查看文章

Koopman Operator is a linear but infinite-dimensional operator defined for a nonlinear dynamical system and captures full information of the system. We present a formulation in Reproduced Kernel Hirbert Space (RKHS) for modeling a nonlinear dynamic system in order to develop relevant linear estimators. The KO is represented as a linear estimator in RKHS, and its parameters are determined using the well-known Gaussian process models. This leads to structures useable in modeling and nowcasting that account for the nonlinear behavior of the system.