Control model generation device and control model generation method
By generating control models through observations and model inference, the problem of inaccurate control model selection caused by lack of knowledge preparation is solved, and high-precision control model generation is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- MITSUBISHI ELECTRIC CORP
- Filing Date
- 2023-09-25
- Publication Date
- 2026-04-17
AI Technical Summary
In existing technologies, it is difficult to generate the motion equations of the controlled object when there is insufficient knowledge preparation, which leads to inaccurate selection of the control model and affects the control accuracy.
By obtaining observations, inferring from the state-space model, and inferring from the upper bound model, a control model representing the motion equations of the controlled object is generated. The control model is then generated using the state-space model and the upper bound model, ensuring that the error between the model output and the linear approximation curve is within the upper bound.
In the absence of sufficient knowledge preparation, the motion equations of the controlled object can be generated with high precision, thereby improving the accuracy of the control model selection and the control precision.
Smart Images
Figure CN121889736A_ABST
Abstract
Description
Technical Field
[0001] This disclosure relates to a control model generation apparatus and a control model generation method. Background Technology
[0002] There exists a control model generation device that generates a control model that represents the motion equations of a control object with nonlinear characteristics.
[0003] As such a control model generation device, Patent Document 1 discloses a control device having multiple control model candidates that are different from each other. Each control model candidate is prepared in advance.
[0004] The control device calculates the error between the output of each control model candidate and the output of the controlled object, and selects any one of the multiple control model candidates based on the error calculation results.
[0005] Patent Document 1: Japanese Patent Application Publication No. 5-303408
[0006] If the person preparing control model candidates does not have sufficient knowledge related to the actions of the controlled object, it is usually difficult to prepare control model candidates that can represent the motion equations of the controlled object.
[0007] The control device disclosed in Patent Document 1 has the problem of needing to prepare multiple control model candidates in advance. For example, if none of the multiple control model candidates represents the motion equation of the controlled object, then even if the control device selects any one of the multiple control model candidates, it may sometimes be unable to control the controlled object with high precision. Summary of the Invention
[0008] This disclosure is made to solve the aforementioned problems, with the aim of obtaining a control model generation device that allows a person with sufficient knowledge of the actions of the controlled object to generate a control model representing the motion equations of the controlled object without pre-preparing multiple control model candidates.
[0009] The control model generation apparatus disclosed herein includes: an observation acquisition unit that acquires multiple observations of the output of a controlled object having nonlinear characteristics; a state-space model inference unit that infers a state-space model representing a linear approximate curve related to the multiple observations acquired by the observation acquisition unit; and an upper bound model inference unit that calculates the error, i.e., the inference error, between each observation acquired by the observation acquisition unit and the linear approximate curve represented by the state-space model inferred by the state-space model inference unit, and infers an upper bound model representing the upper bound of the inference error. Furthermore, the control model generation apparatus includes a control model generation unit that uses the state-space model inferred by the state-space model inference unit and the upper bound model inferred by the upper bound model inference unit to generate a control model representing the motion equations of the controlled object.
[0010] According to this disclosure, a control model representing the motion equations of the controlled object can be generated without pre-preparing multiple control model candidates, provided that a person with sufficient knowledge of the actions of the controlled object has sufficient knowledge of them. Attached Figure Description
[0011] Figure 1 This is a diagram showing the configuration of the control model generation device involved in Implementation Method 1.
[0012] Figure 2 This is a hardware configuration diagram showing the hardware of the control model generation device involved in Embodiment 1.
[0013] Figure 3 It is a hardware configuration diagram of a computer when the control model generation device is implemented by software or firmware.
[0014] Figure 4 This is an explanatory diagram illustrating the relationship between the controller with the installed control model CM and the linear system containing the state-space model Jz and the upper bound model Hz.
[0015] Figure 5A This is a graph representing an example of multiple observations f(z) of the output of the controlled object OB, the state-space model Jz, and the upper bound model Hz. Figure 5B Is to make Figure 5A The curve is rotated so that the state-space model Jz becomes a curve on the horizontal axis.
[0016] Figure 6 This is a flowchart illustrating the control model generation method, which is a processing step of the control model generation device.
[0017] Figure 7 This is a diagram showing the configuration of the control model generation device involved in Embodiment 2.
[0018] Figure 8This is a hardware configuration diagram showing the hardware of the control model generation device involved in Embodiment 2.
[0019] Figure 9A as well as Figure 9B These are explanatory diagrams of the model selection unit 9, which represents the selection of any one of the M controllers 31-1 to 31-Mz, 31-m.
[0020] Figure 10 It represents the M partial state spaces JK1~JK contained in the state space JK where there are multiple observations f(z). M Explanatory diagram. Detailed Implementation
[0021] Hereinafter, in order to illustrate this disclosure in more detail, the manner in which this disclosure is carried out will be described with reference to the accompanying drawings.
[0022] Implementation method 1.
[0023] Figure 1 This is a diagram showing the configuration of the control model generation device involved in Implementation Method 1.
[0024] Figure 2 This is a hardware configuration diagram showing the hardware of the control model generation device involved in Embodiment 1.
[0025] Figure 1 The control model generation device shown includes an observation acquisition unit 1, a state space model inference unit 2, an upper bound model inference unit 3, and a control model generation unit 4.
[0026] For example, the observation acquisition part 1 is obtained by Figure 2 The observed value acquisition circuit 11 shown is implemented.
[0027] The observation acquisition unit 1 acquires multiple observations f(z) of the output of the control object OB, which has nonlinear characteristics. The control object OB can be a known control object or an unknown control object.
[0028] The observation acquisition unit 1 outputs multiple observation values f(z) to the state-space model inference unit 2 and the upper bound model inference unit 3, respectively.
[0029] State-space model inference unit 2, for example, is derived from Figure 2 The state-space model inference circuit 12 shown is implemented.
[0030] The state-space model inference unit 2 obtains multiple observations f(z) from the observation acquisition unit 1.
[0031] The state-space model inference unit 2 infers the state-space model Jz representing the linear approximate curve involved in multiple observations f(z).
[0032] The state-space model inference unit 2 outputs the state-space model Jz to the upper bound model inference unit 3 and the control model generation unit 4, respectively.
[0033] Upper bound model inference part 3, for example, by Figure 2 The upper bound model inference circuit 13 shown is implemented.
[0034] The upper bound model inference unit 3 obtains multiple observation values f(z) from the observation value acquisition unit 1 and obtains the state space model Jz from the state space model inference unit 2.
[0035] The upper bound model inference unit 3 calculates the inference error |f(z)-Jz|, which is the error between each observation value f(z) and the linear approximation curve represented by the state space model Jz.
[0036] The upper bound model inference unit 3 infers the upper bound model Hz that represents the upper bound of the inference error |f(z)-Jz|.
[0037] Specifically, the upper bound model inference unit 3 uses a loss function ζ(t) that represents the difference between the result obtained by multiplying the squared value of the upper bound by a constant α less than 1 and the squared value of the inference error |f(z)-Jz| to infer the upper bound model Hz.
[0038] The upper bound model inference unit 3 outputs the upper bound model Hz to the control model generation unit 4.
[0039] Control model generation unit 4, for example, is composed of Figure 2 The control model generation circuit 14 shown is implemented.
[0040] The control model generation unit 4 obtains the state space model Jz from the state space model inference unit 2 and the upper bound model Hz from the upper bound model inference unit 3.
[0041] The control model generation unit 4 uses the state-space model Jz and the upper bound model Hz to generate a control model CM that represents the motion equations of the controlled object OB.
[0042] Specifically, the control model generation unit 4 generates a control model CM whose output of the control model has an error below the upper bound of the linear approximation curve.
[0043] The control model CM generated by the control model generation unit 4 is installed on the controller that controls the controlled object OB.
[0044] exist Figure 1 In this design, the observation acquisition unit 1, the state-space model inference unit 2, the upper bound model inference unit 3, and the control model generation unit 4, which are components of the control model generation device, are respectively composed of… Figure 2The dedicated hardware implementation is shown. That is, the control model generation device is envisioned to be implemented by the observation acquisition circuit 11, the state space model inference circuit 12, the upper bound model inference circuit 13, and the control model generation circuit 14.
[0045] The observation acquisition circuit 11, the state space model inference circuit 12, the upper bound model inference circuit 13, and the control model generation circuit 14 are respectively equivalent 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 circuit obtained by combining them.
[0046] The components of a control model generation device are not limited to being implemented by dedicated hardware; a control model generation device can also be implemented by software, firmware, or a combination of software and firmware.
[0047] Software or firmware is stored as a program in a computer's memory. A computer refers to the hardware that executes programs, such as CPU (Central Processing Unit), GPU (Graphics Processing Unit), central processing unit, processing unit, arithmetic unit, microprocessor, microcomputer, processor, or DSP (Digital Signal Processor).
[0048] Figure 3 It is a hardware configuration diagram of a computer when the control model generation device is implemented by software or firmware.
[0049] When the control model generation device is implemented by software or firmware, the program for causing the computer to execute each processing step in the observation acquisition unit 1, the state space model inference unit 2, the upper bound model inference unit 3, and the control model generation unit 4 is stored in the memory 21. Furthermore, the computer's processor 22 executes the program stored in the memory 21.
[0050] In addition, Figure 2 The image shows an example where the various components of the control model generation device are implemented using dedicated hardware. Figure 3 The example shown is of a control model generation device implemented by software or firmware. However, this is just one example; it is also possible that some components of the control model generation device are implemented by dedicated hardware, while the remaining components are implemented by software or firmware.
[0051] Figure 4 This is an explanatory diagram illustrating the relationship between the controller with the installed control model CM and the linear system containing the state-space model Jz and the upper bound model Hz.
[0052] exist Figure 4 The diagram illustrates the case where the output x of the state-space model Jz is assigned to the controller, and the output u of the controller is assigned to the linear system.
[0053] In addition, Figure 4 The diagram illustrates the case where the output p of the upper bound model Hz is assigned to the disturbance, and q, as the disturbance, is assigned to the linear system.
[0054] Figure 5A This is a graph representing an example of multiple observations f(z) of the output of the controlled object OB, the state-space model Jz, and the upper bound model Hz.
[0055] Figure 5A The example shown is a two-dimensional space with multiple observations f(z) as the output of the control object OB, represented by the z-axis and x-axis. However, this is just an example; the state space can also be three-dimensional, for example.
[0056] Figure 5B Is to make Figure 5A The curve is rotated so that the state-space model Jz becomes a curve on the horizontal axis.
[0057] The motion equation of the controlled object OB, point x, is shown in equation (1) below. In the specification document, due to the electronic application, the symbol “·” cannot be marked on the character x, so it is described as point x.
[0058]
[0059] In equation (2), T is the mathematical symbol for transpose.
[0060] The relationship between the multiple observations f(z) of the output of the controlled object OB, the state space model Jz, and the upper bound model Hz is shown in the following equation (3).
[0061]
[0062] The state-space model Jz is shown in equation (4) below. Furthermore, the inference 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 shown in equation (5) below. The upper bound model Hz is shown in equation (6) below.
[0063]
[0064] In equations (4) and (6), A, B, C, and D are arbitrary matrices.
[0065] The loss function L of the state-space model Jz J As shown in equation (7) below. Additionally, the loss function L of the upper bound model Hz... H As shown in equation (8) below.
[0066]
[0067] In equation (9), ζ(t) is the loss function representing the difference between the square of the upper bound p(t) multiplied by a constant α and the square of the inference error q(t). α is a constant less than 1.
[0068] Next, for Figure 1 The operation of the control model generation device shown is explained.
[0069] Figure 6 This is a flowchart illustrating the control model generation method, which is a processing step of the control model generation device.
[0070] The observation acquisition unit 1 acquires multiple observation values f(z) from the external source, representing the output of the control object OB, which has nonlinear characteristics. Figure 6 Step ST1).
[0071] The observation acquisition unit 1 outputs multiple observation values f(z) to the state-space model inference unit 2 and the upper bound model inference unit 3, respectively.
[0072] The state-space model inference unit 2 obtains multiple observations f(z) from the observation acquisition unit 1.
[0073] like Figure 5A as well as Figure 5B As shown, the state-space model inference unit 2 infers the state-space model Jz(representing the linear approximate curve involved in multiple observations f(z)). Figure 6 Step ST2).
[0074] The state-space model Jz representing the linear approximate curve involved in multiple observations f(z) is shown in equation (4).
[0075] Since the processing of inferring the state-space model Jz is a well-known technique, detailed explanations are omitted. For example, the state-space model Jz can be inferred by searching for the linear approximation curve that minimizes the expected value of the error vector representing the error between multiple observations f(z) and the linear approximation curve.
[0076] The state-space model inference unit 2 outputs the state-space model Jz to the upper bound model inference unit 3 and the control model generation unit 4, respectively.
[0077] The upper bound model inference unit 3 obtains multiple observation values f(z) from the observation value acquisition unit 1 and obtains the state space model Jz from the state space model inference unit 2.
[0078] As shown in equation (5), the upper bound model inference unit 3 calculates the inference error q, which is the error between each observation value f(z) and the linear approximation curve represented by the state-space model Jz. Figure 6 Step ST3).
[0079] As shown in equation (6), the upper bound model inference unit 3 infers the upper bound model Hz, which represents the upper bound of the inference error q. Figure 6 Step ST4).
[0080] Specifically, as shown in equation (9), the upper bound model inference unit 3 uses a loss function ζ(t) to infer the upper bound model Hz, which represents the difference between the result obtained by multiplying the squared value of the upper bound p by a constant α less than 1 and the squared value of the inference error q. The loss function ζ(t) is a function that outputs a vector that suppresses the norm of the error vector representing the difference from top to bottom.
[0081] The upper bound model inference unit 3 outputs the upper bound model Hz to the control model generation unit 4.
[0082] The control model generation unit 4 obtains the state space model Jz from the state space model inference unit 2 and the upper bound model Hz from the upper bound model inference unit 3.
[0083] As shown in equation (3), the control model generation unit 4 uses the state-space model Jz and the upper bound model Hz to generate a control model CM representing the motion equations of the controlled object OB. Figure 6 Step ST5).
[0084] f(z) that satisfies equation (3) is equivalent to the control model CM that represents the motion equation of the controlled object OB.
[0085] The control model CM generated by the control model generation unit 4 is installed on the controller that controls the controlled object OB.
[0086] Therefore, the error between the output of the control model CM and the output of the state-space model Jz can be considered as the error input to the controlled 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.
[0087] In Embodiment 1 described above, the control model generation apparatus is configured to include: an observation acquisition unit 1, which acquires multiple observations of the output of a control object having nonlinear characteristics; a state-space model inference unit 2, which infers a state-space model representing a linear approximate curve related to the multiple observations acquired by the observation acquisition unit 1; and an upper bound model inference unit 3, which calculates the error, i.e., the inference error, between each observation acquired by the observation acquisition unit 1 and the linear approximate curve represented by the state-space model inferred by the state-space model inference unit 2, and infers an upper bound model representing the upper bound of the inference error. Furthermore, the control model generation apparatus includes a control model generation unit 4, which uses the state-space model inferred by the state-space model inference unit 2 and the upper bound model inferred by the upper bound model inference unit 3 to generate a control model representing the motion equations of the control object. Therefore, the control model generation apparatus can generate a control model representing the motion equations of the control object without requiring multiple control model candidates to be prepared in advance, provided that a person sufficiently related to the action of the control object is available to do so.
[0088] Implementation method 2.
[0089] In Embodiment 2, a control model generation apparatus having a state space segmentation unit 5 that divides a state space containing multiple observation values f(z) into multiple spaces, i.e., partial state spaces, will be described.
[0090] Figure 7 This is a configuration diagram showing the control model generation device involved in Embodiment 2. Figure 7 In the middle, due to with Figure 1 The same reference numerals indicate the same or equivalent parts, therefore detailed descriptions are omitted.
[0091] Figure 8 This is a hardware configuration diagram showing the hardware of the control model generation device involved in Embodiment 2. Figure 8 In the middle, due to with Figure 2 The same reference numerals indicate the same or equivalent parts, therefore detailed descriptions are omitted.
[0092] Figure 7 The control model generation device shown includes an observation acquisition unit 1, a state space segmentation unit 5, a state space model inference unit 6, an upper bound model inference unit 7, a control model generation unit 8, and a model selection unit 9.
[0093] State space partitioning part 5, for example, is composed of Figure 8 The state space partitioning circuit 15 shown is implemented.
[0094] The state space partitioning unit 5 obtains multiple observations f(z) from the observation acquisition unit 1.
[0095] The state space partitioning unit 5 partitions the state space JK, which contains multiple observations f(z), into multiple spaces, namely partial state spaces JK1~JK. M M is an integer greater than or equal to 2.
[0096] State-space model inference unit 6, for example, is derived from Figure 8 The state-space model inference circuit 16 shown is implemented.
[0097] The state-space model inference unit 6 obtains multiple observations f(z) from the observation acquisition unit 1.
[0098] State-space model inference section 6 Inference part state-space model Jz m This part of the state-space model Jz m The state space JK is the portion of the state space after being divided by the state space partitioning part 5 from multiple observations f(z). m The observations f in (m=1, ..., M) m (z) is a state-space model represented by a linear approximation curve.
[0099] State-space model inference unit 6 infers the state-space models Jz of each part. m The outputs are respectively sent to the upper bound model inference unit 7 and the control model generation unit 8.
[0100] Upper bound model inference part 7, for example, by Figure 8 The upper bound model inference circuit 17 shown is implemented.
[0101] The upper bound model inference unit 7 obtains multiple observations f(z) from the observation acquisition unit 1, and obtains the partial state space models Jz from the state space model inference unit 6. m (m=1, ..., M).
[0102] Upper bound model inference unit 7 calculates inference error q m The inference error q m The state spaces JK are the various parts. m The observed value f in m (z) and the state-space model of each part Jz m The error of the linear approximation curve.
[0103] The upper bound model inference unit 7 infers the inference as a representation of each inference error q. m The upper bound of the upper bound model of the Hz model m (m=1, ..., M).
[0104] Upper bound model inference unit 7 will use the upper bound model Hz of each part. m Output to control model generation unit 8.
[0105] Control model generation unit 8, for example, is composed of Figure 8 The control model generation circuit 18 shown is implemented.
[0106] The control model generation unit 8 obtains a partial state-space model Jz from the state-space model inference unit 6. m (m=1, ..., M), obtain a partial upper bound model Hz from the upper bound model inference part 3. m (m=1, ..., M).
[0107] The control model generation unit 8 uses a partial state-space model Jz m and some upper bound models Hz m To generate JK with a partial state space m The corresponding control model CM m (m=1, ..., M).
[0108] The control model CM generated by the control model generation unit 8 m It is installed in the controller 31-m (m=1, ..., M) described later.
[0109] Model selection section 9, for example, is composed of Figure 8 The model selection circuit 19 shown is implemented.
[0110] Model selection unit 9 selects from the M partial state spaces JK1~JK generated by control model generation unit 8. M Each corresponding control model CM m Choose any control model from (m=1, ..., M).
[0111] Specifically, the model selection unit 9 calculates the M partial state spaces JK1~JK. M Each corresponding control model CM m Output and Partial State Space Model Jz m The error of the linear approximation curve represented Δ e m (m=1, ..., M).
[0112] Model selection unit 9 is based on M errors Δe1~Δe M The calculation results are used to select any control model.
[0113] exist Figure 7 In this design, the observation acquisition unit 1, state space segmentation unit 5, state space model inference unit 6, upper bound model inference unit 7, control model generation unit 8, and model selection unit 9, which are components of the control model generation device, are respectively composed of... Figure 8The dedicated hardware implementation is shown. That is, the control model generation device is envisioned to be implemented by an observation acquisition circuit 11, a state space partitioning circuit 15, a state space model inference circuit 16, an upper bound model inference circuit 17, a control model generation circuit 18, and a model selection circuit 19.
[0114] The observation acquisition circuit 11, the state space partitioning circuit 15, the state space model inference circuit 16, the upper bound model inference circuit 17, the control model generation circuit 18, and the model selection circuit 19 are respectively equivalent to, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an ASIC, an FPGA, or a circuit composed of these.
[0115] The components of a control model generation device are not limited to being implemented by dedicated hardware; a control model generation device can also be implemented by software, firmware, or a combination of software and firmware.
[0116] When the control model generation device is implemented by software or firmware, the program for causing the computer to execute each processing step in the observation acquisition unit 1, state space segmentation unit 5, state space model inference unit 6, upper bound model inference unit 7, control model generation unit 8, and model selection unit 9 is stored in... Figure 3 The memory 21 shown. Furthermore, Figure 3 The processor 22 shown executes the program stored in memory 21.
[0117] In addition, Figure 8 The image shows an example where the various components of the control model generation device are implemented using dedicated hardware. Figure 3 The example shown is an example of a control model generation device implemented by software or firmware. However, this is just one example; it is also possible to control some components of the model generation device to be implemented by dedicated hardware, while the remaining components are implemented by software or firmware.
[0118] Figure 9A as well as Figure 9B These are explanatory diagrams of the model selection unit 9, which represents the selection of any one of the M controllers 31-1 to 31-M, 31-m.
[0119] Controllers 31-1 to 31-M are controllers that control the controlled object OB.
[0120] Controller 31-m (m=1, ..., M) and partial state space Jz m correspond.
[0121] Figure 10 It represents the M partial state spaces JK1~JK contained in the state space JK where there are multiple observations f(z). MExplanatory diagram. Figure 10 An example with M=3 is shown.
[0122] Next, for Figure 7 The operation of the control model generation device shown is explained.
[0123] The observation acquisition unit 1 acquires multiple observation values f(z) from the outside as the output of the control object OB, which has nonlinear characteristics.
[0124] The observation acquisition unit 1 outputs multiple observations f(z) to the state space partitioning unit 5, the state space model inference unit 6, and the upper bound model inference unit 7, respectively.
[0125] The state space partitioning unit 5 obtains multiple observations f(z) from the observation acquisition unit 1.
[0126] like Figure 10 As shown, the state space partitioning unit 5 partitions the state space JK containing multiple observations f(z) into multiple spaces, namely partial state spaces JK1~JK. M .
[0127] The state-space model inference unit 6 obtains multiple observations f(z) from the observation acquisition unit 1.
[0128] The state-space model inference unit 6 determines the existence of multiple observations f(z) in the state space JK of each part. m The observed values f of (m=1, ..., M) m (z).
[0129] State-space model inference unit 6 infers pairs of states existing in part of the state space JK. m The observed value f m The state-space model represented by the linear approximation curve involved in (z) is also known as the partial state-space model Jz. m .
[0130] Partial state-space model Jz m As shown in equation (10) below. Due to the inference of the partial state-space model Jz m The processing itself is a well-known technique, so detailed explanations are omitted.
[0131] The state-space model inference unit 6 will infer part of the state-space model Jz m (m=1, ..., M) are output to the upper bound model inference unit 7 and the control model generation unit 8, respectively.
[0132]
[0133] In equation (10), A m B mIt is any matrix.
[0134] The upper bound model inference unit 7 obtains multiple observations f(z) from the observation acquisition unit 1, and obtains the partial state space models Jz from the state space model inference unit 6. m (m=1, ..., M).
[0135] Upper bound model inference unit 7 calculates inference error q m The inference error q m The state spaces JK are the various parts. m The observed value f in m (z) and the state-space model of each part Jz m The error of the linear approximation curve represented. Inference error q m The calculation is performed according to the following formula (11).
[0136] The upper bound model inference unit 7 infers the inference as a representation of each inference error q. m The upper bound of the upper bound model of the Hz model m (m=1, ..., M). Partial upper bound model Hz m The inference process for (m=1, ..., M) is performed according to the following equation (12).
[0137] Upper bound model inference unit 7 will partially implement the upper bound model Hz m (m=1, ..., M) are output to the control model generation unit 8.
[0138]
[0139] In equation (12), C m D m It is any matrix.
[0140] The control model generation unit 8 obtains a partial state-space model Jz from the state-space model inference unit 6. m (m=1, ..., M), obtain a partial upper bound model Hz from the upper bound model inference part 3. m (m=1, ..., M).
[0141] As shown in equation (13) below, the control model generation unit 8 uses a partial state-space model Jz m and some upper bound models Hz m To generate JK with a partial state space m The corresponding control model CM m .
[0142] f satisfies equation (13) m (z) is equivalent to a partial state space JK mThe corresponding control model CM represents the motion equations of the controlled object OB. m .
[0143] The control model CM generated by the control model generation unit 8 m It is installed on the controller 31-m (m=1, ..., M).
[0144]
[0145] Model selection unit 9 selects from M partial state spaces JK1~JK M Each corresponding control model CM m Choose any control model CM from (m=1, ..., M). m .
[0146] Specifically, as shown in equation (14) below, the model selection unit 9 calculates the M partial state spaces JK1~JK M Each corresponding control model CM m Output u m With partial state-space model Jz m The error Δe of the linear approximation curve is represented M (m=1, ..., M).
[0147]
[0148] Model selection unit 9 is based on M errors Δe1~Δe M The calculation results are derived from M control models CM1~CM m Choose any one of the control models CM m .
[0149] Specifically, the model selection unit 9 has errors Δe1~Δe M Determine the minimum error Δe MIN .
[0150] Model selection unit 9 selects from M control models CM1~CM m Among them, choose the one with the smallest error Δe MIN The corresponding control model CM m .
[0151] The control model CM selected by the model selection unit 9 is installed from among M controllers 31-1 to 31-M. m The controller 31-m controls the controlled object OB.
[0152] exist Figure 7 In the control model generation device shown, the model selection unit 9 selects from M control models CM1 to CM2. mAmong them, choose the one with the smallest error Δe MIN The corresponding control model CM m However, this is just one example; the model selection unit 9 can also be used in practical applications to select a range with the minimum error Δe without any problems. MIN Control model CM for errors other than those mentioned above m Specifically, the model selection unit 9 can select the control model CM with the second smallest error. m Alternatively, select the control model CM with the third smallest error. m .
[0153] In the above-described embodiment 2, the system is configured to include a model selection unit 9 that selects any one control model from among the control models generated by the control model generation unit 8 that correspond to each of the plurality of partial state spaces. Figure 7 The control model generation device shown. Therefore, Figure 7 The control model generation device shown is Figure 1 The control model generation device shown is similar, capable of generating a control model representing the motion equations of the controlled object without pre-preparing multiple control model candidates, even when a person with sufficient knowledge of the actions related to the controlled object is involved. Furthermore, it is comparable to... Figure 1 Compared to the control model generation device shown, it can improve the control accuracy of the controller.
[0154] exist Figure 7 In the control model generation device shown, the model selection unit 9 calculates and generates M partial state spaces JK1~JK. M Each corresponding control model CM m Output and Partial State Space Model Jz m The error Δe of the linear approximation curve is represented M (m=1, ..., M), and based on M errors Δe1~Δe M The calculation results are used to select any control model. However, this is just an example; the model selection unit 9 can also select from the M partial upper bound models Hz1~Hz inferred by the upper bound model inference unit 7. m Among them, the partial upper bound model with the smallest slope of the upper bound of the gain is determined as the partial upper bound model Hz. m From M control models CM1~CM m The selected and determined upper bound model Hz m The corresponding control model CM m .
[0155] This disclosure allows for free combination of various embodiments or modification of any constituent elements of each embodiment, or omission of any constituent elements in each embodiment.
[0156] Industrial availability
[0157] This disclosure applies to control model generation apparatus and control model generation method.
[0158] Explanation of reference numerals in the attached figures
[0159] 1...Observation acquisition unit; 2...State-space model inference unit; 3...Upper bound model inference unit; 4...Control model generation unit; 5...State-space partitioning unit; 6...State-space model inference unit; 7...Upper bound model inference unit; 8...Control model generation unit; 9...Model selection unit; 11...Observation acquisition circuit; 12...State-space model inference circuit; 13...Upper bound model inference circuit; 14...Control model generation circuit; 15...State-space partitioning circuit; 16...State-space model inference circuit; 17...Upper bound model inference circuit; 18...Control model generation circuit; 19...Model selection circuit; 21...Memory; 22...Processor; 31-1~31-M...Controller.
Claims
1. A control model generation device, characterized in that, have: The observation acquisition unit acquires multiple observations of the output of the control object, which has nonlinear characteristics. The state-space model inference unit infers a state-space model representing a linear approximate curve involved by multiple observations obtained by the observation acquisition unit. The upper bound model inference unit calculates the error, i.e., the inference error, between each observation obtained by the observation acquisition unit and the linear approximate curve represented by the state space model inferred by the state space model inference unit, and infers an upper bound model representing the upper bound of the inference error. as well as The control model generation unit uses the state-space model inferred by the state-space model inference unit and the upper bound model inferred by the upper bound model inference unit to generate a control model representing the motion equations of the controlled object.
2. The control model generation device according to claim 1, characterized in that, The upper bound model inference unit uses a loss function that represents the difference between the result obtained by multiplying the squared value of the upper bound by a constant less than 1 and the squared value of the inference error to infer the upper bound model.
3. The control model generation device according to claim 1, characterized in that, The control model generation unit generates a control model whose output is below the upper bound and whose error is the error between the output of the control model and the linear approximation curve.
4. The control model generation device according to claim 1, characterized in that, The system includes a state space partitioning unit that divides the state space containing multiple observations acquired by the observation acquisition unit into multiple spaces, i.e., partial state spaces. The state-space model inference unit infers a partial state-space model, which represents a state-space model of the linear approximate curves involved in the observations existing in each partial state space after the state space is divided by the observation acquisition unit from multiple observations acquired by the observation acquisition unit. The upper bound model inference unit calculates the error between the observed values in each partial state space and the linear approximation curve represented by each partial state space model, i.e., the inference error, and infers a partial upper bound model that serves as the upper bound of the upper bound model representing each inference error. The control model generation unit uses the partial state space models inferred by the state space model inference unit and the partial upper bound models inferred by the upper bound model inference unit to generate a control model that represents the motion equations of the controlled object corresponding to each partial state space.
5. The control model generation device according to claim 4, characterized in that, It includes a model selection unit that selects any one control model from the control models generated by the control model generation unit that correspond to each of the plurality of partial state spaces.
6. The control model generation device according to claim 5, characterized in that, The model selection unit calculates the error between the output of the control model corresponding to each of the plurality of partial state spaces and the linear approximation curve, and selects any control model based on the calculation results of the error between the output of each control model and the linear approximation curve.
7. A method for generating a control model, characterized in that, The observation acquisition unit acquires multiple observations of the output of the controlled object, which exhibits nonlinear characteristics. The state-space model inference unit infers a state-space model representing the linear approximate curve involved by the multiple observations obtained by the observation acquisition unit. The upper bound model inference unit calculates the error, i.e., the inference error, between each observation obtained by the observation acquisition unit and the linear approximate curve represented by the state-space model inferred by the state-space model inference unit, and infers an upper bound model representing the upper bound of the inference error. The control model generation unit uses the state-space model inferred by the state-space model inference unit and the upper bound model inferred by the upper bound model inference unit to generate a control model representing the motion equations of the controlled object.
Citation Information
Patent Citations
Control method and control device
JP1993303408A