Many-objective firefly algorithm for solving large-scale sparse optimization problems

The multi-objective firefly algorithm is difficult to ensure the sparsity of the Pareto optimal solutions when dealing with large-scale sparse multi-objective optimization problems, and when the objective dimension of the optimization problem is too large, it will also lead to the failure of Pareto dominance and the slowdown of convergence. In view of this, this paper proposes a many-objective firefly algorithm based on dynamic scoring and neighborhood search (SMaOFA). The algorithm generates sparse initial population based on the dual-coding hybrid ensemble, and proposes a dynamic scoring strategy, which dynamically updates the decision variable score at each round of iteration to provide prior knowledge for subsequent iterations to ensure the sparsity of the solution set. According to the concept of fuzzy dominance and the Euclidean distance between fireflies, a neighborhood search strategy is proposed, which discards the influence of the full attraction model on the convergence speed of the algorithm, and avoids the failure of Pareto dominance caused by the large objective dimension. The linear adjustment factor is introduced to improve the position update formula of fireflies and improve the search ability of the population. Experimental results show that the proposed algorithm has efficient performance when dealing with large-scale sparse multi-objective optimization problems. © 2024 Northeast University. All rights reserved.