Identification of non-linear time series via kernels (Q4787950)

With a nonlinear time series as observation of some dynamical system, the authors show that it is possible to reconstruct the qualitative behaviour of the dynamical system. This provides a general approach for making accurate short-term predictions. The time series reconstruction can be viewed as a function approximation problem which can be addressed by reproducing kernel Hilbert spaces. This approach is described and related to Bayesian estimation with Gaussian priors. An explicit formulation of the model as an input-dependent autoregressive model is possible. The linearization is exact and the autoregressive coefficients can be calculated explicitly at each step. A concrete benchmark nonlinear problem is numerically solved.