Distributed, robust and adaptive model predictive control (MPC)
Supervisor: Dr Paul Trodden
Model predictive control (MPC) is a popular advanced control technique that solves a constrained optimal control problem, on-line, at each sampling instant. The first control input of the optimized sequence is applied, and the optimization is repeated at each subsequent time step, inducing feedback. MPC excels in situations where it is prohibitively difficult to determine an explicit control law off-line: for example, when the system to be controlled is subject to constraints, delays or nonlinearities. The theoretical foundations for MPC are, by now, mature and its applications are widespread; however, some outstanding challenges remain, most notably in distributed and adaptive forms of MPC. Outstanding students are, therefore, sought to research on topics related to distributed, adaptive and also robust MPC.
In distributed (or decentralized) MPC, the conventional MPC problem is decomposed and distributed to several control agents that make decisions locally and independently. This paves the way for the application of MPC to large-scale systems, since the computational bottleneck is removed. The basic challenge is how to coordinate the distributed decision making of agents so that stability of the overall system is maintained, and system-wide performance is acceptable. Many approaches have been proposed, and for an increasingly wide class of systems, but many problems are still open: for example, can the natural sub-optimality of distributed MPC be quantified or bounded? How can distributed MPC methods be made robust to the failure of existing subsystems or the addition of new subsystems? How robust are distributed MPC-controlled systems to cyber attacks?
In adaptive MPC, the dynamics of the system to be controlled are unknown and/or changing over time -- in either a continuous or a discrete (switched) way. The controller must learn or update a model of the system while the latter is being controlled. While seemingly straightforward, this raises several technical and theoretical difficulties, including how to guarantee the properties of stability and constraint satisfaction while probing the system and learning a new model.
Prospective candidates should have an excellent first degree (I or II.i) and/or Masters degree in a mathematical or engineering-related subject. A background in control/systems theory and optimization is desirable.
Tel: +44 (0)114 222 5679