JP2023114204A - Control device, control method and program - Google Patents

Abstract

To enable model predictive control on an edge device, at the same time, automatically adjust a control parameter required for the model predictive control and enable recovery of model identification.SOLUTION: A control device that outputs an amount of operation for a control target and causes an amount of control of the control target to follow a target value sequentially repeats in each predetermined control cycle: calculation of an estimated value of the model parameter by a model parameter estimation unit; calculation of a control parameter by a control parameter calculation unit; calculation of a target deviation by a target deviation calculation unit; calculation of the correction target deviation by a correction target deviation calculation unit; calculation of an amount of change in the amount of operation by an operation change amount calculation unit; and calculation of a new amount of operation by an operation amount calculation unit.SELECTED DRAWING: Figure 2

Description

The present invention relates to a control device, control method and program.

Control devices such as temperature control devices, PLCs (Programmable Logic Controllers), DCSs (Distributed Control Systems), and control devices mounted on personal computers and built-in control devices are widely used in industry.

In addition, PID (Proportional-Integral-Differential) control, model predictive control, internal model control, LQG (Linear-Quadratic-Gaussian) control, Various control methods such as H2 control and H∞ control are known.

Model predictive control is a method of obtaining a desired response by sequentially performing optimization calculations using a state space model of the controlled object and a future time response model. Patent document 1). For example, as an industrial application of standard model predictive control that executes a numerical optimization algorithm online, application to control of an air conditioning system is known (for example, Patent Document 1).

Further, there has been proposed a control device that determines a new manipulated variable based on a corrected target deviation, which is the difference between a predicted value of the controlled variable according to changes in the manipulated variable in the past up to the present and a target value. (For example, Patent Document 2 and Patent Document 3).

In addition, as a method of identifying parameters for a linear prediction model, a sequential least squares method, that is, a RLS (Recursive Least Squares) method, a Kalman filter method, etc. are known (for example, Non-Patent Document 2, Non-Patent Document 3 ).

Furthermore, techniques for determining PID control parameters by a self-tuning method are known (for example, Non-Patent Document 4 and Non-Patent Document 5).

JP 2015-108499 A WO2016/092872 Japanese Patent Application Laid-Open No. 2020-21411

Yang M. Matiejovsky, "Model Predictive Control: Optimal Control Under Constraints", Tokyo Denki University Press, 2005 Shuichi Adachi, "System Identification for Control by MATLAB", Tokyo Denki University Press, 1996 Shuichi Adachi, "Advanced System Identification for Control by MATLAB", Tokyo Denki University Press, 2004 Toru Yamamoto, Kenzo Fujii, "Design of self-tuning control system for practical use", Transactions of the Institute of Electrical Engineers of Japan, Vol. 123, No. 9 (2003) Toru Yamamoto, "New Development of Self-tuning Method Generalized Minimum Dispersion Control and PID Control", Instrumentation and Control, Vol.40, No.10 (2001)

The techniques described in Non-Patent Literature 4 and Non-Patent Literature 5 are promising as techniques that can realize parameter tuning on edge devices (for example, controllers such as PLCs) that have relatively scarce computational resources. There are the following issues.

・Since PID control has only three parameters with a degree of freedom, it may not be possible to match the ideal response.

・When using generalized minimum variance control, it is necessary to give a reference model function that fully considers the plant time constant in advance.

On the other hand, model predictive control has a relatively high degree of freedom in the plant response model, and can handle plants with long dead times and plants with inverse responses that PID control is not good at, but it has the following problems. .

・Because it is a model-based control method, it is necessary to identify the model of the controlled object in advance.

・It is difficult to perform this model identification and control response engineering (for example, preliminary work such as simulation and data analysis to determine control parameters such as control gain) on edge devices with relatively scarce computational resources. .

In addition, when model identification (learning (or estimating) of model parameters) is performed by the RLS method or the like, there are the following problems.

・Since model identification is generally a non-linear algorithm, the results depend on the initial values of the model parameters, and unexpected values may occur.

- During the learning process, the learning state may become unstable due to some factor (for example, disturbance).

Therefore, it is necessary to make model identification recoverable in case the model parameters fall into unexpected values or the learning state becomes unstable.

An embodiment of the present invention has been made in view of the above points, and enables model predictive control on an edge device, automatically adjusts control parameters necessary for model predictive control, and enables model identification. It is intended to be recoverable.

In order to achieve the above object, a control device according to one embodiment outputs a manipulated variable to a controlled object and causes the controlled variable of the controlled object to follow a target value, wherein the controlled variable and the manipulated variable A model parameter configured to calculate an estimated value of the model parameter of the plant response function, which is a function representing the plant response model of the controlled object and a function including the model parameter, based on an estimation unit, a control parameter calculation unit configured to calculate control parameters including a control gain and a look-ahead length based on the plant response function, and a difference between the target value and the current control amount. a target deviation calculation unit configured to calculate a target deviation; and a calculation of the control amount after the look-ahead length has elapsed, based on the plant response function and the amount of change in the past manipulated variable up to the present time. a corrected target deviation calculating unit configured to calculate a corrected target deviation obtained by correcting the target deviation using a predicted value; and a change amount of the manipulated variable calculated by multiplying the corrected target deviation by the control gain. a manipulated variable calculation unit configured to calculate a new manipulated variable by adding the manipulated variable change amount and the current manipulated variable; a reset processing unit configured to initialize the model parameters in response to an input of a predetermined reset signal, wherein the model parameter estimation unit calculates estimated values of the model parameters; calculation of the control parameter by the parameter calculation unit; calculation of the target deviation by the target deviation calculation unit; calculation of the corrected target deviation by the corrected target deviation calculation unit; Calculation of the amount of change and calculation of the new manipulated variable by the manipulated variable calculator are sequentially repeated at each predetermined control cycle.

While enabling model predictive control on the edge device, it is also possible to automatically adjust the control parameters necessary for model predictive control and recover model identification.

It is a figure which shows an example of the hardware constitutions of the control apparatus which concerns on this embodiment. It is a figure showing an example of functional composition of a control device concerning this embodiment. It is a figure for demonstrating an example of operation|movement of a plant response function. 4 is a flowchart for explaining an example of calculation processing of a plant response function; 7 is a flowchart for explaining an example of model parameter estimation processing; FIG. 5 is a diagram for explaining an example of the operation of a correction target deviation calculator; 7 is a flowchart for explaining an example of reset processing; FIG. 10 is a flowchart for explaining an example of model parameter/covariance matrix resetting; FIG. FIG. 10 is a diagram for explaining an example of the operation of an operation change amount calculator; 4 is a flowchart for explaining an example of control parameter calculation processing; FIG. 4 is a diagram showing a step response of a plant to be controlled in Example 1; FIG. 11 is a diagram (part 1) showing a control response in Example 1; FIG. 10 is a diagram (part 1) showing changes in model parameters in Example 1; FIG. 10 is a diagram (part 1) showing changes in control parameters in the first embodiment; FIG. 2 is a diagram (part 2) showing a control response in Example 1; FIG. 10 is a diagram (part 2) showing changes in model parameters in the first embodiment; FIG. 4 is a diagram (part 2) showing transition of control parameters in the first embodiment; 3 is a diagram (3) showing a control response in Example 1. FIG. FIG. 10 is a diagram (part 3) showing changes in model parameters in Example 1; FIG. 10 is a diagram (part 3) showing transitions of control parameters in the first embodiment; FIG. 10 is a diagram showing control responses in Example 2; FIG. 10 is a diagram showing changes in model parameters in Example 2; FIG. 10 is a diagram showing changes in control parameters in Example 2;

An embodiment of the present invention will be described below. In the following, given a target value for an arbitrary plant to be controlled, the manipulated variable for making the controlled variable follow the target value by model predictive control is calculated, and the model parameters for updating the model are calculated. A control device 10 that performs estimation and calculation of control parameters necessary for model predictive control and that is capable of recovering model identification will be described. Here, it is assumed that the control device 10 according to the present embodiment is, for example, an edge device (PLC, DCS, etc.) with less computational resources than a PC (personal computer) or the like. Note that model identification is estimating (or may be called learning) model parameters.

The control device 10 according to the present embodiment is based on an arbitrary target value r, a controlled variable y that indicates the state of the controlled plant 20, a plant response model of the controlled plant 20, and the like. Calculating u as well as estimating model parameters and calculating control parameters. Then, the control device 10 measures the controlled variable y of the controlled plant 20 according to the manipulated variable u, and calculates the next manipulated variable u based on the target value r, the controlled variable y, the plant response model, and the like. , estimates new model parameters and calculates new control parameters. Thus, the control device 10 according to the present embodiment performs the calculation of the manipulated variable u for causing the controlled variable y to follow the target value r, the estimation of the model parameters of the plant response model, and the calculation of the control parameters online. It is repeatedly executed during execution (that is, during control of the controlled plant 20). That is, the control device 10 according to the present embodiment realizes model predictive control without requiring online optimization calculation each time, and automatically adjusts the control parameters necessary for it.

As a result, the control device 10 according to the present embodiment can achieve high control performance by model predictive control even with relatively scarce computational resources, and can eliminate the work required for adjusting control parameters. In addition, since control parameter adjustment work (for example, off-line data analysis, engineering work related to model predictive control such as simulation) is not required, it is possible to reduce the time and cost required for it.

In addition, after the start of operation of the plant 20 to be controlled, the control parameters are automatically adjusted while estimating the model parameters that follow the seasonal changes in the plant characteristics and the temporal changes such as aging deterioration, so that the control performance is stabilized. can be maintained.

Furthermore, it is possible to realize model predictive control that does not require a pre-adjusted reference model function and can handle plants that are not good at PID control (plants with long dead time or plants with inverse response).

In addition to the above, the control device 10 according to the present embodiment can recover model identification (that is, reset model parameters).

As a result, the control device 10 according to the present embodiment, for example, when the model parameter falls into an unexpected value, or when the learning state becomes unstable due to some factor (for example, disturbance, etc.), the model After recovering the identification, it is possible to perform model identification again using the recovered model parameters as initial values.

The controlled variable y is, for example, the temperature of the controlled plant 20, and the target value r is, for example, the set temperature. However, the controlled variable y and the target value r are not limited to the temperature and the set temperature, and any controlled variable in the plant 20 to be controlled and the target target value of the controlled variable can be used.

<Hardware Configuration of Control Device 10>
First, the hardware configuration of the control device 10 according to this embodiment will be described with reference to FIG. FIG. 1 is a diagram showing an example of the hardware configuration of a control device 10 according to this embodiment.

As shown in FIG. 1 , the control device 10 according to this embodiment has an input device 11 , a display device 12 , an external I/F 13 , a communication I/F 14 , a processor 15 and a memory device 16 . Each of these pieces of hardware is communicably connected via a bus 17 .

The input device 11 is, for example, a keyboard, mouse, touch panel, various physical buttons, and the like. The display device 12 is, for example, a display, a display panel, or the like. Note that the control device 10 may not have at least one of the input device 11 and the display device 12, for example.

The external I/F 13 is an interface with an external device such as the recording medium 13a. Examples of the recording medium 13a include a CD (Compact Disc), a DVD (Digital Versatile Disk), an SD memory card (Secure Digital memory card), a USB (Universal Serial Bus) memory card, and the like.

Communication I/F 14 is an interface for connecting control device 10 to a communication network. The processor 15 is, for example, various arithmetic units such as a CPU (Central Processing Unit) and an MPU (Micro-Processing Unit). The memory device 16 is, for example, various storage devices such as SSD (Solid State Drive), RAM (Random Access Memory), ROM (Read Only Memory), and flash memory.

Note that the hardware configuration shown in FIG. 1 is an example, and the control device 10 may have another hardware configuration. For example, the control device 10 may have a plurality of processors 15 and a plurality of memory devices 16, or may have various hardware other than the illustrated hardware.

<Functional Configuration of Control Device 10>
Next, the functional configuration of the control device 10 according to this embodiment will be described with reference to FIG. FIG. 2 is a diagram showing an example of the functional configuration of the control device 10 according to this embodiment.

As shown in FIG. 2, the control device 10 according to the present embodiment includes a model parameter estimation unit 101, a measurement unit 102, a differentiator 103, an operation amount update unit 104, a reset processing unit 105, and a timer 106. have These are realized by, for example, one or more programs installed in the control device 10 causing the processor 15 or the like to execute processing. Further, the control device 10 according to this embodiment has a prediction time series storage unit 107 . The predicted time series storage unit 107 is implemented by, for example, the memory device 16 included in the control device 10 .

The measurement unit 102 measures (observes) the controlled variable y of the controlled plant 20 at each control cycle _Tc . Then, the measurement unit 102 outputs the latest value of the measured control amount y as the control amount current value _y0 . The controlled variable y of the controlled plant 20 is determined according to the manipulated variable u and the disturbance v. For example, when the controlled variable y is temperature, the disturbance v may be a drop or rise in the outside air temperature.

In addition, the measurement unit 102 acquires (observes) the manipulated variable u output from the manipulated variable updating unit 104 at each control cycle _Tc , and converts the latest value of the acquired manipulated variable u to the current manipulated variable u ₀ output as

The differentiator 103 outputs the difference (deviation) between the target value r and the current controlled variable value _y0 as the target deviation _e0 . The target deviation e ₀ (t) at time t is calculated by e ₀ (t)=r(t)−y ₀ (t). Note that in this embodiment, the target value is constant, that is, r(t)=constant.

The manipulated variable update unit 104 outputs the manipulated variable u for the controlled plant 20 every control cycle _Tc . Here, the manipulated variable updater 104 includes a control parameter calculator 111 , a corrected target deviation calculator 112 , an manipulated variable calculator 113 , and an adder 114 .

The control parameter calculator 111 receives the plant response function {S _θ (t)}, calculates the control gain k _I and the look-ahead length T _p as control parameters, and applies the control gain k _I to the operation change amount calculator 113 . , and the look-ahead length _Tp to the correction target deviation calculator 112, respectively.

The corrected target deviation calculator 112 calculates the plant response function {S _θ (t)}, the target deviation e ₀ (t), and the operation change amount time series { du(t)} and the look-ahead length T _p , a corrected target deviation e ^* (t) is calculated by correcting the target deviation e ₀ (t). At this time, the corrected target deviation calculation unit 112 uses the prediction time series (the time series of the predicted value of the control amount y up to a certain time in the future) stored in the prediction time series storage unit 107 to calculate the corrected target deviation Compute e ^* (t). The details of the calculation method of the corrected target deviation e ^* (t) will be described later.

The operation change amount calculator 113 calculates the operation change amount du(t) based on the corrected target deviation e ^* (t) and the control gain _kI . The manipulation change amount calculation unit 113 calculates and outputs the manipulation change amount du(t) in the order of du(t−3T _c ), du(t−2T _c ), and du(t−T _c ), for example. The amount of change in operation du is the amount by which the amount of operation u changes in each control cycle _Tc .

The adder 114 adds the current value u ₀ of the operation amount output from the measurement unit 102 and the operation change amount du output from the operation change amount calculation unit 113 to calculate a new operation amount _us . The adder 114 then outputs this new manipulated variable u to the controlled plant 20 . A new manipulated variable u is calculated by u(t)=u ₀ +du(t)=u(t−T _c )+du(t).

The model parameter estimating unit 101 inputs the current controlled variable value y ₀ (t) and the current manipulated variable u ₀ (t) for each control cycle T _c , and calculates the model of the plant response function {S _θ (t)}. Calculate and output the estimated value θ _est of the parameter. This model parameter estimate θ=θ _est is set to the plant response function {S _θ (t)}. A plant response function {S _θ (t)} is a function including a model parameter θ, and is a plant response model of the plant 20 to be controlled. Hereinafter, the model parameter θ set in the plant response function {S _θ (t)} is also referred to as “model parameter set value θ”. Note that the initial value θ ₀ of the model parameter may be input when calculating the estimated value θ _est of the model parameter.

The reset processing unit 105 resets the model parameters of the plant response function {S _θ (t)} and the predicted time series stored in the predicted time series storage unit 107 when the reset signal is input. This recovers the model identification.

The timer 106 operates the model parameter estimating unit 101, the measuring unit 102, and the manipulated variable updating unit 104 every control cycle _Tc . That is, the timer 106 operates as an operation trigger for the model parameter estimating unit 101, the measuring unit 102, and the manipulated variable updating unit 104 every control cycle _Tc . The control period _Tc is a period for controlling the plant 20 to be controlled, and its value is set in advance.

The predicted time series storage unit 107 stores the predicted time series calculated by the corrected target deviation calculator 112 (time series of predicted values of the controlled variable y up to a certain future time).

Note that the control device 10 according to the present embodiment has a display control unit that causes the display device 12 to sequentially display, for example, model parameters and control parameters (control gain, look-ahead length) calculated for each control cycle _Tc . Alternatively, it may have a display control unit that sequentially displays these model parameters and control parameters on a display device of a terminal such as a PC connected via a communication network. As a result, users and administrators of the control device 10 (hereinafter also referred to as “users, etc.”) can know the trend of model parameters and control parameters, and the identification state of the plant response model and the adjustment state of the control parameters. can be confirmed. In addition, it is also possible to find signs of sudden fluctuations in the plant response model.

<Operation of plant response function {S _θ (t)}>
Next, the operation of the plant response function {S _θ (t)} will be described with reference to FIG. FIG. 3 is a diagram for explaining an example of the operation of the plant response function {S _θ (t)}.

As shown in FIG. 3, the plant response function {S _θ (t)} is a unit step response S Output _θ (t). The unit step response is the response (that is, the output of the plant response model of the controlled plant 20) when the manipulated variable u is the unit step input.

<Calculation of plant response function {S _θ (t)}>
Next, the process of calculating the unit step response S _θ (t) of the plant response function {S _θ (t)} will be described with reference to FIG. FIG. 4 is a flowchart for explaining an example of calculation processing of the plant response function {S _θ (t)}. In FIG. 4, it is assumed that the model parameter set value θ and the time t are input.

Here, in this embodiment, as a calculation model of the plant response function S _θ (t), autoregression of the past N points for the controlled variable y, and moving average of the current value and the past M−1 points for the manipulated variable u are used. A case where the used ARMA (autoregressive moving average) model is adopted will be described. Note that N and M are set in advance by, for example, a user or the like. However, the ARMA model is only an example, and the present invention is not limited to this. For example, other models such as the ARMAX model may be adopted.

At this time, the model parameter setting value θ is set as follows.

^θ =[−a ₁ ,−a ₂ , . _. . ,−a _N ,b ₀ ,b ₁ ,b ₂ , .
Here, T represents transposition.

Each element of the model parameter setting value θ is a coefficient of the ARMA model shown below.

y(k)=-a ₁ y(k-1)-a ₂ y(k-2)-...-a _N y(k-N)+b ₀ u(k)+b ₁ u(k-1) +b ₂ u(k−2)+ . . . +b _M−1 u(k−M+1)
where k is an index, starting from k=1. Note that the ARMA model described above is only an example, and for example, a model that considers the disturbance v of past L points (L is a preset integer equal to or greater than 1) may be used.

Step S11: The manipulated variable update unit 104 sets τ as a variable representing time, φ(k) as a state vector at index k, and y(k) as a predicted control amount value, and sets the time τ, the state vector φ(0), and the control Initialize the quantity prediction value y(0).

Here, the state vector φ(k) is expressed as follows.

φ(k)=[y(k-1), y(k-2), ..., y(kN), u(k), u(k-1), u(k-2), , u(k−M+1)] ^Τ
For example, when considering the disturbance v of L points in the past, the state vector φ(k) is a state having L disturbances v(k) as elements in addition to y(k) and u(k). Just expand it to a vector.

For example, the manipulated variable updating unit 104 initializes τ=0 and y(0)=0, and φ(0)=[0, 0, . , 0] ^T. That is, φ(0) is initialized to 1 only for the element corresponding to u(k) of φ(k) and to 0 for the other elements. This means that it is set as an initial value when simulating a unit step response.

Step S12: Next, the manipulated variable updating unit 104 updates the state vector φ(k) based on the control amount predicted value y(k−1) and the state vector φ(k−1). For example, the manipulated variable updating unit 104 updates the state vector φ(k) as follows.

φ(k)=[y(k-1), y(k-2), ..., y(kN), u(k), u(k-1), u(k-2), , u(k−M+1)] ^Τ
At this time, the control amount prediction value y(k-1) is set to y(k-1), and 1 is set to u(k). , y(k-N) and u(k-1), u(k-2), . (k-1) (that is, y(k-2), ..., y(kN) and u(k-1), contained in the state vector φ(k-1) u(k−2), . . . , u(k−M+1) respectively).

Step S13: Next, the manipulated variable updating unit 104 calculates the controlled variable predicted value y(k) based on the model parameter setting value θ and the state vector φ(k). The manipulated variable updating unit 104 calculates the controlled variable predicted value y(k) by, for example, y(k)=φ(k) ^Tθ .

Step S14: Next, the manipulated variable updating unit 104 updates the time τ to τ+ΔT and updates the index k to k+1. Here, ΔT represents the time width of one step of the ARMA model. The time step ΔT of the plant response function can be changed according to the control cycle _Tc . Since the ARMA model is a discrete model, it is preferable to set ΔT= _Tc .

Step S15: Next, the manipulated variable updating unit 104 determines whether or not τ≧t. Then, if it is not determined that τ≧t (NO in step S15), the manipulated variable updating unit 104 returns to step S12. As a result, steps S12 to S14 are repeatedly executed until τ≧t.

On the other hand, if it is determined that τ≧t (YES in step S15), the manipulated variable updating unit 104 terminates the process. This gives the final calculated y(k) as the unit step response S _θ (t) (that is, S _θ (t)=y(k) is the step response function {S _θ (t)} is calculated as the unit step response at time t of ).

<Estimation of model parameter θ>
Next, the process of estimating the model parameter θ of the plant response function {S _θ (t)} (that is, the process of calculating the model parameter estimated value θ _est ) will be described with reference to FIG. FIG. 5 is a flowchart for explaining an example of model parameter estimation processing. In FIG. 5, a case will be described in which the model parameter estimated value θ _est at a certain index k is calculated assuming that the current controlled variable value y ₀ and the manipulated variable current value u ₀ are input. Also, the value of the control amount one time ago (that is, y ₀ (t−T _c ) when the current time is t) is called the control amount previous value and is denoted by y ₋₁ . Note that this index k is a value independent of the index k used to calculate the plant response function in FIG. 4, and represents an index during model parameter estimation processing.

Step S21: The model parameter estimation unit 101 determines whether or not to initialize the model parameter estimated value θ _est (k) and the covariance matrix P(k). Here, as a case where it is determined to be initialized, for example, when the model parameter estimated value θ _est is calculated for the first time (that is, when the model parameter estimated value θ _est is calculated for the first time), the user or the like issues an initialization instruction. and the like.

If it is determined to initialize the model parameter estimation value θ _est (k) and the covariance matrix P(k) (YES in step S21), the model parameter estimation unit 101 proceeds to step S22. On the other hand, if it is determined not to initialize the model parameter estimated value θ _est (k) and the covariance matrix P(k) (NO in step S21), the model parameter estimation unit 101 proceeds to step S23.

Step S22: If it is determined in step S21 above to initialize the model parameter estimate θ _est (k) and the covariance matrix P(k), the model parameter estimation unit 101 initializes k=0 and , θ _est (0)=θ ₀ and P(0)=γI (where γ>0). Here, as θ ₀ , a model parameter setting value θ that has already been set in the plant response function {S _θ (t)} may be used, or an initial value assumed in advance by a user or the like may be used. Alternatively, model parameter estimated values θ _est =θ _est (k−1) obtained by reset processing, which will be described later, may be used. Also, I may be a unit matrix or any predetermined matrix.

Step S23: The model parameter estimator 101 updates the state vector φ(k) based on the state vector φ(k−1), the current value u ₀ of the manipulated variable, and the previous value y ₋₁ of the controlled variable. That is, model parameter estimating section 101 updates state vector φ(k) as follows, for example.

φ(k)=[y(k-1), y(k-2), ..., y(kN), u(k), u(k-1), u(k-2), , u(k−M+1)] ^Τ
At this time, y(k−1) is set to the previous controlled variable value y ₋₁ , and u(k) is set to the current manipulated variable value u ₀ . , y(k-N) and u(k-1), u(k-2), . (k-1) (that is, y(k-2), ..., y(kN) and u(k-1), contained in the state vector φ(k-1) u(k−2), . . . , u(k−M+1) respectively).

For example, when considering the disturbance v of L points in the past, as described above, the state vector φ(k) is added to y(k) and u(k), and L disturbances v(k) as elements, and estimate the covariance matrix and model parameters for the disturbance.

Step S24: Next, the model parameter estimating unit 101 calculates the prediction error ε(k) based on the model parameter estimated value θ _est (k−1), the state vector φ(k), and the control amount current value y ₀ . calculate. Note that the model parameter estimated value θ _est (k−1) is the estimated value of the model parameter θ previously estimated (that is, at k−1).

The model parameter estimation unit 101, for example, calculates the prediction error ε(k) as follows.

y(k)= _y0
ε(k) _{=y(k)−φ(k) Τθest} ⁽ k−1)
Step S25: Next, the model parameter estimation unit 101 updates the covariance matrix P(k). Model parameter estimation section 101 updates covariance matrix P(k), for example, as follows.

ここで、Ｐ（ｋ－１）は前回（つまり、ｋ－１のとき）得られた共分散行列である。また、λは０＜λ≦１を取る忘却係数であり、予め設定された値である。忘却係数λは過去データを忘却するための係数であり、０に近いほど急激に過去データの影響が減少し、１に近いほど過去データの影響が保持される。

where P(k-1) is the covariance matrix obtained last time (that is, at k-1). Also, λ is a forgetting coefficient that takes 0<λ≦1 and is a preset value. The forgetting coefficient λ is a coefficient for forgetting past data.

Step S26: Next, the model parameter estimation unit 101 determines whether or not to update the model parameter estimated value θ _est . Here, as a case where it is determined to update the model parameter estimated value θ _est , for example, when the model parameter estimated value θ _est has not been updated until a predetermined period has passed since the initialization in step S22. , a case where an update instruction is given by a user or the like.

If it is determined to update the model parameter estimated value θ _est (YES in step S26), the model parameter estimation unit 101 proceeds to step S27. On the other hand, if it is not determined to update the model parameter estimated value θ _est (NO in step S26), the model parameter estimation unit 101 sets θ _est (k)=θ _est (k−1) and then step Without executing S27, the process proceeds to step S28.

Step S27: If it is determined in step S26 to update the model parameter estimation value θ _est , the model parameter estimation unit 101 updates the model parameter estimation value θ _est (k). The model parameter estimating unit 101 updates the model parameter estimated value θ _est (k), for example, as follows.

ステップＳ２８：そして、モデルパラメータ推定部１０１は、モデルパラメータ推定処理の実行時インデックスｋをｋ＋１に更新する。

Step S28: Then, the model parameter estimation unit 101 updates the execution time index k of the model parameter estimation process to k+1.

<Operation of Correction Target Deviation Calculation Unit 112>
Next, the operation of the corrected target deviation calculator 112 will be described with reference to FIG. FIG. 6 is a diagram for explaining an example of the operation of the corrected target deviation calculator 112. As shown in FIG.

As shown in FIG. 6, the corrected target deviation calculator 112 calculates the plant response function {S _θ (t)}, the target deviation e ₀ (t), the operation variation time series {du(t)}, and the look-ahead When the length T _p is input, a value that is predicted to change the control amount after the lapse of T _p from the current time t is calculated as the look-ahead response correction value y _n (t) due to the amount of change in the past operation. Note that the look-ahead length _Tp is calculated by the control parameter calculator 111 .

Then, the corrected target deviation calculator 112 outputs a corrected target deviation e ^* (t) obtained by correcting the target deviation e ₀ (t) with the look-ahead response correction value y _n (t). Here, the corrected target deviation e ^* (t) is calculated by e ^* (t)=r(t)-(y ₀ (t)+y _n (t))=e ₀ (t)-y _n (t) be done. Therefore, as an example, a method of calculating the look-ahead response correction value y _n (t) using the predicted time series stored in the predicted time series storage unit 107 will be described below.

A predicted value of the controlled variable y at time s predicted at time t is defined as a generalized predicted value y _n,C (s|t), which is calculated as follows.

ここで、Ｍは、一般化予測値の計算に使用するモデルの長さ（モデル区間）である。

where M is the length of the model (model interval) used to calculate the generalized predictor.

At this time, the prediction time series storage unit 107 stores the generalized prediction values y _n,C (s|t) from the time t−Δt to the future time t+ _Tb as a time series, where t is the current time. It is assumed that there is That is, the prediction time series storage unit 107 stores y _n,C (t−Δt|t), y _n,C (t|t), y _n,C (t+Δt|t) _{, .} Assume that _C (t+T _b |t) is stored. Here, T _b is a constant that determines the length (time series length) of the generalized prediction values y _n,C stored in the prediction time series storage unit 107, and T _b =N ₁ Δt (N ₁ is any determined positive integer). Also, Δt is the prediction interval, and Δt= _Tc .

When the generalized predicted value y _n,C is used, the look-ahead response predicted value y _{n,A (} t), which is the predicted value of the control amount y at the look-ahead time t+T _p based only on the operation change amount time series {du(t)}, is , y _n,A (t)=y _n,C (t+T _p |t). Further, the free response prediction value y _n,B (t), which is the prediction value of the control amount y at the current time t based only on the operation change amount time series {du(t)}, is y _n,B (t)=y _{n , C} (t|t).

On the other hand, the look-ahead response correction value y _n (t) can be calculated as y _n (t)=y _n,A (t)−y _n,B (t). Therefore, by using the prediction time-series storage unit 107, the corrected target deviation calculator 112 can calculate the corrected target deviation e ^* (t) with a small amount of calculation and a small amount of memory.

Here, the predicted time-series storage unit 107 is updated by the correction target deviation calculation unit 112 each time a new operation change amount du(t) is calculated by the operation change amount calculation unit 113 . An example of a method for updating the prediction time series storage unit 107 will be described below.

A generalized predicted value y _n,C (t+mΔt|t) ahead of mΔt predicted at time t is expressed as follows.

このため、時刻ｔ＋Δｔで予測したｍΔｔ先における一般化予測値ｙ_ｎ，Ｃ（ｔ＋Δｔ＋ｍΔｔ｜ｔ＋Δｔ）は、以下で表される。

Therefore, the generalized predicted value y _n,C (t+Δt+mΔt|t+Δt) after mΔt predicted at time t+Δt is expressed as follows.

すなわち、時刻ｔ＋Δで予測したｍΔｔ先における一般化予測値ｙ_ｎ，Ｃ（ｔ＋Δｔ＋ｍΔｔ｜ｔ＋Δｔ）は、ｙ_ｎ，Ｃ（ｔ＋Δｔ＋ｍΔｔ｜ｔ＋Δｔ）＝Ｓ（ｍΔｔ）ｄｕ（ｔ＋Δｔ）＋ｙ_ｎ，Ｃ（ｔ＋（ｍ＋１）Δｔ｜ｔ）となる。

That is, the generalized prediction value y _n,C (t+Δt+mΔt|t+Δt) ahead of mΔt predicted at time t+Δ is yn _,C (t+Δt+mΔt|t+Δt)=S(mΔt)du(t+Δt)+y _n,C (t+( m+1) Δt|t).

Therefore, the generalized predicted value y n _{, C} after mΔpredicted at time t+Δt is the operation change amount du It can be updated as a value affected by (t+Δ). Note that such an update can be performed by the method described in Patent Document 2 or Patent Document 3, for example.

<Reset processing>
Next, processing for resetting the model parameters of the plant response function {S _θ (t)} and the prediction time series stored in the prediction time series storage unit 107 will be described with reference to FIG. FIG. 7 is a flowchart for explaining an example of reset processing. Note that the reset processing shown in FIG. 7 is executed when a reset signal is input to the reset processing unit 105 . A reset signal is input from, for example, a user, another program, another device, or the like. It is assumed below that a reset signal is input to the reset processing unit 105 .

Step S31: The reset processing unit 105 determines whether or not the reset signal is ON. The fact that the reset signal is ON means that the reset signal is a signal representing a predetermined value (for example, "1"). On the other hand, when the reset signal is a signal representing a value other than the predetermined value, the reset signal is said to be OFF. For example, when the model parameter θ becomes an unexpected value, or when the learning state of the model parameter θ becomes unstable, a reset signal indicating ON is input to the reset processing unit 105 . In addition, for example, when the control performance is degraded due to nonlinearity or the like (that is, when the divergence between the target value r and the control amount y becomes large), a reset signal indicating ON is input to the reset processing unit 105. may be Alternatively, for example, a reset signal indicating ON may be input to the reset processing unit 105 in response to a change in the operation mode of the controlled plant 20 (for example, a change in operation mode such as preparation, reaction, delivery, etc.), A reset signal indicating ON may be input to the reset processing unit 105 by some other logic.

If it is determined that the reset signal is ON (YES in step S31), the reset processing unit 105 proceeds to step S32. On the other hand, if it is not determined that the reset signal is ON (NO in step S31), the reset processing unit 105 ends the reset processing without performing anything.

Step S32: The reset processing unit 105 resets the model parameter θ=θ _est and the covariance matrix P. That is, the reset processing unit 105 instructs the model parameter estimation unit 101 to reset the model parameter θ and the covariance matrix P. Details of this step will be described later.

Step S<b>33 : The reset processing unit 105 resets the predicted time series stored in the predicted time series storage unit 107 . That is, the reset processing unit 105 obtains the generalized prediction values y _n,C (t−Δt|t), y _n,C (t|t), y _n,C ( t+Δt|t), . . . y _n,C (t+T _b |t) are all initialized to zero. As a result, the influence of the operation change amount du before the current time t is eliminated.

<Reset of model parameter θ and covariance matrix P>
Details of the process of resetting the model parameter θ=θ _est and the covariance matrix P in step S32 of FIG. 7 will be described with reference to FIG. FIG. 8 is a flowchart for explaining an example of model parameter/covariance matrix resetting. 8 is executed when the reset processing unit 105 instructs to reset the model parameter θ and the covariance matrix P. FIG.

Step S41: The model parameter estimation unit 101 initializes θ _est (0)=0 and P(0)=γI. Since no prior information is used, the model parameter estimate θ _est (0) is initialized to a zero vector, and the covariance matrix P(0) is initialized to a factor γ times the identity matrix I (where γ>0). An arbitrary positive value can be used for γ, but if the value of γ is set to a value close to 0, sudden fluctuations in the initial model parameter estimated values can be suppressed.

Step S42: Next, the model parameter estimator 101 generates a state vector φ(0) representing a steady state. The model parameter estimation unit 101, for example, generates a state vector φ(0) representing a steady state as follows.

φ(0)= _[ y ₀ , y ₀ , . . . , y ₀ , u ₀ , u _{0 ,} ^.
That is, _y0 is set to the elements corresponding to y(k-1), y(k-2), ..., y(kN), and u(k), u(k-1), Set u ₀ to the elements corresponding to u(k−2), . . . , u(k−M+1). This is obtained by _generating a state vector by updating the state vector in step S23 of FIG _. is. The state vector φ(0) representing the steady state is hereinafter also referred to as the steady state vector.

Step S43: Next, the model parameter estimation unit 101 initializes k=1. Note that this index k is a value independent of the index k used in FIGS.

Step S44: Next, the model parameter estimation unit 101 determines whether or not k≧K. Here, K is a preset value.

If it is determined that k≧K (YES in step S44), the model parameter estimation unit 101 proceeds to step S49. On the other hand, if it is not determined that k≧K, the model parameter estimation unit 101 proceeds to step S45.

Step S45: If k≧K is not determined in step S44, the model parameter estimation unit 101 calculates the model parameter estimated value θ _est (k−1), the steady-state vector φ(0), and the current control amount. Calculate the prediction error ε(k) based on the value y ₀ . Note that the model parameter estimated value θ _est (k−1) is the estimated value of the model parameter θ previously estimated (that is, at k−1).

The model parameter estimation unit 101, for example, calculates the prediction error ε(k) as follows.

y(0)= _y0
ε(k)=y( _{0)−φ(0) Τθest} ⁽ k−1)
It should be noted that since steady-state prediction is performed in the reset process, only the model parameter estimated value θ _est depends on the index k.

Step S46: Next, the model parameter estimation unit 101 updates the covariance matrix P(k). Model parameter estimation section 101 updates covariance matrix P(k), for example, as follows.

ここで、Ｐ（ｋ－１）は前回（つまり、ｋ－１のとき）得られた共分散行列である。また、λは０＜λ≦１を取る忘却係数であり、予め設定された値である。なお、忘却係数λは、図５のステップＳ２５と同様の値を用いることが好適である。

where P(k-1) is the covariance matrix obtained last time (that is, at k-1). Also, λ is a forgetting coefficient that takes 0<λ≦1 and is a preset value. Incidentally, it is preferable to use the same value as in step S25 of FIG. 5 as the forgetting factor λ.

Step S47: Next, the model parameter estimator 101 updates the model parameter estimated value θ _est (k). The model parameter estimating unit 101 updates the model parameter estimated value θ _est (k), for example, as follows.

ステップＳ４８：次に、モデルパラメータ推定部１０１は、インデックスｋをｋ＋１に更新する。これにより、ｋ≧Ｋと判定されるまで、上記のステップＳ４６～ステップＳ４８が繰り返し実行され、その結果、モデルパラメータθが、定常状態ベクトルφ（０）に対するモデルパラメータ推定値θ_ｅｓｔ（Ｋ－１）にリセットされることになる。

Step S48: Next, the model parameter estimation unit 101 updates the index k to k+1. As a result, the above-described steps S46 to S48 are repeatedly executed until it is determined that k≧K, and as a result, the model parameter θ becomes the model parameter estimated value θ _est (K−1 ) will be reset.

Step S49: If k≧K is determined in step S44 above, the model parameter estimation unit 101 reinitializes the covariance matrix P to P(0)=γI. This is because the reset process calculates the model parameter estimated value θ _est for the steady-state vector φ(0), so the covariance matrix P is generally not a correct value, so reinitialization is performed.

<Operation of Operation Variation Calculation Unit 113>
Next, the operation of the operation change amount calculation unit 113 will be described with reference to FIG. FIG. 9 is a diagram for explaining an example of the operation of the operation change amount calculation unit 113. As shown in FIG.

As shown in FIG. 9, when the correction target deviation e ^* (t) and the control gain _kI are input, the ^operation change amount calculation unit 113 calculates the control parameter It is multiplied by the control gain _kI calculated by the calculator 111 to output the operation change amount du(t). That is, the operation change amount calculation unit 113 calculates the operation change amount du(t) by du(t)=k _I ×e ^* (t).

However, if the result of multiplying the corrected target deviation e ^* (t) by the control gain _kI exceeds the upper limit du _max , the operation change amount calculator 113 sets du _max to the operation change amount du(t). . Similarly, when the result of multiplying the corrected target deviation e ^* (t) by the control gain _kI is below the lower limit value du _min , the operation change amount calculation unit 113 regards du _min as the operation change amount du(t). do. This makes it possible to provide a limiter for the upper and lower limits. It should be noted that du _max and du _min may be set each time so that the manipulated variable u obtained by adding the manipulated variable current value _u0 and the manipulated variable amount du by the adder 114 does not deviate from the predetermined upper and lower limits. good.

<Calculation of control parameters>
Next, processing for calculating the control gain _kI and the look-ahead length _Tp as control parameters will be described with reference to FIG. FIG. 10 is a flowchart for explaining an example of control parameter calculation processing. In FIG. 10, it is assumed that the adjustment coefficients α and β are input in addition to the plant response function {S _θ (t)}. The adjustment coefficients α and β are preset values, preferably 0<α≦1 and 0<β≦1.

Step S51: The control parameter calculator 111 calculates the look-ahead length T _p based on the plant response function {S _θ (t)} and the adjustment coefficient β. Using the _{plant response} function {S _θ (t)}, _the control parameter calculator 111 calculates the adjustment coefficient β The time point at which the plant response function value becomes equal to the value obtained by multiplying by is the look-ahead length _Tp time point. That is, the control parameter calculator 111 calculates (searches) the look-ahead length _Tp as follows.

Find T _p , where S _θ (T _p ) = β x S _θ (T _max )
When the response of the closed loop is slow, the speed of the response can be adjusted by setting the adjustment coefficient β to a smaller value.

Further, the control parameter calculator 111 calculates the gain _gp at the look-ahead length _Tp as follows.

g _p = S _θ (T _p )
Thus, the control gain _gp is calculated based on the look-ahead length _Tp . As a result, for example, even when the dead time and time constant of the controlled plant 20 are long, it is possible to respond more quickly.

Note that T _max may be a value that is sufficient for the plant response function to converge, or may be a value determined based on memory restrictions or the like. Alternatively, the maximum value of the period that can be predicted by the corrected target deviation calculator 112 may be set to _Tmax .

Step S52: Then, the control parameter calculator 111 calculates the control gain _kI based on the gain _gp and the adjustment coefficient α at the look-ahead length _Tp . The control parameter calculator 111 calculates the control gain _kI as follows.

k _I =α×(1/g _p )
That is, the control gain _kI is obtained by multiplying the reciprocal of the gain _gp at the look-ahead length _Tp by the adjustment coefficient α.

If there is a lot of overshoot in the closed-loop response, the amount of overshoot can be adjusted by setting the adjustment coefficient α to a smaller value.

<Expansion of plant response model>
A plant response model represented by an ARMA model, an ARMAX model, or the like can be expanded to a model to which an offset term (constant term) representing a steady state is added.

For example, when represented by an ARMA model, the plant response model can be extended to a model with an offset term as follows.

y(k)=-a ₁ y(k-1)-a ₂ y(k-2)-...-a _N y(k-N)+b ₀ u(k)+b ₁ u(k-1) +b ₂ u(k−2)+ . . . +b _M−1 u(k−M+1)+c
Here, c is an offset term and represents the steady state of the plant 20 to be controlled.

When considering a plant response model with the above offset term, by adding an offset term to the end of the model parameter θ and adding 1 as an element corresponding to the offset term to the end of the state vector φ(k), , the present embodiment can be similarly applied.

For example, when the plant response model is represented by the ARMA model with the above offset term, _the model parameters _{are θ=[−a 1 , −a 2 ,} . ^, _. , u(k), u(k−1), u(k−2), . . . , u(k−M+1), 1] ^τ .

Thereby, even if the controlled variable and the manipulated variable have an offset, the offset can be estimated at the same time.

When calculating the unit step response S _θ (t) of the plant response function {S _θ (t)} in FIG. need not be added.

<Example 1>
Next, Example 1 will be described. In this embodiment, a controlled plant 20 represented by a plant response model having an offset term is assumed.

The plant response model of the plant 20 to be controlled in this embodiment is y(k)=θ ₁ y(k−1)+θ ₂ y(k−2)+θ ₃ u(k)+θ ₄ u( k−1)+θ ₅ u(k−2)+θ ₆ . where θ ₆ is the offset term.

At this time, the state vector is expressed as φ(k)=[y(k-1), y(k-2), u(k), u(k-1), u(k-2), 1] ^T and the model parameters are expressed as θ(k)=[θ ₁ (k), θ ₂ (k), θ ₃ (k), θ ₄ (k), θ ₅ (k), θ ₆ (k)] ^T be done. Let the true values of the model parameters excluding the offset term _θ6 be θ(k)=[1.73, −0.737, −0.00185, 0.00415, 0.0117] ^T . Also, if the controlled variable estimated value is y _est , the controlled variable estimated value is calculated by y _est (k)=φ(k) ^Τ θ(k−1).

FIG. 11 shows the unit step response of the controlled plant 20 in this embodiment. As shown in FIG. 11, the unit step response has a shape that increases smoothly.

Also, the upper limit of the manipulated variable is 90, and the lower limit is 0. The adjustment coefficients were α=0.95 and β=0.3. The forgetting factor was set to λ=0.95.

Under the above settings, the model parameter estimating unit 101 estimates the model parameter θ and the control parameter calculating unit 111 calculates the control parameter at each control cycle _Tc . Experiments for controlling the controlled plant 20 with the quantity u were conducted in three cases.

(first case)
The first case is a case in which a reset signal indicating ON is input to the reset processing unit 105 at time t=0.

FIG. 12 shows the control response (target value r, controlled variable y, and manipulated variable u at that time) in this case, FIG. 13 shows the transition of the model parameter .theta., and FIG. 14 shows the transition of the control parameter.

In the transition of the model parameter θ, after the reset process is executed at time t=0, the model parameter changes significantly at the timing when the target value r changes. In addition, the control parameters _kI and _Tp also change accordingly. In the control response, although an overshoot is seen at the beginning, the controlled variable y follows the target value r, and it can be confirmed that the follow-up control by the control device 10 according to the present embodiment is sufficiently functioning.

(second case)
The second case is a case in which a reset signal indicating ON is input to the reset processing unit 105 at times t=0 and t=400.

FIG. 15 shows the control response (target value r, controlled variable y, and manipulated variable u at that time) in this case, FIG. 16 shows the transition of the model parameter .theta., and FIG. 17 shows the transition of the control parameter.

In transition of the model parameter θ, the reset process is executed at time t=0 and time t=400, and the control parameter is changed accordingly. Although the reset process is executed during control at time t=400, the controlled variable y follows the target value r, indicating that the follow-up control by the control device 10 according to the present embodiment is sufficiently functioning. I can confirm.

(third case)
A third case is a case in which a reset signal indicating ON is input to the reset processing unit 105 at time t=400 with the initial value (at t=0) of the model parameter θ as the zero vector.

FIG. 18 shows the control response (target value r, controlled variable y, and manipulated variable u at that time) in this case, FIG. 19 shows the transition of the model parameter .theta., and FIG. 20 shows the transition of the control parameter.

At time t=0, the model parameter θ starts from a zero vector state, and the control response reveals a deviation between the controlled variable y and the target value r. This is because the initial value of the model parameter θ is estimated far from the true value, resulting in insufficient prediction accuracy.

In the transition of the model parameter θ, the reset process was executed at time t=400, and the control parameter also changed accordingly. After the reset process is executed, the controlled variable y follows the target value r, and it can be confirmed that the follow-up control by the control device 10 according to the present embodiment is sufficiently functioning.

<Example 2>
Next, Example 2 will be described. In this embodiment, the plant 20 to be controlled is assumed to be represented by a plant response model that does not have an offset term.

The plant response model of the plant 20 to be controlled in this embodiment is y(k)=θ ₁ y(k−1)+θ ₂ y(k−2)+θ ₃ u(k)+θ ₄ u( k−1)+θ ₅ u(k−2).

At this time, the state vector is expressed as φ(k)=[y(k-1), y(k-2), u(k), u(k-1), u(k-2)] ^Τ , The model parameters are expressed as θ(k)=[θ ₁ (k), θ ₂ (k), θ ₃ (k), θ ₄ (k), θ ₅ (k)] ^T . Other settings are the same as in the first embodiment.

Under the above settings, the model parameter estimating unit 101 estimates the model parameter θ and the control parameter calculating unit 111 calculates the control parameter at each control cycle _Tc . An experiment was conducted to control the controlled plant 20 with the quantity u.

In this experiment, a reset signal indicating ON is input to the reset processing unit 105 at time t=0 and time t=600.

FIG. 21 shows the control response (target value r, controlled variable y, and manipulated variable u at that time) in this experiment, FIG. 22 shows the transition of the model parameter .theta., and FIG. 23 shows the transition of the control parameter.

In transition of the model parameter θ, the reset process is executed at time t=0 and time t=600, and the control parameter is changed accordingly. Although the reset process is executed during control at time t=600, the controlled variable y follows the target value r, and the offset of the control deviation is improved compared to the first embodiment. It can be confirmed that the follow-up control by the control device 10 according to the embodiment functions sufficiently.

<Summary>
As described in the above two examples, the control device 10 according to the present embodiment, for example, when the model parameter falls into an unexpected value, or when the learning state becomes unstable due to some factor, etc. In addition, reset processing can quickly recover learning for model identification.

In addition, the following (1) to (3) can be realized by the control device 10 according to the present embodiment.

(1) A consistent initial value for the model parameter θ can be given. This is because the model parameter estimated value θ _est =θ _est (K−1) obtained by the reset process can be used as the initial value.

(2) It is possible to deal with the nonlinearity of the controlled plant 20 (however, the nonlinearity that can be locally regarded as linear). This is because a reset signal indicating ON can be input to the reset processing unit 105 to reset the model parameter θ and the like when the control performance is degraded due to nonlinearity.

(3) It is possible to respond to changes in the operation mode of the controlled plant 20 . This is because the model parameter θ and the like can be reset by inputting a reset signal indicating ON corresponding to a change in the operation mode to the reset processing unit 105 .

The present invention is not limited to the specifically disclosed embodiments described above, and various variations, modifications, combinations with known techniques, etc. are possible without departing from the scope of the claims. be.

10 control device 11 input device 12 display device 13 external I/F
13a recording medium 14 communication I/F
15 processor 16 memory device 17 bus 20 plant to be controlled 101 model parameter estimating unit 102 measuring unit 103 differentiator 104 manipulated variable update unit 105 reset processing unit 106 timer 107 prediction time series storage unit 111 control parameter calculation unit 112 correction target deviation calculation unit 113 operation variation calculator 114 adder

Claims (10)

A control device that outputs a manipulated variable for a controlled object and causes the controlled variable of the controlled object to follow a target value,
calculating an estimated value of the model parameter of the plant response function, which is a function that represents the plant response model of the controlled object and that includes model parameters, based on the controlled variable and the manipulated variable; a model parameter estimator comprising
a control parameter calculation unit configured to calculate control parameters including a control gain and a look-ahead length based on the plant response function;
a target deviation calculator configured to calculate a target deviation, which is a difference between the target value and the current controlled variable;
calculating a corrected target deviation in which the target deviation is corrected by the predicted value of the controlled variable after the look-ahead length has elapsed, based on the plant response function and the amount of change in the past manipulated variable up to the present A correction target deviation calculation unit configured,
an operation change amount calculation unit configured to calculate an amount of change in the operation amount by multiplying the correction target deviation by the control gain;
a manipulated variable calculation unit configured to calculate a new manipulated variable by adding the amount of change in the manipulated variable and the current manipulated variable;
a reset processing unit configured to initialize the model parameters in response to input of a predetermined reset signal;
has
Calculation of estimated values of the model parameters by the model parameter estimator, calculation of the control parameters by the control parameter calculator, calculation of the target deviation by the target deviation calculator, and calculation of the target deviation by the corrected target deviation calculator Calculation of a corrected target deviation, calculation of the amount of change in the manipulated variable by the manipulated variable calculator, and calculation of the new manipulated variable by the manipulated variable calculator are sequentially repeated for each predetermined control cycle. Control device. The corrected target deviation calculator,
calculating a predicted value of the controlled variable up to a predetermined time in the future as a predicted time series based on the plant response function and the amount of change in the manipulated variable in the past up to the present, and storing the predicted value in a storage unit;
The corrected target deviation is calculated by calculating a predicted value of the control amount after the look-ahead length has elapsed based on the predicted time series stored in the storage unit,
The reset processing unit
2. The control device according to claim 1, configured to initialize the model parameters and the prediction time series stored in the storage unit. The reset processing unit
3. The control device according to claim 2, configured to initialize each prediction value included in the prediction time series to zero. The reset processing unit
4. The control device according to any one of claims 1 to 3, configured to initialize said model parameters when a reset signal indicating ON is input. The reset processing unit
After setting the model parameters to predetermined initial values in response to the input of the reset signal, the model parameters are estimated assuming that the current control amount and the current operation amount continue for a predetermined period. 5. The control device according to any one of claims 1 to 4, configured to cause the model parameter estimator to calculate the value. The plant response function is an ARMA model or an ARMAX model,
The model parameter estimator,
6. The control device according to any one of claims 1 to 5, configured to calculate said estimated value using coefficients of said ARMA model or ARMAX model as said model parameters. The plant response function is
7. The control device according to any one of claims 1 to 6, which is a function having model parameters including an offset term representing a steady state of said controlled object. The model parameter estimator,
8. Estimated values of the model parameters are calculated by estimating a covariance matrix and estimating the model parameters based on a recursive least squares method including a forgetting element. The control device according to any one of 1. A control device that outputs a manipulated variable for a controlled object and causes the controlled variable of the controlled object to follow a target value,
A model for calculating estimated values of the model parameters of a plant response function, which is a function representing the plant response model of the controlled object and includes model parameters, based on the controlled variable and the manipulated variable. a parameter estimation procedure;
a control parameter calculation procedure for calculating control parameters including a control gain and a look-ahead length based on the plant response function;
a target deviation calculation procedure for calculating a target deviation that is the difference between the target value and the current controlled variable;
Correction target for calculating a correction target deviation obtained by correcting the target deviation by using the predicted value of the control amount after the look-ahead length has elapsed, based on the plant response function and the amount of change in the past operation amount up to the present time. a deviation calculation procedure;
an operation change amount calculation procedure for calculating an amount of change in the operation amount by multiplying the correction target deviation by the control gain;
a manipulated variable calculation procedure for calculating a new manipulated variable by adding the amount of change in the manipulated variable and the current manipulated variable;
a reset processing procedure for initializing the model parameters in response to input of a predetermined reset signal;
and run
calculation of the model parameter estimated value by the model parameter estimation procedure; calculation of the control parameter by the control parameter calculation procedure; calculation of the target deviation by the target deviation calculation procedure; Calculation of a corrected target deviation, calculation of the amount of change in the manipulated variable by the procedure for calculating the amount of operation change, and calculation of the new manipulated variable by the procedure for calculating the manipulated variable are sequentially repeated at each predetermined control cycle. control method. A control device that outputs a manipulated variable for a controlled object and causes the controlled variable of the controlled object to follow a target value,
A model for calculating estimated values of the model parameters of a plant response function, which is a function representing the plant response model of the controlled object and includes model parameters, based on the controlled variable and the manipulated variable. a parameter estimation procedure;
a control parameter calculation procedure for calculating control parameters including a control gain and a look-ahead length based on the plant response function;
a target deviation calculation procedure for calculating a target deviation that is the difference between the target value and the current controlled variable;
Correction target for calculating a correction target deviation obtained by correcting the target deviation by using the predicted value of the control amount after the look-ahead length has elapsed, based on the plant response function and the amount of change in the past operation amount up to the present time. a deviation calculation procedure;
an operation change amount calculation procedure for calculating an amount of change in the operation amount by multiplying the correction target deviation by the control gain;
a manipulated variable calculation procedure for calculating a new manipulated variable by adding the amount of change in the manipulated variable and the current manipulated variable;
a reset processing procedure for initializing the model parameters in response to input of a predetermined reset signal;
and
calculation of the model parameter estimated value by the model parameter estimation procedure; calculation of the control parameter by the control parameter calculation procedure; calculation of the target deviation by the target deviation calculation procedure; The calculation of the corrected target deviation, the calculation of the amount of change in the manipulated variable by the procedure for calculating the amount of operation change, and the calculation of the new manipulated variable by the procedure for calculating the manipulated variable are sequentially repeated for each predetermined control cycle. ,program.

Images