Seminar Details
2014-10-28 (14.00) : Multi-objective dynamic optimization: recent developments and applications
At Euler building (A002)
Organized by Mathematical Engineering
Speaker :
Filip Logist (KUL)
Abstract :
Many processes are dynamic in nature and their optimisation involves the determination of optimal time-varying input policies. This results in optimisation problems with the dynamic system as one of the constraints (i.e., so-called dynamic optimization or optimal control problems). Moreover, multiple and conflicting objectives often arise in practice, giving rise to a set of mathematically equivalent Pareto optimal solutions instead of one single solution. For instance in view of sustainable development, not only more profitable (economically sustainable), but also cleaner (environmentally sustainable) and safer (socially sustainable) processes and operations are aimed at. To enhance decision making in practice, it is important to enable the efficient and accurate generation of optimal alternatives (i.e., the Pareto optimal solutions) and the evaluation of the trade-offs between them.
In this talk recent developments in the domain of multi-objective dynamic optimization will be reviewed with a focus on scalarization based multi-objective approaches. Developments include (i) robust multi-objective optimal control, (ii) multi-objective mixed-integer optimal control, (iii) multi-objective dynamic optimal experiment design, (iv) multi-objective optimisation of periodically operated distributed parameter systems and (v) multi-objective tuning of nonlinear model predictive controllers (NMPC).
Examples include, e.g., (i) the control of (bio)chemical reactors in view of conflicting conversion, yield, productivity and/or energy objectives, while accounting for uncertainty, (ii) the driving of a car with gear shifts in view of minimising both the travelling time and the fuel consumption, (iii) the determination of optimal cyclic steady-states for catalytic tubular reactors in view of conversion and energy recovery, (iv) the determination of dynamic profiles in view of different optimal experiment design criteria for estimating microbial kinetics, or (v) the online optimal control of the Tennessee Eastman benchmark in view of trading off several tracking and economic objectives.