실시간 PID 파라미터 추정을 위한 신경회로망 제어기 설계

Abstract

본 논문에서는 정확한 수학적 모델을 구하지 못하는 경우와 제어기의 정확한 파라미터 설계가 어려운 PI제어기의 단점을 극복하고자, 오류역전파 알고리즘을 적용한 신경회로망의 직접제어기로 BLDC 전동기 속도제어기의 PI 파라미터를 자기 동조하는 방법을 사용하였다. 활성화 함수는 절대치 양극성 함수를 적용하여 응답특성을 향상 시켰다. 제안된 활성화 함수와 알고리즘으로 시뮬레이션과 실험 결과를 통하여 타당성을 확인하였으며, 벡터제어 시스템의 적용 시 기존의 PI제어기 방식과 비교하여 BLDC 전동기의 속도제어의 응답특성 및 속응성의 양호한 제어 성능을 확인하였다.; PID Controller has been most generally used in the industrial process control in spite of developing various modern control methods. And the various researches have been made to improve the performance of PID Controller. Several different methods such as Ziegler-Nichols Tuning Method, Astrom and Hagglund`s Selftuning Method have been proposed for tuning of PID Controller. Recently, many methods have been proposed in Self-tuning of PID Controller using neural networks. But the methods developed so far have used plant Jaccobian when they learned the neural networks. Also, this paper is improved the leaning efficiency of neural networks using plant Jaccobian. So these methods had shortcomings that were slow adjust against plants, which has a noise and nonlinear. Recently, adaptive control skill by neural network is doing research. This neural network shows good gain control from output response of a channel, even if there are only changing of set point without any information of processor and phasing noise. And it is because this model process the changing of optional input-output data, which are being based on parallel dispersion processing that real time processing is possible. And it shows good performance about linear and nonlinear calculation. This treatise tries to insist that the characteristic of neural networks can adapt against any kinds of noise or operation circumstances. Therefore, this paper is used to PI parameter Self-tuning of Induction motor speed controller for direct controller of neural networks to EBPA (Error Back- Propagation Algorithm). And applied new Absoluteness Tangent Hyperbolic function is conducted to response characteristic. So, nonlinear function and algorithm is improved through the simulation and experiment result. The performance of the Self-tuning controller is compared with that of the PI controller tuned by conventional method.