Type prediction apparatus and type prediction method

Abstract

To highly accurately predict a mounting probability for each model of a component mountable on a new model having no information of past items.SOLUTION: The model prediction device predicts a future component demand for each model on the basis of a past record, for a product including a model in which a lower limit value and an upper limit value of each specification item are set in advance for each model and a predetermined component can be mounted in different types. A learning data generation unit configured to generate learning data using attribute values of each specification item recorded for each past case, preset attribute values for each model, and a provision specification range preset for each model type, a prediction model generation unit configured to generate a model for predicting a mounting probability for each model type based on the learning data, and a mounting probability calculation unit configured to calculate a mounting probability for each model type based on a provision specification range of a new model for which there is no past record, a mountable model for each component, and the prediction model.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a type prediction device and a type prediction method. [Background technology]

[0002] With made-to-order products, production begins after an order is placed in accordance with customer requirements, but long-lead parts must be stocked in advance to meet delivery deadlines. However, because the part type is not determined until the product design is complete, stocking up is difficult, resulting in stockouts and excess inventory. On the other hand, in the case of elevators, for example, some specification information, such as the size and speed of the car, is determined before the design details are finalized, so a method has been adopted in which the type of long-lead parts is predicted from this specification information and stocked in advance.

[0003] For example, Patent Document 1 discloses that "the latest order project information is read and the model of the incomplete project is predicted based on the specification information and past shipping information," and that "in making the prediction, more specifically, past projects are read and candidate specifications for predicting the model are selected from the vast number of specifications, specifications to be used for the prediction are selected from the narrowed down candidates, a prediction model is constructed, and the specification information is used to predict the model of the incomplete parts in the latest order project" (paragraph 0085). [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Publication No. 2023-68360 Summary of the Invention [Problem to be solved by the invention]

[0005] In the technology disclosed in Patent Document 1, it is difficult to build a prediction model for new models having a new specification range that does not exist in existing models, since there is no information on past cases.

[0006] An object of the present invention is to predict with high accuracy the mounting probability for each model of a part that can be mounted on a new model for which there is no information on past cases. [Means for solving the problem]

[0007] In order to solve the above-mentioned problems, the present invention provides a type prediction device that predicts future demand for parts for each type based on past performance for products in which lower and upper limit values ​​for each specification item are set in advance for each model and which include models in which different types can be installed for specified parts. The device includes: a learning data generation unit that generates learning data using the product model, specification values ​​for each specification item, and type of each part, attribute values ​​for a plurality of attribute items set in advance for each type, and lower and upper limit values ​​for the plurality of specification items set in advance for each model, all recorded for each past case; a prediction model generation unit that generates a prediction model that predicts the installation probability for each model, type, and specification value of each specification item, based on the learning data; and a mounting probability calculation unit that calculates the installation probability for each type and specification value of each specification item for a new model based on the lower and upper limit values ​​of each specification item for a new model for which no past performance exists, the types that can be installed for each part, and the prediction model. [Effects of the Invention]

[0008] According to the present invention, it is possible to predict with high accuracy the probability of mounting, for each model, a part that can be mounted on a new model for which there is no information on past cases.

[0009] Problems, configurations, and effects other than those described above will become clear from the following description of the embodiments. [Brief explanation of the drawings]

[0010] [Figure 1] FIG. 2 is a diagram showing an example of functional blocks of the entire model prediction system. [Figure 2] FIG. 4 is a diagram showing an example of a data structure of search target information. [Figure 3] FIG. 4 is a diagram showing an example of a data structure of case information. [Figure 4]FIG. 10 is a diagram showing an example of a data structure of case-specific specification information. [Figure 5] FIG. 10 is a diagram showing an example of a data structure of case-specific on-board model information. [Figure 6] FIG. 10 is a diagram showing an example of a data structure of case-specific model information. [Figure 7] FIG. 10 is a diagram showing an example of a data structure of model-specific provided specification range information. [Figure 8] FIG. 4 is a diagram showing an example of a data structure of model-specific model list information. [Figure 9] FIG. 4 is a diagram showing an example of a data structure of model predicted specification information. [Figure 10] FIG. 4 is a diagram showing an example of a data structure of model-specific attribute information. [Figure 11] FIG. 4 is a diagram showing an example of a data structure of learning data information. [Figure 12] FIG. 10 is a diagram showing an example of a data structure of mounting probability distribution information. [Figure 13] FIG. 10 is a diagram showing an example of a screen displayed by an output unit. [Figure 14] 10 is a flowchart showing an example of processing by a model prediction device. [Figure 15] FIG. 10 is a diagram showing an example of the data structure of input variable information. DETAILED DESCRIPTION OF THE INVENTION

[0011] Hereinafter, an embodiment of the present invention will be described with reference to the drawings.

[0012] In this specification, a "project" refers to an order from a customer, and its registered information consists of the customer name, delivery destination, etc. Furthermore, in this specification, a "product" refers to a model in which lower and upper limits for each specification item are set in advance for each model, and includes models that can be equipped with different models of specified parts, and is produced using a made-to-order production method. In the following explanation, an elevator (especially an elevator) is used as an example of a product, but the invention can also be applied to other products.

[0013] 1 is a diagram showing an example of the overall functional blocks of a type prediction system 100 according to this embodiment. The type prediction system 100 of this embodiment includes a type prediction device 101, a user terminal 103, and a database 104, which are connected to each other via a network 102.

[0014] The model prediction device 101 is an information processing device such as a PC or server computer that predicts future part demand for each model based on past performance, and is equipped with a memory unit 110, a calculation unit 130, an input unit 140, and an output unit 150.

[0015] The storage unit 110 stores search target information 111, case information 112, case-specific specification information 113, case-specific mounted model information 114, case-specific model information 115, model-specific provided specification range information 116, model-specific model list information 117, model predicted specification information 118, and model-specific attribute information 119, which are acquired from the database 104 by an information acquisition unit 131 described later. Furthermore, the storage unit 110 also stores learning data information 120 generated by a learning data generation unit 132 described later, and installation probability distribution information 121 calculated by an installation probability calculation unit 134 described later. Note that the storage unit 110 may also store a prediction model generated by a prediction model generation unit 133 described later.

[0016] The user terminal 103 is an information processing device such as a PC (Personal Computer). A user issues a processing execution instruction to the type prediction device 101 through the user terminal 103. The user terminal 103 also has a function of displaying information output by the type prediction device 101 to the user.

[0017] The database 104 is, for example, a database or storage device that accumulates data from a system such as ERP (Enterprise Resources Planning), or a similar system. Although omitted from Fig. 1, the database 104 stores information such as project information 112, specification information by project 113, installed model information by project 114, model information by project 115, provided specification range information by model 116, model-by-model model list information 117, model predicted specification information 118, and model attribute information 119.

[0018] The network 102 communicatively connects the model prediction device 101, the user terminal 103, and the database 104. The network 102 is, for example, a local area network (LAN), a wide area network (WAN), a virtual private network (VPN), or a communication network that uses general public lines such as the Internet in part or in whole.

[0019] 2 is a diagram showing an example of the data structure of search target information 111. The search target information 111 stores information selected by the user as a search target, and is composed of part 111a and model 111b. Part 111a indicates the target part to be predicted, such as a "motor." Model 111b indicates the target model to be predicted, such as "model C."

[0020] FIG. 3 is a diagram showing an example of the data structure of the project information 112. The project information 112 stores information about projects that have been ordered in the past, and is composed of a project number 112a, a delivery destination 112b, and a delivery date 112c. The project number 112a indicates a number that identifies the project, such as "001." The delivery destination 112b indicates the project name of the delivery destination, such as "XX Building." The delivery date 112c indicates the deadline for delivery of the elevator product to the delivery destination, such as "2023 / 01 / 11." For example, if the project number 112a in FIG. 3 is "001," this indicates that the delivery destination 112b is the "XX Building" project, and the delivery date 112c is "2023 / 01 / 11."

[0021] FIG. 4 is a diagram showing an example of the data structure of the project-specific specification information 113. The project-specific specification information 113 stores specification information for each past project, and is composed of a project number 113a, a specification item 113b, and a specification value 113c. The project number 113a indicates a number that identifies the project, such as "001." The specification item 113b indicates a specification item name that identifies the specification information, such as "load weight." The specification value 113c indicates a specification value of the specification item, such as "500." For example, if the project number 113a in FIG. 4 is "01" and the specification item 113b is "load weight," the specification value 113c is "500."

[0022] FIG. 5 is a diagram showing an example of the data structure of the project-specific mounted model information 114. The project-specific mounted model information 114 stores part model information to be mounted for each past project, and is composed of a project number 114a, a part 114b, and a model 114c. The project number 114a indicates a number to identify the project, such as "001." The part indicates a part name to identify the part, such as "motor." The model indicates the model name of the part name to be mounted for each project, such as "M01." For example, if the project number 114a in FIG. 5 is "001" and the part 114b is "motor," the model 114c indicates that a model of "M01" will be mounted.

[0023] FIG. 6 is a diagram showing an example of the data structure of the case-specific model information 115. The case-specific model information 115 stores information about the model installed for each past case, and is composed of a case number 115a and a product model 115b. The case 115a indicates a number that identifies the case, such as "001." The product model indicates a model name that identifies the model, such as "model A." For example, if the case number 115a in FIG. 6 is "001," this indicates that the product model 115b is "model A."

[0024] FIG. 7 is a diagram showing an example of the data structure of model-specific provided specification range information 116. The model-specific provided specification range information 116 pre-stores the provided range of specification information for each model of elevator, and is composed of product model 116a, specification item 116b, lower limit 116c, and upper limit 116d. Product model 116a indicates a model name that identifies the model, such as "Model A." Specification item 116b indicates a specification item that identifies the specification information for each model, such as "load weight." Lower limit 116c indicates the lower limit of the provided specification range in the specification information for the specification item. Upper limit 116d indicates the upper limit of the provided specification range in the specification information for the specification item. For example, if product model 116a in FIG. 7 is "Model A" and specification item 116b is "load weight," lower limit 116c, which is the lower limit of the provided specification range, is "300," and upper limit 116d, which is the upper limit of the provided specification range, is "2,000."

[0025] FIG. 8 is a diagram showing an example of the data structure of the model-specific model list information 117. The model-specific model list information 117 stores in advance candidate models that can be installed for each model and part, and is composed of product model 117a, part 117b, and model 117c. The product model 117a indicates a model name that identifies the model, such as "Model A." The part 117b indicates a part name that identifies the part, such as "Motor." The model 117c indicates a model name that identifies the model, such as "M01." For example, if the product model 117a in FIG. 8 is "Model A" and the part 117b is "Motor," the candidates for the model 117c that can be installed are "M01," "M02," "M03," and "M04." A model not listed here indicates that the model or part cannot be installed.

[0026] FIG. 9 is a diagram showing an example of the data structure of model prediction specification information 118. Model prediction specification information 118 stores specification information that serves as a clue for predicting the mounting probability for each part, and is composed of part 118a and specification item 118b. Part 118a indicates a part name that identifies the part, such as "motor." Specification item 118b indicates a specification item name that identifies specification information, such as "load weight." For example, when part 118a in FIG. 9 is a "motor," specification information that serves as a clue for predicting the mounting probability is indicated by specification item 118b, which indicates that "load weight," "speed," etc.

[0027] FIG. 10 is a diagram showing an example of the data structure of model-specific attribute information 119. Model-specific attribute information 119 stores attribute information for each model in advance, and is composed of part 119a, model 119b, attribute item 119c, and attribute value 119d. Part 119a indicates a part name that identifies the part, such as "motor." Model 119b indicates a model name that identifies the model, such as "M01." Attribute item 119c indicates the item name of attribute information for each model, such as "motor capacity" and "voltage." Attribute value 119d indicates an attribute value of the attribute item for each model, such as "2." For example, if part 119a in FIG. 10 is "motor," model 119b is "M01," and attribute item 119c is "motor capacity," the attribute value 119d is "2."

[0028] 11 is a diagram showing an example of the data structure of the learning data information 120. The learning data information 120 stores learning data for predicting the installation probability for each model, and is composed of a case number 120a, a product model 120b, a part 120c, a model 120d, specification information (load weight) 120e, specification information (speed) 120f, attribute information (motor capacity) 120g, attribute information (voltage) 120h, a provided specification lower limit value (load weight) 120i, a provided specification upper limit value (load weight) 120j, a provided specification lower limit value (speed) 120k, a provided specification upper limit value (speed) 120l, a provided specification lower limit value (travel) 120m, a provided specification upper limit value (travel) 120n, and an installation probability 120o. The case number 120a indicates a case name, such as "001," that identifies the case. Product model 120b indicates a model name that identifies the model, such as "Model A." Part 120c indicates a part name that identifies the part, such as "Motor." Model 120d indicates the model name of the part to be installed for each project, such as "M01." Specification information (load weight) 120e and specification information (speed) 120f indicate specification values ​​for each specification item for project No. 120a, such as "500" and "45." Attribute information (motor capacity) and attribute information (voltage) indicate attribute values ​​for each attribute item of model 120d adopted for project No. 120a, such as "2" and "100." The provided specification lower limit (load weight) 120i, the provided specification upper limit (load weight) 120j, the provided specification lower limit (speed) 120k, the provided specification upper limit (speed) 120l, the provided specification lower limit (distance) 120m, and the provided specification upper limit (distance) 120n indicate the provided specification range for each product model 120b for each project number 120a, such as "300", "2,000", "20", "100", "0", and "100". The installation probability 120o indicates whether or not the model 120d will be installed for each project number 120a, such as "100%" or "0%". If it will be installed, it indicates "100%," and if it will not be installed, it indicates "0%".For example, the case number 120a is "001", the product model 120b is "Model A", the part 120c is "Motor", the model number 120d is "M01", the specification information (load weight) 120e is "500", the specification information (speed) 120f is "45", the attribute information (motor capacity) 120g is "2", the attribute information (voltage) 120h is "100", and the provided specification lower limit (load weight) 120 If i is "300", the upper limit of the specification provided (load weight) 120j is "2,000", the lower limit of the specification provided (speed) 120k is "20", the upper limit of the specification provided (speed) 120l is "100", the lower limit of the specification provided (travel) 120m is "0", and the upper limit of the specification provided (travel) 120n is "100", then the loading probability 120o is "100%, meaning that it will be loaded".

[0029] FIG. 12 is a diagram illustrating an example of the data structure of the mounting probability distribution information 121. The mounting probability distribution information 121 stores the predicted results of the mounting probability for each model, and is composed of a product model 121a, a part 121b, a model 121c, specification information (load weight) 121d, specification information (speed) 121e, attribute information (motor capacity) 121f, attribute information (voltage) 121g, a provided specification lower limit (load weight) 121h, a provided specification upper limit (load weight) 121i, a provided specification lower limit (speed) 121j, a provided specification upper limit (speed) 121k, a provided specification lower limit (travel) 121l, a provided specification upper limit (travel) 121m, and a mounting probability 121n. The product model 121a indicates a model name, such as "Model C," that identifies the model. The part 121b indicates a part name, such as "motor," that identifies the part. The model 121c indicates the model name of the part, such as "M01." The specification information (load weight) 121d and the specification information (speed) 121e indicate specification values ​​for each specification item, such as "300" and "40." The attribute information (motor capacity) 121f and the attribute information (voltage) 121g indicate attribute information for each model 121c, such as "2" and "100." The provided specification lower limit (load weight) 121h, the provided specification upper limit (load weight) 121i, the provided specification lower limit (speed) 121j, the provided specification upper limit (speed) 121k, the provided specification lower limit (stroke) 121l, and the provided specification upper limit (stroke) 121m indicate the provided specification range for each product model 121a, such as "300," "2,000," "40," "110," "0," and "200." The installation probability 121n indicates the installation probability for the model 121c, such as "72%." For example, if the product model 121a is "Model C", the part 121b is "Motor", the model number 121c is "M01", the specification information (load weight) 121d is "300", the specification information (speed) 121e is "40", the attribute information (motor capacity) 121f and the attribute information (voltage) 121g are "2" and "100", the provided specification lower limit value (load weight) 121h is "300", the provided specification upper limit value (load weight) 121i is "2,000", the provided specification lower limit value (speed) 121j is "40", the provided specification upper limit value (speed) 121k is "110", the provided specification lower limit value (travel) 121l is "0", and the provided specification upper limit value (travel) 121m is "200", this indicates that the installation probability is "72%".

[0030] Returning to the description of the type prediction device 101 in FIG. 1, FIG. 1 conceptually illustrates the functions executed by the calculation unit 130 as an information acquisition unit 131, a learning data generation unit, a prediction model generation unit, and an installation probability calculation unit 134. These functions are realized by a program stored in a memory (not shown). The program may be provided by being pre-installed in a ROM or the like, or may be provided or distributed by being recorded on a computer-readable recording medium such as a CD-ROM in the form of an installable or executable file. Furthermore, the program may be stored on a computer, such as a database 104 connected to the network 102, and provided or distributed by being downloaded via the network 102.

[0031] The information acquisition unit 131 acquires each piece of information stored in the database 104 via the network 102, and also acquires each piece of information necessary for calculations in the learning data generation unit 132, the prediction model generation unit 133, and the installation probability calculation unit 134 from the memory unit 110, and stores the calculation results in the memory unit 110.

[0032] The learning data generation unit 132 generates learning data using, for example, case-specific specification information 113, case-specific mounted model information 114, case-specific model information 115, model-specific provided specification range information 116, and model-specific attribute information 119. Specifically, the learning data is generated using, for example, (1) the product model, specification values ​​of each specification item, and model of each part recorded for each past case, (2) attribute values ​​of multiple attribute items set in advance for each model, and (3) lower and upper limit values ​​of multiple specification items set in advance for each model. The learning data generated by the learning data generation unit 132 is stored in the storage unit 110 as learning data information 120.

[0033] The prediction model generation unit 133 generates a prediction model that predicts the installation probability for each model, type, and specification value of each specification item based on the learning data information 120.

[0034] The mounting probability calculation unit 134 predicts the mounting probability by inputting the search target information 111, the model-by-model list information 117, etc., into the prediction model generated by the prediction model generation unit 133. Specifically, in addition to the prediction model, the mounting probability for each model and specification value of each specification item for the new model is calculated based on (1) the model (new model) and parts selected by the user as the prediction target and for which no past performance has existed, (2) the lower and upper limits of each specification item of the new model, and (3) a list of models that can be mounted for each part. The mounting probability calculated by the mounting probability calculation unit 134 is stored in the storage unit 110 as mounting probability distribution information 121.

[0035] The input unit 140 is displayed and operated on a screen, for example, and accepts information input to the model prediction device 101 via an input device such as a keyboard, touch panel, or mouse. The output unit 150 outputs screen information including the results of predetermined processing to an output device such as a display.

[0036] 13 is a diagram showing a screen displaying the selection of search target information and the analysis (prediction) results, as an example of a screen displayed by the output unit 150. The mounting probability distribution analysis screen 1500 includes an input information area 1501 where the user inputs the search target information 111, and an analysis result area 1502 that displays the analysis results, which will be described later. The input information area 1501 is composed of a model field 150a, a parts field 150b, and a target model / parts search button 150c, while the analysis result area 1502 is composed of a parts field 150d, a specification item field 150e, a model field 150f, a specification combination number field 150g, a loading weight field 150h, a speed field 150i, and a mounting probability field 150j.

[0037] The model column 150a is a column where the user registers the model they want to check. The parts column 150b is a column where the user registers the parts they want to check. The target model / parts search button 150c is a button for searching the analysis results of the model and parts registered in the model column 150a and parts column 150b described above. The parts column 150d shows the part name registered in the parts column 150b. The specification item column 150e displays specification items that are effective when predicting the model of the part registered in the parts column 150b. The model column 150f shows the model information of the model registered in the model column 150a. The specification combination number column 150g shows the number of specification combinations. The load weight column 150h and speed column 150i display all combinations of specification information within the range of specifications provided for the model selected in the model column 150a for each specification item displayed in the specification item column 150e. The installation probability column 150j displays the predicted installation probability of the model column 150f for each specification combination number column 150g selected by the user.

[0038] Next, the flow of processing executed by the model prediction device 101 according to this embodiment will be described with reference to the flowchart of FIG.

[0039] The following series of processes is based on the premise that the database 104 stores case information 112, case-specific specification information 113, case-specific mounted model information 114, case-specific model information 115, model-specific specification range information 116, model-specific model list information 117, model-predicted specification information 118, and model-specific attribute information 119, and the process is started, for example, when the model prediction device 101 is called from the user terminal 103.

[0040] First, in step S1, the input unit 140 accepts search target information 111 from the user and stores it in the storage unit 110. When the user selects a search target, an input information area 1501 in the analysis screen 1500 of FIG. 13 is used. Specifically, the user inputs information into the model field 150a and the parts field 150b and taps the target model / parts search button 150c, thereby generating the search target information 111. In the search target information 111 of FIG. 2, "motor" is selected as the part 111a and "model C" is selected as the model 111b, and therefore the following description will be given assuming that the same parts and model have been selected.

[0041] Next, in step S2, the information acquisition unit 131 extracts from the storage unit 110 the model of a part previously installed as the per-project model information 115 and the model of the target part previously installed as the per-project installation model information 114. For example, in the case of FIG. 6, the model of a part previously installed would be "Model A," "Model B," etc., and in the case of FIG. 5, if the target part is a "motor," the model of a part previously installed would be "M01," "M02," "M03," etc. The following processes of S3 and S4 are repeated for all combinations of model and model extracted in this way.

[0042] In step S3, the information acquisition unit 131 extracts, from the information stored in the storage unit 110, the provided specification range information corresponding to the model extracted in step S2, the attribute information corresponding to the model extracted in step S2, and the actual specification information adopted in past cases, and the learning data generation unit 132 stores the extracted information in the learning data information 120. The processing in step S3 will be specifically described below.

[0043] First, the information acquisition unit 131 extracts the cases to be installed for each model from the case-specific model information 115 in Fig. 6, and the learning data generation unit 132 stores them in the case number 120a and the product model 120b in the learning data information 120 in Fig. 11. Specifically, in Fig. 6, when the product model 115b is "Model A," cases with case numbers 115a of "001," "003," and "004" are extracted. Next, in the learning data information 120 in Fig. 11, "001," "003," and "004" are stored in the case number 120a, and "Model A" is stored in the product model 120b.

[0044] Furthermore, the learning data generation unit 132 stores information about the target part selected by the user in part 120c in the learning data information 120 of Fig. 11. Specifically, since the part selected by the user this time is "motor," "motor" is stored in part 120c in the learning data information 120 of Fig. 11.

[0045] Next, the information acquisition unit 131 extracts the type to be mounted as the target part for each project from the project-specific mounting type information 114 in Fig. 5, and the learning data generation unit 132 stores it in the type 120d in the learning data information 120 in Fig. 11. Specifically, if the project number 120a is "001" and the part 120c is "motor" in the learning data information 120 in Fig. 11, when the project number 114a is "001" and the part 114b is "motor" in the project-specific mounting type information 114 in Fig. 5, the type 114c is "M01", and therefore "M01" is stored in the type 120d in the learning data information 120 in Fig. 11.

[0046] Next, the information acquisition unit 131 extracts specification information that serves as a clue for prediction from the model prediction specification information 118 in Fig. 9, and the learning data generation unit 132 adds this as a column name of the specification information in the learning data information 120 in Fig. 11. Specifically, in the learning data information 120 in Fig. 11, the part 120c whose case number 120a is "001" is "motor." Here, when the part 118a is "motor" in the model prediction specification information 118 in Fig. 9, the specification items 118b are "load weight" and "speed," and therefore columns of "specification information (load weight) 120e" and "specification information (speed) 120f" are added to the specification information in the learning data information 120 in Fig. 11.

[0047] Next, the information acquisition unit 131 extracts actual specification information adopted in past cases from the case-specific specification information 113 of Fig. 4, and the learning data generation unit 132 stores the specification values ​​of the specification information (load weight) 120e and the specification information (speed) 120f in the learning data information 120 of Fig. 11. Specifically, in the case-specific specification information 113 of Fig. 4, when the case number 113a is "001" and the specification item 113b is "load weight", the specification value 113c is "500", so when the case number 120a is "001" in the learning data information 120 of Fig. 11, the specification information (load weight) 120e becomes "500". Similarly, in the case-specific specification information 113 of FIG. 4, if the case number 113a is "001" and the specification item 113b is "speed", the specification value 113c is "45", so in the learning data information 120 of FIG. 11, if the case number 120a is "001", the specification information (speed) 120f is "45".

[0048] Next, the information acquisition unit 131 extracts attribute information corresponding to the model from the model-specific attribute information 119 of FIG. 10, and the learning data generation unit 132 stores attribute values ​​in attribute information (motor capacity) 120g and attribute information (voltage) 120h in the learning data information 120 of FIG. 11. Specifically, in the learning data information 120 of FIG. 11, when the case-specific number 120a is "001," the model 120d is "M01." Here, in the model-specific attribute information 119 of FIG. 10, when the model 119b is "M01" and the attribute item 119c is "motor capacity," the attribute value is "2." Therefore, in the learning data information 120 of FIG. 11, when the case number 120a is "001," the attribute information (motor capacity) 120g becomes "2." Similarly, in the model-specific attribute information 119 of FIG. 10, if the model 119b is "M01" and the attribute item 119c is "voltage", the attribute value is "100", so in the learning data information 120 of FIG. 11, if the case number 120a is "001", the attribute information (voltage) 120h becomes "100".

[0049] Next, the information acquisition unit 131 extracts the provided specification range information corresponding to the model from the model-specific provided specification range information 116 in Fig. 7, and the learning data generation unit 132 stores the specification ranges in the provided specification lower limit value (load weight) 120i to the provided specification upper limit value (travel) 120n in the learning data information 120 in Fig. 11. Specifically, in the learning data information 120 in Fig. 11, when the case number 120a is "001", the product model 120b is "model A". Here, in the model-specific provided specification range information 116 of Figure 7, if the product model 116a is "Model A" and the specification item 116b is "Load weight", the lower limit value 116c is "300" and the upper limit value 116d is "2,000", so in the learning data information 120 of Figure 11, if the case number 120a is "001" and the product model 120b is "Model A", the provided specification lower limit value (load weight) 120i is "300" and the provided specification upper limit value (load weight) 120j is "2,000". Similarly, in the model-specific provided specification range information 116 of FIG. 7, when the product model 116a is "model A" and the specification item 116b is "speed," the lower limit 116c is "20" and the upper limit 116d is "100," so in the learning data information 120 of FIG. 11, when the case number 120a is "001" and the product model 120b is "model A," the provided specification lower limit (speed) 120k is "20" and the provided specification upper limit (speed) 120l is "100." Similarly, in the model-specific provided specification range information 116 of FIG. 7, when the product model 116a is "Model A" and the specification item 116b is "Journey", the lower limit 116c is "0" and the upper limit 116d is "100". Therefore, in the learning data information 120 of FIG. 11, when the case number 120a is "001" and the product model 120b is "Model A", the provided specification lower limit (journey) 120m is "0" and the provided specification upper limit (journey) 120n is "100".

[0050] Finally, the learning data generation unit 132 stores "100%" in the loading probability 120o in the learning data information 120 in Fig. 11. This is because the loading probability 120o for all jobs that have actually been shipped out can be considered to be 100%. This concludes the explanation of step S3.

[0051] In step S4, the learning data generation unit 132 stores, as dummy data, the specification range information corresponding to the model extracted in step S2, the attribute information corresponding to the model extracted in step S2, and the specification information that has not been adopted in past cases in the learning data information 120. The processing in step S4 will be specifically described below.

[0052] In step S3 described above, for example, for the case number 120a "001" in the learning data information 120 of FIG. 11, "Model A" is stored as the product model 120b, and "500" and "45" are stored as the specification information (load weight) 120e and the specification information (speed) 120f, respectively. Meanwhile, in step S4, specification values ​​within the ranges of the lower and upper limits of the specification items of the model (Model A) (the range of 300 to 2,000 for load weight and 20 to 100 for speed) that have not been used in past cases are selected and stored as dummy cases in the learning data information 120. For example, the learning data generation unit 132 stores the case number 120a with the specification information (load weight) 120e of "300" and the specification information (speed) 120f of "20" as "dummy 001" in the learning data information 120 of FIG. 11, as specification values ​​for which no past performance exists. Here, the product model 120b, part 120c, and model 120d are "Model A," "Motor," and "M01." Furthermore, the information from the attribute information (motor capacity) 120g to the provided specification upper limit value (stroke) 120n stores the same data as the case number 120a "001," in which the product model 120b, part 120c, and model 120d have the same values. Finally, the learning data generation unit 132 stores "0%" in the mounting probability 120o in the learning data information 120 in FIG. 11. This concludes the explanation of step S4.

[0053] In step S5, it is confirmed whether the processes in steps S3 and S4 have been completed for all the combinations of model and type extracted in step S2, and if there are any combinations that have not been processed, the process returns to step S3. This process is repeated until all the combinations have been processed, at which point the learning data information 120 is completed and stored in the storage unit 110.

[0054] Next, in step S6, the prediction model generation unit 133 generates a prediction model that predicts the installation probability for each model, type, and specification value of each specification item based on the learning data information 120. Specifically, the prediction model generation unit 133 uses information from specification information (load weight) 120e to provided specification upper limit value (travel) 120n in the learning data information 120 in FIG. 11 as explanatory variables and installation probability 120o as an objective variable, and learns the relationship between the explanatory variables and the objective variable. For example, the relationship between the explanatory variables and the objective function can be calculated probabilistically using any method, such as a Gaussian regression model, Bayesian linear regression, or kernel ridge regression, and the specific method is not limited.

[0055] Next, in step S7, the information acquisition unit 131 extracts candidate models that can be installed on the target parts of the target new model from the model-specific model list information 117 of FIG. 8. For example, in the model-specific model list information 117 of FIG. 8, if the product model 117a is "model C" and the part 117b is "motor," "M01," "M03," "M11," and the like are extracted as the installable models 117c. Note that the models "M01" and "M03" are existing models that can be installed on models other than "model C," while the model "M11" is a new model that can be installed only on "model C." The processing of the following step S8 is repeated for all of the models extracted in this way.

[0056] In step S8, the installation probability calculation unit 134 generates the input variable information 122 shown in Fig. 15 using the attribute information corresponding to the model extracted in step S7, the provided specification range information corresponding to the new model, and the expected combination of specification values ​​within the provided specification range. The processing in step S8 will be specifically described below.

[0057] First, the mounting probability calculation unit 134 stores the new model and target part selected by the user and the model candidates extracted in step S7 in the product model 121a, part 121b, and model 121c, respectively, in the input variable information 122 of Fig. 15. For example, "model C" is stored in the product model 121a, "motor" in the part 121b, and "M01" in the model 121c.

[0058] Next, the information acquisition unit 131 extracts specification information that serves as a clue for prediction from the model prediction specification information 118 in Fig. 9, and the mounting probability calculation unit 134 adds it as a column name of the specification information in the input variable information 122 in Fig. 15. Specifically, in the model prediction specification information 118 in Fig. 9, when the part 118a is a "motor", the specification item 118b is a "load weight" and a "speed", so columns of "specification information (load weight) 121d" and "specification information (speed) 121e" are added to the input variable information 122 in Fig. 15.

[0059] Next, the mounting probability calculation unit 134 stores multiple expected combinations of specification values ​​within the provided specification range as specification information in the input variable information 122 of Fig. 15. Here, according to the model-specific provided specification range information 116 of Fig. 7, for "Model C," which is a new model, the lower limit value of the specification item "load weight" is "300" and the upper limit value is "2,000," and the lower limit value of the specification item "speed" is "40" and the upper limit value is "110." Therefore, (2,000-300+1) x (110-40+1) = 120,771 combinations of specification values ​​are stored in the specification information (load weight) 121d and the specification information (speed) 121e of the input variable information 122 of Fig. 15. For example, the mounting probability calculation unit 134 stores data in which the specification information (load weight) 121d is "300" and the specification information (speed) 121e is "40" as one of the expected combinations of specification values ​​in the first position of the model "M01" in the input variable information 122 of FIG. 15. Thereafter, the mounting probability calculation unit 134 stores data in which the specification information (load weight) 121d is "300" and the specification information (speed) 121e is "41" as another expected combination of specification values ​​in the second position of the model "M01" in the input variable information 122 of FIG. 15. Thereafter, data is stored in a similar manner up to the 120,771st position of the model "M01". Note that in this embodiment, the specification value combinations are assumed in increments of "1", but the increment size is not limited to "1" and may be, for example, "5" or "10".

[0060] Next, the information acquisition unit 131 extracts attribute information corresponding to the model from the model-specific attribute information 119 of Fig. 10, and the mounting probability calculation unit 134 stores attribute values ​​in attribute information (motor capacity) 121f and attribute information (voltage) 121g in the input variable information 122 of Fig. 15. Specifically, in the model-specific attribute information 119 of Fig. 10, if the model 119b is "M01" and the attribute item 119c is "motor capacity," the attribute value is "2," so that the attribute information (motor capacity) 121f in the input variable information 122 of Fig. 15 becomes "2." Similarly, in the model-specific attribute information 119 of Fig. 10, if the model 119b is "M01" and the attribute item 119c is "voltage," the attribute value is "100," so that the attribute information (voltage) 121g in the input variable information 122 of Fig. 15 becomes "100."

[0061] Next, the information acquisition unit 131 extracts provided specification range information corresponding to the model from the model-specific provided specification range information 116 of Fig. 7, and the mounting probability calculation unit 134 stores the specification range in the provided specification lower limit value (load weight) 121h to the provided specification upper limit value (travel) 121m in the input variable information 122 of Fig. 15. Specifically, in the model-specific provided specification range information 116 of Fig. 7, when the product model 116a is "model C" and the specification item 116b is "load weight", the lower limit value 116c is "300" and the upper limit value 116d is "2,000", so in the input variable information 122 of Fig. 15, the provided specification lower limit value (load weight) 121h is "300" and the provided specification upper limit value (load weight) 121i is "2,000". 7, when the product model 116a is "model C" and the specification item 116b is "speed," the lower limit 116c is "40" and the upper limit 116d is "110," so that the provided specification lower limit (speed) 121j is "40" and the provided specification upper limit (speed) 121k is "110" in the input variable information 122 of FIG. 15. Similarly, when the product model 116a is "model C" and the specification item 116b is "travel," the lower limit 116c is "0" and the upper limit 116d is "200," so that the provided specification lower limit (travel) 121l is "0" and the provided specification upper limit (travel) 121m is "200" in the input variable information 122 of FIG. 15. This concludes the explanation of step S8.

[0062] In step S9, it is confirmed whether the processing of step S8 has been completed for all of the types extracted in step S7, and if there are any types that have not been processed, the process returns to step S8. This processing is repeated, and when the processing for all types has been completed, the input variable information 122 is completed.

[0063] Next, in step S10, the mounting probability calculation unit 134 calculates the mounting probability for each model candidate that can be mounted on the new model and the specification value of each specification item, based on the completed input variable information 122 and the prediction model generated in step S6. Specifically, the mounting probability is calculated using the prediction model generated in step S6, with information from the specification information (loading weight) 121d to the provided specification upper limit value (travel) 121m in the input variable information 122 in Fig. 15 as explanatory variables, and the calculation result is stored as the mounting probability 121n in the mounting probability distribution information 121 in Fig. 12. For example, in the first row of the mounting probability distribution information 121 in FIG. 12, the product model 121a is "Model C", the part 121b is "Motor", the model 121c is "M01", the specification information (loading weight) 121d is "300", the specification information (speed) 121e is "40", the attribute information (motor capacity) 121f is "2", the attribute information (voltage) 121g is "100", and the provided specification lower limit When (load weight) 121h is "300", the provided specification upper limit value (load weight) 121i is "2,000", the provided specification lower limit value (speed) 121j is "40", the provided specification upper limit value (speed) 121k is "110", the provided specification lower limit value (travel) 121l is "0", and the provided specification upper limit value (travel) 121m is "200", the loading probability 121n is "72%".

[0064] Next, in step S11, the output unit 150 transmits the mounting probability distribution information 121 calculated in step S10 to the user terminal 103 via the network 102. The analysis result area 1502 in Fig. 13 is configured to display model-predicted specification information 1503 and mounting probability distribution information 1504. The model-predicted specification information 1503 is displayed in list format and is composed of a parts column 150d and a specification item column 150e. The mounting probability distribution information 1504 is displayed in list format and is composed of a model column 150f, a specification combination number column 150g, a loading weight column 150h, a speed column 150i, and a mounting probability column 150j.

[0065] Here, the parts column 150d displays the part names registered in the parts column 150b. The specification item column 150e extracts records whose parts 118a column in the model predicted specification information 118 matches the parts in the parts column 150d, and displays the specification item information in the extracted specification item column 118b. The model 150f displays model 117c, whose model column 150a and parts column 150b registered by the user match the product model 117a and parts 117b in the model-specific model list information 117. The specification combination number column 150g displays the combination number of the specification information shown in the specification item column 150e. The load weight column 150h and speed column 150i show the specification values ​​of the specification items shown in the specification item column 150e, and all specification values ​​within the range of provided specifications for the model registered in the model column 150a are displayed. In the loading probability column 150j, based on the loading probability distribution information 121, the predicted results of the loading probability for the specification values ​​in the loaded weight column 150h and the speed column 150i are displayed.

[0066] As described above, according to this embodiment, it is possible to predict the installation probability for each type of installable part, even for new models for which there is no information on past projects. Furthermore, the installation probability can be predicted not only for existing models but also for new models. Furthermore, if the range of specifications provided for a new model is expanded compared to existing models, the installation probability for existing models may differ between the existing model and the new model. However, this embodiment makes it possible to predict the installation probability for existing models, including the expanded range of specifications.

[0067] In the above-described embodiment, the learning data generation unit 132, the prediction model generation unit 133, and the mounting probability calculation unit 134 acquire predetermined information from the storage unit 110 and store calculation results in the storage unit 110 via the information acquisition unit 131. However, the learning data generation unit 132, the prediction model generation unit 133, and the mounting probability calculation unit 134 may be configured to exchange information directly with the storage unit 110 without going through the information acquisition unit 131.

[0068] In the above embodiment, the type prediction device 101 not only calculates the installation probability of a type but also generates learning data and a prediction model, but the type prediction device 101 may also only calculate the installation probability of a type. In this case, the learning data and the prediction model are generated in advance outside the type prediction device 101 and stored in the database 104, and the type prediction device 101 receives the prediction model from the database 104 when calculating the installation probability of a type. [Explanation of symbols]

[0069] 100... type prediction system, 101... type prediction device, 102... network, 103... user terminal, 104... database, 110... storage unit, 130... calculation unit, 140... input unit, 150... output unit, 111: search target information, 112: case information, 113: case-specific specification information, 114: case-specific installation type information, 132... learning data generation unit, 133... prediction model generation unit, 134: installation probability calculation unit

Claims

1. A model prediction device that predicts future demand for parts for each model based on past performance for products including models in which lower and upper limits for each specification item are set in advance for each model and different models can be installed in predetermined parts, The product model, specification values ​​for each specification item, and model number of each part recorded for each past case, Attribute values ​​of a plurality of attribute items that are preset for each model; Lower and upper limits for multiple specification items that are preset for each model, a training data generation unit that generates training data using a prediction model generation unit that generates a prediction model that predicts the installation probability for each type, model, and specification value of each specification item based on the learning data; a mounting probability calculation unit that calculates a mounting probability for each model and specification value of each specification item for a new model based on lower and upper limit values ​​of each specification item of a new model for which no past performance exists, models that can be mounted for each part, and the prediction model; A model prediction device comprising:

2. 2. The model prediction device according to claim 1, The model prediction device is characterized in that the learning data generation unit generates the learning data by also using information on specification items that are preset for each part and serve as clues for prediction.

3. 3. The model prediction device according to claim 2, If the product is an elevator and the part is a motor, The model prediction device is characterized in that the specification items that serve as clues for prediction include load weight and speed.

4. 2. The model prediction device according to claim 1, the learning data generation unit also generates, as the learning data, dummy data corresponding to specification values ​​that are within a range between a lower limit value and an upper limit value of a specification item of a predetermined model and that have no past performance data; The type prediction device is characterized in that the installation probability calculation unit also calculates the installation probability corresponding to a specification value that is within the range of a lower limit value and an upper limit value of a specification item of a predetermined model and has no past performance.

5. 5. The model prediction device according to claim 4, A model prediction device characterized in that models that can be installed in a specified part of a new model include existing models that can be installed in models other than the new model, and new models that can be installed only in the new model.

6. 6. The model prediction device according to claim 5, The product is an elevator, The new model's specifications, such as load weight or speed, have a different range from other models. a mounting probability calculation unit that, when calculating the mounting probability of the existing model motor for a new model, also calculates the mounting probability corresponding to the specification values ​​included in the different specification ranges.

7. A model prediction method for predicting future demand for parts for each model based on past performance for a product including a model in which lower and upper limits for each specification item are set in advance for each model and different models can be installed in a predetermined part, comprising: The product model, specification values ​​for each specification item, and model number of each part recorded for each past case, Attribute values ​​of a plurality of attribute items that are preset for each model; Lower and upper limits for multiple specification items that are preset for each model, generating training data using generating a prediction model that predicts the installation probability for each model, type, and specification value of each specification item based on the learning data; a step of calculating the probability of installation for each model and specification value of each specification item of a new model based on the lower limit value and the upper limit value of each specification item of a new model for which no past performance exists, the model that can be installed for each part, and the prediction model; A model prediction method comprising:

Images