Abstract: A genetic algorithm is introduced into the process of automated hardware/software partition in reconfigurable computing. As population may premature or converge too slowly at later process of standard genetic algorithm, niching technique is exploited to protect the diversity of genes. Adaptive genetic algorithm operators（crossover and mutation） are designed to vary according individual’s fitness. The algorithm is implemented 50 times randomly. As experiment demonstrating, the improved genetic algorithm has higher probability to find out the global best solution than standard genetic algorithm. It also finds out the best partition with smaller generations than standard genetic algorithm.

Key words: reconfigure computing, hardware/software partition, genetic algorithm, niching

摘要： 为实现可重构计算中的软硬件任务自动划分，引入了遗传算法来搜寻最优解。为解决标准遗传算法可能出现种群早熟和种群进化后期收敛速度慢的问题，使用了小生境技术来保护种群中基因的多样性。设计了能够随适应度自动改变的自适应遗传算子（杂交算子和变异算子）。对算法进行了50次随机实验，并对结果进行分析。实验表明，改进后的遗传算法搜寻到全局最优任务划分的概率和搜寻到最优任务划分时的进化代数都要优于标准遗传算法。

关键词: 可重构计算, 任务划分, 遗传算法, 小生境技术