基于概率分布估计算法的进化规划码书设计算法

Codebook Design Optimization Based on EDA-EP

提出一种以最近邻划分变异为搜索策略,并以EP(进化规划)与EDA(概率密度估计算法)相结合的混合进化方法作为搜索引擎的新型码书设计算法.在最近邻划分上,引入最近邻划分控制因子作为进化算法的染色体表示,实现最近邻划分变异,从而改变质心运动轨迹.染色体与矢量同维,编码空间相对较小,并且进化操作易于控制和实现.在混合进化方法中,EDA为EP提供了最优个体的搜索方向,加速了算法的收敛速度.实验结果表明该方法是能有效提高码书性能的一种优化方法.

This paper proposed a novel search strategy of the nearest neighbor partition mutation, and a new codebook design algorithm via the hybrid engine combining the evolutionary programming （EP） with the estimation of distribution algorithm （EDA）. The nearest neighbor partition control vector is introduced as the chromosome for implementing the nearest neighbor partition mutation to control the centroid track. Compared with other methods using GAs, the search space is relative small since the size of chromosome equals to the dimension of training vector, and easy for manipulation. In the proposed hybrid algorithm, the search direction is estimated by EDA, thereby the convergence rate is accelerated. The experimental results show the new codebook design algorithm makes better performance on improving the codebook quality.