A dynamic multi-objective evolutionary algorithm using dual-space prediction and surrogate-based sampling

Area of Science:

• Optimization
• Evolutionary Computation
• Artificial Intelligence

Background:

• Dynamic multi-objective optimization problems (DMOEPs) require algorithms to track Pareto-optimal solutions amidst environmental changes.
• Existing prediction-based dynamic multi-objective evolutionary algorithms (DMOEAs) often use single-space prediction or identical models for both decision and objective spaces, limiting effectiveness in complex dynamics.
• Sampling methods in DMOEAs can lead to significant computational burdens due to excessive function evaluations.

Purpose of the Study:

• To propose a novel dynamic multi-objective evolutionary algorithm (DS-DMOEA) that efficiently adapts to environmental changes.
• To address the limitations of existing DMOEAs in capturing distinct space dynamics and managing computational load.

Main Methods:

• DS-DMOEA employs a dual-space prediction strategy: a weight vector-based method for the objective space and a geodesic flow kernel method for the decision space.
• A surrogate-based sampling strategy is utilized to generate high-quality initial populations for new environments by training surrogate models on historical data.
• The predicted and sampled populations are combined to form an optimized initial population for the evolving environment.

Main Results:

• DS-DMOEA was rigorously tested against nine state-of-the-art DMOEAs on 19 benchmark problems.
• The algorithm demonstrated effectiveness across three distinct environmental change patterns.
• Experimental results validated the superior performance of the proposed DS-DMOEA.

Conclusions:

• The proposed DS-DMOEA effectively adapts to dynamic multi-objective optimization problems through its dual-space prediction and surrogate-based sampling strategies.
• The algorithm overcomes limitations of existing methods by capturing complex dynamics in both decision and objective spaces while managing computational cost.
• DS-DMOEA represents a significant advancement in efficiently handling dynamic multi-objective optimization challenges.