Differential Evolution With Underestimation-Based Multimutation Strategy.

Area of Science:

• Optimization Algorithms
• Computational Intelligence
• Evolutionary Computation

Background:

• Differential Evolution (DE) performance is heavily influenced by its mutation strategy.
• Selecting an appropriate mutation strategy for specific problems or during different stages of the search process is challenging.
• Existing multimutation techniques may lose effective strategies or incur extra computational costs.

Purpose of the Study:

• To propose a novel underestimation-based multimutation strategy (UMS) for Differential Evolution (DE).
• To enhance the performance and adaptability of DE algorithms across various problems.
• To address the limitations of existing mutation strategy selection methods in DE.

Main Methods:

• A multimutation strategy (UMS) generates multiple candidate offspring for each target individual using diverse mutation approaches.
• A cost-effective abstract convex underestimation model estimates the quality of candidate offspring.
• The most promising candidate solution, based on the underestimation value, is selected as the final offspring.

Main Results:

• The proposed UMS integrates multiple mutation strategies without discarding any, ensuring equal generation probability.
• UMS avoids additional function evaluations by filtering candidate solutions using underestimation values.
• Empirical evaluations on CEC 2013/2014 benchmark sets and a real-world problem demonstrate performance improvements in DE variants.

Conclusions:

• The underestimation-based multimutation strategy (UMS) effectively enhances the performance of Differential Evolution algorithms.
• UMS offers a robust and efficient approach to mutation strategy selection in DE.
• The proposed method shows significant potential for improving optimization across diverse computational problems.