JP6901037B1 - Control devices, control methods and programs - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a method of estimating parameters for model update while realizing model predictive control on an edge device. SOLUTION: This is a control device that makes a controlled amount of a controlled object follow a target value, and when a target value time series and a look-ahead length indicating a time width for pre-reading the target value are input, a plurality of target values Among them, the target value look-ahead means for acquiring the look-ahead target value, the look-ahead target deviation calculation means for calculating the look-ahead target deviation which is the difference between the read-ahead target value and the current control amount of the controlled object, and the plant response of the controlled object. A correction target that calculates the correction target deviation that indicates the difference between the predicted value of the control amount after the look-ahead length and the look-ahead target value based on the plant response function that represents The deviation calculation means, the operation amount calculation means for calculating a new operation amount based on the correction target deviation, the control amount, and the estimated value of the model parameter included in the plant response function are calculated based on the operation amount. It has a model parameter estimation means. [Selection diagram] Fig. 1

Description

The present invention relates to control devices, control methods and programs.

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

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

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

In addition, a control device that determines a new manipulated variable based on the predicted value of the convergence value of the controlled variable according to changes in the past manipulated variable up to the present and the correction target deviation, which is the difference between the target value and the target value, has been proposed. (For example, Patent Document 2). A design support device has also been proposed that displays on a graph an area in which a control logical formula indicating the relationship between the current value of the target deviation and the amount of change in the manipulated variable is established (for example, Patent Document 3).

Further, as a method for identifying parameters for a linear prediction model, a sequential least squares method, that is, an RLS (Recursive Least Squares) method, a Kalman filter method, and the like are known (for example, Non-Patent Document 2 and Non-Patent Document 3). ).

Further, for example, in the optimum operation of a power plant, a control method is also known in which a characteristic parameter is estimated from a physical model of a power generation element and an input state quantity, and an optimum load distribution is determined from the obtained characteristic parameter.

Japanese Unexamined Patent Publication No. 2015-108499 International Publication No. 2016/092872 International Publication No. 2015/060149

Yang M. Matieyovsky, "Optimal Control under Model Predictive Control Constraints", Tokyo Denki University Press, 2005 Shuichi Adachi, "System Identification for Control by MATLAB", Tokyo Denki University Press, 1996 Shuichi Adachi, "Identification of Advanced Systems for Control by MATLAB", Tokyo Denki University Press, 2004

However, since the conventional model predictive control needs to repeatedly execute sequential optimization calculations, it is often executed on a PC (personal computer) having abundant computing resources such as a CPU (Central Processing Unit) and memory. It was difficult to execute on an edge device (for example, a control device such as a PLC (Programmable Logic Controller)), which has many relatively scarce computing resources.

In addition, since model predictive control is a control method based on a model, it is necessary to identify the model to be controlled in advance, but even if the model is once identified, it depends on factors such as aging and seasonal fluctuation of the controlled object. It may change.

One embodiment of the present invention has been made in view of the above points, and an object of the present invention is to estimate parameters for updating a model while realizing model prediction control on an edge device.

In order to achieve the above object, the control device according to the embodiment of the present invention is a control device that outputs an operation amount for a control target and causes the control amount of the control target to follow a target value. When the target value time series, which is a time series, and the look-ahead length indicating the time width for pre-reading the target value are input, among the plurality of target values included in the target value time series, the target value after the look-ahead length A pre-reading target deviation calculating means for calculating a pre-reading target deviation, which is the difference between the pre-reading target value and the current control amount of the controlled object, and the pre-reading target deviation calculating means A correction target deviation indicating the difference between the predicted value of the control amount after the look-ahead length and the look-ahead target value based on the plant response function representing the plant response and the amount of change in the operation amount in the past up to the present. The plant response function is based on the correction target deviation calculating means for calculating the above, the operation amount calculating means for calculating a new operation amount based on the correction target deviation, the control amount, and the operation amount. It is characterized by having a model parameter estimation means for calculating an estimated value of the included model parameters.

It is possible to estimate the parameters for model update while realizing model predictive control on the edge device.

It is a figure which shows an example of the structure of the control device which concerns on 1st Embodiment. It is a figure for demonstrating an example of the operation of a plant response function. It is a flowchart for demonstrating an example of the calculation process of a plant response function. It is a flowchart for demonstrating an example of the estimation process of a model parameter. It is a figure for demonstrating an example of the operation of the target value look-ahead part. It is a figure (the 1) for demonstrating an example of the operation of the look-ahead response correction part. It is a figure (the 2) for demonstrating an example of the operation of the look-ahead response correction part. It is a flowchart for demonstrating an example of the update process executed by the prediction time series update part. It is a figure for demonstrating an example of the update of a prediction time series. It is a figure for demonstrating an example of the operation of the operation change amount calculation part. It is a figure which shows an example of the hardware composition of the control device which concerns on one Embodiment. It is a figure which shows an example of the structure of the control device which concerns on 2nd Embodiment. It is a flowchart for demonstrating an example of a control amount estimation process. It is a figure which shows an example of the structure of the control device which concerns on 3rd Embodiment. It is a figure which shows an example of the structure of the control device which concerns on 4th Embodiment. It is a figure for demonstrating the step response in Example 1. FIG. It is a figure for demonstrating the setting at the time of model parameter estimation in Example 1. FIG. It is a figure for demonstrating the operation amount and control amount in Example 1. FIG. It is a figure for demonstrating the model parameter estimated value and control quantity estimated value in Example 1. FIG. It is a figure for demonstrating the operation amount and control amount in Example 2. FIG. It is a figure for demonstrating the model parameter estimated value and control quantity estimated value in Example 2. FIG.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. In the following, the control device 10 that calculates the operation amount for making the control amount follow the target value and estimates the model parameter for model update when the target value time series from the present to the future is input. explain. The control device 10 according to each embodiment described below is, for example, an edge device such as a PLC.

The control device 10 according to each embodiment described below has a control amount y indicating a target value time series {r (t)} which is time series data of an arbitrary target value, a state of the controlled target plant 20, and a controlled target. The operation amount u for the controlled plant 20 is calculated based on the plant response model of the plant 20, and the model parameters for model update are estimated. Then, the control device 10 measures the controlled variable y of the controlled variable plant 20 according to the manipulated variable u, and based on the target value time series {r (t)}, the controlled variable y, the plant response model, and the like, the following The manipulated variable u is calculated and new model parameters are estimated. As described above, the control device 10 according to each embodiment described below calculates the operation amount u for making the control amount y follow the target value time series {r (t)}, and the model parameters of the plant response model. Is repeatedly executed during online execution (that is, during control of the controlled plant 20).

The control device 10 according to each embodiment described below realizes high control performance of model prediction control, but does not require online optimization calculation every time, so that it can be realized with a small amount of calculation and a small amount of memory. At the same time, it becomes possible to estimate the model parameters of the plant response model. The control device 10 according to each embodiment can estimate the parameters of the plant response model by using, for example, the operation amount and the control amount of the other existing control devices. Therefore, for example, immediately after the installation of the control device 10, the initial model parameters of the plant response model are estimated using the operation amount and the control amount of the other control device, and after the start of operation, seasonal fluctuations and aging of the controlled plant 20 are performed. It is possible to estimate model parameters that follow changes over time such as deterioration.

The control amount y includes, for example, the temperature of the controlled plant 20, and the target value r includes, for example, the set temperature. However, the control amount y and the target value r are not limited to the temperature and the set temperature, and an arbitrary control amount in the controlled plant 20 and a target target value of the control amount can be used.

[First Embodiment]
<Configuration of control device 10>
First, the configuration of the control device 10 according to the present embodiment will be described with reference to FIG. FIG. 1 is a diagram showing an example of the configuration of the control device 10 according to the first embodiment.

As shown in FIG. 1, the control device 10 according to the present embodiment includes a model parameter estimation unit 101, a target value look-ahead unit 102, a measurement unit 103, a difference device 104, an operation amount update unit 105, and a timer 106. And have. Each of these functional units is realized, for example, by a process in which one or more programs installed in the control device 10 are executed by a CPU or the like.

When the target value time series {r (t)} and the look-ahead length T _{p are input for each predetermined control cycle T c} , the target value look-ahead unit 102 sets _{the time t + T p} after the look-ahead length from the current time t. The target value r (t + T _p ) is output. The look-ahead length T _p is the time length for determining the _{read-ahead target value r (t + T p} ) in the target value time series {r (t)}. Hereinafter, the target value r (t + T _p ) is also referred to as a “look-ahead target value r (t + T _p )”.

The measuring unit 103 measures the controlled variable y of the controlled plant 20 for each _{control cycle T c.} Then, the measurement unit 103 outputs the latest value of the measured controlled variable y as the _{current controlled variable y 0.} The controlled variable y of the controlled plant 20 is determined according to the manipulated variable u and the disturbance v. Examples of the disturbance v include a decrease or increase in the outside air temperature when the control amount y is a temperature.

Further, the measurement unit 103 acquires the operation amount u output from the operation amount update unit 105, and outputs the latest value of the acquired operation amount u as the operation amount current value u ₀ .

The difference device 104 outputs the difference (deviation) between the read-ahead target value r (t + T _p ) output from the target value look-ahead unit 102 and the control amount current value y _{0 as the target deviation e 0} . The target deviation e ₀ is _{calculated by e 0} = r (t + T _p ) −y ₀ (t). Hereinafter, the target deviation e _{0 is also referred} to as “look-ahead target deviation e ₀ ”.

The operation amount update unit 105 outputs the operation amount u for the controlled plant 20 for each _{control cycle T c.} The operation amount update unit 105 includes a look-ahead response correction unit 111, an operation change amount calculation unit 112, and an adder 113.

The look-ahead response correction unit 111 is _{time-series data of the plant response function {S θ} (t)}, the look-ahead target deviation e ₀ (t), the look-ahead length T _p, and the change amount du of the past manipulated variable u. Based on the operation change amount time series {du (t)}, the correction target deviation e ^* _{(t) corrected for the look-ahead target deviation e 0} (t) is calculated. The details of the calculation method of the correction target deviation e ^{* (t) will be described later.} Here, the plant response function {S _θ (t)} is a function including the model parameter θ, and is a plant response model of the controlled plant 20. Hereinafter, the change amount du of the operation amount u is also referred to as the “operation change amount du”.

The operation change amount calculation unit 112 calculates the operation change amount du based on ^{the correction target deviation e *} (t) calculated by the look-ahead response correction unit 111 for each _{control cycle T c.} The operation change amount calculation unit 112 calculates and outputs the operation change amount du (t) in the order of, for example, du (t-3T _c ), du (t-2T _c ), and du (t-T _c). The operation change amount du is an amount in which the operation amount u changes for each _{control cycle T c.}

The adder 113 adds the operation amount current value u ₀ output from the measurement unit 103 and the operation change amount du output from the operation change amount calculation unit 112 to calculate a new operation amount u. Then, the adder 113 outputs this manipulated variable u to the controlled plant 20. This manipulated variable u is calculated by u (t) = u ₀ + du (t) = u (t−T _c ) + du (t).

The timer 106 operates the target value look-ahead unit 102 and the measurement unit 103 for each _{control cycle T c.} Since the target value look-ahead unit 102 and the measurement unit 103 _{operate in each control cycle T c} , the model parameter estimation unit 101 and the manipulated variable update unit 105 also operate in each _{control cycle T c.}

When the control amount current value y ₀ , the operation amount current value u _0, and the model parameter θ set in the current plant response function {S _θ (t)} are input to the model parameter estimation unit 101, _{The estimated value θ est} of the model parameter of the plant response function {S _θ (t)} is calculated and output. In the following, the model parameter θ set in the current plant response function {S _θ (t)} is also referred to as “model parameter current value θ”, and the model parameter of the plant response function {S _θ (t)}. The estimated value θ _{est is also referred} to as “model parameter estimated value θ _est ”.

<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. 2 is a diagram for explaining an example of the operation of _{the plant response function {S θ (t)}.}

As shown in FIG. 2, the plant response function {S _θ (t)} receives a unit step response S at time t after the lapse of time t from the initial time 0 when the model parameter set value θ and the time t are input. _{Output θ} (t). The unit step response is a response when the manipulated variable u is used as a unit step input (that is, the output of the plant response model of the controlled plant 20).

<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. 3 is a flowchart for explaining an example of the calculation process of the plant response function {S _{θ (t)}.} In FIG. 3, it is assumed that the model parameter set value θ and the time t are input.

Here, in the present embodiment, _{as a calculation model of the plant response function S θ} (t), ARMA (autoregressive) using the autoregressive of the past N points for the controlled variable y and the moving average of the past M points for the manipulated variable u. The case where the moving average) model is adopted will be described. Note that N and M are set in advance by, for example, a user or an administrator of the control device 10 (hereinafter, also referred to as “user or the like”).

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

とする。モデルパラメータ設定値θの各要素は数２で示す、ＡＲＭＡモデルの係数である。

And. Each element of the model parameter setting value θ is a coefficient of the ARMA model shown by Equation 2.

ここで、ｋはインデックスである。

Here, k is an index.

Step S11: The manipulated variable update unit 105 initializes the time τ and the state vector φ (0), where the variable representing the time is τ and the state vector at the index k is φ (k). Further, the manipulated variable update unit 105 initializes the index k to k = 0.

Here, the state vector φ (k) is

と表される。

It is expressed as.

The operation amount update unit 105 initializes, for example, τ = 0, and at the same time,

と初期化する。すなわち、φ（０）はｕ（ｋ−１）に相当する要素のみ１、他の要素は０に初期化される。これは、単位ステップ応答を模擬する際の初期値として設定していることを意味する。

And initialize. That is, φ (0) is initialized to 1 only for the element corresponding to u (k-1), and 0 for the other elements. This means that it is set as the initial value when simulating the unit step response.

Step S12: Next, the manipulated variable update unit 105 calculates the controlled variable predicted value y (k) based on the model parameter set value θ and the state vector φ (k). The manipulated variable update unit 105 calculates the controlled variable predicted value y (k) by, for example, y (k) = φ (k) ^Τθ. Here, Τ represents transpose.

Step S13: Next, the manipulated variable update unit 105 updates the state vector φ (k + 1) at the next index (that is, k + 1) based on the controlled variable predicted value y (k) and the state vector φ (k). To do. The operation amount update unit 105 updates the state vector φ (k + 1) as follows, for example.

ここで、ｙ（ｋ）には上記のステップＳ１２で計算した制御量予測値を設定し、ｕ（ｋ）には１を設定する。また、ｙ（ｋ−１），・・・，ｙ（ｋ−Ｎ＋１）及びｕ（ｋ−１），・・・，ｕ（ｋ−Ｍ＋１）には、状態ベクトルφ（ｋ）と同じ値を設定する（すなわち、状態ベクトルφ（ｋ）に含まれるｙ（ｋ−１），・・・，ｙ（ｋ−Ｎ＋１）及びｕ（ｋ−１），・・・，ｕ（ｋ−Ｍ＋１）をそれぞれ設定する。）。

Here, the control amount predicted value calculated in step S12 above is set in y (k), and 1 is set in u (k). Further, y (k-1), ..., y (k-N + 1) and u (k-1), ..., U (k-M + 1) have the same values as the state vector φ (k). Set (that is, y (k-1), ..., y (k-N + 1) and u (k-1), ..., U (k-M + 1) included in the state vector φ (k). Set each.).

Step S14: Next, the manipulated variable update unit 105 updates the time τ to τ + ΔT and the index k to k + 1. Here, ΔT represents the time width of one step of the ARMA model. ΔT can be arbitrarily set according to the time constant of the controlled plant 20, but it is preferable that the _{ΔT is the same as the control cycle Tc, for example.}

Step S15: Next, the manipulated variable update unit 105 determines whether or not τ ≧ t. Then, when it is not determined that τ ≧ t (NO in step S15), the manipulated variable update unit 105 returns to step S12. As a result, steps S12 to S14 are repeatedly executed until τ ≧ t.

On the other hand, when it is determined that τ ≧ t (YES in step S15), the manipulated variable update unit 105 ends the process. As a result, the finally calculated y (k) is _{obtained as the unit step response S θ} (t) (that is, S _θ (t) = y (k) is the step response function {S _θ (t)}. It is calculated as a unit step response at time t).

<Estimation of model parameter θ>
Next, a process of estimating the model parameter θ of the plant response function {S _{θ (t)} (that is, a process of calculating the model parameter estimated value θ est} ) will be described with reference to FIG. FIG. 4 is a flowchart for explaining an example of the estimation process of the model parameter θ. Note that FIG. 4 describes a case where the _{model parameter estimated value θ est} at a certain index k is calculated assuming that the controlled variable current value y ₀ , the manipulated variable current value u _0, and the model parameter set value θ are input. To do. The index k is a value independent of the index k in the calculation of the plant response function, and represents the run-time index of the model parameter estimation process.

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), an initialization instruction is given by the user or the like. For example.

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

Step S22: When it is determined in step S21 above that the model parameter estimated value θ _est (k) and the covariance matrix P (k) are initialized, the model parameter estimation unit 101 determines that the model parameter estimated value θ _est (0) = θ _0. And P (0) = I, and the run-time index k of the model estimation process is initialized to k = 1. Here, as θ ₀ , the model parameter setting value θ already set in the plant response function {S _θ (t)} may be used, or the initial value assumed in advance by the user or the like may be used. However, a fixed initial value such as a vector in which all elements are 0 may be used. Further, I may be an identity matrix or an arbitrary matrix determined in advance.

Step S23: The model parameter estimation unit 101 calculates a 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. The model parameter estimated value θ _est (k-1) is an estimated value of the model parameter θ estimated last time (that is, at the time of k-1). Further, the state vector φ (k) is the state vector updated last time (that is, the state vector obtained in step S13 of FIG. 3 when k-1 or the state obtained in step S27 described later when k-1). Vector).

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

y (k) = y ₀
ε (k) = y (k) −φ (k) ^Τ θ _est (k-1)
Step S24: Next, the model parameter estimation unit 101 updates the covariance matrix P (k). The model parameter estimation unit 101 updates the covariance matrix P (k) by, for example, the following equation (2).

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。

Here, P (k-1) is the covariance matrix obtained last time (that is, at the time of k-1).

Step S25: 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} is not updated until a predetermined period elapses from the time of initialization in step S22 above. , When an update instruction is given by a user or the like.

When it is determined that the model parameter estimated value θ _{est is} to be updated (YES in step S25), the model parameter estimation unit 101 proceeds to step S26. On the other hand, if it is not determined that the model parameter estimated value θ _{est is} to be updated (NO in step S25), the model parameter estimation unit 101 proceeds to step S27 without executing step S26.

Step S26: The model parameter estimation unit 101 _{updates the model parameter estimated value θ est} (k). _{The model parameter estimation unit 101 updates the model parameter estimation value θ est} (k) by, for example, the following equation (3).

ステップＳ２７：モデルパラメータ推定部１０１は、制御量現在値ｙ_０と操作量現在値ｕ_０と状態ベクトルφ（ｋ）とに基づいて、次のインデックス（つまり、ｋ＋１）における状態ベクトルφ（ｋ＋１）を更新する。モデルパラメータ推定部１０１は、例えば、以下により状態ベクトルφ（ｋ＋１）を更新する。

Step S27: The model parameter estimation unit 101 determines the state vector φ (k + 1) at the next index (that is, k + 1) based on _{the controlled variable current value y 0} , the manipulated variable current value u _{0, and the state vector φ (k).} To update. The model parameter estimation unit 101 updates the state vector φ (k + 1) as follows, for example.

ここで、ｙ（ｋ）には制御量現在値ｙ_０を設定し、ｕ（ｋ）には操作量現在値ｕ_０を設定する。また、ｙ（ｋ−１），・・・，ｙ（ｋ−Ｎ＋１）及びｕ（ｋ−１），・・・，ｕ（ｋ−Ｍ＋１）には、状態ベクトルφ（ｋ）と同じ値を設定する（すなわち、状態ベクトルφ（ｋ）に含まれるｙ（ｋ−１），・・・，ｙ（ｋ−Ｎ＋１）及びｕ（ｋ−１），・・・，ｕ（ｋ−Ｍ＋１）をそれぞれ設定する。）。このように、共分散行列の更新とモデルパラメータ推定値θ_ｅｓｔの算出とは逐次最小２乗法に基づく処理で実現される。ただし、逐次最小２乗法は一例であって、例えば、カルマンフィルタ法が用いられてもよい。

Here, the control amount current value y ₀ is set in y (k), and the operation amount current value u ₀ is set in u (k). Further, y (k-1), ..., y (k-N + 1) and u (k-1), ..., U (k-M + 1) have the same values as the state vector φ (k). Set (that is, y (k-1), ..., y (k-N + 1) and u (k-1), ..., U (k-M + 1) included in the state vector φ (k). Set each.). As described above, the update of the covariance matrix and _{the calculation of the model parameter estimated value θ est} are realized by the processing based on the sequential least squares method. However, the sequential least squares method is an example, and for example, the Kalman filter method may be used.

As another example of updating the covariance matrix P (k) in step S24, the model parameter estimation unit 101 calculates the above equation (2) and then uses the following equation (4) to calculate the covariance matrix. P (k) may be updated.

ここで、０≦α≦１である。また、ｄｉａｇ（｛Ｐ_ｉｉ（ｋ）｝_ｉ）は行列Ｐ（ｋ）の対角成分のみで構成される行列（非対角成分以外は０）を表す。

Here, 0 ≦ α ≦ 1. Further, diag ({P _ii (k)} _i ) represents a matrix composed of only diagonal components of the matrix P (k) (0 except for the off-diagonal components).

The above α is a diagonalization adjustment parameter, and the closer it is to 1, the greater the effect of reducing the absolute value of the off-diagonal component of the covariance matrix P (k). That is, the above equation (4) means that the correction for reducing the absolute value of the off-diagonal component of the covariance matrix P (k) is performed. With such a correction, it is possible to stabilize the calculation of the covariance matrix P (k) (that is, the calculation of the covariance matrix P (k) can be converged faster and the calculation accuracy is improved. It is possible).

Step S28: The model parameter estimation unit 101 updates the run-time index k of the model parameter estimation process to k + 1.

<Operation of target value look-ahead unit 102>
Next, the operation of the target value look-ahead unit 102 will be described with reference to FIG. FIG. 5 is a diagram for explaining an example of the operation of the target value look-ahead unit 102.

As shown in FIG. 5, when the target value time series {r (t)} and the look-ahead length T _{p are input, the target value look-ahead unit 102 has a look-ahead target at the time t + T p} after the look-ahead length from the current time t. The value r (t + T _p ) is output. In this way, the target value look-ahead unit 102 outputs the target value r (t + T _{p ) at the time t + T p} after the look-ahead length in the target value time series {r (t)}.

The example shown in FIG. 5 shows a case where the target value time series {r (t)} is represented by a straight line, but the present invention is not limited to this. The target value time series {r (t)} may be represented by an arbitrary curve, rectangle, or the like. In particular, the target value time series {r (t)} may be represented by a curve that changes periodically according to the time t. Further, the target value time series {r (t)} may be set in advance, or future target values may be updated at any time. Further, as the future target value, for example, the current target value may be used assuming that the current target value r (t) is constant and continuous, or the current target value is assumed to change at a constant speed. The value r (t) may be a target value at which _{the time t + T p after the look-ahead length is reached.}

<Operation of look-ahead response correction unit 111>
Next, the operation of the look-ahead response correction unit 111 will be described with reference to FIG. FIG. 6 is a diagram (No. 1) for explaining an example of the operation of the look-ahead response correction unit 111.

As shown in FIG. 6, the look-ahead response correction unit 111 includes a plant response function {S _θ (t)}, a look-ahead target deviation e ₀ (t), a look-ahead length T _p, and an operation change amount time series {du ( When t)} is input, first, the _{value predicted that the current control amount y 0} changes after T _p due to the past operation change amount du is calculated as the look-ahead response correction value y _n (t). For example, when the current time is t, the past operation change amount du is expressed as du (t-T _c ), du (t-2T _c ), or the like.

Next, the look-ahead response correction unit 111 calculates a correction target deviation e ^* _{(t) obtained by correcting the look-ahead target deviation e 0} (t) with the look-ahead response correction amount y _n (t), and the calculated correction target deviation e ^*. (T) is output. As shown in FIG. 6, the correction target deviation e ^* (t) is e ^* (t) = r (t + T _p ) − (y ₀ (t) + y _n (t)) = e ₀ (t) −y _n. It is represented by (t).

Hereinafter, a method of calculating the look-ahead response correction value y _n (t) will be described. Predicted values of the controlled variable y (predicted read-ahead response value) y _{n, A} (t) at the _{look-ahead time t + T p} only by the operation change amount time series {du (t)}

とする。ここで、Ｍ´は先読み応答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。

And. Here, M'is the length (model interval) of the model used for calculating the read-ahead response predicted value.

_{Further, the predicted values (free response predicted values) y n, B} (t) of the controlled variable y at the current time t based only on the operation change amount time series {du (t)} are set.

とする。ここで、上記と同様に、Ｍ´は自由応答予測値の計算に使用するモデル区間である。

And. Here, similarly to the above, M'is a model interval used for calculating the free response predicted value.

Then, the look-ahead response correction unit 111 sets the difference between the look-ahead response prediction values y _{n, A} (t) and the free response prediction values y _{n, B} (t) as the look-ahead response correction value y _n (t). That is, the look-ahead response correction unit 111 has y _n (t) = y _{n, A} (t) −y _{n, B} (t).

Further, as an example, a case where the look-ahead response correction unit 111 uses the prediction time-series storage unit 121 for storing the prediction time series _{to calculate the look-ahead response correction value y n} (t) will be described with reference to FIG. 7. .. FIG. 7 is a diagram (No. 2) for explaining an example of the operation of the look-ahead response correction unit 111. The prediction time series storage unit 121 can be realized by using, for example, a storage device such as an auxiliary storage device or a RAM (Random Access Memory).

As shown in FIG. 7, the predicted time series storage unit 121 has _{predicted values y n, C} (t−Δt | t) of the controlled variable y _{from the time t−Δt to the future time t + T b, where t is the current time.} ), Y _{n, C} (t | t), y _{n, C} (t + Δt | t), ..., y _{n, C} (t + T _b | t) are stored as a time series. Note that T _b is a constant that determines the length (time length) of the _{predicted values y n and C} stored in the predicted time series storage unit 121 _{, and T b} = N'Δt (N'is an arbitrary positive value. Integer). Further, Δt is a prediction interval, and it is assumed that _{Δt = T c.} Hereinafter, the predicted values y _{n and C} are also referred to as “generalized predicted values y _{n and C} ”.

_{The generalized predicted values y n, C} (s | t) at the time s predicted at the time t are

で算出される。

It is calculated by.

_{When the generalized predicted values y n and C} stored in the predicted time series storage unit 121 are used, the above-mentioned free response predicted values y _{n and B} (t) become generalized predicted values y _{n and C} (t | t). They agree, and the read-ahead response predicted values y _{n, A} (t) described above match the generalized predicted values y _{n, C} (t + T _p | t).

Therefore, the look-ahead response correction unit 111 uses the generalized predicted values y _{n, C} stored in the predicted time series storage unit 121, and y _n (t) = y _{n, A} (t) −y _{n, B} ( The look-ahead response correction value y _n (t) can be calculated by t) = y _{n, C} (t + T _p | t) −y _{n, C (t | t).} Thereby, the correction target deviation e ^* (t) = e ₀ (t) -y _n (t) can be calculated and output. In this way, by using the prediction time series storage unit 121, the look-ahead response correction unit 111 can calculate the correction target deviation e ^* with a small amount of calculation and a small amount of memory.

Further, the prediction time series storage unit 121 is updated by the prediction time series update unit 122 every time the operation change amount du (t) is input to the look-ahead response correction unit 111. _{Here, the generalized predicted values y n, C} (t + m'Δt | t) at m'Δt ahead predicted at time t, where m'is a natural number, are

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

It is represented by. _{Therefore, the generalized predicted values y n, C} (t + Δt + m'Δt | t + Δt) at the m'Δt destination predicted at the time t + Δt are

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

Will be. _{That is, the generalized predicted values y n, C} (t + Δt + m'Δt | t + Δt) at the m'Δt destination predicted at the time t + Δt _{are y n, C} (t + Δt + m'Δt | t + Δt) = S _θ (m'Δt) du ( t + Δt) + y _{n, C} (t + (m ′ + 1) Δt | t).

Therefore, Generalized Predictive value _{y n} at M'derutati destination predicted at time _{t + Δt, C (t +} Δt + m'Δt | t + Δt) is predicted at time t (m'+ 1) Δt destination Generalized Predictive value _{y n, C} Is updated as a value to which the influence of the operation change amount du (t + Δt) at the time t + Δt is added.

<Update process executed by the predicted time series update unit 122>
Here, a process in which the prediction time series update unit 122 updates the data held in the prediction time series storage unit 121 will be described with reference to FIG. FIG. 8 is a flowchart for explaining an example of the update process executed by the prediction time series update unit 122. The following describes the case of updating Generalized Predictive value y _n at time t stored in the predicted time series storage unit _121, a _C, Generalized Predictive value y _n at time t + Delta] _t, the _C.

_{Step S41: The predicted time series update unit 122 shifts the generalized predicted values y n and C} stored in the predicted time series storage unit 121 at the time t by the time Δt.

For example, as shown in FIG. 9A, y _{n, C} (t−Δ | t), y _{n, C} (t | t), y _{n, C} (t + Δ | t), ..., Y _{n , C} (t + T _p | t), ..., y _{n, C} (t + T _b | t) are stored in the prediction time series storage unit 121, and their relative positions m'are -1, 0, 1, 1, respectively. ..., N'. In this case, the prediction time series update unit 122 sets the relative position m'of the generalized prediction values y _{n and C} of the relative position m'≥ 0 to -1. That is, the predicted time series updating unit 122, the relative position m'= a _{0 y n, C (t |} t) to the relative position m'= -1, a relative position m'= 1 _{y n,} C ( t + Δt | t) is sequentially shifted to the relative position m'= 0. After that, similarly, the prediction time series update unit 122 _{shifts each y n , C} (t + T _b | t) at the relative position m ′ = N ′ to the relative position m ′ = N ′ -1. _{, C} relative positions m'are shifted in order.

Step S42: The predicted time series update unit 122 updates the generalized predicted values y _{n and C} at the final time (that is, the relative position m ′ = N ′). _{The generalized predicted values y n, C} (t + N'Δt + Δt | t) at the final time t + N'Δt + Δt predicted at the time t can be, for example, the generalized predicted values y _{n, C} before the time t + N'Δt.

により推定する。これは、時刻ｔでの時刻ｔ＋Ｎ´Δｔ−（ｊ−１）Δｔにおける一般化予測値ｙ_ｎ，Ｃに対して重みｗ_ｊを掛けて和を取った式である。

Estimated by. This is an equation obtained by multiplying the _{generalized predicted values y n and C} at the time t + N'Δt− (j-1) Δt at the time t by the weight w _{j and summing them.}

For the generalized predicted values y _{n, C} (t + N'Δt + Δt | t), for example, when the generalized predicted values y _{n, C} of the time t + N'Δt predicted at the time t are used as they are, that is, y. _{It is also conceivable that n, C} (t + N'Δt + Δt | t) = y _{n, C} (t + N'Δt | t). Further, for example, in consideration of the velocity change, the Generalized Predictive value _{y n, C} at time t + N'Δt, time t + N'Δt-Δt Generalized Predictive value _{y n of} the _C extrapolated,

を用いる場合も考えられる。例えば、この数１６で、ｒ＝１とすれば、時刻ｔ＋Ｎ´Δｔにおける一般化予測値ｙ_ｎ，Ｃと、時刻ｔ＋Ｎ´Δｔ−Δｔにおける一般化予測値ｙ_ｎ，Ｃとの差がΔｔだけ継続すると推定することに相当する。

It is also possible to use. For example, in recent 16, if r = 1, the generalized predictive value _{y n} at time t + _N'Δt, and _C, the time t + N'Δt-Δt Generalized Predictive value _{y n} at _the difference between _C only Delta] t Equivalent to presuming to continue.

As a result, as shown in FIG. 9B, the relative position N'of the predicted time series storage unit 121 _{is updated to y n, C} (t + T _b + Δt | t) = y _{n, C} (t + N'Δt + Δt | t). Will be done.

Step S43: The prediction time series update unit 122 reflects the influence of the latest operation change amount du (t + Δt). That is, as shown in FIG. 9 (c), the predicted time series update unit 122 has the generalized predicted values y _{n, C} (t + (m'+ 1) Δt | t] from the relative positions m'= 0 to N'. ), S _θ (m'Δt) du (t + Δt) is added.

_{Specifically, S θ} (0) du (t + Δt) is added to _{the generalized predicted values y n, C} (t + Δt | t) at the relative position m ′ = 0. _{Further, S θ} (Δt) du (t + Δt) is added to _{the generalized predicted values y n, C} (t + 2Δt | t) at the relative position m ′ = 1. The same applies when m'≧ 3.

Step S44: The predicted time series update unit 122 sets the generalized predicted values y _{n and C} stored in the predicted time series storage unit 121 as the predicted time series at time t + Δt. That is, according to the above S43, each generalized predicted value y _{n, C} from _{m'= 0 to N'is y n, C} (t + Δt + m'Δt | t + Δt) = S _θ (m'Δt) du (t + Δt). It can be expressed as + y _{n, C} (t + (m'+ 1) Δt | t).

Therefore, if t'= t + Δt, as shown in FIG. 9D, the generalized predicted values of _{the relative position m'= -1 are y n, C} (t'−Δt | t'), and the relative position m. The generalized predicted value of ′ = 0 can be expressed as y _{n, C} (t ′ | t ′), and the generalized predicted value of the relative position m ′ = 1 _{can be expressed as y n, C} (t ′ + Δt | t ′). Similarly, when m'≧ 2, it _{can be expressed as y n, C} (t'+ m'Δt | t'). _{As a result, the generalized predicted values y n and C} stored in the predicted time series storage unit 121 at time t are updated to the generalized predicted values y _{n and C at time t + Δt.}

<Operation of operation change amount calculation unit 112>
Next, the operation of the operation change amount calculation unit 112 will be described with reference to FIG. FIG. 10 is a diagram for explaining an example of the operation of the operation change amount calculation unit 112.

As shown in FIG. 10, when the correction target deviation e ^* (t) is input ^{, the operation change amount calculation unit 112 operates by multiplying the correction target deviation e *} (t) by a predetermined control gain. The amount of change du (t) is calculated, and the calculated du (t) is output. For example, when the integrated gain k _I is used as the predetermined control gain, the operation change amount du (t) is _{calculated by du (t) = k I} × e ^* (t).

However, when the correction target deviation e ^* (t) is multiplied by a predetermined control gain and the upper limit value du _max of the operation change amount du is exceeded, the operation change amount calculation unit 112 sets the du _max to the operation change amount du. Let it be (t). Similarly, when the correction target deviation e ^* (t) is multiplied by a predetermined control gain and the result is less _{than the lower limit value du min} of the operation change amount du, the operation change amount calculation unit 112 sets the du _min as the operation change amount. Let it be du (t).

<Hardware configuration of control device 10>
Next, the hardware configuration of the control device 10 according to the embodiment will be described with reference to FIG. FIG. 11 is a diagram showing an example of the hardware configuration of the control device 10 according to the embodiment.

As shown in FIG. 11, the control device 10 according to the embodiment includes an input device 201, a display device 202, an external I / F 203, a communication I / F 204, a ROM (Read Only Memory) 205, and a RAM 206. , CPU 207 and auxiliary storage device 208. Each of these hardware is connected to each other by a bus 209 so as to be able to communicate with each other.

The input device 201 is, for example, various buttons, a touch panel, a keyboard, a mouse, or the like, and is used to input various operations to the control device 10. The display device 202 is, for example, a display or the like, and displays various processing results by the control device 10. The control device 10 does not have to have at least one of the input device 201 and the display device 202.

The external I / F 203 is an interface with an external device. The external device includes a recording medium 203a and the like. The control device 10 can read or write the recording medium 203a via the external I / F 203. The recording medium 203a includes, for example, an SD memory card, a USB memory, a CD (Compact Disk), a DVD (Digital Versatile Disk), and the like. One or more programs that realize each functional unit of the control device 10 may be stored in the recording medium 203a.

The communication I / F 204 is an interface for the control device 10 to perform data communication with another device. One or more programs that realize each functional unit of the control device 10 may be acquired (downloaded) from a predetermined server or the like via the communication I / F 204.

The ROM 205 is a non-volatile semiconductor memory that can retain data even when the power is turned off. The RAM 206 is a volatile semiconductor memory that temporarily holds programs and data. The CPU 207 is an arithmetic unit that reads programs and data from the auxiliary storage device 208 or ROM 205 into the RAM 206 and executes various processes.

The auxiliary storage device 208 is, for example, an HDD (Hard Disk Drive), an SSD (Solid State Drive), or the like, and is a non-volatile memory for storing programs and data. The programs and data stored in the auxiliary storage device 208 include, for example, one or more programs that realize each functional unit of the control device 10, an OS (Operating System) that is basic software, and various applications that operate on the OS. There are programs etc.

The control device 10 according to the embodiment can realize the various processes described above by having the hardware configuration shown in FIG. Note that FIG. 11 shows a hardware configuration example in which the control device 10 is realized by one computer, but the present invention is not limited to this, and the control device 10 may be realized by a plurality of computers. ..

[Second Embodiment]
Next, the second embodiment will be described. In the second embodiment, a case where the control amount is estimated using the _{model parameter estimated value θ est will be described.} As a result, it is possible to obtain an estimated value of the controlled variable when the model parameter estimated value θ _est is set in the plant response function {S _θ (t)}, so that, for example, the user or the like compares it with the actual controlled variable. This makes it possible to determine whether or not this model parameter estimated value θ _est should be set in the plant response function {S _{θ (t)}.} In other words, the user or the like can easily visually judge the quality _{of the current plant response function {S θ} (t)} by comparing the estimated value of the control amount with the actual control amount. become.

In the second embodiment, the differences from the first embodiment will be mainly described, and the description of the components substantially the same as those of the first embodiment will be omitted or simplified.

<Configuration of control device 10>
First, the configuration of the control device 10 according to the present embodiment will be described with reference to FIG. FIG. 12 is a diagram showing an example of the configuration of the control device 10 according to the second embodiment.

As shown in FIG. 12, the control device 10 according to the present embodiment further includes a control amount estimation unit 107. The control amount estimation unit 107 is realized, for example, by a process in which one or more programs installed in the control device 10 are executed by a CPU or the like.

The control quantity estimation unit 107 is based on the model parameter estimation value θ _est output by the model parameter estimation unit 101, the control quantity current value y _0, and the operation quantity current value u _0, and the control quantity estimation value y _est. Is calculated and output. Further, the control amount estimation unit 107 initializes the calculation (specifically, the initialization of the state vector φ (k)) according to the value of the estimation initialization flag. In the following, it represents the control amount of the estimated value _{y est} to as "control amount estimation value _{y est".}

<Estimation of control amount>
Next, processing for estimating the control quantity (i.e., the process of calculating the control amount estimation value y _est), is described with reference to FIG. 13. FIG. 13 is a flowchart for explaining an example of the control amount estimation process. In FIG. 14, the case where the control quantity estimate y _est at a certain index k is calculated assuming that the model parameter estimated value θ _est , the control quantity current value y _0, and the operation quantity current value u _{0 are input.} explain. The index k is a value independent of the index k in the calculation of the plant response function, and represents a run-time index of the control amount estimation process.

Step S51: The control amount estimation unit 107 determines whether or not the estimation initialization flag is ON. When the estimation initialization flag is ON, the state vector φ (k + 1) is initialized without estimating the control amount. ON or OFF of the estimation initialization flag may be appropriately set by, for example, a user or the like.

When it is determined that the estimation initialization flag is ON (YES in step S51), the control amount estimation unit 107 proceeds to step S52. On the other hand, if it is not determined that the estimation initialization flag is ON (NO in step S51), the control amount estimation unit 107 proceeds to step S53.

Step S52: When it is determined in step S51 that the estimation initialization flag is ON, the control amount estimation unit 107 initializes the state vector φ (k + 1) and ends the process. The control amount estimation unit 107 initializes the state vector φ (k + 1) as follows, for example.

ここで、ｙ（ｋ）には制御量現在値ｙ_０を設定し、ｕ（ｋ）には操作量現在値ｕ_０を設定する。また、ｙ（ｋ−１），・・・，ｙ（ｋ−Ｎ＋１）及びｕ（ｋ−１），・・・，ｕ（ｋ−Ｍ＋１）には、状態ベクトルφ（ｋ）と同じ値を設定する（すなわち、状態ベクトルφ（ｋ）に含まれるｙ（ｋ−１），・・・，ｙ（ｋ−Ｎ＋１）及びｕ（ｋ−１），・・・，ｕ（ｋ−Ｍ＋１）をそれぞれ設定する。）。

Here, the control amount current value y ₀ is set in y (k), and the operation amount current value u ₀ is set in u (k). Further, y (k-1), ..., y (k-N + 1) and u (k-1), ..., U (k-M + 1) have the same values as the state vector φ (k). Set (that is, y (k-1), ..., y (k-N + 1) and u (k-1), ..., U (k-M + 1) included in the state vector φ (k). Set each.).

When the state in which the estimation initialization flag is ON continues for a certain period of time, each element of the state vector φ (k + 1) is subjected to the control amount and the operation amount (that is, the control amount) measured by the measurement unit 103 in the above step S52. It is initialized to the observed value of the observed value and the observed value of the manipulated variable.

Step S53: When it is not determined in step S51 that the estimation initialization flag is ON, the control amount estimation unit 107 estimates _{the control amount based on the model parameter estimated value θ est} and the state vector φ (k). Calculate the value _yest (k). The control quantity estimation unit 107 calculates the control quantity estimation value y _est (k) by _{, for example, y est} (k) = φ (k) ^Τ θ _est.

Step S54: Next, the controlled variable estimation unit 107 is based on the controlled variable estimated value y _est (k) calculated in step S53 above, the manipulated variable current value u _0, and the state vector φ (k). , Update the state vector φ (k + 1) at the next index (ie, k + 1). The control amount estimation unit 107 updates the state vector φ (k + 1) as follows, for example.

ここで、ｙ（ｋ）には制御量推定値ｙ_ｅｓｔ（ｋ）を設定し、ｕ（ｋ）には操作量現在値ｕ_０を設定する。また、ｙ（ｋ−１），・・・，ｙ（ｋ−Ｎ＋１）及びｕ（ｋ−１），・・・，ｕ（ｋ−Ｍ＋１）には、状態ベクトルφ（ｋ）と同じ値を設定する。このように、制御量推定値ｙ_ｅｓｔ（ｋ）を用いて状態ベクトルφ（ｋ＋１）を更新することで、上記のステップＳ５３での制御量推定値の計算を、制御量の観測値及び操作量の観測値のうちの操作量の観測値のみから行うことが可能となる（つまり、制御量の観測値を用いずに制御量推定値を計算することが可能となる。）。

Here, to set the y (k) controlling the amount of the estimated value _{y est} (k) is the, the u (k) setting the manipulated variable current value _{u 0.} Further, y (k-1), ..., y (k-N + 1) and u (k-1), ..., U (k-M + 1) have the same values as the state vector φ (k). Set. In this way, by updating the state vector φ (k + 1) using the controlled variable estimated value _yest (k), the calculation of the controlled variable estimated value in step S53 described above can be performed by observing and manipulating the controlled variable. It is possible to perform the operation only from the observed value of the manipulated variable among the observed values of (that is, it is possible to calculate the estimated value of the controlled variable without using the observed value of the controlled variable).

Step S55: The control amount estimation unit 107 updates the run-time index k of the control amount estimation process to k + 1.

[Third Embodiment]
Next, the third embodiment will be described. In the third embodiment, a case where the model parameter set value θ set in the plant response function {S _θ (t)} is updated by the model parameter estimated value θ _{est will be described.}

In the third embodiment, the differences from the first embodiment will be mainly described, and the description of the components substantially the same as those of the first embodiment will be omitted or simplified.

<Configuration of control device 10>
The configuration of the control device 10 according to the present embodiment will be described with reference to FIG. FIG. 14 is a diagram showing an example of the configuration of the control device 10 according to the third embodiment.

As shown in FIG. 14, the control device 10 according to the present embodiment further includes a model parameter update unit 108. The model parameter update unit 108 is realized, for example, by a process in which one or more programs installed in the control device 10 are executed by a CPU or the like.

The model parameter update unit 108 sets the model parameter estimate value θ _est output by the model parameter estimation unit 101 in the plant response function {S _θ (t)} according to the value of the model update trigger. For example, when the model update trigger is ON, the model parameter update unit 108 sets the model parameter estimated value θ _est in the plant response function {S _θ (t)} (that is, the model parameter estimated value θ _est is set as the model parameter. The set value is θ). For example, when the model update trigger is OFF, the model parameter update unit 108 does nothing.

As a result, when the user or the like determines that control can be performed with higher accuracy by setting the _{model parameter estimated value θ est} in the plant response function {S _{θ (t)}, the model update trigger is used.} By turning on, it is possible to realize more accurate control.

[Fourth Embodiment]
Next, a fourth embodiment will be described. In the fourth embodiment, a case where the model parameter estimated value θ _est is calculated in consideration of the disturbance v will be described.

In the fourth embodiment, the differences from the first embodiment will be mainly described, and the description of the components substantially the same as those of the first embodiment will be omitted or simplified.

<Configuration of control device 10>
The configuration of the control device 10 according to the present embodiment will be described with reference to FIG. FIG. 15 is a diagram showing an example of the configuration of the control device 10 according to the fourth embodiment.

As shown in FIG. 15, the configuration of the control device 10 according to the present embodiment is the same as that of the first embodiment, but the functions of the model parameter estimation unit 101 and the measurement unit 103 are mainly different.

The measurement unit 103 according to the present embodiment also measures the disturbance v and outputs the measured disturbance v to the model parameter estimation unit 101. The disturbance v measured by the measuring unit 103 is also referred to as "observation disturbance v".

Further, the model parameter estimation unit 101 according to the present embodiment receives the _{input of the control amount current value y 0} , the operation amount current value u ₀ , the observation disturbance v, and the model parameter set value θ, and then the model parameter The estimated value θ _est is calculated and output. Here, when the observation disturbance v is used, for example, an ARMAX model (autoregressive moving average model with exogenous variables) may be adopted as the calculation model of _{the plant response function S θ (t).} As an example, when the autoregressive of the past N points is used for the controlled variable y, the moving average of the past M points is used for the manipulated variable u, and the disturbance term of the past L points is used for the observed disturbance v, the plant response function S _θ (t). ) Is expressed by the following equation (5). Note that L is set in advance by, for example, a user of the control device 10.

なお、モデルパラメータ設定値θは、

The model parameter setting value θ is

と表される。

It is expressed as.

Therefore, the state vector φ (k)

と読み替えることで、モデルパラメータ推定部１０１は、図４のステップＳ２１〜ステップＳ２７によりモデルパラメータ推定値θ_ｅｓｔを算出することができる。ただし、上記のステップＳ２７で状態ベクトルφ（ｋ＋１）を更新する際には、ｖ（ｋ）には観測外乱ｖを設定する。これにより、外乱ｖも考慮したモデルパラメータθを推定することが可能となる。

By reading as, the model parameter estimation unit 101 can calculate the _{model parameter estimation value θ est in steps S21 to S27 of FIG.} However, when updating the state vector φ (k + 1) in step S27 above, the observation disturbance v is set in v (k). This makes it possible to estimate the model parameter θ in consideration of the disturbance v.

[Example 1]
Next, Example 1 will be described. In the first embodiment, a case where the controlled plant 20 is controlled by the control device 10 according to the second embodiment will be described.

First, the true plant response is y (t) = _{−a 1} y (t−ΔT) −a ₂ y (n-2ΔT) + b ₁ u (n−ΔT) + b ₂ u (n), where ΔT is the time mesh. It shall be represented by -2ΔT) + δ (t). Here, δ (t) represents noise.

If ΔT = T _c , the plant response function {S _θ (t)} becomes a quadratic ARMA model, and y (k) = _{−a 1} y (k-1) −a ₂ y (k-2) + b. _{It can be expressed as 1} y (k-1) + b ₂ y (k-2).

At this time, the state vector is

と表せる。

Can be expressed as.

Model parameter estimates

と表せる。

Can be expressed as.

The true value of the model parameters is

と表せる。また、制御量推定値は、ｙ_ｅｓｔ（ｋ）＝φ（ｋ）^Τθ（ｋ−１）と表すことができる。

Can be expressed as. Further, the control amount estimated value _{can be expressed as yest} (k) = φ (k) ^Τθ (k-1).

In the first embodiment, the true value of the model parameter is

であるものとする。このときの制御対象プラント２０のステップ応答を図１６に示す。図１６に示すように、実施例１における制御対象プラント２０のステップ応答は、滑らかに増加する形状となっている。

Suppose that The step response of the controlled plant 20 at this time is shown in FIG. As shown in FIG. 16, the step response of the controlled plant 20 in the first embodiment has a shape that smoothly increases.

In the first embodiment, the controlled plant 20 is operated in batch, and as shown in FIG. 17, three batches (batch 1, batch 2, batch 3) are assumed to be continuous. At this time, in the first embodiment, as shown in FIG. 17, the covariance matrix is initialized and the model parameters are initialized at _{time t 0 before the start of operation.}

The initial values of the model parameters are

とした。

And said.

The initial value of the covariance matrix is

とした。

And said.

Further, the diagonalization adjustment parameter was set to α = 0.7.

In the first embodiment, as shown in FIG. 17, the control device 10 is operated. That is, after the start of the batch, the update of the covariance matrix is continuously performed, and at the time t ₁ when the batch 1 is started and the covariance matrix is updated to some extent, the model parameter estimation value update is turned ON (that is,). By setting YES in step S25 of FIG. 4), the model parameter estimated value is continuously updated thereafter. Further, some degree model parameter estimates are updated, the ON estimation initialization flag at t ₂ the batch 1 is completed, and initializes the state vector. After that, the estimation initialization flag is turned off, and the control amount of batch 2 is estimated during batch 2. Similarly, the estimated initialization flag to ON at the time point t ₃ when batch 2 has been completed, to initialize the state vector. After that, the estimation initialization flag is turned off, and the control amount of batch 3 is estimated during batch 3.

At this time, it is assumed that the manipulated variable u and the controlled variable y are as shown in FIGS. 18 (a) and 18 (b), respectively. That is, in the first embodiment, it is assumed that the control amount y fluctuates according to the operation amount u, and the control amount y has one peak. In this case, the model parameter estimated value θ _est and the controlled variable estimated value y _est are as shown in FIG. The upper part of FIG. 19 is the model parameter estimated value θ _est , and the lower part is the controlled variable estimated value y _est . Note that θ ₁ ^* , θ ₂ ^* , θ ₃ ^*, and θ ₄ ^* are _{the true value of θ 1} , the true value of θ ₂ , the true value of θ ₃ , and the true value of θ ₄ , respectively.

≪Batch 1≫
First, the covariance matrix is updated from time to time 0 to 50, and the model parameter estimated value is not updated. The update of the model parameter estimated value is started from time 50, but the value of the model parameter estimated value changes during batch 1 and settles down to almost the true value near time 230 near the end of batch 1. After the end of batch 1, the estimation initialization flag is turned ON to initialize the state vector.

≪Batch 2≫
The model parameter estimates continue to be updated, but are stable at values that are close to constant. Further, when the control amount estimated value _yest calculated by the control amount estimation unit 107 and the control amount y (that is, the observed value of the control amount) measured by the measurement unit 103 are compared, the peak positions are almost the same. However, the controlled variable estimated value _yest is slightly higher than the observed value y.

≪Batch 3≫
The model parameter estimates continue to be updated, but are stable at values that are close to constant. The amount of fluctuation is smaller than that of batch 2. Further, when the control amount estimated value _yest calculated by the control amount estimation unit 107 and the control amount y (observed value y) measured by the measurement unit 103 are compared, the peak positions are substantially the same. Compared with batch 2, the _{yest is slightly lower than the observed value y at the peak position, but the controlled variable estimated value yest} and the observed value y match well at the latter half of the time 200 to 230. It can be said that it is.

[Example 2]
Next, Example 2 will be described. In the second embodiment, the manipulated variable u and the controlled variable y are different from those of the first embodiment, and other than that, they are the same as those of the first embodiment. In the second embodiment, it is assumed that the manipulated variable u and the controlled variable y are shown in FIGS. 20 (a) and 20 (b), respectively. That is, in the second embodiment, it is assumed that the control amount y fluctuates according to the operation amount u, and the control amount y has two peaks. In this case, the model parameter estimated value θ _est and the controlled variable estimated value y _est are as shown in FIG. The upper part of FIG. 21 is the model parameter estimated value θ _est , and the lower part is the controlled variable estimated value y _est . Note that θ ₁ ^* , θ ₂ ^* , θ ₃ ^*, and θ ₄ ^* are _{the true value of θ 1} , the true value of θ ₂ , the true value of θ ₃ , and the true value of θ ₄ , respectively.

≪Batch 1≫
First, the covariance matrix is updated from time to time 0 to 50, and the model parameter estimated value is not updated. The update of the model parameter estimated value is started from time 50, but the value of the model parameter estimated value changes during batch 1 and settles down to almost the true value near time 230 near the end of batch 1. After the end of batch 1, the estimation initialization flag is turned ON to initialize the state vector.

≪Batch 2≫
The model parameter estimates continue to be updated, but are stable at values that are close to constant. Further, when the control amount estimated value _yest calculated by the control amount estimation unit 107 and the control amount y measured by the measurement unit 103 are compared, the values are almost close to each other, although there is a slight deviation at the peak position. ing.

≪Batch 3≫
The model parameter estimates continue to be updated, but are stable at values that are close to constant. The amount of fluctuation is smaller than that of batch 2. Further, when the control amount estimated value _yest calculated by the control amount estimation unit 107 and the control amount y (observed value y) measured by the measurement unit 103 are compared, although there is a slight deviation at the peak position, The values are almost close.

The present invention is not limited to the above-described embodiment disclosed specifically, and various modifications, changes, combinations, and the like can be made without departing from the scope of claims.

10 Control device 20 Control target plant 101 Model parameter estimation unit 102 Target value look-ahead unit 103 Measuring unit 104 Difference device 105 Operation amount update unit 106 Timer 111 Look-ahead response correction unit 112 Operation change amount calculation unit 113 Adder

Claims (10)

A control device that outputs an operation amount for a controlled object and causes the controlled amount of the controlled object to follow a target value.
When the target value time series, which is the time series of the target value, and the look-ahead length indicating the time width for pre-reading the target value are input, the look-ahead among the plurality of target values included in the target value time series is described. A target value look-ahead means for acquiring a look-ahead target value indicating a long-term target value, and
A look-ahead target deviation calculating means for calculating a look-ahead target deviation, which is the difference between the look-ahead target value and the current control amount of the controlled object,
After the look-ahead length, based on the plant response function, which is a function representing the plant response model to be controlled and includes model parameters, and the amount of change in the operation amount in the past up to the present. A correction target deviation calculation means for calculating a correction target deviation indicating a difference between the predicted value of the control amount of the above and the look-ahead target value, and
An operation amount calculation means for calculating a new operation amount based on the correction target deviation, and
A model parameter estimation means that calculates an estimated value of a model parameter included in the plant response function based on the controlled variable and the manipulated variable.
Have a,
Acquisition of the look-ahead target value by the target value look-ahead means, calculation of the look-ahead target deviation by the look-ahead target deviation calculation means, calculation of the correction target deviation by the correction target deviation calculation means, and operation amount calculation means. A control device characterized in that the calculation of a new operation amount and the calculation of an estimated value of the model parameter by the model parameter estimation means are sequentially repeated for each control cycle. Based on the estimated value calculated by the model parameter estimation means and the manipulated variable, with the control amount at the time when the flag taking either the ON or OFF value is turned ON as the initial value, after the above time point. Control quantity estimation means, which calculates the estimated value of the control quantity of
The control device according to claim 1, wherein the control device comprises. The control amount estimation means is
The estimated value of the controlled variable is calculated using the estimated value calculated by the model parameter estimating means, and the estimated value of the controlled variable is calculated using the model parameter included in the plant response function.
The control device according to claim 2. When the value of the update trigger is ON based on the update trigger that takes either an ON or OFF value, the estimated value calculated by the model parameter estimation means is used as a parameter included in the plant response function. Model parameter update means to set,
The control device according to any one of claims 1 to 3, wherein the control device has. The model parameter estimation means
Based on the controlled variable, the manipulated variable, and the measured value of the disturbance with respect to the controlled object, the estimated value of the model parameter included in the plant response function is calculated.
The control device according to any one of claims 1 to 4, wherein the control device is characterized by the above. The plant response function is represented by an ARMA model or a step response model calculated from the ARMAX model.
The model parameter is a coefficient of the ARMA model or ARMAX model.
Any one of claims 1 to 5, wherein the predicted value of the control amount after the look-ahead length is calculated by calculating the step response at the time point after the look-ahead length by the ARMA model or the ARMAX model. The control device described in. The model parameter estimation means
The covariance matrix is updated and the estimated values of the model parameters are calculated by the sequential least squares method or the Kalman filter method.
The control device according to any one of claims 1 to 6, wherein the control device is characterized by the above. The model parameter estimation means
In updating the covariance matrix, a correction process is performed to reduce the absolute value of the off-diagonal terms of the covariance matrix.
The control device according to claim 7. A computer that outputs the manipulated variable for the controlled object and causes the controlled variable of the controlled object to follow the target value.
When the target value time series, which is the time series of the target value, and the look-ahead length indicating the time width for pre-reading the target value are input, the look-ahead among the plurality of target values included in the target value time series is described. A target value look-ahead procedure for acquiring a look-ahead target value that indicates a long-term target value, and
A look-ahead target deviation calculation procedure for calculating a look-ahead target deviation, which is the difference between the look-ahead target value and the current control amount of the control target, and
After the look-ahead length, based on the plant response function, which is a function representing the plant response model to be controlled and includes model parameters, and the amount of change in the operation amount in the past up to the present. The correction target deviation calculation procedure for calculating the correction target deviation indicating the difference between the predicted value of the control amount and the look-ahead target value, and
An operation amount calculation procedure for calculating a new operation amount based on the correction target deviation, and
A model parameter estimation procedure for calculating an estimated value of a model parameter included in the plant response function based on the controlled variable and the manipulated variable.
The execution,
According to the acquisition of the look-ahead target value by the target value look-ahead procedure, the calculation of the look-ahead target deviation by the look-ahead target deviation calculation procedure, the calculation of the correction target deviation by the correction target deviation calculation procedure, and the operation amount calculation procedure. A control method characterized in that a new calculation of the operation amount and a calculation of an estimated value of the model parameter by the model parameter estimation procedure are sequentially repeated for each control cycle. A computer that outputs the manipulated variable for the controlled object and causes the controlled variable of the controlled object to follow the target value.
When the target value time series, which is the time series of the target value, and the look-ahead length indicating the time width for pre-reading the target value are input, the look-ahead among the plurality of target values included in the target value time series is described. Pre-reading target value indicating the long-term target value Target value pre-reading means,
A look-ahead target deviation calculating means for calculating a look-ahead target deviation, which is the difference between the look-ahead target value and the current control amount of the controlled object.
After the look-ahead length, based on the plant response function, which is a function representing the plant response model to be controlled and includes model parameters, and the amount of change in the operation amount in the past up to the present. Correction target deviation calculation means for calculating the correction target deviation indicating the difference between the predicted value of the control amount and the look-ahead target value.
An operation amount calculation means for calculating a new operation amount based on the correction target deviation,
A model parameter estimation means that calculates an estimated value of a model parameter included in the plant response function based on the controlled variable and the manipulated variable.
To function as,
Acquisition of the look-ahead target value by the target value look-ahead means, calculation of the look-ahead target deviation by the look-ahead target deviation calculation means, calculation of the correction target deviation by the correction target deviation calculation means, and operation amount calculation means. A program that sequentially repeats the calculation of the new operation amount and the calculation of the estimated value of the model parameter by the model parameter estimation means for each control cycle.

Images