodiment, the regression coefficient calculating section 12 is used for calculation of the regression coefficients, the number of purchase in the past predetermined period of time t of the product n (= 1,2 ..., N) , Y ~ n (t) (= Y ~ ₁ (t), Y ~ ₂ (t), ..., Y ~ _n (t), ..., Y ~ _N (t)). This predetermined period t may be a one-day interval, an hour interval, or a one-minute interval. Then, when calculating the regression coefficient, the total number of purchases Y _n (t) purchased in such a predetermined period t is read from the purchase number database 112.

<Variation factor>
Then, since the regression coefficient calculation unit 12 uses the calculation of the regression coefficient, the data X that can be past variation factors of M types (m = 1, 2,..., M) for the product n (= 1, 2,..., N). _nm (t) is read from each database in the storage unit 11 every predetermined period t (t = 1, 2,..., T).

That is, the regression coefficient calculation unit 12 reads the following data from the storage unit 11 as past data that is a variation factor for the product n (= 1, 2,..., N).
X _n1 (0), X _n1 (1), ..., X _n1 (t), ..., X _n1 (T)
X _n2 (0), X _n2 (1), ..., X _n2 (t), ..., X _n2 (T)
......
X _nM (0), X _nM (1), ..., X _nM (t), ..., X _nM (T)

  Here, in order to solve the simultaneous equations, the total number T of the predetermined period needs to be M + 1 or more. In other words, it is necessary to satisfy the relationship T ≧ M + 1 (t = 0, 1, 2,..., T) (m = 1, 2,..., M).

<Regression coefficient calculation>
Next, a regression coefficient calculation method will be described. First, when the past data X _nm (t) serving as a predictor is used as a known regression coefficient calculation method, the estimated purchase number Y of the product n in a predetermined period t (t = 0, 1,..., T) _n (t) can be shown by the regression equations (2-0) to (2-T) shown below.

Y _n (0) = b _n0 + b _n1 X _n1 (0) + b _n2 X _n2 (0) +... + B _nM X _nM (0) (2-0)
Y _n (1) = b _n0 + b _n1 X _n1 (1) + b _n2 X _n2 (1) +... + B _nM X _nM (1) (2-1)
...
Y _n (t) = b _n0 + b _n1 X _n1 (t) + b _n2 X _n2 (t) +... + B _nM X _nM (t) (2-t)
...
Y _n (T) = b _n0 + b _n1 X _n1 (T) + b _n2 X _n2 (T) +... + B _nM X _nM (T) (2-T)

At this time, in the multiple regression analysis, when E () representing an expected value is used, the square sum Q of the residuals can be expressed by Expression (3). _Y˜n (t) is the actual number of purchases in a predetermined period t.

_{Q = E (Y n (t} ) -Y ~ n (t)) 2 ... (3)

Then, M + 1 (m = 0, 1, 2,..., M) regression coefficients b _nm (b _nm = b _n0 , b _n1 , b _n2 ,..., B _nm , for the constant term of the product n and the variation factor X _nm ..., B _nM ) are derived so that Q in the equation (3) is minimized. FIG. 4 is a diagram schematically showing multiple regression analysis. In other words, the regression curve of FIG. 4 is obtained to approximate each measured value.

And in the conventional method, by fixing the regression coefficient b _nm of the regression equations (2-0) to (2-T), and substituting the prediction data X _nM that becomes a variation factor in the period to be predicted, predicted value Y _n future purchase number of the product n is calculated from equation (4) below.
Y _n = b _n0 + b _n1 X _n1 + b _n2 X _n2 +... + B _nM X _nM (4)

  By the way, in the multiple regression analysis mentioned above, as shown by Formula (4), the influence of the interaction of other goods is not included at all. However, in reality, the interaction of each product occurs.

  As the increase / decrease pattern of the number of purchases between the two products A-B, there are the following four types of interactions. That is, (1) when the purchase number of the product A increases, the purchase number of the product B increases, (2) when the purchase number of the product A increases, the purchase number of the product B decreases, and (3) the product A When the number of purchases decreases, the number of purchases of the product B increases. (4) When the number of purchases of the product A decreases, the number of purchases of the product B decreases.

  And these four types of interactions should just be described with respect to all the prediction object goods groups. Therefore, in the present embodiment, in the regression equation for calculating the number of purchases of the product A, all the purchase numbers other than A in the same period are shown as variation factors of the number of purchases of the product A.

Therefore, in the present embodiment, the number of past purchases Y ~ ₁ (t), Y ~ ₂ (t), ..., Y ~ _n- of the other products in the product group excluding the product _{n. 1 (t), Y ~ n} + 1 (t), ..., Y 1, Y 2 corresponding to _{Y ~ n (t), ...} , Y n-1, Y n + 1, ..., a Y _n, goods It is used as a purchase quantity variation factor in the regression equation of n.

Then, the purchase number of items n fluctuation factor _{Y 1, Y 2, ...,} Y n1, Y n + 1, ..., (N-1) a purchase number regression coefficients for _{Y N a n1, a n2,} ..., It is defined as an _(n-1) , an _{(n + 1)} , ..., _anN . Then, as the regression equation for calculating the purchase number Y _n of the product n, all changes factors, purchasing rate variability factor, and the formula (5) using the regression coefficient thereof can be set.

_{Y n = a n1 Y 1 +} a n2 Y 2 + ... + a n (n1) Y n1 + a n (n + 1) Y n + 1 + ... + a nN Y N + b n0 + b n1 X n1 + b n2 X _n2 + ... + b _nM X _nM (5)

The total number of regression coefficients in equation (5) is (N−1) + (M + 1) = N + M. Therefore, the past data set of the fluctuation factors and purchase quantity fluctuation factors extracted from the storage unit 11 is also set to N + M or more. That is, Y ~ ₁ (t), Y ~ ₂ (t), ..., Y ~ _n-1 (t), Y ~ _{n + 1} (t), ..., Y ~ _N (t) and _Xn1 (t ), X _n2 (t),..., X _nM (t) (t = 0, 1, 2,..., T), it is necessary that T ≧ N + M−1.

When the condition of T ≧ N + M−1 is satisfied, the regression coefficient calculation unit 12 uses the regression equation (5) and a known regression coefficient calculation method to purchase all N + M purchases for the product n. number regression coefficients _{a n1, a n2, ...,} a n (n1), a n (n + 1), ..., a nN and regression coefficient _{b n0, b n1, b n2} , ..., can be calculated b _nM .

  The predicted purchase number calculation unit 13 is based on the regression coefficient and purchase number regression coefficient calculated by the regression coefficient calculation unit 12, and the purchase number and variation factor data for each product stored in the storage unit 11. For each product, a predicted number of purchases indicating the number of purchases that are likely to be purchased within a predetermined period after the above-mentioned past predetermined period, in other words, after the present is calculated.

  The predicted purchase number calculation unit 13 uses the following equation (6) to calculate the predicted purchase number of the product n included in the N products that constitute the product group.

b _n0 + b _n1 X _n1 + b _n2 X _n2 +... + b _nM X _nM = −a _n1 Y ₁ −a _n2 Y ₂ −... −an _(n−1) Y _n−1 + Y _n −an _{(n + 1) )} Y _{n + 1} −... −a _nN Y _N (6)

Then, the predicted purchase quantity calculation unit 13 applies the data that becomes a variation factor for each product for a predetermined period after the present to X _n1 ,..., X _nM of the formula (6), and then sets the formula (6) to N calculated kinds of generated for each product, based on the generated n equations, the predicted purchasing number of n each product constituting the product group _{Y 1, Y 2, ...,} Y n, ..., the Y _n To do.

<Estimated purchases>
First, before explaining the estimation method of the future purchase quantity, the past purchase quantity estimation technique will be explained. First, for each of the N products that make up the product group, using the variation factor, purchase number variation factor, regression coefficient, purchase number regression coefficient, etc. as described above, regression equation groups (7-1) to (7) -N) can be set.

Y ₁ = a ₁₂ Y ₂ + a ₁₃ Y ₃ + ... + a _1N Y _N + b ₁₀ + b ₁₁ X ₁₁ + b ₁₂ X ₁₂ + ... + b _1M X _1M (7-1)
Y ₂ = a ₂₁ Y ₁ + a ₂₃ Y ₃ + a ₂₄ Y ₄ + ... + a _{2 N} Y _N + b ₂₀ + b ₂₁ X ₂₁ + b ₂₂ X ₂₂ + ... + b _2M X _2M (7-2)
...
_{Y n = a n1 Y 1 +} a n2 Y 2 + ... + a n (n1) Y n1 + a n (n + 1) Y n + 1 + ... + a nN Y N + b n0 + b n1 X n1 + b n2 X _n2 + ... + b _nM X _nM (7-n)
...
Y _N = a _N1 Y ₁ + a _N2 Y ₂ + ... + a _{N (N-2)} Y _N-2 + a _{N (N-1)} Y _N-1 + b _n0 + b _n1 X _n1 + b _n2 X _n2 + ... + b _nM X _nM (7-N)

  In the regression equation groups (7-1) to (7-N), if T ≧ N + M−1 is satisfied for each past predetermined period (t = 0, 1, 2,..., T), the right side Are all known, the following equations (8-1) to (8-N) can be derived.

Y ₁ (t) = a ₁₂ Y- ₂ (t) + a ₁₃ Y- ₃ (t) + ... + a _1N Y- _N (t) + b ₁₀ + b ₁₁ X ₁₁ (t) + b ₁₂ X ₁₂ (t) + + B _1M X _1M (t) (8-1)
_{Y 2 (t) = a 21} Y ~ 1 (t) + a 23 Y ~ 3 (t) + a 24 Y ~ 4 (t) + ... + a 2N Y ~ N (t) + b 20 + b 21 X 21 (t) + b ₂₂ X ₂₂ (t) + ... + b _2M X _2M (t) (8-2)
...
Y _n (t) = a _n1 Y ~ ₁ (t) + a _n2 Y ~ ₂ (t) + ... + an _(n-1) Y ~ _n-1 (t) + an _{(n + 1)} Y ~ _{n + 1} (t) +... + An _n Y ~ _N (t) + b _n0 + b _n1 X _n1 (t) + b _n2 X _n2 (t) + ... + b _nM X _nM (t) (8-n)
...
Y _N (t) = a _N1 Y ~ ₁ (t) + a _N2 Y ~ ₂ (t) + ... + a _{N (N-2)} Y ~ _N-2 (t) + a _{N (N-1)} Y ~ _{N- 1} (t) + b _n0 + b _n1 X _n1 (t) + b _n2 X _n2 (t) +... + B _nM X _nM (t) (8−N)

The past purchase quantity estimated value Y _n (t) can be calculated under the condition that the square sum Q of the residuals in the multiple regression analysis is minimized.

By the way, in the calculation of the predicted number of purchases for a predetermined period in the future for the product n, all regression coefficients a _nk (n ≠ k) and b _{nm on} the right side of the regression equation groups (7-1) to (7-N) are determined. _Therefore, data X _nM that becomes a variation factor in the predetermined future period can be derived. However, the right-hand side of Y _k (k ≠ n), since Y _k is to mean the predicted value of the future purchases number of other products other than products n, is unknown at future time, the only equation (6), product It is not possible to calculate the predicted number of future purchases for n.

Therefore, in order for the forecast purchase number calculation unit 13 according to the present embodiment to calculate, terms relating to the variation factor X _nm and the regression coefficients b _n0 , b _n1 , b _n2 ,..., B _{nm ,.} B _n = b _n0 + b _n1 X _n1 + b _n2 X _n2 +... + B _nM X _nM Then, by transforming the regression equation groups (7-1) to (7-N) using the formula, a simultaneous purchase number prediction equation group (simultaneous purchase number prediction equation group (N) of N products constituting the product group ( 9-1) to (9-N) can be constructed.

B ₁ = Y ₁ −a ₁₂ Y ₂ −a ₁₃ Y ₃ −... −a _1N Y _N (9-1)
B ₂ = −a ₂₁ Y ₁ −Y ₂ −a ₂₃ Y ₃ −a ₂₄ Y ₄ ... −a _2N Y _N (9-2)
...
_{B n = -a n1 Y 1 -a} n2 Y 2 - ... -a n (n1) Y n1 + Y n -a n (n + 1) Y n + 1 - ... -a nN Y N ... ( 9-n)
...
B _N = −a _N1 Y ₁ −a _N2 Y ₂ −... −a _{N (N−2)} Y _N−2 −a _{N (N−1)} Y _N−1 −Y _N (9−N)

Here, in the simultaneous purchase number prediction formula groups (9-1) to (9-N), a _nn = 1 (n = 1, 2,..., N), B _n = −a _n1 Y ₁ −a _n2 Y ₂ −... −an _(n−1) Y _n−1 + an _n Y _n −an _{(n + 1)} Y _{n + 1} −... _{−an N} Y _N 1) ~ (9-n) is unknown Y _n is the n, the number of equations is also the n. Therefore, if the matrix composed of the coefficients a _ij (i, j = 1, 2,..., N) is regular, the predicted purchase quantity calculation unit 13 uses a known simultaneous linear equation solving method to calculate the product group. future predictions purchase number for the n items constituting _{Y 1, Y 2, ...,} Y n, ..., it can be calculated Y _n.

Then, vector b ^T = (B ₁ , B ₂ ,..., B _n ,..., B _N ), vector y ^T = (Y ₁ , Y ₂ ,..., Y _n ,..., Y _N ) and matrix A = If {a _ij } (i, j = 1, 2,..., N), the simultaneous purchase quantity prediction formula group (9-1) to (9-N) is Ay = b, and if modified, y = A ^{−. 1} b (10) can be derived.

Then, the predicted purchase quantity calculation unit 13 can calculate the vector y indicating the future predicted purchase quantity when the vector b is arbitrarily received after calculating the inverse matrix A ⁻¹ which is a coefficient matrix.

The vector b is composed of a variation factor X _nm or the like, and in other words includes a sales price, a day of the week, a holiday, a temperature, a precipitation amount, a sales type, and the like. For this reason, the forecast purchase number calculation unit 13 accepts the input of the vector b for various conditions on the reference date as the target of the demand prediction, so that the purchase number prediction for the sales price on the reference day and the temperature ( The forecast of the number of purchases can be calculated for the predicted temperature.

  As a result, the demand prediction apparatus 1 according to the present embodiment allows the optimal sales price (sale price) of the product to maximize the sales for any product, or the optimal sales price (sale price) to maximize the sales of all products. ) And the like can be calculated.

  The input receiving unit 14 receives an input from the business support terminal 3 or the like. For example, the input receiving unit 14 receives input of data serving as a variable factor of the sales state (scheduled selling price, etc.) and purchase environment (sales form, etc.) on the reference date that is the target of demand prediction from the business support terminal 3. Further, the input receiving unit 14 receives input of control data for instructing to re-predict from the business support terminal 3 by changing a variable factor such as a planned selling price. Here, by setting the vector b derived from the variation factor that has received the input in the predicted purchase number calculation unit 13, the predicted sales number can be calculated based on the sales state, the purchase environment, and the like.

  Next, the process which concerns on the demand prediction in the system containing the demand prediction apparatus 1 of this embodiment is demonstrated. FIG. 5 is a flowchart showing a procedure of the above-described processing in the system according to the present embodiment.

  First, the regression coefficient calculation unit 12 of the demand prediction apparatus 1 sets a reference date that is a target of demand prediction (step S11). Subsequently, the regression coefficient calculation unit 12 of the demand prediction device 1 sets a range of actual data to be referred to from the storage unit 11 (the latest actual result and the previous year actual result) based on the reference date set in Step S11 (Step S12). . Here, for example, the reference date may be set in accordance with an instruction via the business support terminal 3, or after the business of the current day is completed, the date of the next day (year / month / day) is automatically set as the reference date. It is good also as a form set to.

  Next, the regression coefficient calculation unit 12 of the demand prediction apparatus 1 uses the pattern described above for the latest results and the previous year results used for calculating the regression coefficient and the purchase number regression coefficient of each product according to the presence / absence or trend of the result data. Selection is made from P1 to P3 (step S13). And the regression coefficient calculation part 12 calculates a regression coefficient and a purchase number regression coefficient from the latest results and the results of the previous year based on the selected pattern (step S14). For products for which neither the previous year's results nor the latest results exist, or for products with very few past results, the regression coefficient calculation unit 12 calculates the average value of the results data for the entire product category to which the product belongs or the product The regression coefficient and the purchase number regression coefficient are calculated using the result data of similar products similar to.

  Subsequently, the input receiving unit 14 of the demand prediction device 1 receives an input of a variation factor (data) on the reference date from the business support terminal 3 or the like (step S15). The weather data on the reference date is extracted from the storage unit 11 as the weather forecast data, and therefore does not need to be input, but may be input via the business support terminal 3 or the like. In addition, when the variation factor on the reference date is held in advance in the storage unit 11 or the like, the input reception unit 14 may be configured to receive an input by reading the variation factor.

  The predicted purchase number calculation unit 13 of the demand prediction device 1 calculates the predicted purchase number of each product based on the variation factor that has received the input, the calculated regression coefficient, and the purchased number regression coefficient (step S16). Not only the predicted number of purchases of each product but also the total of sales and profits may be calculated.

  Next, the input reception unit 14 of the demand prediction apparatus 1 determines whether or not control data instructing re-prediction is input from the business support terminal 3 or the like (step S17). Here, when an input of control data instructing to change and predict again by changing the variation factor is received from the business support terminal 3 or the like (step S17; Yes), the procedure returns to step S15 to change the variation factor. Then, the predicted purchase number is calculated again.

  In step S17, when the control data for instructing the re-prediction is not input, that is, the instruction data for allowing the prediction result is input (step S17; No), the business system 4 uses the prediction calculated by the demand prediction device 1. Based on the purchase data (predicted number of purchases), the current planned selling price, etc., an order is placed for the product (step S18), and the process is terminated.

  In this process, the calculation of the predicted number of purchases is terminated according to an instruction from the user, but the present invention is not limited to this. For example, a reference value such as the predicted number of purchases or sales amount may be determined in advance, and if the prediction result satisfies the reference value, processing such as ordering may be performed without waiting for an instruction from the user.

  As described above, according to the demand prediction apparatus 1 according to the present embodiment, the regression coefficient calculation unit 12 for the seasonal product and the sale product, the date before the previous year corresponding to the reference date that is the target of the demand prediction. A regression coefficient is calculated based on each average value of the number of purchases in a predetermined period obtained by dividing a predetermined period from the future direction to the future direction, and using the regression coefficient and various fluctuation factors on the reference date, The number of purchases for each product to be purchased is calculated as the predicted number of purchases. As a result, it is possible to moderately capture trends in seasonal products and seasonal products with a lot of special sales, and improve the prediction accuracy of demand forecasts related to seasonal products and special sales products. It can be carried out.

  As mentioned above, although embodiment of this invention was described, the said embodiment was shown as an example and is not intending limiting the range of invention. The above-described embodiment can be implemented in various other forms, and various omissions, replacements, and changes can be made without departing from the spirit of the invention. The above-described embodiments and modifications thereof are included in the scope and gist of the invention, and are included in the invention described in the claims and equivalents thereof.

DESCRIPTION OF SYMBOLS 1 Demand prediction apparatus 11 Memory | storage means 12 Regression coefficient calculation means 13 Predicted purchase number calculation means 14 Acceptance means

JP 2012-150671 A

Claims (6)

Storage means for storing past purchase number of each product and variation factor information obtained by quantifying a variation factor that fluctuates the purchase number as actual data;
An accepting means for accepting an input of a reference date to be a target of demand forecast;
Of the performance data stored in the storage means, first performance data for a predetermined period closest to the reference date, and second for a predetermined period from the date before the previous year corresponding to the reference date to the future direction. A regression coefficient calculating means for calculating a regression coefficient indicating the degree to which the variable factor information contributes to the fluctuation in the number of purchases based on the actual data;
Using the regression coefficient calculated by the regression coefficient calculating means and the variable factor information on the reference date, the predicted purchase number calculating means for calculating the number of purchases for each product purchased on the reference date as the predicted purchase number When,
With
The regression coefficient calculation means is a divided period obtained by dividing a period related to the second actual data by a predetermined number for a product in which the difference in the number of purchases between the first actual data and the second actual data is a predetermined value or more. Calculating the regression coefficient based on the average value of the number of purchases in
A demand forecasting device characterized by that. The regression coefficient calculating means can arbitrarily set a period related to the second performance data.
The demand prediction apparatus according to claim 1, wherein: The regression coefficient calculating means sets the period related to the second performance data to 70 days,
The demand forecasting device according to claim 2 characterized by things. The regression coefficient calculating means can arbitrarily set the number of divisions of the period related to the second performance data.
The demand prediction apparatus according to any one of claims 1 to 3, wherein The regression coefficient calculating means sets the number of divisions of the period related to the second performance data as two divisions,
The demand forecasting device according to claim 4 characterized by things. Computer
Storage means for storing past purchase number of each product and variation factor information obtained by quantifying a variation factor that fluctuates the purchase number as actual data;
An accepting means for accepting an input of a reference date to be a target of demand forecast;
Of the performance data stored in the storage means, first performance data for a predetermined period closest to the reference date, and second for a predetermined period from the date before the previous year corresponding to the reference date to the future direction. A regression coefficient calculating means for calculating a regression coefficient indicating the degree to which the variable factor information contributes to the fluctuation in the number of purchases based on the actual data;
Using the regression coefficient calculated by the regression coefficient calculating means and the variable factor information on the reference date, the predicted purchase number calculating means for calculating the number of purchases for each product purchased on the reference date as the predicted purchase number When,
Function as
The regression coefficient calculation means is a divided period obtained by dividing a period related to the second actual data by a predetermined number for a product in which the difference in the number of purchases between the first actual data and the second actual data is a predetermined value or more. Calculating the regression coefficient based on the average value of the number of purchases in
A program characterized by that.

Images