Dynamic multiobjective optimization with varying number of objectives assisted by dynamic principal component analysis

Dynamic multi-objective optimization problems, which are equipped with the increment or decrement number of time-varying objective functions, have been hardly researched in recent decades. Different from other dynamism handling approaches, we propose a new framework incorporated with the dynamic principal component analysis technique, which embeds the acquired knowledge of Pareto optimal set during the evolutionary search process. As the environmental changes occur, the dynamic principal component analysis technique learns the global structure of Pareto optimal set incrementally as newly generated data are collected to depict the manifold contour. In addition, this method constructs high-quality solutions on the basis of obtained knowledge, which in turn captures the main structure of previous solutions. We undertake comprehensive experiments in which the benchmark instances are given with a varying number of objective functions and the computational values are assessed with respect to performance metrics. The obtained statistical findings with 70% improvement fully demonstrate that our proposed algorithm is efficient and effective for solving dynamic multi-objective optimization problems.