Information processing device

Abstract

To generate a model of an appropriate physical phenomenon.SOLUTION: An information processing device 1 includes a storage section 10 storing library information, a detection result acquisition section 111, a regression equation generation section 112, a correction section 113, a regression equation regeneration section 114, and an output control section 115. The correction section 113 corrects generation probabilities and a hyperparameter of machine learning on the basis of loss functions of linear regression equations. The regression equation regeneration section 114 extracts nonlinear basis functions on the basis of the corrected generation probabilities from sub-libraries including a plurality of nonlinear basis functions, generates a plurality of linear regression equations obtained by combining the nonlinear basis functions of the plurality of types of sub-libraries, estimates coefficients of the plurality of linear regression equations by machine learning using the corrected hyperparameter, and calculates the loss functions of the plurality of linear regression equations.SELECTED DRAWING: Figure 2

Description

[Technical field]

[0001] An embodiment of the present invention relates to an information processing device. [Background technology]

[0002] One technique for modeling physical phenomena is to apply the function identification problem, a type of machine learning, to obtain a mathematical model that describes a physical phenomenon from time-series data. [Prior art documents] [Non-patent literature]

[0003] [Non-Patent Document 1] SL Brunton, JL Proctor, JN Kutz, “Discovering governing equations from data by sparse identification of nonlinear dynamical systems”, Proc. Natl. Acad. Sci., 113 (2016), pp. 3932-3937 Summary of the Invention [Problem to be solved by the invention]

[0004] For example, if one were to use the technology described in Non-Patent Document 1 to generate a thermal circuit network model as a model of a physical phenomenon, there is a possibility that it would not be possible to generate an appropriate model of the physical phenomenon, as the heat transfer phenomena are so diverse that the search space becomes broad and learning becomes unstable.

[0005] Therefore, the aim is to generate an appropriate model of the physical phenomenon. [Means for solving the problem]

[0006] The information processing device according to the embodiment includes a library storage unit, a detection result acquisition unit, a regression equation generation unit, a correction unit, a regression equation regeneration unit, and an output control unit. The library storage unit stores a plurality of types of sub-libraries including a nonlinear basis function based on a dependent variable or an independent variable, and a generation probability of each of the plurality of nonlinear basis functions included in each sub-library. The detection result acquisition unit acquires the detection result of the sensor. The regression equation generation unit performs a calculation using the detection result and the nonlinear basis function of the sub-library, extracts a nonlinear basis function based on the generation probability from the sub-library including the plurality of nonlinear basis functions, generates a plurality of linear regression equations that are equations for calculating dependent variables by combining the nonlinear basis functions of the plurality of types of sub-libraries, estimates coefficients of the linear regression equations by machine learning, and calculates a loss function of the linear regression equation using the calculation result. If a predetermined condition is not met, the correction unit corrects the generation probability and the hyperparameters of the machine learning based on the loss function of the linear regression equation. The regression equation regeneration unit extracts a nonlinear basis function from a sub-library including multiple nonlinear basis functions based on the revised generation probability, generates multiple linear regression equations combining the nonlinear basis functions of the multiple types of sub-libraries, estimates coefficients of the linear regression equations by machine learning using the revised hyperparameters, and calculates a loss function of the linear regression equation. The output control unit outputs a linear regression equation selected from the multiple linear regression equations generated by the regression equation generation unit or the regression equation regeneration unit to the output unit when a condition is met. [Brief description of the drawings]

[0007] [Figure 1] FIG. 1 is a schematic diagram illustrating an example of an information processing system. [Diagram 2] FIG. 2 is a block diagram illustrating an example of a functional configuration of the information processing device. [Figure 3A] FIG. 3A is a diagram showing sub-library definition information. [Figure 3B] FIG. 3B is a diagram showing generation probability information. [Figure 4] FIG. 4 is a diagram illustrating an example of the sensor information. [Figure 5A] FIG. 5A is a diagram showing the state of internal processing. [Figure 5B] FIG. 5B is a diagram showing an example of output of information on a linear regression equation. [Figure 6] FIG. 6 is a flowchart showing an example of the flow of information processing executed by the information processing device. [Figure 7A] FIG. 7A is a diagram showing an example of temperature prediction using a machine learning model that does not take into account a physical model. [Figure 7B] FIG. 7B is a diagram showing an example of temperature prediction using a machine learning model that takes into account a physical model. [Figure 7C] FIG. 7C is a diagram showing an example of temperature prediction based on the linear regression equation generated according to this embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0008] Hereinafter, an information processing device according to an embodiment will be described in detail with reference to the accompanying drawings. Note that the present invention is not limited to the embodiment.

[0009] FIG. 1 is a schematic diagram illustrating an example of an information processing system 100. As shown in FIG.

[0010] The information processing system 100 includes an information processing device 1 and an electronic device 2. The information processing device 1 and the electronic device 2 are connected to each other so as to be able to transmit and receive data or signals.

[0011] The electronic device 2 is a device that includes one or more components and is driven by supplied power. The electronic device 2 is applied to, for example, various devices that include one or more electronic components mounted in a single rack. Specifically, the electronic device 2 is applied to various electronic devices such as digital broadcast transmitters, data relay devices, computers, and servers.

[0012] The electronic device 2 includes a housing 25 that houses various components.

[0013] The housing 25 is a main body of the electronic device 2, and is an exterior that houses various parts and devices. In this embodiment, the housing 25 will be described as an example in which the inside is a hollow box-shaped member.

[0014] A plurality of components and a plurality of sensors are disposed inside the housing 25. In the present embodiment, a form in which a plurality of components and a plurality of sensors are disposed inside the housing 25 will be described as an example.

[0015] The components 21 and 22 are electronic components. The electronic components are, for example, components that are driven in response to supplied power. The electronic components include heat-generating components that generate heat by being driven in response to supplied power. Note that "driving" includes both electrical driving and mechanical driving. Electrical driving includes, for example, processing by a processor such as a CPU (Central Processing Unit). Mechanical driving includes, for example, driving of a motor. Note that the components 21 and 22 are heat-generating components.

[0016] The heat generating components 21 and 22 are, for example, processors such as a CPU and a GPU (Graphics Processing Unit). Note that the components 21 and 22 are not limited to a CPU or a GPU as long as they are components that generate heat by being driven in response to supplied power. For example, the components 21 and 22 may be a field effect transistor (FET), an intelligent power module (IPM), a motor, an electronic circuit, or the like.

[0017] The parts 23 and 24 are metal blocks made of aluminum, etc. The parts 23 and 24 have a cooling function.

[0018] A plurality of sensors are disposed in the electronic device 2. The plurality of sensors measure, for example, physical quantities of environmental changes at positions P1 to P5. For example, the plurality of sensors detect physical quantities such as temperature, current, voltage, and wind speed at positions P1 to P5, and output the detection results as detection values. The physical quantities and detection values ​​are expressed as numerical values ​​indicating, for example, temperature, current, voltage, wind speed, etc. Note that pressure, the number of revolutions of an air-cooling fan, etc. may also be included.

[0019] The above-mentioned multiple sensors are, for example, a temperature sensor, a flow rate sensor, a current sensor, a voltage sensor, and the like.

[0020] The above-mentioned multiple sensors may be disposed in positions capable of measuring environmental changes inside or outside the housing 25. These multiple sensors may be built into the product, or may be attached externally only when measurements are made.

[0021] The information processing device 1 is a device that generates a model for predicting the temperature of an electronic device based on the state of the electronic device 2. The information processing device 1 and each of the multiple sensors provided in the electronic device 2 are connected to be able to exchange data or signals. The information processing device 1 may be further connected to electronic devices including various other information processing devices other than the sensors mounted on the electronic device 2 to be able to exchange data or signals. For example, the information processing device 1 may be connected to at least one of the multiple sensors and multiple components to be able to exchange data or signals. The information processing device 1 is, for example, a server or a workstation.

[0022] For example, the information processing device 1 may transmit data acquired from a plurality of sensors in a batch to a remote information processing device via a storage medium or the cloud.

[0023] Next, an example of the functional configuration of the information processing device 1 will be described.

[0024] FIG. 2 is a block diagram showing an example of a functional configuration of the information processing device 1. As shown in FIG.

[0025] The information processing device 1 and the multiple sensors are connected so as to be able to transmit and receive data or signals. As described above, the information processing device 1 may be connected so as to be able to transmit and receive data or signals to and from various electronic devices other than the multiple sensors. The information processing device 1 of this embodiment is a device that generates a model based on a thermal network method. The information processing device 1 outputs, as the model, a linear regression equation capable of outputting the temperature of the electronic device 2.

[0026] The information processing device 1 includes a storage unit 10, a control unit 11, and an output unit 12. The control unit 11, the storage unit 10, and the output unit 12 are connected to each other so as to be able to exchange data or signals.

[0027] The storage unit 10 stores various types of data. The storage unit 10 is, for example, a storage medium such as a known HDD (hard disk drive). In this embodiment, the storage unit 10 stores library information 101 in advance.

[0028] Library information 101 includes sub-library definition information, which is information defining multiple types of sub-libraries including non-linear basis functions, and generation probability information, which is information on the generation probability of each of the multiple non-linear basis functions included in each sub-library.

[0029] First, the sub-libraries will be described with reference to FIG. 3A. Each sub-library is for generating a linear regression equation indicating a thermal model. FIG. 3A is a diagram showing sub-library definition information. As shown in FIG. 3A, the sub-libraries include a thermal conduction sub-library, a radiation sub-library, a forced convection sub-library, a natural convection sub-library, and a heat generation sub-library. Each sub-library includes one or more nonlinear basis functions. Note that the sub-libraries are not limited to those shown in FIG. 3A, and a phase change sub-library that takes phase change into consideration may be defined.

[0030] The T shown in Figure 3A i , T j is the temperature at any one of positions P1 to P5.i-j is the temperature difference between two points. v is the velocity (wind speed). V is the voltage. i1, i2 are the currents. Thus, the sub-library contains nonlinear basis functions based on the dependent variable (temperature) and the independent variables (velocity, current, voltage).

[0031] As shown in Fig. 3A, each sub-library includes one or more nonlinear basis functions. In addition, the forced convection sub-library and the natural convection sub-library include multiple nonlinear basis functions with different exponents.

[0032] In the above thermal model, the law of conservation of energy that holds at the nodes (temperature measurement points) is expressed by the nodal equation shown in the following equation (1).

[0033]

number

[0034] In the above formula (1), R is thermal resistance, T is temperature, Q is the amount of heat generated, and Δt is the time interval. Also, m is the subscript for time.

[0035] By modifying the above formula (1), we obtain the following formula (2).

[0036]

number

[0037] As the right side of equation (2) shows, if nonlinear basis functions indicating Q / C or ΔT / RC can be prepared, an appropriate linear regression equation can be derived. When dimensions and physical properties do not change, such as during maintenance and operation, the nonlinear basis functions of the thermal conduction sublibrary, radiation sublibrary, forced convection sublibrary, and natural convection sublibrary are proportional to ΔT / RC. The heat generation sublibrary is proportional to Q / C. The information processing device 1 of this embodiment generates a linear regression equation conforming to the thermal model by using a sublibrary defined in the sublibrary definition information.

[0038] The library information 101 further includes information on the generation probability, which is the probability of being selected from a plurality of nonlinear basis functions included in the same sub-library.

[0039] Next, the generation probability will be described with reference to Fig. 3B. Fig. 3B is a diagram showing generation probability information. As shown in Fig. 3B, a sub-library, an index, and a generation probability are associated with each other.

[0040] 2, the output unit 12 outputs various information. In this embodiment, the output unit 12 outputs information of a model for predicting the temperature of an electronic device. The information of the model for predicting the temperature of an electronic device will be described in detail later.

[0041] The output unit 12 has at least one of a display function for displaying various information and a communication function for communicating data with an external device. The external device is a device provided outside the electronic device 2. The electronic device 2 and the external device may be capable of communicating with each other via a network or the like. For example, the output unit 12 is configured by combining at least one of a known display device and a known communication device.

[0042] Next, the control unit 11 will be described.

[0043] The control unit 11 includes a detection result acquisition unit 111 , a regression equation generation unit 112 , a correction unit 113 , a regression equation regeneration unit 114 , and an output control unit 115 .

[0044] The detection result acquisition unit 111, the regression equation generation unit 112, the correction unit 113, the regression equation regeneration unit 114, and the output control unit 115 are realized, for example, by one or more processors. For example, each of the above units may be realized by having a processor such as a CPU execute a program, that is, by software. Each of the above units may be realized by a processor such as a dedicated IC, that is, by hardware. Each of the above units may be realized by using both software and hardware. When multiple processors are used, each processor may realize one of the units, or may realize two or more of the units.

[0045] The detection result acquisition unit 111 acquires the detection results detected by a plurality of sensors. For example, the detection results are acquired by measuring the temperature at positions P1 to P5, the wind speed at position P5, the current at positions P1 and P2, and the voltage at position P1 at predetermined time intervals, as shown in Fig. 1. The detection result acquisition unit 111 may also acquire information on the number of revolutions and voltage of the cooling fan, and calculate the wind speed based on this information.

[0046] An example of data acquired by the detection result acquisition unit 111 is shown in FIG. 4. FIG. 4 is a diagram showing an example of sensor information. As shown in FIG. 4, for example, data measured every 0.5 seconds is acquired. Temp1 to Temp5 shown in FIG. 4 are temperatures at positions P1 to P5. v shown in FIG. 4 is wind speed at position P5. i1 and i2 are currents at positions P1 and P2. V shown in FIG. 4 is voltage at position P1. The detection result acquisition unit 111 may register the acquired detection results in the storage unit 10.

[0047] Returning to FIG. 2, the regression equation generation unit 112 generates a plurality of regression equations. Specifically, the regression equation generation unit 112 first differentiates Temp1 to Temp5 for each acquired time with respect to time. The regression equation generation unit 112 also generates matrix data by inputting each data acquired by the detection result acquisition unit 111 into the nonlinear basis function of each sub-library. In this way, the regression equation generation unit 112 performs calculations using the detection results of the sensor and the nonlinear basis functions.

[0048] Next, the regression equation generation unit 112 extracts nonlinear basis functions from a sub-library containing multiple nonlinear basis functions based on the generation probability, and combines the nonlinear basis functions of multiple types of sub-libraries to generate multiple linear regression equations for calculating each of Temp1 to Temp5.

[0049] When extracting a nonlinear basis function based on the generation probability from a sub-library including a plurality of nonlinear basis functions, the regression equation generating unit 112 preferentially extracts an exponent having a high generation probability according to the generation probability information.

[0050] Next, the regression equation generation unit 112 obtains coefficients of the above-mentioned multiple linear regression equations by sparse estimation, a known technique. For example, the regression equation generation unit 112 obtains coefficients by the least squares method, and then sets coefficients equal to or less than a predetermined threshold (hyperparameter) to zero. The regression equation generation unit 112 also performs the least squares method again on candidate functions (nonlinear basis functions) having remaining non-zero coefficients to update the coefficients, and sets coefficients equal to or less than a predetermined threshold to zero. The regression equation generation unit 112 repeats the process of performing the least squares method again on candidate functions to update the coefficients and set coefficients equal to or less than the threshold to zero multiple times.

[0051] Note that the regression equation generating unit 112 may obtain coefficients by other known machine learning methods other than sparse estimation.

[0052] Furthermore, the regression equation generating unit 112 calculates a loss function using a plurality of linear regression equations and the above-mentioned calculation results. The method of calculating the loss function is realized by a known method. For example, the regression equation generating unit 112 generates a loss function based on an error between a result of inputting matrix data obtained by inputting each data acquired by the detection result acquiring unit 111 into a nonlinear basis function of each sub-library into the generated plurality of linear regression equations, and a result of differentiating Temp1 to Temp5 for each time with respect to time. Note that the regression equation generating unit 112 may calculate a loss function based not only on the above-mentioned error, but also on the simplicity of the equation.

[0053] When the conditions described below are not met, the correction unit 113 corrects the hyperparameters and generation probability used when calculating the coefficients of the linear regression equation based on the loss function of the linear regression equation. The hyperparameter correction method may be a conventional method (for example, the method described in Non-Patent Document 1). When the loss function of each linear regression equation falls below a predetermined threshold, the correction unit 113 adds a certain value to the hyperparameter. On the other hand, when the loss function of each linear regression equation is equal to or greater than the threshold, the correction unit 113 subtracts a certain value from the hyperparameter.

[0054] The correction unit 113 increases the generation probability of a sub-library constituting a linear regression equation with the smallest loss function among the loss functions of a plurality of linear regression equations generated with the same hyperparameters, and decreases the generation probability of other sub-libraries. Note that the method of correcting the generation probability may be, for example, to further increase the generation probability of a sub-library with a high generation probability. In other words, the generation probability may be corrected so as to narrow down the nonlinear basis functions to be extracted.

[0055] The regression formula regeneration unit 114 extracts nonlinear basis functions from a sub-library containing multiple nonlinear basis functions based on the generation probability after correction by the modification unit 113, generates multiple linear regression formulas by combining the nonlinear functions of multiple types of sub-libraries, estimates coefficients of the multiple linear regression formulas by sparse estimation using the hyper-parameters after correction by the modification unit 113, and calculates loss functions of the multiple linear regression formulas.

[0056] If the regression equation regeneration unit 114 has not generated a linear regression equation a predetermined number of times, the correction unit 113 corrects the hyperparameters and generation probabilities again. Then, the regression equation regeneration unit 114 calculates loss functions of a plurality of linear regression equations based on the hyperparameters and generation probabilities corrected again. In this way, the control unit 11 corrects the hyperparameters and generation probabilities a predetermined number of times, and calculates the loss function using the corrected hyperparameters and generation probabilities.

[0057] When a predetermined condition is met, the output control unit 115 outputs a linear regression formula selected from the multiple linear regression formulas to the output unit 12. An example of the condition is the number of times the regression formula regeneration unit 114 has generated a linear regression formula. In addition, the above condition may be that the loss function of the multiple linear regression formulas calculated by the regression formula generation unit 112 or the loss function of the multiple linear regression formulas calculated by the regression formula regeneration unit 114 is equal to or less than a threshold value.

[0058] In addition, the output control unit 115 outputs information on the multiple selected regression formulas selected according to the ranking based on the loss function of the multiple linear regression formulas to the output unit. Here, an example in which the output control unit 115 outputs multiple linear regression formulas will be described with reference to Figs. 5A and 5B.

[0059] Fig. 5A is a diagram showing the state of internal processing. Specifically, Fig. 5A is a diagram in which the number of iterations (number of corrections) for multiple linear regression equations generated by the regression equation regeneration unit 114, the regression equation number indicating the identification number of the linear regression equation, and the loss function are associated with each other.

[0060] As shown in Figure 5A, the loss function "XXXX" of the linear regression equation with the "number of iterations" of "2" and the "regression formula No." of "2" is ranked first (the value indicated by the loss function is the smallest), the loss function "XXXX" of the linear regression equation with the "number of iterations" of "2" and the "regression formula No." of "3" is ranked second, and the loss function "△△△△" of the linear regression equation with the "number of iterations" of "2" and the "regression formula No." of "4" is ranked third.

[0061] Fig. 5B is a diagram showing an example of output of information on a linear regression formula. The output control unit 115 outputs the loss function and the linear regression formula information to the output unit 12 based on the ranking of the loss function shown in Fig. 5A. Here, the linear regression formula information is information indicating a linear regression formula, such as information indicating simultaneous equations of nodes or linear combinations of basis functions. The output control unit 115 may output the linear regression formula information on a separate screen.

[0062] The output control unit 115 may select the linear regression equation with the smallest loss function from among the multiple linear regression equations.

[0063] The output control unit 115 may transmit and output the selected linear regression equation to another information processing device, so that the other information processing device can perform temperature prediction calculations using the linear regression equation.

[0064] Next, the flow of information processing executed by the information processing device 1 will be described.

[0065] FIG. 6 is a flowchart showing an example of the flow of information processing executed by the information processing device 1.

[0066] The detection result acquisition unit 111 acquires the detection results of a plurality of sensors (step S1). Next, the regression equation generation unit 112 uses the detection results of the plurality of sensors to differentiate Temp1 to Temp5 acquired for each time with respect to time, and generates matrix data by inputting each data acquired by the detection result acquisition unit 111 into the nonlinear basis function of each sub-library. In this way, the regression equation generation unit 112 performs arithmetic processing of the nonlinear basis function using the detection results of the sensors (step S2).

[0067] The regression equation generation unit 112 extracts nonlinear basis functions from a sub-library including a plurality of nonlinear basis functions based on generation probability, and generates a plurality of linear regression equations for calculating each of Temp1 to Temp5 by combining the nonlinear basis functions of the plurality of types of sub-libraries. In addition, the regression equation generation unit 112 obtains coefficients of the plurality of linear regression equations by sparse estimation (step S3).

[0068] The regression equation generating unit 112 calculates a loss function using the multiple linear regression equations and the above-mentioned calculation results (step S4). If the linear regression equations have not been generated a predetermined number of times (step S5: No), the modifying unit 113 modifies the generation probability and the hyperparameters based on the loss function of the linear regression equation (step S6). For example, when the loss function of each linear regression equation falls below a predetermined threshold, the modifying unit 113 adds a certain value to the hyperparameter. In addition, the modifying unit 113 increases the generation probability of a sub-library constituting a linear regression equation with the smallest loss function among the loss functions of multiple linear regression equations generated with the same hyperparameters, and decreases the generation probability of other sub-libraries.

[0069] The regression equation regeneration unit 114 extracts nonlinear basis functions from a sub-library including multiple nonlinear basis functions based on the generation probability after correction by the correction unit 113, generates multiple linear regression equations combining the nonlinear functions of multiple types of sub-libraries, and estimates coefficients of the multiple linear regression equations by sparse estimation using the hyperparameters after correction by the correction unit 113 (step S7). The regression equation regeneration unit 114 also calculates loss functions of the multiple linear regression equations (step S8), and proceeds to step S5.

[0070] In step S5, when the regression equation regeneration unit 114 generates a linear regression equation a predetermined number of times, the output control unit 115 outputs the linear regression equation selected from the multiple linear regression equations to the output unit 12 (step S9). Then, this routine ends.

[0071] As described above, the information processing device 1 of this embodiment includes a memory unit 10 that stores library information, a detection result acquisition unit 111, a regression equation generation unit 112, a correction unit 113, a regression equation regeneration unit 114, and an output control unit 115.

[0072] The memory unit 10 stores library information 101 including sub-library definition information, which is information on multiple types of sub-libraries including nonlinear basis functions based on dependent variables or independent variables, and generation probability information, which is information on the generation probability of each of the multiple nonlinear basis functions included in each sub-library.

[0073] The detection result acquisition unit 111 acquires the detection results of the multiple sensors. The regression equation generation unit 112 calculates nonlinear basis functions of a sub-library using the detection results, extracts nonlinear basis functions from a sub-library including multiple nonlinear basis functions based on generation probability, generates multiple linear regression equations by combining the nonlinear basis functions of multiple types of sub-libraries, estimates coefficients of the multiple linear regression equations by machine learning, and calculates loss functions of the multiple linear regression equations using the calculation results.

[0074] The correction unit 113 corrects the generation probability and the hyperparameters of the machine learning based on the loss function of the linear regression formula. The regression formula regeneration unit 114 extracts a nonlinear basis function from a sub-library including a plurality of nonlinear basis functions based on the corrected generation probability, generates a plurality of linear regression formulas combining the nonlinear basis functions of the plurality of types of sub-libraries, estimates coefficients of the plurality of linear regression formulas by machine learning using the corrected hyperparameters, and calculates the loss function of the plurality of linear regression formulas.

[0075] When a predetermined condition is met, the output control unit 115 outputs the linear regression equation selected from the multiple linear regression equations to the output unit 12.

[0076] In this way, the information processing device 1 extracts nonlinear basis functions based on the generation probability from a sub-library containing multiple nonlinear basis functions, generates multiple linear regression formulas that combine the nonlinear basis functions of multiple types of sub-libraries, and regenerates the linear regression formulas while tuning the generation probability and machine learning hyperparameters until the specified conditions are met.

[0077] In this case, the information processing device 1 regenerates the linear regression equation while tuning the generation probability and the hyperparameters of machine learning, and narrows down the linear regression equation to be generated by changing the generation probability. As a result, the information processing device 1 can output an appropriate linear regression equation while narrowing the search space. In other words, the information processing device 1 can generate an appropriate model of a physical phenomenon.

[0078] Here, an example of temperature prediction based on information acquired from a sensor will be described with reference to the graphs of Figs. 7A to 7C. Fig. 7A is a diagram showing an example of temperature prediction by machine learning without considering a physical model. Fig. 7A includes graphs G1 and G2. Graph G1 is a prediction graph of temperature Temp1 predicted by the result of deep learning using measurement data by a sensor. Graph G2 is a graph of correct data for temperature Temp1. The vertical axis is temperature, and the horizontal axis is time. As shown in Fig. 7A, there is a discrepancy between the prediction graph and the graph of correct data.

[0079] FIG. 7B is a diagram showing an example of temperature prediction using a machine learning model that takes into account a known physical model. FIG. 7B includes graphs G3 and G4. Graph G3 is a graph of temperature Temp1 predicted based on a physical model without a sub-library like the present embodiment. Graph G4 is a graph of correct data for temperature Temp1. As shown in FIG. 7B, there is a deviation between the predicted graph and the correct data graph, although not as large as in FIG. 7A. Also, as shown in FIG. 7B, the deviation increases as time passes, indicating that this method is not suitable for long-term prediction.

[0080] FIG. 7C is a diagram showing an example of temperature prediction based on a linear regression equation generated by this embodiment. FIG. 7C includes graphs G5 and G6. Graph G5 is a graph of temperature Temp1 predicted based on a sub-library such as this embodiment. Graph G6 is a graph of correct data for temperature Temp1. As shown in FIG. 7C, the predicted graph and the correct data graph almost match. In this way, the information processing device 1 of this embodiment can output a linear regression equation that can make a more appropriate prediction by generating a linear regression equation using a sub-library.

[0081] Furthermore, the information processing device 1 estimates coefficients of a plurality of linear regression equations by sparse estimation. A linear regression equation is composed of many basis functions, and most of the coefficients are 0. Therefore, by using sparse estimation, the coefficients can be appropriately estimated.

[0082] Furthermore, the information processing device 1 generates a linear regression equation, which is an equation for calculating temperature, with temperature as a dependent variable. Since there are many different heat transfer phenomena, the search space tends to be wide when calculating temperature, but since the above-mentioned library information 101 is stored and the linear regression equation is generated / regenerated using the library information 101, it is possible to generate an appropriate model of the thermal phenomenon.

[0083] Furthermore, the information processing device 1 stores a plurality of types of sub-libraries including nonlinear basis functions of independent variables corresponding to speed, current, and voltage. In this manner, the information processing device 1 stores nonlinear basis functions related to speed, current, and voltage that are highly related to temperature, and generates a linear regression equation using the nonlinear basis functions, so that an appropriate model of a thermal phenomenon can be generated.

[0084] In addition, the information processing device 1 stores nonlinear basis functions for multiple types of sub-libraries, including a thermal conduction sub-library, a radiation sub-library, a forced convection sub-library, a natural convection sub-library, and a heat generation sub-library.Since the information processing device 1 stores sub-libraries corresponding to nodal equations, it can generate an appropriate model of thermal phenomena.

[0085] Furthermore, the output control unit 115 of the information processing device 1 outputs information on the selected plurality of linear regression formulas according to the ranking based on the loss function of each of the plurality of linear regression formulas to the output unit 12. In this manner, the information processing device 1 outputs information on the selected plurality of linear regression formulas according to the ranking based on the loss function, thereby enabling the user of the information processing device 1 to output information that can be compared and considered.

[0086] In the above embodiment, the information processing device 1 generates a linear regression equation for a thermal model. However, the information processing device 1 may generate a linear regression equation for a model of other physical phenomena (e.g., electrical resistance, physical deformation amount).

[0087] Although some embodiments of the present invention have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These novel embodiments can be implemented in various other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their modifications are included in the scope and spirit of the invention, and are included in the scope of the invention and its equivalents described in the claims. [Explanation of symbols]

[0088] 1 information processing device, 2 electronic device, 10 memory unit, 11 control unit, 12 output unit, 111 detection result acquisition unit, 112 regression equation generation unit, 113 correction unit, 114 regression equation regeneration unit, 115 output control unit.

Claims

1. A method for extracting a function for each of a plurality of sub-libraries, the function being included in a plurality of types of sub-libraries, the function being based on a dependent variable or an independent variable, based on a generation probability that the function is selected; generating a plurality of linear regression equations, which are equations for calculating the dependent variables, by combining the functions extracted from the different types of sub-libraries; Calculating a loss function for each of the plurality of linear regression equations; A regression equation generating unit; an output control unit that outputs one or more linear regression equations selected from the plurality of linear regression equations when a condition based on each loss function of the plurality of linear regression equations is satisfied; The function is a function included in any one of a plurality of types of sublibraries including a sublibrary of heat conduction, a sublibrary of radiation, a sublibrary of forced convection, a sublibrary of natural convection, and a sublibrary of heat generation. Information processing device.

2. A modification unit that modifies the generation probability and the hyperparameters based on the loss function; a regression equation regeneration unit that regenerates a linear regression equation by combining the functions extracted based on the modified generation probability; Further comprising: The information processing device according to claim 1 .

3. the dependent variable is a dependent variable corresponding to temperature; The information processing device according to claim 1 .

4. The independent variables are independent variables corresponding to speed, current and voltage. The information processing device according to claim 1 .

5. The output control unit outputs information on the plurality of linear regression equations selected based on a ranking based on each loss function of the plurality of linear regression equations to an output unit. The information processing device according to claim 1 .

6. The processing circuit includes: A step of extracting a function for each of the sub-libraries from a plurality of functions based on a dependent variable or an independent variable, the functions being included in a plurality of types of sub-libraries, based on a generation probability that the function is selected; generating a plurality of linear regression equations for calculating the dependent variables by combining the functions extracted from different types of the sub-libraries; Calculating a loss function for each of the plurality of linear regression equations; outputting one or more linear regression equations selected from the plurality of linear regression equations when a condition based on a loss function of each of the plurality of linear regression equations is satisfied; The function is a function included in any one of a plurality of types of sublibraries including a sublibrary of heat conduction, a sublibrary of radiation, a sublibrary of forced convection, a sublibrary of natural convection, and a sublibrary of heat generation. program.

7. An information processing method executed by an information processing device, comprising: A step of extracting a function for each of the sub-libraries from a plurality of functions based on a dependent variable or an independent variable, the functions being included in a plurality of types of sub-libraries, based on a generation probability that the function is selected; generating a plurality of linear regression equations for calculating the dependent variables by combining the functions extracted from different types of the sub-libraries; Calculating a loss function for each of the plurality of linear regression equations; outputting one or more linear regression equations selected from the plurality of linear regression equations when a condition based on a loss function of each of the plurality of linear regression equations is satisfied; The function is a function included in any one of a plurality of types of sublibraries including a sublibrary of heat conduction, a sublibrary of radiation, a sublibrary of forced convection, a sublibrary of natural convection, and a sublibrary of heat generation. Information processing methods.

Images