A Tool for Control Research Using Evolutionary Algorithm That Generates Controllers with a Pre-Specified Morphology

Abstract

In control research and design it is frequently necessary to explore, evaluate, tune and compare many control strategies. These activities are assisted by software tools of increasing complexity; however, even with the existing high performance tools these activities are very time consuming due to they imply hundred if not thousand of simulations. If the process of doing such simulations is not automated it can be a very time consuming task. There has been proposed evolutionary algorithms (EA) that in the search for an optimal control automatically generate many control structures. However, the space of possible controllers for any dynamical system is huge. Hence it is mandatory to restrict the search space. The best way to restrict the controller search space is to let the designer influence the search direction. In this paper we propose a software tool for control research that has as its main part an EA that produce only controllers having a pre-specified morphology. By specifying a controller morphology the designer can influence the search direction without losing the exploration capability of evolutionary algorithms. The EA is endowed with a cost function tailored for fast evaluation of closed-loop controller performance. The use of the tool is illustrated by searching an sliding mode and similar controllers for an unstable linear and two nonlinear systems.