JP4339769B2 - Prediction model selection device, prediction model selection method, and program - Google Patents

Description

  The present invention relates to a prediction model selection device, a prediction model selection method, and a program for selecting one prediction model from a plurality of prediction models.

  As one of the inventory management systems for merchandise, there is known an inventory management system of an ordering point system in which a certain amount is ordered when the inventory amount of the merchandise cuts the ordering point inventory. The “order point inventory” is a stock quantity obtained by adding a safety stock quantity to a demand forecast value within a procurement period of the product predicted based on a past demand actual value.

  In general, when appropriate demand forecasting is possible, the variation in the forecast error between the demand forecast value and the actual demand value is smaller than the variation in the actual demand value itself. It is known that the product rate can be realized (for example, see Patent Document 1). For example, as a method of calculating a safety stock amount for absorbing fluctuations in demand, a degree of variation in prediction error as shown in Expression (1) is used. In Expression (1), i indicates each past time point when the demand prediction or the like is performed.

  Here, the “safety factor” is determined by the allowable shortage rate. The larger the safety factor, the smaller the shortage probability. In addition, the “procurement period” is a period from when an order is placed until the product arrives. As a unit of the procurement period, a forecast unit period (demand forecast cycle) of demand forecast is used. For example, when the demand forecast is performed daily, the unit is a day, when the demand forecast is performed weekly, the week is used, and when it is performed monthly, the unit is the month.

  However, since the degree of variation in future prediction errors is unknown, in general, the degree of variation in past prediction errors depends on the assumption that the degree of variation in past prediction errors is the same as the degree of variation in future prediction errors. It is used. In addition, the smaller the degree of prediction error variation, the smaller the safety stock amount and the smaller the stock amount. Therefore, it is necessary to reduce the variation in prediction error as much as possible. Therefore, selecting a prediction model with the smallest possible variation in prediction error is a major issue for reducing inventory.

  On the other hand, as an index (evaluation value) at the time of evaluating the prediction accuracy, generally, a standard deviation of a past prediction error or a mean square error is often used. Therefore, it is appropriate to use the standard deviation of the past prediction error or the mean square error as an index when selecting one prediction model from a plurality of prediction models.

  For example, a method is considered in which an evaluation value for a certain period is calculated based on an error between a predicted theoretical value and an actual value obtained by making a prediction with reference to a past actual value, and a model with the best evaluation value is adopted. (See Patent Document 1). In this method, variance or standard deviation is adopted as the evaluation value.

Also, calculate the prediction error from the past prediction amount and the actual amount stored in the database instead of the prediction theoretical value, calculate the evaluation value from the prediction error for a certain period, and select the prediction model with the best evaluation value. There is also a method of selecting (see, for example, Patent Document 2). In this method, the mean square error is adopted as the evaluation value.
JP 2000-250888 A JP 2001-014295 A

  However, in the method described in Patent Document 1, the standard deviation of the error between the predicted theoretical value and the actual value is used as an index, and the past predicted amount is not evaluated. That is, the method described in Patent Document 1 substitutes a prediction theoretical value close to the past prediction amount in order to suppress the troublesome work of obtaining the past prediction amount. However, since the predicted theoretical value is merely a substitute for the predicted amount, the past predicted error is not necessarily minimized even if the standard deviation of the predicted theoretical value error is the smallest. Therefore, there is a possibility that other prediction models that have not been selected have a smaller degree of variation in past prediction errors, and there has been a problem that an optimal prediction model is not necessarily selected.

  Further, the method described in Patent Document 2 calculates an evaluation value based on past prediction errors. However, for example, as the forecast amount for July, the forecast amount predicted based on the results until June, the forecast amount predicted based on the results until May, and the forecast amount based on the results until April A plurality of types of prediction amounts can be considered depending on when the prediction amount and the prediction amount of the same cycle (for example, the same month) are predicted. If it is said case, normally, the prediction amount of July estimated based on the performance until the nearest June is used. However, when obtaining the safety stock quantity in the order point inventory quantity two cycles ahead of the result reference end cycle, it may be desirable that the accuracy of the predicted quantity in the two cycles ahead is better than the one cycle ahead.

  That is, there is a great need for selecting a prediction model that does not necessarily target one cycle ahead and that has a high prediction amount accuracy after two cycles. However, the method described in Patent Document 2 has a problem that it does not take into account the error in the predicted amount two cycles ahead.

  On the other hand, in inventory management, the degree of variation in prediction error is set on a cycle basis. For example, when the procurement period is N cycles longer than one cycle, the degree of variation in accumulated prediction error is √N × (degree of variation in prediction error for one cycle). This assumption is derived from the assumption that the prediction error follows a normal distribution, but actually includes various external factors (noise). For this reason, the distribution of the prediction error is not a normal distribution, and the variation degree of the cumulative prediction error of N cycles is often not √N × (the degree of variation of the prediction error of one cycle), and an appropriate safety stock can be set. The possibility of excess inventory and shortage was increasing.

  Therefore, when the procurement period is not a single cycle, there is a problem whether or not the evaluation value of the prediction error of one cycle may be used as an index. In particular, when calculating the order quantity in the regular order system, a predicted quantity of two cycles or more is required, and therefore, the degree of variation in prediction errors of a plurality of cycles is essential.

  On the other hand, in the method described in Patent Document 1, since a prediction theoretical value is used, only a single cycle prediction error can be evaluated, and the evaluation of the cumulative prediction error is not considered at all. In addition, the method described in Patent Document 2 does not mention at all the evaluation of the cumulative prediction error of a plurality of cycles.

  The present invention has been made in consideration of such circumstances, and in consideration of the delivery lead time in inventory management, etc., by appropriately evaluating the prediction accuracy of a single or a plurality of cycles, from a plurality of prediction models. An object of the present invention is to provide a prediction model selection device, a prediction model selection method, and a program capable of suitably selecting one prediction model.

In order to solve the above problems, a prediction model selection device according to the present invention comprises the following arrangement. That is,
A prediction model selection device that selects one prediction model from a plurality of prediction models,
First storage means for storing the actual amount of demand for the item in the storage device;
As a prediction in a plurality of future cycles at a past time point, a predicted amount of demand for the item calculated from an actual amount of demand for the item using each of the plurality of prediction models is stored in the storage device for each prediction model. Second storage means for storing
Cycle setting means for setting, for each item, a setting value for determining an accumulation target start cycle indicating a start cycle and an accumulation target end cycle indicating an end cycle of the accumulation period of the prediction amount stored in the storage device When,
Determining means for determining the accumulation target start cycle and the accumulation target end cycle according to a set value corresponding to an item for which the prediction model is selected, set by the cycle setting means;
As the predicted amount of each cycle included in the accumulation period specified by the accumulation target start cycle and the accumulation target end cycle determined by the determination means, the demand of the item calculated using each of a plurality of prediction models First calculation means for acquiring a predicted amount from the storage device and calculating a cumulative predicted amount obtained by calculating a total of the acquired predicted amounts for each of the plurality of prediction models;
The actual amount of the item in each cycle included in the accumulation period specified by the accumulation target start cycle and the accumulation target end cycle determined by the determination unit is acquired from the storage device, and the total of the acquired actual amounts Calculating a cumulative performance amount of the actual amount in the cumulative period,
A prediction error calculating means for calculating a prediction error of each item of each of the plurality of prediction models by taking a difference between the cumulative performance amount of the actual amount and the cumulative prediction amount of the prediction amount by each of the plurality of prediction models ; ,
Evaluation period setting means for setting an evaluation period for evaluating the prediction model;
Based on the prediction error calculated by the prediction error calculation means as the prediction error of each of the plurality of prediction models in the evaluation period set by the evaluation period setting means, the evaluation value of each prediction model in the evaluation period is calculated. An evaluation value calculating means for calculating by an evaluation formula,
And a prediction model selection unit that selects the prediction model having the best evaluation value from the plurality of prediction models for each item.

Preferably, the evaluation value calculating unit is configured such that the prediction error calculating unit takes a difference between the accumulated predicted amount of the predicted amount and the accumulated actual amount of the actual amount of each cycle included in the accumulated period. calculating a mean square error of the accumulated error amount calculated as a prediction error as the evaluation value, the prediction model selection means selects one of the prediction model in which the evaluation value is minimized.

Further preferably, the evaluation value calculation means is calculated by the prediction error calculation means taking a difference between the accumulated predicted amount of the predicted amount and the accumulated actual amount of the actual amount of each cycle included in the accumulation period. by dividing the accumulated amount of error and the cumulative predicted amount of the predictors, the mean square error of the accumulated error rate calculated as the prediction error calculated by said evaluation value, the prediction model selection means, the evaluation value Select one prediction model that minimizes.

The evaluation value calculating means, the prediction error calculating means, calculated as the prediction error by taking the difference between the accumulated actual amount of the actual amount and the cumulative predicted amount of the predicted amount of each cycle included in the cumulative period calculating a standard deviation of the accumulated error amount as the evaluation value, the prediction model selection means selects one of the prediction model which the evaluation value is minimized.

Still further preferably, the cycle setting means sets at least one of the accumulation target start cycle or the accumulation target end cycle as numerical data indicating the number of cycles before the actual reference end cycle. characterized in that it.

  Still preferably, the cycle setting means allows the accumulation target start cycle and the accumulation target end cycle to be set as the same cycle.

Furthermore, in order to solve the above problems, the prediction model selection method according to the present invention comprises the following arrangement. That is,
A prediction model selection method in a prediction model selection device that selects one prediction model from a plurality of prediction models,
A first storage step in which the first storage means of the prediction model selection device stores the actual amount of demand for the item in the storage device;
The second storage means of the prediction model selection device uses the demand of the item calculated from the actual amount of demand of the item using each of the plurality of prediction models as prediction in a plurality of future cycles at a past time point. A second storage step of storing the predicted amount in the storage device for each prediction model;
A setting value for the cycle setting means of the prediction model selection device to determine an accumulation target start cycle indicating a start cycle and an accumulation target end cycle indicating an end cycle of the accumulation period of the prediction amount stored in the storage device , A cycle setting step for setting for each item,
A determination step in which the determination unit of the prediction model selection device determines the accumulation target start cycle and the accumulation target end cycle according to setting values corresponding to an item for which the prediction model is selected, which is set in the cycle setting step. When,
The first calculation means of the prediction model selection device includes a plurality of predictions as prediction amounts of each cycle included in the accumulation period specified by the accumulation target start cycle and the accumulation target end cycle determined by the determination step. A predicted amount of demand for the item calculated using each model is acquired from the storage device, and a cumulative predicted amount obtained by calculating a total of the acquired predicted amounts is calculated for each of the plurality of predicted models. A calculation process;
The second calculation means of the prediction model selection device calculates the actual amount of the item in each cycle included in the accumulation period specified by the accumulation target start cycle and the accumulation target end cycle determined by the determination step. A second calculation step of calculating a cumulative actual amount of the actual amount in the cumulative period by acquiring from the storage device and calculating the total of the acquired actual amount;
The prediction error calculation means of the prediction model selection device takes the difference between the accumulated actual amount of the actual amount and the accumulated predicted amount of the predicted amount by each of the plurality of prediction models, so that the each of the plurality of prediction models A prediction error calculation step of calculating an item prediction error;
An evaluation period setting step in which the evaluation period setting means of the prediction model selection device sets an evaluation period for evaluating the prediction model;
Based on the prediction error calculated in the prediction error calculation step, as the prediction error of each of the plurality of prediction models in the evaluation period set by the evaluation value calculation means of the prediction model selection device in the evaluation period setting step, An evaluation value calculating step of calculating an evaluation value of each of the prediction models in the evaluation period by an evaluation arithmetic expression;
A prediction model selection unit of the prediction model selection device includes a prediction model selection step of selecting, for each item, a prediction model having the best evaluation value from the plurality of prediction models.

Furthermore, a computer program according to the present invention for achieving the above object comprises the following arrangement. That is,
A computer program for causing a computer to function as a prediction model selection device that selects one prediction model from a plurality of prediction models,
The computer,
First storage means for storing the actual amount of demand for the item in the storage device;
As a prediction in a plurality of future cycles at a past time point, a predicted amount of demand for the item calculated from an actual amount of demand for the item using each of the plurality of prediction models is stored in the storage device for each prediction model. Second storage means for storing
Cycle setting means for setting, for each item, a setting value for determining an accumulation target start cycle indicating a start cycle and an accumulation target end cycle indicating an end cycle of the accumulation period of the prediction amount stored in the storage device When,
Determining means for determining the accumulation target start cycle and the accumulation target end cycle according to a set value corresponding to an item for which the prediction model is selected, set by the cycle setting means;
As the predicted amount of each cycle included in the accumulation period specified by the accumulation target start cycle and the accumulation target end cycle determined by the determination means, the demand of the item calculated using each of a plurality of prediction models First calculation means for acquiring a predicted amount from the storage device and calculating a cumulative predicted amount obtained by calculating a total of the acquired predicted amounts for each of the plurality of prediction models;
The actual amount of the item in each cycle included in the accumulation period specified by the accumulation target start cycle and the accumulation target end cycle determined by the determination unit is acquired from the storage device, and the total of the acquired actual amounts Calculating a cumulative performance amount of the actual amount in the cumulative period,
A prediction error calculating means for calculating a prediction error of each item of each of the plurality of prediction models by taking a difference between the cumulative performance amount of the actual amount and the cumulative prediction amount of the prediction amount by each of the plurality of prediction models; ,
Evaluation period setting means for setting an evaluation period for evaluating the prediction model;
Based on the prediction error calculated by the prediction error calculation means as the prediction error of each of the plurality of prediction models in the evaluation period set by the evaluation period setting means, the evaluation value of each prediction model in the evaluation period is calculated. An evaluation value calculating means for calculating by an evaluation formula,
For each item, a prediction model selection device including: a prediction model selection unit that selects the prediction model having the best evaluation value from the plurality of prediction models is caused to function.

  According to the present invention, it is possible to appropriately select one prediction model from a plurality of prediction models by appropriately evaluating the prediction accuracy of a single or a plurality of cycles in consideration of a delivery lead time in inventory management. .

  Hereinafter, with reference to drawings, the prediction model selection device concerning the embodiment of the present invention and the details of the selection operation are explained. In the following embodiment, the evaluation cycle is “month”.

  FIG. 1 is a configuration diagram of a prediction model selection device according to an embodiment of the present invention. As illustrated in FIG. 1, the prediction model selection device according to the present embodiment includes a database 100, an evaluation parameter setting unit 110, a prediction result evaluation value calculation unit 120, and a prediction model selection unit 130.

  The database (DB) 100 includes an item master 101, an evaluation parameter DB 102, a prediction model information DB 103, a demand result DB 104, and a prediction result DB 105. The item master 101 stores an allowable shortage rate, a procurement period, and the like for each product. The evaluation parameter DB 102 stores evaluation parameters such as an accumulation target start cycle, an accumulation target end cycle, an evaluation period, and an evaluation method. The prediction model information DB 103 stores various prediction models that can be predicted by the prediction model selection device (for example, moving average with variation, exponential smoothing with variation, winters without variation, etc.). In the demand record DB 104, the past record of demand record by product and period is accumulated. In the prediction result DB 105, past demand prediction result data of each prediction model is accumulated.

  The evaluation parameter setting unit 110 sets the accumulation target start cycle, the accumulation target end cycle, the evaluation period, the evaluation method, and the like stored in the evaluation parameter DB 102. The prediction result evaluation value calculation unit 120 calculates an evaluation value for each of a plurality of prediction models for the accumulation target cycle and the evaluation target period set by the evaluation parameter setting unit 110. The prediction model selection unit 130 selects an optimal prediction model based on the evaluation value calculated by the prediction result evaluation value calculation unit 120.

  FIG. 15 is a block diagram showing a detailed configuration of an information processing apparatus (computer) that realizes a prediction model selection apparatus according to an embodiment of the present invention. In FIG. 15, the CPU 201 comprehensively controls each device connected to the system bus 212. The RAM 202 functions as a main memory, work area, temporary save area, and the like for the CPU 201. Further, the ROM 203 stores a boot program.

  The input control unit 204 controls input from the input unit 205 realized by a pointing device such as a keyboard or a mouse. The display control unit 206 controls display by the display unit 207 realized by a CRT monitor, a liquid crystal monitor, or the like. Furthermore, the external memory control unit 208 includes various applications such as an operation system (OS) that is a control program of the CPU 201, programs for realizing various functions of the prediction model selection device, the various databases (DB), various files, and the like. Controls access to the external memory 209 that stores various data such as various master files and user files.

  The external memory 209 further stores various tables for realizing various functions of the prediction data selection device. Examples of the external memory 209 include a hard disk (HD), a flexible disk (FD), a compact flash (registered trademark) that can be connected to a PCMCIA card slot via an adapter, smart media, and the like. In the processing in this embodiment, a program recorded in the external memory 209 by the CPU 201 is loaded into the RAM 202 as necessary, and is executed by the CPU 201 based on control by the program.

  In addition, the communication I / F control unit 210 executes communication control processing with an external device via a network 211 such as a LAN or the Internet.

  FIG. 2 is a diagram illustrating an example of demand record data (time series data) stored in the demand record DB 104. FIG. 2 shows actual demand values for each product from January 2001 to March 2004. FIG. 3 is a diagram showing the actual demand data (time series data) shown in FIG. 2 as a line graph. As shown in FIG. 2 and FIG. 3, in the present embodiment, an example of a product with a demand record that causes a significant drop in the amount of goods delivered in a specific month and no special tendency in other months is given as an example.

  FIG. 4 is a diagram illustrating an example of past demand prediction result data accumulated in the prediction result DB 105. FIG. 5 is an image diagram for explaining the prediction history of the past demand prediction result data shown in FIG. As shown in FIG. 4 and FIG. 5, the prediction result in which the result reference end month is September 2003 is four prediction results from October 2003 to January 2004 based on the result data until September 2003. The result of the monthly forecast is shown. As shown in FIGS. 4 and 5, it can be seen that there are data with different performance reference end cycles even with the same product code and the same scheduled delivery month.

FIG. 6 is a diagram for explaining an example of the relationship between the result reference period and the prediction period. In FIG. 6, for example, an actual reference period of 24 months from October 2001 to September 2003 is used for a prediction period from October 2003 to January 2004. Similarly, a performance reference period of 24 months from November 2001 to October 2003 is used for the forecast period from November 2003 to February 2004. In this way, in the example shown in FIG. 6, it can be seen that the result reference period differs depending on the timing to be predicted and is shifted to the right by one cycle. Therefore, even if it is the same scheduled delivery month , it can be said that the predicted amount may be different if the result reference end cycle is different.

  FIG. 7 is a flowchart for explaining an example of the operation procedure of the prediction model selection device according to the embodiment of the present invention. First, parameters α1 and α2 indicating the cycle of the accumulation target prediction amount (in this embodiment, month) are set (step S701). Here, α1 indicates the accumulation target start cycle, and α2 indicates the accumulation target end cycle. For the set α1 and α2, this means that the cumulative prediction error from the cycle of the result reference end cycle + α1 to the cycle of the result reference end cycle + α2 is set as one of the evaluation targets.

  FIG. 8 is a diagram illustrating an example of a screen for setting an accumulation target cycle, an evaluation period, an evaluation method, and the like in step S701. That is, on the screen shown in FIG. 8, an evaluation period 801, an accumulation target start cycle 802, an accumulation target end cycle 803, an evaluation method 804, and a target model (registered model) 805 can be set for each product code.

  The evaluation period 801 is an evaluation implementation target cycle from the latest cycle with a track record. Further, the accumulation target start cycle 802 corresponds to the cycle serial number from the result reference end cycle, and is an evaluation target for the accumulated predicted amount from the accumulation target start cycle 802 ahead of the result reference end cycle to the accumulation target end cycle 803 ahead. It means that. The evaluation method 804 indicates an evaluation method in which each cumulative prediction error at the time of performing each prediction or a function value of the cumulative prediction error is converted into an evaluation value. Examples of the evaluation method include a prediction error, a standard deviation of a prediction error rate, a square error average, and the like. The target model 805 is a part for setting a plurality of prediction models, and one prediction model is selected from the set prediction models.

  Next, the first prediction model of the target prediction model group for calculating the prediction amount is set (step S702). Note that i indicates the number of the target prediction model. For example, when there are three target prediction models, the numbers are 1, 2, and 3 sequentially. Subsequently, the counter (i) of the result reference end cycle for calculating the cumulative prediction amount is initialized to 1 (step S703).

  Thereafter, based on i set in step S703, an actual result reference end cycle for calculating the cumulative prediction amount is determined (step S704). Then, a cumulative prediction amount of the prediction amount of the designated cycle having the designated achievement reference end cycle is obtained (step S705).

  FIG. 9 is a diagram for explaining the process of calculating the cumulative prediction amount in step S705. Here, it is assumed that the accumulation target start cycle is set to 1 and the accumulation target end cycle is set to 2. For example, in the predicted amount in which January, which is two cycles (C cycle) in March, which is the latest cycle with actual results, is the actual reference end month, February, which corresponds to the first cycle that is the accumulation target start cycle α1 The predicted amount (3010) of March and the predicted amount (3050) of March corresponding to the second cycle, which is the accumulation target end cycle α2, are acquired, and the sum of the two months of February and March is calculated to be 6060.

  Next, the actual cumulative amount in the same cycle period as the cumulative predicted amount obtained in step S705 is obtained (step S706). FIG. 10 is a diagram for explaining the process of calculating the cumulative actual amount from the past actual amount in step S706. As shown in FIG. 9, in step S706, the past cumulative period is shifted one by one to calculate the cumulative amount of the actual amount (cumulative actual amount). When the total of the actual results for the same two months described with reference to FIG. 9 is calculated, 6112 is obtained.

  Then, an accumulated prediction error is obtained from the accumulated predicted amount obtained in step S705 and the accumulated actual amount obtained in step S706 (step S707). That is, the difference between the cumulative predicted amount and the cumulative actual amount in the same period compared in February and March is 52, and this is the cumulative prediction error when the actual reference end month is January 2003. Thereafter, the process proceeds to calculation of a prediction error in the actual reference end cycle calculated based on the counter (j + 1) of the next actual reference end cycle (step S708). In other words, in the case of the above example, the cumulative reference error for January and February is calculated in the same manner as December 2003 when the actual reference end month is one month ago.

  Here, it is determined whether or not all prediction errors for the evaluation period have been calculated (step S709). As a result, when all the prediction errors for the evaluation period have not been calculated, that is, when j is equal to or less than the evaluation period (Yes), the process returns to step S704 and the above process is repeated. As described above, when the above-described operation is repeated by the number of evaluation period values, the cumulative prediction error of each result reference end month can be calculated. FIG. 11 is a diagram illustrating a cumulative prediction error that is an error between the cumulative prediction amount calculated using FIGS. 9 and 10 and the cumulative actual amount.

  On the other hand, when all the prediction errors for the evaluation period are calculated in step S112, that is, when j is larger than the evaluation period (No), the process proceeds to step S710.

  In step S710, an evaluation value is calculated from each accumulated prediction error for the evaluation period. For example, when the cumulative prediction error as shown in FIG. 11 is obtained, the mean square error is obtained from the cumulative prediction error of each actual reference end month as shown in the following equation.

  Then, the process proceeds to the next prediction model (i + 1) (step S711). Here, it is determined whether or not there is a next prediction model, that is, whether or not evaluation values of all prediction models have been calculated (step S712).

As a result, if not all the prediction models have been calculated, it is determined that there is a next prediction model (Yes), and the process returns to step S703 to perform the above-described processing on the next prediction model to calculate an evaluation value. To do. On the other hand, when all the prediction models are calculated, it is determined that there is no next prediction model (No), and the prediction model having the best evaluation value is selected based on the calculated evaluation value of each prediction model (step S7). 13 ).

  FIG. 12 is a diagram showing a list of evaluation values calculated for each target prediction model shown in FIG. In this embodiment, the smallest “exponential smoothing with fluctuation” prediction model is selected from the evaluation values shown in FIG.

  FIG. 13 is a diagram illustrating an example when the accumulation target start cycle and the accumulation target end cycle are the same second cycle in the process of calculating the cumulative prediction amount in step S705. When the accumulation target start cycle and the accumulation target end cycle are the same, the prediction amount of one target cycle is the evaluation target.

  FIG. 14 is a diagram illustrating a prediction error function value when the prediction error function is calculated as “prediction error ÷ prediction amount” in the diagram illustrated in FIG. 13. This function is a ratio of the prediction error to the prediction amount, and indicates how much the prediction is deviated from the prediction amount. This function is effective, for example, when it is assumed that the fluctuation range of the outgoing quantity is large and that the predicted fluctuation width is proportional to the magnitude of the outgoing quantity. The evaluation value of the prediction error function value can be calculated as follows.

  As described above, the case where the prediction error and the mean square error of the prediction error rate are used as the evaluation value has been described as the evaluation method. As described above, as the prediction error evaluation method, for example, the prediction error, the standard deviation of the prediction error rate, etc. Of course, it is also possible to use.

  Although the embodiments have been described in detail above, the present invention can take embodiments as, for example, a system, an apparatus, a method, a program, or a storage medium (recording medium). The present invention may be applied to a system composed of a single device or an apparatus composed of a single device.

  In the present invention, a software program (in the embodiment, a program corresponding to the flowchart shown in the figure) that realizes the functions of the above-described embodiment is directly or remotely supplied to the system or apparatus, and the computer of the system or apparatus Is also achieved by reading and executing the supplied program code.

  Accordingly, since the functions of the present invention are implemented by computer, the program code installed in the computer also implements the present invention. In other words, the present invention includes a computer program itself for realizing the functional processing of the present invention.

  In that case, as long as it has the function of a program, it may be in the form of object code, a program executed by an interpreter, script data supplied to the OS, or the like.

  As a recording medium for supplying the program, for example, floppy (registered trademark) disk, hard disk, optical disk, magneto-optical disk, MO, CD-ROM, CD-R, CD-RW, magnetic tape, nonvolatile memory card ROM, DVD (DVD-ROM, DVD-R) and the like.

  As another program supply method, a client computer browser is used to connect to an Internet homepage, and the computer program of the present invention itself or a compressed file including an automatic installation function is downloaded from the homepage to a recording medium such as a hard disk. Can also be supplied. It can also be realized by dividing the program code constituting the program of the present invention into a plurality of files and downloading each file from a different homepage. That is, a WWW server that allows a plurality of users to download a program file for realizing the functional processing of the present invention on a computer is also included in the present invention.

  In addition, the program of the present invention is encrypted, stored in a storage medium such as a CD-ROM, distributed to users, and key information for decryption is downloaded from a homepage via the Internet to users who have cleared predetermined conditions. It is also possible to execute the encrypted program by using the key information and install the program on a computer.

  In addition to the functions of the above-described embodiments being realized by the computer executing the read program, the OS running on the computer based on the instruction of the program is a part of the actual processing. Alternatively, the functions of the above-described embodiment can be realized by performing all of them and performing the processing.

  Furthermore, after the program read from the recording medium is written in a memory provided in a function expansion board inserted into the computer or a function expansion unit connected to the computer, the function expansion board or The CPU or the like provided in the function expansion unit performs part or all of the actual processing, and the functions of the above-described embodiments are realized by the processing.

It is a block diagram of the prediction model selection apparatus which concerns on one Embodiment of this invention. It is a figure which shows an example of the demand performance data (time series data) accumulate | stored in demand performance DB104. It is the figure which displayed the demand performance data (time series data) shown in FIG. 2 with the line graph. It is a figure which shows an example of the past demand prediction result data accumulate | stored in prediction result DB105. It is an image figure for demonstrating the prediction history of the past demand prediction result data shown in FIG. It is a figure for demonstrating an example of the relationship between a performance reference period and a prediction period. It is a flowchart for demonstrating an example of the operation | movement procedure of the prediction model selection apparatus which concerns on one Embodiment of this invention. It is a figure which shows an example of the screen for setting parameters for evaluation, such as an accumulation object cycle, in step S701. It is a figure for demonstrating the process which calculates the accumulation estimated amount in step S705. It is a figure for demonstrating the process in which accumulated performance amount is calculated from past performance amount in step S706. It is a figure which shows the cumulative prediction error which is the difference | error of the cumulative prediction amount calculated and demonstrated using FIG. 9 and FIG. 10, and a cumulative performance amount. It is a figure which shows the list of the evaluation values calculated with respect to each object prediction model shown in FIG. It is a figure which shows the example in case the accumulation object start cycle and accumulation object end cycle are the same 2nd cycles in the process of calculating the accumulation prediction amount in step S705. It is a figure which shows the function value of a prediction error at the time of calculating the function of a prediction error as "prediction error / prediction amount" in the figure shown in FIG. It is a block diagram which shows the detailed structure of the information processing apparatus (computer) which implement | achieves the prediction model selection apparatus which concerns on one Embodiment of this invention.

Explanation of symbols

100 database (DB)
101 Item master 102 Evaluation parameter DB
103 Prediction model information DB
104 Demand results DB
105 Prediction result DB
110 Evaluation Parameter Setting Unit 120 Prediction Result Evaluation Value Calculation Unit 130 Prediction Model Selection Unit

Claims (8)

A prediction model selection device that selects one prediction model from a plurality of prediction models,
First storage means for storing the actual amount of demand for the item in the storage device;
As a prediction in a plurality of future cycles at a past time point, a predicted amount of demand for the item calculated from an actual amount of demand for the item using each of the plurality of prediction models is stored in the storage device for each prediction model. Second storage means for storing
Cycle setting means for setting, for each item, a setting value for determining an accumulation target start cycle indicating a start cycle and an accumulation target end cycle indicating an end cycle of the accumulation period of the prediction amount stored in the storage device When,
Determining means for determining the accumulation target start cycle and the accumulation target end cycle according to a set value corresponding to an item for which the prediction model is selected, set by the cycle setting means;
As the predicted amount of each cycle included in the accumulation period specified by the accumulation target start cycle and the accumulation target end cycle determined by the determination means, the demand of the item calculated using each of a plurality of prediction models First calculation means for acquiring a predicted amount from the storage device and calculating a cumulative predicted amount obtained by calculating a total of the acquired predicted amounts for each of the plurality of prediction models;
The actual amount of the item in each cycle included in the accumulation period specified by the accumulation target start cycle and the accumulation target end cycle determined by the determination unit is acquired from the storage device, and the total of the acquired actual amounts Calculating a cumulative performance amount of the actual amount in the cumulative period,
A prediction error calculating means for calculating a prediction error of each item of each of the plurality of prediction models by taking a difference between the cumulative performance amount of the actual amount and the cumulative prediction amount of the prediction amount by each of the plurality of prediction models ; ,
Evaluation period setting means for setting an evaluation period for evaluating the prediction model;
Based on the prediction error calculated by the prediction error calculation means as the prediction error of each of the plurality of prediction models in the evaluation period set by the evaluation period setting means, the evaluation value of each prediction model in the evaluation period is calculated. An evaluation value calculating means for calculating by an evaluation formula,
A prediction model selection device comprising: a prediction model selection unit that selects, for each item, a prediction model having the best evaluation value from the plurality of prediction models. The evaluation value calculation means includes
A mean square error of the cumulative error amount calculated as the prediction error by the prediction error calculation means taking the difference between the cumulative prediction amount of the prediction amount and the cumulative actual amount of the actual amount of each cycle included in the cumulative period. calculated as the evaluation value,
The prediction model selection device according to claim 1, wherein the prediction model selection unit selects one prediction model that minimizes the evaluation value. The evaluation value calculation means includes
The prediction error calculating means, the cumulative prediction of the predicted amount of accumulated error amount calculated by taking the difference between the accumulated actual amount of the actual amount and the cumulative predicted amount of the predicted amount of each cycle included in the cumulative period by dividing by the mean square error of the accumulated error rate calculated as the prediction error calculated by said evaluation value,
The prediction model selection device according to claim 1, wherein the prediction model selection unit selects one prediction model that minimizes the evaluation value. The evaluation value calculation means includes
The prediction error calculating means, the standard deviation of accumulated error amount calculated as the prediction error by taking the difference between the accumulated actual amount of the actual amount and the cumulative predicted amount of the predicted amount of each cycle included in the cumulative period calculated as the evaluation value,
The prediction model selection device according to claim 1, wherein the prediction model selection unit selects one prediction model that minimizes the evaluation value . The cycle setting means includes
5. At least one of the accumulation target start cycle and the accumulation target end cycle is set as numerical data indicating the number of cycles before the actual reference end cycle. The prediction model selection device according to any one of the above. The cycle setting means includes
The prediction model selection device according to claim 5, wherein the accumulation target start cycle and the accumulation target end cycle are allowed to be set as the same cycle. A prediction model selection method in a prediction model selection device that selects one prediction model from a plurality of prediction models,
A first storage step in which the first storage means of the prediction model selection device stores the actual amount of demand for the item in the storage device;
The second storage means of the prediction model selection device uses the demand of the item calculated from the actual amount of demand of the item using each of the plurality of prediction models as prediction in a plurality of future cycles at a past time point. A second storage step of storing the predicted amount in the storage device for each prediction model;
A setting value for the cycle setting means of the prediction model selection device to determine an accumulation target start cycle indicating a start cycle and an accumulation target end cycle indicating an end cycle of the accumulation period of the prediction amount stored in the storage device , A cycle setting step for setting for each item,
A determination step in which the determination unit of the prediction model selection device determines the accumulation target start cycle and the accumulation target end cycle according to setting values corresponding to an item for which the prediction model is selected, which is set in the cycle setting step. When,
The first calculation means of the prediction model selection device includes a plurality of predictions as prediction amounts of each cycle included in the accumulation period specified by the accumulation target start cycle and the accumulation target end cycle determined by the determination step. A predicted amount of demand for the item calculated using each model is acquired from the storage device, and a cumulative predicted amount obtained by calculating a total of the acquired predicted amounts is calculated for each of the plurality of predicted models. A calculation process;
The second calculation means of the prediction model selection device calculates the actual amount of the item in each cycle included in the accumulation period specified by the accumulation target start cycle and the accumulation target end cycle determined by the determination step. A second calculation step of calculating a cumulative actual amount of the actual amount in the cumulative period by acquiring from the storage device and calculating the total of the acquired actual amount;
The prediction error calculation means of the prediction model selection device takes the difference between the accumulated actual amount of the actual amount and the accumulated predicted amount of the predicted amount by each of the plurality of prediction models, so that the each of the plurality of prediction models A prediction error calculation step of calculating an item prediction error;
An evaluation period setting step in which the evaluation period setting means of the prediction model selection device sets an evaluation period for evaluating the prediction model;
Based on the prediction error calculated in the prediction error calculation step, as the prediction error of each of the plurality of prediction models in the evaluation period set by the evaluation value calculation means of the prediction model selection device in the evaluation period setting step, An evaluation value calculating step of calculating an evaluation value of each of the prediction models in the evaluation period by an evaluation arithmetic expression;
A prediction model selection means of the prediction model selection device comprises: a prediction model selection step for selecting, for each item, a prediction model having the best evaluation value from the plurality of prediction models. Selection method. A computer program for causing a computer to function as a prediction model selection device that selects one prediction model from a plurality of prediction models,
The computer,
First storage means for storing the actual amount of demand for the item in the storage device;
As a prediction in a plurality of future cycles at a past time point, a predicted amount of demand for the item calculated from an actual amount of demand for the item using each of the plurality of prediction models is stored in the storage device for each prediction model. Second storage means for storing
Cycle setting means for setting, for each item, a setting value for determining an accumulation target start cycle indicating a start cycle and an accumulation target end cycle indicating an end cycle of the accumulation period of the prediction amount stored in the storage device When,
Determining means for determining the accumulation target start cycle and the accumulation target end cycle according to a set value corresponding to an item for which the prediction model is selected, set by the cycle setting means;
As the predicted amount of each cycle included in the accumulation period specified by the accumulation target start cycle and the accumulation target end cycle determined by the determination means, the demand of the item calculated using each of a plurality of prediction models First calculation means for acquiring a predicted amount from the storage device and calculating a cumulative predicted amount obtained by calculating a total of the acquired predicted amounts for each of the plurality of prediction models;
The actual amount of the item in each cycle included in the accumulation period specified by the accumulation target start cycle and the accumulation target end cycle determined by the determination unit is acquired from the storage device, and the total of the acquired actual amounts Calculating a cumulative performance amount of the actual amount in the cumulative period,
A prediction error calculating means for calculating a prediction error of each item of each of the plurality of prediction models by taking a difference between the cumulative performance amount of the actual amount and the cumulative prediction amount of the prediction amount by each of the plurality of prediction models; ,
Evaluation period setting means for setting an evaluation period for evaluating the prediction model;
Based on the prediction error calculated by the prediction error calculation means as the prediction error of each of the plurality of prediction models in the evaluation period set by the evaluation period setting means, the evaluation value of each prediction model in the evaluation period is calculated. An evaluation value calculating means for calculating by an evaluation formula,
The computer program for functioning as a prediction model selection apparatus provided with the prediction model selection means which selects the one prediction model with the said best evaluation value from these prediction models for every said item.

Images