From lust to brainwave: fifteen years of persistent excitation

摘要

Adaptive predictive controllers based on truncated orthonormal series representation of the process dynamics were first proposed in the mid 1980's. Those schemes have since been extended to stochastic, multivariable and even nonlinear systems, using various predictive or robust control techniques. Several advantages have been claimed to result from the use of an orthonormal series representation of process dynamics, particularly in an adaptive control framework. The Laguerre model is an output-error structure, is linear in the parameters, preserves convexity for the identification problem. It eliminates the parameter drift due to the influence of unmodelled dynamics on the nominal model, is stable, and robustly stabilizable as long as the numodelled dynamics are stable. This effectively solves the so-called admissibility problem and makes the Lagurre structure particularly suitable for adaptive control applications in the process industries. Various adaptive controllers based on the Laguerre representation have been developed, and used in industrial applications, in diverse application such as pulping and bleasching, TiO_2 pigment production, lime production, glass furnaces, gas desulfuration, fibre glass production, and beer brewing. In particular, considerable experience has been gained through the development of two commercial "general-purpose" adaptive controllers. This experience will be presented and discussed. The validity of the advantages claimed for the Laguerre methodology will be scrutinized. Finally, the ingredients necessary to the design of a robust adaptive process controller will then be reviewed and iterative control will be discussed in this framework.