A Study of the Effect of Different Parameter Assignment Strategies on the Performance of ACLPSO

Adaptive comprehensive learning particle swarm optimization (ACLPSO) is a variant of particle swarm optimization (PSO) we previously proposed, which achieves a good balance between exploration and exploitation. In ACLPSO, the interval scaling coefficient s and the learning probability tradeoff coefficient v are crucial parameters influencing the convergence performance of the algorithm. ACLPSO requires manual setting of these two parameters to handle different benchmark functions, but there are no specific rules for assigning their values. This paper proposes different assignment strategies for s and v, allowing them to periodically change according to different function curves or adaptively adjust based on the change rate of global best fitness value, resulting in different combinations of s and v at different stages of the iteration, including the most appropriate combination of s and v at that stage. In this paper, these parameter assignment strategies are compared on different benchmark functions, and the experimental results show that five parameter assignment strategies can achieve the global optimum or a near optimum on certain benchmark functions. Although the final results of the five assignment strategies are overall not better than the ACLPSO under the most appropriate combination of s and v, this study lays the foundation for our subsequent research on ACLPSO to jump out of local optima.