Abstract

Large-scale global optimization (LSGO) problems are known as hard problems for many evolutionary algorithms (EAs). LSGO problems are usually computationally costly, thus an experimental analysis for choosing an appropriate algorithm and its parameter settings is difficult or sometimes impossible. In this study, we have investigated the performance of novel EA for LSGO based on Adaptive Differential Evolution with Success-History (SHADE) and cooperative coevolution (CC) with random adaptive grouping (RAG). SHADE is a self-adaptive DE algorithm. RAG approach is able to identify effective combinations of variables (subcomponents) in the problem decomposition stage. Thus, the proposed approach contains only two controlled parameters: the population size and the number of subcomponents. We have evaluated the performance of CC-SHADE-RAG with different settings on the IEEE CEC LSGO benchmark. The experimental results show that the approach outperforms some state-of-the-art LSGO techniques. CC-SHADE-RAG performs better with a small number of subcomponents and large size of population. We have performed statistical analysis for the experimental results and have established that the population size has greater effect on the algorithm performance than the number of subcomponents. This information can be used for further development of a completely self-adaptive LSGO technique by introducing an adaptive population sizing scheme in CC-SHADE-RAG.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call