A FRAMEWORK FOR INTEGRATED REACTIVE SCHEDULING OF PLANT OPERATIONS AND MAINTENANCE

Abstract

<p>This thesis addresses a reactive (closed-loop) scheduling framework for integrated<br />scheduling of process operation and maintenance. Ethylene plant furnace scheduling is<br />chosen as the process example because it is concerned with optimally scheduling the<br />furnace operations and its periodic maintenance shutdowns. The main reason for<br />choosing this example is that it addresses a special class of scheduling problems where<br />the operations and maintenance have strong interactions and so integrated decision<br />making becomes necessary.</p> <p>The major goal of this research was to select an appropriate closed-loop<br />framework and develop a reactive scheduling system. With the rolling horizon approach<br />being the most suitable method for closed-loop schedule implementation, a Model<br />Predictive Control (MPC) framework is chosen in this thesis. The presence of integer<br />variables in the scheduling model made the reactive scheduling systems resemble a<br />hybrid MPC problem.</p> <p>The research was performed in two phases. In the first phase, an open-loop<br />scheduling model was formulated as a Mixed Integer Linear programming problem<br />(MILP) using discrete time representation. To ensure the tractability of the model,<br />tailored formulation methods (disjunctive reactor modelling, constraint reformulation and tightening constraints) and efficient solution heuristics (two-stage temporal<br />decomposition heuristic) were developed in this thesis. In the second phase, the openloop scheduling model was applied in the prototype closed-loop framework to develop a reactive scheduling system to assist engineers make appropriate decisions in a timely manner.</p> <p>The automated scheduling system developed in this thesis was tested for several<br />scenarios and proved to have significant benefits over manual scheduling procedures,<br />confirmed that the scheduling model is tractable and achieved feasible solutions for all<br />the scenarios considered, including large problems with multiple feeds, multiple reactors and a long (90-day) scheduling horizon. We conclude that the technology is appropriate for further improvement and eventual application in the industry.</p>