Yang Zhonghua, Cai Yuanli, Li Ge, Wang Peng, Chen Yu
Faculty of Electronics and Information Engineering, Xi'an Jiaotong University, Xian, 710049, Shaanxi, China.
College of Systems Engineering, National University of Defense Technology, Changsha, 410073, Hunan, China.
Sci Rep. 2025 Aug 12;15(1):29533. doi: 10.1038/s41598-025-13215-9.
To address the shortcomings of the gravitational search algorithm, such as its tendency to fall into local optima, slow convergence, and low solution accuracy, this paper proposes a gravitational search algorithm based on multi-strategy cooperative optimization. The proposed algorithm balances global exploration and local exploitation. In the early iterations, particle positions are primarily updated using the original gravitational force, preserving the inherent characteristics of the gravitational search algorithm. In the later stages, particles with better fitness values are updated using a globally optimal Lévy random walk strategy to enhance local search capabilities, while particles with poorer fitness values are updated using the sparrow algorithm follower strategy. This approach increases the exploration of the particles in unexplored local areas, further improving the local exploitation abilities of the algorithm. Finally, the lens-imaging opposition-based learning strategy generates opposite solutions for particles at different stages, increasing population diversity, expanding the search range, and enhancing the global search performance of the algorithm. An effectiveness analysis and algorithm comparison tests were carried out on 24 typical complex benchmark functions. The performance analysis results show that the multi-strategy collaborative optimization method effectively leverages both the global and local search abilities of the algorithm, improving the accuracy of its solutions and stability. Compared with other GSA-based algorithms and advanced intelligent algorithms, the proposed algorithm exhibits superior solution accuracy, convergence speed, and stability, making it an efficient GSA-based algorithm. In addition, the proposed algorithm was applied to three engineering design optimization problems to verify its applicability to real-world scenarios.
为解决引力搜索算法存在的易陷入局部最优、收敛速度慢和求解精度低等缺点，本文提出一种基于多策略协同优化的引力搜索算法。所提算法平衡了全局探索和局部开发能力。在迭代初期，粒子位置主要通过原始引力进行更新，保留引力搜索算法的固有特性。在后期阶段，对适应度值较好的粒子采用全局最优的莱维随机游走策略进行更新，以增强局部搜索能力，而对适应度值较差的粒子则采用麻雀算法跟随者策略进行更新。这种方法增加了粒子在未探索局部区域的探索，进一步提高了算法的局部开发能力。最后，基于透镜成像的反向学习策略为不同阶段的粒子生成相反解，增加种群多样性，扩大搜索范围，提高算法的全局搜索性能。对24个典型复杂基准函数进行了有效性分析和算法比较测试。性能分析结果表明，多策略协同优化方法有效利用了算法的全局和局部搜索能力，提高了解的精度和稳定性。与其他基于引力搜索算法的算法和先进智能算法相比，所提算法具有更高的求解精度、收敛速度和稳定性，是一种高效的基于引力搜索算法的算法。此外，将所提算法应用于三个工程设计优化问题，验证了其在实际场景中的适用性。
