An Adaptive Multipopulation Differential Evolution With Dynamic Population Reduction

Area of Science:

• Computational Intelligence
• Optimization Algorithms

Background:

• Evolutionary algorithms are crucial for solving real-world optimization problems.
• Premature convergence and stagnation are significant challenges in evolutionary computation.

Purpose of the Study:

• To introduce a novel differential evolution algorithm, sTDE-dR, designed to enhance search quality and overcome convergence issues.
• To improve the efficiency and robustness of optimization processes.

Main Methods:

• The sTDE-dR algorithm clusters populations into multiple tribes, employing diverse mutation and crossover strategies.
• A competitive success-based scheme dynamically manages tribe lifecycle and participation ratios.
• Adaptive schemes within each tribe control scaling factors and crossover rates.
• A dynamic population reduction method is incorporated.

Main Results:

• The proposed sTDE-dR algorithm demonstrated superior performance on the CEC2014 benchmark dataset.
• Comparative analysis showed the robustness of sTDE-dR against state-of-the-art algorithms.
• The competitive success-based scheme effectively guided the search towards optimal solutions.

Conclusions:

• The sTDE-dR algorithm offers an effective approach to address premature convergence and stagnation in optimization.
• The tribal clustering and adaptive strategy ensemble contribute to robust and efficient evolutionary computation.
• The method shows significant promise for complex real-world optimization tasks.