A dynamic multi-objective evolutionary algorithm using center and multi-direction prediction strategies.

Area of Science:

• Optimization
• Computational Intelligence
• Algorithm Design

Background:

• Dynamic multi-objective optimization problems (DMOPs) present challenges due to shifting Pareto fronts (PF) and Pareto sets (PS).
• Existing strategies often focus on accelerating convergence and maintaining population diversity in dynamic environments.
• Prediction strategies are common but face efficiency limitations.

Purpose of the Study:

• To develop a more efficient prediction model for dynamic multi-objective optimization.
• To improve the performance of algorithms in dynamic environments by enhancing prediction accuracy.
• To address the key challenge of increasing prediction efficiency in dynamic optimization.

Main Methods:

• A novel prediction model is proposed, utilizing the rank sums of individuals.
• A metric is defined based on the position difference of individuals across two adjacent environments to classify change types.
• The prediction strategy is tailored based on identified environment change types.

Main Results:

• The proposed algorithm was compared against five state-of-the-art approaches.
• Experiments were conducted on 20 benchmark instances of dynamic multi-objective problems.
• The algorithm demonstrated effective convergence and distribution properties in dynamic settings.

Conclusions:

• The developed prediction strategy effectively handles dynamic multi-objective optimization problems.
• The approach shows promise for improving algorithmic performance in environments with changing objectives or constraints.
• The method offers a viable solution for enhancing prediction efficiency in dynamic optimization contexts.