Evolutionary programming based on uniform design with application to multiobjective optimization

Abstract

Pareto-optimality is one of the important methods to solve multiobjective optimization problems. It is desirable to find as much as possible Pareto-optimal solutions, and it is also highly expected to find the ones scattered uniformly over the Pareto frontier such that a variety of compromise solutions can be provided to the decision maker. For this purpose, an evolutionary programming algorithm, called evolutionary programming based on uniform design (UDEP), is proposed in this paper. Uniform design technique is used to define some fitness functions which can guide the search evenly toward the Pareto frontier. In order to overcome premature and provide as much as possible candidate solutions evenly scattered in the whole search space, uniform design technique, variable region search, as well as niche technique are used. Uniform design makes it possible to explore the search space evenly, while variable region search and niche technique help to keep diversity of the population. Their combination improves the search ability of EP (evolutionary programming) significantly. Many numerical experimental results show the usefulness of the proposed method.