JPH0264802A - Optimum parameter adjustment device for controller - Google Patents

Abstract

PURPOSE:To eliminate the needs for cutting the closed loop of a real machine and to eliminate a dynamic characteristic test by setting an evaluation function which can directly evaluate control performance in the state of a normal operation time and obtaining a parameter in which the value of the evaluation function becomes minimum. CONSTITUTION:The evaluation function value Jp of a plant 63 is calculated from the ratio of both dispersion values of control deviations epsilon and (t), and known disturbances (load command values) (w) and (t), and the evaluation function value of a model, in which an independent variable is set to be the proportional gain of a PI controller 61 and a set rate is obtained by using the algorithm of a dynamic neutral learning model. When the deviation (v) of the evaluation function value JM of the model and the evaluation function value Jp of the plant is ¦v¦<=delta (delta is a pre scribed value), the parameter (proportional gain, reset and the like) which can make the evaluation function value JM of the model minimum is obtained by an optimization method. Here, the values are outputted as optimum parameters x1* and X2*, and they are used for correcting the set point of the PI controller 61. Thus, the closed loop of the real machine is prevented from being cut and the dynamic characteristic test is also eliminated.

Description

Detailed Description of the Invention [Industrial Field of Application] The present invention is directed to an optimal noyarameter a1 for a controller used in a control device for thermal norocess of a female ylanolant, nuclear plant, etc.
This relates to a four-way adjustment device.

[Conventional technology]

FIG. 6 shows a meter adjustment device 64 that is an optimal example of a conventional controller.
It is a configuration diagram when connected to an actual machine consisting of a PI controller 6, a control switch 62, and a Gra/Toro 3. Figure 6 PI
When adjusting the 6th and 9th parameter of the regulator, after switching from A (automatic) to M (manual) with the full switch (automatic and manual switching) 62, input the step manual force u(t). , control my(ri) of Grantro 3 is recorded, and the parameters of P■, J, !1 section 6) are determined based on the response waveform. 64 indicates a conventional adjustment device. Examples of this method include the Ziegler-Nichols method that can be found in textbooks, and the method described in Japanese Patent Publication No. 58001/1983.

[Problem to be solved by the invention]

In the conventional method described above, a dynamic characteristic test for optimum parameter adjustment must be performed after the PI sea bream switch 61 is disconnected from the closed loop system, which may be dangerous in terms of plant operation. In addition, it is necessary to replace the response waveform obtained as a result of the dynamic characteristic test with a mathematical model in some way, but the approximation error at that time greatly affects the determination of the meter, and it is not possible to obtain optimal control performance. Sometimes I can't.

Therefore, the present invention provides a PI controller for /? In order to determine the parameter, there is no need to cut the closed-loop system of the actual machine, there is no need to perform a dynamic characteristic test without adjustment, and the response waveform obtained as a result of the dynamic characteristic test is put into a mathematical model in some way. Optimal controller that does not require repair - motherboard meter 14
The purpose is to provide a complete set up.

[Means to solve the problem]

In order to achieve the above object, the present invention provides control deviations and known disturbances (load command values) of a plant controlled by a PI controller.
, as well as the current setting value (the setting value is updated) of the P114 section device, and determines the optimum setting value of the PI controller, the control having the following functions: Optimal f parameter adjustment device for equipment.

(1) Plant evaluation function creation function Calculates the variance values of both the control deviation and known disturbance in a specified time range, and calculates the plant evaluation function JP from the ratio.

(2) Model evaluation function creation function S: Independent variables are
The structure of the evaluation function JM of the model, which is the proportional gain and reset rate of the I% node, is determined in advance, and the evaluation function JM of the model is determined using only the coefficients in the model using the algorithm of the dynamic neural learning model. .

(3) Find the deviation between the evaluation function JP and the JMO value, and if the absolute value of the deviation is equal to or smaller than the specified value, the evaluation function JM of the model can identify the evaluation function J of the plant. As a result, we will move on to the dirameter search function using the following optimization method. If not, return to item (2) and perform learning again.

(4) Parameter search function using optimization method: Find parameters (proportional gain, reset rate) that can minimize the model evaluation function value JM using optimization methods such as nonlinear programming, and use those values as the optimal parameters. Output.

It is.

[Effect]

According to the optimal parameter adjustment device for a controller according to the present invention, /4' parameter: J! It is possible to obtain information for adjustment from measured values during normal operation without conducting dynamic characteristic tests for adjustment, and also to set an evaluation function that can directly evaluate control performance under normal operation conditions. Since we used an optimization method to find the parameter that minimizes its value, the effectiveness of the obtained maternal meter is excellent.

〔Example〕

Embodiments of the present invention will be described below with reference to the drawings. Figure 1 shows the optimal parameter adjustment device itl and PI of the invention.
Since it consists of the adjuster 61 and the grandeur 3, the dynamic characteristic test for adjustment distortion is not necessary in FIG. 1 as in FIG. 6. Instead, jVi (load command value) W(t) of known disturbance during normal operation, control deviation 1(1) and PIBI
Current No. III of the four-section device 61 (proportional gain x1 (tj)
, reset rate X2 (t j)) must be taken into the adjustment device 1.

Note that the outputs of the adjusting device 1,640 in FIGS. 1 and 6 are both the 1&appropriate parameter value of the PI adjuster 6. Also,
tj is a discrete time.

The internal functions of the adjustment device f1 of the present invention are as shown in FIG. This is the control deviation ε(1) of the plant 63 controlled by the PT controller 61 and the current setting value (
This device is capable of determining the optimal setting value of the PI controller 6 while incorporating rC (setting value is updated), and has the following functions.

Plant evaluation function creation function unit 2: Calculate the variance values of both the control deviation C (Ri and known disturbance w(t) and use the ratio to calculate the evaluation function of the plant 63, 1st shift JpCJ) in the specified time range
Calculate.

Model evaluation function creation function unit 5: The construction of the model evaluation function value JM (tj) with the independent variables as the proportional gain of the PI controller 6 and the reset rate is determined in advance, and only the coefficients in the model are determined. The evaluation function value JmCtj) of the model is determined using the Alco 0 rhythm of the dynamic neural learning model. Specifically, as a comparison result of the comparator 4, which will be described later, if the absolute value of the deviation v(tj) of the subtractor 3 is larger than the specified value δ, the deviation v(J) and the current parameter x1(t
j) and x2(J) to calculate the model evaluation function value JM(tj
).

Subtractor 3: Finds the deviation v(J) between the evaluation function value JM(tj) of the model and the evaluation function value Jp(tj) of the plant.

Comparator 4: If the absolute value of the deviation v(J) from $, calculator 3 is equal to or smaller than the specified value δ, the evaluation function value JM(tj) of the model is equal to the evaluation function value JpDj of the plant. ) can be identified, the parameter search mechanism part 6 using the optimization method described below using the following optimization method.
Move to. If not, the process returns to the model evaluation function creation function section 5 described above and is trained again.

Parameter search function unit 6 using an optimization method: When the comparison result of the comparator 4 is 1v (δ in November), nonlinear programming is performed to find the parameters (proportional gain, reset rate) that can minimize the model evaluation function value Jm (tj). The value is determined by an optimization method such as the method, and the value is output as the optimal Norameter: 'C4''*X2''.The outputted optimal Norameter x1
``+X2 umbrella is used to correct the setting value of PI controller 6.

Next, the functions of each part shown in FIG. 2 will be specifically explained.

First, as shown in FIG. 3, the plant evaluation function creation function unit 2 calculates the variance λ2(tj) of the control deviation ε(1) and the known disturbance between J-1 and tj, that is, during L time. W
(A similar variance value σ2(tj) is obtained, and the evaluation function Jp(tj) of the plant is obtained using the following equation.

Jp(tj)”λ2(tj)/σ(tj)
(1) Next, the model evaluation function generator 1tg5 calculates the model evaluation function value JM(jj) using the following equation.

Jm(J)=Σαt(J)−yt(J)
1211=1 Here, v(t'') =Jp(tj) Jig(tj)
(41αi (tj) = αt(t
j-1)+yt(tj)v(tj) and "i p 1"
"L 21"'16 +! 51.

FIG. 4 shows the above relational expression in a block diagram.

Therefore, J, J+1. As the process progresses, the learning process converges to the following equation, which is independent of time.

Therefore, by substituting a time-independent (method relationship) to yi on the right side of equation (6), it can be rearranged into equation (8).

3N(tj) and X2(tj) are the proportional gains and reset rates of the PI regulator 6 between J-1 and 210.

Jm=f (Xl * X2)
+82 Next, A? by optimization method? In the scillameter search function 6, 1. xl for which rM is the minimum,
x2 is determined by an optimization method such as nonlinear programming.

As a result, the obtained values are approximately optimal parameters.

In addition, f("11X2) and its optimal nrammeter value X+
The relationship of "+X2" is shown in FIG.

According to the embodiments of the present invention described above, the following effects can be obtained.

(2) There is no need to cut the closed loop system of the actual machine in order to determine the parameters of the PI controller 6, and there is no need to perform a dynamic characteristic test for fctA adjustment.

■ The evaluation function of the plant 63 is approximated by the function shown in equation (2). As a result, the evaluation function is 1 and the regulator is 2
Since the relationship between the ramimes can be clarified and the weight α1(tj) in equation (2) is automatically determined by the circuit shown in Figure 4, the identification accuracy can be improved when modeling the evaluation function of the plant 63. It will be planned.

〔Effect of the invention〕

According to the present invention, in order to determine the seven cell parameters of the P114j45 device, there is no need to force the closed loop system of the actual machine, there is no need to perform a dynamic characteristic test for adjustment, and there is no need to perform a dynamic characteristic test for adjustment. Accordingly, it is possible to provide an optimal parameter adjustment device for a controller that does not require any method to replace the response waveform obtained by the calculation into a mathematical model.

[Brief explanation of the drawing]

FIG. 1 is a diagram showing a situation in which the optimal parameter adjustment device for a controller according to the present invention is connected to an actual machine, and FIG. 2 explains the internal functions of the optimal parameter adjustment device f1 for a controller according to the present invention shown in FIG. Figure 3 is a diagram showing a specific example of the measured values of the actual machine required by the plant evaluation function creation function unit 2 in Figure 2, and Figure 4 is the model evaluation function creation function unit in Figure 2. 5
and a specific circuit diagram of the subtracter 3, and FIG. 5 shows the function fCx1. x2), and FIG. 6 is a connection relationship diagram between a conventional controller optimum parameter adjustment device and an actual machine. l...Optimum parameter adjustment device d12 according to the invention...
・Plant evaluation function creation function unit, 3...Subtractor, 4
... Comparator, 5... Model evaluation function creation function section,
6... Parameter search function unit using optimization method. Xi, X2 Figure 2 Figure 2

Claims (1)

[Claims] While taking in the control deviation and known disturbance (load command value) of the plant controlled by the PI controller, as well as the current set value of the PI controller (the set value is updated), P
An optimal parameter adjustment device for a controller, which is a device that can determine the optimal setting value of an I controller, and has the following functions. (1) Plant evaluation function creation function unit: Finds the variance values of both the control deviation and the known disturbance in a specified time span, and calculates the plant evaluation function J_P from the ratio. (2) Model evaluation function creation function unit: The independent variable is
The structure of the evaluation function J_M of the model, which is the proportional gain and reset rate of the I regulator, is determined in advance, and the evaluation function J_M of the model is determined using only the coefficients in the model using a dynamic neural learning model algorithm. (3) Find the deviation between the values of the evaluation functions J_P and J_M,
If the absolute value of the value is equal to or smaller than the specified value, it is assumed that the evaluation function J_M of the model has identified the evaluation function J_P of the plant, and the process is transferred to the parameter search function section using the next optimization method. do. If not, return to item (2) and perform learning again. (4) Parameter search function using optimization method: Find parameters (proportional gain, reset rate) that can minimize the model evaluation function value J_M using optimization methods such as nonlinear programming, and use those values as the optimal parameters. Output.