Comparative mixing for DSMGA-II
Marcin M. Komarnicki, Michal W. Przewozniczek, Tomasz M. Durda · 2020
Dependency Structure Matrix Genetic Algorithm-II (DSMGA-II) is a recently proposed optimization method that builds the linkage model on the base of the Dependency Structure Matrix (DSM). This model is used during the Optimal Mixing (OM) operators, such as the Restricted Mixing (RM) and the Back Mixing (BM). DSMGA-II was shown to solve theoretical and real-world optimization problems effectively. In this paper, we show that the effectiveness of DSMGA-II and its improved version, namely Two-edge Dependency Structure Matrix Genetic Algorithm-II (DSMGA-IIe), is relatively low for NK-landscape problems. Thus, we propose the Comparative Mixing (CM) operator that extends the RM operator. The CM operator modifies the linkage information obtained from the DSM-based linkage model by comparing the receiver individual with a randomly selected member of the population. Such modification enables DSMGA-II to solve NK-landscape problems effectively and does not limit DSMGA-II performance on most problems for which it was already shown effective.