JPS62263502A - Process control device - Google Patents

Abstract

PURPOSE:To accurately presume a coifficient included in a process model by outputting the presuming arithmetic value of the new coefficient corresponding to a measuring data column by a presuming arithmetic part only when the determinant of a detecting matrix is larger than the prescribed value. CONSTITUTION:A detecting part 15 inputs measuring data colums [y(t+N)-y (t)-y(t-N)] and [u0(t+N-M)-u0(t)-u0(t-M)] obtained out of the input output of a process from a storage part 11, calculates a square root matrix D by using a data reliability detecting matrix M formed from this and executes the reliability detection of measuring data. When the reliability of the measuring data is judged to be satisfactory, the measuring data column (phic, etc.,) is used and an unknown parameter theta is presumed and calculated in a presuming arithme tic part 112. When the measuring data column is decided to the defective data, another new measuring data are fetched and continuously, testing is executed. during this, the output of the unknown parameter theta of the presuming arithmetic part 12 holds the previous determining value.

Description

Detailed Description of the Invention 3. 51j III Description of the Invention (Industrial Application Field) The present invention relates to improvement of control characteristics of a process control device equipped with a III constant function of coefficients included in a process model. It is.

(Prior Art) Examples of the present invention will be described in detail below with reference to the drawings. First of all, we will explain the basic principles of the conventional process system 12II equipment.
1. The mathematical model of the process is expressed by the following equation, where u (
t) is the 7082 person horn at time t, y(1) is the output of the same process, ai and bJ are the weight system.(1
) In the equation, q-1 is a delay parameter, and when the equation (1) is expanded and expressed, the following equation is obtained.

Ylj)=at y(t-1> az y(t-
2)--aNV(t-N)+bou(t)+bIu(
t 1 ) +・−+bMtJ (t M) −
(2) That is, the current process output y(t> is the past process output sequence V (t-1), V (t-2),
....

y (t-N) and known input sequences u(t), u(t-1)
, . . . , u(tM).

u < t) such that the output of the process expressed by the above mathematical model is the desired output y"(t) is determined as follows. In other words, the desired output y of the process output y(t) "Deviation from (t) e (t) - y (t) - y"
(t)...(3) squared e2(t)=(y(t)-y" (te)2...(4
) is obtained using the least squares method using the following equation.

... (5) If the coefficient aI + bJ + bO is known, (5)
Although the intended purpose can be achieved using uo(j> in the equation, in reality the coefficient aL*bJ*bo is often unknown and needs to be estimated.

- the unknown coefficients ai, t)j + b6 are estimated as follows. From equation (2), each time 1.1-←1゜・
..., the following equation holds true at t+N.

V(t)=-at y(t-1) a2y(t 2
>--aNy (t-N)+bou u (t)+b+
u(t -1> 10b2 u (t-2) 10-+bM
u (t-M) ...(
6o) y(t+1>-al y(t) a 2
y (t-1)−-ar<V(t-N+1)+bo
u (t+1)+b, u (t)+b2u (t-1
)+・+bHu(t−M+1)・
...(61)V (t+N) --at V
(t-1+N) -82V(t2+N>-a
Ny(t)+bou(t+N)+b, LJ(t-
1+N)+b2 Turtle J(t-2+N) +------
)bM u (t-M+N) ・
(6N>If the above equations (6o) to (6N) are expressed in matrix form, the following equations are obtained.

Y-Φθ...(7) However, Φ-...(9). Here, the evaluation function Je = (Y-0 go) (Y-0 go)...(11)
(However, AT does not indicate the transposed matrix to the matrix) The estimated implantation of θ that minimizes is △ e= l)” @)−1(1)T Y
−(12) (where Δ゛I accepts the inverse matrix of matrix A).

In equation (5), the coefficient aζ + bJ + bO is
C △ C Key point of e found by equation (12) 1a+ + bj+ bOr
If replaced, the optimal operation amount Oo (j) can be found using the following formula: 1
('' at = 0. (0) is an appropriate initial estimate orthogonal i
(0), Intimidation J (0), Go. Determine using (0).

FIG. 5 is a block diagram showing a conventional self-tuning process controller that implements the above-mentioned control algorithm. The solid line shows the flow of online information, and the dash-dotted line shows the flow of offline information.

1 is a process controlled by the process b 2 IQ UHa, and 2 is a process controlled by this computer loss control III device 1. In the process control device 1, 11 is a storage section for storing input/output measurement data of the process 2, and 12 is this storage section 1.
The coefficients included in the mathematical model of process 2 are calculated based on the measured data string of process 1.
13 is a JFT constant operation unit that outputs a constant output, and 13 outputs the desired output according to the specification input (according to the desired process). 5); 14 is the output of this reference model determining unit 13;

This is a manipulated variable calculation unit which inputs the coefficient output J'3J of the estimation calculation unit 12 and the measurement data of the input and output of the process 2, performs the operation lu, and outputs it to the process.

The operation of each part of the process control 8 having the above configuration will be explained below. Measurement data uo of input and output of block 1 and process 2
+V is stored in the measurement data storage section 11 as a badge data, and the estimation calculation section 12 performs the calculation of equation (12) using the measurement data string in the measurement data storage section 11 to estimate the unknown coefficient e. Desired output y” (usually gQ constant value)
is determined by the reference model determination unit 13 according to the specification input, and the operation amount calculation unit 14 calculates the operation Fik LJ using equation (13) from the desired output y8 and the coefficient output table of the estimation calculation unit 12. o is determined and output to process 2. This cycle is repeated thereafter. However, the calculation by the estimation calculation section 12 is executed only when the operating state changes.

(Problem to be Solved by the Invention) However, in the process control 8i installation as described above, the measurement data string [y(shi -← N
) ~ y(t) ~ V(t-N>], [
u(wo-tN-M) ~u (t) ~u (t-M)
] becomes lower, the estimation error of unknown parameter θ is 0−e−θ (14), and uo Ij) becomes correct LJ Q”(
1) by ΔuO(t).

uo (t) This ΔUo(j) naturally deteriorates controllability. FIG. 6 is a time chart showing the process response when the controllability of the process control device deteriorates in this way.

The response A is the operation fJ' due to the estimation error of the unknown parameter θ
When the l iQ difference Δ( ) is barely within the stable range of the closed loop system, the response is when the manipulated variable error Δ( ) is out of the stable range [11] due to a large estimation error.

The present invention has been made in order to solve the above-mentioned problems, and aims to improve the Luri control characteristics by using a bus control 1A trade equipped with a function of estimating coefficients included in a process model. The purpose is to

(Measures IIQ, B and Actions for Solving Problems) The process control device according to the present invention includes a storage section for storing process input/output measurement data, and inputs the measurement data string from this storage section and performs the measurement. a verification section that compares the value of the determinant IMI of a verification matrix M composed of a data string with a predetermined value to determine the reliability of the measurement data string; and nFt measurement data output from the storage section. column/) s r et al.; and an estimation calculation part that estimates the coefficients included in the mathematical I del of the process, and only when the determinant IM+ of the test matrix M is larger than a predetermined value, the estimation calculation part It is calculated and outputted that the configuration is configured to output a new estimated distilled value of the coefficient corresponding to the measured data string.

(Embodiments of the Invention) Before entering into the description of the embodiments, the basic principle thereof will first be explained. As explained in the prior art section, the unknown parameter θ is estimated using equation (12).

Union (Φ7Φ) - 薯Φ"Y ... (12) (
Equation 12) simply means that ΦTY is υj by Φ7Φ. Therefore, if ΦTΦ is a small linear matrix, the estimation accuracy of e will be greatly degraded. This will be illustrated with the following simple example. Consider a case where the output y is given by the following equation based on the now known input X.

y=ax+b (16) However, a and b are unknown parameters. (16) is converted into (7
) is expressed in the same format as the following equation.

(17) However, x(1) and x(2) are input data at time 1.2, and V(1) and y(2) are the same output data. At this time, (Φ7) -1 becomes the following equation.

(Φ1Φ)' = (x (1) −x (2) )−
2-...(18) Therefore, from equations (18) and (12), the parameters θ-(a,
b) Estimation of 'FrJ degree is min/n(7)(x(1)-×
(2)) It can be seen that when 2 is a very small value of 1 or less, No. 13 is buried in noise, resulting in poor S/N ratio and deterioration. That is, the estimation accuracy improves as the input data x(1> and x(2) are farther apart from each other, such that X(1)-X(2) becomes larger).

Extending the above idea in general, the measurement data y(tt>
, y(ij), LJ(if), LJ(j
j ) etc. is large, the estimation accuracy of the unknown parameter e improves. FIG. 3 is an explanatory diagram showing specifically which time domain data should be used from the output response waveform. For t1≦t≦ts, V(tt
>, V(tt), and V(is) have very good variations. However, when t>t3, the 1st shift of y(1) hardly changes. Therefore, in order to accurately estimate the unknown parameter θ, it is necessary to use the measurement data sequence [V, LJ] where t1≦t≦15.

In order to express the above concept quantitatively, we define 11 test columns M as shown in the following equation.

M-ΦTΦ... (one) If the serial expression of this real symmetric matrix M is thicker than the value 9 specified in advance in the specifications, it is determined that the measured data string forming Φ is good.t Nawara IMI≧9...(20) is the judgment condition, and the measured data IJ that does not satisfy this condition
is picked up.

(19) (20> In order to perform the test with high accuracy using the formula as is, it is necessary to use a computer with a crystal n length of about 32 bits, but if the following modification is carried out, it is 16 bits P.
i! An equivalent test can be performed using an ia ri machine with a finite word length. First, a real symmetric matrix M is decomposed into square root rows tlJD as follows.

M-DD” (21) Here, D is a lower triangular matrix, and 1]7 is an upper triangular matrix.

Using this square root matrix, a determination operation similar to equation (20) is performed as follows. That is, when +01≧S (22) (however, S7FT), the measured data is good; otherwise, it is considered bad data. The above square root matrix 1] is the following proviso +=
1.2. ..., N←(M+1)Dl, i) 10 However, j<i Full D(i, k)) However, j=i+1. i+2. ..., N+ (M+1
>... (24) Fig. 1 is a diagram illustrating an embodiment of a process control device according to the present invention for executing the above-described process control. The same parts as those in the apparatus shown in FIG. 5 are arranged in the same manner, and a description thereof will be omitted. In the process control device 10, 15
is a test section which inputs the measurement data output from the storage section 11, performs a test, and outputs the result to the estimation performance section 12 according to the result.

The operation of the process control device configured as described above is verified by the verification unit 1.
The explanation will focus on 5. Other parts I'' are the same as in the case of the apparatus shown in FIG. FIG. 2 is a flowchart for explaining the operation of the apparatus shown in FIG.

The verification unit 15 generates a measurement data string [y (t+N) ~ y (t) ~ y (t-N)], [1
-1o (t+N-M)~uo (j)~U.

(t-M)] from the storage unit 11, and using the data reliability test matrix (also simply called test matrix) M of equation (21) created from this, (23>(24) equation 13 days square 1
fi 11 columns are calculated, and the reliability of the measured data is tested using equation (22). Measured data when formula (22) is satisfied is judged to have good reliability, and the next stage estimation calculation unit 12 uses this (11 constant data sequence (ΦC etc.)
2) Estimating the unknown parameter O based on the formula. (
If the equation (22) is not satisfied, the measured data string is determined to be defective data and is discarded, new measurement data is taken in, and verification is continued until the equation (22) is satisfied. During this time, the unknown parameter output of the estimation calculation unit 12 retains the previously determined value. Below, the optimal operation 5 is performed from the unknown parameter θ output as in the device shown in Figure 5.
1 u O is calculated, and the optimal operation that realizes the index of resource saving, energy saving, etc. given as the specification input is i-? )
.

FIG. 4 is a time chart showing the process response when the above process control device is used. As is clear from the figure, the optimal operation FJi U o obtained from the unknown parameter 0, which is determined with good accuracy <+n, results in a good response waveform C of 1!7.

According to the process control device configured as described above, unknown parameters can be estimated with high accuracy, so the optimum operation amount can be determined accurately, and resource saving, energy saving operation, etc. can be easily realized.

Data verification also ensures highly reliable process operation at all times.

In the above example, the measurement data sequence [y(t+N) ~ y(t) ~ y(t-N>] used for data verification is obtained by removing high-frequency noise using a low-pass filter or the like.
After filter processing [y(a-tN) ~ y(t) ~ y (
tN)] may also be used.

Further, in the above embodiment, data verification and parameter HfI determination are performed by off-line 4r batch data processing, but data verification and parameter m determination may also be performed online using input/output data 9+1.

In addition, in the above embodiment, the measurement data was verified using unknown parameters) in order to increase the reliability of the Ni constant.
The present invention is not limited to this, and can also be applied to, for example, plant abnormality diagnosis.

(Effects of the Invention) As described above, according to the present invention, it is possible to estimate the coefficients included in the process model with high accuracy, and to realize a process-controlled txt+ device with improved control characteristics.

[Brief explanation of drawings]

FIG. 1 is a block configuration diagram showing an example of a one-cusp hC of the process control device according to the present invention, FIG. 2 is a flowchart for explaining the operation of Sjα in FIG. An explanatory diagram for explaining the operation, FIG. 4 is a time tsade without showing an example of the response of the device in FIG. 1, FIG. 5 is a configuration block diagram showing a conventional process control MW, and FIG. This is a timeline showing an example of a response. 2... Process, 10... Process 1 control device, 1
DESCRIPTION OF SYMBOLS 1... Storage part, 12... Estimation calculation part, 15... Verification part, O... Estimation calculation value.

Claims (3)

[Claims] (1) A storage unit that stores input/output measurement data of the process, and inputs the measurement data string from this storage unit and sets the value of the determinant |M| of the verification matrix M composed of the measurement data string to a predetermined value. a testing unit that determines the reliability of the measured data string by comparing it with the measured data string; and an estimation calculation unit that estimates coefficients included in the mathematical model of the process from the measured data string output from the storage unit, The method is characterized in that the estimation calculation section outputs a new estimated calculation value of the coefficient corresponding to the measurement data sequence only when the determinant |M| of the matrix M is larger than a predetermined value. Process control equipment. (2) The test section is the determinant of the square root matrix D of the test matrix M |D
2. The process control device according to claim 1, wherein the reliability of the measurement data is determined by comparing | with a predetermined value. (3) The process control device according to claim 1, which calculates and outputs the manipulated variable based on the coefficient output, process input/output measurement data, and setting data output from the estimation calculation section.