비선형 시스템의 모델링을 통한 퍼지 제어기 설계

Abstract

퍼지 제어에서 일반적 제어규칙의 수정방법은 성능 지수표에 의해 과거의 추론된 출력에 보정 함으로써 과거상태의 규칙을 수정하거나 생성한다. 그러나 이러한 방법은 규칙의 수정이나 생성을 직접 제어대상의 실험을 통하여 이루어져야하며, 또한 원하는 성능의 제어규칙을 얻기 위해서는 많은 수정작업이 거쳐야한다. 본 논문에서는 제어대상의 입·출력값과 퍼지이론을 통하여 제어대상을 모델링하며, 모델링된 제어대상을 통해 제어규칙을 수정하거나 생성한다. 실제 시스템에 최종 제어 값을 적용함으로써, 실 시?뵀謗? 대한 적은 수정작업으로 안정된 제어기를 구현하였다.

In fuzzy controllers, it is not easy to optimize linguistic control rules, the input-output gains, and membership functions. The trial-and-error method is often used to adapt these factors, but it needs many effort and time. The control method using simple look-up tables is very efficient in on-line system control, because it needs a few effort to calculate output values.

In this paper, a control method that is based on look-up table control strategy, that can optimize control rules is proposed. To optimize the control rules, real plant are modeled by gradient descent training method and then control rules are modified. After the modifying, the control rules are applied to real plant, then the final rule optimizing learning is performed to minimize the difference between the real plant and modeling plant.

By the simulation with nonlinear system, it is shown that the controller using the proposed method can be constructed easily, and that it works stably.

In fuzzy controllers, it is not easy to optimize linguistic control rules, the input-output gains, and membership functions. The trial-and-error method is often used to adapt these factors, but it needs many effort and time. The control method using simple look-up tables is very efficient in on-line system control, because it needs a few effort to calculate output values.

In this paper, a control method that is based on look-up table control strategy, that can optimize control rules is proposed. To optimize the control rules, real plant are modeled by gradient descent training method and then control rules are modified. After the modifying, the control rules are applied to real plant, then the final rule optimizing learning is performed to minimize the difference between the real plant and modeling plant.

By the simulation with nonlinear system, it is shown that the controller using the proposed method can be constructed easily, and that it works stably.