基于GRU神经网络的数据中心能耗预测模型研究

Research on Predicting Model of Energy Consumption in Data Center Based on GRU Neural Network

数据中心产生的巨大能耗给经济和环境带来了压力,对能耗进行分析和预测可为其能耗效率的优化提供重要依据。考虑到包括室外天气、内部数据中心CPU负载等影响数据中心能耗的复杂因素,文章提出了基于门控循环单元(Gated Recurrent Unit,GRU)神经网络的数据中心能耗预测模型。首先,通过对数据中心能耗数据进行分析,提取了与数据中心能耗最强相关的特征,并将这些时序特征数据作为输入进行模型训练。然后,提出一种结合神经网络(Artificial Neural Network,ANN)和GRU的网络模型(ANN-GRU)来预测数据中心能耗。最后,在仿真环境下基于真实轨迹数据进行了验证和分析。结果表明,与支持向量回归(Support Vector Regression,SVR)、LinearSVR、ANN模型相比,ANN-GRU具有更高的预测精度。

The huge energy consumption of data centers brings pressure to economy and environment.Analysis and prediction of energy consumption is a promising way to reduce data center energy consumption.Energy consumption is affected by many complicated factors,including outdoor temperature,CPU load and etc.In this paper,a data center energy prediction model based on ANN(Artificial Neural Network)and GRU(Gated Recurrent Unit)is proposed.First,we analyze characteristics of energy consumption and select features that are related to energy consumption as much as possible.These time series characteristic data are used as input for model training.Then,the combination of ANN and GRU networks are used to predict energy consumption.Finally,based on the real trajectory data,it is verified and analyzed in the simulation environment.The experimental simulation results show that compared with Linear SVR,SVR,and ANN,the proposed method has higher prediction accuracy.