CMACGSA: Improved Gravitational Search Algorithm Based on Cerebellar Model Articulation Controller for Optimization

Abstract

Metaheuristic algorithms have gained significant attention in recent years for addressing complex and challenging optimization problems, especially in engineering. These algorithms often take inspiration from natural phenomena, systems or biological behaviour to find optimal solutions. Recent advances in the field often involve hybrid methods that combine several algorithms to improve performance. This study introduces an improved Gravitational Search Algorithm, named CMACGSA, which incorporates the Cerebellar Model Articulation Controller (CMAC)-a neural network model-to enhance the performance of Gravitational Search Algorithm (GSA). By employing the CMAC neural network, CMACGSA dynamically learns the masses of particles/agents of GSA, enabling a learning-driven approach to mass computation. Additional enhancements include Lévy mutation, boundary control methods and an error handling mechanism, which together improve the robustness and adaptability of the algorithm. The effectiveness of CMACGSA is demonstrated through extensive testing on a set of 2D CEC 2014 benchmark functions, where it significantly outperforms the original GSA. Further evaluations on multidimensional CEC 2014 test problems, including 30-dimensional cases, reveal improved performance over widely used optimization algorithms and state-of-the-art (SOTA) algorithms. Furthermore, CMACGSA consistently achieves top-tier average performance metrics when benchmarked against four well-established GSA variants. The applicability of the algorithm is further validated by engineering design problems where it demonstrates outstanding performance, confirming its value in solving complex engineering challenges.