| Project Detail |
Most modern technological applications of social relevance, such as renewable smart-grids and autonomous cars, consist of systems with complex dynamics operating under fast sampling. These processes exhibit behaviours that can only be accurately represented by means of nonlinear models (thus, complex) with sampling periods of micro to milliseconds, which means that decision-making should be done under such time. These systems only operate satisfactorily when adequate control methodologies are used, taking into account performance goals, operational constraints and physical metrics. Thus, this project focuses on the development of real-time optimisation for control, using the Model Predictive Control (MPC) framework.
MPC is a very established approach, centred on generating decision actions by solving an optimisation problem. Over the last decades, considerable scientific effort has been devoted to the study of MPC, expounding its vast pertinence. Yet, applying MPC with nonlinear models issues increased digital complexity, typically incompatible with real-time environments. The standard practice consists in applying approximated solvers, which are not reproducible nor optimal.
Recent works have shown an alternative to render real-time capable algorithms: designing nonlinear MPC using the Linear Parameter Varying (LPV) toolkit. Even though the theoretical side of LPV MPC is presumably established, and several results indicate competitiveness against state-of-the-art solvers, the translation to a practically-viable tool is not yet rendered universal. That is, a ready-for-application MPC solver, using LPV representations, is not yet available. Accordingly, the main goal of this project is to develop a comprehensive real-time optimisation solver, based on LPV MPC. Such a tool would ensure optimality, representation exactness, and generate control laws within the microsecond range, contributing as a technological advance oriented to modern application of social concern. |
