Kernel identification using quasilinearization (Q1263549): Difference between revisions

The author considers a system defined by a nonlinear integro-differential equation in time. The inner structure of the system is contained in the behavior of the displacement kernel. Generally this inner structure is known completely, leaving to the analyst the task of obtaining a solution to the equation. The author considers, instead, the inverse problem in which the behavior of the system is observed over a finite time interval and he is seeking to determine the structure of the displacement kernel such that the solution of the problem best fits the observed data in the least squared sense. Using the technique of quasilinearization, one tries to establish an inner structure of the system by determining the kernel which best fits the observations. The displacement kernel will be defined as the solution of a nonlinear differential equation with unknown initial conditions and system parameters. The form of the differential equation is established from the particular problem under study and defines the inner structure of the problem to within arbitrary initial conditions and system parameters. Using quasilinearization, a sequence of quasilinear integro-differential equations is defined generating a set of functions whose unknowns are determined through a least squares fit to the observations at each iteration. A numerical example is given in which three parameters defining an exponential kernel were identified.