As a form of evolutionary computation, particle swarm optimization is less effective in large scale optimization since it is unable to effectively balance exploration and exploitation. To address this problem, first, a learning structure decoupling exploration and exploitation is proposed. This helps simultaneously and independently managing exploration and exploitation in different components. Second, following the proposed learning structure, two novel learning strategies are developed. On the one hand, a local sparseness degree measurement in fitness landscape is proposed to estimate the congestion and distribution of particles, based on which an exploration strategy is built by guiding particles to sparse areas. On the other hand, an adaptive exploitation strategy is developed which can effectively adjust the fitness differences between exemplars and updated particles during the optimization process by employing a multi-swarm strategy and an adaptive sub-swarm size adjustment. Finally, by embedding the two learning strategies into the proposed learning structure, an adaptive particle swarm optimizer with decoupled exploration and exploitation is proposed. Thanks to the novel balancing strategy of exploration and exploitation, the two functions in the proposed algorithm can be independently and simultaneously managed. Furthermore, theoretical analyses are put forward to prove the convergence and computational complexity of the proposed algorithm. Comprehensive experiments are conducted based on the large scale optimization benchmarks from CEC 2010 and CEC 2013 and six state-of-the-art large scale optimization evolutionary algorithms, the results demonstrate the effectiveness of the proposed learning strategies and the competitive performance of the proposed algorithm.

作为进化计算的一种形式，粒子群优化在大规模优化中效果不佳，因为它无法有效地平衡勘探与开发。为了解决这个问题，首先，提出了一种将探索与开发分离的学习结构。这有助于同时独立地管理不同组件中的勘探和开发。其次，遵循提出的学习结构，开发了两种新颖的学习策略。一方面，提出了适合度景观中的局部稀疏度测量方法，以估计颗粒的拥塞和分布，在此基础上，建立了将颗粒引导到稀疏区域的探索策略。另一方面，提出了一种自适应利用策略，通过采用多群策略和自适应亚群大小调整算法，可以在优化过程中有效地调整样本和更新粒子之间的适应度差异。最后，通过将两种学习策略嵌入到所提出的学习结构中，提出了一种具有解耦探索与开发的自适应粒子群优化器。得益于新颖的勘探与开发平衡策略，该算法中的两个功能可以独立，同时管理。并通过理论分析证明了该算法的收敛性和计算复杂性。