A Contour Method in Population-based Stochastic Algorithms

Abstract

Inspired by the contours in topography, this paper proposes a contour method for the population-based stochastic algorithms to solve the problems with continuous variables. Relying on the existed population, the contour method explores the landscape of the fitness function in the search space, which leads to effective speculation about the positions of the potential optima. The contour method is embedded into every generation of the simple genetic algorithm (SGA) for efficiency examination. The genetic algorithm with the contour method is first realized in a two-dimensional space, where the contours in topography can be directly used. Then the proposed contour method is modified to adapt high dimensional space. Numerical optimization experiments are carried out on ten benchmark functions of two and thirty dimensions. Results show that the genetic algorithm with the contour method can outperform the SGA in both solution quality and convergence speed.