Plenary lecture by Alberto Bemporad at the 4th IEEE Conference on Control Technology and Applications (CCTA 2020), Montreal, Canada, August 25, 2020.
Slides: [ Ссылка ]
ABSTRACT:
Model Predictive Control (MPC) is widely recognized as a very effective advanced control technique for multivariable systems subject to constraints on input and output variables. At each sampling step, MPC selects a sequence of manipulated inputs that, according to predictions made by a dynamical model of the controlled process, optimizes a certain performance index under constraints. The main challenges in applying MPC to real industrial problems are (i) how to get suitable prediction models, especially directly from experimental data, (ii) how to solve the optimization problem associated with MPC on line with high throughput, limited machine precision, limited memory resources, and within hard real-time constraints, and (iii) how to help industrial calibrators tuning the design knobs of the controller. In my talk I will describe the main concepts of MPC and present some recent developments in embedded optimization for MPC and in calibration of MPC from data. I will also highlight the main advantages of MPC over more classical methods, in terms of better multi-actuator coordination, reduced setup time and calibration effort, and easier maintenance and portability of the design. Finally, I will provide evidence of how MPC has now become a viable technology for production, thanks to the availability of specialized design and real-time software, giving concrete examples of MPC controllers already in mass production in the automotive industry.
