Engineering applications of multi-objective evolutionary algorithms: A test suite of box-constrained real-world problems

Traditionally, the performance of novel multi-objective optimization algorithms is evaluated on artificial test problems, which are constructed under the consideration of some features observed in real-world problems. However, artificial test problems do not possess all the properties that real-world applications have. Therefore, the evaluation of algorithms on artificial test problems can overestimate the conclusions of an algorithm regarding its performance in real-world applications. In this respect, this paper presents a collection of multi -objective real-world problems taken from different disciplines. The multi-objective problems are suggested to complement the performance evaluation of evolutionary algorithms considering real-life applications without the need to be an expert on the concerned disciplines in which they lie. This study analyzes the conflict between objectives from the correlation point of view of each real-world problem. Notably, this investigation inquires on the real-world multi-objective problems regarding the Pareto front shapes, which have been one of the main difficulties for selection mechanisms in multi-objective evolutionary algorithms. Additionally, the performance of nine state-of-the-art multi-objective evolutionary algorithms based on the main principles of evolutionary computation is analyzed. Although it is impractical to evaluate all the existing algorithms available in the multi-objective evolutionary community, this investigation gives insights into the performance of the main multi-objective approaches considered in this study.

Automated summary

Traditionally, novel multi-objective optimization algorithms are evaluated on artificial test problems, which lack the properties of real-world applications. This paper presents a collection of multi-objective real-world problems from different disciplines to complement the evaluation of evolutionary algorithms. The study analyzes the conflict between objectives for each real-world problem and compares the performance of different multi-objective evolutionary algorithms.