Data-driven Multiobjective Particle Swarm Optimization based on Data Augmentation Strategy

Abstract

In some offline data-driven optimization problems, only small data can be gathered from real applications, which may decrease the reliability of surrogate models. To overcome the issues mentioned above, a data augmentation strategy based on generative adversarial networks (GANs) is adopted to improve the performance of data-driven multiobjective particle swarm optimization (DDMOPSO). Besides the fitting information of the original data, the distribution information is also considered to create synthetic data. The synthetic data is utilized to increase the accuracy of surrogate models. Therefore, a novel offline data-driven multiobjective particle swarm optimization based on data augmentation strategy (DDMOPSO-A) is proposed in this paper. The experiment results show that the proposed algorithm is superior to four competitive algorithms on seven benchmark problems.