Simulation model estimation method and estimation device
Patent Information
- Application Number
- JP2021173159
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2021-10-22
- Publication Date
- 2025-04-25
- Estimated Expiration
- 2041-10-22
AI Technical Summary
Existing simulation models often fail to achieve desired prediction accuracy due to incomplete understanding of event mechanisms, necessitating a technology that can enhance prediction accuracy even when sufficient accuracy is not attainable.
A simulation model estimation method that incorporates a known function and an unknown function, along with a noise term, to compensate for errors and uncertainties in the simulation model, ensuring the output values fall within a predetermined range and maintaining continuity under similar input conditions.
This approach allows for accurate prediction of events with unclear mechanisms by constructing a simulation model that improves prediction accuracy and handles uncertainties effectively.
Smart Images
Figure 00000010_0000 
Figure 00000011_0000 
Figure 00000012_0000
Abstract
Description
Technical field
[0001] The present disclosure relates to a simulation model estimation method and an estimation device. [Background technology]
[0002] With the progress of simulation technology, many phenomena can be predicted by calculation. However, it is almost impossible for the simulation-predicted values to perfectly match the actual measurements. Possible reasons for this include that actual values cannot be measured correctly, and that the simulation model is incomplete due to errors or omissions. When prediction is performed using a simulation model, parameter values are often adjusted so that the difference between the predicted value and the actual value is minimized using the method of least squares, etc., using the parameters of the simulation model as variables. However, if the mechanism of event progression is not fully understood, there is a possibility that the desired prediction accuracy cannot be achieved even if parameters are adjusted using many predicted values and measured values. be. As a related technique, Patent Literature 1 discloses a method of quantitatively estimating an error between a predicted value and a measured value caused by parameters of a simulation model. [Prior art documents] [Patent document]
[0003] [Patent document 1] Japanese Patent Application Laid-Open No. 2011-106970 [Outline of the invention] [Problems to be solved by the invention]
[0004] Although it is possible to predict events with a certain level of accuracy using a simulation model, there is a demand for a technology that enables event prediction with the desired accuracy even when sufficient prediction accuracy cannot be obtained. .
[0005] The present disclosure provides a simulation model estimation method and an estimation device that can solve the above problems. [Means for solving the problem]
[0006] According to one embodiment of the present disclosure, a method for estimating a simulation model comprises: in relation to observing a first set of variables containing one or more variables, a second set containing one or more variables; A simulation model that outputs an output vector having the second variable group as an element when an input vector having the first variable group as an element is input when the variable group is observed is a known function A step of setting to be represented by a known function and an unknown function that is an unknown function; and an output vector when the input vector is input to the known function and the first variable group that is an element of the input vector obtaining an error vector between an observation vector having the observed second variable group as an element; and an output value output by the unknown function when the input vector is input to the unknown function and the error vector. is within a predetermined allowable range, and the variation of the output value of the unknown function when inputting the input vectors belonging to a similar group to the unknown function is within a predetermined range. and estimating.
[0007] When a second variable group containing one or more variables is observed in relation to the observation of a first variable group containing one or more variables, the estimating apparatus A simulation model that outputs an output vector having the second variable group as an element when an input vector having the variable group as an element is input is represented by a known function that is a known function and an unknown function that is an unknown function. , an output vector when the input vector is input to the known function, and the second variable group observed with respect to the first variable group, which is an element of the input vector, as elements means for obtaining an error vector between an observed vector and a difference between an output value output by said unknown function when said input vector is input to said unknown function and said error vector is within a predetermined allowable range, and is similar to and means for estimating the unknown function so that the fluctuation of the output value of the unknown function when the input vector belonging to the group to which the unknown function belongs is input to the unknown function is within a predetermined range.
Effect of the invention
[0008] According to the simulation model estimation method and estimation device described above, events can be predicted with a certain level of accuracy by using a known simulation model. can be improved. [Brief description of the drawing]
[0009]
Figure 1
[0010] <Embodiment> (System configuration) The estimation device of the present disclosure will be described below with reference to FIGS. 1 to 6. FIG. The estimator 10 estimates a simulation model of an event whose mechanism has not been fully elucidated. A simulation model is, for example, a computational model that simulates the evolution of events. As shown in FIG. 1, the estimation device 10 includes a data acquisition unit 11, an input reception unit 12, a calculation unit 13, an output unit 14, and a storage unit 15.
[0011] The data acquisition unit 11 acquires input data to be input to the simulation model and observation data (actual measurement values) of prediction target events observed with respect to the input data, and records them in the storage unit 15 .
[0012] The input reception unit 12 receives various types of information input by the user and records the received information in the storage unit 15 . For example, the input reception unit 12 receives information on known functions and unknown functions. A known function is a known computational model that simulates the evolution of an event. A known function is, for example, a trained model or a formula commonly used to predict the event. A known function can predict an event to be predicted with a certain degree of accuracy or higher. However, if the mechanism of event progress is not fully elucidated, the prediction accuracy is limited. For example, the known function may contain errors or omissions due to the type of learning data used to construct the known function, the amount of data being inadequate, the bias, and the like. In contrast, in the present embodiment, the known function is assumed to be incomplete and erroneous. Then, an unknown function is prepared to absorb the error between the predicted value by the known function and the actual measured value, and the known function plus the unknown function is set as a simulation model. The unknown function is represented by a trained model, formula, or database, and may contain a noise term. The input receiving unit 12 acquires information necessary for constructing this unknown function (eg, rough function shape, input / output information of a trained model, learning method, etc.).
[0013] The calculator 13 estimates the simulation model. For example, the calculation unit 13 calculates an error between the output data output by the known function when the input data is input to the known function and the observed data, and calculates an unknown function that compensates for the calculated error. At this time, the calculation unit 13 groups the input data to be input to the unknown function by category, and calculates the unknown function for each group. Further, the calculation unit 13 assumes that the unknown functions have continuity, and calculates the unknown functions so as to output the same level of output data for the input data of the same group.
[0014] The output unit 14 outputs the simulation model estimated by the calculation unit 13, the parameters of the unknown function, and the like to a display device or an electronic file. The storage unit 15 stores various information such as input data, observation data, known functions, unknown functions, and simulation models.
[0015] Next, a simulation model according to this embodiment will be described with reference to FIGS. 2A and 2B. An example of a simulation model is shown in FIG. 2A. The simulation model 20 of FIG. 2A includes a known function 21, an unknown function 22 and a noise term 23. Assuming that the known function 21 is f(x), the unknown function 22 is Δ1(x), and the noise term 23 is Δ2, the simulation model F can be expressed by the following equation (1). F=f(x)+Δ1(x)+Δ2 (1) Here, x^=(X1, X2, . . . ) is input data and is a vector having one or more variables as elements. The output data y^=(Y1, Y2, . . . ) of the simulation model F is also a vector having one or more variables as elements. Since f(x) is a known function 21, the parameters of f(x) are known. Δ1(x) is, for example, a function represented by a formula such as ax+b, and the parameters (a and b) of Δ1(x), which is the unknown function 22, are unknown. Δ2 is a constant or a value that varies according to a normal distribution or the like, and the magnitude of Δ2, the spread of distribution, and the like are unknown. Δ2 is, for example, white noise that appears as a similar value over the entire range of input data. Δ2 is set and adjusted by the user. The calculation unit 13 calculates the output data y^ output by the simulation model 20 when the input data x^ is input to the simulation model 20, and the measured values of the prediction target event actually observed for the input data x^. Comparing the observed data z^=(Z1, Z2, ...) including . More specifically, the calculation unit 13 subtracts the sum of the output value of the known function 21 when the input data x^ is input to the known function 21 and the noise term Δ2 from the observed data z^. Let the value obtained by the subtraction be the error vector α. The calculation unit 13 calculates the values of the parameters a and b such that the value output by Δ1(x) when the input data x^ is input to Δ1(x) is equal to the error vector α. Note that Δ1(x) may be a function with all or part of the input data x^ as a variable, or additionally (or instead) of the output value of f(x), which is a known function 21. It may be a function whose whole or part is a variable. Also, for calculating the parameters of the unknown function 22, a known method such as the method of least squares can be used.
[0016] Another example of the simulation model is shown in FIG. 2B. A simulation model 20' shown in FIG. 2B includes a known function 21, an unknown function 22 constructed from a trained model or database, and a noise term 23. When the input data x^ is input to the simulation model 20, the known function 21 obtains the input data x^ and calculates and outputs the output value. The unknown function 22 calculates the output value using a trained model or database when all or part of the output value of the known function 21 or all or part of the input data x^ in addition to (or instead of) the output value is input. Print that value. A value obtained by adding the noise term 23 to the output value of the unknown function 22 becomes the output data y^. The calculation unit 13 compares the output data y^ and the observation data z^, and constructs a trained model (unknown function 22) so that the error between the output data y^ and the observation data z^ approaches zero. conduct. If it is difficult to express the unknown function 22 by a formula, a trained model or database can be used as the unknown function 22. FIG. For example, in the case of a trained model, the calculation unit 13, for example, learns the relationship between the output value of the known function 21 and the value obtained by subtracting the noise term 23 from the observed data z^ by machine learning or the like, and builds a trained model. do. In the case of a database, the calculation unit 13 associates, for example, the output value of the known function 21 with the value obtained by subtracting the noise term 23 from the observed data ẑ, and calculates this correspondence as a database. Also, a data group used as learning data when constructing a trained model may be used as a database. For example, if the output value of the known function 21 is in the range (1~10, 1~10,...), the observed data z^ is in the range (5~10, 5~10,...) , and if the output value of the known function 21 is in the range of (11~20, 11~20,...), then the observed data z^ is in the range of (15~20, 15~20,...) If there is, the calculation unit 13 associates (1 to 10, 1 to 10, . . . ) with (5 to 10, 5 to 10, . ) and (15~20, 15~20, ) are associated with each other, and the known function 21 identifies the values in the range (1~10, 1~10, ). When outputting, it outputs values with a range of (5 to 10, 5 to 10, ...), and the known function 21 outputs values in the range of (11 to 20, 11 to 20, ...) The unknown function 22 may be constructed so as to output values (15 to 20, 15 to 20, . . . ) when not.). In the case of an event whose mechanism has not been fully elucidated, the variables that influence the outcome of the event (input data) are unknown, and the variables that explain the circumstances in which the event occurred (output data) are unknown. Therefore, it may be difficult to construct effective functions (formulas), learning models, etc. between input variables and output variables that are currently confirmed to be related to phenomena. Even in such a case, for example, by using the data group representing the output value of the known function 21 and the error between the input data x^ and the observed data z^ as a database (unknown function 22), the prediction accuracy of the known function 21 can be improved. can help you improve.
[0017] Thus, in this embodiment, the existing simulation model (known function 21) can explain the mechanism to some extent, but is incomplete. A new simulation model is constructed by adding an unknown function 22 and a noise term 23 that absorb the error of the measured values. To make the simulation model effective, it is necessary to calculate the unknown function 22 with high accuracy. Next, the processing for calculating the unknown function 22 will be described in more detail with reference to FIGS. 3 and 4. FIG. In estimating the unknown function 22, it is assumed that the unknown function 22 fluctuates according to the input conditions, and that the unknown function 22 has continuity under similar input conditions. I do.
[0018] FIG. 3 is a diagram explaining grouping of input data according to the embodiment. For example, assume that input data 1 to 6 are obtained. Calculation unit 13 groups input data 1-6 into groups 1-3 according to a predetermined criterion. For example, when the input data x^ consists of variables such as fluid temperature, flow velocity, pressure, etc., and grouping is performed by focusing on the temperature, the calculation unit 13 Accordingly, the input data x^ are classified into groups 1 to 3. For example, when the input data x^ is composed of variables such as the age, sex, weight, presence or absence of underlying diseases, etc., of the patient with an infectious disease, the calculation unit 13, for example, inputs according to the age of the input data. Classify data. Note that the number of groups to be classified is not limited to three, and may be two or four or more. Also, the number of variables to be focused on is not limited to one, and may be two or more. For example, the classification may be based on the temperature and flow velocity of the fluid, or the patient's age, gender, and presence or absence of underlying disease. A user who has knowledge of the phenomenon can arbitrarily set which variable should be focused on for classification. For example, the user sets the classification so that the input data can be classified with a granularity that allows the unknown function 22 to maintain continuity in the same group. Prior to the calculation of the unknown function 22, the input data are grouped based on their properties to classify and set the input conditions for the unknown function 22. FIG.
[0019] Note that here, the data input to the unknown function 22 is described as the input data x^, but if the data input to the unknown function 22 includes the output value of the known function 21, the output value of the known function 21 is also grouped. make a change.
[0020] FIG. 4 is a diagram explaining a method of calculating an unknown function according to the embodiment. The calculation unit 13 performs (a) the unknown function 22 so that the output value of the unknown function 22 when the input data is input to the unknown function 22 is equal to the error between the observed data and (the output value of the known function + the noise term 23). 22, the unknown function 22 fluctuates according to the input conditions, and based on the assumption that the unknown function 22 has continuity under similar input conditions, (b) when the input conditions are similar, the same An unknown function 22 is calculated to output a level output value. Regarding (b), for example, when input data classified into the same group are input to the unknown function 22, the input conditions are regarded as similar (classified so as to be regarded as such). Further, the same level of output value means that the output values are similar values (values within a predetermined range). For example, the difference between output data 2 when input data 2 belonging to group 1 is input to unknown function 22 and output data 3 when input data 3 is input to unknown function 22 is unknown so that the difference is within a predetermined range. Adjust the parameters of function 22, etc. Similarly, for group 2, calculation unit 13 determines that the difference between output data 1 when input data 1 is input to unknown function 22 and output data 5 when input data 5 is input to unknown function 22 is a predetermined difference. The unknown function 22 is calculated so that the difference between the output data 4 and the output data 6 of the group 3 is within a predetermined range. A function is said to have continuity if it outputs a value of the same level when the same level of value is input. Note that continuity of the function across groups is not required, assuming that the unknown function 22 fluctuates according to the input conditions. For example, output data 1, output data 2, and output data 4 do not have to be at the same level. For example, the method of least squares or the like can be used to calculate the unknown function 22 . If it is difficult to calculate the unknown function 22 that satisfies both (a) and (b) above, the noise term 23 may be adjusted so that the continuity of the unknown function 22 can be maintained. Although the unknown function 22 fluctuates for each input group, the noise term 23 is set to the same value across the input groups.
[0021] (motion) Next, with reference to FIG. 5, the flow of estimation processing of the simulation model 20 will be described. FIG. 5 is a flowchart illustrating an example of simulation model estimation processing according to the embodiment. The user sets the known function 21, the unknown function 22, and the classification of input data (step S1). For example, if the unknown function 22 is a mathematical formula, the user can create a formula with the parameters of the formula converted into variables, and if the unknown function 22 is a trained model or a database, input and output variables, input data variables used for classification and settings such as their thresholds are input to the estimation device 10 . The input reception unit 12 acquires these settings and records them in the storage unit 15. FIG.
[0022] Next, the data acquisition unit 11 acquires the input data x^ and the observation data z^ which are learning data used for estimation of the simulation model 20 (step S2). The input data x^ and the observation data z^ are paired, and the observation data z^ is the measured value of the variable representing the event that occurred in relation to the measurement of the input data x^. The data acquisition unit 11 acquires multiple sets of input data x^ and observation data z^, and records each set of input data x^ and observation data z^ in association with each other in the storage unit 15 .
[0023] Next, the calculation unit 13 classifies the input data acquired in step S2 (step S3). The calculation unit 13 classifies the input data into groups of the same category according to, for example, temperature, age, etc., based on the classification settings set in step S1.
[0024] Next, the user sets the noise term 23 (step S4). The noise term 23 is set as constant regardless of the group of input data. The noise term 23 is set as a constant, a range of values, a normal distribution, or the like. The input reception unit 12 receives the setting of the noise term 23 and records this information in the storage unit 15. FIG.
[0025] Next, the calculator 13 calculates the error between the predicted value by the known function 21 and the observed data (step S5). For each group, the calculation unit 13 reads from the storage unit 15 a combination of the input data x̂ and the observation data ẑ belonging to that group, and inputs the input data x̂ to the known function 21 . The calculator 13 calculates the error between the output value of the known function 21 and the observed data z^. For example, the calculation unit 13 may calculate an error vector obtained by subtracting the observation data z^ from the output value of the known function 21, or may calculate a value obtained by squaring each element of the error vector. Next, the calculator 13 calculates the unknown function 22 for each group (step S6). (a) The calculation unit 13 compares the value obtained by adding the noise term to the output value of the unknown function 22 and the error calculated in step S5, and calculates the unknown function 22 so that the difference between the two falls within a predetermined range. calculate. (b) In addition, the calculation unit 13 receives the input data x^ of the same group (or the output value of the known function 21 when the input data x^ is input), and the fluctuation of the output value of the unknown function 22 is predetermined. Calculate the unknown function 22 so that it falls within the range of . Calculation unit 13 calculates unknown function 22 that satisfies both conditions (a) and (b) for each of all groups. Next, the calculation unit 13 determines whether or not the unknown function 22 has been calculated for each group (step S7). If the unknown function 22 could not be calculated (Step S7; No), there is a possibility that the setting of the noise term 23 failed to calculate the unknown function 22 that satisfies both conditions (a) and (b). , the processing after step S4 is repeated. If the calculation of the unknown function 22 is successful (step S7; Yes), the calculator 13 estimates the simulation model 20 (step S8). For example, the calculation unit 13 sets the known function 21 and the calculated unknown function 22 as the simulation model 20 (not including the noise term 23). In this case, the sum of the output value when the input data x^ is input to the known function 21 and the output value when the input data x^ (or the output value of the known function 21, etc.) is input to the unknown function 22 is Predicted value. As a result, events whose mechanisms have not been fully elucidated can be predicted with high accuracy.
[0026] Note that the calculation unit 13 may set the simulation model 20 including the known function 21, the unknown function 22, and the noise term. Further, in step S4, the noise term 23 is set constant throughout all groups, but the noise term 23 may be set by adjusting the magnitude of the noise term 23 for each group.
[0027] Conventionally, a known simulation model is assumed to be correct, and the parameters of the simulation model are adjusted so that the error between predicted values and actual values is minimized. However, in this embodiment, the simulation model is assumed to be imperfect. Then, assuming that the event to be predicted has a mechanism that is not dominant but partially unknown, an unknown function 22 is added to the known function 21 to explain the part where the mechanism is unknown. Also, it is assumed that a certain amount of white noise exists in the error between the predicted value by the known function 21 and the observed data, and that the value of the unknown function 22 is obtained by subtracting this noise from the error. Then, based on the assumption that the unknown function 22 fluctuates according to the input conditions, and that the unknown function 22 has continuity under similar input conditions, the unknown function 22 that compensates for the error between the observed data z^ and the predicted value is calculate. As a result, according to the present embodiment, it is possible to accurately predict an event whose mechanism has not been sufficiently clarified. For example, with regard to people who had the opportunity to be around patients with infectious diseases, data on their attributes and behavioral histories were classified into groups based on their spatial proximity to infected patients. A group classified by focusing on the temporal proximity after the person was infected and a group classified by focusing on each person's attributes (age, lifestyle, etc.), and each person was actually infected. The unknown function 22 is calculated by the method of the present embodiment by comparing with observation data including whether or not it has been done. By using the known function 21 and the unknown function 22 for estimating the probability of infection, it is possible to accurately predict the number of infected people. As another example, the simulation model estimated by the method of the present embodiment can also be used for failure prediction of plants, equipment, and the like.
[0028] In the above embodiment, the simulation model and the known function 21 are models for predicting the development of events in the future. However, it may be one that estimates other events at the same time or one that estimates past events. For example, a simulation model estimated by the method of the present embodiment may be used for real-time device abnormality monitoring.
[0029] FIG. 6 is a diagram illustrating an example of a hardware configuration of an estimation device according to the embodiment; Computer 900 includes CPU 901 , main memory device 902 , auxiliary memory device 903 , input / output interface 904 and communication interface 905 . The estimation device 10 described above is implemented in a computer 900 . Each function described above is stored in the auxiliary storage device 903 in the form of a program. The CPU 901 reads the program from the auxiliary storage device 903, develops it in the main storage device 902, and executes the above processing according to the program. Also, the CPU 901 reserves a storage area in the main memory 902 according to the program. Also, the CPU 901 secures a storage area in the auxiliary storage device 903 for storing the data being processed according to the program.
[0030] A program for realizing all or part of the functions of the estimating device 10 is recorded in a computer-readable recording medium, and the program recorded in this recording medium is read by a computer system and executed, thereby performing each functional unit. may be processed by The "computer system" here includes hardware such as an OS and peripheral devices. In addition, the "computer system" includes the homepage providing environment (or display environment) if the WWW system is used. The term "computer-readable recording medium" refers to portable media such as CDs, DVDs, and USBs, and storage devices such as hard disks built into computer systems. Further, when this program is distributed to the computer 900 via a communication line, the computer 900 receiving the distribution may develop the program in the main storage device 902 and execute the above process. Further, the program may be for realizing part of the functions described above, or may be capable of realizing the functions described above in combination with a program already recorded in the computer system. .
[0031] Although several embodiments of the present disclosure have been described above, all these embodiments are provided by way of example and are not intended to limit the scope of the invention. These embodiments can be implemented in various other forms, and various omissions, replacements, and modifications can be made without departing from the scope of the invention. These embodiments and their modifications are included in the scope and spirit of the invention, as well as the scope of the invention described in the claims and equivalents thereof.
[0032] <Appendix> The simulation model estimation method and estimation apparatus described in each embodiment are understood as follows, for example.
[0033] (1) The method for estimating a simulation model according to the first aspect is that, in relation to observation of a first variable group containing one or more variables, a second group containing one or more variables is observed. A simulation model 20 that outputs an output vector having the second variable group as an element when an input vector having the first variable group as an element is input when the variable group is observed is a known function. (S1) setting to be represented by a known function 21 and an unknown function 22, which is an unknown function; (S5) obtaining an error vector between an observation vector having the second variable group observed with respect to the variable group of 1 and an error vector of the unknown function when the input vector is input to the unknown function The difference between the output value and the error vector is within a predetermined allowable range, and the fluctuation of the output value of the unknown function when the input vectors belonging to a similar group are input to the unknown function is within a predetermined range. and (S6) estimating the unknown function such that
[0034] As a result, it is possible to estimate a simulation model that accurately predicts and estimates events whose mechanism is not sufficiently clarified. Since a simulation model is constructed using a known function that can be roughly predicted, it is possible to construct a simulation model with a small amount of data and a small amount of calculation.
[0035] (2) A simulation model estimation method according to a second aspect, which is the simulation model estimation method of (1), further comprising the step of classifying the input vectors into similar groups, wherein the unknown function estimating the unknown function for each group.
[0036] In the conventional method, the parameters of the simulation model are adjusted so as to minimize the error between the predicted value by the simulation model (known function 21) and the observed data for all input data. On the other hand, in the second aspect, the simulation model is changed according to the input conditions. As a result, it is possible to estimate a highly accurate simulation model without sacrificing prediction accuracy by forcibly adjusting parameters to match all data.
[0037] (3) A simulation model estimation method according to a third aspect is the simulation model estimation method of (1) to (2), wherein the unknown function is learned based on the relationship between the input vector and the error vector. pre-existing model or database. Alternatively, the unknown function is a database for outputting the output vector having elements having values within a predetermined range corresponding to the range of values of the input vector when the input vector having elements having values within a predetermined range is input. be.
[0038] As a result, even if the unknown function 22 cannot be modeled by a formula, the unknown function 22 can be calculated. It should be noted that, for example, a method of building a trained model by learning the relationship between the input data and the observed data by machine learning without using the known function 21 is also conceivable. Many types of large amounts of training data are required to construct a trained model. On the other hand, according to the third aspect, a model that compensates for the error between the predicted value and the observed data is constructed in combination with the known function 21 capable of predicting the event with a certain degree of accuracy or higher. It is possible to construct a trained model with high accuracy even with the type and amount of training data specified.
[0039] (4) A simulation model estimation method according to a fourth aspect is the simulation model estimation method of (1) to (3), wherein the unknown function is a mathematical formula in which at least part of the input vector is a variable. and the step of estimating the unknown function estimates the parameters of the equation. As a result, the unknown function 22 can be calculated when the unknown function 22 can be modeled by a formula.
[0040] (5) A simulation model estimation method according to a fifth aspect is the simulation model estimation method of (1) to (4), wherein in the setting step, the simulation model is combined with the known function and the Set as the sum of the unknown function and the noise term. Errors between predicted values and measured values by known simulation models generally contain noise as well as the effects of unknown mechanisms that have not been fully elucidated. By estimating the unknown function 22 in consideration of this noise, the unknown function 22 can be calculated with high accuracy.
[0041] (6) The estimation device according to the sixth aspect, in relation to the observation of the first variable group containing one or more variables, the second variable group containing one or more variables A simulation model that outputs an output vector having the second variable group as an element when an input vector having the first variable group as an element is input when observed is a known function and an unknown function. an unknown function that is a function of; an output vector when the input vector is input to the known function; means for obtaining an error vector from an observation vector having a second variable group as elements; means for estimating the unknown function so that the variation of the output value of the unknown function when the input vector belonging to the similar group is input to the unknown function is within a predetermined range. have. [Description of symbols]
[0042] 11...data acquisition unit, 12...input reception unit, 13...calculation unit, 14...output unit, 15...storage unit, 900...computer, 901...CPU, 902 Main memory 903 Auxiliary memory 904 I / O interface 905 Communication interface
Claims
1. a step of setting a simulation model that outputs an output vector having the second variable group as an element when an input vector having the first variable group as an element is input, in a case where a second variable group having one or more variables is observed in connection with the observation of a first variable group including one or more variables, so as to be expressed by a known function that is a known function and an unknown function that is an unknown function; a step of calculating an error vector between an output vector when the input vector is input to the known function and an observation vector having elements of the second set of variables observed with respect to the first set of variables that is an element of the input vector; a step of estimating the unknown function so that, when the input vector is input to the unknown function, a difference between the output value of the unknown function and the error vector falls within a predetermined tolerance range, and a fluctuation in the output value of the unknown function falls within a predetermined range when the input vector belonging to a similar group is input to the unknown function; A method for estimating a simulation model having the following:
2. classifying the input vectors into similar groups; and In the step of estimating the unknown function, the unknown function is estimated for each of the groups. The method for estimating a simulation model according to claim 1 .
3. The unknown function is a trained model or database based on the relationship between the input vector and the error vector. The method for estimating a simulation model according to claim 1 or 2.
4. the unknown function is expressed by a mathematical expression that uses at least a part of the input vector as a variable, and the step of estimating the unknown function includes estimating parameters of the mathematical expression. The method for estimating a simulation model according to any one of claims 1 to 3.
5. In the setting step, the simulation model is set as a sum of the known function, the unknown function, and a noise term. A method for estimating a simulation model according to any one of claims 1 to 4.
6. a means for setting a simulation model that outputs an output vector having the second variable group as an element when an input vector having the first variable group as an element is input, in a case where a second variable group having one or more variables is observed in association with the observation of a first variable group including one or more variables, so as to be expressed by a known function that is a known function and an unknown function that is an unknown function; means for calculating an error vector between an output vector when the input vector is input to the known function and an observation vector having elements of the second set of variables observed with respect to the first set of variables which are elements of the input vector; means for estimating the unknown function such that, when the input vector is input to the unknown function, a difference between the output value of the unknown function and the error vector falls within a predetermined tolerance, and when the input vector belonging to a similar group is input to the unknown function, a fluctuation in the output value of the unknown function falls within a predetermined range; An estimation device having: