Remaining amount prediction system and remaining amount prediction program

Abstract

To provide a remaining amount prediction system and a remaining amount prediction program that can improve the accuracy of prediction of the remaining amount of a consumable item.SOLUTION: A remaining amount prediction system comprises: a remaining amount prediction unit that predicts the date on which a remaining amount of toner becomes 0% by using a remaining amount prediction model as a machine learning model that predicts the remaining amount of toner used in an image forming apparatus; and a model training unit that executes training of the remaining amount prediction model. When executing the training of the remaining amount prediction model, the model training unit adjusts a parameter of the remaining amount prediction model so as to minimize Loss, which indicates the magnitude of deviation between the remaining amount of toner predicted by the remaining amount prediction model and correct answer data relative to the entirety of a plurality of pieces of teacher data. The model training unit receives a user's input of prediction date evaluation as evaluation of the date predicted by the remaining amount prediction unit to the date on which the remaining amount of toner actually becomes 0% or less, and changes a method for calculating Loss according to the input prediction date evaluation (S203-S208).SELECTED DRAWING: Figure 17

Description

[Technical Field]

[0001] The present invention relates to a remaining amount prediction system and a remaining amount prediction program for predicting the remaining amount of consumables used in electronic devices. [Background technology]

[0002] BACKGROUND ART Conventionally, there is known a remaining amount prediction system that predicts the remaining amount of a consumable item used in an electronic device using a machine learning model that predicts the remaining amount of the consumable item (see, for example, Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 2020-151879 Summary of the Invention [Problem to be solved by the invention]

[0004] However, conventional remaining amount prediction systems have a problem in that they are unable to improve the accuracy of predictions of remaining amounts of consumables using machine learning models.

[0005] SUMMARY OF THE INVENTION It is therefore an object of the present invention to provide a remaining amount prediction system and a remaining amount prediction program that can improve the accuracy of prediction of remaining amounts of consumables. [Means for solving the problem]

[0006] The remaining quantity prediction system of the present invention comprises a remaining quantity prediction unit that predicts the day on which the remaining quantity of a consumable used in an electronic device will fall below a specific amount using a remaining quantity prediction model as a machine learning model that predicts the remaining quantity of the consumable, and a model learning unit that performs learning of the remaining quantity prediction model.When performing learning of the remaining quantity prediction model, the model learning unit adjusts the parameters of the remaining quantity prediction model so that the loss, which indicates the magnitude of deviation between the correct data and the remaining quantity of the consumable predicted by the remaining quantity prediction model for all multiple training data, is minimized, and the model learning unit is characterized in that a user inputs a predicted day evaluation as an evaluation of the day predicted by the remaining quantity prediction unit for the day on which the remaining quantity of the consumable will actually fall below the specific amount, and changes the method of calculating the loss in accordance with the input predicted day evaluation.

[0007] With this configuration, the remaining amount prediction system of the present invention changes the method of calculating loss depending on the prediction date evaluation input by the user, thereby improving the accuracy of prediction of the remaining amount of consumables.

[0008] In the remaining quantity prediction system of the present invention, the predicted date evaluation may include at least "just right," indicating that the date predicted by the remaining quantity prediction unit is appropriate compared to the date on which the remaining quantity of the consumable will actually be below the specific amount, and "early," indicating that the date predicted by the remaining quantity prediction unit is inappropriate and early compared to the date on which the remaining quantity of the consumable will actually be below the specific amount, and when "early" is input as the predicted date evaluation, when the remaining quantity of the consumable predicted by the remaining quantity prediction model is smaller than the correct data, the model learning unit may calculate the loss by weighting the difference between the correct data and the remaining quantity of the consumable predicted by the remaining quantity prediction model more heavily than when the remaining quantity of the consumable predicted by the remaining quantity prediction model is larger than the correct data.

[0009] With this configuration, when "early" is input as the prediction date evaluation, the remaining quantity prediction system of the present invention calculates the loss by weighting the difference between the correct data and the remaining quantity of the consumable predicted by the remaining quantity prediction model more heavily than when the remaining quantity of the consumable predicted by the remaining quantity prediction model is greater than the correct data, thereby improving the accuracy of predictions of the remaining quantity of the consumable.

[0010] In the remaining quantity prediction system of the present invention, when "just right" is input as the predicted date evaluation, the model learning unit calculates multiple squared errors as the square of the difference between the correct data and the remaining quantity of the consumable predicted by the remaining quantity prediction model, and calculates the square root of the average of all the calculated squared errors as the Loss; when "early" is input as the predicted date evaluation, the model learning unit calculates multiple squared errors, increases the squared error when the remaining quantity of the consumable predicted by the remaining quantity prediction model is smaller than the correct data, and calculates the Loss as the square root of the average of all the squared errors when the remaining quantity of the consumable predicted by the remaining quantity prediction model is not smaller than the correct data and all the squared errors after the increase when the remaining quantity of the consumable predicted by the remaining quantity prediction model is smaller than the correct data.

[0011] With this configuration, when "early" is input as the prediction date evaluation, the remaining quantity prediction system of the present invention calculates multiple squared errors as the square of the difference between the correct data and the remaining quantity of the consumable predicted by the remaining quantity prediction model, increases the squared error when the remaining quantity of the consumable predicted by the remaining quantity prediction model is smaller than the correct data, and calculates the loss as the square root of the average of all squared errors when the remaining quantity of the consumable predicted by the remaining quantity prediction model is not smaller than the correct data and all squared errors after the increase when the remaining quantity of the consumable predicted by the remaining quantity prediction model is smaller than the correct data, thereby improving the accuracy of predicting the remaining quantity of the consumable.

[0012] In the remaining quantity prediction system of the present invention, the predicted date evaluation may include at least "just right," indicating that the date predicted by the remaining quantity prediction unit is appropriate relative to the date on which the remaining quantity of the consumable will actually be below the specific amount, and "late," indicating that the date predicted by the remaining quantity prediction unit is inappropriate and late relative to the date on which the remaining quantity of the consumable will actually be below the specific amount, and when "late" is input as the predicted date evaluation, when the remaining quantity of the consumable predicted by the remaining quantity prediction model is larger than the correct data, the model learning unit may calculate the loss by assigning a greater weight to the difference between the correct data and the remaining quantity of the consumable predicted by the remaining quantity prediction model compared to when the remaining quantity of the consumable predicted by the remaining quantity prediction model is smaller than the correct data.

[0013] With this configuration, when "late" is input as the prediction date evaluation, the remaining quantity prediction system of the present invention calculates the loss by weighting the difference between the correct data and the remaining quantity of the consumable predicted by the remaining quantity prediction model more heavily when the remaining quantity of the consumable predicted by the remaining quantity prediction model is larger than the correct data, compared to when the remaining quantity of the consumable predicted by the remaining quantity prediction model is smaller than the correct data, thereby improving the accuracy of prediction of the remaining quantity of the consumable.

[0014] In the remaining quantity prediction system of the present invention, when "just right" is input as the predicted date evaluation, the model learning unit calculates multiple squared errors as the square of the difference between the correct data and the remaining quantity of the consumable predicted by the remaining quantity prediction model, and calculates the square root of the average of all the calculated squared errors as the Loss; when "late" is input as the predicted date evaluation, the model learning unit calculates multiple squared errors, increases the squared error when the remaining quantity of the consumable predicted by the remaining quantity prediction model is greater than the correct data, and calculates the Loss as the square root of the average of all the squared errors when the remaining quantity of the consumable predicted by the remaining quantity prediction model is not greater than the correct data and all the squared errors after the increase when the remaining quantity of the consumable predicted by the remaining quantity prediction model is greater than the correct data.

[0015] With this configuration, when "late" is input as the prediction date evaluation, the remaining quantity prediction system of the present invention calculates multiple squared errors as the square of the difference between the correct data and the remaining quantity of the consumable predicted by the remaining quantity prediction model, increases the squared error when the remaining quantity of the consumable predicted by the remaining quantity prediction model is greater than the correct data, and calculates the loss as the square root of the average of all squared errors when the remaining quantity of the consumable predicted by the remaining quantity prediction model is not greater than the correct data and all squared errors after the increase when the remaining quantity of the consumable predicted by the remaining quantity prediction model is greater than the correct data, thereby improving the accuracy of predicting the remaining quantity of the consumable.

[0016] The remaining quantity prediction program of the present invention causes a computer to implement a remaining quantity prediction unit that predicts the day on which the remaining quantity of a consumable used in an electronic device will fall below a specific amount using a remaining quantity prediction model as a machine learning model that predicts the remaining quantity of the consumable used in the electronic device, and a model learning unit that performs learning of the remaining quantity prediction model.When performing learning of the remaining quantity prediction model, the model learning unit adjusts the parameters of the remaining quantity prediction model so that the loss, which indicates the magnitude of deviation between the correct data and the remaining quantity of the consumable predicted by the remaining quantity prediction model for all multiple training data, is minimized.The model learning unit is characterized in that a user inputs a predicted day evaluation as an evaluation of the day predicted by the remaining quantity prediction unit for the day on which the remaining quantity of the consumable will actually fall below the specific amount, and changes the method of calculating the loss in accordance with the input predicted day evaluation.

[0017] With this configuration, the computer executing the remaining quantity prediction program of the present invention changes the method of calculating loss depending on the prediction date evaluation entered by the user, thereby improving the accuracy of predictions of remaining quantities of consumables. [Effects of the Invention]

[0018] The remaining amount prediction system and remaining amount prediction program of the present invention can improve the accuracy of prediction of remaining amounts of consumables. [Brief explanation of the drawings]

[0019] [Figure 1] 1 is a block diagram of a system according to an embodiment of the present invention. [Figure 2] FIG. 2 is a block diagram of the remaining amount prediction system shown in FIG. 1 when configured by one computer. [Figure 3] 3 is a diagram showing an example of the remaining amount of toner for each day indicated in the actual measurement value data shown in FIG. 2. FIG. [Figure 4] 3 is a diagram showing an example of a method for evaluating an empty day predicted by the remaining amount prediction unit shown in FIG. 2 against an actual empty day. FIG. [Figure 5]FIG. 3 is a diagram showing an example of predicted date evaluation information shown in FIG. 2. [Figure 6] FIG. 2 is a block diagram of the image forming apparatus shown in FIG. 1 when it is an MFP. [Figure 7] FIG. 2 is a block diagram of the user terminal shown in FIG. [Figure 8] 2 is a sequence diagram of the operation of the system shown in FIG. 1 when the remaining amount prediction system records the actual measured value of the remaining amount of toner in the image forming apparatus. [Figure 9] 3 is a flowchart of the operation of the remaining amount prediction system shown in FIG. 2 when predicting the toner empty date of the image forming apparatus. [Figure 10] 3 is a diagram showing an example of the remaining amount of toner for each day indicated in the actual measurement value data shown in FIG. 2 and the predicted remaining amount of toner one day after the latest date indicated in the actual measurement value data. FIG. [Figure 11] FIG. 3 is a diagram showing an example of the remaining amount of toner for each day shown in the actual measurement value data shown in FIG. 2 and the predicted remaining amount of toner for each day up to two days after the latest date shown in the actual measurement value data. [Figure 12] 3 is a flowchart of the operation of the remaining amount prediction system shown in FIG. 2 when generating a remaining amount prediction model for the image forming apparatus. [Figure 13] 3 is a diagram showing an example of a target variable and one day's or more remaining amount of toner that serves as a basis for an explanatory variable in training data for generating the remaining amount prediction model shown in FIG. 2. FIG. [Figure 14] 10 is a flowchart of the operation of the remaining amount prediction system shown in FIG. 2 when calculating Loss in a state where a prediction day evaluation has not been input by a user. [Figure 15] FIG. 15 is a diagram showing an example of a square error calculated in the operation shown in FIG. [Figure 16] FIG. 2 is a sequence diagram of the operation of the system shown in FIG. 1 when a predicted date evaluation is input. [Figure 17] 10 is a flowchart of the operation of the remaining amount prediction system shown in FIG. 2 when calculating Loss in a state where a predicted day evaluation has been input by a user. [Figure 18]18 is a diagram showing an example of squared errors processed in the operation shown in FIG. 17 when the predicted day evaluation is a negative value. FIG. [Figure 19] 18 is a diagram showing an example of squared errors processed in the operation shown in FIG. 17 when the predicted day evaluation is a positive value. FIG. DETAILED DESCRIPTION OF THE INVENTION

[0020] Hereinafter, embodiments of the present invention will be described with reference to the drawings.

[0021] First, the configuration of a system according to an embodiment of the present invention will be described.

[0022] FIG. 1 is a block diagram of a system 10 according to the present embodiment.

[0023] As shown in FIG. 1, system 10 includes a remaining amount prediction system 20 that predicts the remaining amount of toner, which is a consumable item used in an image forming apparatus, which is an electronic device. Remaining amount prediction system 20 can predict the empty day, which is the day when the remaining amount of toner in the image forming apparatus will be 0%, i.e., the day when the toner will be empty. Remaining amount prediction system 20 may be configured by a single computer such as a PC (Personal Computer), or may be configured by multiple computers. Remaining amount prediction system 20 may also be configured on the cloud.

[0024] The system 10 includes an image forming device 30, such as an MFP (Multifunction Peripheral) or a dedicated printer. The system 10 can include at least one other image forming device having a similar configuration to the image forming device 30.

[0025] The system 10 includes a user terminal 40 used by a user. The system 10 may include at least one other user terminal having a similar configuration to the user terminal 40. The user terminal may be configured as a PC.

[0026] The remaining amount prediction system 20 and the image forming device can communicate with each other via a network 11 such as a LAN (Local Area Network) or the Internet. The remaining amount prediction system 20 and the user terminal can communicate with each other via the network 11, for example.

[0027] FIG. 2 is a block diagram of a remaining amount prediction system 20 configured by one computer.

[0028] As shown in Figure 2, the remaining amount prediction system 20 includes an operation unit 21, which is an operation device such as a keyboard or mouse through which various operations are input; a display unit 22, which is a display device such as an LCD (Liquid Crystal Display) that displays various information; a communication unit 23, which is a communication device that communicates with external devices via a network such as a LAN or the Internet, or directly via a wired or wireless connection without using a network; a memory unit 24, which is a non-volatile memory device such as a semiconductor memory or an HDD (Hard Disk Drive) that stores various information; and a control unit 25 that controls the entire remaining amount prediction system 20.

[0029] The storage unit 24 can store a remaining amount prediction program 24a for predicting the remaining amount of toner in the image forming apparatus 30. The remaining amount prediction program 24a may be installed in the remaining amount prediction system 20 during the manufacturing stage of the remaining amount prediction system 20, or may be additionally installed in the remaining amount prediction system 20 from an external storage medium such as a USB (Universal Serial Bus) memory, or may be additionally installed in the remaining amount prediction system 20 from a network.

[0030] The memory unit 24 can store a remaining amount prediction model 24b as a machine learning model that predicts the remaining amount of toner one day after a base latest date, which is the latest date among dates corresponding to one or more days' worth of remaining toner amounts that serve as the basis for explanatory variables input to the machine learning model. The remaining amount prediction model 24b is a remaining amount prediction model for the image forming device 30. The memory unit 24 can store a remaining amount prediction model similar to the remaining amount prediction model 24b for each image forming device. The explanatory variables of the remaining amount prediction model may include the remaining amount of toner on a specific day, such as the remaining amount of toner on the base latest date, the remaining amount of toner one day before the base latest date, or the remaining amount of toner two days before the base latest date. The explanatory variables of the remaining amount prediction model may also include a specific value calculated based on the remaining amount of toner on a specific day, such as the average value of the remaining amount of toner on the base latest date, the remaining amount of toner one day before the base latest date, and the remaining amount of toner two days before the base latest date.

[0031] The storage unit 24 can store actual measurement value data 24c indicating the actual daily measurement value of the remaining amount of toner in the image forming device 30. Similarly, the storage unit 24 can store actual measurement value data indicating the actual daily measurement value of the remaining amount of toner in the image forming device for each image forming device.

[0032] FIG. 3 is a diagram showing an example of the remaining amount of toner for each day indicated in the actual measurement data 24c.

[0033] As shown in Fig. 3, the actual measurement data 24c shows the date and the actual measurement value of the remaining amount of toner for each day. In the example shown in Fig. 3, the actual measurement values ​​of the remaining amount of toner for eight days from January 1 to January 8 are shown. The remaining amount of toner is expressed as a percentage indicated by an integer between 100 and 0.

[0034] As shown in FIG. 2, the memory unit 24 can store predicted day evaluation information 24d that indicates, for each image forming device, an evaluation of the empty day predicted by the remaining amount prediction unit 25a (described later) relative to the actual empty day (hereinafter referred to as the "predicted day evaluation").

[0035] FIG. 4 is a diagram illustrating an example of a method for predictive date evaluation.

[0036] As shown in FIG. 4 , the predicted date rating is implemented, for example, by an integer between −5 and 5. If the user determines that the empty date predicted using the remaining quantity prediction model is appropriate for the actual empty date, i.e., just right, the user should enter 0 as the predicted date rating. If the user determines that the empty date predicted using the remaining quantity prediction model is inappropriate and early for the actual empty date, the user should enter a positive value as the predicted date rating. The larger the difference between the empty date predicted using the remaining quantity prediction model and the actual empty date, the larger the predicted date rating should be. If the user determines that the empty date predicted using the remaining quantity prediction model is inappropriate and late for the actual empty date, the user should enter a negative value as the predicted date rating. The larger the difference between the empty date predicted using the remaining quantity prediction model and the actual empty date, the smaller the predicted date rating should be.

[0037] FIG. 5 is a diagram showing an example of the predicted date evaluation information 24d.

[0038] The predicted date evaluation information 24d shown in Fig. 5 indicates, for each image forming device, a device ID as identification information for the image forming device and a predicted date evaluation of the empty date predicted using the remaining amount prediction model for the image forming device. Some information is omitted from the predicted date evaluation information 24d shown in Fig. 5.

[0039] 2 includes, for example, a CPU (Central Processing Unit), a ROM (Read Only Memory) that stores programs and various data, and a RAM (Random Access Memory) that serves as a memory used as a work area for the CPU of the control unit 25. The CPU of the control unit 25 executes programs stored in the storage unit 24 or the ROM of the control unit 25.

[0040] The control unit 25 executes the remaining amount prediction program 24a to realize a remaining amount prediction unit 25a that predicts the empty date using a remaining amount prediction model, and a model learning unit 25b that learns the remaining amount prediction model.

[0041] FIG. 6 is a block diagram of the image forming apparatus 30 in the case of an MFP.

[0042] As shown in FIG. 6, the image forming apparatus 30 includes an operation unit 31, which is an operation device such as a button through which various operations are input; a display unit 32, which is a display device such as an LCD that displays various information; a printer 33, which is a printing device that prints images on a recording medium such as paper; a remaining amount sensor 34 that detects the amount of toner remaining in a toner container that stores toner consumed by the printer 33; a scanner 35, which is a reading device that reads images from a document; a communication unit 36, which is a communication device that communicates with an external device via a network such as a LAN or the Internet, or directly via a wired or wireless connection without using a network; a memory unit 37, which is a non-volatile memory device such as a semiconductor memory or HDD that stores various information; and a control unit 38 that controls the entire image forming apparatus 30.

[0043] The printer 33 has a replaceable toner container.

[0044] The remaining amount of toner detected by the remaining amount sensor 34 becomes 100% when a used toner container in the image forming apparatus 30 is replaced with a new toner container.

[0045] The control unit 38 includes, for example, a CPU, a ROM that stores programs and various data, and a RAM as memory used as a work area for the CPU of the control unit 38. The CPU of the control unit 38 executes programs stored in the storage unit 37 or the ROM of the control unit 38.

[0046] FIG. 7 is a block diagram of the user terminal 40. As shown in FIG.

[0047] As shown in Figure 7, the user terminal 40 includes an operation unit 41, which is an operation device such as a keyboard or mouse through which various operations are input; a display unit 42, which is a display device such as an LCD that displays various information; a communication unit 43, which is a communication device that communicates with external devices via a network such as a LAN or the Internet, or directly via a wired or wireless connection without using a network; a memory unit 44, which is a non-volatile memory device such as a semiconductor memory or HDD that stores various information; and a control unit 45 that controls the entire user terminal 40.

[0048] The control unit 45 includes, for example, a CPU, a ROM that stores programs and various data, and a RAM as memory used as a work area for the CPU of the control unit 45. The CPU of the control unit 45 executes programs stored in the storage unit 44 or the ROM of the control unit 45.

[0049] The operation of the system 10 will now be described.

[0050] First, the operation of the remaining amount prediction system 20 when recording the actual measured value of the remaining amount of toner in the image forming apparatus will be described.

[0051] FIG. 8 is a sequence diagram of the operation of the remaining amount prediction system 20 when the remaining amount prediction system 20 records the actual measured value of the remaining amount of toner in the image forming device 30.

[0052] As shown in FIG. 8, the image forming device 30 transmits the actual measured value of the remaining amount of toner each day to the remaining amount prediction system 20, for example, by transmitting the latest actual measured value of the remaining amount of toner detected by the remaining amount sensor 34 before the end of the previous day after the end of each day to the remaining amount prediction system 20 (S101).

[0053] When the actual measurement value of the remaining amount of toner is transmitted from the image forming device 30, the remaining amount prediction unit 25a of the remaining amount prediction system 20 records the transmitted actual measurement value in the actual measurement value data 24c of the image forming device 30 (S102).

[0054] The above describes the operation of the remaining amount prediction system 20 when it records the actual measured value of the remaining amount of toner in the image forming device 30, but the operation of the system 10 is similar when it records the actual measured value of the remaining amount of toner in an image forming device other than the image forming device 30.

[0055] Next, the operation of the remaining amount prediction system 20 when predicting the toner empty date of the image forming apparatus will be described.

[0056] FIG. 9 is a flowchart of the operation of the remaining amount prediction system 20 when predicting the toner empty date of the image forming device 30.

[0057] As shown in FIG. 9, the remaining amount predicting unit 25a generates explanatory variables for the remaining amount prediction model 24b based on the remaining amount of toner for each day indicated in the actual measurement value data 24c (S121).

[0058] When the processing of S121 is completed, the remaining amount prediction unit 25a inputs the explanatory variables generated just before into the remaining amount prediction model 24b, thereby predicting the remaining amount of toner one day after the latest base date of the explanatory variables generated just before (S122).

[0059] FIG. 10 is a diagram showing an example of the remaining amount of toner for each day indicated in the actual measurement value data 24c and the predicted remaining amount of toner one day after the latest date indicated in the actual measurement value data 24c.

[0060] If the remaining amount of toner from January 1st to January 8th shown in the actual measurement data 24c used in S121 is as shown in Figure 3, the remaining amount prediction unit 25a, in the first S122 of the operation shown in Figure 9, predicts the remaining amount of toner one day after January 8th, which is the latest base date of the explanatory variables generated in the previous S121, i.e., the remaining amount of toner on January 9th, as shown in Figure 10, for example.

[0061] As shown in FIG. 9, when the process of S122 is completed, the remaining amount predicting unit 25a determines whether the remaining amount of toner predicted in the immediately preceding S122 is 0% or less (S123).

[0062] If the remaining amount prediction unit 25a determines in S123 that the remaining amount of toner predicted in the previous S122 is not 0% or less, it generates explanatory variables for the remaining amount prediction model 24b based on the daily remaining amount of toner shown in the actual measurement data 24c and the daily remaining amount of toner predicted in all S122 in the current operation shown in Figure 9 (S124), and executes the processing of S122.

[0063] FIG. 11 is a diagram showing an example of the remaining amount of toner for each day indicated in the actual measurement value data 24c and the predicted remaining amount of toner for each day up to two days after the latest date indicated in the actual measurement value data 24c.

[0064] If the remaining amount of toner from January 1st to January 8th shown in the actual measurement data 24c used in S121 and the remaining amount of toner on January 9th predicted in the first S122 of the current operation shown in Figure 9 are as shown in Figure 10, the remaining amount prediction unit 25a will predict, in the second S122 of the current operation shown in Figure 9, the remaining amount of toner one day after January 9th, which is the latest base date of the explanatory variables generated in the previous S124, i.e., the remaining amount of toner on January 10th, as shown in Figure 11, for example.

[0065] As shown in FIG. 9, when the remaining amount prediction unit 25a determines in S123 that the remaining amount of toner predicted in the previous S122 is 0% or less, it determines the date one day after the most recent base date of the explanatory variable generated immediately before as the toner empty date (S125), and terminates the operation shown in FIG. 9.

[0066] The above describes the operation of the remaining amount prediction system 20 when predicting the toner empty date of the image forming device 30, but the operation is similar when the remaining amount prediction system 20 predicts the toner empty date of an image forming device other than the image forming device 30.

[0067] Next, the operation of the remaining amount prediction system 20 when generating a remaining amount prediction model for an image forming apparatus will be described.

[0068] FIG. 12 is a flowchart of the operation of the remaining amount prediction system 20 when generating the remaining amount prediction model 24b for the image forming device 30.

[0069] As shown in FIG. 12, the model learning unit 25b generates training data for generating the remaining amount prediction model 24b (S141).

[0070] For example, if the actual measurement data 24c shown in Figure 3 exists, the remaining amount of toner of one day or more that forms the basis of the objective variable and the explanatory variable in the training data for generating the remaining amount prediction model 24b can be generated as shown in Figure 13, for example.

[0071] FIG. 13 is a diagram showing an example of a target variable and a remaining amount of toner equivalent to one day or more that serves as a basis for an explanatory variable in training data for generating the remaining amount prediction model 24b.

[0072] As shown in FIG. 13, the objective variable in the training data for the remaining amount prediction model 24b, in which the remaining amount of toner on January 1, 100, is the only explanatory variable, is the remaining amount of toner on January 2, 99.

[0073] The objective variable in the training data for the remaining amount prediction model 24b, in which the remaining amounts of toner on January 1st and January 2nd, 100 and 99, respectively, are the explanatory variables, is the remaining amount of toner on January 3rd, 98.

[0074] The objective variable in the training data for remaining amount prediction model 24b, in which the remaining toner amounts on January 1st, January 2nd, and January 3rd, respectively, are 100, 99, and 98, are the explanatory variables, and the remaining toner amount on January 4th, 97, is the objective variable.

[0075] The objective variable in the training data for remaining amount prediction model 24b, whose explanatory variables are 100, 99, 98, and 97, which represent the remaining toner amounts on January 1st, January 2nd, January 3rd, and January 4th, respectively, is 96, which represents the remaining toner amount on January 5th.

[0076] The objective variable in the training data for remaining amount prediction model 24b, whose explanatory variables are 100, 99, 98, 97, and 96, which are the remaining toner amounts on January 1st, January 2nd, January 3rd, January 4th, and January 5th, respectively, is 95, which is the remaining toner amount on January 6th.

[0077] The objective variable in the training data for remaining amount prediction model 24b, whose explanatory variables are 100, 99, 98, 97, 96, and 95, which represent the remaining toner amounts on January 1st, January 2nd, January 3rd, January 4th, January 5th, and January 6th, respectively, is 94, which represents the remaining toner amount on January 7th.

[0078] The objective variable in the training data for remaining amount prediction model 24b, whose explanatory variables are 100, 99, 98, 97, 96, 95, and 94, which represent the remaining toner amounts on January 1st, January 2nd, January 3rd, January 4th, January 5th, January 6th, and January 7th, respectively, is 93, which represents the remaining toner amount on January 8th.

[0079] The model learning unit 25b generates a large amount of training data such as the training data described above in S141.

[0080] As shown in FIG. 12, after the process of S141 is completed, the model learning unit 25b generates the remaining amount prediction model 24b by performing training using the training data generated in S141 (S142).

[0081] The model learning unit 25b may generate the remaining amount prediction model 24b of the image forming device 30 by training using not only the teacher data generated based on the actual measurement value data 24c of the image forming device 30, but also teacher data generated based on the actual measurement value data of an image forming device other than the image forming device 30. When the model learning unit 25b generates the remaining amount prediction model 24b using teacher data generated based on the actual measurement value data of an image forming device other than the image forming device 30, the model learning unit 25b may generate the remaining amount prediction model 24b by giving more importance to the teacher data generated based on the actual measurement value data of an image forming device of the same model as the image forming device 30 than to the teacher data generated based on the actual measurement value data of an image forming device of a different model from the image forming device 30.

[0082] In the processing of S142, the model learning unit 25b adjusts the parameters of the remaining amount prediction model 24b so that the objective variable of the training data for all of the multiple training data, i.e., the loss indicating the magnitude of deviation between the correct data and the remaining amount of toner predicted by the remaining amount prediction model 24b, is minimized.

[0083] When the process of S142 is completed, the model learning unit 25b ends the operation shown in FIG.

[0084] Figure 14 is a flowchart of the operation of the remaining amount prediction system 20 when calculating Loss when the predicted day evaluation of the empty day predicted using the remaining amount prediction model 24b (hereinafter referred to as the "predicted day evaluation of the remaining amount prediction model 24b") has not been entered by the user.

[0085] As shown in FIG. 14, the model learning unit 25b inputs the explanatory variables of the teacher data into the remaining amount prediction model 24b to predict the remaining amount of toner one day after the latest base date of the explanatory variables (S161).

[0086] When the process of S161 is completed, the model learning unit 25b calculates a square error based on the objective variable of the training data, i.e., the correct data, and the remaining amount of toner predicted in S161 (S162). That is, the model learning unit 25b calculates the square of the difference between the correct data and the predicted remaining amount of toner corresponding to this correct data.

[0087] FIG. 15 is a diagram showing an example of the square error calculated in the operation shown in FIG.

[0088] As shown in FIG. 15, the model learning unit 25b calculates the square error for each correct answer data.

[0089] 14, when the process of S162 is completed, the model learning unit 25b calculates the square root of the average value of all the squared errors calculated in S162, that is, the RMSE (Root Mean Squared Error) as the loss (S163). For example, if the squared errors calculated in S162 are those shown in FIG. 15, the average value of all the squared errors calculated in S162 is (1+16+25+36+49+49+49) / 7, and therefore the loss calculated in S163 is √(32.142 ).

[0090] When the process of S163 is completed, the model learning unit 25b ends the operation shown in FIG.

[0091] The above describes the operation of the remaining amount prediction system 20 when generating a remaining amount prediction model 24b for the image forming device 30, but the operation of the remaining amount prediction system 20 when generating a remaining amount prediction model for an image forming device other than the image forming device 30 is also similar.

[0092] Next, the operation of the system 10 when a predicted day evaluation is input will be described.

[0093] FIG. 16 is a sequence diagram of the operation of the system 10 when the prediction date evaluation of the remaining amount prediction model 24b is input.

[0094] The user can input the start of the predicted date evaluation of remaining amount prediction model 24b into user terminal 40 from operation unit 41 of user terminal 40. For example, if a new replacement toner container is delivered so that it arrives to the user of image forming apparatus 30 approximately a specific number of days before the empty date predicted by remaining amount prediction model 24b, the user may input the start of the predicted date evaluation of remaining amount prediction model 24b into user terminal 40 from operation unit 41 of user terminal 40 when the new toner container arrives.

[0095] In addition, as a mechanism for delivering a new toner container so that a new replacement toner container arrives to the user of the image forming device 30 approximately a specific number of days before the empty date predicted by the remaining amount prediction model 24b, a mechanism may be used in which the remaining amount prediction system 20 automatically orders a new toner container from a system that delivers toner containers at a timing corresponding to the empty date predicted by the remaining amount prediction model 24b so that a new replacement toner container arrives to the user of the image forming device 30 approximately a specific number of days before the empty date predicted by the remaining amount prediction model 24b.

[0096] When the control unit 45 of the user terminal 40 is instructed to start the prediction date evaluation of the remaining amount prediction model 24b, the control unit 45 instructs the remaining amount prediction system 20 to start the prediction date evaluation of the remaining amount prediction model 24b, as shown in FIG. 16 (S181).

[0097] When the model learning unit 25b of the remaining amount prediction system 20 is instructed in S181 to start the predicted date evaluation of the remaining amount prediction model 24b, it instructs the user terminal 40 to display a screen for the predicted date evaluation of the remaining amount prediction model 24b (hereinafter referred to as the "predicted date evaluation screen") (S182).

[0098] When the display of the prediction date evaluation screen is instructed in S182, the control unit 45 of the user terminal 40 displays the prediction date evaluation screen on the display unit 42 (S183). Therefore, the user can input the prediction date evaluation of the remaining amount prediction model 24b to the prediction date evaluation screen via the operation unit 41 of the user terminal 40.

[0099] In addition, if a new toner container is delivered so that it arrives at the user of the image forming apparatus 30 approximately a certain number of days before the empty date predicted by the remaining amount prediction model 24b, the user may determine the predicted date evaluation of the remaining amount prediction model 24b based on the actual delivery date of the new toner container. For example, if the user determines that the actual delivery date of the new toner container is appropriate for the user's desired delivery date, i.e., just right, the user may enter 0 as the predicted date evaluation of the remaining amount prediction model 24b. If the user determines that the actual delivery date of the new toner container is inappropriate and too early for the user's desired delivery date, the user may enter a positive value as the predicted date evaluation of the remaining amount prediction model 24b. The greater the difference between the actual delivery date of the new toner container and the user's desired delivery date, the larger the value may be entered as the predicted date evaluation of the remaining amount prediction model 24b. If the user determines that the actual delivery date of the new toner container is inappropriate and later than the delivery date desired by the user, the user can simply input a negative value as the predicted date evaluation of the remaining amount prediction model 24b.The larger the difference between the actual delivery date of the new toner container and the delivery date desired by the user, the smaller the number can be input as the predicted date evaluation of the remaining amount prediction model 24b.

[0100] After processing S183, when the predicted date evaluation of the remaining amount prediction model 24b is entered into the predicted date evaluation screen displayed in S183, the control unit 45 of the user terminal 40 notifies the remaining amount prediction system 20 of the predicted date evaluation entered into the predicted date evaluation screen (S184).

[0101] When the model learning unit 25b of the remaining amount prediction system 20 is notified of the predicted day evaluation of the remaining amount prediction model 24b in S184, it associates the predicted day evaluation notified in S184 with the device ID of the image forming device 30, which is the image forming device that is the target of the remaining amount prediction model 24b, and stores it in the predicted day evaluation information 24d (S185).

[0102] The above describes the operation of system 10 when a predicted date evaluation of remaining amount prediction model 24b of image forming device 30 is input, but the operation of system 10 is similar when a predicted date evaluation of a remaining amount prediction model of an image forming device other than image forming device 30 is input.

[0103] Next, the operation of the remaining amount prediction system 20 when learning a remaining amount prediction model will be described.

[0104] The model learning unit 25b of the remaining amount prediction system 20 executes an operation similar to the operation shown in Fig. 12 at a specific timing with respect to the remaining amount prediction model 24b of the image forming device 30. For example, the model learning unit 25b may execute an operation similar to the operation shown in Fig. 12 with respect to the remaining amount prediction model 24b of the image forming device 30 every six months.

[0105] However, when the user has input the prediction date evaluation of the remaining amount prediction model 24b, the model learning unit 25b calculates the loss by the operation shown in FIG. 17, not by the operation shown in FIG.

[0106] FIG. 17 is a flowchart of the operation of the remaining amount prediction system 20 when calculating the loss in a state where the prediction date evaluation of the remaining amount prediction model 24b has been input by the user.

[0107] As shown in FIG. 17, the model learning unit 25b inputs the explanatory variables of the teacher data into the remaining amount prediction model 24b to predict the remaining amount of toner one day after the latest base date of the explanatory variables (S201).

[0108] When the process of S201 is completed, the model learning unit 25b calculates a square error based on the objective variable of the training data, i.e., the correct data, and the remaining amount of toner predicted in S201 (S202). That is, the model learning unit 25b calculates the square of the difference between the correct data and the predicted remaining amount of toner corresponding to this correct data.

[0109] When the processing of S202 is completed, the model learning unit 25b determines the predicted day evaluation of the remaining amount prediction model 24b of the image forming device 30 in the predicted day evaluation information 24d, i.e., the predicted day evaluation associated with the device ID of the image forming device 30 (S203).

[0110] When the model learning unit 25b determines in S203 that the prediction date evaluation of the remaining amount prediction model 24b in the prediction date evaluation information 24d is a negative value indicating "late," the model learning unit 25b processes the squared error calculated in S202 by multiplying the squared error calculated in S202 when the predicted remaining amount of toner is greater than the correct data by a coefficient corresponding to the prediction date evaluation of the remaining amount prediction model 24b in the prediction date evaluation information 24d (S204). That is, the model learning unit 25b weights the squared error according to the prediction date evaluation. Here, the coefficients corresponding to the prediction date evaluation of the remaining amount prediction model 24b in the prediction date evaluation information 24d are 1.5, 1.4, 1.3, 1.2, and 1.1 when the prediction date evaluation of the remaining amount prediction model 24b in the prediction date evaluation information 24d is -5, -4, -3, -2, and -1, respectively.

[0111] FIG. 18 is a diagram showing an example of the squared error processed in the operation shown in FIG. 17 when the prediction date evaluation of the remaining amount prediction model 24b in the prediction date evaluation information 24d is a negative value.

[0112] 18 is an example in which the predicted date evaluation of remaining amount prediction model 24b in predicted date evaluation information 24d is −1. In the example shown in FIG. 18, the predicted remaining amount of toner for January 1st is greater than the correct data, so the squared error for January 1st is multiplied by 1.1.

[0113] 17, when the process of S204 is completed, model learning unit 25b calculates, as Loss (S205), the square root of the average value of all square errors calculated in S202 that have not been processed in S204 and all square errors calculated in S202 that have been processed in S204. For example, when the square errors processed in S204 are those shown in FIG. 18, the average value of all square errors calculated in S202 that have not been processed in S204 and all square errors calculated in S202 that have been processed in S204 is (1.1 + 16 + 25 + 36 + 49 + 49 + 49) / 7, and therefore the Loss calculated in S205 is √(32.157 ).

[0114] If the model learning unit 25b determines in S203 that the prediction day evaluation of the remaining amount prediction model 24b in the prediction day evaluation information 24d is 0, which means "just right," the model learning unit 25b calculates the square root of the average value of all the squared errors calculated in S202, i.e., the RMSE, as the loss (S206). For example, if the squared errors calculated in S202 are those shown in Fig. 15, the average value of all the squared errors calculated in S202 is (1 + 16 + 25 + 36 + 49 + 49 + 49) / 7, and therefore the loss calculated in S206 is √(32.142...).

[0115] When the model learning unit 25b determines in S203 that the prediction date evaluation of the remaining amount prediction model 24b in the prediction date evaluation information 24d is a positive value indicating "early," the model learning unit 25b processes the squared error calculated in S202 by multiplying the squared error calculated in S202 when the predicted remaining amount of toner is smaller than the correct data by a coefficient corresponding to the prediction date evaluation of the remaining amount prediction model 24b in the prediction date evaluation information 24d (S207). That is, the model learning unit 25b weights the squared error according to the prediction date evaluation. Here, the coefficients corresponding to the prediction date evaluation of the remaining amount prediction model 24b in the prediction date evaluation information 24d are 1.1, 1.2, 1.3, 1.4, and 1.5 when the prediction date evaluation of the remaining amount prediction model 24b in the prediction date evaluation information 24d is 1, 2, 3, 4, and 5, respectively.

[0116] FIG. 19 is a diagram showing an example of the squared error processed in the operation shown in FIG. 17 when the prediction date evaluation of the remaining amount prediction model 24b in the prediction date evaluation information 24d is a positive value.

[0117] The example shown in Fig. 19 is a case where the prediction date evaluation of remaining amount prediction model 24b in prediction date evaluation information 24d is 1. In the example shown in Fig. 19, the predicted remaining toner amounts for January 2nd, January 3rd, January 4th, January 5th, January 6th, and January 7th are smaller than the correct data, so the squared errors for January 2nd, January 3rd, January 4th, January 5th, January 6th, and January 7th are multiplied by 1.1.

[0118] 17, when the process of S207 is completed, model learning unit 25b calculates, as Loss (S208), the square root of the average value of all squared errors calculated in S202 that have not been processed in S207 and all squared errors calculated in S202 that have been processed in S207. For example, when the squared errors processed in S207 are those shown in FIG. 19, the average value of all squared errors calculated in S202 that have not been processed in S207 and all squared errors calculated in S202 that have been processed in S207 is (1 + 17.6 + 27.5 + 39.6 + 53.9 + 53.9 + 53.9) / 7, and therefore the Loss calculated in S208 is √(35.342 ).

[0119] When the process of S205, S206, or S208 is completed, model learning unit 25b ends the operation shown in FIG.

[0120] The above describes the operation of the remaining amount prediction system 20 when learning the remaining amount prediction model 24b of the image forming device 30, but the operation of the remaining amount prediction system 20 when learning the remaining amount prediction model of an image forming device other than the image forming device 30 is also similar.

[0121] As described above, remaining amount prediction system 20 changes the method of calculating loss in accordance with the prediction date evaluation input by the user (S203 to S208), thereby improving the accuracy of prediction of the remaining amount of toner.

[0122] When "early" is input as the prediction date evaluation ("positive value" in S203), remaining amount prediction system 20 assigns a greater weight to the difference between the correct data and the remaining amount of toner predicted by remaining amount prediction model 24b when the remaining amount of toner predicted by remaining amount prediction model 24b is smaller than the correct data (S207) compared to when the remaining amount of toner predicted by remaining amount prediction model 24b is larger than the correct data, and calculates the loss (S208), thereby improving the accuracy of prediction of remaining amount of toner.

[0123] When "early" is input as the prediction date evaluation ("positive value" in S203), the remaining amount prediction system 20 calculates multiple squared errors as the square of the difference between the correct data and the remaining amount of toner predicted by the remaining amount prediction model 24b (S202), increases the squared error when the remaining amount of toner predicted by the remaining amount prediction model 24b is smaller than the correct data (S207), and calculates the loss as the square root of the average value of all squared errors when the remaining amount of toner predicted by the remaining amount prediction model 24b is not smaller than the correct data and all squared errors after the increase when the remaining amount of toner predicted by the remaining amount prediction model 24b is smaller than the correct data (S208), thereby improving the accuracy of prediction of the remaining amount of toner.

[0124] When "late" is input as the prediction date evaluation ("negative value" in S203), and the remaining amount of toner predicted by the remaining amount prediction model 24b is greater than the correct data, the remaining amount prediction system 20 assigns a greater weight to the difference between the correct data and the remaining amount of toner predicted by the remaining amount prediction model 24b (S204) and calculates the loss (S205) compared to when the remaining amount of toner predicted by the remaining amount prediction model 24b is smaller than the correct data, thereby improving the accuracy of the prediction of the remaining amount of toner.

[0125] When "late" is input as the prediction date evaluation ("negative value" in S203), the remaining amount prediction system 20 calculates multiple squared errors as the square of the difference between the correct data and the remaining amount of toner predicted by the remaining amount prediction model 24b (S202), increases the squared error when the remaining amount of toner predicted by the remaining amount prediction model 24b is greater than the correct data (S204), and calculates the loss as the square root of the average of all squared errors when the remaining amount of toner predicted by the remaining amount prediction model 24b is not greater than the correct data and all squared errors after the increase when the remaining amount of toner predicted by the remaining amount prediction model 24b is greater than the correct data (S205), thereby improving the accuracy of prediction of the remaining amount of toner.

[0126] In this embodiment, the remaining amount prediction system 20 accepts the predicted day evaluation on an 11-point scale. However, the remaining amount prediction system 20 may accept the predicted day evaluation on a scale other than 11 points.

[0127] The remaining amount prediction system 20 may employ coefficients different from those described in this embodiment in S204 or S207.

[0128] In this embodiment, the remaining amount prediction system 20 predicts the empty date. However, the remaining amount prediction system 20 may predict the date when the remaining amount of the consumable item will be equal to or less than a specific amount other than 0%.

[0129] In this embodiment, the consumable of the present invention is toner used in an image forming apparatus, but the consumable of the present invention may be a consumable other than toner used in an image forming apparatus, or may be a consumable used in an electronic device other than an image forming apparatus. [Explanation of symbols]

[0130] 20 Remaining amount prediction system (computer) 24a Remaining amount prediction program 24b Remaining amount prediction model 25a Remaining amount prediction section 25b Model Learning Section 30 Image forming equipment (electronic equipment)

Claims

1. a remaining amount prediction unit that predicts the date when the remaining amount of the consumable will be equal to or less than a specific amount by using a remaining amount prediction model as a machine learning model that predicts the remaining amount of the consumable used in the electronic device; a model learning unit that executes learning of the remaining amount prediction model; Equipped with the model learning unit, when learning the remaining amount prediction model, adjusts parameters of the remaining amount prediction model so as to minimize a loss indicating a magnitude of deviation between correct answer data and the remaining amount of the consumable predicted by the remaining amount prediction model for all of a plurality of training data; The remaining quantity prediction system is characterized in that the model learning unit receives a predicted day evaluation from the user as an evaluation of the day predicted by the remaining quantity prediction unit for the day on which the remaining quantity of the consumable will actually be below the specified amount, and changes the method of calculating the loss in accordance with the input predicted day evaluation.

2. The predicted date evaluation includes: "Just right" indicating that the day predicted by the remaining amount prediction unit is appropriate for the day when the remaining amount of the consumable will actually be equal to or less than the specific amount; "Early" indicates that the date predicted by the remaining amount prediction unit is inappropriate and early with respect to the date when the remaining amount of the consumable will actually be equal to or less than the specific amount. exists at least, The remaining quantity prediction system described in claim 1, characterized in that when ``early'' is input as the predicted date evaluation, when the remaining quantity of the consumable predicted by the remaining quantity prediction model is smaller than the correct data, the model learning unit calculates the loss by weighting the difference between the correct data and the remaining quantity of the consumable predicted by the remaining quantity prediction model more heavily than when the remaining quantity of the consumable predicted by the remaining quantity prediction model is larger than the correct data.

3. when the prediction date evaluation is "just right," the model learning unit calculates a plurality of squared errors as the squares of the differences between the correct answer data and the remaining amount of the consumable predicted by the remaining amount prediction model, and calculates the square root of the average value of all the calculated squared errors as the Loss; The remaining quantity prediction system described in claim 2, characterized in that when ``early'' is input as the predicted date evaluation, the model learning unit calculates multiple squared errors, increases the squared error when the remaining quantity of the consumable predicted by the remaining quantity prediction model is smaller than the correct data, and calculates the Loss as the square root of the average of all the squared errors when the remaining quantity of the consumable predicted by the remaining quantity prediction model is not smaller than the correct data and all the squared errors after the increase when the remaining quantity of the consumable predicted by the remaining quantity prediction model is smaller than the correct data.

4. The predicted date evaluation includes: "Just right" indicating that the day predicted by the remaining amount prediction unit is appropriate for the day when the remaining amount of the consumable will actually be equal to or less than the specific amount; "Late" indicates that the day predicted by the remaining amount prediction unit is inappropriate and late with respect to the day when the remaining amount of the consumable will actually be equal to or less than the specific amount. exists at least, The remaining quantity prediction system described in claim 1, characterized in that when ``late'' is input as the predicted date evaluation, when the remaining quantity of the consumable predicted by the remaining quantity prediction model is greater than the correct data, the model learning unit calculates the loss by weighting the difference between the correct data and the remaining quantity of the consumable predicted by the remaining quantity prediction model more heavily than when the remaining quantity of the consumable predicted by the remaining quantity prediction model is smaller than the correct data.

5. when the prediction date evaluation is "just right," the model learning unit calculates a plurality of squared errors as the squares of the differences between the correct answer data and the remaining amount of the consumable predicted by the remaining amount prediction model, and calculates the square root of the average value of all the calculated squared errors as the Loss; The remaining quantity prediction system described in claim 4, characterized in that when ``late'' is input as the predicted date evaluation, the model learning unit calculates multiple squared errors, increases the squared error when the remaining quantity of the consumable predicted by the remaining quantity prediction model is greater than the correct data, and calculates the Loss as the square root of the average of all the squared errors when the remaining quantity of the consumable predicted by the remaining quantity prediction model is not greater than the correct data and all the squared errors after the increase when the remaining quantity of the consumable predicted by the remaining quantity prediction model is greater than the correct data.

6. a remaining amount prediction unit that predicts the date when the remaining amount of the consumable will be equal to or less than a specific amount by using a remaining amount prediction model as a machine learning model that predicts the remaining amount of the consumable used in the electronic device; a model learning unit that executes learning of the remaining amount prediction model; By implementing this on a computer, the model learning unit, when learning the remaining amount prediction model, adjusts parameters of the remaining amount prediction model so as to minimize a loss indicating a magnitude of deviation between correct answer data and the remaining amount of the consumable predicted by the remaining amount prediction model for all of a plurality of training data; The remaining quantity prediction program is characterized in that the model learning unit receives a predicted day evaluation from the user as an evaluation of the day predicted by the remaining quantity prediction unit for the day on which the remaining quantity of the consumable will actually be below the specified amount, and changes the method of calculating the loss in accordance with the input predicted day evaluation.

Images