Multiscale modelling and identification of a class of lattice dynamical systems

A new multiscale modelling framework is introduced to describe a class of lattice dynamical systems (LDS), which can be used to model natural systems involving multiphysics and the multi-resolution facets of a single spatio-temporal dynamical system. The emphasis of the paper is on the multi-resolution facets, with respect to the spatial domain, of a single spatio-temporal dynamical system by using a Haar wavelet decomposition technique. A multiscale identification method for such systems is then proposed, which can be considered as a dual of the multigrid method. The proposed identification method involves three steps: the system dynamics at some specific scale of interest are identified using a recursive least-squares algorithm; the residual is then projected onto coarser scales using Haar wavelets and the parameter estimation errors are minimized; and finally a coarse correction procedure is applied to the original scale. An outstanding advantage of the proposed identification method is a saving on the computational costs. Numerical examples are provided to demonstrate the application of the proposed new approach.