A multiresolution wavelet based subspace identification

This paper presents multiresolution subspace identification as an extension of classical subspace modeling, thus inheriting features of robust subspace identification with added advantages of wavelet based modeling enabling multiresolution state-space model development. Identification of a noisy process in presence of mild nonlinearity can be approximated by estimating multiple multiresolution time invariant models. Parameter estimation in projection space at appropriate scales is achieved using least squares method. The efficacy of the proposed approach has been demonstrated by modeling nuclear reactor in prediction as well as simulation environment. It is shown that root mean squared error reduces significantly as compared to their single scale counterparts providing better modeling performances.