1,721,266 research outputs found
Partitioning datasets based on equalities among parameters
No full textWhen a phenomenon is described by a parametric model and multiple datasets are available, a key problem in statistics is to discover which datasets are characterized by the same parameter values. Equivalently, one is interested in partitioning the family of datasets into blocks collecting data that are described by the same parameters. Because of noise, different partitions can be consistent with the data, in the sense that they are accepted by generalized likelihood ratio tests with a given confidence level. Given the fact that testing all possible partitions is a computationally unaffordable task, we propose an algorithm for finding all acceptable partitions while avoiding to test unnecessary ones. The core of our method is an efficient procedure, based on partial order relations on partitions, for computing all partitions that verify an upper bound on a monotone function. The reduction of the computational burden brought about by the algorithm is analyzed both theoretically and experimentally. Applications to the identification of switched systems are also presented
Single-linkage clustering for optimal classification in piecewise affine regression
No full textWhen performing regression with piecewise affine maps, the most challenging task is to classify the data points, i.e. to correctly attribute a data point to the affine submodel that most likely generated it. In this paper, we consider a regression scheme similar to the one proposed in (Ferrari-Trecate et al., 2001,2003) that reduces the classification step to a clustering problem in presence of outliers. However, instead of the K-means procedure adopted in (Ferrari-Trecate et al., 2001,2003), we propose the use of single-linkage clustering that estimates automatically the number of submodels composing the piecewise affine map. Moreover we prove that, under mild assumptions on the data set, single-linkage clustering can guarantee optimal classification in presence of bounded noise
System identification for biological systems
No full textEditorial for the special issue on systems identification for biological systems
Computation of Observability Regions for Piecewise Affine Systems: A Projection-Based Algorithm
No full textIn this paper we consider the problem of computing sets of observable states for discrete-time, piecewise affine systems. When the maximal set of observable states is full-dimensional, we provide an algorithm for reconstructing it up to a zero measure set. The core of the method is a quantifier elimination procedure that, in view of basic results on piecewise linear algebra, can be performed via the projection of polytopes on subspaces. We also provide a necessary condition on the minimal length of the observability horizon in order to expect a full-dimensional set of observable states. Numerical experiments highlight that the new procedure is considerably faster than the one proposed in (Ferrari-Trecate and Gati, 2004)
Fuzzy systems with overlapping Gaussian concepts: Approximation properties in Sobolev norms
No full textIn this paper the approximating capabilities of fuzzy systems with overlapping Gaussian concepts are considered. The target function is assumed to be sampled either on a regular gird or according to a uniform probability density. By exploiting a connection with Radial Basis Functions approximators, a new method for the computation of the system coefficients is provided, showing that it guarantees uniform approximation of the derivatives of the target function
Plug-and-play distributed model predictive control with coupling attenuation
No full textWe consider the control of a large-scale system composed of state-coupled linear subsystems that can be added or removed offline. In this paper, we present plug-and-play (PnP) design methods based on model predictive control, meaning that (i) the design of a local controller requires information from parent subsystems only, (ii) the plugging in/out of a subsystem triggers at most the redesign of controllers associated to subsystems coupled to it, and (iii) plug-in/out operations are automatically denied if they compromise the stability of the overall system or constraint satisfaction. We advance previously proposed PnP decentralized control schemes by introducing a distributed control architecture that exploits communication between coupled subsystems. New controllers embody coupling attenuation terms that make PnP design applicable even when existing synthesis method are not. The main features of our approach are illustrated considering thePnP design of controllers for regulating the frequency of multiple generators in power networks
Observability analysis and state observers for automotive powertrains with backlash: a hybrid system approach.
No full textIn this paper, the observability properties of automotive powertrains with backlash are analysed. We model the powertrain as a hybrid system in the piecewise affine form and use measurements of the torque and the angular speed of the engine for computing the maximal set of observable states. This set, that is usually non-convex and disconnected, captures in a precise way how the main variables and parameters of the driveline influence the possibility of estimating the shaft twist. Then, we show how to exploit the knowledge of observable states in order to build computationally efficient deadbeat observers for the reconstruction of the powertrain states
Call for papers: Special issue on systems identification for biological systems
No full textExperimental techniques in molecular biology have led to the production of enormous amounts of data on the
dynamics of cellular processes. The availability of time series data characterizing genomic, proteomic and metabolic
systems must be complemented with formal methods for identifying quantitative models of networks of interactions.
Reverse-engineering of regulatory networks is a central issue in modern biology because, beside enabling the
computer-based simulation of biological systems, it promotes the understanding of cell functioning and underlies
the design of interventions of biotechnological or biomedical relevance. However, standard system identification
techniques are unlikely to work out of the box since they must cope with (1) the complexity and the high nonlinearity
of biological systems; (2) the quality and type of available biological data; (3) the stochastic nature of chemical
interactions; and (4) the interaction of discrete events and continuous dynamics.
The aim of this special issue is to present some very recent achievements in system identification tailored to the
reconstruction of biological processes
Hybrid identification methods for the reconstruction of genetic regulatory networks
No full textThis tutorial paper considers the problem of reconstructing Genetic Regulatory Networks (GRNs) from gene expression measurements. Among various modeling frameworks that have been proposed for these biological systems, we focus on PieceWise Affine (PWA) models because of their ability to capture both the switching behavior of genes and the continuous dynamics of molecule concentrations. PWA models of GRNs have a special structure that must be preserved by the identification process. In the paper, we discuss the new challenges that this constraint raises in the field of hybrid identification. As an example, we summarize recently proposed methods for detecting switches in gene expression profiles and for reconstructing multiple PWA models consistent with the data. We also present the results obtained by applying these algorithms to synthetic data produced by PWA models of the GRN governing the carbon starvation response in E. coli
- …
