摘要
软硬件划分是软硬件协同设计中的关键步骤,为NP难问题。分布估计算法可以解难优化问题,具有很好的全局搜索能力,但存在局部搜索能力差、种群多样性易失问题。针对此问题,对分布估计算法进行改进,对精英解进行克隆选择以加强局部搜索能力,对概率模型进行修正以改善种群多样性损失问题。同时,针对划分问题提出一种不可行解的修复方法。将改进后的分布估计算法应用于软硬件划分问题,并与现有算法做比较,结果表明所提算法在不同的约束条件下均可获得更好的优化结果。
Hardware/software （HW/SW） partitioning is a crucial step in embedded system co-design,also an NP hard problem. Estimation of distribution algorithm is good in globe search, hut poor in local search and is prone to premature convergence because of diversity loss. An improved estimation of distribution algorithm was proposed to solve HW/SW partitioning problem. The improved algorithm clones and searches the promising solutions to strengthen the local searching ability, and corrects the probability model to improve the diversity loss. A method of repairing infeasible solutions was also proposed. Simulation was carried out. And the comparisons with existing algorithm demonstrate the effectiveness of the improved estimation of distribution algorithm in solving HW/SW partitioning problem.
出处
《计算机科学》
CSCD
北大核心
2014年第9期285-289,共5页
Computer Science
基金
国家自然科学基金(61271143
60871080)资助
关键词
分布估计算法
软硬件划分
精英克隆
概率模型修正
不可行解修复
Estimation of distribution algorithm
Hardware/software partitioning
Dominance clone
Probability model correction
Infeasible solutions repair
