Adaptive Population Differential Evolution with Dual Control Strategy for Large-Scale Global Optimization Problems

In the greedy selection operator of differential evolution (DE), the trial solution will be selected into the new population only if it is better than the original target solution. Otherwise, the generated solution is simply eliminated. However, in most cases, these eliminated solutions may still be promising, and it will waste the computing resources to directly ignore them. Especially for the large-scale global optimization (LSGO) problems, it is important to make full use of all generated solutions and to enhance the population diversity in the limited fitness evaluation budget. To address this issue, an adaptive population DE, termed as APDE, is proposed with dual control strategy. Firstly, a population increasing (pop_inc) strategy is proposed for giving the opportunity to the generated trial solutions to survive in the population even though they are not good enough. Secondly, to avoid the gradual expansion of the population due to the pop_inc strategy, a population decreasing (pop_dec) strategy is proposed based on the "degradation value" designed for solutions. In the end of every iteration, if the degradation value of a solution is large, it represents the solution has a worse fitness value or has no improvement for a long time, and this solution will be deleted. In this way, the population size can be kept within a certain range. The test functions in CEC'2013 on LSGO are used to verify the performance of APDE. The experiment shows that APDE generally outperforms the original DE and two state-of-the-art LSGO algorithms.