An adaptive dual-population evolutionary paradigm with adversarial search: Case study on many-objective service consolidation

Abstract

Optimizing many conflicting objectives simultaneously is one of the most challenging topics in the multi-criterion decision-making. This paper develops a dual-population co-evolutionary paradigm for solving many-objective service selection problems. It evolves two co-evolving populations separately with different scalarizing functions (SFs) and adversarial search orientations in parallel. In particular, one population, driven by convergence-oriented SF with ideal point, pulls the solutions toward the Pareto front; the other one, driven by diversity-oriented SF with nadir point, pushes the solutions backward from the nadir point. Accordingly, the search behaviors of the two populations are arguably complement to each other. Moreover, corner solutions and angle-based similarity are employed to enhance the coverage of population as widely as possible, the interaction and collaboration among populations are leveraged by a carefully crafted elitism pairing strategy. A series of experimental studies have been performed on challenging real-world service composition problems. Empirical results have demonstrated the competitiveness of our proposal against the state-of-the-art peers.