A Fast Dynamic Evolutionary Multiobjective Algorithm via Manifold Transfer Learning.

Area of Science:

• Optimization
• Artificial Intelligence
• Computational Science

Background:

• Dynamic multiobjective optimization problems (DMOPs) present challenges in tracking evolving Pareto-optimal fronts.
• Existing transfer learning (TL) methods for DMOPs are often computationally intensive and time-consuming.

Purpose of the Study:

• To develop a more efficient and accurate method for solving DMOPs.
• To reduce the computational cost associated with existing TL-based approaches.

Main Methods:

• Proposes a memory-driven manifold TL-based evolutionary algorithm for DMOPs (MMTL-DMOEA).
• Combines memory mechanisms to preserve past best individuals with manifold TL for predicting optimal individuals.
• Utilizes elites from past experience and future predictions to form the initial population.

Main Results:

• Significantly improves solution quality at the initial stage of optimization.
• Reduces computational cost compared to existing methods.
• Achieves a two-order-of-magnitude improvement in computational speed while enhancing solution quality.

Conclusions:

• The MMTL-DMOEA effectively addresses the challenges of DMOPs.
• The proposed method offers a superior balance of speed and solution quality compared to state-of-the-art algorithms.
• This approach provides a computationally efficient and accurate solution for dynamic multiobjective optimization.