Reviewing and Benchmarking Parameter Control Methods in Differential Evolution.

Area of Science:

• Computational Intelligence
• Optimization Algorithms
• Evolutionary Computation

Background:

• Numerous parameter control methods (PCMs) exist for differential evolution (DE) algorithms, but they are often embedded within complex frameworks, obscuring their individual characteristics and performance.
• A lack of systematic analysis hinders understanding of the similarities and differences between various PCMs for DE.

Purpose of the Study:

• To conduct an in-depth review and large-scale benchmarking of 24 distinct PCMs for the scale factor and crossover rate in DE algorithms.
• To systematically extract and describe PCMs, facilitating a clearer understanding of their individual properties.
• To evaluate the performance of DE algorithms utilizing these PCMs under standardized conditions.

Main Methods:

• Extraction and systematic description of 24 PCMs from their original complex DE algorithms.
• Large-scale benchmarking of DE algorithms incorporating the 24 PCMs and 16 variation operators.
• Performance evaluation across 24 diverse black-box benchmark functions.

Main Results:

• Identification of PCMs that demonstrate high performance when integrated into a standardized DE framework, irrespective of their original algorithmic context.
• Comparative analysis of the 24 PCMs against an oracle-based model to assess the potential for future improvements.
• Detailed insights into the performance characteristics of individual PCMs under 16 different experimental conditions.

Conclusions:

• The study provides a clear understanding of the relative strengths and weaknesses of various PCMs for differential evolution.
• Benchmarking results offer guidance on selecting effective PCMs for specific optimization tasks.
• The research highlights areas for potential advancement in the design of parameter control strategies for DE algorithms.