Analysis and improvement of GSA's optimization process

Abstract Gravitational search algorithm (GSA) is one of the heuristic algorithms proposed in recent years, which is inspired by the law of universal gravitation between masses. However, many practical applications and researches show that when the region affected by global optimum occupies less search space, GSA is prone to fall into local optimum, especially when the optimal value is close to the boundary of the search space and there are sub-optimal solutions in the center of the space. By microscopic analysis of the particle motion process of GSA in the above optimization situation, we find that the Kbest mechanism and the characteristic of central convergence are the two main factors affecting the GSA optimization performance. In this paper, an improved algorithm called Balanced Gravitational Search Algorithm is proposed, in which the balance operator is designed to solve two inherent problems in GSA. Then the proposed method is firstly tested on 10 benchmark functions provided by CEC 2020 compared with the state-of-the-art variant algorithms of the GSA and other typical meta-heuristics. Further, the algorithms are tested and compared on the real-world optimization problems including CEC 2011 real-world optimization problems and the Multi-Layer Neural Network (MLNN) training problem based on wine dataset. The simulation results show that BGSA can significantly improve the optimization performance of GSA and it can be a good choice for solving real-world optimization problems.