| This dissertation presents a methodology for cost weighting normalization in optimal predictive control. The cost functions formulated for optimal control typically involve multiple terms associated with inputs and outputs over different windows of time. Based on gain relationships, normalization factors are devised to unify the scaling of different cost terms. The method is first developed for single-input and single-output (SISO) and multi-input and single-output (MISO) systems, and then extended to multi-input and multi-output (MIMO) systems. The normalization makes weighting factors in an optimal control cost function comparable. For SISO and MISO systems, the weighting factors are exactly comparable; and for MIMO systems, they are in the same order of magnitude. This greatly simplifies weighting factor selection and thus facilitates controller tuning. The normalization method was applied in adaptive control of a lab scale fluid system and a food extrusion system, both with significant time delays. Both simulation and experimental results are presented to demonstrate the effectiveness of the methodology. |
