基于深度学习的数据中心负载能耗预测模型研究  

Research on data center load energy consumption forecasting models based on deep learning

作　　者：刘芊 周泉 叶晓江[1] LIU Qian;ZHOU Quan;YE Xiao-jiang

机构地区：[1]武汉工程大学,湖北武汉430205

出　　处：《节能》2024年第11期98-100,共3页Energy Conservation

摘　　要：为了提高数据中心负载能耗预测的精度,提出一种基于深度学习的数据中心负载能耗预测模型。该模型基于历史负载数据,并结合数据中心的环境参数,利用长短期记忆(LSTM)神经网络进行预测。选取某数据中心的实际能耗数据集,按照8∶1∶1的比例随机划分训练集、验证集和测试集,从而能验证模型的预测效果。结果显示,与传统的时间序列预测方法相比,该模型的预测精度和稳定性明显提升。该模型的平均绝对误差(MAE)和均方根误差(RMSE)分别比自回归积分滑动平均(ARIMA)模型降低了36.2%和34.2%,平均绝对百分比误差(MAPE)也降低了24.4%。To enhance the accuracy of energy consumption forecasting for data centers,a deep learning-based model for predicting data center load energy consumption is proposed.This model is based on historical load data and integrates environmental parameters of the data center,utilizing long short-term memory(LSTM)neural networks for forecasting.A real energy consumption dataset from a certain data center was selected and randomly divided into training,validation,and testing sets in a ratio of 8:1:1,which allows for the verification of the model's forecasting effectiveness.The results show that compared to traditional time series forecasting methods,the model's prediction accuracy and stability have significantly improved.The model's mean absolute error(MAE)and root mean square error(RMSE)were reduced by 36.2%and 34.2%,respectively,compared to the autoregressive integrated moving average(ARIMA)model,and the mean absolute percentage error(MAPE)was also reduced by 24.4%.

关 键 词：数据中心 负载能耗预测 深度学习 LSTM神经网络 环境参数 

分 类 号：TK018[动力工程及工程热物理]