Solving Many-Objective Optimization Problems by a Pareto-Based Evolutionary Algorithm With Preprocessing and a...

Area of Science:

• Computational Intelligence
• Optimization Theory
• Algorithm Design

Background:

• Pareto-based algorithms are standard for multiobjective optimization but struggle with numerous objectives.
• Existing methods falter as Pareto dominance becomes ineffective with >3 objectives.
• Diversity estimators in current algorithms are biased towards dominance resistance solutions (DRSs).

Purpose of the Study:

• To develop a novel Pareto-based algorithm for many-objective optimization problems.
• To overcome the limitations of traditional Pareto-based approaches in high-dimensional objective spaces.
• To improve the performance of multiobjective optimization algorithms by addressing DRSs and balancing convergence/diversity.

Main Methods:

• A new Pareto-based algorithm incorporating an interquartile range method for preprocessing solution sets.
• Implementation of a penalty mechanism with alternating selection and penalty operations to manage convergence and diversity.
• Comparative analysis against five state-of-the-art algorithms on diverse benchmarks with 3-15 objectives.

Main Results:

• The proposed algorithm effectively preprocesses solution sets to eliminate dominance resistance solutions.
• The penalty mechanism successfully balances convergence towards the Pareto front and population diversity.
• Experimental results demonstrate strong performance across various test functions and objective numbers.

Conclusions:

• The novel Pareto-based algorithm offers a robust solution for many-objective optimization problems.
• The integration of interquartile range preprocessing and a penalty mechanism significantly enhances algorithm performance.
• The proposed method generally outperforms existing state-of-the-art algorithms in handling complex multiobjective scenarios.