Elite Opposition-Based Differential Evolution for Solving Large-Scale Optimization Problems and Its Implementation on GPU

Recently, the interests of solving large-scale optimization problems have increased in the field of evolutionary algorithms. This paper presents a novel differential evolution, namely EOBDE, to solve these kinds of problems by using elite opposition-based learning strategy. In the proposed algorithm, the opposite solutions of some selected elite individuals from the current population are generated at a certain probability for generation jumping. Then a corresponding opposite population is constructed to compete with the current population for providing more chances of finding out the global optimum. This approach is helpful to obtain a tradeoff between exploration and exploitation ability of DE. As another contribution, a parallel version of the proposed algorithm is implemented on Graphics Processing Units (GPU) based on CUDA platform for accelerating computing speed. The experiments are carried out on a set of representative problems with D=500 and 1000. The results of EOBDE are compared with other four state-of-the-art evolutionary algorithms in order to investigate the performance, which show that our proposed algorithm outperform the compared algorithms in terms of solution accuracy. Also the parallel version based on GPU shows promising performance in terms of the computational time.