绿色数据中心不完备能耗大数据填补及分类算法研究

发布时间：2019-05-11 19:24

【摘要】：随着云计算和大数据时代的到来,大规模数据中心在全球范围内得到了广泛的部署.但大规模数据中心的高能耗仍然是当今亟待解决的问题.为解决这一问题,通常采用太阳能等可再生绿色新能源为数据中心供电.绿色数据中心能够根据新能源的变化配合市电为数据中心提供高效、低能耗且稳定的电能供给,这是数据中心发展的趋势.文中针对数据收集不完善和断电等因素会造成一定程度的数据缺失情况,提出了一种基于完备相容类的不完备大数据填补算法,来填补数据中心的缺失数据;针对绿色数据中心能耗大数据的不稳定、间歇性和随时变化等特点,提出了一种基于离散弱相关的决策森林并行分类算法,通过对数据中心能耗大数据并行分类,来指导供电方式,以利于高效节能和延长电池寿命;此外进一步提出了一种增量更新决策森林的算法,来增量更新分类模型,该算法能够保障分类模型不断适应数据变化,防止分类准确率随时间而下降,从而避免电池频繁充放电,以保证稳定供电.整体来说,文中提出了一种数据中心能耗大数据管理模型,该模型针对大规模绿色数据中心的能源供给相关问题,运用不完备能耗大数据的填补、能耗大数据的并行分类、分类模型更新这三方面的技术方法,动态调控太阳能和市电供电端口,为数据中心提供高效、低能耗且稳定的电能供给.最后,采用绿色数据中心真实的能耗相关大数据集进行实验,实验结果说明文中提出的能耗大数据管理模型,能够帮助绿色数据中心有效管理太阳能和其他资源来配合市电提供稳定且充足的电能供应,从而为整个数据中心服务体系提供高效的能源服务.
[Abstract]:With the advent of cloud computing and the big data era, large-scale data centers have been widely deployed worldwide. However, the high energy consumption of large-scale data centers is still an urgent problem to be solved. To solve this problem, renewable green energy such as solar energy is usually used to power the data center. According to the change of new energy, green data center can provide efficient, low energy consumption and stable power supply for data center according to the change of new energy, which is the development trend of data center. In view of the lack of data caused by imperfect data collection and power outage, an incomplete big data filling algorithm based on complete compatibility class is proposed to fill the missing data in the data center. In view of the instability, intermittent and constant variation of energy consumption in green data center, a parallel classification algorithm of decision forest based on discrete weak correlation is proposed. Big data parallel classification of energy consumption in data center is carried out. To guide the power supply mode, in order to facilitate high efficiency and energy saving and prolong the battery life; In addition, an algorithm of incremental updating decision forest is proposed to update the classification model incrementally, which can ensure that the classification model can adapt to the change of data and prevent the classification accuracy from decreasing with time. So as to avoid frequent charge and discharge of batteries to ensure stable power supply. On the whole, this paper proposes a data center energy consumption big data management model, which aims at the energy supply related problems of large-scale green data center, which is filled by incomplete energy consumption big data and the parallel classification of energy consumption big data. The classification model updates these three technical methods, dynamically controls the solar energy and municipal power supply ports, and provides efficient, low energy consumption and stable power supply for the data center. Finally, the real energy consumption related big data set of green data center is used to carry on the experiment. The experimental results illustrate the big data management model of energy consumption put forward in the paper. It can help the green data center to manage solar energy and other resources effectively to provide stable and adequate power supply to the city power, thus providing efficient energy services for the whole data center service system.
【作者单位】：武汉理工大学计算机科学与技术学院;佛罗里达大学电气与计算机工程系;
【基金】：国家自然科学基金(61303029) 教育部留学回国基金([2012]1707) 美国国家自然科学基金资助
【分类号】：TP308

【参考文献】