Linear Multivariable System Identification: Multi-experiments Case

International audienceIn this paper, we propose a method for identifying the linear model of a system in the case of multi-experiments. The method is based on the minimization of output error cost function of all the input/output data sets simultaneously. The implementation of the proposed method is based on a local parameterization of the linear state space model in order to minimize the number of gradient search iterations. The optimization process is initialized by an extension of a classical subspace method. We show that, the estimated linear models provided by the proposed method are more accurate than that obtained by the actual methods of linear systems identification. Moreover, we figure out that, the method can handle the case of short multi-experiments by increasing the model's vector of parameters to also include the initial conditions of the internal states