Date of Award
Master of Engineering (ME)
The current industry standard in real-time optimization (RTO) is the two-step method. In this approach, mismatch between the plant and process model is compensated for by continuously updating a subset of the parameters in the process model. It is suitably resistant to measurement noise, however it is not guaranteed to move toward the plant optimum if structural plant-model mismatch exists. Due to this deficiency, a number of alternative methods have been developed over the years, including ISOPE and modifier adaptation. These methods, however, utilize plant derivative information, which must be estimated because a precise plant model is typically not known in practice. This makes these methods particularly susceptible to measurement noise. Therefore, in this thesis, the development of an RTO technology which is both optimum seeking and resistant to measurement noise is considered.
This research can be separated into two parts. In the first phase, the current state-of-the-art modifier adaptation algorithm is modified by employing Broyden's method to estimate the plant output derivatives. A pair of deficiencies of Broyden's method are then detailed, and a modification to the algorithm, designed to mitigate these deficiencies, is proposed. This consists of the inclusion of additional constraints in the model-based optimization problem, designed to limit both offset and variance in the Broyden derivative estimates. Since the new algorithm possesses two distinct goals, optimality and the accuracy of the Broyden estimates, it is referred to as dual modifier adaptation.
In the second phase of this research, the design of dual modifier adaptation systems is considered. The design methodology is built around the design cost criterion, a metric which had previously been developed for the two-step approach of RTO. The calculation procedure for the metric is adapted in this research in order to address dual modifier adaptation systems. In addition, an approach designed to compute the constraint back-off necessary to ensure a certain level of feasibility is developed.
The concepts discussed in both the first and second phases of the research are illustrated using the Williams-Otto Reactor case study. This is a benchmark problem that has been used in the RTO literature for many years. A more involved case study, a propane furnace, is introduced in the last main chapter of this thesis. Both the performance of the dual modifier adaptation algorithm itself and the design of dual modifier adaptation systems are discussed for this case study.
Rodger, Eric, "Dual Modifier Adaptation Methodology For the On-line Optimization of Uncertain Processes" (2010). Open Access Dissertations and Theses. Paper 4336.
McMaster University Library