改进BP神经网络在软件能耗分析中的应用  

Application of Improved BP Neural Network in Software Energy Consumption Analysis

作　　者：邬小龙[1] 郭兵[1] 沈艳[2] 

机构地区：[1]四川大学计算机学院,成都610064 [2]成都信息工程学院控制工程学院,成都610225

出　　处：《计算机工程》2012年第10期185-187,共3页Computer Engineering

基　　金：国家自然科学基金资助项目(61073045);四川省杰出青年科技基金资助项目(2010JQ0011)

摘　　要：提出以软件特征量为基础的嵌入式软件体系结构级能耗建模方法。利用软件特征量与嵌入式软件能耗之间存在非线性函数关系的特点,采用基于改进遗传算法的BP神经网络算法进行训练拟合。从初始群体、编码、模拟退火算子3个方面对遗传算法进行改进以增强拟合的效果。实验结果证明,基于该算法的能耗模型预测值与实际能耗值的误差较小。This paper proposes a modeling of energy consumption in embedded software architecture level based on the software characteristic.It contains the idea that there is non-linear relationship between software characteristic and software energy consumption.And it presents a new BP neural network algorithm based on improved genetic algorithm to train and simulate the function.This algorithm takes three aspects into consideration to improve the genetic algorithm in order to enhance the simulation effect.Experimental results show that the deviation between the predictive value of energy consumption model based on this algorithm and the actual energy consumption value is smaller.

关 键 词：能耗建模 软件特征量 软件体系结构 遗传算法 BP神经网络 模拟退火算法 

分 类 号：TP39[自动化与计算机技术—计算机应用技术]