An Improved Backtracking Search Algorithm for Global Optimization

Population-based optimization algorithms have been widely applied to optimization problems in many fields and have achieved encouraging computational results. To improve the performance of population-based optimization algorithms, it is necessary to adopt targeted improvement techniques to improve the algorithms, which is also a major direction for the current development of population-based optimization algorithms. To improve the performance of population-based optimization algorithms, this paper proposes a new technique for improving metaheuristic algorithms from the perspective of average operators, which defines six averaging operators called global average position, upper average position, lower average position, random global average position, random upper average position, and random lower average position. These six average operators define the average position from different perspectives. To verify the validity of the defined average operators, they are applied to the classic backtracking search algorithm (BSA), and the average operator-driven backtracking search algorithm (AOBSA) is proposed to overcome the disadvantage of BSA’s single search strategy being prone to getting stuck in local optima. To investigate the performance of the AOBSA, it is used to solve 30 challenging test functions with 30 and 50 dimensions, along with two classical engineering design problems. The experimental results demonstrate the excellent performance of the AOBSA in solving numerical problems and engineering design problems. This also demonstrates the effectiveness of the defined average operators and their potential to be applied to other population-based optimization algorithms.