手机应用程序能耗的估计算法  

An Energy Consumption Estimation Algorithm for Applications of Mobile Phones

作　　者：唐家博 王宇然 程茹洁 陆建[1] 蒋厚明[2] 胡牧[2] 吴佳 

机构地区：[1]东南大学信息科学与工程学院,江苏南京210096 [2]国网电力科学研究院,江苏南京211000 [3]国网冀北电力有限公司,北京100053

出　　处：《计算机与现代化》2018年第1期1-7,共7页Computer and Modernization

基　　金：国家电网公司科技项目(SGTYHT/14-JS-188);国家自然科学基金资助项目(61401086)

摘　　要：针对智能手机的能耗进行建模,将手机能耗分为4个部分,分别是系统能耗、应用能耗、周期性跳变和白噪声,在此基础上,提出应用能耗的估计算法,分析应用能耗所包含的函数执行能耗,实现高能耗函数的代码定位。本文提出函数执行能耗的估计算法,主要包括系统能耗的估计、周期性跳变带来的噪声去除以及函数的平均功耗估计。在去除周期性跳变带来的噪声时,结合机器学习方法,提出基于DBSCAN的改进算法,通过扫参进行参数选取,实现数据集的聚类和去噪。通过对多台手机进行测试检验并与trace文件结合,实验结果表明本模型在应用能耗估计上的平均误差为5.58%,在函数执行功耗、代码段功耗计算上具有较好的精度和泛化能力。The paper is to establish a model of the energy consumption of mobile phones. The energy consumption of mobile phones is divided into four different parts,namely,system energy consumption,application energy consumption,periodic jump and white noise. Based on the model established,the energy consumption estimation algorithm for applications is proposed to analyze the energy consumption of the functions included in the application,and to locate the code segment which consumes energy abnormally. The model for the energy consumption mainly consists of the energy consumption evaluation of the operation system,the removal of noise arisen caused by the periodic jump,and the evaluation of functional average energy consumption. The denoising model for the noise caused by the periodic jump is based on DBSCAN algorithm,and the hyper parameters of the denoising model are selected by scanning hyper parameters combined with the theory of machine learning. Based on the results of the model above and verification of varied mobile phones combined with trace files,the average error is only 5. 58% when evaluating the energy consumption of mobile phones,which means the model has good generalization ability in the calculation of energy consumption of functions and code segments.

关 键 词：DBSCAN 机器学习 应用能耗 函数能耗 代码段能耗 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]