Dynamic Multi-objective Estimation of Distribution Algorithm based on Domain Adaptation and Nonparametric Estimation

Although dynamic optimization and multi-objective optimization have made considerable progress individually, solving dynamic multi-objective optimization problems remains a monumental challenge since their multiple, conflicting objectives could change over time. In this paper, we propose a Domain Adaptation and Nonparametric Estimation-based Estimation of Distribution Algorithm, called DANE-EDA, to solve dynamic multi-objective optimization problems. Notable features of the proposed algorithm include the importance sampling, nonparametric density estimation, probabilistic prediction, and a domain adaptation technique seamlessly unified under an innovative framework. The design takes full advantage of the powerful Monte-Carlo method and transfer learning technique. This kind of combination will help the proposed algorithm to maintain a delicate exploration-exploitation trade-off from temporal and spatial perspectives. At the same time, it will help the proposed algorithm to overcome the shortcomings caused by transfer learning, specifically, the loss of the diversity. After proving convergence and analyzing the computational complexity of the DANE-EDA, we compare the proposed method with nine EDAs or dynamic multi-objective optimization algorithms on twelve different test instances. The experimental results affirm the effectiveness of the proposed method in addressing dynamic multi-objective optimization problems.