Tuesday, May 31, 2016, DTU Compute have three very interesting lectures related to prediction and optimization based control. The lectures will be given at DTU Building 308 Auditorium 13. The titles, speakers and short abstracts are given below. Everybody is welcome. Feel free to distribute this invitation to potentially interested colleagues and students.
09:30-10:15
Distributed MPC: a comprehensive overview and some recent advances
Gabriele Pannocchia, University of Pisa, Italy
Abstract:
Distributed model predictive control refers to a class of predictive control architectures in which a number of local controllers manipulate a subset of inputs to control a subset of outputs (states) composing the overall system. Different levels of communication and (non)cooperation exist, al- though in general the most compelling properties can be established only for cooperative schemes, those in which all local controllers optimize local inputs to minimize the same plantwide objective function. Starting from state-feedback algorithms for constrained linear systems, extensions and recent advances are discussed to cover output feedback, reference target tracking, and offset-free control. An outlook of future directions is finally presented.
10:15-11:00
Model-based online optimization and control with imperfect models
Sebastian Engell, TU Dortmund, Germany
Abstract:
The economics and energy and resource efficiency of chemical production processes can be significantly improved by employing model-based optimizing control and model-based real-time optimization. However, plant models never are perfect, and the effort to develop very accurate models may be prohibitively high. Coping with a mismatch between the model that is used in the optimization or optimizing control and the behaviour of the real plant therefore is of crucial importance in model-based optimization and control. The talk will discuss several approaches to implement optimizing control and real-time optimization using models which describe the plant behaviour approximately, but not accurately.
We will first discuss the online optimization of the operating points of chemical plants using a combination of models and measured data from previous operating points. The measured data is used to iteratively correct the optimization model by means of the so-called modifier adaptation. In this context, we propose a novel way of estimating the modifiers – the empirical plant gradients- which is based on ideas from derivative-free optimization.
Secondly, we discuss multi-stage optimizing control which is a systematic way to achieve optimal performance and robust constraint satisfaction in the presence of model uncertainties. In contrast to other approaches, the presence of feedback in the control loop which enables corrections of the predicted control moves after new information has been obtained is taken into account in the optimization by computing so-called recourse actions.
13:00-13:50
Industrial PhD lecture:
Economic Model Predictive Control for Spray Drying Plants
Lars Norbert Petersen, GEA Process Engineering A/S & DTU
14:00-16:00 PhD defense