Recursive Subspace Identification Based on Principal Component Analysis

The problem of recursive subspace identification of state-space models is considered in this paper. A new recursive algorithm based on SA-PCA (stochastic approximation-principal component analysis) is proposed to estimate a basis of the extended observability matrix in the noise-free case. Besides, a recursive algorithm based on RLS (Recursive Least-Squares) is proposed to estimate the system matrices. The algorithm is evaluated by a simulation study.