Increasing robustness of differential evolution by passive opposition

Abstract

The differential evolution (DE) algorithm is known to be fairly robust among various global optimization algorithms. However, the application of this algorithm to an extensive set of test functions shows DE fails at least partially on 19% of the test functions, and completely on 6.9%of the test functions. The opposition-based DE in the literature, while aimed to improve the efficiency of DE, has a robustness even worse than that of the classic DE. In this paper we describe a new variant of DE called “passive oppositional differential evolution” (PODE) which utilizes opposition in such a way that a set of opposite vectors, while not part of the population, are used in mutation to gain diversity. Numerical experiments show that compared to DE and ODE, it has a much better robustness and a similar speed of convergence.