Estimate of process compositions and plantwide control from multiple secondary measurements using artificial neural networks

• Published in: Computers & chemical engineering Vol. 27; no. 1; pp. 55 - 72
• Main Authors: Yeh, Tzu-Ming, Huang, Ming-Chieh, Huang, Chi-Tsung
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.01.2003
• Subjects: Artificial neural networks; Composition estimator; Inferential control; Model predictive control; Plantwide control; Tennessee Eastman plant; Composition estimator; Plantwide control; Model predictive control; Inferential control; Tennessee Eastman plant; Artificial neural networks
• ISSN: 0098-1354; 1873-4375
• DOI: 10.1016/S0098-1354(02)00152-7

The primary object of this exploration is to infer process compositions from other measurements in place of on-line analyzers under plantwide consideration using the predicted capability of artificial neural networks (ANN) dynamically. The Tennessee Eastman (TE) plant was employed for the investigation. First, transient data of process compositions and secondary measurements, such as temperature, pressure, level, and flow rate were obtained. Then, a composition estimator was developed using these data by the training and testing of recurrent ANN models based on a cross-validation technique. Several plantwide control techniques including classic PI control and supervisory model predictive control (MPC) to regulate process compositions in the TE plant using ANN estimators were investigated. Simulation results have demonstrated that an ANN estimator with laboratory calibration can reliably estimate process compositions on-line from secondary measurements. In addition, plantwide control using ANN estimators, either in the classic PI control or in the neural MPC, can perform reliable composition control.