Control model generation device and control model generation method

Abstract

This control model generation device is configured to comprise: an observation value acquisition unit (1) that acquires a plurality of observation values that are outputs of a controlled object having nonlinear characteristics; a state space model estimation unit (2) that estimates a state space model representing a linear approximation curve for the plurality of observation values acquired by the observation value acquisition unit (1); and an upper-bound model estimation unit (3) that calculates estimation errors, which are errors between the observation values acquired by the observation value acquisition unit (1) and the linear approximation curve represented by the state space model estimated by the state space model estimation unit (2), and estimates an upper-bound model representing the upper bounds of the estimation errors. The control model generation device also comprises a control model generation unit (4) that uses the state space model estimated by the state space model estimation unit (2) and the upper-bound model estimated by the upper-bound model estimation unit (3) to generate a control model representing an equation of motion of the controlled object.

Description

Control model generating device and control model generating method

[0001] The present disclosure relates to a control model generation device and a control model generation method.

[0002] There is a control model generation device that generates a control model that represents an equation of motion of a controlled object having nonlinear characteristics. For example, Patent Document 1 discloses an example of such a control model generation device, which is a control device that includes a plurality of mutually different control model candidates. Each control model candidate is prepared in advance. The control device calculates an error between the output of each control model candidate and the output of the controlled object, and selects one of the control model candidates from the plurality of control model candidates based on the error calculation result.

[0003] Japanese Patent Application Publication No. 5-303408

[0004] Unless a person preparing control model candidates has sufficient knowledge about the behavior of the controlled object, it is generally difficult to prepare control model candidates that represent the equation of motion of the controlled object. The control device disclosed in Patent Document 1 has a problem in that multiple control model candidates must be prepared in advance. For example, if none of the multiple control model candidates represent the equation of motion of the controlled object, the control device may not be able to control the controlled object with high accuracy even if it selects one of the multiple control model candidates.

[0005] The present disclosure has been made to solve the above-mentioned problems, and aims to provide a control model generation device that enables a person with sufficient knowledge of the behavior of a controlled object to generate a control model that represents the equation of motion of the controlled object, without having to prepare multiple control model candidates in advance.

[0006] A control model generation device according to the present disclosure includes an observation value acquisition unit that acquires multiple observation values ​​that are outputs of a controlled object having nonlinear characteristics, a state-space model estimation unit that estimates a state-space model that represents a linear approximation curve related to the multiple observation values ​​acquired by the observation value acquisition unit, and an upper-bound model estimation unit that calculates an estimation error that is the error between each observation value acquired by the observation value acquisition unit and the linear approximation curve represented by the state-space model estimated by the state-space model estimation unit, and estimates an upper-bound model that represents an upper bound of the estimation error.The control model generation device also includes a control model generation unit that generates a control model that represents an equation of motion of the controlled object using the state-space model estimated by the state-space model estimation unit and the upper-bound model estimated by the upper-bound model estimation unit.

[0007] According to the present disclosure, a person with sufficient knowledge of the behavior of a controlled object can generate a control model that represents the equation of motion of the controlled object without having to prepare multiple control model candidates in advance.

[0008] 5A is a graph showing an example of a plurality of observation values ​​f(z), which are outputs of a control object OB, the state space model Jz, and the upper-bound model Hz. FIG. 5B is a graph showing the graph of FIG. 5A rotated so that the state space model Jz is on the horizontal axis. FIG. 5B is a flowchart showing a control model generation method, which is a processing procedure of the control model generation device. FIG. 5A is a hardware configuration diagram showing the hardware of the control model generation device according to the second embodiment. FIG. 9A and FIG. 9B are explanatory diagrams showing a model selection unit 9 that selects one of M controllers 31-1 to 31-M, namely, a controller 31-m. FIG. 9B shows M partial state spaces JK included in a state space JK in which a plurality of observation values ​​f(z) exist. １ ~JKＭ FIG.

[0009] In order to explain the present disclosure in more detail, embodiments of the present disclosure will be described below with reference to the accompanying drawings.

[0010] Embodiment 1. Fig. 1 is a configuration diagram showing a control model generation device according to embodiment 1. Fig. 2 is a hardware configuration diagram showing hardware of the control model generation device according to embodiment 1. The control model generation device shown in Fig. 1 includes an observation value acquisition unit 1, a state space model estimation unit 2, an upper-bound model estimation unit 3, and a control model generation unit 4.

[0011] The observation value acquisition unit 1 is realized, for example, by an observation value acquisition circuit 11 shown in FIG. 2 . The observation value acquisition unit 1 acquires a plurality of observation values ​​f(z), which are outputs of a control object OB having nonlinear characteristics. The control object OB may be a known control object or an unknown control object. The observation value acquisition unit 1 outputs the plurality of observation values ​​f(z) to each of the state space model estimation unit 2 and the upper-bound model estimation unit 3.

[0012] The state space model estimation unit 2 is realized by, for example, a state space model estimation circuit 12 shown in Fig. 2. The state space model estimation unit 2 acquires a plurality of observation values ​​f(z) from the observation value acquisition unit 1. The state space model estimation unit 2 estimates a state space model Jz that represents a linear approximation curve relating to the plurality of observation values ​​f(z). The state space model estimation unit 2 outputs the state space model Jz to each of the upper-bound model estimation unit 3 and the control model generation unit 4.

[0013] The upper-bound model estimation unit 3 is realized, for example, by an upper-bound model estimation circuit 13 shown in FIG. 2 . The upper-bound model estimation unit 3 acquires multiple observed values ​​f(z) from the observed value acquisition unit 1 and acquires a state space model Jz from the state space model estimation unit 2. The upper-bound model estimation unit 3 calculates an estimation error |f(z)-Jz|, which is the error between each observed value f(z) and a linear approximation curve represented by the state space model Jz. The upper-bound model estimation unit 3 estimates an upper-bound model Hz that represents an upper bound of the estimation error |f(z)-Jz|. Specifically, the upper-bound model estimation unit 3 estimates the upper-bound model Hz using a loss function ζ(t) that indicates the difference between the squared value of the upper bound and the squared value of the estimation error |f(z)-Jz|, obtained by multiplying the squared value of the upper bound by a constant α that is equal to or less than 1. The upper-bound model estimation unit 3 outputs the upper-bound model Hz to the control model generation unit 4.

[0014] The control model generation unit 4 is realized by, for example, the control model generation circuit 14 shown in FIG. 2. The control model generation unit 4 acquires the state space model Jz from the state space model estimation unit 2 and acquires the upper bound model Hz from the upper bound model estimation unit 3. The control model generation unit 4 generates a control model CM that represents an equation of motion of the control object OB using the state space model Jz and the upper bound model Hz. Specifically, the control model generation unit 4 generates, as the control model CM, a control model in which the error between the output of the control model and the linear approximation curve is equal to or less than an upper bound. The control model CM generated by the control model generation unit 4 is implemented in a controller that controls the control object OB.

[0015] 1, it is assumed that each of the components of the control model generation device, namely, an observation value acquisition unit 1, a state space model estimation unit 2, an upper-bound model estimation unit 3, and a control model generation unit 4, is realized by dedicated hardware as shown in Fig. 2. That is, it is assumed that the control model generation device is realized by an observation value acquisition circuit 11, a state space model estimation circuit 12, an upper-bound model estimation circuit 13, and a control model generation circuit 14. Each of the observation value acquisition circuit 11, the state space model estimation circuit 12, the upper-bound model estimation circuit 13, and the control model generation circuit 14 corresponds to, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an ASIC (Application Specific Integrated Circuit), an FPGA (Field-Programmable Gate Array), or a combination thereof.

[0016] The components of the control model generation device are not limited to those realized by dedicated hardware, and the control model generation device may be realized by software, firmware, or a combination of software and firmware. Software or firmware is stored as a program in the memory of a computer. The computer refers to hardware that executes a program, and includes, for example, a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), a central processing unit, a processing unit, an arithmetic unit, a microprocessor, a microcomputer, a processor, or a DSP (Digital Signal Processor).

[0017] 3 is a hardware configuration diagram of a computer when the control model generation device is realized by software, firmware, etc. When the control model generation device is realized by software, firmware, etc., a program for causing a computer to execute the respective processing procedures of the observation value acquisition unit 1, the state space model estimation unit 2, the upper-bound model estimation unit 3, and the control model generation unit 4 is stored in a memory 21. A processor 22 of the computer then executes the program stored in the memory 21.

[0018] 2 shows an example in which each of the components of the control model generation device is realized by dedicated hardware, while Fig. 3 shows an example in which the control model generation device is realized by software, firmware, etc. However, this is merely an example, and some of the components in the control model generation device may be realized by dedicated hardware, and the remaining components may be realized by software, firmware, etc.

[0019] Fig. 4 is an explanatory diagram showing the relationship between a controller that implements the control model CM and a linear system that includes both the state space model Jz and the upper bound model Hz. Fig. 4 shows that the output x of the state space model Jz is provided to the controller, and the output u of the controller is provided to the linear system. Fig. 4 also shows that the output p of the upper bound model Hz is provided to the disturbance, and the disturbance q is provided to the linear system.

[0020] 5A is a graph showing an example of a plurality of observed values ​​f(z), which are the output of the control object OB, a state space model Jz, and an upper bound model Hz. FIG. 5A shows an example of a two-dimensional space represented by the z-axis and the x-axis, in which the state space containing the plurality of observed values ​​f(z), which are the output of the control object OB, exists. However, this is merely an example, and the state space may be, for example, a three-dimensional space. FIG. 5B is a graph obtained by rotating the graph of FIG. 5A so that the state space model Jz is on the horizontal axis.

[0021] The equation of motion xdot of the controlled object OB is expressed as the following equation (1): In the specification document, due to electronic filing, the symbol "." cannot be placed above the letter x, so it is written as xdot.

[0022] In equation (2), T is a mathematical symbol indicating transposition.

[0023] The relationship between the multiple observed values ​​f(z) that are the outputs of the control object OB, the state space model Jz, and the upper bound model Hz is expressed by the following equation (3).

[0024]

[0025] The state space model Jz is expressed as in the following equation (4): Furthermore, the estimation error q, which is the error between the observed value f(z) and the linear approximation curve represented by the state space model Jz, is expressed as in the following equation (5): The upper bound model Hz is expressed as in the following equation (6):

[0026] In equations (4) and (6), A, B, C, and D are arbitrary matrices.

[0027] Loss function L of state space model Jz Ｊ is expressed as the following equation (7). In addition, the loss function L of the upper bound model Hz Ｈ is expressed as the following equation (8).

[0028] In equation (9), ζ(t) is a loss function that indicates the difference between the square of the upper bound p(t) multiplied by a constant α and the square of the estimation error q(t), where α is a constant equal to or less than 1.

[0029] Next, a description will be given of the operation of the control model generating device shown in Fig. 1. Fig. 6 is a flowchart showing a control model generating method, which is a processing procedure of the control model generating device.

[0030] The observation value acquisition unit 1 acquires from the outside a plurality of observation values ​​f(z), which are outputs of a control object OB having nonlinear characteristics (step ST1 in FIG. 6 ). The observation value acquisition unit 1 outputs the plurality of observation values ​​f(z) to the state space model estimation unit 2 and the upper-bound model estimation unit 3, respectively.

[0031] The state space model estimator 2 acquires multiple observed values ​​f(z) from the observed value acquirer 1. As shown in FIGS. 5A and 5B , the state space model estimator 2 estimates a state space model Jz representing a linear approximation curve related to the multiple observed values ​​f(z) (step ST2 in FIG. 6 ). The state space model Jz representing the linear approximation curve related to the multiple observed values ​​f(z) is expressed as shown in Equation (4). The process of estimating the state space model Jz itself is a well-known technique, and detailed description thereof will be omitted. For example, the state space model Jz can be estimated by using a method of searching for a linear approximation curve that minimizes the expected value of an error vector representing the error between the multiple observed values ​​f(z) and the linear approximation curve. The state space model estimator 2 outputs the state space model Jz to each of the upper-bound model estimator 3 and the control model generator 4.

[0032] The upper-bound model estimation unit 3 acquires multiple observed values ​​f(z) from the observed value acquisition unit 1 and acquires the state space model Jz from the state space model estimation unit 2. The upper-bound model estimation unit 3 calculates an estimation error q, which is the error between each observed value f(z) and the linear approximation curve represented by the state space model Jz, as shown in equation (5) (step ST3 in FIG. 6 ). The upper-bound model estimation unit 3 estimates an upper-bound model Hz representing the upper bound of the estimation error q, as shown in equation (6) (step ST4 in FIG. 6 ). Specifically, the upper-bound model estimation unit 3 estimates the upper-bound model Hz using a loss function ζ(t) representing the difference between the squared value of the upper bound p multiplied by a constant α equal to or less than 1 and the squared value of the estimation error q, as shown in equation (9). The loss function ζ(t) is a function that outputs a vector that suppresses the norm of the error vector representing the difference from above. The upper-bound model estimation unit 3 outputs the upper-bound model Hz to the control model generation unit 4.

[0033] The control model generation unit 4 acquires the state space model Jz from the state space model estimation unit 2 and the upper bound model Hz from the upper bound model estimation unit 3. The control model generation unit 4 generates a control model CM representing the equation of motion of the control object OB using the state space model Jz and the upper bound model Hz, as shown in equation (3) (step ST5 in FIG. 6 ). f(z) satisfying equation (3) corresponds to the control model CM representing the equation of motion of the control object OB. The control model CM generated by the control model generation unit 4 is implemented in a controller that controls the control object OB. This constructs a controller in which the error between the output of the control model CM and the output of the state space model Jz is regarded as an error input to the control object OB. The upper bound of the error between the output of the control model CM and the output of the state space model Jz is the upper bound represented by the upper bound model Hz.

[0034] In the above-described first embodiment, the control model generation device is configured to include an observation value acquisition unit 1 that acquires multiple observation values ​​that are outputs of a controlled object having nonlinear characteristics, a state space model estimation unit 2 that estimates a state space model that represents a linear approximation curve related to the multiple observation values ​​acquired by the observation value acquisition unit 1, and an upper-bound model estimation unit 3 that calculates an estimation error that is the error between each observation value acquired by the observation value acquisition unit 1 and the linear approximation curve represented by the state space model estimated by the state space model estimation unit 2, and estimates an upper-bound model that represents an upper bound of the estimation error. The control model generation device also includes a control model generation unit 4 that generates a control model that represents an equation of motion of the controlled object using the state space model estimated by the state space model estimation unit 2 and the upper-bound model estimated by the upper-bound model estimation unit 3. Therefore, the control model generation device allows a person with sufficient knowledge of the operation of the controlled object to generate a control model that represents an equation of motion of the controlled object without having to prepare multiple control model candidates in advance.

[0035] Second Embodiment In a second embodiment, a control model generation device will be described that includes a state space division unit 5 that divides a state space in which a plurality of observed values ​​f(z) exist into a plurality of partial state spaces.

[0036] Fig. 7 is a configuration diagram showing a control model generation device according to embodiment 2. In Fig. 7, the same reference numerals as in Fig. 1 indicate the same or corresponding parts, and detailed description thereof will be omitted. Fig. 8 is a hardware configuration diagram showing hardware of the control model generation device according to embodiment 2. In Fig. 8, the same reference numerals as in Fig. 2 indicate the same or corresponding parts, and detailed description thereof will be omitted. The control model generation device shown in Fig. 7 includes an observation value acquisition unit 1, a state space division unit 5, a state space model estimation unit 6, an upper-bound model estimation unit 7, a control model generation unit 8, and a model selection unit 9.

[0037] The state space division unit 5 is realized by, for example, a state space division circuit 15 shown in Fig. 8. The state space division unit 5 acquires a plurality of observed values ​​f(z) from the observed value acquisition unit 1. The state space division unit 5 divides the state space JK in which the plurality of observed values ​​f(z) exist into a plurality of partial state spaces JK １ ~JK Ｍ M is an integer of 2 or more.

[0038] The state space model estimation unit 6 is realized by, for example, a state space model estimation circuit 16 shown in Fig. 8. The state space model estimation unit 6 acquires a plurality of observation values ​​f(z) from the observation value acquisition unit 1. The state space model estimation unit 6 estimates each partial state space JK after division by the state space division unit 5 from the plurality of observation values ​​f(z). ｍ (m = 1, ..., M) ｍ A partial state space model Jz, which is a state space model representing a linear approximation curve related to (z) ｍ The state space model estimation unit 6 estimates each partial state space model Jz ｍ are output to the upper-bound model estimation unit 7 and the control model generation unit 8, respectively.

[0039] The upper-bound model estimating unit 7 is realized by, for example, an upper-bound model estimating circuit 17 shown in Fig. 8. The upper-bound model estimating unit 7 acquires a plurality of observed values ​​f(z) from the observed value acquiring unit 1, and estimates the respective partial state space models Jz from the state space model estimating unit 6. ｍ (m=1, . . . , M). The upper bound model estimation unit 7 obtains each partial state space JKｍ The observations f that exist in ｍ (z) and each partial state space model Jz ｍ The estimated error q is the error from the linear approximation curve represented by ｍ The upper bound model estimation unit 7 calculates the respective estimation errors q ｍ The partial upper bound model Hz is an upper bound model that represents the upper bound of ｍ (m=1, . . . , M). The upper bound model estimation unit 7 estimates each partial upper bound model Hz ｍ is output to the control model generation unit 8.

[0040] The control model generating unit 8 is realized by, for example, a control model generating circuit 18 shown in FIG. 8. The control model generating unit 8 receives the partial state space model Jz from the state space model estimating unit 6. ｍ (m=1, . . . , M) is obtained, and the partial upper bound model Hz ｍ (m=1, . . . , M). The control model generation unit 8 obtains the partial state space model Jz ｍ and partial upper bound model Hz ｍ Using the above, the partial state space JK ｍ Control model CM corresponding to ｍ (m=1, . . . , M) is generated. The control model CM generated by the control model generation unit 8 ｍ are implemented in controllers 31-m (m=1, . . . , M) described later.

[0041] The model selection unit 9 is realized by, for example, a model selection circuit 19 shown in FIG. 8. The model selection unit 9 selects M partial state spaces JK １ ~JK Ｍ The control model CM corresponding to each of the above ｍ (m=1, . . . , M) to select one control model. Specifically, the model selection unit 9 selects one control model from M partial state spaces JK １ ~JK Ｍ The control model CM corresponding to each of the above ｍ and the output of the partial state space model Jz ｍ The error Δe from the linear approximation curve ｍ(m=1, . . . , M). The model selection unit 9 calculates M errors Δe １ ~Δe Ｍ Based on the calculation result of (a), one of the control models is selected.

[0042] 7, it is assumed that each of the components of the control model generation device, namely, the observation value acquisition unit 1, the state space division unit 5, the state space model estimation unit 6, the upper-bound model estimation unit 7, the control model generation unit 8, and the model selection unit 9, is realized by dedicated hardware as shown in Fig. 8. That is, it is assumed that the control model generation device is realized by an observation value acquisition circuit 11, a state space division circuit 15, a state space model estimation circuit 16, an upper-bound model estimation circuit 17, a control model generation circuit 18, and a model selection circuit 19. Each of the observation value acquisition circuit 11, the state space division circuit 15, the state space model estimation circuit 16, the upper-bound model estimation circuit 17, the control model generation circuit 18, and the model selection circuit 19 corresponds to, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an ASIC, an FPGA, or a combination thereof.

[0043] The components of the control model generation device are not limited to those realized by dedicated hardware, and the control model generation device may be realized by software, firmware, or a combination of software and firmware. When the control model generation device is realized by software, firmware, or the like, a program for causing a computer to execute the respective processing procedures of the observation value acquisition unit 1, the state space division unit 5, the state space model estimation unit 6, the upper-bound model estimation unit 7, the control model generation unit 8, and the model selection unit 9 is stored in a memory 21 shown in Fig. 3. Then, a processor 22 shown in Fig. 3 executes the program stored in the memory 21.

[0044] 8 shows an example in which each of the components of the control model generation device is realized by dedicated hardware, while Fig. 3 shows an example in which the control model generation device is realized by software, firmware, etc. However, this is merely an example, and some of the components in the control model generation device may be realized by dedicated hardware, and the remaining components may be realized by software, firmware, etc.

[0045] 9A and 9B are explanatory diagrams showing a model selection unit 9 that selects one of the M controllers 31-1 to 31-M, 31-m. Each of the controllers 31-1 to 31-M controls a control object OB. The controller 31-m (m=1, ..., M) is a controller that selects one of the M controllers 31-1 to 31-M. ｍ FIG. 10 shows M partial state spaces JK included in a state space JK where multiple observation values ​​f(z) exist. １ ~JK Ｍ 10 shows an example where M=3.

[0046] Next, the operation of the control model generation device shown in Fig. 7 will be described. The observation value acquisition unit 1 acquires from the outside a plurality of observation values ​​f(z), which are outputs of a control object OB having nonlinear characteristics. The observation value acquisition unit 1 outputs the plurality of observation values ​​f(z) to the state space division unit 5, the state space model estimation unit 6, and the upper-bound model estimation unit 7, respectively.

[0047] The state space division unit 5 acquires a plurality of observed values ​​f(z) from the observed value acquisition unit 1. As shown in FIG. 10 , the state space division unit 5 divides the state space JK in which the plurality of observed values ​​f(z) exist into a plurality of partial state spaces JK １ ~JK Ｍ Divide into.

[0048] The state space model estimation unit 6 acquires a plurality of observation values ​​f(z) from the observation value acquisition unit 1. The state space model estimation unit 6 estimates the partial state space JK ｍ (m = 1, ..., M) ｍThe state space model estimation unit 6 identifies the partial state space JK ｍ The observations f that exist in ｍ A partial state space model Jz, which is a state space model representing a linear approximation curve related to (z) ｍ The partial state space model Jz is estimated. ｍ is expressed as the following equation (10): ｍ The process of estimating the partial state space model Jz is a well-known technique, and therefore a detailed description thereof will be omitted. ｍ (m=1, . . . , M) is output to the upper-bound model estimation unit 7 and the control model generation unit 8, respectively.

[0049] In formula (10), A ｍ , B ｍ is an arbitrary matrix.

[0050] The upper bound model estimator 7 acquires a plurality of observed values ​​f(z) from the observed value acquirer 1, and estimates each partial state space model Jz from the state space model estimator 6. ｍ (m=1, . . . , M). The upper bound model estimation unit 7 obtains each partial state space JK ｍ The observations f that exist in ｍ (z) and each partial state space model Jz ｍ The estimated error q is the error from the linear approximation curve represented by ｍ Calculate the estimated error q ｍ The calculation process of each estimation error q ｍ The partial upper bound model Hz is an upper bound model that represents the upper bound of ｍ (m = 1, ..., M) is estimated. Partial upper bound model Hz ｍ The estimation process for (m=1, . . . , M) is performed according to the following equation (12). ｍ (m=1, . . . , M) is output to the control model generating unit 8.

[0051] In formula (12), C ｍ , D ｍ is an arbitrary matrix.

[0052] The control model generation unit 8 receives the partial state space model Jz from the state space model estimation unit 6. ｍ (m=1, . . . , M) is obtained, and the partial upper bound model Hz ｍ (m=1, . . . , M). The control model generation unit 8 obtains the partial state space model Jz ｍ and partial upper bound model Hz ｍ Using the above, the partial state space JK ｍ Control model CM corresponding to ｍ Generate f that satisfies equation (13). ｍ (z) is the partial state space JK ｍ A control model CM representing the equation of motion of the control object OB corresponding to ｍ The control model CM generated by the control model generating unit 8 corresponds to ｍ is implemented in the controller 31-m (m=1, . . . , M).

[0053]

[0054] The model selection unit 9 selects M partial state spaces JK １ ~JK Ｍ The control model CM corresponding to each of the above ｍ (m=1, . . . , M) ｍ Specifically, the model selection unit 9 selects M partial state spaces JK １ ~JK Ｍ The control model CM corresponding to each of the above ｍ Output u ｍ and the partial state space model Jz ｍ The error Δe from the linear approximation curve ｍ (m=1, . . . , M) is calculated.

[0055]

[0056] The model selection unit 9 selects M errors Δe １ ~Δe Ｍ Based on the calculation results, M control models CM １ ~CM ＭSelect one of the control models CM ｍ Specifically, the model selection unit 9 selects the error Δe １ ~Δe Ｍ Among them, the smallest error Δe ＭＩＮ The model selection unit 9 identifies M control models CM １ ~CM Ｍ Among these, the minimum error Δe ＭＩＮ Control model CM corresponding to ｍ The control model CM selected by the model selection unit 9 from among the M controllers 31-1 to 31-M is selected. ｍ The controller 31-m that implements the above controls the control object OB.

[0057] In the control model generating device shown in FIG. 7, a model selection unit 9 selects M control models CM １ ~CM Ｍ Among these, the minimum error Δe ＭＩＮ Control model CM corresponding to ｍ However, this is only an example, and the model selection unit 9 selects the smallest error Δe within a range that does not cause any practical problems. ＭＩＮ Control model CM corresponding to errors other than ｍ Specifically, the model selection unit 9 may select the control model CM having the second smallest error. ｍ Or, select the control model CM with the third smallest error. ｍ may be selected.

[0058] In the second embodiment described above, the control model generation device shown in Fig. 7 is configured to include a model selection unit 9 that selects one control model from among the control models corresponding to each of a plurality of partial state spaces, which are generated by the control model generation unit 8. Therefore, similar to the control model generation device shown in Fig. 1, the control model generation device shown in Fig. 7 allows a person with sufficient knowledge about the operation of the controlled object to generate a control model that represents an equation of motion of the controlled object without having to prepare a plurality of control model candidates in advance, and can also improve the control accuracy of the controller compared to the control model generation device shown in Fig. 1.

[0059] In the control model generation device shown in FIG. 7, the model selection unit 9 selects M partial state spaces JK １ ~JK Ｍ The control model CM corresponding to each of the above ｍ and the output of the partial state space model Jz ｍ The error Δe from the linear approximation curve ｍ (m=1, . . . , M) and calculate M errors Δe １ ~Δe Ｍ However, this is merely an example, and the model selection unit 9 selects one of the control models based on the calculation results of the M partial upper bound models Hz estimated by the upper bound model estimation unit 7. １ ~Hz Ｍ Among them, the partial upper bound model Hz with the smallest upper bound slope, which is the gain of the partial upper bound model, is ｍ M control models CM １ ~CM Ｍ The partial upper bound model Hz ｍ Control model CM corresponding to ｍ may be selected.

[0060] In addition, the present disclosure allows for free combination of the respective embodiments, modification of any of the components of the respective embodiments, or omission of any of the components of the respective embodiments.

[0061] The present disclosure is suitable for a control model generation device and a control model generation method.

[0062] 1 Observation value acquisition unit, 2 State space model estimation unit, 3 Upper bound model estimation unit, 4 Control model generation unit, 5 State space division unit, 6 State space model estimation unit, 7 Upper bound model estimation unit, 8 Control model generation unit, 9 Model selection unit, 11 Observation value acquisition circuit, 12 State space model estimation circuit, 13 Upper bound model estimation circuit, 14 Control model generation circuit, 15 State space division circuit, 16 State space model estimation circuit, 17 Upper bound model estimation circuit, 18 Control model generation circuit, 19 Model selection circuit, 21 Memory, 22 Processor, 31-1 to 31-M Controllers.

Claims

1. A control model generation device comprising: an observation value acquisition unit that acquires multiple observation values ​​which are outputs of a controlled object having nonlinear characteristics; a state space model estimation unit that estimates a state space model which represents a linear approximation curve relating to the multiple observation values ​​acquired by the observation value acquisition unit; an upper bound model estimation unit that calculates an estimation error which is the error between each observation value acquired by the observation value acquisition unit and the linear approximation curve represented by the state space model estimated by the state space model estimation unit, and estimates an upper bound model which represents an upper bound of the estimation error; and a control model generation unit that generates a control model which represents an equation of motion of the controlled object using the state space model estimated by the state space model estimation unit and the upper bound model estimated by the upper bound model estimation unit.

2. The control model generating device according to claim 1, characterized in that the upper bound model estimation unit estimates the upper bound model using a loss function indicating the difference between the result of multiplying the squared value of the upper bound by a constant not greater than 1 and the squared value of the estimated error.

3. The control model generating device according to claim 1, characterized in that the control model generating unit generates, as the control model, a control model in which the error between the output of the control model and the linear approximation curve is equal to or smaller than the upper bound.

4. A control model generation device as claimed in claim 1, further comprising a state space division unit which divides a state space in which a plurality of observation values ​​acquired by the observation value acquisition unit exist into a plurality of partial state spaces, the state space model estimation unit estimating a partial state space model which is a state space model representing a linear approximation curve relating to observation values ​​existing in each partial state space after division by the state space division unit among the plurality of observation values ​​acquired by the observation value acquisition unit, the upper bound model estimation unit calculating an estimation error which is the error between the observation values ​​existing in each partial state space and the linear approximation curve represented by each partial state space model, and estimating a partial upper bound model which is an upper bound model representing an upper bound of each estimation error, and the control model generation unit generating a control model which represents an equation of motion of the controlled object corresponding to each partial state space using each partial state space model estimated by the state space model estimation unit and each partial upper bound model estimated by the upper bound model estimation unit.

5. The control model generating device according to claim 4, further comprising a model selection unit which selects one of the control models generated by the control model generating unit and corresponding to each of the plurality of partial state spaces.

6. The control model generating device according to claim 5, characterized in that the model selection unit calculates an error between the output of a control model corresponding to each of the plurality of partial state spaces and the linear approximation curve, and selects one of the control models based on the calculation result of the error between the output of each control model and the linear approximation curve.

7. A control model generation method, in which an observation value acquisition unit acquires a plurality of observation values ​​which are outputs of a controlled object having nonlinear characteristics, a state space model estimation unit estimates a state space model which represents a linear approximation curve relating to the plurality of observation values ​​acquired by the observation value acquisition unit, an upper bound model estimation unit calculates an estimation error which is the error between each of the observation values ​​acquired by the observation value acquisition unit and the linear approximation curve represented by the state space model estimated by the state space model estimation unit, and estimates an upper bound model which represents an upper bound of the estimation error, and a control model generation unit generates a control model which represents an equation of motion of the controlled object using the state space model estimated by the state space model estimation unit and the upper bound model estimated by the upper bound model estimation unit.

Images