一种动态电源管理预测算法的设计及仿真

针对基于马尔可夫模型的预测式动态电源管理算法（DPMPA）对大型样本数据预测精度低的问题,提出了一种具备自反馈功能的内嵌式马尔可夫模型（RMM）的DPMPA。该算法基于分层迭代思想,对满足马尔可夫性质的大型数据进行聚类,再使用马尔可夫算法对构建出的迭代数据模型：上层抽象数据模型和底层实例数据模型进行训练。引入反馈函数φ（i）,控制转换概率矩阵更新频率,保证预测精度范围。依此,编制了自反馈内嵌式马尔可夫模型DPMPA的Matlab程序。应用该程序对无线热点访问次数进行仿真预测,得出不同训练样本数对后期样本的预测精度的影响,对比马尔可夫算法和自适应学习树（ALT）算法预测结果表明,基于该自反馈RMM预测式动态电源管理算法对于大型样本数据预测精度比前者高5%,后者高10%。预测精确度的提高,将更有利于马尔可夫算法的DPM系统功耗控制。

Design and simulation of prediction algorithm for dynamic power management

For the problem of low prediction accuracy that dynamic power management prediction algorithm(DPMPA) based on Markov model(MM) predicts large scale sample data,DPMPA of RMM with auto-feedback function is proposed.The algorithm can meet the cluster of large scale data with Markov property.The Markov algorithm was adopted to train the iteration data models(upper layer Abstract data model and bottom layer data model).Meanwhile,a feedback function was introduced to control the update frequency of transition probability matrix,and thus guarantee the prediction accuracy in a reasonable range.Afterwards,the Matlab program of DPMPA of AFRMM was compiled.This program was applied to conducting the simulation prediction for the access times of each wireless access point(WAP).The forecast accuracy based on learning different amount samples was obtained.Compared the prediction results of Markov algorithm and adaptive learning tree(ALT) algorithm,the prediction accuracy of AFRMM DPMPA is 5% higher than the former method and 10% higher than the latter method.The accuracy improvement is benefit for the power consumption control of DPM system by Markov algorithm.