A Multimodal Approach for Evolutionary Multi-objective Optimization (MEMO): Proof-of-Principle Results

Most evolutionary multi-objective optimization (EMO) methods use domination and nichepreserving 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 the multi-objective or many-objective optimization problem and develop a niche-based evolutionary algorithm (MEMO) to find multiple Pareto-optimal solutions in a single simulation run. Proof-of-principle results on two to 10-objective unconstrained and constrained problems using our proposed multimodal approach are able to find hundreds of Pareto-optimal solutions. The proposed MEMO approach is also compared with stateof-the-art evolutionary multi/many-objective optimization methods. MEMO is then applied to a number of engineering design problems. Results are promising and provide a new direction for solving multiand many-objective optimization problems.