Concerning system identification, I am currently active in the following areas:

·         Subspace model identification: subspace methods have been an active area of research since the early 1990s and represent now a valid and reliable alternative to classical prediction error minimization methods, particularly when dealing with multivariable problems. I have contributed to this area in various directions, e.g., by providing the first analysis of the uncertainty associated with the computed estimates for the model parameters (using ideas from computational statistics) and by taking part in the development of recursive versions for this class of identification algorithms. In the development of recursive algorithms, the emphasis was initially on computational efficiency and numerical reliability, while more recently the theoretical issues related with the asymptotic convergence and performance of such methods are also being studied (in cooperation with Dr. Guillaume Mercere, University of Poitiers, France).

·         Identification of Linear Parametrically Varying (LPV) models: the LPV paradigm has been proposed in the robust control community as a means to formulate gain-scheduled controller design problems in a technically sound way. While LPV control design is now a fairly mature area, the development of algorithms for the corresponding system identification problems has not followed equally well, mostly because of the complexity of the underlying theory. In this area the current research aims at the development of identification methods for state space LPV models.

·         Identification of nonlinear models: I am currently involved in two projects related to nonlinear identification. The first one, developed in cooperation with Prof. Luigi Piroddi (DEI-POLIMI), aims at extending to nonlinear problems some classical and well known results from the linear identification literature concerning the so-called user choices in system identification. In particular, we have provided the first theoretical analysis of the effect of data and error prefiltering and of the choice of sampling interval on the model accuracy. The second project, in cooperation with Prof. Fabio Previdi (University of Bergamo, Italy), has led to the development of a novel model class for nonlinear identification, which is given by a feedback interconnection of a linear time-varying model with a static nonlinearity. Such models provide a very flexible approach to black box nonlinear identification, and have been successfully employed in a number of applications.

·         Integrated modeling, simulation and identification: object-oriented modeling, symbolic manipulation and parameter estimation are the main ingredient of a research effort, in cooperation with Dr. Francesco Casella (DEI-POLIMI) to define a new class of control-oriented modeling tools capable of combining prior physical knowledge and information from experiments.