Model Predictive Control

CU Engineering Building

Comic-style illustration of the principle behind Model Predictive Control

Original artwork by , commissioned by Marco M. Nicotra.

Model Predictive Control (MPC) is a modern feedback law that generates the control signal by solving an optimal control problem at each sampling time. This approach is capable of maximizing a certain performance criteria while simultaneously ensuring the non-violation of system constraints.

 

Despite its great potential, the pervasive use of MPC in industrial applications is somewhat limited by the ability to solve the optimal control problem in real-time. As such, a very active area of research is the development of algorithms and implementation strategies that reduce the computational cost required by the MPC.

Due to its formulation, Model Predictive Control is particularly suited for applications that push for elevated performances while having to comply with strict safety requirements

.


Hypersampled Model Predictive Control

An underlying assumption in MPC is that the sampling time of the controller is the same as the discretization time used in the prediction model. However, there may be advantageous to introduce a distinction between the two. Notably, the sampling time should be as short as possible to improve the reactiveness of the control law, whereas the discretization time should be as large as possible to reduce the number of optimization variables for a fixed prediction horizon. This project investigates the idea using notions from numerical methods, optimal control, and digital control.

Students: Yaashia GautamSai Abhishek Aravind

Funding:  (Award Number: 2046212)


Real-Time Model Predictive Control

An effective way of reducing the computational requirements of MPC is to treat the iterative solver as a dynamic system that converges over multiple time steps rather than attaining the exact solution at every sampling instance. Research in this field requires a combination of optimization and control theory to study the interactions between the solver iterations and the plant dynamics. The transient dynamics of this interconnection are addressed by the Real-Time Iteration Governor.

Collaborators: 1, 2
1. University of Michigan
2. University of British Columbia

Students: Terrence Skibik

Funding:  (Award Number: 1904441)