Batched Data-Driven Evolutionary Multiobjective Optimization Based on Manifold Interpolation

Multiobjective optimization problems are ubiquitous in real-world science, engineering, and design problems. It is not uncommon that the objective functions as a black box, evaluation of which usually involve time-consuming and/or costly physical experiments. Data-driven evolutionary can be used to search for set nondominated tradeoff solutions, where expensive approximated surrogate model. In this article, we propose framework implementing batched data-driven multiobjective (EMO). so general any off-the-shelf EMO algorithms applied plug-in manner. There two unique components: 1) based on Karush–Kuhn–Tucker conditions, manifold interpolation approach explores more diversified solutions with convergence guarantee along Pareto-optimal 2) batch recommendation reduces computational time process by evaluating multiple samples at parallel. Comparing against seven state-of-the-art surrogate-assisted algorithms, experiments 168 benchmark test problem instances various properties application hyper-parameter fully demonstrate effectiveness superiority our proposed framework, featured faster stronger resilience front shapes.