Adaptive Multi-optimiser Cooperative Co-evolution for Large-Scale Optimisation
Nasser R. Sabar, Ayad Mashaan Turky, Andy Song · 2019
Large-scale optimisation is of great importance as most real-world optimisation problems involve a large number of decision variables. Cooperative co-evolution (CC) approaches are successful in this field as they can tackle thousands of decision variables. CC methods consist of decomposition, optimisation of decomposed sub-components and cooperative combination based on the `divide-and-conquer' mechanism. Sub-components are optimised cooperatively before the merge. Most of existing CCs optimise different sub-components using the same optimisation algorithm. In this study we propose an adaptive approach based on multi-optimiser cooperative co-evolution for large-scale optimisation problems. As different sub-components may have different characteristics and different contributions towards the global fitness function, they should to be handled differently considering the diversity and the quality of each sub-component. To achieve this effect our proposed approach incorporates four different optimisers adaptively according to the relative contribution. In each optimisation cycle, an adaptive selection strategy is used to allocate the most suited optimiser to each sub-component using quality-diversity measure and the historical performance of the algorithms. The performance of the proposed approach has been evaluated on large-scale optimisation benchmark problems. The results confirmed the effectiveness of the proposed adaptive approach as it can achieve comparable or better performance on almost of test cases compared to state-of-the-art methods.