Cooperative coevolutionary algorithm with resource allocation strategies to minimize unnecessary computations

Abstract

In this paper, we propose a new computational resource allocation (CRA)-based cooperative coevolutionary (CC) algorithm, called ECCRA. To effectively allocate the computational resources into the sub-problems, ECCRA compensatively utilizes various strategies: (i) evaluate a degree of contribution for each sub-problem; (ii) extricate the stagnant sub-problems from any local minimums; (iii) allocate individuals adaptively according to a size of each sub-problem and its contribution; (iv) prune the unpromising sub-problems from the evolution process; and (v) utilize the multi-armed bandit (MAB)based selection method to choose various sub-problems extensively. In the experiments, ECCRA achieved best optimization results for the CEC'2010 benchmark problems. Moreover, ECCRA showed notable optimization performance for imbalanced problems in the CEC'2013 benchmark suite. Thus, we found that ECCRA could considerably outperform the existing CRA-based CC algorithms in terms of the optimization quality and convergence.