Abstract: Multi-objective genetic algorithms are effective in solving SoC system level synthesis. These available algorithms can only attain part of the whole Pareto front, because of the worse local searching ability, the convergence speed is slow. In order to overcome these disadvantages, an updated multi-objective genetic algorithm, self-adaptive niching genetic algorithm, is proposed, which integrates the niching technique, self-adaptively adjusts the sub-populations’ size and the probabilities of crossover and mutation by using the previous population’s diversity information. It can effectively increase the solutions coverage and improve the convergence speed. Using the updated algorithm, this paper succeeds in optimizing the synthesis of video codec. Result indicates that the new algorithm can rapidly acquire more Pareto optimal solutions and proves the superiority of the algorithm.

Key words: System Level Synthesis(SLS), self-adaptive, niching, multi-objective optimization

摘要： 多目标遗传算法是解决SoC系统级综合问题的有效途径之一，但现有的遗传算法只能求得非劣解集前沿的一部分，局部搜索能力差，收敛速度较慢。该文通过结合小生境技术，根据种群往代的多样性信息，自适应地确定子种群的规模和交叉、变异的概率，提出一种自适应小生境遗传算法，有效提高解集的覆盖率，加快收敛速度。以视频编解码的系统级综合为例，证明该算法可以较快地产生较多非 劣解。

关键词: 系统级综合, 自适应, 小生境技术, 多目标优化

CLC Number: 

• TP18