Evolutionary Dynamic Multiobjective Optimization via Learning From Historical Search Process.

Area of Science:

• Computational Intelligence
• Optimization Algorithms
• Evolutionary Computation

Background:

• Dynamic multiobjective optimization problems (DMOPs) present challenges in maintaining fast convergence and solution diversity.
• Prediction-based evolutionary algorithms are popular but face limitations in adaptability and reliance on historical data quality.
• Existing methods struggle with the generalizability of predictors across diverse problems.

Purpose of the Study:

• To propose a novel knowledge learning strategy for effective change response in DMOPs.
• To address the limitations of prediction-based approaches in dynamic optimization.
• To enhance both convergence speed and diversity maintenance in evolutionary algorithms.

Main Methods:

• Developed a knowledge learning strategy that reacts to environmental changes by learning from the historical search process.
• Introduced a method for extracting valuable knowledge from previous search experiences.
• Compared the proposed strategy against state-of-the-art algorithms through comprehensive experiments.

Main Results:

• The proposed knowledge learning strategy demonstrated superior performance compared to existing algorithms.
• Significant improvements were observed in solution quality and computational efficiency.
• The extracted knowledge effectively accelerated convergence and introduced diversity for future environments.

Conclusions:

• The knowledge learning strategy offers a robust alternative to prediction-based methods for DMOPs.
• This approach enhances the adaptability and performance of evolutionary algorithms in dynamic environments.
• The findings suggest a promising direction for tackling complex optimization challenges.