Neural-Network Modeling and Optimization of a Batch Polymerization Reactor

• Main Authors: Bo-Cchang Hung, 洪伯昌
• Other Authors: Jyh-Shyong Chang
• Format: Others
• Language: zh-TW
• Published: 2000
• Online Access: http://ndltd.ncl.edu.tw/handle/76306383677010160750

碩士 === 大同大學 === 化學工程研究所 === 88 === A simulated verification and validation of the neural-network rate-function (NNRF) approach to modeling nonlinear dynamic systems is provided. The NNRF modeling scheme utilizes some a priori process knowledge (the relationships between the measurements and the states of the dynamic system) and experimental data to develop a dynamic neural-network model. Applicability of the proposed neural-network model is examined via the reliability of the computed optimal temperature trajectory in driving the free-radical polymerization reaction to a prescribed molecular weight distribution (MWD). For MWD control via the modified two-step method, profile of instantaneous average chain length to give the desired MWD is estimated first followed by tracking the computed profile of instantaneous average chain length via a conventional proportional-integral (PI) controller, one can obtain the optimal temperature trajectory easily. From the quality of the achieved end product, the proposed NNRF modeling approach can be applied in dynamic modeling a complex free radical polymerization reaction based on the available measurements directly. The NNRF modeling approach can overcome the difficulties that may be encountered in the nonlinear parameter estimation for obtaining an identified physical model.