Component demand prediction device

Abstract

The purpose of the present invention is to provide technology whereby demand can be accurately predicted even for a component having a small number of received order history data. A component demand prediction method according to the present invention separates an order reception log into order reception, usage purpose data and usage purpose, and component data, and predicts an interval between orders received for a component item for each usage purpose on the basis of the order reception and the usage purpose data (see fig. 5).

Description

Parts demand forecasting device

[0001] The present invention relates to a technology for predicting demand for parts.

[0002] For companies that manufacture and sell machinery, including construction machinery, it is important to accurately forecast the demand or order volume for parts used in repairing machinery. High forecast accuracy allows them to reduce safety stock and thereby reduce parts inventory costs.

[0003] The following Patent Document 1 describes a technology for segmenting customers. Specifically, it describes a technology for segmenting parts into seasonal products, semi-seasonal products, non-seasonal products, and unexpected order products, and forecasting demand using different methods for each segment.

[0004] US2008 / 0082386A1

[0005] To accurately forecast parts demand, a certain amount of order history data is required. However, for intermittent demand items (e.g., parts with annual shipments of 100 or less), the amount of historical data is small, making demand forecasting difficult. The amount of data becomes even smaller when segmenting customers or looking at changes in demand trends.

[0006] The present invention has been made in consideration of the above-mentioned problems, and aims to provide a technology that can accurately predict demand even for parts with a small number of order history data items.

[0007] The parts demand forecasting method according to the present invention breaks down an order log into order-purpose data and purpose-purpose data, and forecasts the order interval for parts items for each purpose of use based on the order-purpose data.

[0008] According to the parts demand forecasting method of the present invention, it is possible to accurately forecast demand even for parts with a small number of order history data items. Other problems, configurations, and effects will become clear from the description of the following embodiments.

[0009] 1 is a schematic diagram illustrating the basic concept of a parts demand forecasting method according to a first embodiment. FIG. 2 is an example of visually presenting forecast results according to the first embodiment. FIG. 3 is a configuration diagram of a parts demand forecasting device 100 according to the first embodiment. FIG. 4 is a flowchart illustrating a conventional parts demand forecasting method. FIG. 5 is an overall flow diagram illustrating a method in which a parts order model building unit F00 predicts order intervals for each part. FIG. 6 shows an example of the configuration and data of a parts order log D01. FIG. 7 is an example of an order x parts matrix D02. FIG. 8 is an example of an order x parts matrix D03. FIG. 9 is a flow diagram illustrating processing procedures by a matrix decomposition unit F01. FIG. 10 is a flow diagram illustrating processing procedures by a repair log extraction unit F02. FIG. 11 is a flow diagram illustrating processing procedures by an order interval model building unit F03. FIG. 11 is a flow diagram illustrating processing procedures by a parts order model building unit F04. FIG. 12 is a flow diagram illustrating processing procedures by a customer segment extraction unit F05. FIG. 13 is an overall flow diagram illustrating another method in which the parts order model building unit F00 predicts order intervals for each part. FIG. 14 is an example of a user interface provided by the parts demand forecasting device 100. FIG. 15 is an example of a user interface provided by the parts demand forecasting device 100.

[0010] <First Embodiment: Basic Concept> Figure 1 is a schematic diagram illustrating the basic concept of a parts demand forecasting method according to a first embodiment of the present invention. In this example, assume that multiple customers (customer 1 to customer 4 in Figure 1) each place orders for various parts (ordered parts 1 to 4 in Figure 1), and that a parts supplier receives these orders. The parts supplier wishes to accurately perform demand forecasting (synonymous with order forecasting or order forecasting) for each part. This embodiment relates to such a demand forecasting method.

[0011] It is generally difficult to accurately predict which parts each customer will order. Therefore, in this embodiment, the prediction process is divided into two stages. As a prerequisite for prediction, it is assumed that order history data for each part has been acquired in advance for each customer. White ellipses in Figure 1 indicate that history data exists. Black ellipses indicate data derived by prediction when history data does not exist.

[0012] It is assumed that a customer orders parts when repairs to machinery are to be performed using those parts. For example, when repairing a hydraulic excavator, a type of construction machinery, it is assumed that the customer orders parts to perform multiple repairs at once, such as an "understructure replacement" and an "oil change." Therefore, in this embodiment, the interval between repairs (e.g., the interval between understructure repairs) is predicted, rather than the interval between parts orders. This increases the amount of data that can be used for prediction. Furthermore, customers are grouped (209, 210) by repair item (e.g., customer 209 who tends to perform understructure repairs). However, since there is no record of the type of repairs performed by the customer, the repair items are inferred based on the order record data.

[0013] In the example shown in Figure 1, two repair items are assumed: Repair 1 (211) "undercarriage replacement" and Repair 2 (212) "oil change." Customers who tend to perform each repair item are grouped (209, 210). This allows customers to be classified into two customer groups 209, 210. Furthermore, for each repair item, the probability of parts being ordered for that repair item is predicted. In other words, first the probability that a customer will perform the repair item is predicted, and then the probability of parts being ordered for that repair item is predicted. Through this two-stage prediction, the probability that a customer will order parts (and the order interval) is predicted.

[0014] The above prediction process is believed to enable accurate prediction of order intervals even for intermittent demand parts, which have a small amount of order history data. This is because predicting repair items as new data items increases the amount of data. Furthermore, by classifying customers by repair purpose, it is possible to estimate customer characteristics (e.g., those who perform maintenance diligently, those who use specialized construction machinery, etc.), which can then be used to create advice for promoting parts sales.

[0015] 2 is an example of visually presenting the prediction results according to this embodiment. The horizontal axis indicates the order date, and the vertical axis indicates the number of parts ordered. The parts ordered on each order date and their quantities are presented, along with the repair items that use each part. In this way, it is possible to predict the parts ordered for each repair item and present the results.

[0016] <Embodiment 1: System Configuration> Fig. 3 is a configuration diagram of a part demand forecasting device 100 according to embodiment 1. The part demand forecasting device 100 is a device that implements the part demand forecasting method exemplified in Fig. 1. The part demand forecasting device 100 includes a storage device 101, a CPU (Central Processing Unit) 102, a communication unit 103, a memory 104, a display 105, and a keyboard 106.

[0017] The storage device 101 is a storage device that stores parts inventory data 107 and a parts order log D01. These data will be described later. The communication unit 103 communicates with other devices. The display 105 is a device that displays the processing results of the CPU 104, and can present a screen such as that shown in FIG. 2, for example. The keyboard 106 is an interface through which the user inputs instructions to the parts demand prediction device 100.

[0018] The memory 104 is a storage device that stores data used by the CPU 102. The memory 104 stores software modules executed by the CPU 102, including a parts order model construction unit F00, a sales promotion scenario recommendation unit 110, a parts order forecasting unit 108, a safety stock estimation unit 111, and a parts order quantity determination unit 112.

[0019] The parts order model construction unit F00 includes sub-modules, which will be described later. The parts order model construction unit F00 uses these sub-modules and the parts order log D01 to construct a model used to estimate the order intervals of parts. The construction method is outlined in FIGS. 1 and 2. The specific method will be described later.

[0020] The sales promotion scenario recommendation unit 110 uses the order interval model constructed by the parts order model construction unit F00 and the customer classification results to construct advice for parts sales promotion. The order interval model can be a model that describes the order interval for each repair item, or a model that describes the order interval for each part item. Either model can be used.

[0021] The parts order forecasting unit 108 predicts the order interval for each part using the parts order interval model constructed by the parts order model construction unit F00. The safety stock estimation unit 111 estimates the safety stock for each part based on the predicted parts order interval. The parts order quantity determination unit 112 determines the order quantity for each part based on the estimated safety stock and the parts inventory data 107.

[0022] <First Embodiment: Conventional Forecasting Method> Figure 4 is a flowchart illustrating a conventional part demand forecasting method. Conventionally, a part order log D01 is used to construct a part order interval model D08 that mathematically describes the order interval for each part (F3). This model is then used to predict the order quantity for each part (F06). The prediction results are stored in storage device 101 as part order forecast results D09.

[0023] 4 predicts the order interval for each part directly from the parts order log D01. Therefore, it is generally difficult to accurately predict the order interval for intermittently demanded parts, for which there is little order history data. Therefore, the parts demand prediction device 100 attempts to predict part demand using the following method.

[0024] <Embodiment 1: Details of Parts Demand Forecasting Method> Figure 5 is an overall flow diagram illustrating a method by which the parts order model construction unit F00 predicts the order interval for each part. F01 to F06 in Figure 5 are sub-modules provided in the parts order model construction unit F00. The method by which the parts order model construction unit F00 predicts the order interval will be described below with reference to Figure 5.

[0025] The matrix decomposition unit F01 decomposes the parts order log D01 into an order x repair matrix D04 (order - intended use data) and a repair x parts matrix D05 (intended use - parts data). A specific decomposition procedure will be described later with reference to FIG.

[0026] The repair log extraction unit F02 extracts a repair log D06 from the parts order log D01 and the order x repair matrix D04. The contents of the repair log D06 and the extraction method will be described later using FIG. 10. The order interval model construction unit F03 uses the repair log D06 to construct a repair order interval model D07 that mathematically represents the order interval for each repair item. The contents of the repair order interval model D07 and the construction method will be described later using FIG. 11.

[0027] The parts order model construction unit F04 constructs a parts order interval model D08 that mathematically represents the order interval for each part, using the repair order interval model D07 and the repair x part matrix D05. The contents and construction method of the parts order interval model D08 will be described later with reference to FIG.

[0028] The customer segment extraction unit F05 classifies customers using the order x repair matrix D04. The classification results are output as customer segment data D10. A specific classification method will be described later with reference to FIG. 13.

[0029] The prediction unit F06 predicts the order interval for each part using the part order interval model D08, and outputs the result as part order forecast result D09. Since the prediction unit F06 is substantially the same as the part order forecast unit 108, it can also be configured integrally with the part order forecast unit 108.

[0030] Figure 6 shows the structure and example data of the parts order log D01. The parts order log D01 is data that records the number of parts ordered and the orderer (customer) for each order. Multiple orders may be received on the same date, and even orders for multiple parts may be received within a single order. The order ID is an ID that identifies a single order.

[0031] 7 is an example of the order x parts matrix D02. The order x parts matrix D02 (order-parts data) is intermediate data created during processing by the matrix decomposition unit F01. The order x parts matrix D02 is data describing the number of parts ordered for each order described in the parts order log D01, and describes the number of parts for each combination of order ID and part ID. The order x parts matrix D02 can be extracted from the parts order log D01.

[0032] FIG. 8 is an example of the order x part matrix D03. The order x part matrix D03 (order-part data) is intermediate data created in the processing by the matrix decomposition unit F01. The order x part matrix D03 is data obtained by normalizing the order x part matrix D02. The order x part matrix D03 can be decomposed into an order x repair matrix D04 and a repair x part matrix D05 by matrix decomposition. In other words, the order x part matrix D03 is roughly equivalent to the product of the order x repair matrix D04 and the repair x part matrix D05. This point will be explained again later.

[0033] 9 is a flow diagram illustrating the processing procedure performed by the matrix decomposition unit F01. The matrix decomposition unit F01 generates an order x part matrix D02 by generating a histogram of the parts order log D01 (step 901). The order x part matrix D02 is the data described in FIG. 7 and describes the number of parts in each order. The matrix decomposition unit F01 normalizes the order x part matrix D02 to generate an order x part matrix D03 (step 902). The matrix decomposition unit F01 decomposes the order x part matrix D03 into an order x repair matrix D04 and a repair x part matrix D05 (step 903).

[0034] Focus on a certain part described in the orders x parts matrix D02. It is possible to calculate the ratio of the number of parts ordered in each order to the number of parts used in all orders. This ratio is called part importance. By replacing the number of parts with part importance, the orders x parts matrix D02 can be normalized. In step 902, the orders x parts matrix D03 is generated using the above procedure.

[0035] Step 903 can decompose the order x part matrix D03 using a technique such as NMF matrix decomposition (Non-negative Matrix Factorization). NMF matrix decomposition is a technique for decomposing a matrix before decomposition into products of basis vectors and contribution rate vectors. The repair x part matrix D05 corresponds to the basis vectors, and describes the part importance for each combination of repair item and part item. The order x repair matrix D04 corresponds to the contribution rate vector, and describes the repair contribution rate for each combination of order ID and repair item.

[0036] Since the number of vector dimensions is required as a prerequisite for NMF matrix decomposition, the number of repair items must be input. The number of repair items is given so that the product of the order × repair matrix D04 and the repair × part matrix D05 is as equivalent as possible to the order × part matrix D03. For example, the calculation may be repeated while changing the number of repair items until a decomposition result that is most equivalent to the order × part matrix D03 is obtained.

[0037] The matrix decomposition performed in step 903 is performed mechanically so that the product of the order × repair matrix D04 and the repair × part matrix D05 is as equivalent as possible to the order × part matrix D03 (the error between the two is minimized). Therefore, it should be noted that the repair items obtained in step 903 do not necessarily have practical meaning. However, by segmenting customers using repair IDs, it is possible to understand the ordering trends of each customer. Furthermore, since the part IDs belonging to the repair IDs can be obtained from the repair × part matrix D05, it is possible to assign a tentative meaning to the repair IDs using representative part IDs. Therefore, it is possible to plan sales promotion activities using this information.

[0038] 10 is a flow diagram illustrating the processing procedure performed by the repair log extraction unit F02. The repair log extraction unit F02 extracts a repair log D06 from the parts order log D01 and the order x repair matrix D04. The parts order log D01 describes when a part was ordered. In contrast, the repair log D06 describes when a repair was ordered. In other words, the repair log extraction unit F02 converts the part order records described in the parts order log D01 into repair order records.

[0039] The repair log extraction unit F02 merges the parts order log D01 with the order x repair matrix D04 (step 1001). Specifically, for each part ID described in the parts order log D01, the values ​​for each repair ID described in the order x repair matrix D04 are combined horizontally. As a result of the merging, a matrix like the second from the top in the left column of Figure 10 is generated.

[0040] The repair log extraction unit F02 binarizes the matrix generated in step 1001 according to a threshold value (step 1003). Specifically, of the values ​​introduced from the order x repair matrix D04, those that are equal to or greater than a threshold value (e.g., 0.5) are replaced with 1, and those that are less than the threshold value are replaced with 0. The matrix generated by this process is referred to as the order / repair correspondence table D11 (order-purpose correspondence table). The order / repair correspondence table D11 is a data table that describes ordered repair items by order ID.

[0041] The repair log extraction unit F02 converts (reformats) the order / repair correspondence table D11 as necessary (step 1005). For example, if the order / repair correspondence table D11 describes multiple repair items in one row, the data format is reorganized so that each row describes only one repair item. Furthermore, information regarding the order ID (i.e., each order) is deleted. In the example of FIG. 10, the result of this formatting is output as the repair log D06 (purpose order data). Therefore, the repair log D06 is data described by a combination of the order date and the repair item (repair ID), and therefore describes when the repair order was placed.

[0042] 11 is a flow diagram illustrating the processing procedure of the order interval model construction unit F03. The order interval model construction unit F03 calculates the order interval (first order interval) for each repair item based on the repair log D06 (step 1101). The calculation result is stored in the storage device 101 as repair order interval data 1102 (step 1102).

[0043] The order interval model construction unit F03 creates a histogram of the repair order interval data 1102 (step 1103) to create a repair order interval model D07 (step 1104). The repair order interval model D07 is a model that mathematically describes the order interval for each repair item by approximating the histogram of the repair order interval data 1102 by, for example, a gamma distribution. By creating the repair order interval model D07, it is possible to calculate the average order interval for each repair ID, the standard deviation of the order interval, and the like.

[0044] The order interval model construction unit F03 can predict the order interval for each repair item (repair ID) using the repair order interval model D07 (step 1105). The predicted order interval for each repair item can be used as one of the outputs from the parts demand forecasting device 100. For example, this output can be used to strengthen sales promotion activities for a repair ID if the order interval is gradually increasing.

[0045] 12 is a flow diagram illustrating the processing procedure of the parts order model construction unit F04. The parts order model construction unit F04 creates a parts order interval model D08 from the repair order interval model D07, the parts order probability for each repair item, and the parts order log D01. The parts order interval model D08 is a model that mathematically describes the order interval (second order interval) for each part item.

[0046] The parts order model construction unit F04 merges the repair x parts matrix D05, the order / repair correspondence table D11, and the parts order log D01 (step 1201). The order / repair correspondence table D11 describes which repair item (repair ID) has been ordered for each order ID. By merging the repair x parts matrix D05 and the parts order log D01 with the order / repair correspondence table D11, the number of orders for each part item can be obtained for each order ID. The obtained results are set as data 1202.

[0047] The parts order model construction unit F04 aggregates the data 1202 by part ID (part item) (step 1203). Specifically, it calculates the total number of orders for each part ID, and also calculates the number of orders for that part for each repair item. The sum of the part orders for each repair item is equal to the total number of part orders. The obtained results are set as data 1204.

[0048] The parts order model construction unit F04 calculates the parts order probability for each repair item (step 1205). Specifically, the order probability is calculated using the total number of parts orders as the denominator and the number of parts ordered for each repair item as the numerator. The parts order model construction unit F04 calculates a similar order probability for each part item. The calculation result is set as parts order probability data 1206.

[0049] The parts order model construction unit F04 generates a parts order interval model D08 using the parts order probability data 1206 and the repair order interval model D07. The repair order interval model D07 describes a mathematical model of the order interval and order quantity for each repair item (repair ID). By multiplying the repair order interval model D07 for a certain repair ID by the record of the parts order probability data 1206 that corresponds to that repair ID, a mathematical model of the interval and order quantity for each part ID can be obtained.

[0050] In the example of Fig. 12, for example, by multiplying the repair order interval model D07 for repair ID = 1 by the record for repair ID = 1 in the part order probability data 1206, it is possible to obtain mathematical models for the order interval and order quantity for each of part ID = 1, part ID = 2, and part ID = 3. Fig. 12 shows only the model for part ID = 1.

[0051] 13 is a flow diagram explaining the processing procedure by the customer segment extraction unit F05. The customer segment extraction unit F05 calculates a repair contribution rate vector for each customer by statistically processing the repair contribution rate vector, and classifies customers based on this. Since the repair contribution rate vector represents the contribution rate of a repair item to a base vector, it is thought that by statistically processing the repair contribution rate vector, customers can be classified according to the repair item. Customer classification by the customer segment extraction unit F05 is based on this idea.

[0052] The customer segment extraction unit F05 merges the parts order log D01 with the order x repair matrix D04 (step 1301). The order x repair matrix D04 describes the repair contribution rate vector for each order ID. By merging the parts order log D01 with the order x repair matrix D04, it is possible to associate the repair contribution rate vector with the customer ID. The merged result is data 1302.

[0053] The customer segment extraction unit F05 aggregates the repair contribution rate vector portion of the data 1302 for each customer ID and calculates the center of gravity (step 1303). The aggregated result is set as data 1304. Data 1304 is data in which multidimensional vectors of repair items are described for each customer ID. The numerical values ​​of the repair contribution rate vector are normalized so that they have the same scale for each customer ID. For example, when aggregating, they are normalized so that the maximum value is 1.0.

[0054] The customer segment extraction unit F05 clusters the data 1304 in a multi-dimensional vector space for each repair item (repair ID) (step 1305). Customer groups with similar repair contribution rate vector trends are located close to each other in the vector space. By clustering customers with close vector distances as one cluster, customers can be classified by repair item. In other words, customers who tend to order the same repair items can be considered to belong to the same customer segment. The classification results are stored in the storage device 101 as customer segment data D10.

[0055] <Summary of First Embodiment> The parts demand prediction device 100 according to this embodiment decomposes the parts order log D01 into an order x repair matrix D04 and a repair x parts matrix D05, and creates a repair order interval model D07 using the order x repair matrix D04. The repair order interval model D07 describes the order interval for each repair item. Even for intermittently demanded parts with a small number of order history data items, the number of data items can be increased by estimating the repair items. Furthermore, by predicting the order interval for each repair interval using the increased number of data items, prediction accuracy can be improved.

[0056] <Embodiment 2> In the first embodiment, it has been described that the order x repair matrix D04 and the repair x part matrix D05 are each generated once when parts demand forecasting is performed. However, these matrices change at different rates over time. Specifically, the repair x part matrix D05 changes gradually over time due to part generation changes, whereas the order x repair matrix D04 changes relatively rapidly over time depending on the customer's economic situation. Therefore, in the second embodiment of the present invention, a processing example will be described in which the generation frequency of these matrices is adjusted for each matrix. The other configurations are the same as those of the first embodiment.

[0057] 14 is an overall flow diagram illustrating another method by which the parts order model construction unit F00 predicts the order interval for each part. In this embodiment, the matrix decomposition unit F01 generates the repair x part matrix D05 only once, and generates the order x repair matrix D04 multiple times. That is, multiple order x repair matrices D04 are generated for one repair x part matrix D05. For convenience, a subnumber indicating the number of times the matrix has been generated is assigned, and the matrix is ​​represented as repair x part matrix D05_1, order x repair matrix D04_2, and so on.

[0058] The matrix decomposition unit F01 decomposes the parts order log D01 into an order × repair matrix D04_1 and a repair × part matrix D05_1, as in embodiment 1. The parts order model construction unit F04 constructs a parts order interval model D08, as in embodiment 1, using the repair × part matrix D05_1 and the repair order interval model D07.

[0059] The order x repair matrix D04 changes relatively rapidly depending on the customer's economic situation. Therefore, the parts order model construction unit F00 filters the parts order log D01 for each customer and for each period (step 1401). A partial order log 1402 is generated by filtering. The matrix decomposition unit F01 performs the same matrix decomposition as in embodiment 1 on the partial order log 1402. That is, the partial order log 1402 for a specific customer for a specific period is divided into a repair x parts matrix D05_2 and an order x repair matrix D04_2. The repair log extraction unit F02, the order interval model construction unit F03, and the parts order model construction unit F04 use the order x repair matrix D04_2 to perform the same processing as in embodiment 1.

[0060] The above processing makes it possible to construct a parts order interval model D08 that focuses on a specific customer and a specific period. By performing similar processing for each customer and for each period, it is possible to construct a parts order interval model D08 for each combination of customer and period. When a factor that causes a change in the repair x part matrix D05 occurs, such as when a part model change occurs, the entire process of FIG. 14 (or the processing for generating the repair x part matrix D05) may be performed again.

[0061] <Third Embodiment> Fig. 15 is an example of a user interface provided by the parts demand prediction device 100. For example, the parts order model construction unit F00 generates the screen interface of Fig. 15 and displays it on the display 105, thereby providing these user interfaces. The same is true for the example of Fig. 16. Other configurations are the same as those of the first and second embodiments.

[0062] Screen 1501 is an example of displaying order details for a certain customer by order date and order ID. The horizontal axis represents the order date, and the vertical axis represents the order volume for each order ID. Screen 1501 allows users to visually grasp whether an order has been placed and the order volume.

[0063] Screen 1502 is an example of displaying order details for a certain customer by order date, sorted by repair ID. The horizontal axis represents the order date, and the vertical axis represents the order volume for each repair ID. Screen 1502 allows the user to understand the types of orders placed by the customer by repair item (repair ID). Order types include, for example, repairs related to routine inspections and repairs related to the undercarriage. Furthermore, the user can visually understand trends in order types.

[0064] Screen 1503 is an example of displaying the order details for a certain customer by part item (part ID) for each order date. The horizontal axis represents the order date, and the vertical axis represents the order quantity for each part ID. Screen 1503 allows the customer's order details to be visually grasped for each part item (part ID).

[0065] FIG. 16 is an example of a user interface provided by the parts demand prediction device 100. FIG. 16 shows changes in ordering trends for each customer (or each customer group) as changes in orders for repair items over time. The horizontal axis of FIG. 16 represents date and time, and the vertical axis represents repair items for each customer / customer group. The density of the pattern represents the ordering interval. Areas with high density represent short ordering intervals (i.e., many orders), while areas with low density represent long ordering intervals (i.e., few orders). For example, in the ordering trend for repair 1 of customer group 1, the ordering interval in period 1601 is shorter than that in period 1602. Furthermore, the ordering intervals in periods 1603 and 1604 for repair 2 are both longer than that in period 1601.

[0066] By presenting the changes over time for each repair item on the user interface, as shown in Figure 16, it is possible to visually grasp changes in ordering trends for each customer. Customer Group 1 started doing business with us for the first time in fiscal year 2022, and we can see that they have received many orders for Repair 1. Customer Group 2 mainly received Repair 1 in fiscal year 2021, but from fiscal year 2022, Repair 1 disappeared and they began receiving orders for Repair 2 and Repair 3. By visually presenting the changes over time in ordering trends for each repair item in this way, it is possible to effectively plan sales promotion measures. For example, for Customer Group 2, it is possible to aim to increase orders for Repair 1 while ensuring stable orders for Repair 2 and Repair 3.

[0067] <Modifications of the present invention> The present invention is not limited to the above-described embodiment, and includes various modifications. For example, it is possible to replace part of the configuration of one embodiment with the configuration of another embodiment, or to add the configuration of another embodiment to the configuration of one embodiment, without departing from the spirit of the invention. Furthermore, it is possible to add, delete, or replace part of the configuration of each embodiment with another configuration.

[0068] In the above embodiment, any two or more of the parts order model construction unit F00, sales promotion scenario recommendation unit 110, parts order forecasting unit 108, safety stock estimation unit 111, parts order quantity determination unit 112, and parts order quantity determination unit 109 may be constructed as a single module. For example, the parts order model construction unit F00 and the parts order forecasting unit 108 may be constructed as a single forecasting unit.

[0069] In the above embodiment, it is assumed that parts are used for repair. However, the present invention can also be applied to cases where parts are ordered for purposes other than repair. In this case, the "repair" described in the above embodiment can be replaced with "purpose of use." Therefore, the parts demand prediction device 100 predicts part demand for each purpose of use.

[0070] 100: Parts demand forecasting device F00: Parts order model construction unit 108: Parts order forecasting unit 110: Sales promotion scenario recommendation unit 111: Safety stock estimation unit 112: Parts order quantity determination unit

Claims

1. A parts demand forecasting device that forecasts demand for parts, comprising an order interval forecast value calculation unit that calculates a forecast value of the order interval for the parts, wherein the order interval forecast value calculation unit obtains, from a storage device, order-ordered parts data that describes the relationship between orders for parts and the parts items ordered in the orders, creates order-order-purpose data from the order-ordered parts data that describes the relationship between the orders and the intended uses of the ordered parts items, and calculates a first order interval that represents the order interval for the parts items for each intended use based on the order-purpose data, thereby forecasting the demand for the parts.

2. The parts demand forecasting device according to claim 1, wherein said order interval forecast value calculation unit creates said order-order purpose data by decomposing said order-ordered parts data into purpose-of-use-parts data describing the relationship between said purpose of use and said part item, and said order-order purpose data, and when decomposing said order-ordered parts data, said order interval forecast value calculation unit converts said order-ordered parts data into the product of a basis vector and a contribution rate vector, and adopts said basis vector as said purpose-of-use-parts data and adopts said contribution rate vector as said order-order purpose data.

3. The parts demand forecasting device according to claim 2, wherein the order-ordered parts data describes the number of parts ordered in each order for each combination of the order and the part item, and when breaking down the order-ordered parts data, the order interval prediction value calculation unit converts the number into part importance which indicates the ratio of the individual number of parts in each order to the total number of parts in all of the orders, and when converting into the product of the basis vector and the contribution rate vector, the order interval prediction value calculation unit converts the order-ordered parts data after conversion into part importance into the product of the basis vector and the contribution rate vector.

4. The parts demand forecasting device according to claim 1, characterized in that the order interval prediction value calculation unit further converts the order-ordered parts data into intended-use order data that describes order records for the intended uses by integrating the order-ordered parts data and the order-intended-use data, and when calculating the first order interval, the order interval prediction value calculation unit calculates the order interval for each intended use using the intended-use order data, thereby calculating the first order interval.

5. The parts demand forecasting device according to claim 4, wherein, when converting the order-order-ordered parts data into the intended-use order data, the order interval forecast value calculation unit converts the integrated data into an order-order-intended-use correspondence table that indicates the correspondence between the orders and the intended uses, depending on whether a value inherited from the order-intended-use data contained in integrated data obtained by integrating the order-ordered parts data and the order-intended-use data is equal to or greater than a threshold value, and generates the intended-use order data by shaping the order-intended-use correspondence table into a record for each order.

6. The parts demand forecasting device according to claim 4, characterized in that, when calculating the first order interval, the order interval prediction value calculation unit calculates the order interval for each intended use from the order records described in the intended use order data, statistically processes the order interval for each intended use to construct a first mathematical model that mathematically describes the order interval for each intended use, and calculates the first order interval using the first mathematical model.

7. The parts demand forecasting device according to claim 2, characterized in that the order interval prediction value calculation unit further constructs a first mathematical model that mathematically describes the order interval for each of the intended uses, calculates the order probability of the part item for each of the intended uses, constructs a second mathematical model that represents the order interval for each of the part items using the order probability of the part item for each of the intended uses and the first mathematical model, and calculates a second order interval that represents the order interval for each of the part items using the second mathematical model.

8. The parts demand forecasting device according to claim 7, wherein, when calculating the order probability of the part item for each of the intended uses, the order interval prediction value calculation unit uses the order-intended use data to create an order-intended use correspondence table that describes, for each combination of the order and the intended use, whether the order has been placed for the intended use, integrates the order-intended use correspondence table with the intended use-parts data, tabulates the results of integrating the order-intended use correspondence table with the intended use-parts data for each of the part items, and calculates the order probability of the part item for each of the intended uses based on the ratio of the number of orders for the part item for each of the intended uses to the total number of the part items.

9. The parts demand forecasting device according to claim 1, characterized in that the order-ordered parts data describes the orders and the part items for each customer, and the order interval forecast value calculation unit further classifies the customers described in the order-ordered parts data by the intended use by statistically processing the customers by the intended use.

10. The parts demand forecasting device according to claim 9, characterized in that, when classifying the customers, the order interval prediction value calculation unit integrates the order-ordered parts data and the order-purpose use data, statistically processes the results of integrating the order-ordered parts data and the order-purpose use data for each customer, clusters the customers within a vector space of the purposes of use based on the results of the statistical processing, and classifies the customers based on the results of the clustering.

11. The parts demand forecasting device according to claim 7, characterized in that: the order-ordered parts data describes the orders and the part items for each customer; the order interval prediction value calculation unit further generates the order-purpose data for at least either a period or for each customer; and when calculating the second order interval, the order interval prediction value calculation unit calculates the second order interval for at least either a period or for each customer using the order-purpose data generated for at least a period or for each customer.

12. The parts demand forecasting device according to claim 11, characterized in that, when calculating the second order interval, the order interval prediction value calculation unit calculates the second order interval using a single piece of intended use-parts data together with the order-intended use data for each period or for each customer.

13. The parts demand forecasting device according to claim 1, characterized in that the order-ordered parts data describes the orders and the part items for each customer, and the order interval forecast value calculation unit further presents the order history for each purpose of use for each customer.

14. The parts demand forecasting device according to claim 1, wherein said order interval forecast value calculation unit further presents changes over time in the order history for each of said intended uses.

Images