Evolutionary computation and nonlinear programming in multi-model-robust control design

Abstract

An algorithmic parameter tuning methodology for controller design of complex systems is needed. This methodology should offer designers a great degree of flexibility and give insight into the potentials of the controller structure and the consequences of the design decisions that are made. Such a method is proposed here. For an exploratory phase a new pareto-ranked genetic algorithm is proposed to generate an evenly dispersed set of near optimal, global, solutions. By pair-wise preference statements on design alternatives a linear program is set up as a formal means for selecting the solution with best overall designer satisfaction. In a following interactive design phase using nonlinear programming techniques with a priori decisions on allowed quality levels, a best tuning compromise in competing requirements satisfaction is searched for while guaranteeing pareto-optimality. In particular, this two-phase tuning approach allows the designer to balance nominal control performance and multi-model control robustness.