Solving dynamic multi-objective problems with a new prediction-based optimization algorithm.

Area of Science:

• Computational intelligence
• Optimization algorithms
• Dynamic systems

Background:

• Multi-objective optimization problems (MOOPs) are common in real-world scenarios.
• Dynamic environments present challenges for traditional optimization algorithms due to their evolving nature.
• Existing algorithms often struggle to adapt to rapid changes in objective functions or constraints.

Purpose of the Study:

• To develop a novel dynamic multi-objective optimization algorithm.
• To enhance the ability of algorithms to track moving Pareto-optimal sets in changing environments.
• To improve the convergence and diversity of solutions in dynamic multi-objective optimization.

Main Methods:

• Integration of a fitting-based prediction (FBP) mechanism with a regularity model-based multi-objective estimation of distribution algorithm (RM-MEDA).
• Implementation of a prediction-based reaction mechanism with three subpopulations for effective tracking.
• Utilizing a simple linear prediction model, FBP strategy, and a recent sampling strategy for subpopulation generation.

Main Results:

• The proposed algorithm demonstrates competitive effectiveness in dynamic multi-objective optimization.
• Experimental results show superior performance on benchmark functions with diverse dynamic characteristics.
• The three-subpopulation approach effectively tracks the moving Pareto-optimal set.

Conclusions:

• The novel FBP-integrated RM-MEDA algorithm offers a robust solution for dynamic multi-objective optimization.
• The prediction-based reaction mechanism significantly improves adaptation to environmental changes.
• The algorithm provides a promising direction for future research in adaptive optimization techniques.