JP2023009811A - Calculator system, demand prediction method of item, and program - Google Patents

Abstract

To produce a model having high prediction accuracy using a loss function suitable for learning of the model while automatically setting the loss function.SOLUTION: A calculator system manages a plurality of loss functions and a history concerning demand for an item and executes: first processing which accepts information on a condition to be satisfied by an evaluation index; second processing which produces an integrated loss function using a plurality of loss functions multiplied by a weight; third processing which calculates the evaluation index on the basis of demand prediction of the item, obtained by producing a model by using machine learning with the integrated loss function and the history and inputting the history into the model; fourth processing which determines whether or not the condition is satisfied on the basis of the accepted information and the evaluation index; fifth processing which, when the condition is not satisfied, updates the weight on the basis of the evaluation index and executes the third processing; and sixth processing which, when the condition is satisfied, outputs the demand prediction of the item.SELECTED DRAWING: Figure 1

Description

The present invention relates to systems and methods for predicting demand for items.

In the logistics industry, systems are used to forecast item demand in order to plan item inventory, replenishment, and the like. There are multiple methods of demand forecasting because demand units, fluctuation cycles, etc. differ depending on the business field. Therefore, the user needs to select the optimum demand forecast method based on the shipping pattern of the target business field. A technique described in Patent Document 1 is known for this problem.

In Patent Document 1, "The method automatic selection device uses a product sales record storage unit 15 that stores the sales record during the verification period of the product, and a demand prediction method based on the sales record of the product during the verification period. A predicted demand amount calculation unit 11 that calculates the predicted demand amount of, a theoretical inventory calculation unit 12 that calculates the theoretical inventory during the verification period based on the predicted demand amount using an inventory simulation method, the theoretical inventory and the An order forecast error output unit 13 for calculating an order forecast error for each set of a demand forecast method and an inventory simulation method from the difference from the actual sales of the product, and a demand forecast method and inventory corresponding to the product based on the order forecast error. and a method evaluation unit 14 that selects a set of simulation methods.”.

JP 2010-86278 A

That is, in Patent Document 1, the optimum combination for each product (item) is selected from a large number of prediction methods and simulation methods. On the other hand, it is conceivable to use a model generated by machine learning. Models are typically trained using a loss function. The loss function varies depending on the problem to be addressed, the target of prediction, the parameters of the model, and so on. The definition of the loss function has a great influence on the prediction accuracy of the model. Therefore, we need to define an appropriate loss function.

In the case of a model for predicting item demand, various loss functions can be considered depending on the characteristics of the item itself, the environment in which the item is handled, and so on. Therefore, it is difficult to define an appropriate loss function, and the number of man-hours required increases the cost of generating the model.

The present invention realizes a system and method for automatically setting a loss function suitable for model learning and using the loss function to generate a model with high prediction accuracy.

A representative example of the invention disclosed in the present application is as follows. That is, a computer system comprising at least one computer, which manages a plurality of loss functions and a history of item demand, wherein the at least one computer stores information on conditions to be satisfied by the item demand evaluation index. Demand for the item is predicted by a first process of receiving, a second process of generating an integrated loss function using the multiple weighted loss functions, and machine learning using the integrated loss function and the history. a third process for calculating the evaluation index based on the demand forecast for the item obtained by inputting the history into the model; and based on the received information and the evaluation index, a fourth process for determining whether or not the condition is satisfied; and when it is determined that the condition is not satisfied, at least one of the loss functions included in the integrated loss function is multiplied based on the evaluation index. a fifth process of updating the weights and executing the third process; and a sixth process of outputting a demand forecast for the item if it is determined that the condition is satisfied.

According to the present invention, a computer system can automatically set a loss function suitable for model learning and generate a model with high prediction accuracy. Problems, configurations, and effects other than those described above will be clarified by the following description of the embodiments.

1 is a diagram illustrating an example of a configuration of a demand forecasting system of Example 1; FIG. FIG. 2 is a diagram illustrating the details of the functional configuration of the demand forecasting system of Example 1; FIG. 10 is a diagram showing an example of the data structure of demand record management information of Example 1; 4 is a diagram showing an example of the data structure of element loss function information in Example 1. FIG. 4 is a diagram showing an example of the data structure of integrated loss function information of Example 1. FIG. 4 is a diagram illustrating an example of the data structure of evaluation index information in Example 1; FIG. 7 is a flowchart illustrating an example of processing executed by an integrated loss function generation unit according to the first embodiment; 6 is a flowchart illustrating an example of processing executed by a demand forecasting unit according to the first embodiment; 7 is a flowchart illustrating an example of processing executed by an evaluation index calculation unit according to the first embodiment; 8 is a flowchart illustrating an example of processing executed by a function weight updating unit according to the first embodiment; 7 is a flowchart illustrating an example of processing executed by an integrated loss function updating unit according to the first embodiment; FIG. 5 is a diagram showing an example of a screen displayed by the display unit of Example 1; FIG. 5 is a diagram showing an example of a screen displayed by the display unit of Example 1; FIG. 12 is a diagram illustrating the details of the functional configuration of the demand forecasting system of Example 2; FIG. 10 is a diagram showing an example of a screen displayed by the input unit of Example 2;

Hereinafter, embodiments of the present invention will be described with reference to the drawings. However, the present invention should not be construed as being limited to the contents of the examples described below. Those skilled in the art will easily understand that the specific configuration can be changed without departing from the idea or gist of the present invention.

In the configurations of the invention described below, the same or similar configurations or functions are denoted by the same reference numerals, and overlapping descriptions are omitted.

The notations such as “first”, “second”, “third”, etc. in this specification and the like are attached to identify the constituent elements, and do not necessarily limit the number or order.

The position, size, shape, range, etc. of each component shown in the drawings may not represent the actual position, size, shape, range, etc. in order to facilitate understanding of the invention. Therefore, the present invention is not limited to the positions, sizes, shapes, ranges, etc. disclosed in the drawings and the like.

FIG. 1 is a diagram showing an example of the configuration of a demand forecasting system according to the first embodiment.

The demand forecast system is a computer system composed of at least one computer 100 . The computer 100 has a processor 110 , an IO interface 111 , a network interface 112 , a main storage device 113 and a secondary storage device 114 . Each piece of hardware is connected to each other via a bus.

The processor 110 executes programs stored in the main memory device 113 . The processor 110 operates as a functional unit (module) that implements a specific function by executing processing according to a program. In the following description, when processing is described with a functional unit as the subject, it means that the processor 110 is executing a program that implements the functional unit.

The IO interface 111 is an interface for connecting with an external device. The computer 100 connects with the input device 101 and the output device 102 via the IO interface 111 . The input device 101 is a keyboard, mouse, touch panel, or the like. The output device 102 is a display, a printer, or the like.

A network interface 112 is an interface for connecting to an external device via the network 103 . The network 103 is a WAN (Wide Area Network), a LAN (Local Area Network), or the like. The connection format of the network 103 may be wired or wireless.

The main storage device 113 stores programs executed by the processor 110 and data used by the programs. The main memory 113 also includes a work area that is temporarily used by the program. The main storage device 113 is, for example, a DRAM (Dynamic Random Access Memory) or the like.

The main storage device 113 stores programs that implement the input unit 120 , the prediction unit 121 and the display unit 122 . Although the main storage device 113 stores programs (not shown) such as an OS (Operating System), they are omitted in this embodiment.

The input unit 120 receives various inputs from the user via the input device 101 or the network 103 . The prediction unit 121 predicts item demand and outputs a prediction result. The time range for forecasting demand for an item can be, for example, but not limited to, years, months, and weeks. It is assumed that the prediction time range is set in advance. The display unit 122 presents prediction results and the like to the user via the output device 102 . The computer 100 receives data from a user terminal communicably connected via the network 103, thereby accepting input from the user and transmitting an output to be output by the output device 102 to the user terminal. By doing so, the data may be output to the user terminal.

The secondary storage device 114 permanently stores data. The secondary storage devices 114 are, for example, HDDs (Hard Disk Drives) and SSDs (Solid State Drives).

The secondary storage device 114 stores demand performance management information 130 and element loss function information 131 . Although the secondary storage device 114 stores a demand forecast model, it is omitted in this embodiment.

The actual demand management information 130 stores data (shipment history) regarding past shipment amounts of items. The shipping history includes values related to the amount of items shipped during the measurement period. The measurement period is, for example, days, weeks, months, or the like. Note that the measurement period is not limited to these. The data structure of the actual demand management information 130 will be explained using FIG.

The elemental loss function information 131 stores loss functions that differ in at least one of the units of time series to be handled, variables, and formulas. Differences in the formulas represent sums, averages, variances, and the like. The elemental loss function information 131 stores loss functions related to KPIs (Key Performance Indicators) of the target business. The data structure of the element loss function information 131 will be explained using FIG.

As will be described later, in this embodiment, an integrated loss function suitable for a target business field is generated from a plurality of loss functions. In the following description, the loss functions managed by the element loss function information 131 are referred to as "element loss functions".

Note that the demand prediction system may be composed of a plurality of computers 100 . In this case, functional units are distributed and arranged in a plurality of computers 100 . For example, the input unit 120 is arranged in one computer 100 , the prediction unit 121 is arranged in one computer 100 , and the display unit 122 is arranged in one computer 100 .

The programs and data stored in the main storage device 113 may be stored in the secondary storage device 114 . In this case, processor 110 reads programs and data from secondary storage device 114 and stores them in main storage device 113 .

FIG. 2 is a diagram illustrating the details of the functional configuration of the demand prediction system of the first embodiment.

The input unit 120 receives a demand forecast output request for the target item together with the evaluation index target information 220 . The output request includes identification information of the target item. The input unit 120 reads out a data set composed of a group of shipping histories of target items in an arbitrary processing period from the actual demand management information 130 and outputs the data set to the prediction unit 121 . The input unit 120 also outputs the element loss function stored in the element loss function information 131 and the evaluation index target information 220 to the prediction unit 121 .

The evaluation index target information 220 includes the name of the evaluation index and conditions regarding the evaluation index. The condition of the evaluation index is, for example, a specific value or a tendency of change of the evaluation index.

The prediction unit 121 predicts the demand (shipment amount) of the target item during the prediction period, and outputs the prediction result to the display unit 122 . Details of the prediction unit 121 will now be described.

The prediction unit 121 includes an integrated loss function generator 200 , a demand forecaster 201 , an evaluation index calculator 202 , a function weight updater 203 , an integrated loss function updater 204 and a result generator 205 .

The integrated loss function generator 200 generates an initial integrated loss function using multiple element loss functions. Here, the integrated loss function is defined by Equation (1).

F represents the integrated loss function, i represents the ID of the element loss function, _{wi represents the weight, and f i represents} the element loss function. The weight _wi is a non-negative value.

The integrated loss function generator 200 generates an initial integrated loss function by setting initial values to the weights _wi . The integrated loss function generation unit 200 outputs integrated loss function information 210 including the generated integrated loss function to the demand prediction unit 201 .

The demand forecasting unit 201 generates a forecast model using the data set of the shipment history of the target item and the integrated loss function, and calculates the demand forecast value of the target item using the forecast model. The demand forecasting unit 201 outputs demand forecast information 211 including the name of the target item, the date, the demand forecast value, and the weight of the integrated loss function. Note that the data included in the demand forecast information 211 is an example and is not limited to this.

A prediction model is, for example, a neural network, but the present invention is not limited to this.

The evaluation index calculation unit 202 calculates an evaluation index related to work dealing with items. For example, KPIs such as the out-of-stock rate and transportation costs are calculated as evaluation indices. The evaluation index calculation unit 202 calculates the evaluation index by simulating a future business plan using, for example, the demand forecast value and the past shipment volume. It is assumed that an algorithm for simulating a business plan is set in advance in the evaluation index calculation unit 202 . The evaluation index calculator 202 outputs the evaluation index information 212 to the function weight updater 203 . The evaluation index information 212 includes data for specifying the weight of the integrated loss function together with the calculated evaluation index.

The function weight updating unit 203 updates the weights of the element loss functions that make up the integrated loss function based on the calculated evaluation index and evaluation index target information 220 . For example, the function weight updating unit 203 increases the weight of the element loss function that contributes to the improvement of the KPI calculated as the evaluation index, and decreases the weight of the element loss function that does not contribute to the improvement of the KPI. Improving the KPI means, for example, reducing the error between the calculated evaluation index and the target value.

When the calculated evaluation index does not satisfy a predetermined condition or the maximum calculation time is not reached, the function weight update unit 203 outputs the weight update result as the intermediate information 213 to the integrated loss function update unit 204. do. The function weight updating unit 203 updates the integrated loss function information 210 including only the final integrated loss function to the demand forecasting unit 203 when the calculated evaluation index satisfies a predetermined condition or when the maximum calculation time is reached. output to

Based on the intermediate information 213, the integrated loss function updating unit 204 deletes element loss functions that contribute less to the evaluation index from the integrated loss function.

The result generation unit 205 generates prediction result information 214 for displaying the demand prediction value and the final integrated loss function, and outputs it to the display unit 122 .

The display unit 122 presents the prediction result information 214 output from the prediction unit 121 .

As for each functional unit of the computer 100, a plurality of functional units may be combined into one functional unit, or one functional unit may be divided into a plurality of functional units for each function.

FIG. 3 is a diagram showing an example of the data structure of the actual demand management information 130 of the first embodiment.

The actual demand management information 130 stores an entry composed of an item ID 301 , a shipping date 302 and a shipping amount 303 . One entry corresponds to one shipping history. The actual demand management information 130 shown in FIG. 3 stores a shipment history whose measurement period is one day.

The item ID 301 is a field for storing item identification information. A shipping date 302 is a field for storing the shipping date of the item. A shipment amount 303 is a field for storing the shipment amount of the item.

Note that the demand forecast information 211 has the same data structure as the entry of the actual demand management information 130 .

FIG. 4 is a diagram showing an example of the data structure of the element loss function information 131 of Example 1. As shown in FIG.

The element loss function information 131 stores entries composed of ID 401 , name 402 , loss function 403 and description 404 . One entry corresponds to one element loss function.

An ID 401 is a field that stores identification information of an entry. The letter i in the integrated loss function is the value of ID401. Name 402 stores the name of the element loss function. A loss function 403 is a field that stores a formula representing an element loss function. Description 404 is a field that stores the description of the element loss function. Description 404 stores a document describing the element loss function.

FIG. 5 is a diagram showing an example of the data structure of integrated loss function information 210 according to the first embodiment.

Integrated loss function information 210 includes an entry consisting of pattern ID 501 and weight 502 . One entry corresponds to one weight pattern, ie, an integrated loss function.

The pattern ID 501 is a field that stores identification information of weight patterns. A weight 502 is a field that stores the weight of each element loss function stored in the element loss function information 131 .

FIG. 6 is a diagram showing an example of the data structure of the evaluation index information 212 of Example 1. As shown in FIG.

The evaluation index information 212 includes entries composed of pattern IDs 601 and evaluation indexes. One entry corresponds to the evaluation index calculated from the integrated loss function of one weight pattern.

Pattern ID 601 is the same field as pattern ID 501 . The evaluation index 602 is a field that stores the evaluation index calculated by the evaluation index calculation unit 202 . In this embodiment, it is assumed that the item shortage rate is calculated as the evaluation index. Note that the evaluation index 602 may include the name of the evaluation index and the like.

FIG. 7 is a flowchart illustrating an example of processing executed by the integrated loss function generation unit 200 according to the first embodiment.

The integrated loss function generator 200 acquires an element loss function from the element loss function information 131 (step S101).

The integrated loss function generator 200 generates an initial integrated loss function by determining the weight of each element loss function, that is, the weight pattern (step S102). For example, the integrated loss function generator 200 sets the weight of one element loss function to "1" and sets the weight of all other element loss functions to "0". Note that the present invention is not limited to the weighting pattern determination method.

At this time, integrated loss function generator 200 adds an entry to integrated loss function information 210 , sets identification information to pattern ID 501 of the added entry, and sets the determined weight to weight 502 .

The integrated loss function generating unit 200 outputs the initial integrated loss function to the demand forecasting unit 201 (step S103).

Specifically, the integrated loss function generating unit 200 outputs the integrated loss function information 210 to the demand forecasting unit 201 together with the defining formula of the element loss function acquired from the element loss function information 131 .

FIG. 8 is a flowchart illustrating an example of processing executed by the demand forecasting unit 201 of the first embodiment.

The demand forecasting unit 201 acquires a data set of the shipping history of the target item from the actual demand management information 130 (step S201), and also acquires a forecast model (step S202). It is assumed that the target items and the scope of the shipping history to be acquired are specified in advance.

The demand forecasting unit 201 sets an integrated loss function based on the integrated loss function information 210 (step S203).

Specifically, the demand forecasting unit 201 acquires the latest weighting pattern registered in the integrated loss function information 210, and sets the integrated loss function based on the weighting pattern.

The demand forecasting unit 201 learns a forecasting model using the integrated loss function and shipping history (step S204). A known technique may be used for the learning process, so a detailed description thereof will be omitted.

The demand forecasting unit 201 calculates the demand forecast value of the target item using the forecast model and the shipping history (step S205), and outputs the demand forecast value (step S206).

Specifically, when the integrated loss function information 210 is received from the integrated loss function generating unit 200 or the integrated loss function updating unit 204 , the demand forecasting unit 201 outputs the demand forecast information 211 to the evaluation index calculating unit 202 . Also, when receiving the integrated loss function information 210 from the function weight updating unit 203 , the demand forecasting unit 201 outputs the integrated loss function information 210 and the demand forecast information 211 to the result generating unit 205 . Note that the integrated loss function information 210 includes the ID of the weight pattern, not the integrated loss function itself.

Note that the shipping histories used in steps S204 and S205 may be the same data set or different data sets.

FIG. 9 is a flowchart illustrating an example of processing executed by the evaluation index calculation unit 202 according to the first embodiment.

The evaluation index calculator 202 acquires a demand forecast value (step S301), and calculates an evaluation index based on the demand forecast value and the algorithm (step S302). Note that the evaluation index calculation unit 202 may acquire the demand amount included in the shipping history in order to calculate the evaluation index.

The evaluation index calculator 202 outputs the calculated evaluation index to the function weight updater 203 (step S303).

Specifically, the evaluation index calculator 202 outputs the evaluation index information 212 to the function weight updater 203 .

Although only one evaluation index is calculated in this embodiment, a plurality of evaluation indices may be calculated.

FIG. 10 is a flowchart illustrating an example of processing executed by the function weight updating unit 203 of the first embodiment.

The function weight updating unit 203 acquires an evaluation index (step S401), and acquires the evaluation index target information 220 (step S402).

Note that when a plurality of evaluation indices are calculated, the function weight updating unit 203 may integrate the plurality of evaluation indices to calculate a single evaluation index. For example, it is conceivable to integrate multiple evaluation indices using a linear regression model.

The function weight updating unit 203 determines whether or not the condition regarding the evaluation index is satisfied (step S403).

For example, it is determined whether the evaluation index corresponding to the latest integrated loss function is less than a threshold. Also, it is determined whether or not the rate of change of the evaluation index is smaller than the threshold using a plurality of evaluation indexes. The present invention is not limited to the contents of the conditions relating to the evaluation index.

If it is determined that the condition regarding the evaluation index is satisfied, the function weight updating unit 203 proceeds to step S407.

When it is determined that the condition regarding the evaluation index is not satisfied, the function weight updating unit 203 determines whether or not the maximum calculation time has been reached (step S404).

For example, the function weight updating unit 203 determines whether or not the number of weight updates is greater than a threshold.

If it is determined that the maximum calculation time has been reached, the function weight updating unit 203 proceeds to step S407.

If it is determined that the maximum calculation time has not been reached, the function weight updating unit 203 updates the weights of the element loss functions that make up the integrated loss function (step S405).

For example, when a target value is included in the evaluation index target information 220, the function weight updating unit 203 updates the weight so that the error between the latest evaluation index and target value becomes small. Specifically, the function weight updating unit 203 reduces the weight of the element loss functions that do not contribute to the evaluation index and increases the weight of the element loss functions that contribute to the evaluation index. Note that the present invention is not limited to the weight update method.

The function weight update unit 203 outputs the intermediate information 213 including the weight update result to the integrated loss function update unit 204 (step S406).

At step S407, the function weight updating unit 203 outputs the final integrated loss function to the demand forecasting unit 201 (step S407).

Specifically, the function weight update unit 203 leaves only the latest weight pattern in the integrated loss function information 210, and sends the integrated loss function information 210 to the demand prediction unit 201 along with a flag indicating that it is the final result. Output. Note that the function weight updating unit 203 may determine the final integrated loss function based on the evaluation index.

FIG. 11 is a flowchart illustrating an example of processing executed by the integrated loss function updating unit 204 according to the first embodiment.

The integrated loss function updating unit 204 acquires the intermediate information 213 (step S501), and updates the weight smaller than the threshold value to "0" (step S502).

For example, when the weight is smaller than 0.01, the integrated loss function updating unit 204 updates the weight to "0". This operation corresponds to deleting element loss functions that contribute less to the value of the integrated loss function.

The integrated loss function updating unit 204 outputs the updated integrated loss function to the demand forecasting unit 201 (step S503).

Specifically, the integrated loss function updating unit 204 adds an entry to the integrated loss function information 210 , sets identification information to the pattern ID 501 , and sets the update result to the weight 502 .

12A and 12B are diagrams showing examples of screens displayed by the display unit 122 of the first embodiment.

Screen 1200 includes a prediction display field 1201 and an integrated loss function display field 1202 .

The forecast display column 1201 displays the target item, the demand forecast value, the shipping date, and the like. The integrated loss function display column 1202 displays the element loss functions that make up the integrated loss function and their weights as bar graphs. The size of the bar graph corresponds to the size of the weight. Also, in the integrated loss function display field 1202, a description (context) of the prediction result based on the element loss function with the maximum weight is displayed. The user can confirm the magnitude of contribution of each element loss function in the integrated loss function by referring to the integrated loss function display field 1202 .

If the bar graph is clicked, screen 1210 of FIG. 12B is displayed. Screen 1210 displays a time series of forecast results of the forecast model generated using the element loss function corresponding to the clicked bar graph, and a time series of demand history in units corresponding to the element loss function. The horizontal axis corresponds to time, and the vertical axis corresponds to shipment volume. The user can check the effectiveness and prediction accuracy of the element loss function by referring to screen 1210 .

As described above, the demand forecasting system of the first embodiment can automatically set an integrated loss function suitable for learning a forecasting model using a plurality of elemental loss functions. Also, the demand forecasting system can generate a highly accurate forecasting model using the integrated loss function. In addition, the demand forecasting system can present the magnitude of the contribution of the element loss function in the integrated loss function as one of the grounds for forecasting the forecasting model.

It should be noted that the present invention is not limited to models that predict demand for items. Similar techniques and configurations can be applied to a wide variety of models.

The second embodiment differs from the first embodiment in that the user designates the element loss functions and weights to be used. The second embodiment will be described below, focusing on the differences from the first embodiment.

Since the configuration of the demand forecasting system of the second embodiment is the same as that of the first embodiment, the description thereof is omitted. The functional configuration of the demand forecasting system of the second embodiment differs from that of the first embodiment.

FIG. 13 is a diagram illustrating the details of the functional configuration of the demand prediction system of the second embodiment.

The input unit 120 receives a demand forecast output request for the target item. The output request includes identification information of the target item. Also, the input unit 120 displays a screen for designating the element loss functions and weights to be used.

FIG. 14 is a diagram illustrating an example of a screen displayed by the input unit 120 according to the second embodiment.

A screen 1400 includes a selection field 1401 and a setting button 1402 . A selection column 1401 is a column for designating an element loss function and a weight. The selection column 1401 stores entries composed of adoption flags 1411 , names 1412 , loss functions 1413 , descriptions 1414 and weights 1415 . One entry corresponds to one element loss function.

Name 1412 , loss function 1413 and description 1414 are the same fields as name 402 , loss function 403 and description 404 .

The adoption flag 1411 is a field for designating the element loss function to be used. A weight 1415 is a field for entering a weight. The user adds a check to the adoption flag 1411 of the element loss function to be used and inputs a value to the weight 1415 .

A setting button 1402 is an operation button for setting the element loss function and weight specified by the user.

The input unit 120 reads out a data set composed of a group of shipping histories of target items in an arbitrary processing period from the actual demand management information 130 and outputs the data set to the prediction unit 121 . The input unit 120 also outputs the element loss function and weight specified by the user to the prediction unit 121 .

The prediction unit 121 predicts the demand (shipment amount) of the target item and outputs the prediction result to the display unit 122 . Details of the prediction unit 121 will now be described.

The prediction unit 121 includes an integrated loss function generation unit 200 , a demand prediction unit 201 and a result generation unit 205 .

The integrated loss function generation unit 200 of the second embodiment generates an integrated loss function using element loss functions and weights specified by the user. In Example 2, the weights of the element loss functions that make up the integrated loss function are not updated.

The demand forecasting unit 201 of the second embodiment is the same as the demand forecasting unit 201 of the first embodiment. However, the demand forecasting unit 201 of the second embodiment outputs the integrated loss function information 210 and the demand forecast information 211 to the result generating unit 205 .

The result generator 205 of the second embodiment is the same as the result generator 205 of the first embodiment.

The data structures of the integrated loss function information 210, the demand forecast information 211, and the forecast result information 214 of the second embodiment are the same as those of the first embodiment.

The display unit 122 presents the prediction result information 214 output from the prediction unit 121 .

In addition, the present invention is not limited to the above-described embodiments, and includes various modifications. Further, for example, the above-described embodiments are detailed descriptions of the configurations for easy understanding of the present invention, and are not necessarily limited to those having all the described configurations. Moreover, it is possible to add, delete, or replace a part of the configuration of each embodiment with another configuration.

Further, each of the above configurations, functions, processing units, processing means, and the like may be realized by hardware, for example, by designing a part or all of them using an integrated circuit. The present invention can also be implemented by software program code that implements the functions of the embodiments. In this case, a computer is provided with a storage medium recording the program code, and a processor included in the computer reads the program code stored in the storage medium. In this case, the program code itself read from the storage medium implements the functions of the above-described embodiments, and the program code itself and the storage medium storing it constitute the present invention. Examples of storage media for supplying such program code include flexible disks, CD-ROMs, DVD-ROMs, hard disks, SSDs (Solid State Drives), optical disks, magneto-optical disks, CD-Rs, magnetic tapes, A nonvolatile memory card, ROM, or the like is used.

Also, the program code that implements the functions described in this embodiment can be implemented in a wide range of programs or scripting languages such as assembler, C/C++, perl, Shell, PHP, Python, and Java (registered trademark).

Furthermore, by distributing the program code of the software that implements the functions of the embodiment via a network, it can be stored in storage means such as a hard disk or memory of a computer, or in a storage medium such as a CD-RW or CD-R. Alternatively, a processor provided in the computer may read and execute the program code stored in the storage means or the storage medium.

In the above-described embodiments, the control lines and information lines indicate those considered necessary for explanation, and not all control lines and information lines are necessarily indicated on the product. All configurations may be interconnected.

100 computer 101 input device 102 output device 103 network 110 processor 111 IO interface 112 network interface 113 main storage device 114 secondary storage device 120 input unit 121 prediction unit 122 display unit 130 demand performance management information 131 element loss function information 200 integrated loss function generation Unit 201 Demand prediction unit 202 Evaluation index calculation unit 203 Function weight update unit 204 Integrated loss function update unit 205 Result generation unit 210 Integrated loss function information 211 Demand forecast information 212 Evaluation index information 213 Intermediate information 214 Prediction result information 220 Evaluation index target information 1200, 1210, 1400 screen

Claims (12)

A computer system comprising at least one computer,
Manage multiple loss functions and history of item demand,
The at least one calculator comprises:
a first process of receiving information on a condition to be satisfied by the evaluation index relating to demand for the item;
a second process of generating an integrated loss function using the multiple weighted loss functions;
A model for predicting the demand for the item is generated by machine learning using the integrated loss function and the history, and the evaluation index is calculated based on the demand prediction for the item obtained by inputting the history to the model. a third process of calculating
a fourth process for determining whether or not the condition is satisfied based on the received information and the evaluation index;
updating the weight by which at least one of the loss functions included in the integrated loss function is multiplied based on the evaluation index when it is determined that the condition is not satisfied, and executing the third process; 5 processing;
a sixth process of outputting a demand forecast for the item when it is determined that the condition is satisfied;
A computer system characterized by executing A computer system according to claim 1,
A computer system, wherein each of the plurality of loss functions differs in at least one of a unit of time series to be handled, a variable, and a formula. A computer system according to claim 1,
In the sixth process, the at least one computer outputs prediction result information including a demand prediction for the item and a weight of each of the plurality of loss functions. A computer system according to claim 1,
In the fifth processing, the at least one computer deletes the loss function whose updated weight is smaller than a threshold from the integrated loss function. An item demand forecasting method executed by a computer system comprising at least one computer, comprising:
The computer system manages a plurality of loss functions and a history of item demand,
The demand forecasting method for the item includes:
the at least one calculator performing a first process of receiving information on conditions to be satisfied by the evaluation index relating to the demand for the item;
the at least one computer performing a second process of generating a combined loss function using the multiple weighted loss functions;
The at least one computer generates a model for predicting the demand for the item by machine learning using the integrated loss function and the history, and predicts the demand for the item obtained by inputting the history to the model. a step of executing a third process of calculating the evaluation index based on
the at least one computer executing a fourth process of determining whether or not the condition is satisfied based on the received information and the evaluation index;
If it is determined that the condition is not satisfied, the at least one calculator updates the weight multiplied by at least one of the loss functions included in the integrated loss function based on the evaluation index, and a step of executing a fifth process of executing a third process;
If it is determined that the condition is satisfied, the at least one computer executes a sixth process of outputting a demand forecast for the item;
An item demand forecasting method comprising: An item demand forecasting method according to claim 5,
The item demand forecasting method, wherein each of the plurality of loss functions is different in at least one of time-series units, variables, and mathematical expressions. An item demand forecasting method according to claim 5,
In the sixth process, the at least one computer outputs prediction result information including the demand prediction of the item and the weight of each of the plurality of loss functions. An item demand forecasting method according to claim 5,
In the fifth processing, the at least one computer deletes the loss function whose updated weight is smaller than a threshold from the integrated loss function. A program to be executed by a computer,
The calculator is
a processor, a storage device connected to the processor, and a network interface connected to the processor;
Manage multiple loss functions and history of item demand,
Said program
a procedure for executing a first process of receiving information on conditions to be satisfied by the evaluation index relating to demand for the item;
performing a second process of generating an integrated loss function using the multiple weighted loss functions;
A model for predicting the demand for the item is generated by machine learning using the integrated loss function and the history, and the evaluation index is calculated based on the demand prediction for the item obtained by inputting the history to the model. a procedure for executing a third process of calculating
a procedure for executing a fourth process for determining whether or not the condition is satisfied based on the received information and the evaluation index;
updating the weight by which at least one of the loss functions included in the integrated loss function is multiplied based on the evaluation index when it is determined that the condition is not satisfied, and executing the third process; 5 a procedure for performing the process;
a step of executing a sixth process of outputting a demand forecast for the item by the at least one calculator if it is determined that the condition is satisfied;
is executed by the computer. The program according to claim 9,
A program characterized in that each of the plurality of loss functions differs in at least one of units of time series to be handled, variables, and mathematical expressions. The program according to claim 9,
In the sixth process, the program causes the computer to output prediction result information including the demand prediction for the item and the weight of each of the plurality of loss functions. The program according to claim 9,
In the fifth processing, the computer causes the computer to delete the loss function whose updated weight is smaller than a threshold from the integrated loss function.

Images