ProfessorEric Kerrigan

Many control and optimisation problems involve differential equations, constraints and uncertain parameters. My group develops numerical methods for solving these problems, with a particular emphasis on model predictive control.

Model predictive control and numerical optimisation

Model predictive control (MPC) repeatedly solves an optimisation problem to choose a system's control inputs. Its ability to account for constraints, nonlinear dynamics, and uncertainty makes it useful across engineering, but solving these problems quickly and reliably remains a major challenge.

We develop numerical methods to improve the accuracy, reliability and computational efficiency of optimal control and dynamic optimisation. Our work includes novel discretisation methods for solving dynamic optimisation problems, adaptive discretisation methods for problems involving uncertainty, and derivative-based and derivative-free optimisation solvers. Recent work addresses the accuracy of computed trajectories and robust control of nonlinear systems.

Related work: co-design

A related strand of my research uses multi-objective optimisation to design control algorithms, computing resources and physical systems together, examining trade-offs between performance, cost and reliability.

Applications and software

My research is motivated by applications in aerospace, energy systems and information systems:

• Aerospace: flow control and aerodynamic drag reduction, and control and coordination of autonomous aerial vehicles.
• Energy systems: predictive control for energy management and storage, power conversion, and wind turbine load alleviation. Energy management in groups of buildings is one application.
• Information systems: coordinating control, communication and computation in networks of autonomous vehicles, and scheduling tasks on resource-constrained computing systems.

Current EPSRC projects include data-driven coalitional control for building energy systems, and concurrent learning and control of uncertain large-scale phenomena.

ICLOCS, the Imperial College London Optimal Control Software, is a tool for solving numerical optimal control problems. See my publications for the methods and applications in more detail.