摘要
在深入研究目前国际上比较流行的几种系统级电源管理(PM)算法的基础上,利用BP神经网络的非线性映射能力,提出基于BP神经网络的、对任务之间相互间隔时间也就是系统空闲时段的长度进行自适应学习的BPPM算法,具有传统回归PM算法不可比拟的优点。仿真实验表明引入神经网络的电源管理算法较之传统PM算法大大降低了系统级功耗。实现了在不需要建立系统模型、无需预先获得负载统计特性的前提下,通过从系统正常工作产生的数据中不断学习,使系统具有自适应、高效的电源管理能力,以达到降低系统功耗、提高器件可靠性、延长工作寿命的目的。
On the study of several popular system-level power management（PM） algorithm, a new PM policy called BPPM based on the BP neural network is proposed. Using the nonlinear mapping ability of back propagation neural networks to self-learning the interarrival time of the history idle periods of the system, and so BPPM has higher performance than traditional PM techniques. The results of the experiments prove that BPPM policy based on neural network greatly lower the system-level power consumption than traditional ones. Without premise constructing the system model and preacquiring statistical characters of workloads, just by online self-learning based on the data from general working stats, electric systems can achieve efficiently intelligent adaptive PM to reduce the power cousumption and to improve reliability and prolong the lifetime of chips.
出处
《计算机工程》
EI
CAS
CSCD
北大核心
2006年第4期214-216,共3页
Computer Engineering
基金
国家自然科学基金资助项目(90207008)
关键词
电源管理
系统级功牦
神经网络
Power management（PM）
System-level power consumption
Neural network
