Optimization of a Batch Polymerization Reactor Via Hybrid Neural-Network Rate Function Model

• Main Authors: Ying-Ta Sung, 宋英達
• Other Authors: Jyh-Shyong Chang
• Format: Others
• Language: en_US
• Online Access: http://ndltd.ncl.edu.tw/handle/74806847642938332285

碩士 === 大同大學 === 化學工程學系(所) === 92 === A simulated verification and validation of the proposed hybrid neural-network rate-function (HNNRF) approach to modeling a batch polymerization reactor system is provided. In a chemical process, some measurements may not be obtainable easily, and the designed NNRF model does not embed these state variables in the built dynamic model. To overcome this problem, the approximated physical model is combined with the NNRF model to give the hybrid neural-network rate-function (HNNRF) model. In this study, a sequential pseudo-uniform design is used to locate desired experiments to provide the HNNRF model of a batch polymerization reactor with rich information. Transformation of the HNNRF dynamic model into a feed-forward artificial neural network (FANN) static model reduces the computation time in determining the optimal operation conditions base on the random search method. An optimal temperature trajectory and initial loading of the initiator for achieving the molecular weight distribution control can be obtained accordingly.