| Summary: | The optimisation of dynamic systems is of high relevance in chemical engineering as many practical systems can be described by ordinary differential equations (ODEs) or differential algebraic equations (DAEs). The current techniques for solving these problems rigorously to global optimality rely mainly on sequential approaches in which a branch and bound framework is used for solving the global optimisation part of the problem and a verified simulator (in which rounding errors are accounted for in the computations) is used for solving the dynamic constraints. The verified simulation part is the main bottleneck since tight bounds are difficult to obtain for high dimensional dynamic systems. Additionally, uncertainty in the form of, for example, intervals is introduced in the parameters of the dynamic constraints which are also the decision variables of the optimisation problem. Nevertheless, in the verified simulation the accumulation of trajectories that do not belong to the exact solution (overestimation) makes the state bounds overconservative and in the worst case they blow up and tend towards ±∞. In this thesis, methods for verified simulation in global optimisation for dynamic systems were investigated. A novel algorithm that uses an interval Taylor series (ITS) method with enhanced overestimation reduction capabilities was developed. These enhancements for the reduction of the overestimation rely on interval contractors (Krawczyk, Newton, ForwardBackward) and model reformulation based on pattern substitution and input scaling. The method with interval contractors was also extended to Taylor Models (TM) for comparison purposes. The two algorithms were tested on several case studies to demonstrate the effectiveness of the methods. The case studies have a different number of state variables and system parameters and they use uncertain amounts in some of the system parameters and initial conditions. Both of the methods were also used in a sequential approach to address the global optimisation for dynamic systems problem subject to uncertainty. The simulation results demonstrated that the ITS method with overestimation reduction techniques provided tighter state bounds with less computational expense than the traditional method. In the case of the forward-backward contractor additional constraints can be introduced that can potentially contribute significantly to the reduction of the overestimation. Similarly, the novel TM method with enhanced overestimation reduction capabilities provided tighter bounds than the TM method alone. On the other hand, the optimisation results showed that the global optimisation algorithm with the novel ITS method with overestimation reduction techniques converged faster to a rigorous solution due to the improved state bounds.
|
|---|
