JP6657869B2 - Information processing apparatus, model generation program, and model generation method - Google Patents

Description

  The present invention relates to an information processing device, a model generation program, and a model generation method.

  Conventionally, control design is performed. In this control design, the target object of the control design is set as the measurement target, the set values of the parameters used for control of the measurement target are changed, and the state of the measurement target is measured until the state of the measurement target becomes a steady state. I do. In the control design, a model to be measured is generated based on the measured data.

  However, if the measurement is performed after waiting for the state of the measurement target to reach a steady state for each measurement condition, it takes time to collect measurement data. Therefore, a technique for performing quasi-stationary measurement has been proposed. In the quasi-stationary measurement, the set values of the parameters are changed at a speed that can be regarded as steady, and the time series data that can be obtained is used as the data in the steady state.

JP 2008-077376 A JP 2014-002519 A

Takako Shimojo et al., "Soot Modeling of Gasoline Engine Using Quasi-steady State Measurement", 1st Multi-Symposium, Control Division, 2014

  Quasi-stationary measurement is used when there is one parameter. However, a measurement target may use a plurality of parameters for control as the number of devices and the like increases. For this reason, when using a plurality of parameters, it is difficult to generate a model of a measurement target by quasi-stationary measurement.

  Therefore, for example, it is conceivable to perform a quasi-stationary measurement by sequentially performing a change in which only one parameter among a plurality of parameters is changed from a previous measurement point to control a measurement target using a Hilbert curve. .

  By the way, generally, at the time of model generation, each measurement data is equally evaluated.

  Therefore, it is conceivable to improve the prediction accuracy by changing the measurement data used for generating the model according to the purpose of the optimization. For example, it is conceivable to generate a model by making the measurement data dense for a specific area where the accuracy of the model is desired to be high, and coarsening the measurement data for the other area.

  However, even if the measurement data is dense for a specific region, the prediction accuracy of the model may not be improved.

  An object of one aspect is to provide an information processing apparatus, a model generation program, and a model generation method that can improve the prediction accuracy of a model.

  In the first plan, the information processing device has an acquisition unit and a generation unit. The acquisition unit is measurement data of each measurement point sequentially measured along the measurement path curve, the number of measurement points is increased in a specific region of the measurement target range as compared with the region other than the specific region, and For every two sides near the measurement path curve and in which the control parameter change is in the opposite direction, the measurement data of each measurement point that balances the number of measurement points located on the two sides is acquired. The measurement path curve is generated from a Hilbert curve arranged in a normalized space in which a measurement target range of each of a plurality of control parameters related to control of the measurement target is normalized. The generation unit generates a control model of the measurement target based on the acquired measurement data of each measurement point.

  According to one embodiment of the present invention, the prediction accuracy of a model can be improved.

FIG. 1 is an explanatory diagram illustrating an example of a system configuration. FIG. 2 is a diagram illustrating an example of a functional configuration of the information processing apparatus according to the first embodiment. FIG. 3 is a diagram illustrating an example of the normalized space. FIG. 4 is a diagram illustrating an example of a three-dimensional Hilbert curve. FIG. 5 is a diagram illustrating an example of a two-dimensional Hilbert curve from the first place to the third place. FIG. 6 is a diagram for explaining the correspondence between measurement data and measurement conditions. FIG. 7 is a diagram illustrating a change in the change parameter. FIG. 8 is a diagram schematically illustrating a flow of generating a model. FIG. 9 is a flowchart illustrating an example of a procedure of a model generation process according to the first embodiment. FIG. 10A is a diagram illustrating an example of a model. FIG. 10B is a diagram illustrating an example of a model. FIG. 11A is a diagram illustrating an example of measurement points used for generating a model. FIG. 11B is a diagram illustrating an example of measurement points used for generating a model. FIG. 12 is a diagram illustrating an example of grouping of the measurement path curves. FIG. 13 is a diagram illustrating an example of the measurement path curve. FIG. 14 is a flowchart illustrating an example of a procedure of a model generation process according to the second embodiment. FIG. 15 is a diagram illustrating a flow of acquiring measurement data during quasi-stationary measurement. FIG. 16 is a flowchart illustrating an example of a procedure of a model generation process according to the third embodiment. FIG. 17 is an explanatory diagram illustrating an example of a configuration of a computer that executes a model generation program.

  Hereinafter, embodiments of an information processing apparatus, a model generation program, and a model generation method disclosed in the present application will be described in detail with reference to the drawings. Note that the disclosed technology is not limited by the present embodiment. Further, the embodiments described below may be appropriately combined within a range that does not cause inconsistency.

[System configuration]
In the control design, how to control the measurement target is designed using a model generated from measurement data obtained by measuring the measurement target. The system 1 is a system that generates a model to be measured. In the present embodiment, a case will be described as an example where the measurement target is an engine and control design of the engine is performed. In engine control design, in order to determine a numerical value used for controlling the engine, how the state of the engine is changed by operating the engine is measured, and the engine is modeled based on the measured data. For example, in the control design of an engine, a model is generated by measuring how the exhaust gas and the fuel efficiency change due to a change in the valve opening and the fuel injection amount. In the control design of the engine, how to control the engine using the generated model is designed. For example, in the control design of an engine, a valve opening and a fuel injection amount are designed to suppress exhaust gas and fuel consumption while obtaining a required output from a generated model.

  FIG. 1 is an explanatory diagram illustrating an example of a system configuration. As shown in FIG. 1, the system 1 includes an information processing device 10 and an engine 11.

  The engine 11 is an object that has been subjected to control design.

  The information processing device 10 is a device that generates a model of the engine 11. The information processing apparatus 10 is, for example, a computer such as a personal computer or a server computer. The information processing device 10 may be implemented as one computer, or may be implemented by a plurality of computers. In this embodiment, an example in which the information processing apparatus 10 is a single computer will be described.

  The information processing device 10 operates the engine 11 under various measurement conditions. For example, the information processing apparatus 10 changes a measurement condition set as a parameter for controlling the engine 11 and outputs an operation instruction to the engine 11. This parameter is input information to the model in control design. The parameters for controlling the engine 11 include, for example, a valve opening and a fuel injection amount. The parameters are not limited to these.

Further, the information processing device 10 acquires measurement data indicating the state of the engine 11 under various measurement conditions. For example, the information processing apparatus 10 acquires measurement data indicating the state of the engine 11 under various measurement conditions in which only one of a plurality of parameters is changed from the immediately preceding measurement condition. At this time, the information processing apparatus 10 changes the value of any one parameter from the immediately preceding measurement condition at a slow speed that can be regarded as substantially steady for each measurement condition. The information processing apparatus 10 uses the measured data sequentially acquired while changing one of the parameters at a slow speed as data in a steady state. For example, the information processing apparatus 10 changes the valve opening and the fuel injection amount one by one at a slow speed that can be regarded as substantially steady, and acquires measurement data indicating the state of the engine 11. As the measurement data, for example, data of each concentration of NO _x (nitrogen oxide), PM (particulate matter), CO ₂ (carbon dioxide) contained in the exhaust gas, and fuel consumption data are acquired.

[Configuration of Information Processing Device]
Next, an information processing apparatus 10 according to the present embodiment will be described. FIG. 2 is a diagram illustrating an example of a functional configuration of the information processing apparatus according to the first embodiment. As illustrated in FIG. 2, the information processing device 10 includes an external I / F (interface) unit 20, an input unit 21, a display unit 22, a storage unit 23, and a control unit 24. Note that the information processing device 10 may include devices other than the above devices.

  The external I / F unit 20 is an interface that inputs and outputs information to and from other devices. As the external I / F unit 20, various input / output ports such as a USB (Universal Serial Bus) and a network interface card such as a LAN card can be adopted.

The external I / F unit 20 transmits and receives various information to and from other devices via a communication cable (not shown). For example, the external I / F unit 20 outputs an operation instruction specifying a measurement condition to a control device (not shown) that controls the engine 11. The control device operates the engine 11 under the designated measurement conditions. The control device measures the state of the engine 11 at a predetermined cycle. For example, the control device measures the respective concentrations of NO _x , PM, and CO ₂ contained in the exhaust gas of the engine 11 and the fuel consumption at a cycle of 16 ms. Then, the control device outputs data indicating the measured state of the engine 11 in a predetermined cycle as measurement data in association with the measurement time. That is, the control device outputs the measurement data in chronological order every time the measurement is performed. The external I / F unit 20 receives the measurement data output from the control device.

  The input unit 21 is an input device for inputting various information. Examples of the input unit 21 include an input device that receives an input of an operation such as a mouse or a keyboard. The input unit 21 receives input of various types of information related to control design. For example, the input unit 21 receives an input of a measurement target range in which measurement is performed by changing a parameter for each parameter that controls the engine 11. Further, the input unit 21 receives an input of a change speed at which the parameter is changed for each parameter for controlling the engine 11. The input unit 21 receives an input of a period until measurement data corresponding to the measurement condition is obtained for each parameter that controls the engine 11. Hereinafter, a period until the measurement data corresponding to the measurement condition is obtained is also referred to as “dead time”. The input unit 21 receives an operation input from a user and inputs operation information indicating the content of the received operation to the control unit 24.

  The display unit 22 is a display device that displays various information. The display unit 22 includes a display device such as an LCD (Liquid Crystal Display) or a CRT (Cathode Ray Tube). The display unit 22 displays various information. For example, the display unit 22 displays various screens such as an operation screen.

  The storage unit 23 is a storage device that stores various data. For example, the storage unit 23 is a storage device such as a hard disk, a solid state drive (SSD), and an optical disk. The storage unit 23 may be a rewritable semiconductor memory such as a random access memory (RAM), a flash memory, or a non-volatile random access memory (NVSRAM).

  The storage unit 23 stores an OS (Operating System) executed by the control unit 24 and various programs. For example, the storage unit 23 stores various programs including a program for executing a model generation process described later. Further, the storage unit 23 stores various data used in the program executed by the control unit 24. For example, the storage unit 23 stores the measurement path information 30 and the measurement data 31.

  The measurement path information 30 is data storing how to change the measurement conditions set in the parameters for controlling the engine 11. For example, the measurement path information 30 stores a plurality of measurement conditions for the parameters in the order in which they are changed.

  The measurement data 31 is data in which measurement data of the engine 11 is stored. The measurement data 31 stores the measurement data for each measurement condition in association with the measurement condition.

  The control unit 24 is a device that controls the information processing device 10. As the control unit 24, an electronic circuit such as a CPU (Central Processing Unit) and an MPU (Micro Processing Unit), and an integrated circuit such as an ASIC (Application Specific Integrated Circuit) and an FPGA (Field Programmable Gate Array) can be adopted. The control unit 24 has an internal memory for storing programs and control data defining various processing procedures, and executes various processing by using these. The control unit 24 functions as various processing units when various programs operate. For example, the control unit 24 includes a reception unit 40, a measurement condition generation unit 41, an output unit 42, an acquisition unit 43, a storage unit 44, and a model generation unit 45.

  The reception unit 40 performs various receptions. For example, the receiving unit 40 receives various operation instructions related to control design. For example, the receiving unit 40 displays an operation screen related to control design on the display unit 22, and receives an input of an operation instruction instructing start of generation of a model from the operation screen. The receiving unit 40 receives an input of a parameter measurement target range for each parameter for controlling the engine 11 from the operation screen. Further, the receiving unit 40 receives an input of a change speed at which the parameter is changed for each parameter that controls the engine 11. The receiving unit 40 receives an input of a dead time for each parameter for controlling the engine 11. The measurement target range of each parameter is determined by the user, for example, according to the range of the parameter for modeling the engine 11. The change speed of each parameter is set, for example, to a change speed at which the user can observe changes in the state of the engine 11 in advance by individually changing the parameters and change the state of the engine 11 to an almost steady state. Is determined. The dead time of each parameter is determined by the user, for example, by individually changing each parameter and observing a change in the state of the engine 11 in advance by the user.

  The measurement condition generation unit 41 generates a plurality of measurement conditions for a plurality of parameters for controlling a measurement target in a predetermined order. For example, the measurement condition generating unit 41 changes only one of the plurality of parameters from the immediately preceding measurement condition in a space defined by the respective measurement target ranges of the plurality of parameters for controlling the measurement target, and sets the plurality of measurement parameters. Generate measurement conditions. For example, the measurement condition generation unit 41 is a measurement condition in which only one of the plurality of parameters changes from the immediately preceding measurement condition, and comprehensively covers the space defined by the measurement target range of the plurality of parameters. Generate a plurality of changing measurement conditions. For example, the measurement condition generation unit 41 generates a plurality of measurement conditions using a space filling curve arranged in a normalized space obtained by normalizing the measurement target range of each of the plurality of parameters. The space filling curve is a curve that comprehensively changes the space by sequentially changing the value of any one parameter in the space formed by a plurality of parameters. As this space filling curve, for example, a Hilbert curve is used.

  Here, the flow of generating the measurement conditions will be described using a specific example. The measurement condition generation unit 41 normalizes the measurement target ranges of the valve opening and the fuel injection amount for controlling the engine 11 to a range of 0 to 1, respectively. FIG. 3 is a diagram illustrating an example of the normalized space. The normalized space in which the measurement target ranges of the valve opening and the fuel injection amount are normalized is a two-dimensional space in which the valve opening and the fuel injection amount are in the range of 0 to 1. The measurement condition generator 41 arranges the Hilbert curve in the normalized space. FIG. 3 shows an example of a two-dimensional Hilbert curve. The Hilbert curve is one of the fractal figures and is a curve obtained by recursively repeating a U-shaped trajectory, and is one of the space filling curves in which the limit of repetition coincides with the region. The Hilbert curve is a one-stroke curve composed of a combination of changes in one parameter. Since the Hilbert curve is a space-filling curve, a curve formed by performing repetition to some extent exhaustively fills the area. In addition, the Hilbert curve has a portion that changes in the opposite direction in most portions nearby. Hilbert curves can be created in any dimension. FIG. 4 is a diagram illustrating an example of a three-dimensional Hilbert curve.

  An example of a method of forming a two-dimensional Hilbert curve will be described. There are four types of two-dimensional Hilbert curves (RUL (n); DLU (n); LDR (n); URD (n)) due to the difference in starting conditions. n is the number of repetitions (order, order). The Hilbert curve of each order is drawn by rules shown in the following equations (1) to (4).

RUL (n) = URD (n-1) → RUL (n-1) ↑ RUL (n-1) ← DLU (n-1) (1)
DLU (n) = LDR (n-1) ↓ DLU (n-1) ← DLU (n-1) ↑ RUL (n-1) (2)
LDR (n) = DLU (n-1) ← LDR (n-1) ↓ LDR (n-1) → URD (n-1) (3)
URD (n) = RUL (n-1) ↑ URD (n-1) → URD (n-1) ↓ LDR (n-1) (4)

  However, it is assumed that RUL (0) = DLU (0) = LDR (0) = URD (0) = “”. "" Means no operation.

  Specifically, RUL (n) is described. The Hilbert curve RUL (n) divides each side of the square area into 2n equal parts and divides the square area into small square areas. It is composed by tying. FIG. 5 is a diagram illustrating an example of a two-dimensional Hilbert curve from the first place to the third place.

  The Hilbert curve RUL (1) having an order n of 1 is as follows from Expression (1).

RUL (1) = URD (0) → RUL (0) ↑ RUL (0) ← DLU (0)
= → ↑ ←

  The Hilbert curve RUL (1) is obtained by connecting the centers of four small square regions formed by dividing the square region into two equal sides, in the order of lower left, lower right, upper right, and upper left.

  Similarly, a Hilbert curve RUL (2) having an order n of 2 is as shown below.

RUL (2) = URD (1) → RUL (1) ↑ RUL (1) ← DLU (1)
= (↑ → ↓) → (→ ↑ ←) ↑ (→ ↑ ←) ← (↓ ← ↑)
= ↑ → ↓ →→ ↑ ← ↑ → ↑ ←← ↓ ← ↑

  The Hilbert curve RUL (2) is a curve formed by connecting the centers of 16 small square areas formed by dividing the square area into four equal parts on each side in the order of arrows from the lower left.

  The measurement condition generation unit 41 normalizes the measurement target ranges of the valve opening and the fuel injection amount to a range of 0 to 1, respectively. The measurement condition generation unit 41 sets the center of the lower left small square in the normalized two-dimensional normalized space to (0; 0) and the center of the upper right small square to (1; 1). The Hilbert curves are arranged in association with each other.

  The measurement condition generation unit 41 returns the normalized space in which the Hilbert curve is arranged to the original space of the measurement target range for each of the valve opening and the fuel injection amount, and associates the Hilbert curve with the original space. A curve obtained by associating the Hilbert curve with the original space is a measurement path curve indicating a path to be actually measured. The measurement condition generation unit 41 generates measurement conditions along a measurement path curve. Even in the original space, only one parameter of the measurement path curve changes. The measurement condition generation unit 41 generates a measurement condition that changes the value of a changing parameter at a changing speed of the parameter along the measurement path curve. For example, when the change speed of the parameter is 1 per second, the measurement condition generation unit 41 generates a measurement condition that changes the value of the parameter by 1 per second. As a result, only one parameter changes from the immediately preceding measurement condition, and a measurement condition in which each parameter comprehensively changes in a space defined by the measurement target range of each parameter is generated.

  The measurement condition generation unit 41 stores the generated measurement conditions in the measurement path information 30 in the order along the measurement path curve. That is, the measurement condition generation unit 41 stores a plurality of measurement conditions in the measurement path information 30 in association with the order of change.

  The output unit 42 causes the external I / F unit 20 to output an operation instruction specifying a measurement condition to a control device (not shown) that controls the engine 11. For example, the output unit 42 outputs an operation instruction for operating the engine 11 under the measurement conditions in the order of the order stored in the measurement path information 30. The parameters of the engine 11 are changed to the measurement conditions, and the engine 11 operates under the measurement conditions. Any parameter of the engine 11 changes at such a slow speed that it can be regarded as steady based on the operation instruction. In the external I / F unit 20, measurement data indicating the state of the engine 11 whose parameters have been changed to the measurement conditions is received in a time series. The measurement data received in this time series can be regarded as steady state data.

  The acquisition unit 43 acquires measurement data that is received by the external I / F unit 20 and that indicates the state of the engine 11.

  The storage unit 44 stores the measurement data acquired by the acquisition unit 43 in the measurement data 31 in association with the measurement conditions. Here, the engine 11 takes a long time to react even if the measurement conditions are changed, and there is a dead time until the measurement data corresponding to the measurement conditions is acquired.

  Therefore, the storage unit 44 specifies a change parameter that changes under the measurement condition of the operation instruction output from the output unit 42. Then, the storage unit 44 stores the measurement data acquired by the acquisition unit 43 in the measurement data 31 in association with the measurement condition of the specified change parameter before the dead time. FIG. 6 is a diagram for explaining the correspondence between measurement data and measurement conditions. In the example of FIG. 6, the measurement conditions are changed from U (1) to U (6) at times 1 to 6. In addition, measurement data X (1) to X (6) are acquired at times 1 to 6. In the example of FIG. 6, it is assumed that the dead time is 3. In this case, the storage unit 44 stores the measurement data three hours earlier (three steps before) in association with the measurement data. In the example of FIG. 6, U (1) to U (6) and X (4) to X (9) are stored in the measurement data 31 in association with each other. As a result, the measurement data 31 stores the measurement data that can be regarded as a steady state for each measurement condition.

  By the way, in the measurement of this embodiment, the change parameter of the measurement condition may change during the dead time of one parameter. FIG. 7 is a diagram illustrating a change in the change parameter. In the example of FIG. 7, the timing at which the change parameter changes from the first parameter to the second parameter is shown. The measurement data at the timing 50 at which the first parameter is changed is acquired at the timing 51 after the dead time. In the example of FIG. 7, at the timing 51, the second parameter changes. As described above, when the change parameter of the measurement condition changes during the dead time, the measurement data may be affected by the change of the change parameter. For this reason, when the parameter for which the measurement condition changes changes, the storage unit 44 discards the measurement data for the dead time after the parameter changes. In the example of FIG. 7, the storage unit 44 discards measurement data for five measurement cycles, which is the dead time of the first parameter, from the timing at which the change parameter changes, without storing the measurement data in the measurement data 31. This makes it possible to generate a model excluding measurement data that may have been affected by a change in a plurality of parameters. When the measurement data is discarded in this manner, the order n of the Hilbert curve is preferably set to about 3 to 5. This is because the higher the order of the Hilbert curve, the greater the number of change points at which the change parameter changes, and many measurement data are discarded.

The model generation unit 45 generates a model of the engine 11. For example, for each type of measurement data, the model generation unit 45 generates a model by performing machine learning using the measurement conditions stored in the measurement data 31 and the measurement data corresponding to the measurement conditions. For example, the model generation unit 45 weights the surrounding data by regression analysis using a LOLMOT (Local Linear Model Tree) or a Gaussian Process for each of the concentrations of NO _x , PM, and CO ₂ and the fuel consumption. A model is generated using a method of outputting an average. As a result, the concentrations of NO _x , PM, and CO ₂ under various measurement conditions and the fuel efficiency can be predicted from the generated model.

  Here, a description will be given using a specific example. FIG. 8 is a diagram schematically illustrating a flow of generating a model. In the information processing apparatus 10, the receiving unit 40 receives an input of a parameter measurement target range, a parameter change speed, and a parameter dead time for each parameter that controls the engine 11.

  The measurement condition generation unit 41 generates a plurality of measurement conditions for a plurality of parameters for controlling the engine 11 in a predetermined order. For example, the measurement condition generation unit 41 arranges a Hilbert curve in a normalized space obtained by normalizing a measurement target range of each of a plurality of parameters, and sets a measurement condition along a measurement path curve in which the Hilbert curve corresponds to the space. Generate. The measurement condition generation unit 41 stores the measurement conditions generated along the measurement path curve in the measurement path information 30 in association with the order along the measurement path curve.

  The output unit 42 outputs an operation instruction for operating the engine 11 under the measurement conditions in the order of the order stored in the measurement path information 30.

  The acquisition unit 43 acquires measurement data indicating the state of the engine 11.

  The storage unit 44 stores the measurement data acquired by the acquisition unit 43 in the measurement data 31 in association with the measurement conditions. For example, the storage unit 44 specifies a change parameter that changes under the measurement condition of the operation instruction output from the output unit 42. Then, the storage unit 44 stores the measurement data acquired by the acquisition unit 43 in the measurement data 31 in association with the measurement condition of the specified change parameter before the dead time. Further, when the change parameter of the measurement condition changes, the storage unit 44 discards the measurement data for the dead time after the change parameter changes.

  The model generation unit 45 generates a model by performing machine learning using the measurement condition stored in the measurement data 31 and the measurement data corresponding to the measurement condition for each type of the measurement data. This makes it possible to predict measurement data under various measurement conditions for each type of measurement data from the generated model.

  Here, in the conventional quasi-stationary measurement, measurement data can be acquired for only one parameter. The measurement target may have a plurality of parameters to be controlled as the number of devices and the like increases. For this reason, the correlation with other parameters cannot be completely expressed only by using the quasi-stationary state for one parameter. For example, with multiple parameters, the slow speed under quasi-stationary conditions varies depending on the direction of the change. For this reason, when changing a plurality of parameters simultaneously, an infinite speed setting is required, which is practically impossible. Further, when a plurality of parameters are changed at the same time, the dead time differs depending on the direction of the change. For this reason, in the conventional quasi-stationary measurement, it is difficult to collect quasi-stationary data on a plurality of parameters. In some cases, parameters having strong non-linearity exist as parameters, and an accurate model cannot be generated using only measurement data of one parameter.

  On the other hand, the information processing apparatus 10 according to the present embodiment generates a measurement condition for changing any one parameter among a plurality of parameters for controlling the measurement target from the immediately preceding measurement condition. For this reason, the information processing apparatus 10 can easily set the changing speed of the parameter and the dead time of the parameter. In addition, the information processing apparatus 10 according to the present embodiment generates a measurement condition that comprehensively changes in a space based on the measurement target range of each of the plurality of parameters that control the measurement target. For this reason, the information processing apparatus 10 can sufficiently collect measurement data even when a parameter having strong nonlinearity exists, and can generate an accurate model.

[Processing flow]
Next, a flow of a model generation process in which the information processing apparatus 10 according to the present embodiment generates a model will be described. FIG. 9 is a flowchart illustrating an example of a procedure of a model generation process according to the first embodiment. This model generation processing is executed at a predetermined timing, for example, at a timing when an input of an operation instruction to start generation of a model is received from an operation screen.

  As shown in FIG. 9, the receiving unit 40 receives an input of a parameter measurement target range, a parameter change speed, and a parameter dead time for each parameter for controlling the engine 11 from the operation screen (S10).

  The measurement condition generation unit 41 generates a plurality of measurement conditions for a plurality of parameters for controlling the engine 11 in a predetermined order (S11). For example, the measurement condition generation unit 41 arranges a Hilbert curve in a normalized space obtained by normalizing a measurement target range of each of a plurality of parameters, and sets a measurement condition along a measurement path curve in which the Hilbert curve corresponds to the space. Generate. The measurement condition generation unit 41 stores the measurement conditions generated along the measurement path curve in the measurement path information 30 in association with the order along the measurement path curve (S12).

  The output unit 42 outputs an operation instruction for operating the engine 11 under the measurement conditions in the order of the order stored in the measurement path information 30 (S13). The acquisition unit 43 acquires measurement data (S14). The acquisition unit 43 stores the acquired measurement data in the measurement data 31 in association with the measurement conditions (S15). For example, the storage unit 44 stores the acquired measurement data in the measurement data 31 in association with the measurement conditions before the dead time of the change parameter. Further, when the change parameter of the measurement condition changes, the storage unit 44 discards the measurement data for the dead time after the change parameter changes.

  The model generation unit 45 generates a model by performing machine learning using the measurement conditions stored in the measurement data 31 and the measurement data corresponding to the measurement conditions for each type of measurement data (S16), and ends the process. I do.

[effect]
As described above, the information processing apparatus 10 according to the present embodiment sets only one of the plurality of control parameters to one in the space defined by the plurality of control parameters related to the control of the measurement target. By sequentially performing a change that changes from the immediately preceding measurement point, a measurement condition for performing measurement covering the space of the measurement target range is generated. The information processing apparatus 10 sequentially outputs the generated operation instructions of the plurality of measurement conditions. The information processing apparatus 10 acquires measurement data indicating the state of the measurement target whose control parameter has been changed to the measurement condition according to the output operation instruction. Thereby, the information processing apparatus 10 can collect quasi-stationary data on a plurality of parameters. Accordingly, the information processing apparatus 10 can generate a model of the measurement target even when a plurality of parameters are used for controlling the measurement target.

  Further, the information processing apparatus 10 according to the present embodiment uses the measurement path curve generated from the Hilbert curve arranged in the normalized space in which the measurement target range of each of the plurality of control parameters is normalized. Generate As a result, the information processing apparatus 10 can generate a measurement condition in which only one of the control parameters changes from the immediately preceding measurement point and exhaustively changes in the space defined by the measurement target range of each of the plurality of control parameters. .

  Further, the information processing apparatus 10 according to the present embodiment receives a dead time until measurement data corresponding to the measurement condition is obtained. The information processing device 10 stores the acquired measurement data in association with the measurement condition for the time before the dead time. Thereby, the information processing apparatus 10 can associate the measurement condition with the quasi-stationary measurement data corresponding to the measurement condition.

  Further, the information processing apparatus 10 according to the present embodiment discards the measurement data for the dead time when the control parameter whose measurement condition changes changes. Thereby, the information processing apparatus 10 can exclude the measurement data that may have been affected by the change of the plurality of control parameters.

  Further, the information processing apparatus 10 according to the present embodiment moves a point on a space filling curve constructed on a computer and having each of a plurality of control parameters related to control of a measurement target as a variable along the space filling curve. Let it. The information processing apparatus 10 acquires measurement data indicating the state of the measurement target when the measurement target is controlled based on each value of the plurality of control parameters corresponding to the point after the movement, every time the point moves. The information processing device 10 generates a control model of a measurement target based on the acquired plurality of measurement data. Accordingly, the information processing apparatus 10 can generate a model of the measurement target even when a plurality of parameters are used for controlling the measurement target.

  Next, a second embodiment will be described. In a second embodiment, a case will be described in which the measurement data used for generating the model is changed according to the purpose of optimization to improve the prediction accuracy. Since the configurations of the system 1 and the information processing apparatus 10 according to the second embodiment are substantially the same as those of the system 1 and the information processing apparatus 10 according to the first embodiment illustrated in FIGS. 1 and 2, mainly different portions will be described. .

  The acquisition unit 43 acquires measurement data used for generating a model according to the purpose of optimization. For example, the acquisition unit 43 acquires the measurement data of each measurement point from the measurement data 31 by increasing the number of measurement points in a specific region of the measurement target range as compared with the region other than the specific region.

  Here, an example of a change in the model due to the density of the measurement points will be described. FIG. 10A is a diagram illustrating an example of a model. The example of FIG. 10A shows a result of obtaining an approximate function as a model from measurement data of measurement points measured uniformly. Generally, at the time of model generation, measurement data at each measurement point is evaluated equally. In the example of FIG. 10A, the error is small as a whole, but a slight shift occurs near the minimum value because the measurement data at each measurement point is uniformly evaluated.

  FIG. 10B is a diagram illustrating an example of a model. The example of FIG. 10B shows the result of obtaining a function approximating as a model from the measurement data of each measurement point by increasing the number of measurement points near the minimum value other than those other than the minimum value. In the example of FIG. 10B, the error is large around the vicinity of the minimum value, but there is almost no deviation near the minimum value. At the time of model generation, the measurement data at each measurement point is equally evaluated. Therefore, when the density of the measurement points is adjusted, the accuracy of the model is increased in a portion where the density of the measurement points is high. That is, with respect to the optimization calculation, adjusting the density of the measurement points in a specific area makes the model highly accurate in the specific area.

In control design, in order to find the optimal control, it may be better to generate a model with high accuracy for a specific area of the measurement target range, rather than generating a model with a small error for the entire measurement target range. . For example, when trying to reduce NO _x contained in the exhaust gas of the engine 11, the control design measures NO _x as measurement data, and increases the number of measurement points near the minimum value of NO _x compared to other values than the minimum value. Therefore, it is preferable to generate a model with high accuracy around the minimum value.

  By the way, in the quasi-stationary measurement, since the set value of the parameter is changed at a speed that can be regarded as a steady state, hysteresis in which the influence of the previous measurement point remains in the measurement data measured at each measurement point occurs. On the other hand, in the Hilbert curve, as shown in FIGS. For this reason, for example, when a model is generated using all the measurement points of the measurement data 31 as in the first embodiment, the measurement points of the same degree exist in the part that changes in the opposite direction, and the influence of the hysteresis is reduced. Almost offset.

  FIG. 11A is a diagram illustrating an example of measurement points used for generating a model. The measurement points A1 to A5 and the measurement points A11 to A15 in the example of FIG. 11A are portions where a certain parameter changes in the opposite direction. Measurement points A1 to A5 have an upward hysteresis in the measurement data, and measurement points A11 to A15 have a downward hysteresis in the measurement data. When the model is generated using the measurement data of the measurement points A1 to A5 and A11 to A15, the influence of the hysteresis is almost canceled. For example, in the example of FIG. 11A, the portion surrounded by the line L1 includes the same number of measurement points A1 to A3 and the measurement points A13 to A15 whose parameters change in the opposite direction. As a result, the effect of the hysteresis is almost canceled in the generated model, so that the effect of the hysteresis is suppressed to a small value.

  FIG. 11B is a diagram illustrating an example of measurement points used for generating a model. In the example of FIG. 11B, the measurement points A1 to A5 and the measurement points A21 and A22 are portions where a certain parameter changes in the opposite direction. Measurement points A1 to A5 have an upward hysteresis in the measurement data, and measurement points A21 and A22 have a downward hysteresis in the measurement data. When a model is generated using the measurement data of the measurement points A1 to A5, A21, and A22, the influence of the upward hysteresis remains. For example, in the example of FIG. 11B, in the portion surrounded by the line L1, there are more measurement points A1 to A3 having an upward hysteresis than the measurement points A22 having a downward hysteresis. If there is a large difference in the number of measurement points located on two opposite sides where the change in the parameter is opposite, the generated model will have reduced model prediction accuracy. In the case of the example of FIG. 11B, the generated model is affected by upward hysteresis, and the prediction accuracy of the model is reduced.

  The acquisition unit 43 acquires, from the measurement data 31, measurement data of a larger number of measurement points in a specific region of the measurement target range than in regions other than the specific region. In addition, the acquisition unit 43 acquires measurement data of each measurement point on each of two sides near the measurement path curve and in which the parameter change is in the opposite direction, by balancing the number of measurement points located on the two sides. For example, the acquisition unit 43 sets the number of measurement points on two measurement sides corresponding to the switching width of the parameter in which the widths of the two sides facing each other in the measurement path curve change and the parameter change in the opposite direction. Obtain the measurement data of For example, the acquisition unit 43 divides the measurement path curve into groups for every two sides that are near and whose parameter changes are in opposite directions. For example, the acquisition unit 43 divides the two sides into groups for each two sides of the measurement path curve that are closest to each other and whose parameter changes are in opposite directions. When there are a plurality of sides which are closest to the sides of the measurement path curve and whose parameter changes are in opposite directions, the acquiring unit 43 determines the numbers of the sides along the measurement path curve. Then, the acquiring unit 43 divides the side of the measurement path curve and the side having the smallest number among the sides closest to the relevant side and in which the parameter change is in the opposite direction. In addition, the acquisition unit 43 divides the measurement path curve into groups each of which is not adjacent to a side whose parameter change is in the opposite direction.

  FIG. 12 is a diagram illustrating an example of grouping of the measurement path curves. In the example of FIG. 12, an arrow indicates a group in the vicinity of the measurement path curve and every two sides in which the parameter change is in the opposite direction.

  The acquisition unit 43 calculates the average of the measurement data of the measurement points on the side belonging to the group for each group. For example, when the group is a group of two sides in the vicinity of which the parameter changes in the opposite direction, the acquisition unit 43 calculates the average of the measurement data of the measurement points on the two sides. Further, when the parameter change is a group of one side where the side whose parameter change is in the opposite direction is not in the vicinity, the acquisition unit 43 calculates the average of the measurement data of the measurement points on the one side.

  The acquisition unit 43 determines a thinning width according to the average for each group. Then, the acquisition unit 43 extracts the measurement data of the measurement points from the measurement data 31 by thinning out the measurement points by the determined thinning width from the measurement points on the sides belonging to the group for each group.

  Here, a method of determining the thinning width will be described. In the case of the optimization problem of minimization, the acquisition unit 43 decreases the thinning width as the measurement data is smaller, and acquires the measurement data of more measurement points as the measurement data is smaller. In the case of the optimization problem of maximization, the acquisition unit 43 reduces the thinning width as the measurement data increases, and acquires the measurement data of more measurement points as the measurement data increases. In addition, in the case of an optimization problem that emphasizes the vicinity of a predetermined value, the acquisition unit 43 reduces the thinning width as the difference between the measurement data and the predetermined value is smaller, and increases the number of measurement points as the difference between the measurement data and the predetermined value is smaller. Obtain the measurement data of

  For example, let Si be the average of the measurement data of group i. The minimum value p of the average Si of each group can be expressed by the following equation (5). In addition, the maximum value q of the average Si of each group can be expressed by the following equation (6). Further, the minimum of the thinning width is set to k, and the maximum of the thinning width is set to m. The minimum k of the thinning width and the maximum m of the thinning width may be determined in advance, and the receiving unit 40 may display an operation screen related to control design and receive an input.

p = min (Si) (5)
q = max (Si) (6)

  In the case of the optimization problem of minimization, the acquisition unit 43 obtains the thinning width CB of the group i from the following equation (7).

CB = floor (((Si−p) × (mk)) / (q−p)) + k (7)
Here, floor () is an operation for calculating the value of the integer part in ().

  In the case of the optimization problem of maximization, the obtaining unit 43 obtains the thinning width CB of the group i from the following equation (8).

CB = floor (((q−Si) × (mk)) / (q−p)) + k (8)

  In addition, in the case of an optimization problem that emphasizes the vicinity of the predetermined value a, the acquiring unit 43 obtains the average Si ′ of the square of the difference between the measurement data and the specified value a for each group. Note that the average Si 'may be the average of the absolute values of the differences. The minimum value P of the average Si 'of each group can be expressed by the following equation (9). Further, the maximum value Q of the average Si 'of each group can be expressed as in the following equation (10).

P = min (Si ') (9)
Q = max (Si ′) (10)

  In the case of an optimization problem that emphasizes the vicinity of the predetermined value a, the acquisition unit 43 obtains the thinning width CB of the group i from the following equation (11).

CB = floor (((Si′−P) × (mk)) / (Q−P)) + k (11)

  The acquisition unit 43 acquires measurement data of the measurement points from the measurement data 31 by thinning out the measurement points from the measurement points on the sides belonging to the group by the determined thinning width for each group.

  Here, a description will be given using a specific example. FIG. 13 is a diagram illustrating an example of the measurement path curve. FIG. 13 shows a part of the measurement path curve 70. In the quasi-stationary measurement, measurement data indicating the state of the engine 11 is acquired by changing the set value of any one parameter along the measurement path curve 70 at a speed that can be regarded as steady. The measurement path curve 70 is divided into sides for each parameter that changes, and has sides 71 to 74. FIG. 13 shows a measurement point A at which measurement data is obtained on sides 71 to 74. The sides 71 and 73 are in the same group because they are near and the parameter changes are in opposite directions. The sides 72 and 74 are grouped by one side, because there are no sides near the parameter change in the opposite direction.

  The acquisition unit 43 acquires measurement data indicating the state of the engine 11 at each measurement point A from a control device (not shown) that controls the engine 11 via the external I / F unit 20.

  The storage unit 44 stores the measurement data acquired by the acquisition unit 43 in the measurement data 31 in association with the measurement conditions. Further, when a parameter whose measurement condition changes changes, the storage unit 44 discards the measurement data for the dead time after the parameter changes. The measurement data of the measurement point A for the dead time on the sides 71 to 74 in FIG. 13 is discarded.

  Here, the average of the measurement data of the measurement points A1 to A2 and A5 to A6 on the side 71 and the side 73 is denoted as S1. The average of the measurement data at each of the measurement points A3 to A4 on the side 72 is represented as S2. The average of the measurement data of each of the measurement points A7 to A8 on the side 74 is represented as S3.

  For example, S1 = 15, S2 = 10, and S3 = 30. Further, the minimum value of the average Si of each group is p = 10, and the maximum value of the average Si of each group is q = 20. The thinning width is set to 2 to 10 (k = 2, m = 10).

  For example, in the case of the optimization problem of minimization, the thinning width CB1 of the group of the side 71 and the side 73 is obtained from the above equation (7) as follows.

CB1 = floor (((S1-p) × (mk)) / (q-p)) + k
= Floor (((15-10) x (10-2)) / (20-10)) + 2
= Floor (4) +2
= 6

  Similarly, the thinning width CB2 of the group of the side 72 is obtained from the above equation (7) as follows.

CB2 = floor (((S2-p) * (mk)) / (q-p)) + k
= Floor (((10-10) x (10-2)) / (20-10)) + 2
= Floor (0) +2
= 2

  Similarly, the thinning width CB3 of the group of the side 74 is obtained from the above equation (7) as follows.

CB3 = floor (((S3-p) * (mk)) / (q-p)) + k
= Floor (((30-10) x (10-2)) / (20-10)) + 2
= Floor (8) + 2
= 10

  The acquisition unit 43 reads the measurement data of the measurement points at every six measurement points from the measurement data 31 for the sides 71 and 73. The acquisition unit 43 reads the measurement data of the measurement points at every second measurement point from the measurement data 31 for the side 72. The acquisition unit 43 reads the measurement data of the measurement points of the side 74 from the measurement data 31 at every ten measurement points.

  The model generation unit 45 generates a model by performing machine learning using the measurement data of the measurement points read by the acquisition unit 43. Note that the acquisition unit 43 stores the read measurement data of each measurement point in the storage unit 23, and the model generation unit 45 generates a model from the measurement data of each measurement point stored in the storage unit 23. Good.

  As a result of the thinning, the number of measurement points used for generating the model on the side 71 and the side 73 is reduced to 1/6 from before the thinning. On the side 72, the number of measurement points used for generation of the model is １ ／ from the number before the thinning. On the side 74, the number of measurement points used for generation of the model is 1/10 from that before thinning. As a result, the generated model has high prediction accuracy near the side 72 where the average value Si is small. Further, the sides 71 and 73 are thinned out at the same number of measurement points as the number of measurement points, and the number of measurement points located on two sides is balanced, so that the influence of hysteresis is almost canceled. As a result, the generated model has improved prediction accuracy.

  Also, for example, S1 = 15, S2 = 10, and S3 = 30. Further, for example, the emphasis is set to a predetermined value a = 16. The average S1 'of the square of the difference between the average S1 of the measurement data of the sides 71 and 73 and the designated value a is obtained as in the following (12). The average S2 'of the square of the difference between the average S2 of the measurement data of the side 72 and the designated value a is obtained as in the following (13). The average S3 'of the square of the difference between the average S3 of the measurement data of the side 74 and the designated value a is obtained as in the following (14).

S1 ′ = (S1-16) ² = 1 (12)
S2 ′ = (S2-16) ² = 36 (13)
S3 ′ = (S3-16) ² = 16 (14)

  The minimum value P of the average Si 'of each group is set to 1, and the maximum value Q of the average Si' of each group is set to 36. The thinning width is set to 2 to 10 (k = 2, m = 10).

  In the case of an optimization problem that emphasizes the vicinity of the predetermined value a = 16, the thinning width CB1 of the group of the side 71 and the side 73 is obtained from the above equation (11) as follows.

CB1 = floor (((S1′−P) × (mk)) / (Q−P)) + k
= Floor (((1-1) × (10-2)) / (36-1)) + 2
= Floor (0) +2
= 2

  Similarly, the thinning width CB2 of the group of the side 72 is obtained from the above equation (11) as follows.

CB2 = floor (((S2′−P) × (mk)) / (Q−P)) + k
= Floor (((36-1) x (10-2)) / (36-1)) + 2
= Floor (8) + 2
= 10

  Similarly, the thinning width CB3 of the group of the side 74 is obtained from the above equation (11) as follows.

CB3 = floor (((S3′-P) × (mk)) / (QP)) + k
= Floor (((16-1) x (10-2)) / (36-1)) + 2
= Floor (1.9) + 2
= 3

  For the sides 71 and 73, the acquisition unit 43 reads the measurement data of the measurement points every two measurement points from the measurement data 31. The acquisition unit 43 reads the measurement data of the measurement points of the side 72 from the measurement data 31 at every ten measurement points. The acquisition unit 43 reads the measurement data of the measurement points at every third measurement point from the measurement data 31 for the side 74.

  The model generation unit 45 generates a model by performing machine learning using the measurement data of the measurement points read by the acquisition unit 43.

  As a result of the thinning, the number of measurement points used for generating the model on the side 71 and the side 73 becomes か ら of the number before the thinning. On the side 72, the number of measurement points used for generation of the model is 1/10 from that before thinning. On the side 74, the number of measurement points used for generating a model is reduced to 1/3 from before the thinning. As a result, the generated model has high prediction accuracy near the sides 71 and 73 where the average Si is close to the predetermined value a (= 16). Further, the sides 71 and 73 are thinned out at the same number of measurement points as the number of measurement points, and the number of measurement points located on two sides is balanced, so that the influence of hysteresis is almost canceled. As a result, the generated model has improved prediction accuracy.

  Note that the acquisition unit 43 may acquire the measurement data such that the difference between the numbers of the measurement points located on the two sides near and in opposite directions of the parameter change is at most one or less. . For example, as a result of thinning out the measurement points on the two sides in the vicinity and in which the parameter change is in the opposite direction by the determined thinning width, the acquisition unit 43 determines that the number of the measurement points on one side is the measurement on the other side. If the number is larger than the number of points by two or more, the measurement points on one side are further thinned so that the difference in the number of measurement points is at most within one. The measurement points to be thinned out are preferably selected from the start portion or the end portion of the side. For example, the acquisition unit 43 reads the measurement data of the measurement point on one side from the measurement data 31 at the determined number of intervals of the thinning width. Then, when the number of read measurement points on one side becomes the read measurement point on the other side, the acquisition unit 43 ends the reading of the measurement data of the measurement points on one side. As a result, the number of measurement points located on the two sides is substantially equal, and the effect of hysteresis is almost cancelled, so that the generated model has improved prediction accuracy.

[Processing flow]
Next, a flow of a model generation process in which the information processing apparatus 10 according to the present embodiment generates a model will be described. FIG. 14 is a flowchart illustrating an example of a procedure of a model generation process according to the second embodiment. Since the model generation processing according to the second embodiment is partially the same as the model generation processing according to the first embodiment illustrated in FIG. 9, the same parts are denoted by the same reference numerals, and different parts are mainly described. I do.

  The acquisition unit 43 acquires measurement data for model generation from the measurement data 31. For example, the acquisition unit 43 divides the measurement path curve into groups for every two sides that are near and whose parameter changes are in opposite directions. In addition, the acquisition unit 43 divides the measurement path curve into groups each of which is not adjacent to a side whose parameter change is in the opposite direction. The acquisition unit 43 calculates the average of the measurement data of the measurement points on the side belonging to the group for each group. The acquisition unit 43 determines a thinning width according to the average for each group. Then, the obtaining unit 43 obtains measurement data of the measurement points from the measurement data 31 by thinning out the measurement points by the determined thinning width from the measurement points on the sides belonging to the group for each group.

  The model generation unit 45 generates a model by performing machine learning using the measurement data of the measurement points acquired in S20 (S21), and ends the process.

[effect]
As described above, the information processing apparatus 10 according to the present embodiment includes the measurement data of each measurement point sequentially measured along the measurement path curve, where the measurement point is a specific area of the measurement target range. For each of the two sides that are larger than the area other than the area and that are in the vicinity of the measurement path curve and in which the control parameter changes in opposite directions, the number of measurement points located on the two sides is balanced. Obtain measurement data. The information processing device 10 generates a control model of a measurement target based on the acquired measurement data of each measurement point. Thereby, the information processing apparatus 10 can improve the prediction accuracy of the model.

  Further, in the information processing apparatus 10 according to the present embodiment, the difference in the number of measurement points located on the two sides in the vicinity of the measurement path curve and in which the change of the control parameter is opposite in the opposite direction is within 1 at most. To acquire the measurement data of each measurement point. Thereby, in the information processing apparatus 10, the number of measurement points located on two sides is substantially equal, and the effect of hysteresis is almost canceled, so that the prediction accuracy of the generated model is improved.

  In addition, the information processing apparatus 10 according to the present embodiment, from the measurement data sequentially measured by changing the control parameter along the measurement path curve at a speed that can be regarded as steady, from a region other than the specific region for a specific region Also extract measurement data at short intervals. Thereby, the information processing apparatus 10 can generate a control model with high prediction accuracy for a specific area.

  Further, the information processing apparatus 10 according to the present embodiment acquires measurement data of more measurement points as the measurement data is smaller in the case of the optimization problem of minimization, and acquires the measurement data in the case of the optimization problem of maximization. Measurement data of more measurement points is acquired as the size is larger. Thereby, the information processing apparatus 10 can generate a control model that can predict the minimum of the measurement data with high accuracy in the case of the optimization problem of minimization and the maximum of the measurement data in the case of the optimization problem of maximization.

  In addition, in the case of an optimization problem that emphasizes the vicinity of a predetermined value, the information processing apparatus 10 according to the present embodiment acquires measurement data of more measurement points as the difference between the measurement data and the predetermined value is smaller. Thereby, the information processing apparatus 10 can generate a control model that can predict the vicinity of the predetermined value of the measurement data with high accuracy in the case of an optimization problem that emphasizes the vicinity of the predetermined value.

  Next, a third embodiment will be described. In the third embodiment, a case will be described in which measurement data is acquired at a higher density in a specific region than in a region other than the specific region by changing the measurement interval during quasi-stationary measurement. Since the configurations of the system 1 and the information processing apparatus 10 according to the third embodiment are substantially the same as those of the system 1 and the information processing apparatus 10 according to the first embodiment illustrated in FIGS. 1 and 2, mainly different portions will be described. .

  The output unit 42 causes the external I / F unit 20 to output an operation instruction specifying a measurement condition to a control device (not shown) that controls the engine 11. For example, the output unit 42 outputs an operation instruction for operating the engine 11 under the measurement conditions in the order of the order stored in the measurement path information 30. The parameters of the engine 11 are changed to the measurement conditions, and the engine 11 operates under the measurement conditions. Any parameter of the engine 11 changes at such a slow speed that it can be regarded as steady based on the operation instruction. That is, the output unit 42 performs quasi-stationary measurement of the engine 11.

  The acquisition unit 43 causes the external I / F unit 20 to output an instruction to measure the state of the engine 11 to a control device (not shown) that controls the engine 11. Upon receiving the measurement instruction, the control device measures the state of the engine 11 and outputs data indicating the measured state of the engine 11 as measurement data in association with the measurement time. That is, every time the control device receives the measurement instruction, the control device measures the state of the engine 11 and outputs measurement data. The acquisition unit 43 acquires measurement data output from the control device via the external I / F unit 20.

  The acquisition unit 43 adjusts the timing at which the measurement instruction is output during execution of the quasi-stationary measurement of the engine 11 by the output unit 42, and obtains measurement data at a higher density in a specific region than in a region other than the specific region. get.

  Here, the flow of acquisition of measurement data will be described in detail. For example, the acquisition unit 43 divides the measurement path curve into groups for every two sides that are near and whose parameter changes are in opposite directions. In addition, the acquisition unit 43 divides the measurement path curve into groups each of which is not adjacent to a side whose parameter change is in the opposite direction. In addition, the acquisition unit 43 acquires parameter settings used for control. For example, the acquiring unit 43 acquires the setting of the dead time whose input has been received by the receiving unit 40.

  In the third embodiment, since the quasi-stationary measurement is being performed, the minimum value p and the maximum value q of the average Si of each group and the minimum value P and the maximum value Q of the average Si ′ of each group are accurately determined. do not know. However, for example, the user can predict the minimum value p, the maximum value q, the minimum value P, and the maximum value Q based on similar experimental results and experiences in the past. Therefore, the receiving unit 40 displays an operation screen related to control design and receives inputs of the minimum value p, the maximum value q, the minimum value P, and the maximum value Q. The acquisition unit 43 acquires the settings of the minimum value p, the maximum value q, the minimum value P, and the maximum value Q that have been received by the reception unit 40.

  The output unit 42 outputs an operation instruction to operate the engine 11 under the measurement conditions in the order of the order stored in the measurement path information 30 and executes the quasi-stationary measurement of the engine 11. The acquisition unit 43 acquires measurement data while the output unit 42 is performing the quasi-stationary measurement of the engine 11.

  FIG. 15 is a diagram illustrating a flow of acquiring measurement data during quasi-stationary measurement. FIG. 15 shows a part of the measurement path curve 70. The output unit 42 outputs an operation instruction for operating the engine 11 under the measurement conditions along the measurement path curve 70, and executes the quasi-stationary measurement of the engine 11. When the parameter whose measurement condition changes in the quasi-stationary measurement is changed, the acquisition unit 43 determines whether the side being measured in which the parameter is changing is one side where the side whose parameter change is in the opposite direction is not in the vicinity. I do. The sides 72 and 74 in FIG. 15 are one side in which there is no side whose parameter change is in the opposite direction. When the side being measured is one side where the side whose parameter change is in the opposite direction is not in the vicinity, the acquisition unit 43 waits for the lapse of dead time after the changed parameter changes. When the dead time elapses, the acquisition unit 43 outputs a measurement instruction at a measurement cycle for a predetermined period, and acquires measurement data for the start portion 75 of the side being measured. This predetermined period may be set in advance, may be about two to three times the dead time, and may be set by the user. The acquisition unit 43 calculates the average of the measurement data obtained at the start portion 75 of the side being measured, and determines the thinning width according to the average. For example, in the case of the optimization problem of minimization, the acquisition unit 43 calculates the thinning width from the above equation (7). In the case of an optimization problem that emphasizes the vicinity of a predetermined value, the acquisition unit 43 calculates the average of the square of the difference between the average of the measurement data and the predetermined value. Then, the acquisition unit 43 calculates the thinning width from the above equation (11). The acquisition unit 43 outputs a measurement instruction for the remaining part of the side being measured at a timing according to the thinning width. For example, the acquisition unit 43 outputs a measurement instruction in a cycle obtained by multiplying the measurement cycle by the value of the thinning width for the remaining portion 76 of the side being measured, and acquires measurement data according to the thinning width. Thereby, for example, when the value of the thinning width is double, the measurement interval of the measurement point from which the measurement data is acquired is doubled. The storage unit 44 stores the measurement data acquired by the acquisition unit 43 in the measurement data 31 in association with the measurement conditions. Also. The obtaining unit 43 thins out the measurement data obtained at the start portion 75 of the side being measured with the thinning width, and obtains the measurement data of the measurement points at the measurement intervals corresponding to the thinning width. The storage unit 44 stores the measurement data acquired by the acquisition unit 43 in the measurement data 31 in association with the measurement conditions.

  On the other hand, if the side being measured is not one side where the side whose parameter change is in the opposite direction is not in the vicinity, the side being measured is one of the two sides near and whose parameter change is in the opposite direction. . The acquisition unit 43 determines whether the thinning width has been determined for another side of the same group as the side being measured. When the thinning width has not been determined for the other sides, the acquisition unit 43 determines the thinning width from the start portion 75 of the side being measured, as in the case of one side where the parameter change is not in the vicinity of the opposite side. Calculate and acquire measurement data according to the thinning width. The storage unit 44 stores the measurement data acquired by the acquisition unit 43 in the measurement data 31 in association with the measurement conditions.

  If the thinning width has been determined for another side, the acquisition unit 43 waits for the elapse of the dead time after the changing parameter changes. After that, the acquisition unit 43 outputs a measurement instruction at a timing corresponding to the thinning width determined for the other sides from the start-out portion 75 to the end of the side being measured, and obtains measurement data according to the thinning width. The storage unit 44 stores the measurement data acquired by the acquisition unit 43 in the measurement data 31 in association with the measurement conditions.

  Thus, in the third embodiment, when performing the quasi-stationary measurement, the measurement data is acquired at a higher density for a specific region than for a region other than the specific region.

  The model generation unit 45 generates a model by performing machine learning using the measurement condition stored in the measurement data 31 and the measurement data corresponding to the measurement condition for each type of the measurement data. As a result, the generated model has higher prediction accuracy in a specific region. In addition, the two sides near and in which the parameter changes are in opposite directions are thinned out with the same number of measurement points as the number of measurement points, and the number of measurement points located on the two sides is balanced, so that the influence of hysteresis is almost canceled. As a result, the generated model has improved prediction accuracy.

  Next, a flow of a model generation process in which the information processing apparatus 10 according to the present embodiment generates a model will be described. FIG. 16 is a flowchart illustrating an example of a procedure of a model generation process according to the third embodiment. Since the model generation processing according to the third embodiment is partially the same as the model generation processing according to the first embodiment illustrated in FIG. 9, the same parts are denoted by the same reference numerals, and mainly different parts will be described. I do.

  The output unit 42 sequentially outputs operation instructions for operating the engine 11 under the measurement conditions in the order of the order stored in the measurement path information 30, and starts the quasi-stationary measurement of the engine 11 (S30).

  The acquisition unit 43 determines whether or not the parameter whose measurement condition changes in the quasi-stationary measurement has changed (S31). If the changed parameter has not changed (No at S31), the process proceeds to S37 described later.

  On the other hand, when the parameter to be changed has changed (Yes at S31), the acquiring unit 43 determines whether the side being measured is one side having no parameter change in the opposite direction (S32). When the side being measured is one side in which the parameter change is not in the vicinity of the side in the opposite direction (Yes at S32), the acquiring unit 43 waits for the lapse of dead time after the changed parameter is changed (S33). The acquisition unit 43 outputs a measurement instruction at a measurement cycle for a predetermined period, and acquires measurement data for the start portion 75 of the side being measured (S34). The acquisition unit 43 calculates the average of the measurement data obtained at the start portion 75 of the side being measured, and determines the thinning width according to the average (S35). The obtaining unit 43 thins out the measurement data obtained at the start portion 75 of the side being measured at the determined thinning width, and obtains the measurement data of the measurement points at the measurement intervals corresponding to the thinning width (S36). The acquisition unit 43 outputs a measurement instruction at a timing according to the determined thinning width, and acquires measurement data according to the thinning width (S37). If the thinning width is not determined, the process is performed with the thinning width set to a predetermined initial value (for example, 1). The storage unit 44 stores the measurement data acquired in S36 and S37 in the measurement data 31 in association with the measurement conditions (S38).

  On the other hand, when the side under measurement is not one side where the side whose parameter change is in the opposite direction is not in the vicinity (No at S32), the acquiring unit 43 sets the thinning width in another side of the same group as the side under measurement. It is determined whether it has been determined (S39). If the thinning width has not been determined for another side (No at S39), the process proceeds to S33 described above.

  On the other hand, if the thinning width has already been determined for another side (S39: Yes), the acquisition unit 43 waits for the elapse of dead time after the changing parameter changes (S40). The acquisition unit 43 outputs a measurement instruction at a timing corresponding to the determined thinning width determined for the other side, and acquires measurement data according to the thinning width (S41). The storage unit 44 stores the measurement data acquired in S41 in the measurement data 31 in association with the measurement conditions (S42).

  The output unit 42 outputs all the operation instructions of the measurement conditions stored in the measurement path information 30 and determines whether or not the quasi-stationary measurement has been completed (S43). If the quasi-stationary measurement has not been completed (No at S43), the process proceeds to S31 described above.

  On the other hand, when the quasi-stationary measurement is completed (Yes at S43), the model generation unit 45 generates a model by performing machine learning using the measurement conditions stored in the measurement data 31 and the measurement data corresponding to the measurement conditions. (S44), the process ends.

[effect]
As described above, the information processing apparatus 10 according to the present embodiment changes the control parameter along the measurement path curve at a speed that can be regarded as steady, shortens the measurement interval for a specific area, and changes the area other than the specific area. For, the measurement data is acquired by increasing the measurement interval. Thereby, the information processing apparatus 10 can acquire measurement data at a high density for a specific region during the measurement in the quasi-stationary measurement. Further, in the information processing apparatus 10, since only the measurement data of the measurement points used for generating the model is stored in the storage unit 23, the storage area used for storing the measurement data can be reduced.

  [C] Third Embodiment Although the embodiments related to the disclosed apparatus have been described above, the disclosed technology may be implemented in various different forms other than the above-described embodiments. Therefore, another embodiment included in the present invention will be described below.

  For example, in the above-described embodiment, the case where the measurement target is the engine 11 is illustrated. However, it is not limited to these. For example, any measurement object may be used as long as it takes time for the state of the measurement object to stabilize to a steady state after changing the measurement conditions. For example, the measurement target may be an actuator, a plant that performs various types of production, a large-sized machine, or the like.

  Further, in the above-described embodiment, the case where a model is generated for each type of measurement data has been illustrated. However, it is not limited to these. For example, a model that predicts a plurality of types of measurement data by performing machine learning using the measurement conditions stored in the measurement data 31 and the measurement data in a steady state corresponding to the measurement conditions may be generated. For example, the model generation unit 45 may generate one model that predicts all types of measurement data.

  Each component of each device illustrated is a functional concept, and does not necessarily need to be physically configured as illustrated. In other words, the specific state of the distribution / integration of each device is not limited to the one shown in the figure, and all or a part thereof may be functionally or physically distributed / arbitrarily divided into arbitrary units according to various loads and usage conditions. Can be integrated and configured. For example, the processing units of the reception unit 40, the measurement condition generation unit 41, the output unit 42, the acquisition unit 43, the storage unit 44, and the model generation unit 45 may be appropriately integrated. Further, all or any part of each processing function performed by each processing unit can be realized by a CPU and a program analyzed and executed by the CPU, or can be realized as hardware by wired logic. .

[Model generation program]
The various processes described in the above embodiments can also be realized by executing a prepared program on a computer system such as a personal computer or a workstation. Therefore, an example of a computer system that executes a program having the same function as the above embodiment will be described below. First, a model generation program for controlling the driver's alert will be described. FIG. 17 is an explanatory diagram illustrating an example of a configuration of a computer that executes a model generation program.

  As shown in FIG. 17, the computer 400 has a CPU (Central Processing Unit) 410, a HDD (Hard Disk Drive) 420, and a RAM (Random Access Memory) 440. These units 400 to 440 are connected via a bus 500.

  The HDD 420 stores in advance a model generation program 420a that performs the same functions as the above-described reception unit 40, measurement condition generation unit 41, output unit 42, acquisition unit 43, storage unit 44, and model generation unit 45. Note that the model generation program 420a may be separated as appropriate.

  The HDD 420 stores various information. For example, the HDD 420 stores the OS and various data used for determining the order quantity.

  Then, the CPU 410 reads out the model generation program 420a from the HDD 420 and executes the same to execute the same operation as each processing unit of the embodiment. That is, the model generation program 420a performs the same operation as the reception unit 40, the measurement condition generation unit 41, the output unit 42, the acquisition unit 43, the storage unit 44, and the model generation unit 45.

  It is not always necessary to store the above-described model generation program 420a in the HDD 420 from the beginning.

  Further, for example, the model generation program 420a may be stored in a “portable physical medium” such as a flexible disk (FD), a CD-ROM, a DVD disk, a magneto-optical disk, or an IC card inserted into the computer 400. . Then, the computer 400 may read out the program from these and execute it.

  Further, the program is stored in “another computer (or server)” connected to the computer 400 via a public line, the Internet, a LAN, a WAN, or the like. Then, the computer 400 may read out the program from these and execute it.

  Regarding the embodiment including the above-described examples, the following supplementary notes are further disclosed.

(Supplementary Note 1) In a space defined by each of a plurality of control parameters relating to control of a measurement target, a change is sequentially performed in which only one of the plurality of control parameters is changed from the immediately preceding measurement point. By generating a measurement condition for performing a measurement covering the space of the measurement target range,
An output unit that sequentially outputs operation instructions of the plurality of measurement conditions generated by the generation unit,
An acquisition unit that acquires measurement data indicating a state of the measurement target in which the control parameter is changed to a measurement condition by an operation instruction output from the output unit,
An information processing apparatus comprising:

(Supplementary Note 2) The generation unit generates the measurement condition using a measurement path curve generated from a Hilbert curve arranged in a normalized space obtained by normalizing a measurement target range of each of the plurality of control parameters. The information processing apparatus according to claim 1, characterized in that:

(Supplementary Note 3) a receiving unit that receives a period until measurement data corresponding to the measurement condition is acquired;
A storage unit that stores the measurement data acquired by the acquisition unit in association with the measurement condition for the period of time received by the reception unit,
3. The information processing apparatus according to claim 1, further comprising:

(Supplementary Note 4) The information processing apparatus according to Supplementary Note 3, wherein the storage unit discards the measurement data for the period during a change in the control parameter whose measurement condition changes, after the control parameter changes.

(Supplementary Note 5) The acquisition unit is measurement data of each measurement point sequentially measured along the measurement path curve, wherein the measurement point is a specific region of the measurement target range and a region other than the specific region. For each of two sides near the measurement path curve and opposite in control parameter change, the number of measurement points located on the two sides is balanced. The information processing device according to claim 2, wherein the information processing device acquires data.

(Supplementary Note 6) The information processing apparatus according to Supplementary Note 5, wherein the acquisition unit acquires the measurement data of each measurement point assuming that the difference in the number of measurement points located on the two sides is at most one or less. .

(Supplementary Note 7) The acquisition unit may be configured to change the control parameter along the measurement path curve at a speed that can be regarded as a steady state and sequentially measure the measurement data, so that the specific area is more than the area other than the specific area. 7. The information processing device according to appendix 5 or 6, wherein the measurement data is extracted at short intervals.

(Supplementary Note 8) The acquisition unit changes the control parameter along the measurement path curve at a speed that can be regarded as steady, shortens the measurement interval for the specific area, and sets the measurement interval for the area other than the specific area. 7. The information processing apparatus according to appendix 5 or 6, wherein the measurement data is acquired after being lengthened.

(Supplementary Note 9) The acquisition unit acquires the measurement data of more measurement points as the measurement data is smaller in the case of the optimization problem of minimization, and increases as the measurement data is larger in the case of the optimization problem of maximization. 9. The information processing apparatus according to any one of supplementary notes 5 to 8, wherein measurement data of a measurement point is acquired.

(Supplementary Note 10) In the case of an optimization problem that emphasizes the vicinity of a predetermined value, the acquisition unit acquires measurement data of more measurement points as the difference between the measurement data and the predetermined value is smaller. 8. The information processing apparatus according to any one of 8.

(Appendix 11)
In a space defined by the respective measurement target ranges of the plurality of control parameters related to the control of the measurement target, by sequentially performing a change in which only one of the plurality of control parameters is changed from the immediately preceding measurement point, Generate measurement conditions for performing measurement covering the space of the measurement target range,
Output the operation instructions of the generated multiple measurement conditions in order,
A model generation program for executing a process of acquiring measurement data indicating a state of the measurement target in which the control parameter is changed to a measurement condition according to an output operation instruction.

(Supplementary Note 12) The computer
In a space defined by the respective measurement target ranges of the plurality of control parameters related to the control of the measurement target, by sequentially performing a change in which only one of the plurality of control parameters is changed from the immediately preceding measurement point, Generate measurement conditions for performing measurement covering the space of the measurement target range,
Output the operation instructions of the generated multiple measurement conditions in order,
A model generation method, comprising: executing a process of acquiring measurement data indicating a state of the measurement target in which the control parameter has been changed to a measurement condition according to an output operation instruction.

(Supplementary Note 13) A point on a space filling curve constructed on a computer and having each of a plurality of control parameters related to control of a measurement target as a variable is moved along the space filling curve.
For each movement of the point, to obtain measurement data indicating the state of the measurement target when controlling the measurement target based on each value of the plurality of control parameters corresponding to the point after the movement,
A model generation method, wherein a computer executes a process of generating the control model of the measurement target based on a plurality of acquired measurement data.

1 System 10 Information processing device 11 Engine 20 External I / F unit 21 Input unit 22 Display unit 23 Storage unit 24 Control unit 30 Measurement path information 31 Measurement data 40 Acceptance unit 41 Measurement condition generation unit 42 Output unit 43 Acquisition unit 44 Storage unit 45 Model Generator

Claims (8)

Measurement data of each measurement point sequentially measured along a measurement path curve generated from a Hilbert curve arranged in a normalized space obtained by normalizing a measurement target range of each of a plurality of control parameters related to control of a measurement target. Te, the density of the measuring points is higher than in a region other than the specific area at a particular region of the measurement target range and 2 on the measurement path curve change in the near neighbor and control parameters are opposite An acquisition unit that acquires measurement data of each measurement point such that the number of measurement points located on the two sides is balanced for each side;
A generation unit that generates a control model of the measurement target based on the measurement data of each acquired measurement point,
An information processing apparatus comprising: The information processing apparatus according to claim 1, wherein the acquisition unit acquires the measurement data of each measurement point assuming that the difference between the numbers of the measurement points located on the two sides is at most one or less. The acquisition unit is configured to measure the control parameter at a shorter interval than the area other than the specific area from the measurement data sequentially measured by changing the control parameter along the measurement path curve at a speed that can be regarded as steady. The information processing apparatus according to claim 1, wherein data is extracted. The acquisition unit changes the control parameter along the measurement path curve at a speed that can be regarded as steady, shortens the measurement interval for the specific region, and increases the measurement interval for the region other than the specific region for measurement. The information processing apparatus according to claim 1, wherein data is acquired. In the case of the optimization problem of minimization, the acquisition unit compares the values of the measurement data in the entire measurement data so that the ratio of the measurement data having a small value is greater than the ratio of the measurement data having a large value. In the case of an optimization problem of maximization, the value of the measurement data is compared with the entire measurement data, and the ratio of the measurement data having the larger value is larger than the ratio of the measurement data having the smaller value. The information processing apparatus according to any one of claims 1 to 4 , wherein the measurement data is acquired as follows. In the case of an optimization problem that emphasizes the vicinity of a predetermined value, the acquisition unit compares the difference among the entire difference of the measurement data with respect to the predetermined value, and determines that the ratio of the measurement data having the small difference is the measurement having the large difference. The information processing apparatus according to claim 1 , wherein the measurement data is acquired so as to be larger than a data ratio . On the computer,
Measurement data of each measurement point sequentially measured along a measurement path curve generated from a Hilbert curve arranged in a normalized space obtained by normalizing a measurement target range of each of a plurality of control parameters related to control of a measurement target. Te, the specific area of the density of the measuring points the measuring object range is higher than in the region other than the specific area, and 2 on the measurement path curve change in the near neighbor and control parameters are opposite For each side, obtain measurement data of each measurement point so as to balance the number of measurement points located on the two sides,
Based on the measurement data of each acquired measurement point, to generate a control model of the measurement target,
A model generation program for executing a process. Computer
Measurement data of each measurement point sequentially measured along a measurement path curve generated from a Hilbert curve arranged in a normalized space obtained by normalizing a measurement target range of each of a plurality of control parameters related to control of a measurement target. Te, the specific area of the density of the measuring points the measuring object range is higher than in the region other than the specific area, and 2 on the measurement path curve change in the near neighbor and control parameters are opposite For each side, obtain measurement data of each measurement point so as to balance the number of measurement points located on the two sides,
Based on the measurement data of each acquired measurement point, to generate a control model of the measurement target,
A model generation method, characterized by performing a process.