Regression models for prediction and control of processes of unknown structure

Given discrete observations of the input and output values over a period of past history of an unknown controlled process, a minimum order linear stationary difference equation (predictor-controller) is sought which reproduces data in ??-neighborhood of the observations and represents the class of informationnally equivalent regression models for the process. The problem is formulated in Rn and in the l?? (Chebyshev approximation) and l1,?? Banach spaces. Finite linear programming methods are applied to develop effective procedures for model identification.