基于图的保序回归方法  

Graph of Isotonic Regression

作　　者：严昂 吉万鹏 Yan Ang;Ji Wanpeng

机构地区：[1]江苏科技大学深蓝学院 [2]南京邮电大学计算机学院

出　　处：《变频器世界》2022年第11期88-93,共6页The World of Inverters

摘　　要：随着当前科学技术的快速发展以及社会经济的高速增长,数据中心在全球蓬勃发展,针对其核心设施服务器的稳定工作已成为当前亟需解决的课题。服务器工作时会产生热量,为了使机房中的温度保持在行业安全温度以下,每当温度超过基线时就会预警观察研究表明,服务器温度变化趋势波动复杂,易产生温度异常误报情况,使运维人员无法快速的确定故障来源。为此,我们提出了一种基于保序回归图网络的温度预警方法并在多个机房进行了实验研究,采用XGBoost构建模型对数据进行拟合,其次使用保序回归对服务器温度进行预测,最后将各个服务器视为节点,建立网络拓扑图。应用实验表明该方法有效提高了温度预警的精确度,验证了该方法的有效性和泛化性。With the rapid development of science and technology and the fast growth of social economy,data centers are booming around the world.Meanwhile,the subject of stable work of servers,the core facilities of data centers,has become an urgent problem to be solved.Servers generate heat when they work.To keep the temperature in the computer room below industry-safe temperatures,an early warning will be issued whenever the temperature exceeds the baseline.Research shows that the trend fluctuation of server temperature change is complex,and it is easy to generate false alarms of abnormal temperature,which makes it hard for operation and maintenance personnel to quickly determine the source of the fault.To this end,we propose a temperature warning method based on graph network of Isotonic Regression and conduct experimental research in various computer rooms.We use XGBoost to build a model to fit the data,then use Isotonic Regression to predict the server temperature.Finally,each server is regarded as a node,and a network topology map is established.The experiments show that the method can effectively improve the accuracy of temperature warning,which verifies the effectiveness and generalization of the method.

关 键 词：服务器 保序回归 图网络 XGBoost 温度预警 

分 类 号：S624.4[农业科学—园艺学]