Comparative study between three subspace identification algorithms

Since the appearance of subspace system identification in literature different subspace algorithms are available in practice. The most commonly used algorithms are CVA, N4SID and MOESP. Although they all follow the same philosophy, their practical implementation can differ considerably. In the present paper it is our intention to make a direct comparison between these different subspace system identification algorithms. The comparison is made on the basis of 15 publicly available practical data sets on which the algorithms have been applied. The quality measure for the identification we considered is the computational complexity of the method on the one hand and the prediction/simulation error on the other hand. The influence of the identification parameters on the identification quality is studied and analyzed.