A fitness-based two roles adaptive inertia weight particle swarm optimization

Particle Swarm Optimization (PSO) has been widely applied to practical problems. Similar to other evolutionary algorithms, PSO is prone to premature convergence and local optima entrapment. To achieve a dynamic balance between exploration and exploitation, this paper proposes a fitness-based two roles adaptive inertia weight particle swarm optimization (FAIWPSO). The proposed algorithm divides the population into elite and ordinary subgroups according to individual fitness values, and adaptively reduces the number of elite particles as the iteration proceeds, thereby realizing a smooth transition from global exploration to local exploitation. The elite subgroup identifies and guides the population toward promising regions using neighborhood best information, while maintaining population diversity through non-uniform mutation. Guided by the elite subpopulation, the ordinary subpopulation adopts a weighted learning strategy to intensively exploit promising regions. Furthermore, to achieve a better balance between exploration and exploitation, a nonlinear adaptive inertia weight strategy based on both population evolution state and individual differences is introduced. Additionally, a dimension-adaptive Gaussian mutation strategy is developed, which mutates different dimensions of the global best solution depending on the evolutionary stage, enhancing the ability to escape local optima. To evaluate the effectiveness of FAIWPSO, comprehensive experiments were conducted, demonstrating that the proposed strategies substantially enhance the algorithm’s performance. On the CEC2014 test suite, FAIWPSO outperforms seven widely used PSO variants in terms of solution accuracy and stability. On the CEC2018 test suite, its overall performance surpasses that of APGSK-IMODE and MadDE. • FAIWPSO is a new PSO variant with high exploitation and exploration capabilities. • Adaptive inertia weight is adapted to the evolution state and fitness. • Weighted learning strategy based on fitness. • Dimension-adaptive Gaussian mutation avoids local optima via staged perturbation.