JP6159056B2 - Selection program, selection method and selection device - Google Patents

Description

  The present invention relates to a selection program, a selection method, and a selection device.

  Conventionally, in order to perform product production management and inventory management, demand prediction for predicting demand for products has been performed. In the demand prediction, various prediction models that define calculation methods for predicting the demand for goods are used. This prediction model defines a prediction function for calculating a future predicted value from a past actual value. When performing the demand prediction, a prediction function is selected according to the characteristics of the product that is the target of the demand prediction, seasonal variation, and the like, and the predicted value is calculated from the actual value of the product using the selected prediction function.

  As a technique related to demand prediction, there is a technique for selecting a prediction function after verifying an error between a past actual value and a predicted value. This technology uses data for the latest three months among past actual values of a certain product as data for verification. That is, this technique calculates predicted values for the latest three months for each prediction function from data obtained by excluding data for the latest three months from the actual values. Then, this technique verifies an error between the actual value and the predicted value for the latest three months, and selects a prediction function that calculates a predicted value with a small error from the actual value.

JP 2007-293624 A

  By the way, in said demand prediction, improving the prediction precision is calculated | required. However, the inventor has found that, when selecting a prediction function, the prediction accuracy as expected may not be obtained only by selecting a prediction function with a small error from the actual value.

  For example, the period for which the person who sells the product attaches importance differs depending on whether the product for which the demand is to be forecasted is a best seller or a dead one. In other words, the best-selling product is a product that one wants to supply without losing sales opportunities. For this reason, it is expected that the most recent prediction accuracy is high in the demand prediction of the best-selling products. However, in the above-described conventional technology, the expected latest prediction accuracy may not be obtained.

  On the other hand, merchandise that is dying is a product that you want to adjust in a direction that reduces the number of stocks. In order to adjust the number of stocks, it is necessary to have a long-term outlook for demand. Therefore, it is expected that the long-term forecast accuracy is high in the demand forecast for daring products. However, in the above-described conventional technology, there are cases where the expected long-term prediction accuracy cannot be obtained.

  The disclosed technology has been made in view of the above, and an object thereof is to provide a selection program, a selection method, and a selection device that can improve prediction accuracy.

  In one aspect, the technology disclosed in the present application includes a reception unit, a classification unit, a calculation unit, and a selection unit. A reception part receives performance data about a plurality of goods. The classification unit classifies the plurality of products into a plurality of groups according to the sales volume of each product. The calculation unit uses a plurality of prediction functions for predicting future sales volume from past sales volume, and calculates a plurality of predicted values from the actual data. A selection part selects a prediction function according to a group among several prediction functions based on the calculated some predicted value and performance data.

  According to one aspect of the technology disclosed in the present application, there is an effect that the prediction accuracy can be improved.

FIG. 1 is a block diagram illustrating a functional configuration of the selection device according to the first embodiment. FIG. 2 is a diagram illustrating an example of information stored as performance data. FIG. 3A is a diagram for describing a prediction model determined by a determination unit. FIG. 3B is a diagram for describing the prediction model determined by the determination unit. FIG. 3C is a diagram for describing the prediction model determined by the determination unit. FIG. 4 is a diagram for explaining processing of classifying products into groups by the classification unit. FIG. 5 is a diagram for explaining processing of classifying products into groups by the classification unit. FIG. 6 is a diagram for explaining processing of classifying products into groups by the classification unit. FIG. 7 is a diagram for explaining processing in which the classification unit classifies the products into groups. FIG. 8 is a diagram for explaining processing in which the calculation unit calculates a predicted value. FIG. 9 is a diagram for explaining processing in which the calculation unit calculates a predicted value. FIG. 10 is a flowchart illustrating the processing procedure of the selection device according to the first embodiment. FIG. 11 is a diagram for explaining an application example to the data center. FIG. 12 is a diagram illustrating an example of a computer that executes a selection program.

  Embodiments of a selection program, a selection method, and a selection device disclosed in the present application will be described below in detail with reference to the drawings. Note that the present invention is not limited to the embodiments. Each embodiment can be appropriately combined within a range in which processing contents do not contradict each other.

  An example of a functional configuration of the selection device according to the first embodiment will be described. FIG. 1 is a block diagram illustrating a functional configuration of the selection device according to the first embodiment. As illustrated in FIG. 1, the selection device 100 includes a storage unit 110 and a control unit 120. The selection device 100 is connected to the input device 101 and the output device 102. The selection device 100 corresponds to an information processing device such as a personal computer or a workstation.

  The input device 101 is a device that receives input of various types of information, and corresponds to, for example, a keyboard, a mouse, a medium reading device, or a display with a touch panel function. For example, the input device 101 receives an input of performance data 111 indicating the sales volume for each period from a person who uses the selection device 100. The input device 101 outputs various received information input to the receiving unit 121 described later.

  The output device 102 is a device that outputs various types of information, and corresponds to, for example, a display or a monitor. For example, the output device 102 receives a prediction function for each product from the selection unit 125 described later. The output device 102 outputs the received various information.

  The storage unit 110 stores performance data 111. The storage unit 110 corresponds to, for example, a semiconductor memory device such as a RAM (Random Access Memory) or a flash memory, and a storage device such as a hard disk device or an optical disk device.

  The performance data 111 indicates, for example, the sales volume for each period. As an example, the performance data 111 is stored in the storage unit 110 by the receiving unit 121 described later. As another example, the record data 111 is referred to by a determination unit 122, a classification unit 123, a calculation unit 124, and a selection unit 125, which will be described later.

  As one aspect, the performance data 111 indicates time-series data in which a date, a sales quantity, and a profit amount are associated with each other for each of a plurality of products that are targets of demand prediction. Among these, “date” indicates a past date when the product was sold. “Sales volume” refers to the quantity of products sold during one day. “Profit amount” indicates the amount of profit obtained by selling a product during one day. In addition, although the case where sales volume was totaled for every day was illustrated here, the technique of an indication is not limited to this. For example, the sales volume may be totaled every week or may be totaled every year. The unit in which the sales volume is totaled is also referred to as “bucket”. Below, the case where a bucket is one day is demonstrated.

  FIG. 2 is a diagram illustrating an example of information stored as performance data. The performance data 111 shown in FIG. 2 shows, for example, the sales volume of the product A in units of 100,000, and the profit amount obtained by the sales volume in units of 1 million yen. In the first record in FIG. 2, the date “20120220”, the sales quantity “10”, and the profit amount “1” are associated with each other. This indicates that there are 1 million products A sold on February 20, 2012, and the profit obtained thereby is 1 million yen. Further, the date “201202019”, the sales quantity “14”, and the profit amount “2” are associated with the second record in FIG. This indicates that there are 1.4 million products A sold on February 19, 2012, and the profit obtained thereby is 2 million yen. Similarly, the sales data and the profit amount are associated with the past data 111 for other past dates. In FIG. 2, the actual data 111 of the product A is illustrated as an example, but the date, the sales volume, and the profit amount correspond to the actual data 111 of other products that are targets of demand prediction. Attached.

  The control unit 120 includes a reception unit 121, a determination unit 122, a classification unit 123, a calculation unit 124, and a selection unit 125. The function of the control unit 120 can be realized by an integrated circuit such as an application specific integrated circuit (ASIC) or a field programmable gate array (FPGA). Further, the function of the control unit 120 can be realized, for example, by a CPU (Central Processing Unit) executing a predetermined program.

  For example, the reception unit 121 receives the record data 111 indicating the sales quantity for each period for a plurality of products. As an example, the accepting unit 121 accepts performance data 111 in which a date, a sales quantity, and a profit amount are associated with each other for each of a plurality of products. Then, the reception unit 121 stores the received performance data 111 in the storage unit 110.

  For example, the reception unit 121 performs data density analysis on the received performance data 111. Here, the data density analysis is to determine whether or not the received time-series data has a sufficient data density as a processing target for demand prediction. For example, the reception unit 121 detects, for a predetermined period included in the time series data, an intermittent portion indicating a record that does not include sales volume data and an effective portion that indicates a record that does not include sales volume data. And the reception part 121 calculates the ratio of the effective part contained in a predetermined period. Then, the reception unit 121 determines that the time-series data has a sufficient data density as a processing target of the demand prediction when the calculated ratio is equal to or greater than the threshold value, and if the calculated ratio is less than the threshold value, the time-series data includes the demand data. It is determined that there is not enough data density as a prediction processing target. Note that the time series data determined not to have a sufficient data density here is deleted by the receiving unit 121, for example.

  For example, the reception unit 121 performs a variation analysis on the received performance data 111. Here, the variation analysis is to determine whether or not the received time-series data is variation enough to maintain the reliability of demand prediction. For example, the reception unit 121 calculates the standard deviation of the sales quantity for a predetermined period included in the time series data. If the calculated standard deviation is less than the threshold, the reception unit 121 determines that the time-series data has enough variation to maintain the reliability of the demand prediction. It is determined that the variation is such that the reliability of demand prediction cannot be maintained. Note that the time-series data determined to have such a variation that the reliability cannot be maintained is deleted by the receiving unit 121, for example.

  Note that the reception unit 121 does not necessarily perform the data density analysis and the variation analysis. For example, when receiving time-series data for which data density analysis and variation analysis have been performed, the reception unit 121 does not need to perform data density analysis and variation analysis.

  For example, the determination unit 122 determines a prediction model for predicting a future sales quantity from a past sales quantity for each product. Examples of the prediction model include an exponential smoothing model, an expansion rate model, a seasonal adjustment model, and the like. Among them, the exponential smoothing model is a model that is applied when it is predicted from past data that the sales volume of a product will stably increase. The growth rate model is a model applied when it is predicted from past data that a change in the sales volume of a product shows a series of life cycles of a growth period, a stable period, and a decline period. The seasonal adjustment model is a model that is applied when it is predicted from past data that the sales volume of a product shows periodicity. Each prediction model includes a prediction function for calculating a future sales volume from a past sales volume.

  Here, the case where a prediction model is determined by the determination part 122 is demonstrated using FIG. 3A to FIG. 3C. 3A to 3C are diagrams for explaining the prediction model determined by the determination unit 122. FIG. 3A to FIG. 3C illustrate changes over time in the sales volume of a certain product. The vertical axis in FIG. 3 indicates the sales volume, and the horizontal axis indicates time. The product illustrated in FIG. 3A has a characteristic that there is little variation in the change in the sales volume with time, and the sales volume increases stably. For example, the determination unit 122 determines to apply the exponential smoothing model to a product having the characteristics shown in FIG. 3A. The product illustrated in FIG. 3B has a characteristic that after the sales volume temporarily increases, the increase slows down and gradually decreases. For example, the determination unit 122 determines to apply the expansion rate model to a product having characteristics as illustrated in FIG. 3B. In addition, the product illustrated in FIG. 3C has a characteristic that the sales quantity shows periodicity. For example, the determination unit 122 determines to apply the seasonal adjustment model to a product having characteristics as illustrated in FIG. 3C.

  For example, the determination unit 122 determines whether to apply the seasonal adjustment model for each performance data 111 by determining whether each performance data 111 stored in the storage unit 110 has periodicity. As an example, the determination unit 122 superimposes the performance data 111 for each product at a predetermined cycle, and determines that there is a periodicity when there is a correlation with the temporal change of the overlapped sales quantity. It is determined that the seasonal adjustment model is applied to the performance data 111. Here, the period used for determining the presence or absence of periodicity is also referred to as a “cycle parameter”. As the cycle parameter, for example, when the bucket is one day, “7” is designated in order to evaluate the characteristics of the day of the week. Also, for example, when the bucket is one week, “52” is specified to evaluate the seasonal characteristics, and when the bucket is one year, “12” is also used to evaluate the seasonal characteristics. Is specified. The cycle parameter specified here is not limited to this, and an arbitrary value may be specified.

  Further, for example, the determination unit 122 applies the growth rate model by determining whether each performance data 111 stored in the storage unit 110 has a series of life cycles of a growth period, a stable period, and a decline period. Whether or not to do so is determined for each performance data 111. As an example, the determination unit 122 determines that the person has a life cycle when there is a correlation between the actual data 111 for each product and the actual data 111 of the product known to have a life cycle. It is determined that the expansion rate model is applied to the performance data 111. For example, the determination unit 122 determines to apply the seasonal adjustment model to the performance data 111 to which the seasonal adjustment model and the expansion rate model are not applied.

  Note that the prediction model determined by the determination unit 122 is not limited to the above example. Although the case where the determination unit 122 determines a prediction model for each performance data 111 has been described here, the disclosed technique is not limited to this. For example, the determination unit 122 may not determine the prediction model for the performance data 111 in which the prediction model is designated in advance.

  For example, the classification unit 123 classifies a plurality of products into a plurality of groups according to the sales volume of each product. As an example, the classification unit 123 classifies a plurality of products that are targets of demand prediction into the best selling and the most dead according to the sales volume of each product. Of these, “Selling” refers to products that sell well compared to other products among multiple products, and “Sustained” refers to products that sell less than other products among multiple products. Refers to a product. Note that the best selling is also referred to as “net”. The dead muscle is also called “gross”.

  As one aspect, the classifying unit 123 performs ABC analysis to classify a plurality of products into the best selling and the most dead. This “ABC analysis” is an analysis method for comparing the performance data 111 of a plurality of products and assigning importance to each product, and classifying them into three ranks A, B and C in order from the most important one. It is a technique to do. For example, the classification unit 123 performs ABC analysis on a plurality of products using the sales volume and profit amount included in the performance data 111 of each product, and then uses each analysis result to identify each product as a best seller. Classify as dead muscle. The ABC analysis is also referred to as “important analysis”.

  A process in which the classification unit 123 classifies the products into groups will be described with reference to FIGS. 4 to 7. FIG. 4 to FIG. 7 are diagrams for explaining a process in which the classification unit 123 classifies products into groups. Here, a case will be described in which the classification unit 123 classifies the nine types of products from the product A to the product I into the best selling and the dead ones. In addition, it is assumed that the record data 111 of the products A to I is stored in the storage unit 110.

  A process in which the classification unit 123 classifies the products A to I into ranks A to C will be described with reference to FIG. In the example shown in FIG. 4, the sales, the composition ratio, the cumulative total, and the sales rank are shown for each product. Among these, “sales” refers to the sum of the sales volume of each product in a predetermined period. For example, the sales correspond to the sum of the sales quantities for the latest three days in the performance data 111 of FIG. For example, sales are shown in units of 100,000. “Composition ratio” indicates the ratio of sales of each product to the total sales of all products. Further, “cumulative” refers to a cumulative composition ratio obtained by adding the composition ratios of the products in order from the products with the largest composition ratio. “Sales rank” indicates the importance assigned to each product according to the cumulative value. For example, the sales rank “A” indicating the most important of all the products analyzed by the ABC analysis is assigned to the products corresponding to the total of 0% to 70%. Further, for example, a sales rank “B” indicating that the product corresponding to the cumulative total of 70% to 90% is the next most important after the sales rank “A” is assigned. Further, for example, a sales rank “C” indicating that the cumulative total corresponds to 90% to 100% is not important. The predetermined period for calculating the sales is a period for analyzing the importance of the product, and may be arbitrarily set by a person using the selection device 100. In addition, the cumulative value for assigning the sales rank may be arbitrarily set by a person using the selection device 100.

  As illustrated in FIG. 4, for example, the classification unit 123 acquires each result data 111 of the products A to I from the storage unit 110, and calculates the sum of the sales volume for the most recent three days from the result data 111 of each product. Calculated as sales. For example, as illustrated in FIG. 4, the classification unit 123 calculates 10 million pieces of sales of the product A and 20 million pieces of sales of the product B. Subsequently, the classification unit 123 adds the sales of each product to obtain the sum of sales of all the products, and calculates the ratio of the sales of each product to the obtained sum as a component ratio. As an example, the classification unit 123 calculates 3% as the component ratio of the product A and calculates 6% as the component ratio of the product B as illustrated in FIG. Thereafter, the classification unit 123 calculates the total from the product having the largest component ratio to each product by adding the component ratios of each product in order from the product having the largest component ratio. As an example, the classification unit 123 sets the composition ratio “18%” of the product G having the largest composition ratio among the products A to I as the cumulative total of the products G, as illustrated in FIG. Subsequently, the classification unit 123 adds the composition ratio “17%” of the product H having the second largest composition ratio to the product G to the accumulated “18%” of G, and obtains the accumulated “35%” up to the product H. calculate. Similarly, the classification unit 123 calculates the total from each product A to product I by adding the composition ratio of each product in order from the product with the largest composition ratio. Then, the classification unit 123 assigns a sales rank to each product according to the cumulative value. As an example, the classification unit 123 assigns the sales rank “A” to four types of products G, H, I, and F corresponding to a total of 0% to 70%. The classification unit 123 assigns the sales rank “B” to the two types of products E and D corresponding to the cumulative total of 70% to 90%. The classification unit 123 assigns the sales rank “C” to the three types of products C, B, and A corresponding to a total of 90% to 100%. As described above, the classification unit 123 classifies the products A to I into the sales ranks A to C by using the record data 111 of the products A to I. Here, the case where the classification unit 123 classifies the products into the three ranks A to C has been described, but the disclosed technique is not limited to this. For example, the classification unit 123 may classify into four ranks A to D, or may classify into two ranks A or B.

  The process in which the classification unit 123 classifies the products A to I into the ranks A to C will be described with reference to FIG. In the example shown in FIG. 5, the profit, the composition ratio, the cumulative total, and the profit rank are shown for each product. Among these, “profit” refers to the sum of profit amounts of each product in a predetermined period. For example, the profit corresponds to the sum of the profit amounts for the most recent three days in the performance data 111 of FIG. Further, for example, the profit is shown in units of 1 million yen. “Composition ratio” refers to the ratio of the profit of each product to the total profit of all products. Further, “cumulative” refers to a cumulative composition ratio obtained by adding the composition ratios of the products in order from the products with the largest composition ratio. The “profit rank” indicates the importance assigned to each product according to the cumulative value. For example, a profit rank “A” indicating that it is most important among all the products analyzed by ABC analysis is assigned to the products corresponding to the cumulative total of 0% to 70%. Further, for example, a profit rank “B” indicating that it is the next most important product after the sales rank “A” is assigned to products corresponding to a cumulative total of 70% to 90%. Further, for example, a profit rank “C” indicating that it is not important is assigned to a product corresponding to a cumulative total of 90% to 100%. The predetermined period for calculating the profit is a period for analyzing the importance of the product, and may be arbitrarily set by a person using the selection device 100. Further, the cumulative value for assigning the profit rank may be arbitrarily set by a person using the selection device 100.

  As shown in FIG. 5, for example, the classification unit 123 acquires each result data 111 of the products A to I from the storage unit 110, and calculates the sum of the profit amounts for the most recent three days from the result data 111 of each product. Calculated as profit. As an example, the classification unit 123 calculates 20 million yen as the profit of the product A, and calculates 130 million yen as the profit of the product B, as shown in FIG. Subsequently, the classification unit 123 adds the profits of the respective products to obtain the sum of the profits of all the products, and calculates the ratio of the profits of the respective products in the obtained total as a component ratio. As an example, the classification unit 123 calculates 4% as the component ratio of the product A and 26% as the component ratio of the product B as illustrated in FIG. Thereafter, the classification unit 123 calculates the total from the product having the largest component ratio to each product by adding the component ratios of each product in order from the product having the largest component ratio. As an example, as shown in FIG. 5, the classification unit 123 sets the composition ratio “26%” of the product B having the largest composition ratio among the products A to I as the cumulative total of the products B. Subsequently, the classification unit 123 adds the composition ratio “25%” of the product G having the second largest composition ratio to the product B to the cumulative “26%” of B, and obtains the cumulative “50%” up to the product G. calculate. Similarly, the classification unit 123 calculates the total from each product A to product I by adding the composition ratio of each product in order from the product with the largest composition ratio. Then, the classification unit 123 assigns a profit rank to each product according to the cumulative value. As an example, the classification unit 123 assigns a profit rank “A” to three types of products B, G, and F corresponding to a total of 0% to 70%. The classification unit 123 assigns the profit rank “B” to one type of product E corresponding to the cumulative total of 70% to 90%. The classification unit 123 assigns the profit rank “C” to the five types of products H, A, C, I, and D corresponding to the cumulative total of 90% to 100%. As described above, the classification unit 123 classifies the products A to I into the profit ranks A to C by using the record data 111 of the products A to I. Here, the case where the classification unit 123 classifies the products into the three ranks A to C has been described, but the disclosed technique is not limited to this. For example, the classification unit 123 may classify into four ranks A to D, or may classify into two ranks A or B.

  A process in which the classification unit 123 classifies the products A to I into the best-selling and dead lines will be described with reference to FIGS. 6 and 7. FIG. 6 exemplifies the result of tabulating the sales rank of FIG. 4 and the profit rank of FIG. As an example, the first record in FIG. 6 indicates that the sales rank of the product A is “C” and the profit rank is also “C”. FIG. 7 illustrates an example in which the aggregation result of FIG. 6 is applied to a matrix having nine lattices. In the example shown in FIG. 7, the horizontal direction is a sales rank of A to C, and the vertical direction is a profit rank of A to C. As an example, the upper right grid in FIG. 7 includes products F and G whose sales rank and profit rank are both “A”. In this case, for example, the classifying unit 123 classifies the product included in the area 7a of FIG. 7 as the best selling and classifies the product included in the area 7b as “dead streak”. That is, in the example of FIG. 7, the classification unit 123 classifies the three types of products E, F, and G as the best selling and classifies the six types of products A, B, C, D, H, and I as the dead.

  In this way, the classification unit 123 performs ABC analysis and classifies the products A to I into the best selling and the dead one. Then, the classification unit 123 outputs the classified result to the calculation unit 124 and the selection unit 125 described later. Here, the case where the classification unit 123 performs the ABC analysis to classify into the best selling and the dead one has been described, but the disclosed technique is not limited to this. For example, the classification unit 123 may compare the sales volume of each product, and may set the product with the highest sales volume as the best selling and the remaining products as the dead line.

  For example, the calculation unit 124 calculates a plurality of predicted values from the performance data 111 by using a plurality of prediction functions for predicting a future sales quantity from a past sales quantity. This “predicted value” is calculated for a past verification period in order to verify an error from the actual data 111. In other words, for example, the calculation unit 124 calculates the predicted value of the verification period from the actual data 111 of the period excluding the verification period among the actual data 111 of each product.

  As an example, the calculation unit 124 sets the verification period according to the group classified by the classification unit 123. For example, the calculation unit 124 sets a shorter verification period for a group including products with a large sales volume, and sets a longer verification period for a group including products with a small sales volume. Subsequently, the calculation unit 124 sets a plurality of patterns of parameters in the prediction function corresponding to the prediction model determined by the determination unit 122, thereby generating a plurality of patterns of prediction functions. And the calculation part 124 calculates the predicted value of a several pattern for every goods from the performance data 111 of the period except the set verification period using the produced | generated prediction function of the several pattern.

  The process in which the calculation part 124 calculates a predicted value is demonstrated using FIG.8 and FIG.9. FIG. 8 and FIG. 9 are diagrams for explaining processing in which the calculation unit 124 calculates a predicted value. 8 and 9, the vertical axis indicates the sales volume, and the horizontal axis indicates time. A curve 8a in FIG. 8 shows a change over time in the sales quantity of the product F, and a curve 9a in FIG. 9 shows a change over time in the sales quantity of the product A. It is assumed that the product F is classified as a selling source and the product A is classified as a dead line. In addition, an exponential smoothing model is applied to the product A and the product F as a prediction model. The prediction function used in this exponential smoothing model includes a parameter called a smoothing exponent. For example, an arbitrary value from “0.1” to “0.9” is set as the smoothing index. Hereinafter, for convenience of explanation, a prediction function used in the exponential smoothing model is represented as “f (x)”.

  As illustrated in FIGS. 8 and 9, the calculation unit 124 sets a verification period 8 b for the product F that is a hot selling item, and sets a verification period 9 b for the product A that is a dead source. Here, the verification period 8b is characterized by being shorter than the verification period 9b. For example, the verification period 8b is set to 3 days retroactive from the present, and the verification period 9b is set to 7 days retroactive from the present. . Subsequently, the calculation unit 124 sets a plurality of patterns of parameters in the prediction function f (x), thereby generating a plurality of patterns of prediction functions. For example, the calculation unit 124 has the prediction function f1 (x) with the smoothing index set to “0.3”, the prediction function f2 (x) with the smoothing index set to “0.5”, and the smoothing index “0”. .7 "is generated, respectively. In the example illustrated in FIG. 8, the calculation unit 124 uses the prediction functions f1 (x), f2 (x), and f3 (x) to predict the predicted value 8d from the actual data 111 in the period 8c excluding the verification period 8b. , 8e and 8f, respectively. Among these, the prediction value 8d corresponds to the prediction value calculated by the prediction function f1 (x), the prediction value 8e corresponds to the prediction value calculated by the prediction function f2 (x), and the prediction value 8f is the prediction function f3. This corresponds to the predicted value calculated by (x). In the example illustrated in FIG. 9, the calculation unit 124 uses the prediction functions f1 (x), f2 (x), and f3 (x) to predict the predicted value 9d from the actual data 111 in the period 9c excluding the verification period 9b. , 9e and 9f, respectively. Among these, the prediction value 9d corresponds to the prediction value calculated by the prediction function f1 (x), the prediction value 9e corresponds to the prediction value calculated by the prediction function f2 (x), and the prediction value 9f is the prediction function f3. This corresponds to the predicted value calculated by (x).

  As described above, the calculation unit 124 calculates the predicted value of the plurality of patterns for each product from the performance data 111 in the period excluding the verification period, using the prediction function of the plurality of patterns. Here, the case where the calculation unit 124 calculates three patterns of prediction values for the three patterns of prediction functions has been described, but the disclosed technique is not limited to this. For example, the calculation unit 124 may calculate a predicted value of an arbitrary number of patterns by a person using the selection device 100 setting parameters of an arbitrary number of patterns in the prediction function.

  The selection unit 125 selects a prediction function from a plurality of prediction functions based on the plurality of prediction values and the record data 111 according to whether the selling item is the best seller or the dead one. For example, the selection unit 125 selects a prediction function having a smallest error between the predicted value and the actual data in a shorter verification period for a group including a product with a larger sales volume among a plurality of prediction functions. In addition, the selection unit 125 selects a prediction function having the smallest error between the predicted value and the actual data in the longer verification period for a group including a product with a smaller sales volume. At this time, the selection unit 125 verifies errors between the predicted values of the plurality of patterns and the actual data 111 by a calculation method according to the classified group.

  In the example illustrated in FIG. 8, the selection unit 125 verifies an error between the sales volume curve 8a of the product F and the predicted values 8d, 8e, and 8f for the verification period 8b. At this time, the selection unit 125 verifies the error with respect to the commodity F which is the best seller by using the mean absolute deviation (MAD). This MAD is an average value of absolute values of errors between the actual data 111 and the predicted values. The selection unit 125 calculates MAD from the error between the sales quantity curve 8a of the product F and the predicted values 8d, 8e, and 8f. Then, the selection unit 125 selects a prediction function corresponding to the one with the smallest calculated MAD. In the example shown in FIG. 8, since the predicted value with the smallest MAD is the predicted value 8f, the selection unit 125 selects the prediction function f3 (x) corresponding to the predicted value 8f.

  In the example illustrated in FIG. 9, the selection unit 125 verifies the error between the sales volume curve 9a of the product A and the predicted values 9d, 9e, and 9f for the verification period 9b. At this time, the selection unit 125 verifies an error for a product A that is a dead line using a tracking signal (TS). This TS is obtained by dividing the total error between the actual data 111 and the predicted value by MAD. The selection unit 125 calculates TS by dividing the total error between the sales quantity curve 9a of the product A and the predicted values 9d, 9e, and 9f by MAD. Then, the selection unit 125 selects a prediction function corresponding to the one with the smallest calculated TS. In the example shown in FIG. 9, since the predicted value with the smallest TS is the predicted value 9d, the selection unit 125 selects the prediction function f1 (x) corresponding to the predicted value 9d.

  As described above, the selection unit 125 selects a prediction function for each product from a plurality of prediction functions based on the plurality of prediction values and the record data 111 according to whether the item is a selling item or a death item. Then, the selection unit 125 outputs the prediction function selected for each product to the output device 102. Note that the selection unit 125 may specify parameters set in the selected prediction function and output the specified parameters.

  Next, a processing procedure of the selection device 100 according to the first embodiment will be described. FIG. 10 is a flowchart illustrating the processing procedure of the selection device 100 according to the first embodiment. The process illustrated in FIG. 10 is executed, for example, when the reception unit 121 receives the record data 111.

  As illustrated in FIG. 10, for example, when the reception unit 121 receives the performance data 111 of a plurality of products (Yes in step S101), the reception unit 121 stores the performance data 111 of each product in the storage unit 110 (step S102). ).

  Subsequently, the determination unit 122 determines a prediction model (step S103). For example, the determination unit 122 determines which prediction model is to be applied among the exponential smoothing model, the expansion rate model, and the seasonal adjustment model for each of the product performance data 111 stored in the storage unit 110.

  Subsequently, the classifying unit 123 classifies the plurality of products targeted for demand prediction into the best selling and the most dead (step S104). For example, the classification unit 123 uses the sales volume of each product to classify each product into sales ranks A to C. The classification unit 123 classifies each product into the profit ranks A to C using the profit amount of each product. And the classification | category part 123 classify | categorizes each goods into a best selling and a dead line based on a sales rank and a profit rank.

  Subsequently, the calculation unit 124 sets a verification period according to the group classified by the classification unit 123 (step S105). For example, the calculation unit 124 sets a shorter verification period for a group including products with a large sales volume, and sets a longer verification period for a group including products with a small sales volume. Subsequently, the calculation unit 124 sets a plurality of patterns of parameters to the prediction function corresponding to the prediction model determined by the determination unit 122, thereby generating a plurality of patterns of prediction functions (step S106). And the calculation part 124 calculates the predicted value of a several pattern for every goods from the performance data 111 of the period except the set verification period using the produced | generated prediction function of the several pattern (step S107).

  Subsequently, the selection unit 125 selects a prediction function from a plurality of prediction functions based on the plurality of prediction values and the actual data 111 according to whether the selling item is the best selling or the dead one (Step S108). At this time, for example, the selection unit 125 verifies an error using a MAD for a hot-selling product, and selects a prediction function with the smallest error. Further, for example, the selection unit 125 verifies an error using a TS for a deadly product, and selects a prediction function having the smallest error.

  And the selection part 125 outputs the prediction function selected for every goods to the output device 102 (step S109). Note that the selection unit 125 may specify parameters set in the selected prediction function and output the specified parameters.

  Next, effects of the selection device 100 according to the first embodiment will be described. For example, the selection device 100 according to the first embodiment receives the record data 111 for a plurality of products. The selection apparatus 100 classifies a plurality of products into a plurality of groups according to the sales volume of each product. The selection device 100 calculates a plurality of predicted values from the performance data 111 by using a plurality of prediction functions for predicting a future sales volume from the past sales volume. The selection apparatus 100 selects a prediction function according to a group among several prediction functions based on the calculated some predicted value and the performance data 111. FIG. For this reason, the selection apparatus 100 can improve prediction accuracy.

  In addition, for example, the selection apparatus 100 selects a prediction function having a smallest error between the predicted value and the actual data in a short period of time among the plurality of prediction functions, in a group including products with a large sales volume. On the other hand, the selection apparatus 100 selects a prediction function having the smallest error between the predicted value and the actual data in the long term for a group including a product with a smaller sales quantity among a plurality of prediction functions. That is, for example, the selection device 100 selects a prediction function that fits the short-term predicted value to the performance data 111 for the best-selling product. On the other hand, the selection device 100 selects a prediction function that fits the long-term predicted value to the performance data 111 as the product becomes more deadly. Thus, since the selection apparatus 100 selects a prediction function using an appropriate period according to whether it is a best seller or a dead person, it is possible to select an appropriate parameter and improve prediction accuracy.

  In addition, for example, the selection device 100 classifies the plurality of products into the best selling and the most dead according to the sales volume and the profit amount of each product. Then, the selection device 100 selects a prediction function in which the short-term predicted value fits the actual result data 111 for the best-selling product. On the other hand, the selection device 100 selects a prediction function that fits the long-term predicted value to the performance data 111 as the product becomes more deadly. As described above, since the selection device 100 classifies the profitable amount as the best-selling or dead line, for example, based on the management judgment that the sales opportunity may be missed for a product with a large sales amount but a small profit amount. Can be classified.

  Although the embodiments of the present invention have been described so far, the present invention may be implemented in other embodiments besides the above-described embodiments. Therefore, other embodiments will be described below.

  For example, in the above-described embodiment, the case where the performance data 111 includes the profit amount has been described, but it is not always necessary that the profit amount is included. That is, it is only necessary that the date and the sales quantity are associated with the result data 111.

  Further, for example, in the above-described embodiment, the case where the classification unit 123 classifies the merchandise according to the sales volume and the profit amount of each product has been described as the best selling or the dead one, but the disclosed technique is not limited thereto. Absent. For example, the classification unit 123 may classify the item into the best selling or the most dead according to the sales volume of each product without using the profit amount.

  Further, for example, in the above-described embodiment, the case where the classification unit 123 classifies into two groups of selling and dead muscles has been described. However, the disclosed technique is not limited to this. For example, an intermediate line may be provided between the best-selling and dead lines, and it may be classified into three groups. In this case, for example, the calculation unit 124 sets the verification period of the hot selling product to 3 days, sets the verification period of the intermediate selling product to 5 days, and sets the verification period of the deadly selling product to 7 days. . According to this, since the selection apparatus 100 sets an appropriate period according to the group classified not only in detail, but also according to the group classified in detail and selects a prediction function, it can improve prediction accuracy.

  Further, for example, among the processes described in the above embodiments, all or part of the processes described as being automatically performed can be manually performed, or are described as being performed manually. All or part of the processing can be automatically performed by a known method. In addition, the processing procedures, control procedures, specific names, and information including various data and parameters shown in the above-described document and drawings can be arbitrarily changed unless otherwise specified.

  Further, each component of the selection device 100 shown in FIG. 1 is functionally conceptual and does not necessarily need to be physically configured as illustrated. In other words, the specific form of distribution / integration of the selection device 100 is not limited to that shown in the figure, and all or part of the selection apparatus 100 is functionally or physically distributed in arbitrary units according to various loads or usage conditions. -Can be integrated and configured. For example, an external storage device may be connected to the selection device 100 via a network, and the information stored in the storage unit 110 may be stored in the external storage device.

  The selection device 100 can also be realized by installing each function of the selection device 100 in a known information processing device. Known information processing apparatuses correspond to apparatuses such as personal computers and workstations, for example.

[Select prediction model]
For example, in the above-described embodiment, the case where the selection apparatus 100 selects one prediction function from a plurality of patterns of prediction functions in which different parameters are set for a predetermined prediction model has been described. The technology is not limited to this. For example, the selection apparatus 100 can also select one prediction function from a plurality of patterns of prediction functions generated with different prediction models.

  Below, the case where the selection apparatus 100 selects one prediction function from the prediction function of the several pattern produced | generated by the different prediction model is demonstrated using FIG.8 and FIG.9. Here, the curve 8a in FIG. 8 shows the change over time in the sales volume of the product F, and the curve 9a in FIG. It is assumed that the product F is classified as a selling source and the product A is classified as a dead line. In addition, an exponential smoothing model, an expansion rate model, and a seasonal adjustment model are applied to the products A and F as prediction models. In the following, for convenience of explanation, the prediction function used in the exponential smoothing model is expressed as “f (x)”, the prediction function used in the expansion rate model is expressed as “g (x)”, and the seasonal adjustment model The prediction function used is represented as “h (x)”.

  As illustrated in FIGS. 8 and 9, the calculation unit 124 sets a verification period 8 b for the product F that is a hot selling item, and sets a verification period 9 b for the product A that is a dead source. Here, the verification period 8b is characterized by being shorter than the verification period 9b. For example, the verification period 8b is set to 3 days retroactive from the present, and the verification period 9b is set to 7 days retroactive from the present. . In the example illustrated in FIG. 8, the calculation unit 124 uses the prediction functions f (x), g (x), and h (x) to predict the predicted value 8d from the actual data 111 in the period 8c excluding the verification period 8b. , 8e and 8f, respectively. Among these, the prediction value 8d corresponds to the prediction value calculated by the prediction function f (x), the prediction value 8e corresponds to the prediction value calculated by the prediction function g (x), and the prediction value 8f is the prediction function h. This corresponds to the predicted value calculated by (x). In the example illustrated in FIG. 9, the calculation unit 124 uses the prediction functions f (x), g (x), and h (x) to predict the predicted value 9d from the actual data 111 in the period 9c excluding the verification period 9b. , 9e and 9f, respectively. Among these, the prediction value 9d corresponds to the prediction value calculated by the prediction function f (x), the prediction value 9e corresponds to the prediction value calculated by the prediction function g (x), and the prediction value 9f is the prediction function h. This corresponds to the predicted value calculated by (x).

  Subsequently, the selection unit 125 selects, based on the predicted values 8d, 8e, 8f, 9d, 9e, and 9f, and the performance data 111, a prediction function for each product from a plurality of prediction functions according to whether the item is selling or dead. select. In the example illustrated in FIG. 8, the selection unit 125 calculates MAD from the error between the sales volume curve 8a of the product F and the predicted values 8d, 8e, and 8f for the verification period 8b. Then, the selection unit 125 selects a prediction function corresponding to the one with the smallest calculated MAD. In the example shown in FIG. 8, since the predicted value with the smallest MAD is the predicted value 8f, the selection unit 125 selects the prediction function h (x) corresponding to the predicted value 8f. In the example illustrated in FIG. 9, the selection unit 125 calculates TS by dividing the total of the sales volume curve 9a of the product A and the predicted values 9d, 9e, and 9f by MAD for the verification period 9b. Then, the selection unit 125 selects a prediction function corresponding to the one with the smallest calculated TS. In the example shown in FIG. 9, since the predicted value with the smallest TS is the predicted value 9d, the selection unit 125 selects the prediction function f (x) corresponding to the predicted value 9d.

  As described above, the selection device 100 selects one prediction function from a plurality of patterns of prediction functions generated using different prediction models. For this reason, since the selection apparatus 100 selects a prediction function using an appropriate period according to whether it is a best-selling or dead line, it can select an appropriate prediction model and can improve prediction accuracy.

[Data Center]
For example, the selection apparatus 100 may be applied to a data center that provides a service on the cloud. FIG. 11 is a diagram for explaining an application example to the data center. As shown in FIG. 11, the data center 200 is connected to clients 220a, 220b... Via a network 210, for example. The network 210 may employ any type of communication network, such as the Internet or WAN (Wide Area Network), whether wired or wireless. The clients 220a, 220b,... Are distributed and installed in a plurality of institutions. The data center 200 includes server devices 201a, 201b,... And storage devices 202a, 202b,. Here, the clients 220a, 220b,... Are collectively referred to as the client 220 when they are described without distinction. In addition, when the server apparatuses 201a, 201b,. In addition, the storage devices 202a, 202b,. For example, the data center 200 uses the server device 201 to realize the function of the selection device 100 illustrated in FIG. 1 as a virtual machine.

  For example, in the data center 200, a selection VM (Virtual Machine) that is a virtual machine that assumes the function of the selection device 100 executes processing as follows. That is, the selection VM receives the record data 111 for a plurality of products, for example. The selection VM classifies the plurality of products into a plurality of groups according to the sales volume of each product. The selected VM uses a plurality of prediction functions for predicting future sales volume from past sales volume, and calculates a plurality of predicted values from the performance data 111. The selection VM selects a prediction function according to the group from among the plurality of prediction functions based on the plurality of calculated prediction values and the actual data 111. For this reason, the selection VM can improve the prediction accuracy.

[program]
FIG. 12 is a diagram illustrating an example of a computer that executes a selection program. As illustrated in FIG. 12, the computer 300 includes a CPU 301 that executes various arithmetic processes, an input device 302 that receives data input from a user, and a monitor 303. The computer 300 also includes a medium reading device 304 that reads a program or the like from a storage medium, an interface device 305 for connecting to another device, and a wireless communication device 306 for connecting to another device wirelessly. The computer 300 also includes a RAM (Random Access Memory) 307 that temporarily stores various types of information and a hard disk device 308. Each device 301 to 308 is connected to a bus 309.

  The hard disk device 308 stores a selection program having the same functions as the processing units of the reception unit 121, the classification unit 123, the calculation unit 124, and the selection unit 125 illustrated in FIG. The hard disk device 308 stores various data for realizing the selection program.

  The CPU 301 reads out each program stored in the hard disk device 308, develops it in the RAM 307, and executes it to perform various processes. In addition, these programs can cause the computer to function as the reception unit 121, the classification unit 123, the calculation unit 124, and the selection unit 125 illustrated in FIG.

  Note that the above selection program need not necessarily be stored in the hard disk device 308. For example, the computer 300 may read and execute a program stored in a computer-readable recording medium. The computer-readable recording medium corresponds to, for example, a portable recording medium such as a CD-ROM, a DVD disk, a USB (Universal Serial Bus) memory, a semiconductor memory such as a flash memory, a hard disk drive, and the like. Further, the program may be stored in a device connected to a public line, the Internet, a LAN (Local Area Network), a WAN (Wide Area Network), etc., and the computer 300 may read and execute the program therefrom. good.

  The following supplementary notes are further disclosed with respect to the embodiments including the above examples.

(Supplementary note 1)
For multiple products, we accept performance data showing the sales volume for each period,
Classifying the plurality of products into a plurality of groups according to the sales volume of each product;
Using multiple prediction functions to predict future sales volume from past sales volume, calculate multiple prediction values from the actual data,
A selection program that executes each process of selecting a prediction function according to the group from the plurality of prediction functions based on the calculated plurality of prediction values and the actual data.

(Additional remark 2) The said process to select selects the prediction function with the smallest difference | error between the predicted value and performance data in a short period for the group containing the goods with many sales quantities from these several prediction functions, and there are few sales quantities The selection program according to appendix 1, wherein a prediction function having a smallest error between a predicted value and actual data in a long period is selected for a group including a product.

(Additional remark 3) The said classification | category process classifies the said some goods into the some group according to the sales volume and profit amount of each goods,
The selecting process selects a prediction function having a smallest error between the predicted value and the actual data in a short period of time for a group including products having a larger sales volume and profit amount from the plurality of forecast functions, The selection program according to appendix 1, wherein a group including products having a smaller profit amount selects a prediction function having the smallest error between the predicted value and the actual data over a long period of time.

(Supplementary Note 4) The selection program according to Supplementary Note 2 or 3, wherein the selecting process calculates the error by a calculation method according to the classified group.

(Appendix 5) A selection method executed by a computer,
Computer
For multiple products, we accept performance data showing the sales volume for each period,
Classifying the plurality of products into a plurality of groups according to the sales volume of each product;
Using multiple prediction functions to predict future sales volume from past sales volume, calculate multiple prediction values from the actual data,
A selection method comprising: executing each process of selecting a prediction function according to the group from the plurality of prediction functions based on the plurality of calculated prediction values and the actual data.

(Additional remark 6) The process to select selects a prediction function with the smallest difference | error between the predicted value and performance data in a short period for the group containing a product with many sales quantities from these several prediction functions, and there are few sales quantities The selection method according to appendix 5, wherein a prediction function having the smallest error between the predicted value and the actual data in a long period is selected for a group including a product.

(Additional remark 7) The said process to classify | categorize the said some goods into the some group according to the sales quantity and profit amount of each goods,
The selecting process selects a prediction function having a smallest error between the predicted value and the actual data in a short period of time for a group including products having a larger sales volume and profit amount from the plurality of forecast functions, The selection method according to appendix 5, wherein a prediction function having a smallest error between the predicted value and the actual data in a long period is selected for a group including a product having a smaller profit amount.

(Supplementary note 8) The selection method according to supplementary note 6 or 7, wherein in the process of selecting, the error is calculated by a calculation method according to the classified group.

(Additional remark 9) About several goods, the reception part which receives the performance data which show the sales volume for every period,
A classification unit for classifying the plurality of products into a plurality of groups according to the sales volume of each product;
Using a plurality of prediction functions for predicting future sales volume from past sales volume, and calculating a plurality of predicted values from the actual data,
A selection device comprising: a selection unit that selects a prediction function according to the group from the plurality of prediction functions based on the plurality of calculated prediction values and the actual data.

(Additional remark 10) The said selection part selects the prediction function in which the difference | error of the predicted value and performance data in the short term is the smallest in the group containing a product with many sales quantities from these several prediction functions, and goods with few sales volumes The selection apparatus according to appendix 9, wherein a prediction function having a smallest error between the predicted value and the actual data in a long period is selected for a group including.

(Additional remark 11) The said classification | category part classifies the said some goods into the some group according to the sales volume and profit amount of each goods,
The selection unit selects, from the plurality of prediction functions, a prediction function in which a difference between the predicted value and the actual data in a short period is the smallest for a group including a product having a large sales volume and profit amount, and the sales volume and profit The selection apparatus according to appendix 9, wherein a group including products with a small amount of money selects a prediction function having a smallest error between the predicted value and the actual data over a long period of time.

(Additional remark 12) The said selection part calculates the said error with the calculation method according to the classified group, The selection apparatus of Additional remark 10 or 11 characterized by the above-mentioned.

DESCRIPTION OF SYMBOLS 100 Selection apparatus 101 Input apparatus 102 Output apparatus 110 Memory | storage part 111 Performance data 120 Control part 121 Reception part 122 Decision part 123 Classification part 124 Calculation part 125 Selection part 200 Data center 201 Server apparatus 202 Storage apparatus 210 Network 220 Client 300 Computer 301 CPU
302 Input Device 303 Monitor 304 Medium Reading Device 305 Interface Device 306 Wireless Communication Device 307 RAM
308 Hard disk device 309 Bus

Claims (5)

On the computer,
Read the sales performance data of a specific product from the storage unit that stores the sales performance data indicating the change in the sales volume of the product,
Applying a plurality of different types of parameters for the prediction model corresponding to the characteristics of the read sales performance data among the plurality of prediction models, generating a plurality of sales prediction data,
Based on the sales performance data, set a specific period shorter as the number of sales of the product is larger,
Wherein in said certain times of the actual sales data for a particular product, based on said a sales record data a plurality of forecast data, applies any of the parameters of the plurality of types of parameters to the prediction model parameters Select as the
A selection program characterized by causing each process to be executed. The selection of the parameters, the selection program according to claim 1, wherein a specified time period, the difference between the actual sales data and the plurality of forecast data to select a small parameter, it is characterized. The selection program according to claim 1, wherein the specific period is set shorter as the sales profit of the product is larger . A selection method performed by a computer,
Computer
Read the sales performance data of a specific product from the storage unit that stores the sales performance data indicating the change in the sales volume of the product,
Applying a plurality of different types of parameters for the prediction model corresponding to the characteristics of the read sales performance data among the plurality of prediction models, generating a plurality of sales prediction data,
Based on the sales performance data, set a specific period shorter as the number of sales of the product is larger,
Wherein in said certain times of the actual sales data for a particular product, based on said a sales record data a plurality of forecast data, applies any of the parameters of the plurality of types of parameters to the prediction model parameters Select as the
A selection method characterized by executing each process. From the storage unit storing the sales result data indicating the change in the sales volume of the product, the sales result data of the specific product is read, and the prediction model corresponding to the characteristics of the read sales result data among the plurality of prediction models, Applying different types of parameters to generate multiple sales forecast data,
The sales on the basis of the actual data, the set specific period shorter the larger sales number of products, in the particular time period of the actual sales data of the particular product, sales record data and the plurality of forecast data And a selection unit that selects any one of the plurality of types of parameters as a parameter to be applied to the prediction model, and
The selection apparatus characterized by having.

Images