Abstract: Regarding to the evolution characteristics of standard bat algorithm, a bat algorithm with the capability of self-learning and individual variation was proposed. In this proposed algorithm, the best global individual with the self-learning capability could self-optimize within a small range of solutions and lead to other individuals develop deep searching. In addition, the each individual generated a dynamic number variation cluster in proportion its fitness value. According to the rule of greedy selection, the best individual in the variation cluster was selected which protected the excellent individual and avoided the individual degradation. The proposed algorithm made use of the self-learning and individual variation improved the optimization accuracy and convergence speed. The simulation results for the standard test functions show that the improved bat algorithm has significant advantage of high optimization ability and search precision, and can skip from local optimum effectively. The improved bat algorithm has great value to engineering of complex function optimization.

Key words: bat algorithm, self-learning, variation selection, optimization accuracy, multi-dimensional function optimization