Improved sampling of the pareto-front in multiobjective genetic optimizations by steady-state evolution: a pareto...

Area of Science:

• Computational Intelligence
• Optimization Algorithms
• Evolutionary Computation

Background:

• Previous multiobjective genetic algorithms primarily focused on preventing genetic drift.
• The critical issue of convergence towards the Pareto-front received limited attention in prior research.
• Existing methods often rely on complex sharing/niching mechanisms and numerous heuristic parameters.

Purpose of the Study:

• To introduce a simple, steady-state strategy, the Pareto Converging Genetic Algorithm (PCGA), for effective multiobjective optimization.
• To ensure natural sampling of the solution space and advancement towards the unknown Pareto-front.
• To minimize reliance on heuristic parameters and complex procedures like niching.

Main Methods:

• Developed PCGA, a steady-state genetic algorithm strategy.
• Introduced a systematic convergence assessment using histograms of rank.
• Optionally incorporated a nonmigrating island model to enhance diversity and competition.

Main Results:

• PCGA naturally samples the solution space and drives populations towards the Pareto-front.
• The algorithm eliminates the need for sharing/niching, reducing heuristic parameter dependence.
• Demonstrated diverse Pareto-front sampling with significantly less computational effort on benchmark problems.

Conclusions:

• PCGA offers an efficient and effective method for multiobjective optimization, focusing on convergence.
• The proposed convergence assessment method provides a systematic way to evaluate progress towards the Pareto-front.
• PCGA's design, including the optional island model, is well-suited for complex real-world problems prone to local optima.