对于满足带宽约束的低能耗片上网络映射问题,提出一种基于灾变遗传退火的映射算法.该算法以标准遗传算法为基础,引入Boltzmann选择方法,对遗传操作后的较优个体采用多邻域的模拟退火操作进行优化,对处于停滞状态的种群使用灾变操作重新初始化部分较差个体,跳出局部极值.实验结果表明:与标准遗传算法相比,该算法具有优化性能好,收敛速度快的优点,映射结果比混沌遗传算法平均节能21.7％,有效地降低了片上网络系统通信能耗.%A mapping algorithm based on catastrophic genetic annealing is proposed, aiming at the mapping problem of low-energy consumption network-on-chip (NoC) that satisfies bandwidth constraints. Derived from a standard genetic algorithm, the proposed algorithm introduces the Boltzmann selection method and optimizes the optimum individuals upon genetic manipulation by the multi-neighborhood simulated annealing operation. Besides, some poor individuals belonging to the population in stagnation are re-initialized by catastrophic operations to jump out of local extreme values. Experimental results suggest that the presented algorithm has the advantages over the standard genetic algorithm of better optimizing performance and faster convergence. The algorithm can also save 21. 7% energy on average compared to the chaos genetic algorithm in mapping results, thus effectively reducing the energy consumption of NoC system communications.
展开▼
