基于小生境圆锥邻域粒子群的不完备决策表属性约简鲁棒算法

针对基于粒子群的属性约简算法易陷入局部最优、效率不高等问题,充分利用小生境技术在寻求最优解方面优势,提出一种基于小生境圆锥邻域粒子群的不完备决策表属性约简鲁棒算法(NCNPSO-IAR)。该算法通过圆锥分层空间构造小生境半径邻域子集向量,避免过多地依赖于先验领域知识生成小生境半径和早熟收敛,始终保持种群多样性,提高算法收敛速度。另外粒子种群在圆锥解空间充分进行约简集子矢量的协同学习,使属性约简集较好收敛到最优集。相关仿真实验表明:该属性约简优化算法是高效和鲁棒的,适用于不完备、含噪音决策表的属性约简。

Attribute Reduction Robust Algorithm of Incomplete Decision Table Based Niche Conic Neighborhood Particle Swarm Optimization

In order to overcome the premature convergence and poor running efficiency of the attribute reduction algorithm based on particle swarm optimization, based on some special searching optimization advantages of the niche technology, a novel incomplete attribute reduction robust algorithm (named NCNPSO-IAR) of niche conic neighborhood particle swarm optimization was proposed. It could construct niche subvector of neighborhood radius by the layered conic space. The main advantages of the proposed algorithm involves to partition the adaptive niche radius by avoiding depending on the prior domain knowledge and to overcome the premature convergence.It could maintain the diversity of populations, and improve the running converge speed. Further, reduction set subvectors could share some cooperative social cognition in their conic subspaces, so as to get the optimization attribute reduction sets. Experimental results demonstrated that the proposed algorithm is efficient and robust, especially for the incomplete and noisy attribute reduction.