An Enhanced Differential Evolution Algorithm Based on Multiple Mutation Strategies.

Area of Science:

• Computational Intelligence
• Optimization Algorithms
• Metaheuristics

Background:

• Standard differential evolution (DE) algorithms, particularly DE/best/1/bin, suffer from premature convergence.
• This limitation hinders their effectiveness in global optimization over continuous spaces.

Purpose of the Study:

• To propose an enhanced differential evolution (EDE) algorithm that addresses the premature convergence issue.
• To leverage direction guidance from the best individual while avoiding local optima.

Main Methods:

• Introduced opposition-based learning initialization for superior initial solution quality.
• Developed a combined mutation strategy using DE/current/1/bin and DE/pbest/bin/1 for accelerated convergence and reduced clustering.
• Integrated a perturbation scheme to further combat premature convergence.
• Utilized two linear time-varying functions to dynamically select mutation and perturbation strategies.

Main Results:

• Experimental results on twenty-five benchmark functions demonstrate EDE's significant superiority over standard DE.
• Comparative analysis shows EDE is superior or equal to five other state-of-the-art approaches on most benchmark functions.

Conclusions:

• The proposed EDE algorithm effectively enhances global optimization performance.
• EDE successfully mitigates premature convergence and improves upon existing differential evolution methods.