A Multi-Objective Decomposition Based Evolutionary Algorithm Using Adaptive Searching Method
Xiaofang Guo · 2021 17th International Conference on Computational Intelligence and Security (CIS) · 2021
In the research of multi-objective evolutionary algorithms (MOEAs), model-based estimation of distribution algorithms (MEDAs) are one of the important branches to build probabilistic models for generating new candidate solutions. For example, inverse modeling (IM) based MOEA takes advantage of the dependencies from objective space to decision space, and constructs some Gaussian process-based regression (GPR) inverse models for creating offspring. Motivated by the idea of IM-MOEA using trained Gaussian process regression model to generate offspring, an adaptive searching approach (ASIM-MOEA) is proposed in this paper. The estimated range for locating the test input points in GPR is dynamically adjusted according to the different stage of evolution, and a greedy sampling method for choosing the new test samples is also adopted to develop the unexplored sparse area in objective space to improve the diversity. The experimental results show that the proposed strategy achieves a better performance than state-of-the-art compared algorithms.