Multivariable Process Control Using Decentralized Single Neural Controllers

• Main Authors: Yen, Jia-Hwang, 顏家煌
• Other Authors: Chen Chyi-Tsong
• Format: Others
• Language: zh-TW
• Published: 1997
• Online Access: http://ndltd.ncl.edu.tw/handle/28023160993363418209

碩士 === 逢甲大學 === 化學工程研究所 === 85 === In this dissertation, a learning-type multi-loop control system is developedfor interacting multi-input/multi-output industrial process systems. The recently developed single neural controllers are adopted as the decentralized controllers. With a simple parameter tuning algorithm, the single neural controller in each loop is able to learn to control a changing process by merely observing the process output errors in the same loop. To circumvent strong loop interactions, static decouplers are incorporated in the presented scheme. The only a priori knowledge of the controlled plant is the steady state process gains, which can be easily obtained from open-loop test. The presented learning-type multi-loop control system was tested successfully with some typical multivariable processes. Extensive comparisons with decentralized PI controllers were also performed. Simulationresults show that the performances of the proposed nonlinear control strategywere superior to those of conventional PI controllers,mainly due to its learning ability. Based on its simple structure, efficient algorithm, and goodperformance, it is convinced that proposed learning-type decentralized control system has high potential for controlling interacting multivariableindustrial processes, alternative to existing decentralized control strategies.