Enhancing Gaussian Estimation of Distribution Algorithm by Exploiting Evolution Direction With Archive.

Area of Science:

• Artificial Intelligence
• Computational Optimization

Background:

• Estimation of Distribution Algorithms (EDA) are model-based evolutionary algorithms used for global optimization.
• Traditional Gaussian EDA (GEDA) often exhibits premature convergence, limiting search efficiency.

Purpose of the Study:

• To address the premature convergence issue in GEDA.
• To enhance the search efficiency of EDAs.

Main Methods:

• Developed a novel EDA variant, EDA², which incorporates an archive of high-quality historical solutions.
• Utilized this archive to assist in estimating the Gaussian model's covariance matrix, integrating evolution direction information.
• Reduced population size requirements for model estimation.

Main Results:

• EDA² effectively integrates evolution direction information into the model estimation process.
• The method allows for a reduced population size, leading to faster convergence.
• Experimental results on benchmark functions show EDA² is efficient and competitive.

Conclusions:

• EDA² offers an improved approach to global optimization by leveraging historical data.
• The proposed method enhances convergence speed and maintains competitiveness against state-of-the-art algorithms.