Model Structure and Adjustable Parameter Selection for Operations Optimization

Abstract

<p>The value of model-based process optimization systems for competitive advantage in many industries, has been widely recognized. Such model-based optimization systems include Real-Time Optimization, On-Line Optimizing Control, off-line process scheduling, and any other economic process optimization scheme which uses a process model to predict optimal plant operation. The thesis investigates the design of these model-based optimization systems, particularly with respect to model structure and adjustable parameter selection.</p> <p>The main contribution of this work include design phase methods, based on fundamental principles of optimization and statistics theory, for determining whether a model-based optimization system can attain the plant optimum, as well as methods for discriminating between design alternatives. Three necessary conditions for zero-offset from the optimal plant operation are presented. These include Pont-Wise Model Adequacy, Augmented Model Adequacy and Point-Wise Stability. Recognizing that achieving zero-offset from the plant optimum may not always be possible, or may not be the only design objective, a Design Cost method is presented for selecting among design alternatives. This Design Cost method provides a natural "trade off" between offset elimination and variance of the predicted optimal manipulated variable values.</p> <p>Finally, the thesis is completed with a larger-scale case study involving the Williams-Otto Plant [1960]. In the case study selection of a process model and the adjustable parameter set for implementation in closed-loop Real-Time Optimization system is investigated.</p>