1.内蒙古工业大学 新能源学院, 内蒙古 鄂尔多斯 017010
2.内蒙古工业大学,风能太阳能利用技术教育部 重点实验室,呼和浩特 010051
3.内蒙古工业大学,大规模储能技术教育部工程研究中心,呼和浩特 010051
4.内蒙古工业大学,内蒙古自治区新能源与储能技术重点实验室,呼和浩特 010051
5.广西大学 电气工程学院, 南宁 530004
6.北京邮电大学 集成电路学院,北京 100876
朱虹谕(1996),女,副研究员,博士,从事源网荷储分布式灵活性资源协同响应与优化运行等方面的研究,hongyuzhu@imut.edu.cn;
宋凯宇(2003),男,硕士生,从事源网荷储分布式灵活性资源协同优化、算力与电力资源耦合调度技术等方面的研究,20251800864@imut.edu.cn;
武晓栋*(1996),男,讲师,博士,从事算电协同、算力-能源网络安全等方面的研究,xdwu@imut.edu.cn;
黄亚松(2000),男,硕士生,从事数据中心负荷预测方面的研究,2557074236@qq.com;
刘一宁(2001),男,硕士生,从事数据中心、需求响应等方面的研究,yiningliu@st.gxu.edu.cn;
张冬冬(1990),男,教授,博士,从事多元柔性负荷参与新型电力系统调控、算电协同等方面的研究,dongdongzhang@imut.edu.cn;
张杰(* 1972),男,教授,博士生导师,博士,从事云数据中心互联与算力网络等方面的研究,lgr24@bupt.edu.cn。
收稿:2026-07-17,
修回:2026-07-30,
网络首发:2026-08-21,
纸质出版:2026-08-25
移动端阅览
朱虹谕,宋凯宇,武晓栋等.基于优化双层分解与差异化建模的数据中心CPU利用率预测[J].综合智慧能源,2026,48(08):29-43.
ZHU Hongyu,SONG Kaiyu,WU Xiaodong,et al.CPU utilization prediction for data centers based on optimized two-stage decomposition and differentiated modeling[J].BLASTING,2026,48(08):29-43.
朱虹谕,宋凯宇,武晓栋等.基于优化双层分解与差异化建模的数据中心CPU利用率预测[J].综合智慧能源,2026,48(08):29-43. DOI: 10.3969/j.issn.2097-0706.2026.08.003.
ZHU Hongyu,SONG Kaiyu,WU Xiaodong,et al.CPU utilization prediction for data centers based on optimized two-stage decomposition and differentiated modeling[J].BLASTING,2026,48(08):29-43. DOI: 10.3969/j.issn.2097-0706.2026.08.003.
针对数据中心中央处理器(CPU)利用率序列具有非线性、非平稳、多时间尺度耦合及高频扰动显著等特点,提出了一种基于优化双层分解、特征筛选与频段差异化建模的组合预测方法。首先,以谷歌数据中心2021年CPU利用率时序数据为研究对象,配置与其时间戳一致的28项气象环境特征,采用24 h滑动时间窗口构造监督学习样本,并按照时间先后顺序将3月、6月、10月和12月数据划分为训练集、验证集和测试集。其次,采用澳洲野狗优化算法对改进的自适应噪声完全集合经验模态分解参数进行寻优,实现原始序列的一次分解;根据各本征模态函数分量零交叉率的分布特征,将其重构为高频、中频和低频分量;进一步采用足球队训练算法优化变分模态分解参数,对高频分量进行二次分解,以提取细粒度波动特征。然后,采用轻量梯度提升机特征重要性排序与验证集决定系数曲线相结合的两阶段方法,为不同子序列分别确定最优输入特征集合。最后,对变化平稳、趋势性较强的中低频分量采用自适应核极限学习机进行预测,对波动剧烈、时序关系复杂的高频分量采用双向时间卷积网络-双向长短期记忆网络-多头自注意力组合模型进行预测,并将各子序列预测结果叠加重构得到最终的CPU利用率预测值。算例结果表明,两阶段特征筛选后,6月样本的平均绝对误差和均方根误差分别由0.026 0和0.032 5降至0.019 9和0.025 8,决定系数由0.824 3提高至0.887 5;所提模型在不同月份均表现出较高的预测精度和稳定性,其中10月和12月的决定系数分别达到0.963 1和0.903 3。研究结果表明,优化双层分解能够有效降低原始序列复杂度,差异化建模能够充分发挥不同预测模型在各频段上的优势,可为数据中心资源配置、运行状态评估及能耗分析提供参考。
The central processing unit (CPU) utilization series in data centers has the characteristics of nonlinearity, non-stationarity, multi-timescale coupling, and significant high-frequency disturbance. In view of these characteristics, a combined prediction method based on optimized two-stage decomposition, feature selection, and differentiated modeling across frequency bands was proposed. The CPU utilization time-series data of the Google data center from 2021 were selected as the research object. A total of 28 meteorological and environmental features consistent with the timestamps were configured. Supervised learning samples were constructed using a 24 h sliding time window, and data from March, June, October, and December were divided chronologically into training, validation, and test sets. The dingo optimization algorithm (DOA) was used to optimize the parameters of improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN), thereby achieving the primary decomposition of the original series. Based on the distribution characteristics of the zero-crossing rates of the intrinsic mode function (IMF) components, these components were reconstructed into high-, medium-, and low-frequency components. Furthermore, the football team training algorithm (FTTA) was used to optimize the parameters of variational mode decomposition (VMD), and secondary decomposition was performed on the high-frequency components to extract fine-grained fluctuation features. Subsequently, a two-stage approach combining light gradient boosting machine (LightGBM) feature importance ranking with validation-set coefficient of determination curves was used to determine the optimal input feature set for each subsequence. An adaptive kernel extreme learning machine was used to predict the medium- and low-frequency components with relatively stable variations and strong trends, whereas a combined bidirectional temporal convolutional network-bidirectional long short-term memory network-multi-head self-attention model was used to predict the high-frequency components with intense fluctuations and complex temporal relationships. The prediction results of all subsequences were superimposed and reconstructed to obtain the final CPU utilization predicted values. Case study results showed that after two-stage feature selection, the mean absolute error and root mean square error for the June samples decreased from 0.026 0 and 0.032 5 to 0.019 9 and 0.025 8, respectively, while the coefficient of determination increased from 0.824 3 to 0.887 5. The proposed model exhibited high prediction accuracy and stability across different months, with the coefficients of determination reaching 0.963 1 and 0.903 3 in October and December, respectively. The results indicate that optimized two-stage decomposition can effectively reduce the complexity of the original series, while differentiated modeling can fully leverage the advantages of different prediction models across frequency bands, providing a reference for data center resource allocation, operational status assessment, and energy consumption analysis.
MASANET E , LEI N , KOOMEY J . To better understand AI's growing energy use, analysts need a data revolution [J]. Joule , 2024 , 8 ( 9 ): 2427 - 2436 .
朱继忠 , 董瀚江 , 李盛林 , 等 . 数据驱动的综合能源系统负荷预测综述 [J]. 中国电机工程学报 , 2021 , 41 ( 23 ): 7905 - 7923 .
ZHU Jizhong , DONG Hanjiang , LI Shenglin , et al . Review of data-driven load forecasting for integrated energy systems [J]. Proceedings of the CSEE , 2021 , 41 ( 23 ): 7905 - 7923 .
WANG Y Z , HAN T , WEI Y M . Building accurate energy-use statistics for data centers [J]. Engineering , 2026 , 60 ( 5 ): 336 - 342 .
苗磊 , 李擎 , 蒋原 , 等 . 深度学习在电力系统预测中的应用 [J]. 工程科学学报 , 2023 , 45 ( 04 ): 663 - 672 .
MIAO Lei , LI Qing , JIANG Yuan , et al . A survey of power system prediction based on deep learning [J]. Chinese Journal of Engineering , 2023 , 45 ( 04 ): 663 - 672 .
SAXENA D , KUMAR J , SINGH A K , et al . Performance analysis of machine learning centered workload prediction models for cloud [J]. IEEE Transactions on Parallel and Distributed Systems , 2023 , 34 ( 4 ): 1313 - 1330 .
唐新姿 , 顾能伟 , 黄轩晴 , 等 . 风电功率短期预测技术研究进展 [J]. 机械工程学报 , 2022 , 58 ( 12 ): 213 - 236 .
TANG Xinzhi , GU Nengwei , HUANG Xuanqing , et al . Progress on short term wind power forecasting technology [J]. Journal of Mechanical Engineering , 2022 , 58 ( 12 ): 213 - 236 .
BI J , LI S , YUAN H T , et al . Integrated deep learning method for workload and resource prediction cloud systems [J]. Neurocomputing , 2021 , 424 : 35 - 48 .
程乐峰 , 刘奕杭 , 邹涛 . 博弈论视角下智能电网需求响应研究综述 [J]. 综合智慧能源 , 2025 , 47 ( 4 ): 1 - 22 .
CHENG Lefeng , LIU Yihang , ZOU Tao . Review of demand response in smart grids from the perspective of game theory [J]. Integrated Intelligent Energy , 2025 , 47 ( 4 ): 1 - 22 .
张冬冬 , 单琳珂 , 刘天皓 . 人工智能技术在风力与光伏发电数据挖掘及功率预测中的应用综述 [J]. 综合智慧能源 , 2025 , 47 ( 3 ): 32 - 46 .
ZHANG Dongdong , SHAN Linke , LIU Tianhao . Review on the application of artificial intelligence in data mining and wind and photovoltaic power forecasting [J]. Integrated Intelligent Energy , 2025 , 47 ( 3 ): 32 - 46 .
陈琨 , 丁苗 , 刘炬 , 等 . 基于改进灰狼算法优化WLSSVM的短期风功率预测 [J]. 内蒙古电力技术 , 2024 , 42 ( 2 ): 1 - 7 .
CHEN Kun , DING Miao , LIU Ju , et al . Short term wind power prediction based on WLSSVM optimized by improved grey wolf optimization algorithm [J]. Inner Mongolia Electric Power , 2024 , 42 ( 2 ): 1 - 7 .
于海存 , 殷建华 , 李荣丽 , 等 . 基于电网考核指标的火电机组一次调频仿真与优化 [J]. 内蒙古电力技术 , 2022 , 40 ( 5 ): 80 - 85 .
YU Haicun , YIN Jianhua , LI Rongli , et al . Simulation and optimization of primary frequency modulation of thermal power unit based on grid assessment index [J]. Inner Mongolia Electric Power , 2022 , 40 ( 5 ): 80 - 85 .
刘学芳 , 于鲜莉 , 陈波 , 等 . 基于特征评估与改进SOA-DELM的变压器状态预测方法 [J]. 内蒙古电力技术 , 2025 , 43 ( 1 ): 80 - 89 .
LIU Xuefang , YU Xianli , CHEN Bo , et al . Transformer state prediction method based on feature evaluation and improved SOA-DELM [J]. Inner Mongolia Electric Power , 2025 , 43 ( 1 ): 80 - 89 .
ZHANG H F , LI J , YANG H R . Cloud computing load prediction method based on CNN-BiLSTM model under low-carbon background [J]. Scientific Reports , 2024 , 14 ( 1 ): 18004 .
郭威 , 孙胜博 , 陶鹏 , 等 . 基于多元变分模态分解和混合深度神经网络的短期光伏功率预测 [J]. 太阳能学报 , 2024 , 45 ( 4 ): 489 - 499 .
GUO Wei , SUN Shengbo , TAO Peng , et al . Short-term photovoltaic power forecasting based on multivariate variational mode decomposition and hybrid deep neural network [J]. Acta Energiae Solaris Sinica , 2024 , 45 ( 4 ): 489 - 499 .
BENIDIS K , RANGAPURAM S S , FLUNKERT V , et al . Deep learning for time series forecasting: Tutorial and literature survey [J]. ACM Computing Surveys , 2023 , 55 ( 6 ): 1 - 36 .
邵必林 , 纪丹阳 . 基于VMD-SE的电力负荷分量的多特征短期预测 [J]. 中国电力 , 2024 , 57 ( 4 ): 162 - 170 .
SHAO Biling , JI Danyang . Multi-feature short-term prediction of power load components based on VMD-SE [J]. China Electric Power , 2024 , 57 ( 4 ): 162 - 170 .
LIM B , ZOHREN S . Time-series forecasting with deep learning: A survey [J]. Philosophical Transactions of The Royal Society A: Mathmatical , Physical and Engineering Sciences , 2021 , 379 ( 2194 ): 0209 .
钟燕 , 王军 , 宋戈 , 等 . 基于二次重构分解去噪及双向长短时记忆网络的极端天气下超短期电力负荷预测 [J]. 电网技术 , 2025 , 49 ( 11 ): 4791 - 4800 .
ZHONG Yan , WANG Jun , SONG Ge , et al . Ultra-short-term power load prediction under extreme weather based on secondary reconstruction denoising and BiLSTM [J]. Power System Technology , 2025 , 49 ( 11 ): 4791 - 4800 .
MALEKI N , LUNDSTROM O , MUSADDIQ A , et al . Future energy insights: Time-series and deep learning models for city load forecasting [J]. Applied Energy , 2024 , 374 : 124067 .
胡锐 , 乔加飞 , 李永华 , 等 . 基于WOA-VMD-SSA-LSTM的中长期风电预测 [J]. 太阳能学报 , 2024 , 45 ( 9 ): 549 - 556 .
HU Rui , QIAO Jiafei , LI Yonghua , et al . Medium and long term wind power forecast based on WOA-VMD-SSA-LSTM [J]. Acta Energiae Solaris Sinica , 2024 , 45 ( 9 ): 549 - 556 .
RAN P , DONG K , LIU X , et al . Short-term load forecasting based on CEEMDAN and Transformer [J]. Electric Power Systems Research , 2023 , 214 : 108885 .
宋江涛 , 崔双喜 , 樊小朝 , 等 . 基于SGMD-SE与优化TCN-BiLSTM/BiGRU的超短期风功率预测 [J]. 太阳能学报 , 2024 , 45 ( 10 ): 588 - 596 .
SONG Jiangtao , CUI Shuangxi , FAN Xiaochao , et al . Ultra-short-term wind power prediction based on SGMD-SE and optimized TCN-BiLSTM/BiGRU [J]. Acta Energiae Solaris Sinica , 2024 , 45 ( 10 ): 588 - 596 .
SONG K Y , YU Y , ZHANG T F , et al . Short-term load forecasting based on CEEMDAN and dendritic deep learning [J]. Knowledge-Based Systems , 2024 , 294 : 111729 .
韩博 , 李长青 , 刘卫亮 , 等 . 基于二次分解和BiGRU的超短期光伏发电功率预测 [J]. 动力工程学报 , 2025 , 45 ( 1 ): 62 - 69, 79 .
HAN Bo , LI Changqing , LIU Weiliang , et al . Ultra-short-term photovoltaic power forecasting based on secondary decomposition and BiGRU [J]. Journal of Chinese Society of Power Engineering , 2025 , 45 ( 1 ): 62 - 69, 79 .
LI J , DENG D Y , ZHAO J B , et al . A novel hybrid short-term load forecasting method of smart grid using MLR and LSTM neural network [J]. IEEE Transactions on Industrial Informatics , 2021 , 17 ( 4 ): 2443 - 2452 .
DING S F , ZHAO H L , ZHANG Y , et al . Extreme learning machine: Algorithm, theory and applications [J]. Artificial Intelligence Review , 2015 , 44 : 103 - 115 .
ZHENG Y F , CHEN B D , WANG S Y , et al . Mixture correntropy-based kernel extreme learning machines [J]. IEEE Transactions on Neural Networks and Learning Systems , 2022 , 33 ( 2 ): 811 - 825 .
FAN C D , LI G R , XIAO L Y , et al . Short-term power load forecasting in city based on ISSA-BiTCN-LSTM [J]. Cognitive Computation , 2025 , 17 ( 1 ): 1 - 24 .
HUANG Z H , XU W , YU K . Bidirectional LSTM-CRF models for sequence tagging [J]. arXiv preprint arXiv: 1508.01991 , 2015 .
QIN Y , YANG J H , ZHOU J H , et al . A new supervised multi-head self-attention autoencoder for health indicator construction and similarity-based machinery RUL prediction [J]. Advanced Engineering Informatics , 2023 , 56 : 101973 .
0
浏览量
1
下载量
0
CSCD
关联资源
相关文章
相关作者
相关机构
京公网安备11010802024621