Direct optimization of gain scheduled controllers via genetic algorithms

Abstract

A new approach to designing gain scheduled controllers in which controller performance is directly optimized over the operating envelope of the system is presented. Rather than interpolating between point designs, the basic controller form is selected, and then the controller's parameter schedule is computed to minimize the selected cost functional. The resulting optimization problem is difficult and ill behaved but can be solved using genetic algorithms. The feasibility of the approach and the resulting performance improvements are demonstrated by designing several controllers for the longitudinal dynamics of an F-18 aircraft and comparing the results to a traditionally designed controller.