基于CNN-BiLSTM-Attention的工业数据中心IT设备能耗预测模型研究
电子技术应用
宋越1,靳晟1,林栎2,高国强2,郭付展2
1.新疆农业大学 计算机与信息工程学院;2.新疆电子研究所股份有限公司
摘要: IT设备的能耗直接影响到工业数据中心的电力消耗,预测IT设备能耗对优化能源管理和资源规划具有重要意义。然而,由于IT能耗数据呈现出非线性、非平稳的特点,导致预测精度低。对此,结合卷积神经网络CNN、双向长短期记忆网络BiLSTM和注意力机制的优势,分别对IT设备能耗的局部特征、数据中深层次的关键信息进行提取,并根据自测IT设备能耗数据集构建基于CNN-BiLSTM-Attention的能耗预测模型,该模型的R2、MAE和RMSE分别为0.905 3、0.050 4、0.067 3,相较于现有的LSTM、BiLSTM和CNN-BiLSTM模型均有不同程度的提高,说明该模型可以应用于工业数据中心内IT设备能耗的准确预测。
中图分类号:TP391 文献标志码:A DOI: 10.16157/j.issn.0258-7998.246045
中文引用格式: 宋越,靳晟,林栎,等. 基于CNN-BiLSTM-Attention的工业数据中心IT设备能耗预测模型研究[J]. 电子技术应用,2025,51(10):63-68.
英文引用格式: Song Yue,Jin Sheng,Lin Li,et al. Research on energy consumption prediction model of industrial data center IT equipment based on CNN-BiLSTM-Attention[J]. Application of Electronic Technique,2025,51(10):63-68.
中文引用格式: 宋越,靳晟,林栎,等. 基于CNN-BiLSTM-Attention的工业数据中心IT设备能耗预测模型研究[J]. 电子技术应用,2025,51(10):63-68.
英文引用格式: Song Yue,Jin Sheng,Lin Li,et al. Research on energy consumption prediction model of industrial data center IT equipment based on CNN-BiLSTM-Attention[J]. Application of Electronic Technique,2025,51(10):63-68.
Research on energy consumption prediction model of industrial data center IT equipment based on CNN-BiLSTM-Attention
Song Yue1,Jin Sheng1,Lin Li2,Gao Guoqiang2,Guo Fuzhan2
1.College of Computer and Information Engineering,Xinjiang Agricultural University;2.Xinjiang Institute of Electronics Co.,Ltd.
Abstract: The energy consumption of IT equipment directly affects the power consumption of industrial data centers, and predicting the energy consumption of IT equipment is of great significance for optimizing energy management and resource planning. However, due to the non-linear and non-stationary nature of IT energy consumption data, the prediction accuracy is low. In this regard, by combining the advantages of convolutional neural network (CNN), bidirectional long short-term memory network (BiLSTM), and attention mechanism, local features of IT equipment energy consumption and deep key information in the data are extracted separately. Based on the self tested IT equipment energy consumption dataset, an energy consumption prediction model based on CNN-BiLSTM-Attention is constructed. The R2, MAE, and RMSE of this model are 0.905 3, 0.050 4, and 0.067 3, respectively. Compared with existing LSTM, BiLSTM and CNN-BiLSTM models, this model has improved to varying degrees, indicating that this model can be applied to accurate prediction of IT equipment energy consumption in industrial data centers.
Key words : IT energy consumption prediction model;CNN-BiLSTM-Attention;industrial data center;deep learning
引言
数据中心是承载云计算、大数据、移动互联网和智能终端不可或缺的处理数据的设施。随着越来越多的服务和数据“上云”,数据中心的规模在不断扩大、数量在不断增长,因而产生了巨大的能源消耗[1]。随着互联网数字化进程加速推进,预计2024年全国数据中心的耗电量将在3 400亿至3 600亿度之间,其产生的巨大能耗给经济和环境带来了压力,因此构建绿色高效的数据中心[2]迫在眉睫。数据中心的管理者需要通过能耗预测的结果,帮助数据中心更有效地管理能源资源,降低成本和提高能耗[3]。传统的数据中心能耗预测方法通常依赖经验法则和历史数据,这些方法的局限性在于它们难以捕捉到影响能耗的各种复杂因素,如环境参数变化[4]、电压电流以及负载情况变化。因此,这些方法难以在多重因素交互作用且不断变化的条件下对数据中心能耗进行高精度预测。
本文详细内容请下载:
//www.51qz.net/resource/share/2000006808
作者信息:
宋越1,靳晟1,林栎2,高国强2,郭付展2
(1.新疆农业大学 计算机与信息工程学院,新疆 乌鲁木齐 830052;
2.新疆电子研究所股份有限公司,新疆 乌鲁木齐 830052)

