Parts model prediction device

Abstract

To provide a component type prediction device and method for predicting a type with high accuracy by selecting a specification having a high correlation to the type from a large amount of specification information after selecting a long term delivery component having a risk of a missing component and an excessive stock.SOLUTION: A component type prediction device includes a storage part for storing receiving / shipping / stock information recording the number of past receiving / shipping / stock results of components constituting a product, and past item information recording specifications and types of past items, an object component selection part for selecting a component of a prediction object from the receiving / shipping / stock information, a specification grouping part for dividing specifications to be used into a plurality of groups from the past item information, a correlation degree measurement part for extracting a specification having a high use frequency from specification group information to calculate a correlation degree between the specification and a type, a specification selection part for combining specifications having a high correlation degree for each group from the correlation degree information and prediction object item information to select a specification that becomes necessary for prediction, and a type prediction part for predicting a type of the object from specification selection result information.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention is a system for managing parts procurement for products. Part type prediction device Regarding. [Background technology]

[0002] For example, in the elevator business, production begins after receiving an order in response to customer requests, but in order to meet delivery deadlines, it is necessary to stock up on parts, especially those with long lead times, in advance.However, because the part model is not determined until the elevator design is complete, stocking up is difficult, leading to shortages and excess inventory.

[0003] On the other hand, some specification information, such as the size and speed of the elevator car, is determined before the details of the design are finalized. Therefore, there is a technology that predicts the type of parts with long lead times from this specification information and stocks them in advance, and the details of this technology are described in Patent Document 1. Patent Document 1 states that "for specification items for which specification values have been input, the requirements calculation unit 11 calculates the required quantity of parts based on the specification values while referring to the configuration information, and for specification items for which specification values have not been input, the requirements calculation unit 11 calculates the required quantity of parts based on the predicted values while referring to the configuration information." [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Laid-Open No. 2017-528326 Summary of the Invention [Problem to be solved by the invention]

[0005] The technology in Patent Document 1 requires manual selection of parts for which required quantities are calculated. Therefore, if you want to use the technology in Patent Document 1 to improve the stocking of parts that are at risk of being out of stock or having excess inventory, you need to manually determine the risk of each part.

[0006] However, elevators have thousands of different parts, making it impossible to judge them all manually. Furthermore, the required quantity of parts is calculated based on the relationship between the specification values and the product's part configuration.

[0007] For this reason, it is necessary to investigate the relationship between the specification values and the product's component configuration in advance, but if there are many target components, this preliminary investigation takes a lot of time. Furthermore, when the specification values are not yet determined, the model is predicted using the predicted specification values, but the accuracy of the model prediction depends on the accuracy of the specification value prediction, so there is a risk of the prediction accuracy decreasing.

[0008] The present invention has been made in consideration of the above circumstances, and aims to provide a part type prediction device and method that can select long-term delivery parts that are at risk of being out of stock or in excess of stock, and then select specifications that are highly correlated with the type from a huge amount of specification information, thereby predicting the type with high accuracy. [Means for solving the problem]

[0009] In view of the above, in the present invention, there is provided a storage unit for storing information on receiving, shipping and inventory, which records the past number of receiving, shipping and inventory of parts constituting a product, and information on past cases, which records the specifications and models of past cases; a target part selection unit for selecting parts to be forecasted from the information on receiving, shipping and inventory; a specification grouping unit for classifying specifications to be used into a plurality of groups from the past case information; Every Extract frequently used specifications and calculate the correlation between specifications and model and output as correlation information The correlation measurement unit, which measures the correlation between the specifications, combines the highly correlated specifications for each group based on the correlation information and the information on the items to be predicted, and selects the specifications required for prediction. and output as specification selection result information and a model prediction section that predicts the target model based on the specification selection result information. and, Equipped with The specification grouping unit classifies specifications used from past case information into multiple groups by extracting specification items from the past case information, analyzing the similarity of each extracted specification item using natural language processing, and storing similar specification items and similarity scores in a memory unit. Next, similar specifications with a similarity of 0.5 or more are grouped as specifications with the same content but different notations. Furthermore, specification items with the same content but different notations within the past case information are unified, and specification information with similar usage trends is classified using a solution that can classify the characteristics of the target from past trends. The correlation measurement unit calculates the correlation between specifications and models for each group. This is a process in which the degree of correlation between the group's specification information and the model to be predicted is calculated using Cramer's correlation coefficient, and then a non-correlation test is used to eliminate specifications that are used less frequently. The specification selection unit selects the specifications required for prediction by using multiple regression analysis to learn a prediction model for new past case information for each candidate specification combination, with the specification information as the explanatory variable and the model as the objective variable. After learning the prediction model, the loss function of the prediction model for the squared sum error between the predicted value and the actual value is stored in the memory unit as score information. The process is repeated for each group until the score of the specification combination no longer improves, and the specification combination with the smallest score is extracted for each group. The part type prediction device is characterized by the above. [Effects of the Invention]

[0011] According to the present invention, it is easy to identify elevator parts with long lead times that are at risk of shortage or excess inventory. Furthermore, by selecting specifications that are correlated with the model from a vast number of specifications for those parts and predicting the model with high accuracy, it becomes possible to stock parts without the risk of shortage or excess inventory.

[0012] Problems, configurations, and effects other than those described above will become clear from the following description of the embodiments. [Brief explanation of the drawings]

[0013] [Figure 1] FIG. 2 is a diagram showing an example of a functional block configuration of the part type prediction device according to the present embodiment. [Figure 2] FIG. 2 is a diagram showing the flow of processing executed by the part type prediction device 1 in the present embodiment. [Figure 3] FIG. 10 is a diagram showing an example of the data structure of incoming / outgoing / inventory information D1. [Figure 4] FIG. 10 is a diagram showing an example of the data structure of predicted part information D2. [Figure 5] FIG. 10 is a diagram showing an example of the data structure of past case information D3. [Figure 6] FIG. 10 is a diagram showing an example of the data structure of similar specification information D4. [Figure 7] FIG. 10 is a diagram showing an example of the data structure of unified specification information D5. [Figure 8] FIG. 10 is a diagram showing an example of the data structure of new past case information D6. [Figure 9] FIG. 10 is a diagram showing an example of the data structure of specification group information D7. [Figure 10] FIG. 10 is a diagram showing an example of the data structure of correlation degree information D8. [Figure 11] FIG. 10 is a diagram showing an example of the data structure of score information D9. [Figure 12] FIG. 10 is a diagram showing an example of the data structure of specification selection result information D10. [Figure 13] FIG. 10 is a diagram showing an example of the data structure of prediction target project information D11. [Figure 14] FIG. 10 is a diagram showing an example of the data structure of model prediction information D12. [Figure 15] FIG. 10 is a diagram showing an example of an output screen. DETAILED DESCRIPTION OF THE INVENTION

[0014] Hereinafter, an embodiment of the present invention will be described with reference to the drawings.

[0015] In this specification, an order means an order from a customer, and its registration information consists of the customer name, delivery address, etc. However, it is not limited to this.

[0016] In the following explanation, an elevator is used as an example of a product, but the present invention can be widely applied to other products that are made to order and tend to have a wide variety of parts in small quantities, and the effects of the present invention will be particularly evident. [Example]

[0017] An embodiment of a part type prediction device according to the present invention will be described below with reference to the accompanying drawings.

[0018] 1 is a diagram showing an example of functional blocks of a part type prediction device 1 according to an embodiment of the present invention. The part type prediction device 1 of this embodiment is connected to a user terminal 50 used by a user and an external database DB2 in which data is stored, via a network Nw.

[0019] The user terminal 50 is an information processing device such as a PC (Personal Computer). A user issues a processing execution instruction to the part type prediction device 1, which is configured as a computer, through the user terminal 50. The user terminal 50 also has a function of displaying information output by the part type prediction device 1 to the user via the output unit 40. The external database DB2 is, for example, a system such as ERP (Enterprise Resources Planning), or a database or storage device that accumulates data equivalent thereto.

[0020] The network Nw communicatively connects the user terminal 50, the database DB2, and the part type prediction device 1. The network Nw is, for example, any of a local area network (LAN), a wide area network (WAN), a virtual private network (VPN), and a communication network that uses general public lines such as the Internet in part or in whole.

[0021] The part type prediction device 1 is an information processing device such as a PC or a server computer, and includes a storage unit DB1, a calculation unit CPU, an input unit 30, and an output unit .

[0022] The memory unit DB1 stores incoming / outgoing / inventory information D1, predicted parts information D2, past case information D3, similar specification information D4, unified specification information D5, new past case information D6, specification group information D7, correlation degree information D8, score information D9, specification selection result information D10, prediction target case information D11, and model prediction information D12.

[0023] This information can be divided into initial information stored in advance, final product information that is the result of the calculation, and intermediate product information obtained during the calculation. The initial information is incoming / outgoing / inventory information D1, past project information D3, and forecast target project information D11, and is shown by dotted lines in Figure 1. The final product information is forecasted parts information D2 and model forecast information D12, and is shown by thick solid lines in Figure 1. The rest is intermediate product information, and is shown by thin solid lines in Figure 1. Details of this information will be explained in the following explanations.

[0024] The processing contents executed by the CPU can be functionally expressed as having an information acquisition unit 11, a target part selection unit 12, a specification grouping unit 13, a correlation measurement unit 14, a specification selection unit 15, and a model prediction unit 16. Here, only an outline of each function will be explained.

[0025] First, the information acquisition unit 11 acquires information necessary for processing in the target part selection unit 12, specification grouping unit 13, correlation measurement unit 14, specification selection unit 15, and model prediction unit 16 from the memory unit DB1, and stores the processing results in the memory unit DB1.

[0026] At the start of the process, the storage unit DB1 contains the following data as shown in FIG. dotted line As initial information expressed by the above, incoming / outgoing / inventory information D1, past case information D3, and forecast target case information D11 are stored.

[0027] The target part selection unit 12 extracts parts that may be out of stock or in excess stock from all elevator parts based on the incoming / outgoing / inventory information D1, and stores the processing results in the memory unit DB1 as predicted part information D2.

[0028] The specification grouping unit 13 analyzes the similarity of each specification item extracted from the initial information, past case information D3, based on the predicted parts information D2, and then stores the analyzed data in the storage unit DB1 as similar specification information D4. Furthermore, using the similar specification information D4, the item names of specifications that are written differently but have the same content are unified and stored in the storage unit DB1 as unified specification information D5. Thereafter, the specification items of the past case information D3 are unified based on the unified specification information D5 and stored in the storage unit DB1 as new past case information D6. Next, the specifications are classified into multiple groups based on the input trends of the specifications from the new past case information D6 and stored in the storage unit DB1 as specification group information D7.

[0029] The correlation measurement unit 14 sorts the specifications for each group in the specification group information D7 in descending order of the correlation between the specification and the model, and stores the sorted specifications as correlation information D8 in the storage unit DB1. In addition, in order to delete specifications that are used less frequently, the new past case information D6 is used to count the number of past cases corresponding to each specification, and the counted number is stored in the storage unit DB1 as the number of past cases in the correlation information D8.

[0030] The specification selection unit 15 extracts specification combinations for each group in descending order of correlation from the correlation degree information D8, excluding specification information with a small number of past cases. Then, a prediction model is learned from the new past case information D6, and the prediction accuracy of the learned model is stored in the storage unit DB1 as score information D9. Furthermore, the specification combination with the smallest score for each group is extracted from the score information D9, and stored in the storage unit DB1 as specification selection result information D10.

[0031] The type prediction unit 16 learns a prediction model for each specification group from the specification selection result information D10 and the new past case information D6, predicts the part type to be predicted for each case in the prediction target case information D11, and stores the prediction results in the memory unit DB1 as type prediction information D12.

[0032] The input unit 30 is connected to the user terminal 50 and the database DB2 via the network Nw, receives incoming / outgoing / inventory information D1, past case information D3, and forecast target case information D11 from the database DB2, and stores them in the memory unit DB1. As a result, the memory unit DB1 is provided with the initial information necessary for carrying out subsequent processing.

[0033] The output unit 40 outputs the predicted part information D2, the specification group information D7, the correlation degree information D8, the score information D9, and the specification selection result information D1 stored in the storage unit DB1. D10 The model prediction information D12 is transmitted to the user terminal 50, and the result is displayed to the user.

[0034] Next, the flow of processing executed by the part type prediction device 1 in this embodiment will be explained using the flowchart in Fig. 2. Note that the following series of processing is premised on the fact that incoming / outgoing / inventory information D1, past case information D3, and prediction target case information D11 are recorded in the external database DB2, and is started in response to a start command from the user to the user terminal 50, for example.

[0035] In response to a start command from the user, the information acquisition unit 11 acquires the incoming / outgoing / stock information D1 and the past case information D3 from the database DB2 via the network Nw and stores them in the memory unit DB1. As a result, the memory unit DB1 is provided with the initial information necessary to carry out the subsequent processing.

[0036] In this state, in processing step S1, the target part selection unit 12 selects parts that may be out of stock or in excess stock from the incoming / outgoing / inventory information D1 as target parts. The incoming / outgoing / inventory information D1, which is the initial information used at this time, includes the data exemplified in FIG.

[0037] Figure 3 shows an example of the data structure of the receiving, shipping, and inventory information D1. The receiving, shipping, and inventory information D1 stores information on the receiving record D1d, shipping record D1e, inventory record D1f, and safety stock D1g of all elevator parts, along with the part D1a, model D1b, and date D1c.

[0038] Here, part D1a indicates the number information that identifies the part. Model D1b indicates the model number of the part that makes up the part. Date D1c indicates the date on which the actual incoming, outgoing, and inventory data was compiled. Incoming quantity D1d indicates the actual number of parts received. For example, "Model 1" in Figure 3 indicates that 100 units were received on "'21 / 1". Outgoing quantity D1e indicates the actual number of parts shipped. For example, "Model 1" in Figure 2 indicates that 300 units were shipped on "'21 / 1". Inventory quantity D1f indicates the actual number of parts in stock. For example, "Model 1" in Figure 3 indicates that 350 units were in stock on "'21 / 1". Safety stock quantity D1g indicates the safety stock quantity of parts. For example, the safety stock quantity for "Model 1" in Figure 3 is 300 units.

[0039] Processing Steps S1 In determining whether there is a shortage or excess inventory, for example, if the model D1b in Figure 3 is "Model 1" and the date D1c is "'21 / 2~'21 / 6", the inventory quantity D1f is 150~50 unitsThis is less than the "300 units" of the safety stock D1g. In this way, parts whose stock quantity D1f is less than the safety stock quantity D1g are judged to be out of stock. Similarly, if the model D1b in Figure 3 is "Model 2" and the date D1c is "'21 / 1~'21 / 6", the stock quantity D1f is 1000 units ~ 1140 units It can be seen that the safety stock quantity D1f is more than "100 units". In this way, parts whose stock quantity D1f exceeds the safety stock quantity D1f are determined to be potentially in excess of stock. Therefore, model 1 and model 2 are treated as target parts and are stored in the storage unit DB1 as predicted part information D2 in FIG. 4. In the following, the processing of process steps S2 to S9 is repeated for each target part selected in process step S1.

[0040] FIG. 4 shows an example of the data structure of the predicted parts information D2 created in processing step S1. The predicted parts information D2 stores the results of extracting parts that are likely to be out of stock or overstocked from the parts information registered in the incoming / outgoing / inventory information D1 in FIG. 3, and is composed of prediction target parts D2a and part types D2b. The prediction target parts D2a indicate the name of the part to be predicted. The part type D2b indicates the type of the part; for example, if the prediction target part D2a in FIG. 4 is "Part 1," the part type D2b may have types 1, 2, 3, 4, and 5.

[0041] Next, in processing step S2 of the flow of Fig. 2, past case information D3, which is initial information, is read in. Fig. 5 shows an example of the data structure of past case information D3.

[0042] The past project information in Figure 5 stores information about past projects for parts registered in the predicted part information D2, and is composed of project D3a, customer name D3b, delivery date D3c, elevator specification information such as load weight D3d, weight D3e, load D3f, weight D3g, speed D3h, speed D3i, door width D3j, opening direction D3k, and door type D3l, part name D3m, and model name D3n.

[0043] Among these, the item D3a indicates number information for identifying the item, the customer name D3b indicates the name of the customer for each item, and the delivery date D3c indicates the deadline for delivery of the product to the customer.

[0044] Load weight D3d, weight D3e, load D3f, and weight D3g indicate the load weight of the product desired by the customer. For example, if project D3a in Figure 5 is project "1," the customer is looking for a product that can carry up to 500 kg. Note that the load weight is expressed as load weight D3d, weight D3e, load D3f, weight D3g, etc. This is because there are many cases where the specifications are the same but there are variations in the notation, so the specifications can be understood unambiguously even if there are variations in the notation.

[0045] Speed D3h and speed D3i indicate the product speed desired by the customer. For example, if project D3a in Figure 5 is project "1," the customer desires a product with a speed of 120 m / min. Door width D3j indicates the door width of the product desired by the customer. For example, if project D3a in Figure 5 is project "1," the customer desires a product with a door width of 1000 mm. Opening direction D3k indicates the opening direction of the product desired by the customer. For example, if project D3a in Figure 5 is project "2," the customer desires a product with an opening direction of L. Door type D3l indicates the product type of the door to be constructed. For example, if project D3a in Figure 5 is project "1," the customer desires a door of type 1. Part name D3m indicates the name of the target part. For example, if project D3a in Figure 5 is project "1," the part name constitutes part 1. Model name D3n indicates the model type of the target part. For example, if the case D3a in Fig. 5 is case "1", the part model is model 1. Note that in each data item, if no data is entered, it is displayed as "-".

[0046] Next, in processing step S3 of the flow in Figure 2, grouping is performed by specification. This processing is performed because the specification information used for each project differs depending on the ordered model, etc., and so it is necessary to divide the specifications into multiple groups.

[0047] Therefore, in processing step S3, the specification grouping unit 13 extracts specification items from the past case information D3. Then, as a first process, it analyzes the similarity of each extracted specification item using natural language processing or the like, and stores similar specification items and their similarity scores in the storage unit DB1 as similar specification information D4. Furthermore, as a second process, it groups similar specifications in the similar specification information D4, for example, with a similarity of 0.5 or more, as specifications with different notations but the same content, and stores these in the storage unit DB1 as unified specification information D5. Furthermore, as a third process, it uses the unified specification information D5 to unify specification items with the same content but different notations in the past case information D3, and stores the processing results in the storage unit DB1 as new past case information D6. As a fourth process, in the new past case information D6, it groups the specification information based on the tendency of specifications to be used, and stores the processing results in the storage unit DB1 as specification group information D7.

[0048] Regarding the first process, specifically, first, the specification grouping unit 13 focuses on the specification items (load weight D3d, weight D3e, load D3f, weight D3g, speed D3h, speed D3i, door width D3j, opening direction D3k, door type D3l) of the past case information D3 in FIG. 5, analyzes similar specification items, and stores the similar specifications and the degree of similarity in the memory unit DB1 as similar specification information D4 in FIG. 6.

[0049] FIG. 6 shows an example of the data structure of similar specification information D4. The similar specification information D4 stores specifications similar to each specification item registered in the past project information D3, and is composed of specification items D4a, similar specifications D4b, and scores D4c. The specification items D4a indicate the names of the specification items registered in the past project information D3. For example, the specification item D4a in the similar specification information D4 in FIG. 6 indicates data when the specification item in the past project information D3 in FIG. 5 is "load weight D3d." The similar specifications D4b indicate specification items registered in the past project information D3 other than the specification items in the similar specification information D4. For example, when the specification item D4a in the similar specification information D4 in FIG. 6 is "load weight," the similar specifications D4b are the specification items other than the load weight, such as weight D3e, load D3f, weight D3g, and speed D3h, among the specification items in the past project information D3 in FIG. 5. The score D4c indicates the similarity between each specification item D4a and the similar specification D4b as a numerical value. For example, if the specification item in the similar specification information D4 in FIG. 6 is "load weight" and the similar specification is "weight," the score indicates that the similarity between these two specifications is 0.98.

[0050] For example, if the specification item D4a is "load weight," the similarity with other specification items D4a such as "weight," "load," "weight," and "door width" is analyzed using cosine similarity or the like, and the processing results are stored in the storage unit DB1 as similar specification information D4. Here, if the specification item of the similar specification information D4 in Fig. 5 is "load weight," the similarity with the similar specification "weight" is "0.98," and the similarity with the similar specification "load" is "0.83."

[0051] Next, for the second process, similar specification information D4 in FIG. 6 is used to unify specification items that have different specification notations but the same content, and the processing results are stored in storage unit DB1 as unified specification information D5 in FIG.

[0052] 7 shows an example of the data structure of the unified specification information D5. The unified specification information D5 is composed of pre-unification D5a and post-unification D5b. For example, when a similar specification D4b with a similarity (score) D4c of 0.5 or more in the similar specification information D4 in FIG. 6 is considered to be a specification with the same content but different notation, the specifications with the same content as "load weight" in the specification item D4a of the similar specification information D4 in FIG. 6 are "weight," "load," and "weight."

[0053] Thereafter, in the unified specification information D5 in Fig. 7, the grouped similar specifications, "load weight," "load," "weight," and "weight," are stored in the pre-unification D5a, and the first analyzed specification item, "load weight," is stored in the post-unification D5b. Here, the analysis of similar specifications proceeds in the order of the specification items in the past case information D3 in Fig. 5, but if a specification item has already been grouped as a similar specification to a previously analyzed specification item, no analysis is performed.

[0054] for example, Figure 5 If the specification item in the past case information D3 is "weight," the specification item is grouped as a similar specification when analyzing "load weight," and since it has already been stored in the "before unification" column D5a of the unified specification information D5 in Figure 6, no analysis is performed, and the analysis of the next specification item, "speed," begins.

[0055] Next, regarding the third process, the notation of the specification items in the past case information D3 in FIG. 5 is unified using the unified specification information D5 in FIG. 7, and the processing results are stored in the storage unit DB1 as new past case information D6.

[0056] 8 shows an example of the data structure of the new past project information D6. The new past project information D6 stores the results of modifying the specification information in the past project information D3 based on the standardization of the specification notation in the unified specification information D5, and is composed of a project D6a, a customer name D6b, a delivery date D6c, a load weight D6d, a speed D6e, a door width D6f, an opening direction D6g, a door model D6h, a part name D6i, and a model name D6j.

[0057] The content of each item is the same as the item in the past project information D3. For example, in the case of the new past project information D6 in Fig. 8, the specification notations of "load weight," "weight," "load," and "weight" in the past project information D3 in Fig. 5 are unified to "load weight," and the specification notations of "speed" and "speed" are unified to "speed," and all columns before unification are consolidated into the unified column.

[0058] For example, in the past case information D3 of Figure 5, if the case is "1", the specification items with weight "500 kg", load "500 kg", and weight "500 kg" are "loaded weight" according to the unified unified specification information D5 of Figure 7, so the notation is unified to "loaded weight", and the processing result is stored in the memory unit DB1 as new past case information D6.

[0059] Furthermore, regarding the fourth process, the new past matter information D6 is used to find specification information that has similar tendencies in the specifications used. Many The target characteristics are classified based on past trends such as class classification using a solution method that can classify the target characteristics, and the classification results are stored in the storage unit DB1 as specification group information D7.

[0060] FIG. 9 shows an example of the data structure of specification group information D7. The specification group information D7 stores the results of classifying specifications into multiple groups in the new past case information D6 of FIG. 8, and is composed of specification groups D7a and specification information D7b. The specification group D7a indicates the specification group name that identifies the group of specification information. The specification information D7b indicates the specification information that belongs to the specification group. For example, if the specification group in FIG. 9 is "Group 1," the specification information that belongs to it is "load weight," "speed," "door width," and "door type."

[0061] For example, if the case is "1" in the new past case information D6 in Figure 8, the specifications used will be "load weight, speed, door width, door type" because there is no description of the opening direction. Also, if the case is "2" or "3", the specifications used will be "load weight, speed, opening direction, door type" because there is no description of the door width. Based on this tendency of the specifications used, the specifications are divided into specification group 1 of "load weight, speed, door width, door type" and specification group 2 of "load weight, speed, opening direction, door type", and the results are stored in the memory unit DB1 as specification group information D7.

[0062] Returning to FIG. 2, in processing step S4, the correlation measurement unit 14 measures the correlation between specifications and models for each specification group from the new past case information D6 and the specification group information D7, and stores the result as correlation information D8 in the memory unit DB1.

[0063] 10 is an example of the data structure of the correlation degree information D8. The correlation degree information D8 stores the results of analyzing the correlation between the specification information in the new past case information D6 and the part model for each specification group in the specification group information D7, and is composed of a specification group D8a, specification information D8b, a correlation degree D8c, and the number of past cases D8d. The specification group D8a indicates the specification group name that identifies the group of the specification information D8a. The specification information D8a indicates the specification information that belongs to the specification group. The correlation degree D8c indicates the correlation between the specification information in the new past case information D6 and the model of the prediction target.

[0064] For example, if the specification group D8a in FIG. 10 is "Group 1" and the specification information D8b is "load weight," the correlation degree D8c is 0.97. The number of past cases D8d indicates the number of cases of the corresponding specification in the new past case information D6. For example, if the specification group in FIG. 9 is "Group 1" and the specification information is "load weight," the number of past cases is 33.

[0065] Specifically, for example, in FIG. Specification Group D7a If "Group 1", specification information D7bare "load weight," "speed," "door width," and "door type." From the new past case information D6 in FIG. 8, the correlation between these pieces of specification information and the model of the prediction target is calculated using Cramer's correlation coefficient or the like, and stored in the storage unit DB1 as correlation degree information D8. In addition, using the new past case information D6, the number of cases in which each piece of specification information was used is tallied, and stored in the storage unit DB1 as the number of past cases D8d of the correlation degree information D8.

[0066] Next, in processing step S5 of FIG. 2, the information acquisition unit 11 acquires the latest forecast target item information D11 from the database DB2 via the network Nw and stores it in the storage unit DB1. 15 The process of steps S6 to S7 is repeated until the score generated in step S7 (described later) no longer improves.

[0067] In the repeated processing of processing steps S6 to S7, first, in processing step S6, Specification Selection Section 15 extracts specification combination candidates for each specification group from the correlation degree information D8, and stores the processing results in the storage unit DB1 as score information D9.

[0068] Specifically, first, specifications that are used less frequently are removed from the correlation degree information D8 using a non-correlation test or the like. For example, Figure 10 If the specification group of the correlation information D8 is "Group 1," the specification information with the highest correlations, in order from top to bottom, are "load weight," "speed," "door width," and "door type." If the result of using the non-correlation test shows that the number of cases required for prediction is 10, the number of past cases with the specification information "speed" is 8, so this specification is excluded. After excluding specifications that are used less frequently in this way, the specification information with the highest correlations is combined from top to bottom, and the results are stored in the storage unit DB1 as score information D9. For example, in the above example, if the specification combinations are set in order of highest correlation, they would be "load weight," "load weight and door width," and "load weight, door width, and door type." These candidate specification combinations are stored in the storage unit DB1 as score information D9.

[0069] In processing step S7 of the repetitive processing, the specification selection unit 13 learns a type prediction model from the new past case information D6 and the score information D9 in order from the top of the specification combination candidates, calculates the prediction error of the prediction model, and stores the result in the memory unit DB1 as score information D9.

[0070] 11 shows an example of the data structure of score information D9. Score information D9 stores the measurement results of prediction accuracy when a prediction model is trained by combining specifications with high correlations for each specification group from correlation degree information D8 and new past case information D6, and is made up of specification groups D9a, specification combinations D9b, and scores D9c.

[0071] The specification group D9a indicates a specification group name that identifies a group of specification information. D9b indicates the results of combining specification information with high correlation for each specification group from the correlation information D8. Score D9c indicates the prediction error when a prediction model is trained using new past case information D6 based on the specification information of the specification combination, with the smaller the error, the higher the accuracy of the prediction model. For example, if the specification group in Figure 11 is "Group 1" and the specification combination is "load weight," the score is 0.431. Also, if the specification group in Figure 11 is "Group 1" and the specification combination is "load weight, door width," the score is 0.282. Also, if the specification group in Figure 10 is "Group 1" and the specification combination is "load weight, door width, door type," the score is 0.644.

[0072] Specifically, for each candidate specification combination calculated in processing step S6, a prediction model of new past case information D6 is trained using multiple regression analysis or the like, with specification information as the explanatory variable and model as the objective variable. After the prediction model is trained, a loss function of the prediction model, such as the squared sum error between predicted values and actual values, is stored in storage unit DB1 as score information D9. For example, if the specification combination in score information D9 in Figure 11 is "load weight," the score of the loss function is "0.431," so "0.431" is stored as the score.

[0073] Finally, in Figure 2, process steps S6 and S7 are repeated for each specification group until the score of the specification combination no longer improves. For example, if the specification group in score information D9 in Figure 11 is "Group 1," as the number of specification combinations increases, the pattern with the smallest score, "0.282," will be found when the specification combination is "load weight, door width." Therefore, the repetition ends when the score for the next specification combination, "load weight, door width, door type," is calculated.

[0074] 2, in processing step S8, the specification selection unit 13 extracts the specification combination with the smallest score for each specification group from the score information D9, and stores it as specification selection result information D10 in the storage unit DB1. For example, when the specification group of the score information D9 in FIG. 11 is "Group 1," the specification combination with the smallest score is "load weight and door width," and this result is stored in the storage unit DB1 as specification selection result information D10.

[0075] FIG. 12 shows an example of the data structure of the specification selection result D10. The specification selection result D10 stores the result of determining, for each specification group, the combination of specifications with the lowest score among the specification combinations determined by the score information D9, and is composed of a specification group D10a and specifications used for prediction D10b. The specification group D10a indicates the specification group name that identifies the group of specification information. The specifications used for prediction D10b indicates the combination of specifications with the lowest score among the specification combinations determined by the score information D9. For example, if the specification group in FIG. 11 is "Group 1," the specifications used for prediction are the load weight and door width.

[0076] In processing step S9 in FIG. 2, the information acquisition unit 11 acquires the forecast target case information from the database DB2 via the network Nw. D11 is acquired and stored in the storage unit DB1.

[0077] 13 shows an example of the data structure of the forecast target project information D11. The forecast target project information D11 stores registration information for currently received orders and is composed of elevator specification information such as project D11a, customer name D11b, delivery date D11c, load weight D11d, weight D11e, load D11f, weight D11g, speed D11h, speed D11i, door width D11j, opening direction D11k, and door type D11l.

[0078] In a subsequent processing step S9, the type prediction unit 16 learns a type prediction model for each specification group from the prediction target project information D11, the new past project information D6, and the specification selection result information D10, and predicts the type to be used in each prediction target project.Then, the processing result is stored in the storage unit DB1 as type prediction information D12.

[0079] FIG. 14 shows an example of the data structure of type prediction information D12. Type prediction information D12 stores the results of predicting the type to be predicted for each project in prediction target project information D11 after constructing a prediction model for each group from specification selection results D10 and new past project information D6, and is composed of project D12a, specification group D12b, specification D12c, part D12d, and type prediction D12e. Project D12a indicates number information that identifies the project to be predicted. Specification group D12b indicates the specification group of the project to be predicted. For example, project D12a in FIG. 14 is 「101」 In this case, the specification group D12b belongs to group 1. The specification D12c indicates a combination of specifications used when constructing a prediction model. For example, Figure 14 Project D12a 「101」 If the specification group D12b is "Group 1", the specification D12c is the load weight and door width. The part D12d indicates the part to be predicted for each case. The predicted model D12e indicates the predicted model of the part to be used for each case. For example, Figure 14 Project D12a 「101」 In this case, part D12e is part 1 and the predicted model is model 1.

[0080] For example, the specification selection result information in Figure 12 D10If the specification group is "Group 1," the specifications used for prediction are "load weight and door width." Next, past cases whose specification group is "Group 1" are extracted from the new past case information D6 in Figure 8, and the relationship between the specification information "load weight and door width" and the model is learned using multiple regression analysis, etc. The prediction results are then stored in the memory unit DB1 as model prediction information D12.

[0081] The information on the final product obtained as described above is output externally via the output unit 40 in Figure 1 together with the initial information and information on the intermediate products as needed, and is presented to the user on a monitor in a viewable format.

[0082] 15 shows an example of the configuration of a monitor output screen 400 in this case. The output screen 400 shown in the figure displays target part selection results 401, specification selection results 402, and model prediction results 403, which are information as the final deliverables.

[0083] Here, the target part selection results 401 are displayed in a list format, and the list is made up of a parts column 151, a model column 152, a risk column 153, and a selection result column 154. The specification selection results 402 are made up of a case button 155 for selecting a case and a list of the specification selection results of the selected case, and the list is made up of a specification group column 156, a specification information column 157, a correlation degree column 158, a number of past cases column 159, and a specification selection result column 160. The list of model prediction results 403 is made up of a parts column 161, a predicted model column 152, and a specification group column 156. 163 It consists of:

[0084] In this display screen, the parts column 151 displays the names of all parts of the elevator. The model column 152 displays the model names of the parts that each part is made up of. The risk column 153 displays whether the model belonging to the model column 152 is likely to be out of stock or overstocked. The selection result column 154 displays whether the item in the risk column 153 is likely to be out of stock or overstocked. The case number column 155 displays information about the case number specified by the user. The specification group column 156 displays the specification group. The specification information column 157 displays the specification information belonging to the specification group column 156. The correlation degree column 158 displays the degree of correlation between the specification information column 157 and the model. The number of past cases column 159 displays the number of past cases in which the specification information column 157 was used. The specification selection result column 160 displays the result of the determination as to whether or not to use the specification information column 157 for model prediction. In the parts column 161, the part name when the model is predicted based on the specification selection result column 160 is displayed. 163 In the column 160, the model name when the model is predicted based on the specification selection result column 160 is displayed.

[0085] The present invention has been described in detail above based on the embodiments, but to put it simply, the present invention reads the latest order item information and predicts the type of an incomplete order item based on specification information and past shipping information. More specifically, in making the prediction, past orders are read, candidate specifications for predicting the type are selected from a vast number of specifications, specifications to be used for prediction are selected from the narrowed-down candidates, a prediction model is constructed, and the specification information is used to predict the type of incomplete parts in the latest order item.

[0086] To achieve this, it is essential to select candidate specifications from the vast number of specifications to predict the model, and to select specifications to be used for prediction from the narrowed down candidates and build a prediction model.

[0087] For this reason, the selection of candidate specifications is divided into pre-processing and main processing. In the pre-processing, elevators are designed by different designers for each model and distribution channel, and specification descriptions are not standardized, so designer-dependent elements are standardized. For example, even for the same specifications, variations in notation occur depending on the designer, so natural language processing is used to unify duplicate specification information. Also, while specifications used vary depending on the model, classification of models is subjective and not resilient, so the combinations of specifications used are analyzed and classified into multiple groups independent of the designer. In the main processing to select candidate specifications, after pre-processing, infrequently used specifications such as project-specific specifications are excluded for each group, and candidate specifications with a high correlation with the model to be predicted are selected.

[0088] In addition, when constructing the predictive model, for each group, we extracted combinations of specifications to be used for prediction in order of the highest correlation, and repeatedly trained the predictive model until the accuracy of the model was minimized.

[0089] As a result, compared to conventional methods, this invention uses past data to learn the relationship between specification information and models, and by building a model prediction model, it is possible to identify incomplete models, so it can handle even if specifications are added or deleted by learning the prediction model each time.In addition, since the specification information required for prediction is extracted taking correlation into consideration, prediction accuracy can be improved.

[0090] Furthermore, the present invention does not use incomplete specification information, but predicts the part type within the range of completed specifications, so the type is predicted directly from completed specifications, resulting in high accuracy. It can also handle a huge number of specifications.

[0091] As described above, according to the present invention, after selecting parts with long delivery times that are at risk of being out of stock or overstocked, specifications that are highly correlated with the model are selected from a vast amount of specification information, making it possible to predict the model with high accuracy. [Explanation of symbols]

[0092] 1: Part type prediction device 11: Target part selection section 12: Information acquisition department 13: Specification grouping section 14: Correlation measurement unit 15: Specification selection section 16: Type prediction section 30: Input section 40: Output section 50: User terminal 400: Output screen CPU: Arithmetic unit DB2:Database DB1: Storage section Nw: Network 30: Input section 40: Output section D1: Inventory information D2: Predicted parts information D3:Past case information D4: Similar specification information D5: Unified specification information D6: New past case information D7: Specification Group Proposal Information D8: Correlation information D9: Score information D10: Specification selection result information D11: Project information for forecast D12: Model prediction information

Claims

1. a storage unit that stores inventory information that records the past inventory numbers of parts that constitute products, inventory numbers that are received, shipped, and stocked, and past case information that records the specifications and models of past cases; a target part selection unit that selects a part to be predicted from the incoming / outgoing / inventory information; a specification grouping unit that classifies specifications to be used into a plurality of groups based on the past case information; a correlation measurement unit that extracts frequently used specifications for each group, calculates a correlation between the specifications and the model, and outputs the calculated correlation information; a specification selection unit that combines specifications with high correlation for each group based on the correlation degree information and the information on the project to be predicted, selects specifications required for prediction, and outputs the selected specifications as specification selection result information; a model prediction unit that predicts a target model from the specification selection result information, The specification grouping unit performs a process of classifying specifications used from past case information into a plurality of groups by extracting specification items from the past case information, analyzing the similarity of each extracted specification item using natural language processing, storing similar specification items and similarity scores in the storage unit, then grouping similar specifications with a similarity of 0.5 or more as specifications with the same content but different notations, and further unifying specification items with the same content but different notations within the past case information, and classifying specification information with similar usage trends using a solution method that can classify the characteristics of the target from past trends. the process of calculating the degree of correlation between the specifications and the model in the correlation degree measurement unit is a process of calculating, for each of the groups, the degree of correlation between the specification information of the group and the model of the prediction target using Cramer's correlation coefficient, and then excluding specifications that are used less frequently using a non-correlation test; The process of selecting specifications required for prediction in the specification selection unit is a process of: for each candidate combination of specifications, using specification information as an explanatory variable and model as a target variable, learning a prediction model of new past case information using multiple regression analysis; after learning the prediction model, storing a loss function of the prediction model of the square sum error between predicted values and actual values as score information in the storage unit; repeating the process for each group until the score of the specification combination no longer improves; and extracting the specification combination with the smallest score for each group. A part type prediction device characterized by:

2. 2. The part type prediction device according to claim 1, The part type prediction device is characterized in that the target part selection unit selects parts that are likely to be out of stock or in excess stock based on past inventory changes of elevator parts, including inventory receipts, deliveries, and stock levels.

3. 2. The part type prediction device according to claim 1, The part type prediction device is characterized in that the type prediction unit learns a prediction model from specification selection result information, specification group information, past case information, and target part information, and predicts the type of the prediction target case.

Images