Abstract
This paper proposes a novel algorithm for solving continuous complex optimization problems with a relatively low memory consumption. The proposed approach, namely Composed compact Differential Evolution, consists of a set of compact Differential Evolution units which simultaneously search the decision space from various perspectives. A randomization in the virtual population allows the algorithm to behave, on one hand, as a multiple local search with a multi-start logic integrated within it. On the other hand, the compact units communicate among each other by means of a ring topology and propagation of information. More specifically, the most promising elite solutions and scale factor values of each compact unit are migrated to the neighbour unit so that the search of the global optimum is performed. In other words, while each single compact unit performs a local search by exploiting the direction suggested by each elite solution, the entire structure combines the achievement of each local search operation towards the direction of the global search. The proposed algorithm is characterized by a limited memory consumption and is memory-wise equivalent to a population-based algorithm with a small population. Numerical results show that the proposed approach outperforms other compact algorithms and various modern population-based structures.