A Step-Size Adaptive Hill-Climbing Algorithm for Local Search

This paper proposed a step-size adaptive local search (SSALS) algorithm. SSALS is designed for single individual search, and can also be used in multiple individuals search when combined with population-based algorithms. In SSALS, each dimension possesses a basic step size. For each iteration, one or more dimensions are randomly selected to mutate. The basic step size of these dimensions will be adjusted automatically based on the fitness of the new position. The SSALS was compared with five state-of-the-art algorithms by low- and high-dimensional test experiments. A set of 28 benchmark functions in the CEC’2013 test suite on real-parameter optimization is used for low-dimensional experiments, and for high-dimensional experiments, a set of 15 benchmark functions in the CEC’2013 test suite on large-scale global optimization is utilized. The experimental results indicate that the performance of SSALS is significantly higher than the other algorithms, especially in high-dimensional optimization.