文摘
In current model predictive control (MPC) applications, the accuracy of the models utilized is usually themain factor for the final successes. These models, especially for complex multivariable processes, are oftenobtained by system identification. Then, the design of the identification input signals is the key step for asuccessful MPC scheme. In this paper, a new design method is proposed based on a preliminary steady stategain model of a multivariable system and process operability. This is a general design method, and the existingmethods can be viewed as its special cases. Previous publications have only considered an orthogonally rotatedinput signal whose aspect ratios are fixed by the singular values of the gain matrix. The one presented herealso allows oblique projections, and the aspect ratios are calculated by the use of explicit considerations onthe outputs. As a result, the identified model from this method is expected to be more accurate. From theparameter covariance analysis, it is shown that this method leads to the D-optimal design for the gain matrixparameter estimation. Because the preliminary model is usually an approximate one, model uncertainty isexplicitly accounted for in the proposed method. Two examples are given to prove the effectiveness of theproposed method. This paper is a more detailed account of an earlier communication [Zhan, Q.; Georgakis,C. In Proceedings of the 12th IFAC Symposium on System Identication, Santa Barbara, CA, 2000].