An intelligent multi-restart memetic algorithm for box constrained global optimisation

Area of Science:

• Optimization algorithms
• Computational mathematics
• Artificial intelligence

Background:

• Global continuous optimization problems are challenging.
• Existing evolutionary algorithms (EAs) can get trapped in local optima.
• Efficiently exploring search spaces and refining solutions is crucial.

Purpose of the Study:

• To propose a multi-restart memetic algorithm framework for box-constrained global continuous optimization.
• To develop a specific algorithm using an Estimation of Distribution Algorithm (EDA) and the NEWUOA local optimizer.
• To evaluate the proposed algorithm's performance against established EAs.

Main Methods:

• A hybrid framework combining an EA for exploration and a local optimizer for exploitation.
• An adaptive multivariate probability model and multiple sampling strategy to enhance exploration.
• A restart mechanism to escape local optima, utilizing previous search history.

Main Results:

• The proposed EDA-based memetic algorithm demonstrates comparable performance to top EAs, including the CEC2005 winner.
• The algorithm significantly outperforms other EAs in solution quality.
• The computational cost is also competitive, indicating efficiency.

Conclusions:

• The multi-restart memetic algorithm framework is effective for global continuous optimization.
• The specific EDA and NEWUOA implementation offers a robust and efficient optimization tool.
• This approach provides a valuable alternative for complex optimization tasks.