Structural estimation of RKHS models

This paper proposes a new algorithm to estimate the minimal value of the parameter number defining the model developed in Reproducing Kernel Hilbert Space (RKHS) and describing non linear processes. The estimated value which corresponds to the model order is equal to the number of input / output measurements contained in a learning set used to develop the model. The proposed algorithm consists on characterising the nonlinear process by an mth order model, incrementing this order and computing for each value a given criterion. The seaked value is obtained when the computed criterion jumps suddenly.