An algorithm combining the subspace identification methods ORT and CCA

A recent identification study using the Tennessee Eastman Process model revealed that the MOESP-based (multiple-input multiple-output output error state space) algorithm (i.e. implementation) of the subspace identification method ORT (orthogonal decomposition) underachieves given this realistic benchmark example. In this paper, the cause of this unexpected weak performance is analyzed. Based thereon an algorithm combining the idea of the ORT method with the algorithm of the CCA (canonical correlation analysis) method is proposed. Presented simulation results of this algorithm illustrate its enhanced reliability (i.e. determination of a correct model) in comparison to existing methods. This concerns the rejection of arbitrarily colored disturbances within linear systems and the identification of realistic processes like the Tennessee Eastman Process.