A Novel Bat Algorithm using Sobol Sequence for the Initialization of Population

Abstract

Evolutionary and Swarm intelligence algorithms are widely used for global optimization problems. Bat Algorithm is one such algorithm based on a population-based metaheuristic algorithm. Bat Algorithm faces the premature convergence problem in which bats get stuck in local optimum and cannot reach a global optimum safely. The initial population generation of the bats plays an important role to move efficiently in a d-dimensional search space. The Standard Bat Algorithm (StdBA) uses random distribution for the initialization of bats. In this paper, we propose a novel variant of the Bat algorithm, namely, the Bat Algorithm using the Sobol sequence (BASobol). In this variant, the exploration abilities of StdBA are enhanced using Sobol distribution instead of random distribution. Furthermore, the velocity update equation is also modified using Inertia Weight. Ten different standard benchmark functions are compared with the proposed technique to test its validity. Statistical results of BASobol are compared with StdBA and Modified Bat Algorithm (MBA). Superior results are seen by BASobol over seven benchmark functions.