Computational efficiency in predictive control

Predictive control requires the online solution of an optimisation problem which carries both a large online computation as well as the overhead of substantial code. This is an obstacle to applications on smaller and faster systems where perhaps embedded control may be a better option. The most common tool for constraint handling is model predictive control (MPC); these deploy a quadratic programming optimisation on line which limits the use to slow systems. Algorithms dealing with uncertainty and/or nonlinearity require even more burdensome computations. The aim of this project is to look at ways of modifying MPC to allow constraint handling well at the same stage using simple optimisers and hence to overcome the limitation to slow processes and open access to fast systems. There is a desire that any developments be easily extended to cope with both multivariable problems and uncertainty/non-linearity. Some effort should also be directed towards transparency in order to optimise ease of use. A secondary aim is to investigate the potential for embedding these control strategies. The start point of the research could be multi parametric quadratic programming, a recently developed innovation in quadratic programming which parameterises all the possible solutions to a quadratic program. Secondly one should incorporate recent insights into the structure of MPC to optimise efficiency. It is expected that use will be made of recent research results on invariant sets and the closed-loop paradigm.
Suitable students should have a good engineering background.