A multimodal approach for evolutionary multi-objective optimization (MEMO): Proof-of-principle results

Abstract

Most evolutionary multi-objective optimization (EMO) methods use domination and niche-preserving principles in their selection operation to find a set of Pareto-optimal solutions in a single simulation run. However, classical generative multi-criterion optimization methods repeatedly solve a parameterized single-objective problem to achieve the same. Due to lack of parallelism in the classical generative methods, they have been reported to be slow compared to efficient EMO methods. In this paper, we use a specific scalarization method, but instead of repetitive independent applications, we formulate a multimodal scalarization of multiple objectives and develop a niche-based evolutionary algorithm to find multiple Pareto-optimal solutions in a single simulation run. Proof-of-principle results on two to 10-objective problems from our proposed multimodal approach are compared with standard evolutionary multi/many-objective optimization methods.