Design of strong causal fitness functions

摘要

The principal contribution of this paper is the introduction of an alternative quality function for parameter optimisation with Evolution Strategies, combining the advantages of the ITSE criterion and the characteristic value criterion. The integral criterion is divided into error sections that are adjusted on the characteristic values of the signal in the time domain. A smooth fitness landscape is achieved. Time-domain based, user-specified performance requirements are implemented and determine the location of the optimum in the parameter space. By means of weighting the priority of reaching certain performance specifications in the optimum can be defined. The new criterion is applicable to all control system design problems with time domain based performance requirements.