基于线性SVM算法的云数据中心蓄电池状态预测  

Prediction of battery status in cloud data centers based on linear SVM algorithms

作　　者：杨玉丽 李培仁 李学智 马彦楷 李震[2] YANG Yuli;LI Peiren;LI Xuezhi;MA Yankai;LI Zhen(Kunlun Digital Intelligence Technologies Co.Ltd.,Beijing 102206,China;School of Aerospace Engineering,Tsinghua University,Beijing 100084,China)

机构地区：[1]昆仑数智科技有限责任公司,北京102206 [2]清华大学航天航空学院,北京100084

出　　处：《西安工程大学学报》2023年第5期77-82,115,共7页Journal of Xi’an Polytechnic University

基　　金：中国石油数据中心(昌平)信息技术服务运维项目(2022YW007)。

摘　　要：针对运行在云数据中心的蓄电池老化威胁供电可靠性的问题,提出基于机器学习算法的蓄电池状态在线预测方法。采用经典机器学习算法模型预测蓄电池状态,对未提炼特征值的模型与提炼特征值的模型预测准确度进行对比实验。结果表明:未提取特征值的模型预测准确度在71.72%~81.82%,提取特征值后预测准确度提高了10.10%~20.20%,提取特征值能够提高模型预测准确度。基于线性SVM方法提取特征值预测模型优于其他算法,准确度达到96.46%,可以用于在线预测云数据中心蓄电池状态。Valve regulated lead-acid(VRLA)battery aging has threatened power supply reliability in cloud data centers,aimed at this problem,a method of an online battery state prediction method based on machine learning algorithms was proposed.The classic machine learning algorithm model was used to predict the battery status,and the prediction accuracy unextracted and the extracted feature values model was compared.The prediction accuracy of the model without extracted feature values ranges from 71.72%to 81.82%,and after extracting feature values,the prediction accuracy has improved by 10.10%to 20.20%,extracting feature values can improve prediction accuracy.The prediction model based on linear SVM method for extracting feature values is superior to other algorithms,with an accuracy of 96.46%.Experiment results show that machine learning algorithm based prediction models can be used for online VRLA battery state prediction in cloud data centers.

关 键 词：云数据中心 铅酸阀控蓄电池 高斯滤波器 线性判别 机器学习算法 

分 类 号：TM912[电气工程—电力电子与电力传动]