Evolutionary Large-Scale Multiobjective Optimization Via Self-guided Problem Transformation

Songbai Liu,Min Jiang,Qiuzhen Lin,Kay Chen Tan

DOI: https://doi.org/10.1109/cec55065.2022.9870259

2022-01-01

Abstract:The performance of traditional multiobjective evolutionary algorithms (MOEAs) often deteriorates rapidly when using them to solve large-scale multiobjective optimization problems (LMOPs). To effectively handle LMOPs, we propose a large-scale MOEA via self-guided problem transformation. In the proposed optimizer, the original large-scale search space is transferred to a lower-dimensional weighted space by the guidance of solutions themselves, aiming to effectively search in the weighted space for speeding up the convergence of the population. Specifically, the variables of the target LMOP are adaptively and randomly divided into multiple equal groups, and then solutions are self-guided to construct the small-scale weighted space correspondingly to these variable groups. In this way, each solution is projected as a self-guided vector with multiple weight variables, and then new weight vectors can be generated by searching in the weighted space. Next, new offspring is produced by inversely mapping the newly generated weight vectors to the original search space of this LMOP. Finally, the proposed optimizer is tested on two different LMOP test suites by comparing them with five competitive large-scale MOEAs. Experimental results show some advantages of the proposed algorithm in solving the considered benchmarks.