On the implementation of predictive adaptive control by adaptive predictors

摘要

AbstractA convenient parametrization of LQ predictive control laws in terms of predictions only is stressed. This has two desirable consequences in predictive adaptive control. First, it makes the control computation independent of the identified plant representation (be it transversal, lattice or any other). Secondly, it allows modular VLSI implementation by means of standard RLS adaptive predictor modules. In this connection it is shown that control strategies based on the separate identification of multistep‐ahead predictors are more favourable than strategies based on computing multistep‐ahead predictor parameters by analytically extrapolating the identified one‐step‐ahead pr