Composition Estimation for the Tennessee Eastman Process Using Artificial Neural Networks

• Main Authors: Huang, Ming-Chieh, 黃銘傑
• Other Authors: Huang Chi-Tsung
• Format: Others
• Language: zh-TW
• Published: 1997
• Online Access: http://ndltd.ncl.edu.tw/handle/83551951224714179938

碩士 === 東海大學 === 化學工程學系 === 85 === The use of inferential variables to estimate plant compositionsin place ofdirect on-line measuerments was explored in this study.The Tennessee Eastman Process was employed for the investigation ofusing artificial neural network (ANN) models or estimate compositions.Under varied sinusoidal and step inputs in the set points of process controllers, learning and testing data on the Eastman Process wereobtained by process measurements including on-line analyzers, and a suitable ANN topology was selected by the cross validation technique.The recurrent ANN composition estimator was then developed using process variables (i.e., temperature, pressure, and level measurements) and manipulated variables (i.e., flow rates) under plantwide consideration.Finally, the developed ANN model was undertaken several plant tests onthe Eastman Process via computer simulation. Simulation results have demonstrated that the ANN model can reliably estimate the dynamic compositions, even if the model sampling time is dramatically reduced.In addition, more accurate estimation could also be obtained by the ANN model if an on-line analyzer is employed for calibration.