EVENT DETAILS AND ABSTRACT

Title: Nonlinear system identification: A tool for modeling complex biosystems
Speaker: David Westwick
Speaker Info: University of Calgary
Brief Description:
Special Note: More current information may be available at Plan-it Purple
Abstract:

System identification entails the construction of mathematical models of dynamical systems from measurements of their inputs and outputs. Physiological applications of system identification have centered on gaining mechanistic insight from the identified model. Thus, rather than linearizing the model at an operating point, as is often done in control applications, identified models of physiological systems must deal explicitly with their nonlinearities. We will introduce some of the most commonly used models of nonlinear systems: functional expansions such as the Wiener and Volterra series, and block structuredmodels, comprising cascades of linear dynamic elements and memoryless nonlinearities. Recent advances in the identification of both functional expansions and block-structured models will be discussed. First,tests used to determine a block structure will be applied in a statistical framework, using confidence bounds recently derived for least squares estimates of Volterra kernels. Secondly, iterative cross-correlation based methods for fitting block structured models will be replaced with nonlinear optimizations based on a separable least squares decomposition. Finally, this new approach for fitting block structured models will be modified to allow for the use of alternate descriptions of the nonlinear element(s),including piecewise linear functions, neural networks, rational polynomials and cubic splines. Theoretical discussions will be supported by example applications drawn from studies of sensory systems, biomechanics and lung tissue dynamics.