Multiple source transfer learning for dynamic multiobjective optimization

Recently, dynamic multiobjective evolutionary algorithms (DMOEAs) with transfer learning have become popular for solving optimization problems (DMOPs), as the used methods in DMOEAs can effectively generate a good initial population new environment. However, most of them only non-dominated solutions from previous one or two environments, which cannot fully exploit all historical information and may easily induce negative limited knowledge is available. To address this problem, paper presents multiple source method DMOEA, called MSTL-DMOEA, runs procedures to environments. First, select some representative transfer, clustering-based manifold run cluster last environment obtain their centroids, are then fed into model predict corresponding centroids After that, further by using multisource TrAdaboost, above old aiming construct more accurate prediction model. This way, MSTL-DMOEA an better quality The experimental results also validate superiority over several competitive state-of-the-art various kinds DMOPs.