摘要：AbstractThis paper formulates a methodology of on-line subspace identification in wavelet-based multiresolution framework. The proposed strategy integrates proficiency of wavelets for multiscale data representation with the robust parameter estimation ability of subspace identification. The efficacy of proposed technique is demonstrated on point kinetics nuclear reactor taking the effects of temperature and xenon poisoning into consideration. The nonlinear process is approximated by multiple linear state-space models identified at significant scales. Time variation in the process has been taken into account by updating parameters with the arrival of new dataset. It is validated through simulations that the proposed scheme gives better output prediction as compared to other single-scale and multiscale techniques.
