WO2021210107A1 - Model creation device, model creation method, and model creation program - Google Patents

Abstract

The present invention creates a highly accurate model even in a case where time constants of an explained variable and an explanatory variable change according to a condition by: a selection unit (22) selecting, from a data series DB (31) in which each data series of the explanatory variable and the explained variable are stored, a prescribed number of data series of the explanatory variable in another period, which are similar to a data series of the explanatory variable in a prescribed period including a time point to be estimated; and a creation unit (23) creating a model for estimating the explained variable at the time point to be estimated from a data series of the explanatory variable in a prescribed period including the time point to be estimated on the basis of the selected data series of the explanatory variable in the other period and data of the explained variable at a time point corresponding to the time point to be estimated in the other period for a plurality of cases where the widths of the prescribed period are set to be different.

Description

Model creation device, model creation method, and model creation program

The disclosed technology relates to a model creation device, a model creation method, and a model creation program.

Conventionally, there is a technique for creating a machine learning model used when estimating a certain value from other measurement data. For example, when a certain numerical value that is an explained variable is estimated from another numerical value that is an explanatory variable and has a causal relationship with the explained variable but a change rate is different from that of the explained variable, it is an estimation target. A method has been proposed in which data before and after the time point is also used as an explanatory variable (see Non-Patent Document 1). As a result, this method absorbs the difference in the rate of change between the explanatory variable and the explained variable, that is, the difference in the time constant.

The conventional technology can handle cases where the time constant between parameters is constant, but there is a problem that a highly accurate model cannot be created when the time constant between parameters in the model changes depending on the situation. As an example, when predicting the number of people staying from the carbon dioxide (CO2) concentration in a room, not only the time constant differs greatly depending on the CO2 concentration and the change in the number of people, but also the CO2 emitted by the person depends on the ventilation capacity at that time. The degree and time constant that affect the CO2 concentration can change.

The disclosed technology was made in view of the above points, and aims to create a highly accurate model even in cases where the time constants of the explained variable and the explanatory variable change depending on the situation.

The first aspect of the present disclosure is a model creation device, which is similar to the data series of the explanatory variables in a predetermined period including the time to be estimated from the storage unit in which the data series of the explanatory variables and the explained variables are stored. , The data of the explanatory variables in the other period selected by the selection unit in the case of a selection unit for selecting a predetermined number of data series of explanatory variables in other periods and a plurality of cases in which the widths of the predetermined periods are different from each other. Based on the series and the data of the explained variable at the time corresponding to the estimated target time in the other period, the explained from the data series of the explanatory variables in the predetermined period including the estimated target time is explained. Includes a creation unit that creates a model for estimating variables.

The second aspect of the present disclosure is a model creation method executed by a model creation device including a selection unit and a creation unit, in which the selection unit stores data sequences of explanatory variables and explained variables. A predetermined number of explanatory variable data series in other periods similar to the explanatory variable data series in the predetermined period including the estimation target time point are selected from the storage unit, and the creating unit makes the width of the predetermined period different. In the plurality of cases, based on the data series of the explanatory variables in the other period selected by the selection unit and the data of the explained variables at the time corresponding to the estimation target time in the other period. This is a method of creating a model for estimating the explained variable at the estimation target time from the data series of the explanatory variables in the predetermined period including the estimation target time.

The third aspect of the present disclosure is a model creation program, which is a program for causing a computer to function as each part constituting the above-mentioned model creation device.

According to the disclosed technology, it is possible to create a highly accurate model even in the case where the time constants of the explained variable and the explanatory variable change depending on the situation.

It is a block diagram which shows the hardware configuration of a model making apparatus. It is a block diagram which shows the example of the functional structure of the model making apparatus which concerns on 1st Embodiment. It is a flowchart which shows the flow of the model creation process in 1st Embodiment. It is a block diagram which shows the example of the functional structure of the model making apparatus which concerns on 2nd and 3rd Embodiment. It is a figure which shows an example of a data series DB. It is a flowchart which shows the flow of the model creation process in 2nd Embodiment. It is a figure which shows an example of a model DB. It is a flowchart which shows the flow of the model creation process in 3rd Embodiment.

Hereinafter, an example of the embodiment of the disclosed technology will be described with reference to the drawings. The same reference numerals are given to the same or equivalent components and parts in each drawing. In addition, the dimensional ratios in the drawings are exaggerated for convenience of explanation and may differ from the actual ratios.

<First Embodiment>
In the first embodiment, a model for estimating the explained variable at the time of the estimation target time is created based on the data series of the explanatory variables, and a model creation device for estimating the explained variable at the time of the estimation target time is described.

FIG. 1 is a block diagram showing a hardware configuration of the model creation device 10.

As shown in FIG. 1, the model creation device 10 includes a CPU (Central Processing Unit) 11, a ROM (Read Only Memory) 12, a RAM (Random Access Memory) 13, a storage 14, an input unit 15, a display unit 16, and a communication. It has an I / F (Interface) 17. The configurations are connected to each other via the bus 19 so as to be communicable with each other.

The CPU 11 is a central arithmetic processing unit that executes various programs and controls each part. That is, the CPU 11 reads the program from the ROM 12 or the storage 14, and executes the program using the RAM 13 as a work area. The CPU 11 controls each of the above configurations and performs various arithmetic processes according to the program stored in the ROM 12 or the storage 14. In the present embodiment, the ROM 12 or the storage 14 stores a model creation program for executing the model creation process described later.

ROM 12 stores various programs and various data. The RAM 13 temporarily stores a program or data as a work area. The storage 14 is composed of a storage device such as an HDD (Hard Disk Drive) and an SSD (Solid State Drive), and stores various programs including an operating system and various data.

The input unit 15 includes a pointing device such as a mouse and a keyboard, and is used for performing various inputs.

The display unit 16 is, for example, a liquid crystal display and displays various types of information. The display unit 16 may adopt a touch panel method and function as an input unit 15.

The communication interface 17 is an interface for communicating with other devices. For the communication, for example, a wired communication standard such as Ethernet (registered trademark) or FDDI, or a wireless communication standard such as 4G, 5G, or Wi-Fi (registered trademark) is used.

Next, the functional configuration of the model creation device 10 will be described. FIG. 2 is a block diagram showing an example of the functional configuration of the model creation device 10.

As shown in FIG. 2, the data series DB (Database) 31 stores each data series of the explanatory variable and the explained variable acquired by the data acquisition unit 32. The data acquisition unit 32 is, for example, an environment sensor installed in the real world that measures room temperature, CO2 concentration, etc., a motion sensor that counts the number of people existing in a room, software that transmits system logs accumulated in a server, and the like. Is. The explanatory variables may be one type or a plurality of types. The data sequence read from the data sequence DB 31 is input to the model creation device 10. The data series DB 31 is an example of a storage unit of the disclosed technology.

Further, the set value is input to the model creation device 10 from the set value input unit 33. The setting value input unit 33 inputs the setting values related to the creation and selection (details will be described later) of the model used for estimating the explained variable at the time of the estimation target to the model creation device 10. Specific examples of the set values will be described later.

Further, the model creation device 10 outputs the estimation result estimated using the model.

As shown in FIG. 2, the model creation device 10 has an estimation unit 21, a selection unit 22, and a creation unit 23 as functional configurations. Each functional configuration is realized by the CPU 11 reading the model creation program stored in the ROM 12 or the storage 14 and expanding and executing the model creation program in the RAM 13.

The estimation unit 21 passes the information for selecting the data series necessary for creating the model to the selection unit 22, and instructs the selection unit 22 to select the data series for creating the model. Specifically, the estimation unit 21 has a time window in the past direction from the estimation target time point (hereinafter referred to as "previous time window") and a time window in the future direction from the estimation target time point (hereinafter, "after time"). At least one of the windows) is used to set the width of the predetermined period including the estimated target time point. In the present embodiment, a case where the width of a predetermined period is set by using both the front time window and the back time window will be described.

More specifically, the estimation unit 21 acquires the set value input from the set value input unit 33. In the present embodiment, the estimating unit 21, a set value, the estimated target time point t, the minimum value ∂t _1min before time window, after the minimum value ∂t _2min time window before the time window of maximum ∂t _1max, after _{The maximum value ∂t 2max} of the time window, the number of selected data n, and the type M of the model to be created are acquired. Estimation unit 21, the size ∂t ₁ before time window in the range of ∂t _1min ~ ∂t _1max, by changing the size ∂t ₂ after time window in the range of ∂t _2min ~ ∂t _2max, width Sets a plurality of predetermined periods, each of which is different. That is, the estimating unit 21 sets up t-∂t ₁ after t + ∂t ₂ as the predetermined period.

Estimation unit 21, at predetermined time intervals set, the data series DB31, acquires a data series of explanatory variables in the period _{t-∂t 1 ~ t + ∂t} 2, a vector X by combining the acquired data sequence do. When there is only one type of explanatory variable, each data included in the acquired data series becomes each element of the vector X. When there are a plurality of types of explanatory variables, the vector X is a list of all the data included in the data series of the explanatory variables of each type. The estimation unit 21 passes the vector X together with various set values to the selection unit 22.

Further, after the model is created, the estimation unit 21 uses the optimum model in the created model and the data series of the explanatory variables in the predetermined period including the estimation target time point to explain the variable at the estimation target time point. To estimate. Specifically, the estimation unit 21 selects the model with the highest evaluation of accuracy from the models created by the creation unit 23, which will be described later. Estimation unit 21, a predetermined period including the estimated target time point t, the data series of explanatory variables in a predetermined period of set width when creating the selected model _{(t-∂t 1 ~ t +} ∂t 2) Is acquired from the data series DB 31. Then, the estimation unit 21 inputs the vector X in which the acquired data series of the explanatory variables are combined into the selected model, and estimates the explained variable at the estimation target time point t.

The selection unit 22 selects a predetermined number of explanatory variable data series in other periods that are similar to the explanatory variable data series in the predetermined period including the estimation target time point from the data series DB 31.

Specifically, the selection unit 22 selects the data series of the explanatory variables similar to the data series indicated by the vector X passed from the estimation unit 21, and the data of the explained variables corresponding to the data series of the explanatory variables. , Select n data series DB31. For example, the selection unit 22 can use the sum of the average values of the pairwise distances between the data series, the correlation coefficient, the cosine similarity of the vectors, and the like as the similarity between the data series. Then, the selection unit 22 selects a data series of explanatory variables in another period in which the degree of similarity with the vector X satisfies a predetermined condition. The selection unit 22 combines the data series of the selected n explanatory variables into a vector X'. Further, the selection unit 22 selects n data of the explained variables at the time corresponding to the estimation target time point in another period, and combines the data of the selected n explained variables to form a vector Y'. .. Then, the selection unit 22 passes the vectors X'and Y'to the creation unit 23 together with various set values.

The creating unit 23 estimates the explained variable at the time to be estimated based on the data series of the explanatory variables and the data of the explained variables selected by the selection unit 22 for each of a plurality of cases having different widths of the predetermined period. Create a model. Examples of the model include a multiple regression model, a support vector regression, a non-linear regression model such as Random Forest regression, a neural network model, and the like. Specifically, the creating unit 23 learns the relationship between the vector X'and the vector Y'passed from the selection unit 22, and derives the parameters of the model according to the model type M to obtain the model. To create.

The creation unit 23 calculates an evaluation value indicating the accuracy of each of the created models for each of a plurality of cases where the width of the predetermined period is different. For example, the creation unit 23 inputs each of the data series of explanatory variables constituting the vector X'passed from the selection unit 22 into the model, and the error between each of the estimated values and each of the correct answer values. The average value can be calculated as an evaluation value. The correct answer value is the data of the explained variable constituting the vector Y'passed from the selection unit 22, that is, the data of the explained variable at the time corresponding to the estimation target time in another period corresponding to the predetermined period. can do. The creation unit 23 stores each of the created models in the model DB 25 in association with the width of the predetermined period set when the model was created and the calculated evaluation value. The model DB 25 is an example of a model storage unit of the disclosed technology.

Next, the operation of the model creation device 10 according to the first embodiment will be described.

FIG. 3 is a flowchart showing the flow of the model creation process by the model creation device 10. The model creation process is performed by the CPU 11 reading the model creation program from the ROM 12 or the storage 14, expanding it into the RAM 13 and executing the model creation program.

In step S11, the CPU 11 acquires the set value input from the set value input unit 33 as the estimation unit 21. Specifically, CPU 11, as the estimating unit 21, estimation target time t, before the time window minimum ∂t _1min of, after the time window minimum ∂t _2min of the previous time window maximum ∂t _1max, after time The maximum value of the window ∂t _2max , the number of selected data n, and the type M of the model to be created are acquired.

Next, in step S12, the CPU 11 sets the size of the front time window ∂t ₁ to ∂t _{1 min} and the size of the rear time window ∂t ₂ to ∂t _{2 min} as the estimation unit 21.

Next, in step S13, CPU 11 may, as an estimated 21, from the data series DB31, a vector X by combining the data sequence of the explanatory variable in _{t-∂t 1 ~ t + ∂t} 2, together with the various set values, selection Hand over to department 22.

Next, in step S14, the CPU 11, as the selection unit 22, obtains data of the explanatory variable data series in another period and the corresponding explained variable data similar to the vector X passed from the estimation unit 21. Select n items from the series DB 31. Then, as the selection unit 22, the CPU 11 uses the vector X'that combines the data series of the selected n explanatory variables and the vector Y'that combines the data of the selected n explained variables together with various setting values. Hand over to the creation unit 23.

Next, in step S15, the CPU 11 creates the model f by learning the relationship between the vector X'and the vector Y'as the creating unit 23 and deriving the parameters of the model according to the model type M. do.

Next, in step S16, the CPU 11 calculates, for example, the average value of the error between the estimated value and the correct answer value as the evaluation value score of the model f as the creating unit 23. Then, the CPU 11 stores the calculated evaluation value score as the creation unit 23 in the model DB 25 together with the model _{f and the currently set ∂t 1} and ∂t _2.

Next, in step S17, the CPU 11 determines whether or not _{∂t 1} = ∂t _{1max as the estimation unit 21.} When ∂t ₁ = ∂t _1max , the process proceeds to step S19. On the other hand, when ∂t ₁ ≠ ∂t _1max , the process proceeds to step S18, the CPU _{1 adds 1 to ∂t 1} as the estimation unit 21, and the process returns to step S13.

In step S19, the CPU 11 determines whether or not _{∂t 2} = ∂t _{2max as the estimation unit 21.} When ∂t ₂ = ∂t _2max , the process proceeds to step S21. On the other hand, when ∂t ₂ ≠ ∂t _2max , the process proceeds to step S20, and the CPU 1 adds 1 to _{∂t 2} and sets _{∂t 1} to ∂t _{1min as the estimation unit 21.} , The process returns to step S13.

In step S21, the CPU 11 serves as the estimation unit 21 to store the model f having the smallest score, the size ∂t _{1 of} the front time window, and the size of the back time window stored in association with the model f from the model DB 25. to get the ∂t _2. Then, CPU 11, as the estimator 21, the data series DB31, acquires a data series of explanatory variables in the _{t-∂t 1 ~ t + ∂t} 2, to obtain the vector X that combines acquired data series.

Next, in step S22, the CPU 11 serves as the estimation unit 21, and based on the acquired explanatory variable data series X and the model f, the estimated value Y ^ of the explained variable at the estimation target time point t (in FIG. 3, “ "^ (Hat)") is estimated on "Y" and output as the estimation result. Then, the model creation process ends.

As described above, in the model creation device according to the first embodiment, the data series of the explanatory variables in the predetermined period including the estimation target time point is obtained from the storage unit in which the data series of the explanatory variables and the explained variables are stored. Select a predetermined number of similar data series of explanatory variables in other periods. In addition, the model creation device has a data series of explanatory variables in other selected periods and an explained variable at a time point corresponding to an estimation target time point in other periods in a plurality of cases in which the widths of predetermined periods are different from each other. Based on the data of, a model is created to estimate the explained variable at the estimation target time from the data series of the explanatory variables in the predetermined period including the estimation target time. As a result, a highly accurate model can be created even in the case where the time constants of the explained variable and the explanatory variable change depending on the situation.

Further, the model creation device evaluates the accuracy of each of the created models, and stores each of the created models in association with the width of the predetermined period set at the time of creating the model and the evaluation of the accuracy. Keep it. Then, the model creation device selects the model having the highest evaluation of accuracy from the created models. Further, the data series of the explanatory variables in the predetermined period including the estimation target time point and the width set when the selected model was created is input to the selected model to be explained at the estimation target time point. Estimate the variable. As a result, highly accurate estimation can be performed even in the case where the time constants of the explained variable and the explanatory variable change depending on the situation.

<Second Embodiment>
Next, the second embodiment will be described. In the second embodiment, the same configuration as the model creation device 10 according to the first embodiment is applied to the creation and estimation of a model for estimating the number of people staying in the room from the carbon dioxide (CO2) concentration in the room where the ventilator is located. This case will be described. Since the hardware configuration of the model creation device according to the second embodiment is the same as that of the model creation device 10 according to the first embodiment, the description thereof will be omitted.

As shown in FIG. 4, the data series DB 231 stores the data series measured by the CO2 sensor 232A, the ventilator controller 232B, and the number of people counter 232C, which are examples of the data acquisition unit 32 in the first embodiment. .. The CO2 sensor 232A is installed in the room to be measured, and measures and outputs the CO2 concentration in the room at each time. The ventilator controller 232B measures and outputs the output of the inverter of the ventilator (hereinafter referred to as "INV output") at each time as an index indicating the ventilation capacity of the ventilator. The number of people counter 232C detects a person staying in the room with a motion sensor or the like, and counts and outputs the number of people detected at each time.

FIG. 5 shows an example of the data series DB 231. In the example of FIG. 5, the data series DB231 shows the number of people staying in the room to be measured, the CO2 concentration, and the CO2 concentration every 10 minutes from 0:00 on August 1, 2019 to 17:50 on December 1, 2019. The INV output is stored. Also, from 8:00 to 17:50 on December 1, 2019, the CO2 concentration and INV output are memorized every 10 minutes, but the number of people staying cannot be measured, and "null" is memorized. There is. In the second embodiment, "CO2 concentration" and "INV output" are set as explanatory variables, "number of people staying" is set as an explained variable, and the number of people staying in null is the CO2 concentration and INV output at the same time and before and after the same time. Estimate from.

Further, as shown in FIG. 4, a set value is input to the model creation device 210 according to the second embodiment from the input unit 233 which is an example of the set value input unit 33 in the first embodiment. The input unit 233 is an input device such as a keyboard and a touch panel display.

As shown in FIG. 4, the model creation device 210 has an estimation unit 221, a selection unit 222, and a creation unit 223 as functional configurations. Each functional configuration is the same as each functional configuration of the model creation device 10 according to the first embodiment. Hereinafter, in the description of the operation, specific processing of each functional configuration according to the specific example applied in the second embodiment will be described.

Next, the operation of the model creation device 210 according to the second embodiment will be described.

FIG. 6 is a flowchart showing the flow of the model creation process by the model creation device 210. The model creation process is performed by the CPU 11 reading the model creation program from the ROM 12 or the storage 14, expanding it into the RAM 13 and executing the model creation program.

In step S211 the CPU 11 acquires the set value input from the input unit 233 as the estimation unit 221. Specifically, the CPU 11 acquires the following set values as the estimation unit 221.

・ Estimated target date and time t: December 1, 2019 8:00
・ Estimated target (explained variable): Number of people staying ・ Minimum value of previous time window ∂t _1min : 3
・ Maximum value of the previous time window ∂t _1max : 12
・ Minimum value of the after-time window ∂t _2min : 3
・ Maximum value of the later time window ∂t _2max : 12
-Number of selected data n: 50
・ Model type M: Multiple regression

Next, in step S212, the CPU 11 sets the size ∂t ₁ of the front time window to ∂t _{1 min} and the size of the rear time window ∂t ₂ to ∂t _{2 min} as the estimation unit 221.

Next, in step S213, CPU 11 has the estimation unit 221, CO2 from the data sequence DB 231, to _{t-∂t 1 ~ t + ∂t} 2, that is, from 2019 December 1 7:30 until the 8:30 Acquire the data series of concentration and INV output. Then, the CPU 11 passes the vector X, which is a combination of the acquired data series, to the selection unit 222 together with various set values as the estimation unit 221.

Next, in step S214, the CPU 11, as the selection unit 222, has a data series of CO2 concentration and INV output in another period similar to the vector X passed from the estimation unit 221 and data of the corresponding explained variable. And n are selected from the data series DB 231. Specifically, the selection unit 222 is a vector that combines the CO2 concentration data series and the INV output data series in the period defined by each time t', the front time window, and the back time window in the data series DB231. , The distance from the vector X acquired in step S213 is calculated.

For example, assuming that the time t'is 13:30 on October 14, 2019, here, the data series of CO2 concentration and the data series of INV output from 13:00 to 14:00 on October 14, 2019 are used. The combined vector is obtained. Further, the selection unit 222 adds, for example, a value obtained by adding the average value of the pairwise distance for each data series of the CO2 concentration and the INV output data series, and the distance between the vector and the vector X obtained at the time t'dist _t, Calculated as _t'.

As a more specific example, the data series of CO2 concentration when t'= 13:30 on October 14, 2019 is (564,568,579,592,601,599,597), and the data series of INV output. Is (33.5, 35.3, 34.9, 37.6, 37.1, 36.7, 36.0). In this case, dist _{t, t'} = (| 564-489 | + | 568-499 | + | 579-501 | + | 592-499 | + | 601-504 | + | 599-521 | + | 597-527 | + | 33.5-29.6 | + | 35.3-29.7 | + | 34.9-29.9 | + | 37.6-28.7 | + | 37.1-28.6 | + | 36.7-29.6 | + | 36.0-30.1 |) ÷ 14 = 43.2.

The selection unit 222 calculates the dist t, t'for all t'in which the number of people staying in the data series DB 231 is stored within the period defined by the time window _{and t'.} In the example of FIG. 5, t'is the date and time from 0:30 on August 1, 2019 to 7:20 on December 1, 2019. The selection unit 222 selects the top n = 50 vectors as the vector X'in ascending order of _{the calculated dust t, t'.} Further, the selection unit 222 combines n data of the number of people staying at t'corresponding to each vector X'to form a vector Y'. Then, the CPU 11 passes the vector X'and the vector Y'to the creating unit 223 together with various set values as the selection unit 22.

Next, in step S215, the CPU 11 acts as the creation unit 223 as a parameter of the multiple regression model based on X', Y', and the model type M, and the coefficient of the term corresponding to each explanatory variable in the multiple regression equation. Calculate the coefficient vector α that combines. As a result, a multiple regression model (hereinafter referred to as “model α”) represented by using the coefficient vector α is created.

Next, in step S216, the CPU 11 calculates _{the evaluation values score ∂t1 and ∂t2 of the model α as the creation unit 223.} score _{∂t1 and ∂t2} can be, for example, the average value of the errors between the correct answer value Y'and the estimated value estimated based on the corresponding X'and the model α. The CPU 11 stores the calculated score _{∂t1 and} ∂t2 in the model DB 225 together with α, ∂t ₁ and ∂t _{2 as the creation unit 223.}

Next, in step S217, since ∂t ₁ = 3 ≠ ∂t _1max = 12, the process _{proceeds to step S218, ∂t 1} = 4, and returns to step S213 to perform the same process. When this is repeated _{and the process reaches step S217 with ∂t 1} = 12, the process proceeds to step S219.

In step S219, since ∂t ₂ = 3 ≠ ∂t _2max = 12, ∂t ₂ = 4 and ∂t ₁ = ∂t _1min = 3 in step S220, and the process returns to step S213 again. This is repeated, and finally, when ∂t ₁ = ∂t _1max and ∂t ₂ = ∂t _2max , a positive determination is made in both step S217 and step S219, and the process proceeds to step S221. At this point, as in the model DB225 shown in FIG. 7, a model α is created for each combination _{of all ∂t 1} and ∂t _{2 in the range specified by the set value, and the evaluation value is score ∂t1. , ∂t2} is calculated.

Next, in step S221, the CPU 11 serves as the estimation unit 221 from the model DB 225 to _{display the model α having the smallest score ∂t1 and ∂t2 and} the size ∂t of the previous time window stored in association with the model α. _{Obtain 1} and the size ∂t ₂ of the later time window. For example, if _{score ∂t1, ∂t2 = 0.26 at ∂t 1} = 6 and ∂t ₂ = 9 is the minimum, the CPU 11 uses the estimation unit 221 as the model α, ∂t _{1 in that case.} = 6 and ∂t ₂ = 9 are acquired. Then, CPU 11, as the estimator 221, the data series DB 231, acquires the CO2 concentration and the data series of the INV output at _{t-∂t 1 ~ t + ∂t} 2, to obtain the vector X that combines acquired data sequence .. In this example, the CO2 concentration and INV output from 7:00 to 9:30 on December 1, 2019 are acquired and combined to obtain a vector X.

Next, in step S222, the CPU 11 estimates the estimated value Y ^ of the number of people staying at the estimation target time point t based on X and α as the estimation unit 221 and outputs it as the estimation result, and the model creation process ends. do.

As described above, according to the model creation device according to the second embodiment, when the first embodiment is applied to a specific example, the same effect as that of the model creation device according to the first embodiment can be obtained. ..

<Third Embodiment>
Next, the third embodiment will be described. In the third embodiment, in the same specific example as in the second embodiment, a case where an optimum value is determined for the number of selected data n will be described. Since the hardware configuration of the model creation device according to the third embodiment is the same as that of the model creation device 10 according to the first embodiment, the description thereof will be omitted. Further, in the model creation device according to the third embodiment, the same components as those of the model creation device 210 according to the second embodiment are designated by the same reference numerals and detailed description thereof will be omitted.

As shown in FIG. 4, the model creation device 310 according to the third embodiment has an estimation unit 321, a selection unit 222, and a creation unit 223 as functional configurations.

The estimation unit 321 is a range input as a set value instead of the estimation unit 221 in the second embodiment passing a fixed value input from the input unit 233 to the selection unit 222 as the number of selection data n. Each of the selection data number n different from each other is passed to the selection unit 222.

FIG. 8 is a flowchart showing the flow of the model creation process by the model creation device 310. The model creation process is performed by the CPU 11 reading the model creation program from the ROM 12 or the storage 14, expanding it into the RAM 13 and executing the model creation program. In the model creation process of the third embodiment, the same process as the model creation process (FIG. 6) of the second embodiment is assigned the same step number and detailed description thereof will be omitted.

In step S311 the CPU 11 acquires the set value input from the input unit 233 as the estimation unit 321. Specifically, the CPU 11 acquires the following set values as the estimation unit 321.

・ Estimated target date and time t: December 1, 2019 8:00
・ Estimated target (explained variable): Number of people staying ・ Minimum value of previous time window ∂t _1min : 3
・ Maximum value of the previous time window ∂t _1max : 12
・ Minimum value of the after-time window ∂t _2min : 3
・ Maximum value of the later time window ∂t _2max : 12
-Minimum value of the number of selected data n _min : 50
_-Maximum number of selected data n max: 100
・ Model type M: Multiple regression

Regarding the number of selected data, it differs from the second embodiment in that the maximum value and the minimum value are included instead of the fixed value.

Next, in step S312, the CPU 11 sets the size of the front time window ∂t ₁ to ∂t _{1 min} , the size of the rear time window ∂t ₂ to ∂t _{2 min} , and the number of selected data n to n _{min as the estimation unit 321.} Are set respectively.

Next, from steps S213 to S220, the same processing as in the second embodiment is executed. _{Incidentally,} score ∂t1, _{∂t2, n} is in the step _S216, score ∂t1 in the second embodiment _is similar to _∂t2, for selecting the number of data n is also variable, and adding n to the index ..

When both steps S217 and S219 are positive, the process proceeds to step S319. In step S319, the CPU 11 determines whether or not _{n = n max as the estimation unit 321.} When n = n _max , the process proceeds to step S221, and when n ≠ n _max , the process proceeds to step S320.

In step S320, CPU 11 has the estimation unit 321, as well as adding 1 to n, to set the ∂t _1min to ∂t _1, set the ∂t _2min to ∂t _2, the process returns to step S213.

Next, in step S221, the CPU 11 serves as the estimation unit 321 _{to store the model α in which score ∂t1, ∂t2, n} is the smallest from the model DB225, and the size of the previous time window stored in association with the model α. Get ∂t ₁ and the size of the later time window ∂t ₂ . For example, here _{, assuming that score ∂t1, ∂t2, n = 0.21 at ∂t 1} = 4, ∂t ₂ = 8, n = 60 is the minimum, the CPU 11 uses the estimation unit 321 as the estimation unit 321. In that case, the models α, ∂t ₁ = 4, and ∂t ₂ = 8 are acquired.

Hereinafter, the same processing as in the second embodiment is executed, and the model creation processing is completed.

As described above, according to the model creation device according to the third embodiment, in addition to the time window when selecting the data series of the explanatory variables, the optimum value of the number of selected data is also determined from the set range. Since the optimum value of the number of selected data may change depending on the time of estimation, a more accurate model should be created by determining the optimum value of the number of selected data as in the third embodiment. Can be done. In addition, a model created using the optimum number of selected data enables more accurate estimation.

In each of the above embodiments, the case where the front time window and the back time window are increased by 1 has been described, but an increase width may be added as a set value to increase by the set width. Similarly, the number of selected data in the third embodiment may be increased by the set width by adding an increase width as a set value.

Further, in each of the above embodiments, the case where all kinds of explanatory variables in the data series DB are used when calculating the distance between the data series and when creating and estimating the model has been described, but the present invention is not limited to this. .. Only some kinds of explanatory variables selected from the explanatory variables included in the data series DB may be used. In addition, the distance may be calculated, or a model may be created or estimated, using only the data at a part of the data series in the period defined by the time window.

Further, the process of acquiring the data series of the explanatory variables for the predetermined period including the estimation target time point t, which is executed by the estimation unit in each of the above embodiments, may be performed by the selection unit.

Further, various processors other than the CPU may execute the model creation process in which the CPU reads the software (program) and executes the software (program) in each of the above embodiments. In this case, the processors include PLD (Programmable Logic Device) whose circuit configuration can be changed after manufacturing FPGA (Field-Programmable Gate Array), and ASIC (Application Specific Integrated Circuit) for executing ASIC (Application Special Integrated Circuit). An example is a dedicated electric circuit or the like, which is a processor having a circuit configuration designed exclusively for the purpose. Further, the model creation process may be executed by one of these various processors, or a combination of two or more processors of the same type or different types (for example, a plurality of FPGAs and a combination of a CPU and an FPGA). Etc.). Further, the hardware structure of these various processors is, more specifically, an electric circuit in which circuit elements such as semiconductor elements are combined.

Further, in each of the above embodiments, the mode in which the model creation processing program is stored (installed) in the ROM 12 or the storage 14 in advance has been described, but the present invention is not limited to this. The program is a non-temporary storage medium such as a CD-ROM (Compact Disk Read Only Memory), a DVD-ROM (Digital Versailles Disk Online Memory), and a USB (Universal Serial Bus) memory. It may be provided in the form. Further, the program may be downloaded from an external device via a network.

Regarding the above embodiments, the following additional notes will be further disclosed.

(Appendix 1)
Memory and
With at least one processor connected to the memory
Including
The processor
From the storage unit in which each data series of the explanatory variable and the explained variable is stored, a predetermined number of data series of the explanatory variables in other periods similar to the data series of the explanatory variables in the predetermined period including the estimation target time point are selected. ,
For a plurality of cases in which the widths of the predetermined periods are different, the data series of the explanatory variables in the other period selected and the data of the explained variables at the time corresponding to the estimation target time in the other period. A model creation device configured to create a model for estimating the explained variable at the estimation target time from the data series of the explanatory variables in the predetermined period including the estimation target time based on the above.

(Appendix 2)
A non-temporary recording medium that stores a program that can be executed by a computer to perform a modeling process.
The model creation process is
From the storage unit in which each data series of the explanatory variable and the explained variable is stored, a predetermined number of data series of the explanatory variables in other periods similar to the data series of the explanatory variables in the predetermined period including the estimation target time point are selected. ,
For a plurality of cases in which the widths of the predetermined periods are different, the data series of the explanatory variables in the other period selected and the data of the explained variables at the time corresponding to the estimation target time in the other period. A non-temporary recording medium comprising creating a model for estimating an explained variable at an estimated target time from a data series of explanatory variables in the predetermined period including the estimated target time based on the above.

10, 210, 310 Modeling device 11 CPU
12 ROM
13 RAM
14 Storage 15 Input unit 16 Display unit 17 Communication I / F
19 Bus 21, 221 and 321 Estimating unit 22, 222 Selecting unit 23, 223 Creating unit 25, 225 Model DB
31,231 data series DB
32 Data acquisition unit 33 Set value input unit 232A CO2 sensor 232B Ventilator controller 232C Number of people counter

Claims (8)

1. From the storage unit in which each data series of the explanatory variable and the explained variable is stored, a predetermined number of data series of the explanatory variables in other periods similar to the data series of the explanatory variables in the predetermined period including the estimation target time point are selected. Selection part and
   In the case of a plurality of cases in which the widths of the predetermined periods are different from each other, the data series of the explanatory variables in the other period selected by the selection unit and the time points corresponding to the estimation target time points in the other period are explained. A creation unit that creates a model that estimates the explained variable at the estimation target time from the data series of the explanatory variables in the predetermined period including the estimation target time based on the variable data.
   Modeling device including.
2. The selection unit defines the width of the predetermined period using at least one of the time window in the past direction from the estimation target time point and the time window in the future direction from the estimation target time point, and determines the size of the time window. The model creation device according to claim 1, wherein the width of the predetermined period is changed by changing the width of the predetermined period.
3. The model creation device according to claim 1 or 2, wherein the selection unit has different widths of the predetermined period within a range input as a set value.
4. The item according to any one of claims 1 to 3, wherein the selection unit selects the predetermined data series of the explanatory variables in the other period within the range input as the set value. Modeling device.
5. The creation unit evaluates the accuracy of each of the created models, and puts each of the created models in the model storage unit in association with the width of the predetermined period when the model is created and the evaluation of the accuracy. Remember,
   Among the models created by the creation unit, the model with the highest evaluation of accuracy is selected, and the predetermined period including the estimation target time point is in the predetermined period of the width when the selected model is created. The model creation apparatus according to any one of claims 1 to 4, further comprising an estimation unit that inputs a data series of explanatory variables into the selected model and estimates the explained variable at the time of the estimation target.
6. The estimation unit selects the model having the highest evaluation of accuracy from the model storage unit, and based on the width of the predetermined period stored in the model storage unit in association with the selected model. The model creation device according to claim 5, wherein the predetermined period is set for the data series of explanatory variables to be input to the selected model.
7. A model creation method executed by a model creation device including a selection unit and a creation unit.
   From the storage unit in which the data series of the explanatory variable and the explained variable are stored, the selection unit is similar to the data series of the explanatory variable in the predetermined period including the estimation target time point, and the data series of the explanatory variable in another period. Select the specified number and
   In the case where the creation unit has a plurality of cases in which the widths of the predetermined periods are different from each other, the data series of the explanatory variables in the other period selected by the selection unit corresponds to the estimation target time point in the other period. A model creation method for creating a model for estimating an explained variable at an estimated target time point from a data series of explanatory variables in the predetermined period including the estimated target time point, based on the data of the explained variable at the time point to be estimated.
8. A model creation program for causing the computer to function as each part constituting the model creation device according to any one of claims 1 to 6.

Images