摘要
针对云计算中心机柜能耗数据序列化且变化频率高的特点,融合信号处理与深度学习方法,研究基于序列能耗特征提取的能耗预测模型,即基于变分模态分解(VMD)与粒子群优化(PSO)的门控循环单元(GRU)的能耗预测模型.首先,采用VMD算法对机柜设备能耗序列进行特征提取,获得不同中心频率的能耗数据本征模态函数(IMF);其次,基于IMF分量学习训练,构建各分量PSO-GRU预测模型;最后,使用最优模型实现能耗数据IMF函数分量预测及其重构,以序列叠加运算获得机柜的能耗预测结果.经云计算中心机柜设备能耗数据实验表明,基于VMD算法特征提取的PSO-GRU能耗预测模型具有精度较高、通用性强的特点,可为数据机柜及云计算中心的能效优化控制提供支撑.
In view of the high change-frequency of energy consumption data in Cloud Computation Center,this study integrated signal-processing and deep-learning methods,and researched on the feature-extraction and combined-prediction model of energy consumption.A combined-prediction model VMD-PSO-GRU of energy consumption was designed,which was based on variational mode decomposition,particle swarm optimization and gated recurrent unit.Firstly,VMD algorithm was used to extract features of energy consumption sequences of cabinet equipments,which obtained the IMF with different center-frequencies.Secondly,the PSO-GRU prediction model of each component was constructed,basing on IMF component learning-and-training.Finally,the IMF function component prediction and reconstruction of energy consumption data were realized by using the optimal model,and the energy consumption prediction results of the cabinet were obtained by sequence superposition operation.The energy consumption data experiment of cabinet equipment in Cloud Computing Center showed that the PSO-GRU energy consumption prediction model based on VMD algorithm feature extraction had the characteristics of high precision and strong universality,which provided support for the energy efficiency optimization control of data cabinet and Cloud Computing Center.
作者
许晓萍
叶泓伟
XU Xiaoping;YE Hongwei(School of Robotics,Fuzhou Polytechnic,Fuzhou Fujian 350108,China;School of Electrical Engineering and Automation,Fuzhou University,Fuzhou Fujian 350108,China)
出处
《泉州师范学院学报》
2022年第5期70-75,共6页
Journal of Quanzhou Normal University
基金
福州职业技术学院科研项目(FZYKJJJYB202201)
福建省高校产学合作科技项目(2019H6009)
作者简介
许晓萍(1985-),女,福建福州人,讲师,硕士,主要从事智能化电气技术研究.
