Gene Targeting Particle Swarm Optimization for Large-Scale Optimization Problem

With the development of information technology, solving large-scale optimization problems (LSOPs) has become an urgent issue in the artificial intelligence (AI) for solving practical problems. LSOPs are optimization problems with high dimensions and large search space. Although many improved evolutionary computation (EC) algorithms have been proposed as promising AI approaches for dealing with LSOPs, they still face two challenges, which are easily falling into local optima and having slow convergence speed. Focusing on these two challenges, this paper proposes to integrate the gene targeting (GT) technology in the biotechnology with the particle swarm optimization (PSO), so as to propose a gene targeting particle swarm optimization (GTPSO) algorithm. The GTPSO works well in solving LSOPs due to the following three novel designs. Firstly, a Monte Carlo probabilistically targeting strategy is used to determine the poorly-performing dimension(s) of the globally best particle, so as to balance the targeting chance of all the dimensions. Secondly, a dual GT (DGT) strategy is proposed, including two GT strategies called GT1 and GT2, to modify the poorly-performing dimension(s), which can balance diversity and convergence at the same time. Thirdly, a greedy strategy is used to accept the targeted new solution only if it is better than the original globally best particle, so as to ensure the quality of the evolution. In the experiment, the GTPSO algorithm is evaluated on 12 benchmark functions and obtains excellent results, which demonstrates the effectiveness of the GT strategy and the GTPSO algorithm.