DE with Random Vector based Mutatiton for High Dimensional Problems

Abstract

Metaheuristic techniques are the current standard for solving optimization problems. Differential Evolution (DE) is one of the most used because all operations are on real floating point numbers and does not require extra coding. However, the performance shown by DE could decay when applied in problems of high dimensionality. In this paper we present RLSDE, a modified version of DE, based on a random vector as a scaling factor for the differential mutation and the application of a local search operator. These modifications constitute an algorithm capable of solving 100D problems using few computational resources. RLSDE is compared against the results obtained with the classic version of DE and ELSDE (Enchanced Local Search Differential Evolution), showing the performance of the proposal.