Le Laboratoire d’Automatique de l’EPFL a le plaisir de vous inviter aux séminaires selon la liste ci-après. Une mise à jour régulière des informations concernant ces séminaires est disponible à l’adresse sur cette page. En particulier, il est conseillé aux visiteurs externes de vérifier que les séminaires soient dispensés comme prévu ci-dessous.
Where: Salle de séminaire LA-EPFL,ME C2 405 (2nd floor) 1015 Lausanne
When: Friday at 10.15am
25.09.2009 Mr. P. Gujral, Laboratoire d’Automatique (EPFL, Lausanne) http://people.epfl.ch/paman.gujral
Latent-variable regression has found use in many diverse applications such as social science, business studies and engineering. I will focus the attention on the engineering application in spectroscopy (e.g. near-infrared, infrared, Raman, chromatography, nuclear magnetic resonance, mass spectroscopy), where a typical goal is to estimate analyte concentrations from the corresponding spectra. Spectroscopic data is characterized by (i) a high degree of multicollinearity in the spectra, and (ii) a linear relationship between the spectra and the concentrations. These properties make latent-variable regression models, such as PCR and PLSR, ideal candidates for calibration.
02.10.2009 Prof. C. de Prada Moraga, Dpt. of Systems Engineering and Automatic Control (University of Valladolid), http://www.isa.cie.uva.es/~prada/
Optimal operation of hybrid processes.
Processes involving continuous and discrete elements or variable structure are known as hybrid systems, and the associated optimization problems are denoting as mixed integer dynamic optimization (MIDO). The talk provides an overview of some methods and tools available for dealing with MIDO problems using the so-called sequential approach, where the optimization algorithms are combined with a dynamic simulator in order to compute cost functions and constraints. A parameterization is proposed that allows converting the MIDO problem into a NLP one, saving computational time. The talk also examines the problems associated with the evaluation of the gradients when discontinuities take place in the inputs or the model due to its hybrid character. Some real life examples are provided to illustrate the method: A crystallization section of a sugar factory, the optimal start-up of an evaporation station and the optimal operation of a desalination plant.
16.10.2009 Prof. M. Deistler, Econometrics and System Theory, Institute for Mathematical Methods in Economics (Vienna University of Technology), http://www.eos.tuwien.ac.at/Oeko/MDeistler/
System identification: General aspects and structure.
System identification is concerned with obtaining good models from data, i.e. with data-driven modeling. In this contribution, the aim is to explain and discuss ideas, general approaches, theories and algorithms for the identification of multi-input, multi-output linear dynamic systems with stochastic noise. Identification of linear systems is a nonlinear problem, since the function attaching an estimated system to the data is nonlinear and is prototypical also for many parts of identification of nonlinear systems. We discuss problems of structure theory, such as parametrization, estimation of real-valued parameters, (maximum likelihood type estimation and its asymptotic properties) and model selection (order estimation by information criteria).
23.10.2009 Prof. J. Lygeros, Automatic Control Laboratory (ETHZ, Zürich, http://control.ee.ethz.ch/~jlygeros/
Randomized optimization for stochastic systems: Theory and applications.
Simulated annealing, Markov Chain Monte Carlo, and genetic algorithms are all randomized methods that can be used in practice to solve (albeit approximately) complex optimization problems. They rely on constructing appropriate Markov chains, whose stationary distribution concentrates on “good” parts of the parameter space (i.e. near the optimizers). Many of these methods come with asymptotic convergence guarantees that establish conditions under which the Markov chain converges to a globally optimal solution in an appropriate probabilistic sense. An interesting question that is usually not covered by asymptotic convergence results is the rate of convergence: How long should the randomized algorithm be executed to obtain a near optimal solution with high probability? Answering this question allows one to determine a level of accuracy and confidence with which approximate optimality claims can be made, as a function of the amount of time available for computation. In this talk we present some new results on finite sample bounds of this type, primarily in the context of stochastic optimization with expected value criteria using Markov Chain Monte Carlo methods. The discussion will be motivated by the application of these methods to collision avoidance in air traffic management and parameter estimation for biological systems.
30.10.2009 Dr. J.-Y. Favez, Bombardier Transportation (Basel, Switzerland),
http://www.bombardier.com/en/transportation
Instabilities and low frequency oscillations in railway power systems.
It is suspected that new customer demands as well as actual control laws for modern railway vehicles could stimulate instabilities and low frequency oscillations in railway power systems. A rigorous analysis of these issues is exceedingly complicated by the complexity of railway power systems.
In this seminar the focus is set to the following topics: constant power demand and power limitation as a function of line voltage and line frequency. In order to illustrate the system’s complexity, an overview of structures and different components of railway systems will be given. It will be shown that the analysis of subsystems and raw simplifications allow establishing potential factors that induce instabilities and oscillations.
20.11.2009 Prof. A. Rapaport, Research Director, French National Research Institute for Agriculture & French National Research Institute for Computer Sciences and Automatic Control, (INRA & INRIA Montpellier, France), http://www1.montpellier.inra.fr/umr_asb/umr.php?page=rapaport
About minimal-time impulse control of sequential batch reactors with one or more species.
We consider the minimal-time optimal control problem of feeding a tank, where several species compete for a single resource, with the objective to reach a given level of the resource. We allow controls to be bounded measurable functions of time as well as impulses. For the one species case, we show that the immediate one impulse strategy (filling the whole reactor with one single impulse at initial time) is optimal when the growth function is monotonic. For non-monotonic growth functions with one maximum, we show that the singular arc strategy (making the resource reach this maximum as fast as possible and maintain it at this level until the fill is complete) is optimal. These results extend former ones obtained by J. Moreno for the class of measurable controls. For the two species case with monotonic growth functions, we first give conditions under which immediate one impulse strategy is optimal. We also give optimality conditions for the singular arc strategy (at a level that depends on the initial condition) to be optimal. The possibility for the immediate one impulse strategy to be non-optimal, although both growth functions are monotonic, is a surprising result, illustrated with the help of numerical simulations. This is a joined work with P. Gajardo and H. Ramirez, from Univ. of Chile.
04.12.2009 Prof. A. Martinoli, Distributed Intelligent Systems and Algorithms Laboratory (EPFL, Lausanne), http://people.epfl.ch/alcherio.martinoli
Modeling and distributed control methods for multi-robot systems: Achievements and trends at DISAL.
In this talk, I will first highlight the challenges related to the design, control, modeling, performance evaluation and optimization of distributed intelligent systems. I will then shift my focus on one particular instance of such system class: distributed robotic systems. In particular, I will describe selected modeling and distributed control methods that we developed in order to deal with resource-constrained, mobile robotic nodes, which have to operate and coordinate their actions in a shared real environment. I will support the discussion with a few recent case studies concerned with distributed sensing and manipulation missions, in particular using miniature robots of a few centimeters in size. Finally, I will extrapolate a few lessons we learned from such methods and outline the importance to combine them, when appropriate, with data-based methods relying on robust machine-learning algorithms in order to overcome some of their limitations.