Comparative Performance Analysis of Differential Evolution Variants on Engineering Design Problems.

Area of Science:

• Computational Intelligence
• Optimization Algorithms
• Engineering Applications

Background:

• Nature-inspired optimization algorithms, particularly meta-heuristics like Differential Evolution (DE), are vital for complex real-world problems.
• DE has seen numerous successful variations, many recognized in IEEE CEC competitions, yet a comparative performance analysis is lacking.

Purpose of the Study:

• To systematically compare the performance of the original Differential Evolution (DE) algorithm against eight of its IEEE CEC competition-winning variants.
• To identify the most effective DE variants for diverse optimization challenges, including bound-constrained, unconstrained, and mechanical engineering design problems.

Main Methods:

• Evaluated DE and its eight variants on IEEE CEC 2019 and 2020 bound-constrained functions.
• Tested algorithms on one unconstrained problem (IEEE CEC 2011) and five constrained mechanical engineering design problems (four from IEEE CEC 2020).
• Employed statistical analyses, including Friedman's and Wilcoxon's tests, for rigorous performance verification.

Main Results:

• No single Differential Evolution variant demonstrated superior performance across all tested problem types.
• SHADE and ELSHADE-SPACMA emerged as notable performers for the specific mechanical engineering design problems evaluated.

Conclusions:

• The efficacy of Differential Evolution variants is problem-dependent, highlighting the need for careful algorithm selection.
• Further research into specialized DE variants like SHADE and ELSHADE-SPACMA is warranted for complex engineering optimization tasks.