JP2007293624A - Demand prediction method and demand prediction program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a demand prediction method and a demand prediction program for further accurately predicting demand of articles by using a more suitable demand prediction system. <P>SOLUTION: A managing computer 21 acquires order reception result data to be predicted from an order reception result data storage unit 22. The managing computer 21 calculates temporary demand prediction functions of a cumulative secondary system, a cumulative tertiary system, and a cumulative quartic system by using order reception result data excluding data of recent three months from the acquired order reception result data to be predicted. The managing computer 21 calculates a predicted value of the recent three months by using the temporary demand prediction functions, finds differences between the predicted value and order reception results of the respective months, and calculates the total of the differences as an error. The managing computer 21 uses the function of the system having the smallest error to calculate a demand prediction function by using all of the order reception result data, and uses the demand prediction function to calculate predicted values of service parts to be predicted. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a demand prediction method and a demand prediction program for predicting the demand for goods.

  When providing products to customers, it is necessary to manage product inventory appropriately. This product includes not only finished products but also consumables used for finished products, replacement parts due to failure, and the like. And by carrying out accurate inventory management, it is possible to suppress inventory loss due to surplus inventory and opportunity loss due to shortage of inventory.

  Accurate demand forecasting is necessary for accurate inventory management. In such demand prediction, for example, multiple regression analysis is used. In this multiple regression analysis, past results are analyzed to create a prediction formula. However, if the prediction formula once created is continuously used, the error between the predicted value and the actual value may increase. Therefore, it is conceivable to recreate the prediction formula based on the results for each short period, but the load for creating the prediction formula increases.

  For this reason, a sales prediction method for performing sales prediction for each product category in consideration of the variation factor has been proposed (for example, see Patent Document 1). In this sales prediction method, first, a moving average value is calculated by averaging sales results of a product whose sales volume is predicted for a predetermined period. Then, the fluctuation forecast quantity of the sales quantity of the product is calculated from the fluctuation factors that are considered to affect the sales quantity on the forecast sales date. Further, the sales forecast quantity is calculated by correcting the moving average value based on the fluctuation forecast quantity. Thereby, the sales performance in the predetermined period immediately before the sales prediction date can be reflected in the sales prediction, and the followability of the predicted value can be improved.

  However, the demand prediction based on the moving average value has a large prediction error due to a time lag, and it is necessary to prepare a large amount of safety stock. In addition, when calculating the sales forecast quantity by correcting the moving average value based on the forecasted fluctuation quantity, identify the fluctuation factors that may affect the sales quantity on the sales forecast date, and then change the sales quantity of the product. It is necessary to calculate the predicted quantity. However, fluctuations in the sales volume of merchandise are often derived from various factors, and it is not easy to identify them.

  Therefore, a demand prediction method and a demand prediction program for more efficiently and accurately forecasting the demand for goods have been studied (for example, see Patent Document 2). In the technique described in this document, the management computer applies a growth model to the cumulative amount transition of the order quantity, and calculates a trend function for the order result. Next, the management computer calculates the transition of the difference between the order record and the trend function. Then, the periodic intensity of the difference transition is calculated using the periodogram. When the periodicity is determined based on the tuning strength, a periodic function is calculated by applying a quadratic Sin model composed of a quadratic function and a trigonometric function to the difference transition between the order record and the trend curve. Then, the management computer 21 performs demand prediction using a function generated by synthesizing the trend function and the periodic function.

  As described above, there are a plurality of demand prediction methods as a method for predicting the future order quantity from the tendency of the past order results of service parts, such as the techniques described in Patent Documents 1 and 2. In this case, it is necessary to use a more appropriate demand prediction method for service parts from among a plurality of demand prediction methods. For this reason, a standard for selecting which demand forecasting method is necessary.

Here, a method of using the contribution rate, the degree of freedom adjusted contribution rate, and the like, and comparing with the goodness of fit in the range to which the demand prediction method is applied is also conceivable. However, it has been mathematically proven that the higher the order is, the higher the contribution ratio is (see Non-Patent Document 1, for example). For this reason, for example, when a demand prediction method having a different order from the cumulative second-order / third-order / fourth-order method is used, it cannot be evaluated that the fit is good just because the contribution rate is high.
JP 2000-339543 A (first page) JP-A-2004-234471 (first page) Kume Hitoshi and Iizuka Tsuyoshi, “Regression Analysis”, Iwanami Shoten, October 1987, p. 153-155

  On the other hand, there is a degree-of-freedom-adjusted contribution rate created for the purpose of adjusting the difference in order. However, the contribution rate and the degree of freedom adjusted contribution rate indicate the degree of fit within the range in which the demand prediction model is applied, and do not consider the future fit. This is because, when the result of the degree of freedom adjusted contribution rate is compared with the actual result, there is no correlation between the degree of freedom adjusted contribution rate and the degree of prediction fit. Therefore, even if the degree of freedom adjusted contribution rate is high, the accuracy of prediction does not necessarily increase.

  The present invention has been made in order to solve the above-described problems, and an object of the present invention is to select a more appropriate demand prediction method, and to perform a demand prediction method and a demand prediction for performing a more accurate demand prediction for a product. To provide a program.

  In order to solve the above-mentioned problem, the invention described in claim 1 is connected to a record data storage unit that records a product identifier for specifying a product and data related to the actual value of the product corresponding to the time axis in association with each other. Using the management computer, the management computer obtains prediction target specifying data including a product identifier for specifying the product for which the demand prediction is performed, and the product identifier of the product identifier is obtained. The actual value is extracted from the actual result data storage means using the product identifier, and the actual value in the provisional function calculation period excluding the most recent predetermined evaluation period is extracted among the extracted actual values. The temporary demand prediction function calculation process for calculating the temporary demand prediction function according to the method is performed, and the predicted value of the evaluation period is calculated using the calculated temporary demand prediction function. Based on the difference between the value and the actual value of the evaluation period, a method determination stage for determining a method to be adopted, and using the determined method and the actual value extracted from the actual data storage means, the demand forecast is performed. The gist is to execute a predicted value calculating stage for calculating a predicted value of a product to be performed and a demand predicted output stage for outputting the calculated predicted value.

  According to a second aspect of the present invention, in the demand forecasting method according to the first aspect, the evaluation period is set so that a plurality of actual values are included in the evaluation period, and the method determining step includes the provisional determination step. An error calculation stage that calculates an error by summing up the difference between the predicted value calculated using the demand prediction function and the actual value in the evaluation period, and a method that satisfies the condition that the error is minimum are determined. Including the decision stage.

  According to a third aspect of the present invention, in the demand forecasting method according to the first aspect, the method determining step executes the temporary demand prediction function calculation processing in a plurality of temporary function calculation periods having different lengths. The minimum that specifies the method in which the error between the predicted value and the actual value calculated using the temporary demand prediction function calculated by the temporary demand prediction function multiple calculation stage is the smallest in the evaluation function multiple calculation stage and the one evaluation period The gist includes an error method specifying step and a determination step of determining a method that satisfies the condition that the number of times that the error is specified is the largest in the plurality of provisional function calculation periods.

  According to a fourth aspect of the present invention, in the demand forecasting method according to the second or third aspect, in the determination step, when the management computer specifies a plurality of methods satisfying the condition, the plurality of specified methods are determined. The gist is to determine a method with a lower order among the methods.

  The invention according to claim 5 uses the management computer connected to the record data storage means that records the product identifier for specifying the product and the data related to the actual value of the product corresponding to the time axis, and uses the management computer. A demand prediction data including a product identifier for specifying a product for which demand prediction is performed, and using the product identifier for the actual value of the product of the product identifier. Using the actual value in the temporary function calculation period excluding the most recent predetermined evaluation period among the extracted actual values, the actual demand prediction function corresponding to a plurality of methods is extracted from the actual data storage means. The temporary demand prediction function to be calculated is calculated, the predicted value of the evaluation period is calculated using the calculated temporary demand prediction function, and the predicted value and the actual result of the evaluation period A method determining means for determining a method to be adopted based on the difference between the prediction and the prediction for calculating a predicted value of a product for which demand prediction is performed using the determined value and the actual value extracted from the actual data storage means. The gist is to function as a value calculation means and a demand prediction output means for outputting the calculated prediction value.

(Function)
According to the present invention, the management computer calculates a provisional demand prediction function corresponding to a plurality of methods using the actual value in the provisional function calculation period excluding the most recent predetermined evaluation period from the acquired order achievement data. The temporary demand prediction function calculation process is performed, the predicted value of the evaluation period is calculated using the calculated temporary demand prediction function, and the method to be adopted is determined based on the difference between the predicted value and the actual value of the evaluation period. Using the determined method, the management computer calculates and outputs the predicted value of the product for which the demand is predicted using the result extracted from the result data storage means. That is, a method suitable for prediction is selected and used based on the difference between the latest predicted value using a plurality of methods for calculating the temporary demand prediction function and the actual actual value. Therefore, a method that is closest to the latest actual value and has a good fit can be efficiently selected as a method suitable for prediction, and a predicted value can be calculated using this method. For this reason, the demand forecast of goods can be performed more correctly.

According to the present invention, by using a method with the smallest error, it is possible to more accurately and efficiently predict the demand for goods.
According to the present invention, the management computer uses a method in which the number of times that the error is specified to be minimum is the largest in a plurality of temporary function calculation periods having different lengths. For this reason, even if it is the same system, the more suitable demand prediction function system can be determined from the result of a plurality of temporary function calculation periods.

  According to the present invention, when a plurality of methods satisfying the conditions for determination are specified, the management computer determines a method satisfying the conditions for determination and having a low order. If a higher-order method is used, the value of irregular orders will be affected too much, and the predicted value will deviate greatly, and the inherent tendency may not be grasped. Therefore, when the errors are the same, the demand prediction function can be calculated using a steady tendency by using a simple model with a small order.

  According to the present invention, a more appropriate demand prediction method can be selected, and a more accurate demand prediction can be performed for a product.

(First embodiment)
Hereinafter, an embodiment of a demand prediction process embodying the present invention will be described with reference to FIGS. The present embodiment will be described as a demand forecasting method and a demand forecasting program that are used when forecasting demand for service parts based on the actual order of service parts of products provided to customers. Here, the service parts mean parts that are exchanged with service in the event of wear or failure. The service part is a minimum unit for maintaining the function of the product, and the service part includes not only a part but also a unit in which the parts are combined.

  In the present embodiment, as shown in FIG. 1, an order record is input using an order receiving system 10. Further, an order is instructed based on the demand forecast output to the order receiving system 10. The order receiving system 10 is installed in a service parts management department that manages service parts, and receives orders received at sales bases, service bases, etc., and outputs orders for production departments, purchasing departments, and the like.

  The order receiving system 10 is a computer terminal having a function of transmitting data, a function of displaying received data, and the like. The order receiving system 10 includes a CPU, RAM, and ROM (not shown), input means such as a keyboard and mouse, output means such as a display, communication means such as a communication interface, and the like.

  The order receiving system 10 is connected to the demand prediction system 20 via the network N. The demand prediction system 20 is a computer system that performs various data processing relating to demand prediction. The demand prediction system 20 includes a management computer 21.

  The management computer 21 performs data transmission / reception with the order receiving system 10, management processing of various data for executing demand prediction, and the like. This management computer 21 has a CPU, RAM, ROM, etc. (not shown), and includes later-described processes (a result determination stage, a method determination stage including an error calculation stage and a determination stage, a predicted value calculation stage, a demand prediction output stage, etc. Process). Then, by executing the demand prediction program for this purpose, the management computer 21 functions as a method determination means including a result extraction means, an error calculation means and a determination means, a predicted value calculation means, a demand prediction output means, and the like.

  The result extracting unit obtains prediction target specifying data including a product identifier for specifying a product for which demand prediction is performed, and extracts a result value of the product of the product identifier from the result data storage unit using the product identifier. The method determining means calculates a temporary demand prediction function that calculates a temporary demand prediction function corresponding to a plurality of methods using the actual value in the temporary function calculation period excluding the most recent predetermined evaluation period among the extracted actual values. Processing is performed, a predicted value for the evaluation period is calculated using the calculated provisional demand prediction function, and a method to be adopted is determined based on a difference between the predicted value and the actual value for the evaluation period. The error calculating means calculates an error obtained by summing the differences between the predicted value calculated using the temporary demand prediction function and the actual value in the evaluation period. The determining means determines a method that satisfies the condition that the error is minimum. The predicted value calculation means calculates the predicted value of the product for which the demand prediction is performed, using the actual value extracted from the actual result data storage means using the determined method. The demand forecast output means outputs the calculated forecast value.

  Further, the demand prediction system 20 includes an order record data storage unit 22 as a record data storage unit, and the management computer 21 is connected to the order record data storage unit 22.

As shown in FIG. 2, the order record data storage unit 22 stores order record data 220 regarding each service part. In the order record data 220, data on the order quantity (number of orders) of service parts is stored together with data on the service part identifier and order month. This order record data 220 is to open service parts to each service base (
When the received order record is received from the order receiving system 10, the order is set and stored. Accordingly, for each service part, the number of order record data corresponding to the opened month is stored. Specifically, the order record data 220 stores, for each service part, data related to the service part identifier, the order date, and the number of orders received.

In the service part identifier data area, data relating to an identifier for specifying a service part with a received order record is stored.
In the order date data area, data related to the date when the service parts were ordered is stored.

In the order quantity data area, data related to the order quantity of the service parts is stored in the corresponding order date. This order quantity is stored in units of months as a time-axis variable.
Accordingly, when the order results of an example of service parts stored in the order result data storage unit 22 are expressed with the order date as the horizontal axis and the number of orders as the vertical axis, they are displayed as a broken line P as shown in FIG. Can do.

  In the system configured as described above, a processing procedure for performing service part demand prediction will be described with reference to FIG. In the present embodiment, three methods of a cumulative secondary method, a cumulative tertiary method, and a cumulative quaternary method are used as methods for performing demand prediction.

  The demand prediction system 20 first acquires data (prediction target specifying data) for specifying service parts for which demand prediction is performed (step S1-1). Specifically, the management computer 21 of the demand prediction system 20 acquires the service part identifier of the service part for which the demand prediction is performed from the order receiving system 10 via the network N.

  Next, the demand prediction system 20 acquires the order result data to be predicted (step S1-2). Specifically, the management computer 21 extracts the order record data 220 having the service part identifier acquired from the order system 10 from the order record data storage unit 22.

  Next, the demand prediction system 20 calculates a temporary demand prediction function for determining a demand prediction method (step S1-3). Here, first, the management computer 21 specifies the order record in the provisional function calculation period excluding the latest n months (evaluation period) from the acquired order record data. Then, a function that is respectively fitted to the specified order record is calculated using a cumulative secondary method, a cumulative tertiary method, and a cumulative quaternary method. In the present embodiment, for example, n = 3 is set in advance in the memory of the management computer 21, so that the function of each method is calculated using the actual order received excluding the latest three months.

  Here, the case where a temporary demand prediction function is calculated about the service part of the order reception result shown in FIG. 3 is demonstrated. The service part order record data 220 is from the release month to the 45th month. Therefore, the management computer 21 calculates the function of each method as shown in FIG. 5 using the order record data 220 for 42 months (provisional function calculation period) excluding the latest three months. FIG. 5A shows a trend curve a representing a function calculated by the cumulative quadratic method. FIGS. 5B and 5C show a trend curve b representing a function calculated by the cumulative cubic method and a trend curve c representing a function calculated by the cumulative quartic method, respectively.

  Next, the management computer 21 calculates an error in each temporary demand prediction function (step S1-4). In the present embodiment, the error is calculated using the sum of the difference between the predicted value of each month not used for calculating the function and the actual result. Here, the error of each function shown in FIG. 5 will be specifically described with reference to FIG.

  First, the management computer 21 calculates a predicted value for the 43rd month using a function according to each method. Then, the management computer 21 calculates the difference between the predicted value of the 43th month and the number of orders (actual result) of the 43rd month for each method. Next, similarly for the 44th month, the management computer 21 calculates the difference between the predicted value of the 44th month and the order quantity (actual result) of the 44th month. Further, similarly for the 45th month, the management computer 21 calculates the difference between the predicted value of the 45th month and the order quantity (actual result) of the 45th month. Then, the management computer 21 calculates the sum of the difference between the predicted value calculated in the month (the 43rd month to the 45th month) not used for the function calculation and the order quantity (actual value) as an error between the predicted value and the actual value. Calculate as

  Next, the management computer 21 specifies a method with the smallest calculated error. Here, when there is one method with the smallest calculated error (in the case of “YES” in step S1-5), the management computer 21 adopts the method with the smallest error (step S1- 6) The process of step S1-8 is performed. When the error calculated by the cumulative secondary method is the smallest, the management computer 21 adopts this cumulative secondary method.

  On the other hand, when there are two or more methods having the smallest calculated error (in the case of “NO” in step S1-5), the management computer 21 adopts the method having the smallest error and the smallest degree. (Step S1-7) and Step S1-8 are performed. For example, when the error calculated by the cumulative third-order method is the same as the error calculated by the cumulative fourth-order method, and these are smaller than the cumulative second-order method, the error is the smallest and the order is the smallest. As a method, a cumulative tertiary method is adopted.

  Next, the management computer 21 calculates a demand prediction function using the adopted method (step S1-8). Specifically, the management computer 21 calculates the demand prediction function using the method adopted in step S1-6 or S1-7 for all the order record data to be predicted acquired in step S1-2. To do. For example, as shown in FIG. 6, when the cumulative secondary method is the smallest, the demand prediction function is calculated using the cumulative secondary method for the order record data up to the 45th month. FIG. 6 shows a trend curve a1 based on the function calculated here.

  Next, the management computer 21 calculates a predicted value (step S1-9). Specifically, using the demand prediction function calculated in step S1-8, for example, the predicted value of the number of orders received such as the next month or the next month is calculated.

  Then, the management computer 21 outputs the calculated predicted value result to the order receiving system 10 via the network N (step S1-10). In the order receiving system 10, the result of the predicted value is output to output means such as a display. By using this demand prediction, the service parts management department can issue an ordering instruction according to the demand.

According to this embodiment, the following effects can be obtained.
(1) In this embodiment, the management computer 21 calculates the function of each method using the received order results in the provisional function calculation period excluding the most recent n months from the acquired received order result data. Next, the management computer 21 calculates an error in each temporary demand prediction function calculated in this way (step S1-4), and uses the method with the smallest error, and uses the method of the actual demand data acquired in step S1-2. All are used to calculate a demand prediction function (step S1-8). And the management computer 21 calculates a demand prediction value using this demand prediction function (step S1-9), and outputs this result (step S1-10). For this reason, a provisional demand forecast function excluding the last n months is calculated, and the fit of each method in this service part is verified from the difference between the forecast value calculated from the provisional demand forecast function and the last n months. To efficiently select a method suitable for prediction. Therefore, it is possible to more accurately predict demand for service parts.

  (2) In this embodiment, when there are two or more methods with the smallest calculated error (in the case of “NO” in step S1-5), the management computer 21 has the smallest error and the smallest order. Adopting a small number of methods (step S1-7), the processing of the next step S1-8 is performed. For example, there may be a case where an irregular order value is included in the order record. In this case, if a higher-order method is used, the influence of this value becomes too great, and the predicted value is greatly deviated, so that there is a possibility that the original tendency cannot be grasped. Therefore, when it is considered that the error is the same and the accuracy is the same, the demand prediction function can be calculated using a steady tendency by using a simple model with a small order.

(Second Embodiment)
Next, a second embodiment of the present invention will be described with reference to FIGS. In the present embodiment, the same parts as those in the first embodiment are denoted by the same reference numerals, and detailed description thereof is omitted. Here, the present embodiment uses the same system as the first embodiment, but the method for determining the method for calculating the forecast value of the demand forecast is different.

  The management computer 21 according to the present embodiment performs processing described later (processing including a result extraction stage, a method determination stage including a provisional demand prediction function multiple calculation stage and a determination stage, a prediction value calculation stage, a demand prediction output stage, and the like). Then, by executing the demand forecasting program for this purpose, the management computer 21 can be used as a system extracting means, a provisional demand forecasting function plural calculating means, a method determining means including a determining means, a predicted value calculating means, a demand forecast outputting means, etc. Function. Here, in this embodiment, the provisional demand prediction function plural calculation means executes the provisional demand prediction function calculation processing in a plurality of provisional function calculation periods having different lengths. Further, the minimum error method specifying means specifies a method having the smallest error between the predicted value and the actual value calculated using the temporary demand prediction function calculated by the temporary demand prediction function plural calculating means in one evaluation period.

  Hereinafter, a processing procedure in the case where the demand prediction of service parts according to the present embodiment is performed will be described with reference to FIG. In the present embodiment, three methods of a cumulative secondary method, a cumulative tertiary method, and a cumulative quaternary method are used as methods for performing demand prediction, as in the first embodiment.

  Also in the present embodiment, the management computer 21 first acquires prediction target specifying data in the same manner as in step S1-1 in the first embodiment (step S2-1). Next, similarly to step S1-2 of the first embodiment, the management computer 21 acquires order result data to be predicted (step S2-2).

  Then, the management computer 21 calculates a temporary demand prediction function for determining the demand prediction method, and calculates an error in each temporary demand prediction function (step S2-3). In this case, the management computer 21 first calculates the temporary demand prediction function of each method using the order results obtained by excluding the most recent one month,... calculate. That is, the most recent evaluation period is changed, and the temporary demand prediction function of each method in the temporary function calculation period having a different length is calculated. Then, the management computer 21 uses the temporary demand prediction function to calculate a predicted value for the month that was not used for calculating the temporary demand prediction function, and calculates an error based on the difference between the predicted value and the order received in the same month. To do. In this case, when there are a plurality of months that are not used in the calculation of the temporary demand prediction function, the sum of the differences between the predicted value and the actual order value in each month is regarded as an error.

This will be described with reference to an example of service parts with an order receipt shown in FIG. As described above, data from the release month to the 45th month is recorded as the order result data 220 for this service part. Furthermore, the case where n = 5 is assumed here. Therefore, in the present embodiment, the management computer 21 calculates a temporary demand prediction function for determining a method by sequentially removing orders received from the latest one month to five months.

  The management computer 21 first calculates the function of each method using the order record data 220 of 44 months excluding the most recent one month. A predicted value for the 45th month is calculated using the calculated function of each method, and a difference between the predicted value and the actual result is calculated as an error.

  Next, the management computer 21 calculates the function of each method using the order record data 220 of 43 months excluding the latest two months. And the predicted value of the 44th month and the 45th month is calculated using the function of each calculated method. The sum of the difference between the predicted value calculated for the 44th month and the actual value for the 44th month and the difference between the calculated predicted value for the 45th month and the actual value for the 45th month is calculated as an error.

  Similarly, the management computer 21 calculates the function of each method using the order record data 220 of 42 months excluding the latest three months, and uses the calculated function of each method for the 43rd month, the 44th month, Calculate the predicted value for the 45th month. And the difference of these estimated values and the actual value of 43 months-45 months is each calculated, and the sum total of these differences is calculated as an error.

  Furthermore, the management computer 21 calculates the function of each method using the 41-month order record data 220 excluding the latest four months, and uses the calculated function of each method to predict the 42nd to 45th months. Is calculated. Then, the difference between these predicted values and the actual values in the 42nd to 45th months is calculated, and the sum of these differences is calculated as an error.

  Further, the management computer 21 calculates the function of each method using the 40-month order record data 220 excluding the latest five months, and uses the calculated function of each method to predict the predicted values for the 41st to 45th months. Is calculated. And the difference of these prediction values and the actual value of the 41st month-the 45th month is calculated, respectively, and the sum total of these differences is calculated as an error.

  Next, the management computer 21 calculates the number of minimum errors in each method (step S2-4). Specifically, in each method, the management computer 21 specifies the smallest error method among the errors calculated except for the latest one to five months, and counts the smallest number of times.

  For example, when the error shown in the table 100 of FIG. 8 is calculated, the management computer 21 specifies the minimum error method when the most recent one month is excluded as the cumulative fourth order. Further, the management computer 21 specifies the minimum error method when the most recent two months are excluded as cumulative fourth order, and specifies the minimum error method when the latest three months are excluded as cumulative secondary. Furthermore, the minimum error method when the latest 4 months and 5 months are excluded is specified as the cumulative secondary. Then, the management computer 21 counts the minimum error count as 3 times for the cumulative secondary method, 0 times for the cumulative tertiary method, and 2 times for the cumulative quaternary method.

  Next, the management computer 21 adopts the method having the largest minimum error (step S2) when there is one method having the largest number of minimum errors (in the case of “YES” in step S2-5). -6) The next step S2-8 is performed. For example, as shown in FIG. 8, when the minimum error calculated by the cumulative secondary method is the largest, the management computer 21 adopts this cumulative secondary method.

On the other hand, when there are two or more methods having the largest number of minimum errors (“NO” in step S2-5), the management computer 21 has the largest number of times of the smallest error and the least number of orders. Is adopted (step S2-7), and the next step S2-8 is performed. For example, as shown in the table 200 of FIG. 9, when the cumulative secondary method has the minimum error twice as in the cumulative quaternary method, and more than the cumulative tertiary method, The cumulative secondary method is adopted as a method having the largest number of minimum errors and the smallest number of orders.

  Next, the management computer 21 calculates a demand prediction function using the adopted method (step S2-8). Specifically, the management computer 21 uses the method adopted in step S2-6 or S2-7, and calculates the demand prediction function using all of the prediction target order result data acquired in step S2-2. For example, as shown in FIGS. 8 and 9, when the cumulative secondary method is adopted, the cumulative secondary method is used, and the 45th month is used as in step S1-8 of the first embodiment. The demand prediction function is calculated using the received order data.

  Then, the management computer 21 calculates a predicted value (step S2-9). Specifically, using the demand prediction function calculated in step S2-8, for example, the predicted value of the number of orders received such as the next month or the next month is calculated. Next, the management computer 21 outputs the result of the calculated prediction value to the order receiving system 10 via the network N as in Step S1-10 of the first embodiment (Step S2-10). Thus, the demand prediction process is completed.

According to the present embodiment, in addition to the effects similar to the effects described in (1) and (2) of the first embodiment, the following effects can be obtained.
(3) In the present embodiment, in the case of method determination, the management computer 21 first differs in length except for the most recent one month,... The temporary demand prediction function of each method in the temporary function calculation period is calculated. And using this temporary demand prediction function, the predicted value of the month which was not used for calculation of the temporary demand prediction function is calculated, and the error based on the difference between this predicted value and the order record of the same month is calculated. In this case, when there are a plurality of months that are not used in the calculation of the temporary demand prediction function, the sum of the differences between the predicted value and the actual order value in each month is regarded as an error. Then, the management computer 21 calculates the number of times of the minimum error in each method (step S2-4), adopts the method having the largest minimum error (step S2-6), and performs the processing of the next step S2-8. Do. When there are two or more methods having the largest number of minimum errors (in the case of “NO” in step S2-5), the management computer 21 has the largest number of minimum errors and the least number of methods. Is adopted (step S2-7), and the next step S2-8 is performed. Therefore, a more appropriate demand prediction function method can be determined from a plurality of verification results with different provisional demand prediction functions even in the same method.

Moreover, you may change the said embodiment as follows.
In each of the above embodiments, when the demand prediction function calculation process is performed (step S1-8), three methods of the cumulative secondary method, the cumulative tertiary method, and the cumulative quaternary method are used. The method for calculating the demand prediction curve is not limited to these. For example, when there is a periodic variation, a method of a demand prediction function that combines a trend curve and a periodic function may be used. In this case, the management computer 21 calculates a trend curve, obtains a periodic variation function, and calculates a temporary demand prediction function based on the trend function and the periodic variation function. Then, a predicted value is calculated from the provisional demand prediction function, and a method to be adopted is determined from the difference between the predicted value and the order record in the evaluation period. Even in this case, a more appropriate demand prediction method can be selected to perform a more appropriate demand prediction.

In the first embodiment, the management computer 21 calculates an error in each temporary demand prediction function, and adopts a method in which the calculated error is the smallest. In the second embodiment, the management computer 21 calculates the temporary demand prediction function of each method in the temporary function calculation periods having different lengths by excluding the received order results in the evaluation period from the acquired order results. Using the function, a predicted value for the evaluation period is calculated, and an error based on the difference between the predicted value and the actual order received in the same month is calculated. Then, the management computer 21 calculates the number of times of the minimum error in each method, and employs the method having the largest minimum error. Not limited to this, if the estimated value for the evaluation period is calculated using the calculated temporary demand prediction function and the method to be adopted is determined based on the difference between the predicted value and the actual value for the evaluation period, It may be anything.

  For example, weighting may be used so that a method with a low order is used as much as possible. Specifically, the memory of the management computer 21 stores data related to the low-order selection value related to the criterion for adopting the low-order method. For example, in the first embodiment, when there is a method with the highest order than the method with the smallest error, the error of the method with the smallest error and the error of the method with the lesser order are lower. Compare the value with the selected value. As a result of the comparison, the management computer 21 adopts the method having the smallest order when the value of the method having the smallest order is equal to or smaller than the error of the smallest method, and adopts the method having the smallest error in the other cases. As a result, it is possible to adopt a method with a low order as much as possible.

  In addition, for example, in the second embodiment, the error between the predicted value and the actual value calculated by the demand prediction function calculated using the order results excluding the most recent one month, ..., the most recent n months, respectively. Further, a method in which the total value is minimized may be used.

  In the second embodiment, the management computer 21 calculates a temporary demand prediction function for determining a method by sequentially removing orders received from the latest one month to five months. Then, the management computer 21 calculates the number of times of the minimum error in each method (step S2-4), and adopts the method having the largest number of times of the minimum error (steps S2-6 and S2-7). Instead of this, the management computer 21 may calculate the error of each method from the latest evaluation period, and may adopt a method in which the number of times of the minimum error reaches the predetermined number earliest. Specifically, the management computer 21 stores data related to the number of determinations for determining the demand prediction method. The management computer 21 first specifies the minimum error method from the temporary demand prediction function in the temporary function calculation period excluding the shortest evaluation period, and records the minimum error count for this method. Then, the minimum error count is compared with the determination count. If the minimum error count has not reached the determination count, the management computer 21 sets the next longer evaluation period. Then, the management computer 21 specifies the minimum error method from the temporary demand prediction function in the evaluation period, records the minimum error number for this method, and compares the minimum error number with the determined number. In this way, the minimum error count is counted by sequentially changing to a long evaluation period until the minimum error count reaches the determination count. Then, the demand prediction method in which the minimum error count reaches the determination count earliest is used as a method for actually forecasting demand. As a result, it is possible to identify a method with good fitting while utilizing the latest evaluation result earlier.

In the above embodiment, the service part order record was recorded monthly. Not limited to this, for example, orders received every several months, every week, or every several weeks may be used.
In the above embodiment, the demand for service parts is predicted. The target of the demand forecast is not limited to this, and may be a demand forecast for other products as long as it is a demand forecast for a product to which a plurality of demand prediction methods can be applied.

1 is a schematic diagram of a demand prediction system in an embodiment. Explanatory drawing explaining the data structure of the data memorize | stored in the order performance data storage part. The figure of the order receipt result and the order month which shows the specific example of an order receipt record. The flowchart for demonstrating the process sequence of a demand prediction process. It is a figure which shows an order record and a tendency curve, (A) is a cumulative secondary system, (B) is a cumulative tertiary system, (C) shows the tendency curve by a cumulative quaternary system. Explanatory drawing for determining the system used for demand prediction in 1st Embodiment. The flowchart for demonstrating the process sequence of a demand prediction process. Explanatory drawing for determining the system used for demand prediction in 2nd Embodiment. Explanatory drawing for determining the system used for demand prediction in 2nd Embodiment.

Explanation of symbols

  20 ... Demand prediction system, 21 ... Management computer, 22 ... Order record data storage unit as record data storage means.

Claims (5)

A method for predicting demand for a product using a management computer connected to a product data storage means that records a product identifier for specifying a product and data related to a product actual value corresponding to a time axis. ,
The management computer is
Obtaining prediction target specifying data including a product identifier for specifying a product for which demand prediction is performed, and a result extraction stage for extracting a result value of the product of the product identifier from the result data storage means using the product identifier;
Of the extracted actual values, use the actual values in the temporary function calculation period excluding the most recent predetermined evaluation period, perform temporary demand prediction function calculation processing to calculate temporary demand prediction function according to multiple methods, and calculate Calculating a predicted value of the evaluation period using the provisional demand prediction function, and determining a method to be employed based on a difference between the predicted value and the actual value of the evaluation period;
A predicted value calculation step of calculating a predicted value of a product for which demand prediction is performed using a result value extracted from the result data storage means using the determined method;
A demand prediction method characterized by executing a demand forecast output stage for outputting a calculated forecast value. An evaluation period is set so that a plurality of actual values are included in the evaluation period,
The method determining step includes:
An error calculating step of calculating an error obtained by summing a difference between the predicted value calculated using the provisional demand prediction function and the actual value in the evaluation period;
The demand prediction method according to claim 1, further comprising a determination step of determining a method that satisfies a condition that the error is minimum. The method determining step includes:
A provisional demand prediction function multiple calculation stage for executing the temporary demand prediction function calculation processing in a plurality of temporary function calculation periods having different lengths;
A minimum error method specifying step for specifying a method in which the error between the predicted value calculated using the temporary demand prediction function calculated by the temporary demand prediction function multiple calculation step and the actual value is the smallest in the evaluation period of 1;
The demand prediction method according to claim 1, further comprising: a determination step of determining a method that satisfies a condition that the number of times that the error is specified to be the smallest is the largest in the plurality of temporary function calculation periods. .   4. The determination step according to claim 2, wherein, when the management computer specifies a plurality of methods that satisfy the condition, the determination method determines a method having a lower order among the specified methods. 5. Demand forecast method. A program for predicting demand for a product using a management computer connected to a record data storage means that records a product identifier for specifying a product and data related to the actual value of the product corresponding to a time axis. ,
The management computer,
A record extracting unit that acquires prediction target specifying data including a product identifier for specifying a product for which demand prediction is performed, and extracts a track record value of the product of the product identifier from the track record data storage unit using the product identifier;
Of the extracted actual values, use the actual values in the temporary function calculation period excluding the most recent predetermined evaluation period, perform temporary demand prediction function calculation processing to calculate temporary demand prediction function according to multiple methods, and calculate Calculating a predicted value of the evaluation period using the provisional demand prediction function, and determining a method to be adopted based on a difference between the predicted value and the actual value of the evaluation period;
As the predicted value calculation means for calculating the predicted value of the product for which the demand is predicted using the actual value extracted from the actual data storage means using the determined method, and the demand prediction output means for outputting the calculated predicted value A demand forecasting program characterized by functioning.

Images