Artificial intelligence applied to the discrimination of the order of multivariable linear systems

Abstract The problem of order determination of multivariable systems from noisy process data records is studied. A method based on pattern recognition principles is proposed. It is a direct method, in that the model parameters are not estimated at any stage. Rather, learning machines are allowed to extract features from suitably defined finite-dimensional functionals of the observed time functions. A non-parametric training procedure is adopted for determining the decision vectors. The learning-machines method is a totally empirical method of data interpretation. The sole assumption is that a relationship between the data and the defined classes exists. Even this assumption may be investigated by the empirical method itself. The attributes of the learning machines developed are presented in this paper.