KR100310609B1 - Coefficient adjusting apparatus of proportional integral derivative(pid) controller - Google Patents

Abstract

PURPOSE: A coefficient adjusting apparatus of proportional integral derivative(pid) controller is provided to automatically adjust a PID coefficient until rising time, overshot, and correction time as characteristic values of a system output indicating a system attribute arrives at the system output of a desired performance index. CONSTITUTION: A process monitoring station part(100) comprises an evolution calculation part(131), a management part(110), and a fuzzy professional device(120). The evolution calculation part(131) classifies a process to search an optimum regular base with regard to each classified process. The management part(110) interfaces with an outer system or a driver, and generates a control target set value of a control object. The management part(110) displays and manages the set value, a control value, and a process value. The fuzzy professional device(120) performs according to reasoning of a professional technique using a fuzzy rule when satisfying system response specifications of requiring change of a PID coefficient. A process control station part(200) directly collects data from the process via a field bus and controls the collected data. A process part(300) receives a control output value from a constant sampling period in the process control station part(200) and outputs a process value correspond to the received value.

Description

Coefficient Adjustment Device of PID Controller Using Fuzzy Expert Device and Its Control Method

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a coefficient adjusting device of a PID controller using a fuzzy expert device and a control method thereof. In particular, the system is operated by introducing expert techniques to improve the control method of a distributed process control system applied to a plant control system. The present invention relates to a coefficient adjusting device of a PID controller using a fuzzy expert device and to a method of controlling the same.

Proportional Integral Derivative (PID) controller is used for most industrial applications because it is very stable and robust compared to the simplicity of structure. The form of general PID controller is as follows.

-------- (K_c: 비례이득, T_i: 적분이득의 역수, T_d: 미분이득)

-------- (K _c : proportional gain, T _i : inverse of integral gain, T _d : differential gain)

The PID controller determines a control value by operating in a control unit of a control station in charge of substantial control in a distributed control system, and setting of various parameters (proportional, integral, derivative) of the PID controller is performed. It is performed by the monitoring station corresponding to the monitoring part of the distributed control system.

FIG. 1 is a block diagram of a conventional distributed control system according to an embodiment. As shown in FIG. 1, the upper layer station of the distributed control system is in charge of the driver's human interface to monitor the operation state of the entire process. A first through NMS (PMS: Process Monitoring Station) (10A, 10B) which is in charge of monitoring and storing and managing data; First to NPC (PCS: Process Control Staation) units 30A to 30N which perform lower data collection and control directly through a field bus from a process; It is composed of an Advanced Control Station (ACS) 10C of an upper layer including an advanced control algorithm and an algorithm having a special function, which will be described with reference to FIG.

FIG. 2 is a block diagram of a PS / PCS unit to which the conventional PD technique is applied. As shown in FIG. 2, a set value (SV) and a process value (PV) of the first to NPCS units 30A to 30N are shown. : Process Value) is supplied to the adder 31, and the difference thereof, that is, error is supplied to the PID controller 32, from which the control output value (MV) is supplied in a constant sampling period. Outputted to the government unit 40 and the process management and monitor unit 12, and the first to N PMS units 10A and 10B are controlled by the process management and monitor unit 12, and control values and control values and process values. Is displayed and managed and the signal generated indirectly from the external system is selected by the set value generating unit 11 or data is selected by the driver, thereby enabling the interface with the driver.

In order to automatically tune the PID controller to the process, offline tuning and online tuning methods are used. The offline tuning method generates response characteristics (rising time, overshoot, settling time, etc.) that are satisfactory for the PID controller initially determined. If it is not available, it is a method of using the PID controller's coefficients as the final tuned coefficients when the response results of the unit input are collected and collected until the desired response characteristics are obtained. In this case, it may be inconvenient because the driver has to allow a response to a large number of unit inputs, and the online tuning method initially determines the PID controller coefficients, and then updates the PID controller coefficients at each sampling time in response to the results. The coefficient of the PID controller can be set within a short time. A system controller of the coefficients may easily be easily set.

In the conventional apparatus as described above, first, when the process has a large time delay element; Secondly, the actual system does not consider unmodeled dynamics and there is a large amount of nonlinearity that cannot be linearized; Third, the operating point of the process changes according to the change of the set value, and thus the dynamic characteristics of the process are different; Fourth, there is a problem that stable application is difficult in the case where the load variation of the process to be controlled is severe or a disturbance (noise) that cannot be ignored exists in the control output value or the process value.

Therefore, the present invention was devised to solve the above-mentioned conventional problems, and the rise time, overshoot, and settling time which are characteristic values of the system output indicating the property of the system by automatically adjusting the coefficients of the PID controller are desired. It is an object of the present invention to provide a fuzzy expert apparatus and control method that automatically adapts the PID coefficients up to the system output of the figure of merit.

1 is a block diagram of a conventional distributed control system.

Figure 2 is a block diagram of a PS / PS unit applying a conventional PD technique.

Figure 3 is a block diagram of a coefficient adjusting device of the PID controller using the fuzzy expert apparatus of the present invention.

4 is a flow chart of a fuzzy expert apparatus of the present invention.

5 is a block diagram of a relay feedback system.

Fig. 6 is a disk driving function of a relay.

7 is an initial PD coefficient determination table by the Ziegler-Nichols method.

8 is a rule base table according to the present invention.

9 is a goodness of fit of the fuzzy set membership function of FIG. 8;

10 is an optimized rule base table.

11 is a simulation result table using the present invention.

12 is a block diagram showing the configuration of the evolutionary calculation unit in FIG.

13 is an operational flowchart of FIG. 12;

* Description of the symbols for the main parts of the drawings *

100: PMS 110: management unit

120: fuzzy expert apparatus 121: performance index part

122: rule base section 123: pattern extraction section

124: monitor unit 125: fuzzy inference unit

131: Evolutionary Computing Unit 200: PS Part

202: adder 201: PID controller

300: process unit 301: rule definition unit

302: object generation unit 303: performance evaluation unit

304: comparison selection unit 305: process data generation unit

The structure of the present invention for achieving the above object is, as shown in Figure 3, the evolutionary operation unit 131 for classifying the process to find the optimal rule base for each classified process, and interface with an external system or the driver The control unit 110 generates a control target set value of the control target and displays and manages the control value and the process value as well as the set value, and satisfies a system response specification (threshold value of the evaluation index, attribute value) for which the PID coefficient is changed. A PMS unit 100 configured as a fuzzy expert device 120 to perform by reasoning of expert techniques using fuzzy rules; A PC unit 200 which is a lower layer station performing direct data collection and control through a field bus from a process; The PC unit 200 may be configured to include a process unit 300 that receives a control output value output at a predetermined sampling period and outputs a process value corresponding thereto.

The fuzzy expert apparatus 120 includes a pattern extractor 123 for analyzing the transient response after the transient response of the process is finished and outputting the characteristic of the response as a numerical value; A rule base portion 122 having rules for changing controller coefficients; A fuzzy inference unit 125 for updating PID controller coefficients by using the output of the pattern extraction unit 123, desired feature values of each pattern, and a rule base of the rule base unit 122; A performance index unit 121 for stopping updating of the PID coefficients in the fuzzy inference unit 125 when the performance index reaches a certain threshold; Characterized in that it consists of a monitor 124 for displaying the output of the performance index unit 121 and the process unit 300.

The control method of the present invention comprises a first step of optimizing a rulebase using evolutionary computation for a given class of process; A second step of classifying a system by modeling a target process; Determining an initial PID coefficient according to the Ziegler-Nichols method; Determining a threshold value and a weight of the evaluation index; Extracting a pattern using a process value; Calculating an evaluation index, determining whether the threshold is exceeded, and updating the ID coefficient by fuzzy inference; And a seventh step of determining a control output using the updated PD coefficient and storing data of the determined control output and the process value.

Another control method includes a first step of optimizing a rulebase using evolutionary operations for a given class of process; A second step of classifying a system by modeling a target process; Determining an initial PID coefficient according to the Ziegler-Nichols method; Extracting a pattern using a process value for each sampling period; Updating a PD coefficient by fuzzy inference; And a sixth step of determining a control output using the updated PD coefficient and storing data of the determined control output and the process value.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

4 is an operation flowchart of the coefficient adjusting device of the PID controller using the fuzzy expert apparatus of the present invention. As shown in FIG. 4, if the step response is monotonous increase or monotonic increase except for the initial portion, the process may be divided into two types. The first kind is a stable process, and its dynamic characteristics can be expressed as process gain (K _p ), delay time (L), and time constant (T), and the transfer function of a typical stable process is as follows.

------------------ (식1)G (s) = K _p

------------------ (Equation 1)

The second type is an unstable process with an integrator. The dynamic characteristics can be expressed as process gain (K _v ), delay time (L), and time constant (T _v ), and the typical transfer function of an unstable process is as follows.

---------------------- (식2)G (s) =

---------------------- (Equation 2)

을 갖는 공정은 제어하기 쉽고, 큰

을 갖는 공정은 제어하기 힘들다.If you define the normalized delay time without unit as the basis of process classification, the normalized delay time is defined as the ratio of delay time and time constant for the stable process as shown in Equation 3, and the normalized delay time is difficult to control the process. Can be used as a measure of That is, small

Process having a large, easy to control

The process with is difficult to control.

--------------------------------- (식3)θ ₁ =

--------------------------------- (Eq. 3)

Similarly to Equation 3, when the normalized delay time is defined for an unstable process, assuming that the transfer function [G (s)] of Equation 1 is unstable, the transfer function [s × G (s)] of Equation 2 is On the contrary, since it belongs to a stable process, its dynamic characteristics are expressed as delay time (L) and time constant (T _v ), and the normalized delay time for an unstable process with an integrator is defined as follows.

-------------------------------- (식4)θ ₂ =

-------------------------------- (Equation 4)

Since the normalized delay time is used as a measure of difficulty in controlling the process, it is a standard of classification, and according to the normalized delay time, it is classified into four stable processes and two unstable processes as follows.

1) stable process

을 만족하는 공정[Category 1]

Satisfying process

을 만족하는 공정[Category 2]

Satisfying process

을 만족하는 공정[Category 3]

Satisfying process

을 만족하는 공정[Category 4]

Satisfying process

2) unstable process

을 만족하는 공정[Category 5]

Satisfying process

을 만족하는 공정[Category 6]

Satisfying process

The initial PID coefficient is determined according to the Ziegler-Nichols method, and the method of obtaining the PI coefficient is obtained from the relay response.

FIG. 5 is a block diagram of a relay feedback system, FIG. 6 is a disk driving function of a relay, and FIG. 7 is an initial PID coefficient determination table by the Ziegler-Nichols method. When connected, vibration occurs when the following equation is satisfied.

또는

----------- (식5)

or

----------- (Eq. 5)

In Equation 5, N (A) is a describing function of the relay, and the oscillation period (t _c ) and the amplitude (A) of the relay are measured on the Nyquist curve for the target system. The value of a particular point can be known and used to determine the initial gain.

In the high frequency range, a relay loop is fed back into the system with at least 180 ° phase lag. The output of the system oscillates. If the relay amplitude is N≥0 and M≥0, the relay's disk The ice function N (A) is

N (A) =

From here,

b ₁ =

a ₁ =

Therefore, the disk drive function of the relay is as follows.

---------------------------- (식6)

---------------------------- (Eq. 6)

The period of vibration generated by the feedback of the relay is approximately t _c , the critical period of the system (ultimate perriod), and when the amplitude is A, the following equations are obtained using Equations 5 and 6 above.

는 시스템의 임계이득(ultimate gain)이고, 릴레이 되먹임 실험을 통해 얻은

,

값을 이용하여 지글러-니콜스(Ziegler-Nichols)법에 따라 초기의 피아이디 계수를 결정하며, 이때 사용되는 주파수 영역의 지글러-니콜스(Ziegler - Nichols)법은 도7에 도시한 바와 같다.From here

Is the critical gain of the system and is obtained from the relay feedback experiment.

,

The initial PID coefficient is determined according to the Ziegler-Nichols method using the value, and the Ziegler-Nichols method in the frequency domain used at this time is shown in FIG.

In FIG. 3, the fuzzy expert apparatus 120 includes a pattern extraction unit 123, a fuzzy inference unit 125, a rule base unit 122, and the like, depending on whether the control method is an online tuning method or an offline tuning method. How to do it is different.

In the offline tuning method, the pattern extractor 123 analyzes the transient response after the transient response is completed and outputs the characteristic of the response as a numerical value, and the fuzzy inference unit 125 outputs the pattern extractor 123. And the PID controller coefficients are updated by using a rule base (rules for changing controller coefficients) optimized by the specific value and the evolutionary operation desired for each pattern.

In the online tuning method, the pattern extractor 123 differs in that it outputs the characteristic of the response as a numerical value for each sampling period of the process.

In addition, in the offline tuning method, the fuzzy expert apparatus 120 adjusts the coefficients of the PID controller 201 to minimize three property values (rising time, overshoot, and settling time) that the plant response desires when a step input is applied to the plant. Tuning, the pattern extractor 123 extracts three response features after the transient response is over.

These three attribute values are defined as follows.

-------------------------- (식7-1)

-------------------------- (Eq. 7-1)

은 상승시간,

는 오버슈트,

는 정정시간이고,

는 패턴 추출부(123)의 출력이며,

는 각 응답 속성의 원하는 값이고,

는 각 응답 속성에 대한 가중치를 나타낸다.In Equation 7-1

Is the rise time,

Overshoot,

Is the settling time,

Is the output of the pattern extraction unit 123,

Is the desired value for each response attribute,

Denotes a weight for each response attribute.

In the online tuning method, when the stair input is applied to the plant, the plant response tunes the coefficients of the PID controller 201 so that three desired attribute values (error, error rate, and error sum) are minimized, and the pattern extraction unit 123 performs a sampling period. Each of the three response features is extracted. These three attribute values are defined as follows.

e (K) = (y _r (K) -y (K))

Δe (K) = (e (K) -e (K-1)) ------------------------ (Equation 7-2)

=

=

는 오차합, y_r(K)는 설정치, y(K)는 현재의 공정 출력치를 나타낸다.In Equation 7-2, e (K) is an error, Δe (K) is an error rate,

Is the sum of errors, y _r (K) is the set point, and y (K) is the current process output.

FIG. 8 is a rule base table according to the present invention, and FIG. 9 is a goodness of fit of the fuzzy set membership function of FIG. 8. As shown in FIG. 8, the fuzzy rule has three output fuzzy variables. Depending on the output fuzzy variable is applied differently.

,

,

이고, 후건부(consequent)에 해당하는 출력변수는 ΔKc, ΔT_i,

ΔT_d이며, 이때 퍼지집합 변수는 GOOD, BAD이고, 다음과 같은 의미를 갖는다.In the offline tuning method as shown in FIG. 8A, the input fuzzy variable corresponding to the antecedent is

,

,

The output variables corresponding to the consequent parts are ΔKc, ΔT _i,

ΔT _d , where the fuzzy set variables are GOOD and BAD and have the following meanings.

GOOD: Satisfy the desired property

BAD: does not satisfy desired attributes

,

,∼,

)을 사용하였다.That is, if the property is better than the desired property, GOOD, and if not, it means BAD, and the membership function of each set is as shown in FIG. 9A, and as shown in FIG.

,

, ~,

) Was used.

ΔT_d이며, 이때 퍼지집합 변수는 포지티브, 제로, 네거티브이고, 다음과 같은 의미를 갖는다.A rear-line tuning method as in 8b around the conditional input fuzzy variables for the (antecedent) is error, error rates, error sum, and then conditional output variable corresponding to the (consequent) is blood ID coefficient change amount of ΔKc, ΔT _i of _,

ΔT _d , where the fuzzy set variable is positive, zero, negative and has the following meaning.

Positive: Positive Direction

Zero: Zero

Negative: negative direction

,

, ∼,

)을 사용하였는데,

는 실수값인 싱글톤(Singleton)이다.That is, a positive fuzzy set variable with respect to the input means a set of positive inputs, zero means a set of approximately zero inputs, and a negative means a set of negative inputs. The membership function has the form shown in FIG. 9B, and as shown in FIG. 8A, the 27 rules (

,

, ~,

),

Is the real value Singleton.

The coefficients of the PID controller (offline tuning method, online tuning method) are obtained by the following equation.

,

,

는 각각 피아이디 제어기의

번째 계수이고,

,

,

는 각각 피아이디 제어기 계수들을 갱신하는 도3의 퍼지추론부(125)의 출력이다.here,

,

,

Each of the PID controllers

Th coefficient,

,

,

Is the output of the fuzzy inference unit 125 of FIG. 3, which updates the PID controller coefficients, respectively.

The performance index of the entire system is managed by the performance index unit 121 of FIG. 3, which is classified by the offline tuning method and the online tuning method, and the performance index by the offline tuning method is as follows.

----------------------- (식9-1)Figure of merit =

----------------------- (Eq. 9-1)

는 평가지수의 가중치를 나타내며, 성능지수가 어느 임계치에 이를 때 피아이디 계수의 갱신을 중지하도록 하고,

는 패턴 추출부(123)의 출력이다.In the above formula (9-1)

Denotes the weight of the evaluation index, stops updating the PID coefficients when the performance index reaches a certain threshold,

Is the output of the pattern extraction unit 123.

The figure of merit by online tuning method is as follows.

---------------------- (식9-2)Figure of merit =

---------------------- (Eq. 9-2)

는 평가지수의 가중치를 나타내며, 성능지수가 어느 임계치에 이를 때 피아이디 계수의 갱신을 중지하도록 하고, 여기서

은 상승시간(Rise Time),

는 오버슈트(Overshoot),

은 정정시간(Settling Time)이다.In the above formula (9-2)

Denotes the weight of the evaluation index, and when the performance index reaches a certain threshold to stop updating the PID coefficients, where

Is the rise time,

Means overshoot,

Is the settling time.

,

,∼,

)과 도8b에서의 27개의 규칙(

,

, ∼,

)이며, 이를 최적화하기 위해 진화연산(EC : Evolutionary Computation)을 사용하는데, 상기 진화연산은 자연 진화에서의 서로 다른 측면들을 강조하는 유전 알고리즘(GA : Genetic Algorithm), 진화기법(ES : Evolutionary Strategy), 진화 프로그래밍(EP : Evolutionary Programming) 등으로 구분되며, 상기 유전 알고리즘은 염색체 단위에서의 진화를 중요시하고, 상기 진화기법은 개체단위에서의 행동의 변화를, 상기 진화 프로그래밍은 종 단위에서의 행동변화를 중요한 현상으로 여기고 있으며, 이들의 공통적인 개념은 최적의 개체만이 살아남는다는 자연계의 적자 생존의 원리를 최적화기법으로 사용한다.The rule base of FIG. 8 should be optimized to be more adaptive to perform the system operation of the present invention, and the object of optimization is the nine rules (Fig. 8A).

,

, ~,

) And the 27 rules in FIG.

,

, ~,

Evolutionary Computation (EC) is used to optimize this, which is a genetic algorithm (GA) that emphasizes different aspects of natural evolution, and an evolutionary strategy (ES). , Evolutionary Programming (EP), etc., wherein the genetic algorithm emphasizes evolution at the chromosome level, and the evolutionary technique is a change in behavior at the individual level, and the evolutionary programming is a change in behavior at the species level. Are considered to be important phenomena, and their common concept uses the principle of survival of the fittest in nature that only optimal individuals survive.

In general, fuzzy logic is suitable for constructing an expert device that can effectively express the knowledge of experts so that the PID control coefficients satisfy the desired characteristics of the control system, but it is difficult to obtain information systematically from the expert. It provides the ability to optimize the knowledge of expert devices.

,

),

∈{1,∼,P}]로 구성)를 하고, 각각의 부모개체에 대해서 부모개체 평가함수(J)를 계산하는데, 평가함수는 다음 식과 같다.FIG. 12 is a block diagram showing the construction of the evolutionary operation unit in FIG. 3, and FIG. 13 is an operation flowchart of FIG. 12. As shown in FIG. If objects are generated randomly and the number of variables is N, one object is N real arrays [(

,

),

∈ {1, ~, P}], and the parent object evaluation function (J) is calculated for each parent object. The evaluation function is as follows.

----------------- (식10)J =

----------------- (Eq. 10)

는 상수이고,

은 상승시간(Rise Time),

는 오버슈트(Overshoot),

은 정정시간(Settling Time) 이다.From here

Is a constant,

Is the rise time,

Means overshoot,

Is the settling time.

After calculating the evaluation function (J) according to Equation 10, the offspring are generated, and the offspring is generated using the following equation.

=

+

=

+

=

------------ (식11)

=

------------ (Eq. 11)

,

이다.In the above formula (11)

,

to be.

를 계산하고, 부모/자손개체 모두 포함하여 2×P개의 개체에 대해 확률적으로 선택된 비교(Q개의 무작위로 선택된 상대방과 비교)를 하여 자신이 상대방보다 낮은 목적함수를 가지게되면 승(win)수를 한점씩 올리며, 또한 2×P개의 개체 중에서 승수가 높은 P개의 개체의 다음 세대의 부모로서 선택하고, 종료조건을 만족하는지를 판단하여 만족하면 종료하고, 만족하지 못하면 다시 자손개체 생성부터 다시 시작한다.Evaluate function of offspring after creating offspring

If we have a lower objective function than the other party, we compare the randomly selected targets (compared with Q randomly selected opponents) against 2 × P individuals, including both parents and grandchildren. Is increased by one point, and is selected as the parent of the next generation of P entities with a high multiplier among 2 × P entities, and if it satisfies the termination condition, it is terminated if it is satisfied. .

For example, the evolutionary programming was used to optimize the rules (r ₁ to r ₉ ) (r ₁ to r ₂₇ ) of FIG. 8 for the following six processes. Using two process models covering the model and the characteristics of the unstable process,

A) for stable processes

B) unstable processes

The equations include most of the cases occurring in the process and can be applied to various processes.

FIG. 11 is a simulation result table using the present invention. As shown in FIG. 11, the process models were optimized using the following evolutionary programming.

A first step of initializing the population and generating twenty initial populations, ie, parent generations from [-1, 1]; A second step of evaluating the population by the following evaluation function;

(

은 패턴 추출부의 출력, a는 상수)J = a

(

Is the output of the pattern extractor, a is a constant)

A third step of generating a new population, ie, a generation of offspring from the parent generation, by disturbing the parent generation as in the following equation;

Offspring = parent generation + N (0, J)

A fourth step of evaluating the new population; A fifth step of making ten coefficients of evaluation (probably the smallest value of evaluation J) by means of probabilistic selection; A sixth step of sequencing all populations in order of multiplier, and selecting twenty among them; If it is a satisfactory solution, it ends, and if not, it returns to the third step.

The evolution program was applied for 1000 generations to obtain a rule base as shown in FIG. 10, and simulated using the rule base to obtain the result of FIG.

As described above, the apparatus and method for adjusting the coefficients of the PID controller using the fuzzy expert apparatus of the present invention have a value smaller than the rise time and the settling time when the Ziegler-Nichols (ZN) method is applied, In the case of [Category 2], the overshoot is small so that the overshoot is '0' and the rise time and correction time when Ziegler-Nichols (ZN) method is applied. It is effective.

Claims (7)

The evolutionary operation unit that classifies the process and finds the optimal rule base for each classified process, and interfaces with an external system or an operator to generate the control target setpoint for the control object, and displays and manages the control value and the process value as well as the setpoint. And a fuzzy expert device configured to perform by inference of an expert technique using fuzzy rules until a system response specification (a threshold value and an attribute value of an evaluation index) desired to change the PD coefficient is satisfied. A PCS unit, which is a lower layer station that performs data collection and control directly through a field bus from a process; And a control unit configured to receive a control output value output at a predetermined sampling period from the PC unit and output a process value corresponding to the control unit. A first step of optimizing the rulebase using evolutionary operations for a given class of process; A second step of classifying a system by modeling a target process; Determining an initial PID coefficient according to the Ziegler-Nichols method; Determining a threshold value and a weight of the evaluation index; Extracting a pattern using a process value; Calculating an evaluation index, determining whether the threshold is exceeded, and updating the ID coefficient by fuzzy inference; And a seventh step of determining a control output using the updated PD coefficient and storing data of the determined control output and the process value. A first step of optimizing the rulebase using evolutionary operations for a given class of process; A second step of classifying a system by modeling a target process; Determining an initial PID coefficient according to the Ziegler-Nichols method; Extracting a pattern using a process value for each sampling period; Updating a PD coefficient by fuzzy inference; And a sixth step of determining a control output using the updated PD coefficient and storing data of the determined control output and the process value. The apparatus of claim 1, wherein the evolution calculation unit comprises: a rule defining unit defining a rule structure of the expert apparatus; An entity generating unit generating an entity group for each rule; A performance evaluation unit for evaluating a specified performance for each individual; A comparison selection unit which compares each of the individuals evaluated by the performance evaluation unit and selects the parent object based on a defined selection criterion; A coefficient adjusting device of a PID controller using a fuzzy expert device, characterized in that the process data generator for generating data of the process for performance evaluation. The fuzzy expert apparatus of claim 1, further comprising: a pattern extractor configured to analyze the transient response after the transient response of the process is completed and output a characteristic of the response as a numerical value; A rule base portion having rules for changing controller coefficients; A fuzzy inference unit for updating a PID controller coefficient by using the output of the pattern extracting unit, a desired feature value of each pattern, and a rule base of the rule base unit; A performance index unit for stopping updating of the PID coefficients in the fuzzy inference unit when the performance index reaches a certain threshold; And a monitor unit for displaying outputs of the performance index unit and the process unit. 3. The method of claim 2, wherein the sixth step comprises: extracting three response features (rising time, overshoot, correction time) after the transient response is finished; Calculating a figure of merit from the three response features and checking whether the figure of merit satisfies a desired specification; If satisfied by the check of the above step is finished, if not satisfied by using a selected rule updating the PID coefficients through the fuzzy inference coefficient adjustment of the PID controller using a fuzzy expert device, characterized in that Control method. 4. The method of claim 3, wherein the fifth step comprises: extracting three response features (error, error rate, error sum) for each sampling period; Calculating a figure of merit from the three response features and checking whether the figure of merit satisfies a desired specification; If satisfied by the check of the above step is finished, if not satisfied by using a selected rule updating the PID coefficients through the fuzzy inference coefficient adjustment of the PID controller using a fuzzy expert device, characterized in that Control method.

Images