期刊文献+

基于深度自回归循环神经网络的边缘负载预测

Load Prediction with Deep Autoregressive Recurrent Networks in Edge Computing
下载PDF
导出
摘要 为了更好地支持边缘计算服务提供商进行资源的提前配置与合理分配,负载预测被认为是边缘计算中的一项重要的技术支撑.传统的负载预测方法在面对具有明显趋势或规律性的负载时能取得良好的预测效果,但是它们无法有效地对边缘环境中高度变化的负载取得精确的预测.此外,这些方法通常将预测模型拟合到独立的时间序列上,进而进行单点负载实值预测.但是在实际边缘计算场景中,得到未来负载变化的概率分布情况会比直接预测未来负载的实值更具应用价值.为了解决上述问题,本文提出了一种基于深度自回归循环神经网络的边缘负载预测方法(Edge Load Prediction with Deep Auto-regressive Recurrent networks,ELP-DAR).所提出的ELP-DAR方法利用边缘负载时序数据训练深度自回归循环神经网络,将LSTM集成至S2S框架中,进而直接预测下一时间点负载概率分布的所有参数.因此,ELP-DAR方法能够高效地提取边缘负载的重要表征,学习复杂的边缘负载模式进而实现对高度变化的边缘负载精确的概率分布预测.基于真实的边缘负载数据集,通过大量仿真实验对所提出ELP-DAR方法的有效性进行了验证与分析.实验结果表明,相比于其他基准方法,所提出的ELP-DAR方法可以取得更高的预测精度,并且在不同预测长度下均展现出了优越的性能表现. To better support the service providers of edge computing in configuring in advance and allocating resources reasonably,load prediction is considered as an important technical support in edge computing.Traditional load prediction methods may achieve satisfactory prediction results when handling the loads with obvious trends or regularities,but they cannot effectively achieve accurate prediction for highly-variable loads in edge environments.Moreover,these methods usually fit prediction models to independent time series and make single-point real-value load prediction.However,for practical scenarios of edge computing,it would be more valuable for applications to obtain the probability distribution of future loads than directly predict their real values.To solve these problems,we propose an Edge Load Prediction with Deep Autoregressive Recurrent networks(ELP-DAR)method.The ELP-DAR method uses the time-series data of edge loads to train deep autoregressive recurrent networks,which integrates LSTM into the S2S framework to directly predicts all the parameters of the probability distribution at the next time point.Therefore,the ELP-DAR method can efficiently extract essential representations of edge loads and learns their complex patterns,which can achieve accurate prediction of the probability distribution for highly-variable edge loads.Using real-world edge load datasets,the extensive simulation experiments validate the effectiveness of the proposed ELP-DAR method.The results show that the ELP-DAR method achieves higher prediction accuracy than other benchmark methods and exhibits superior performance with various prediction lengths.
作者 陈礼贤 梁杰 黄一帆 陈哲毅 于正欣 陈星 CHEN Lixian;LIANG Jie;HUANG Yifan;CHEN Zheyi;YU Zhengxin;CHEN Xing(College of Computer and Data Science,Fuzhou University,Fuzhou 350116,China;Fujian Provincial Key Laboratory of Networking Computing and Intelligent Information Processing,Fuzhou 350116,China;School of Computing and Communications,Lancaster University,Lancaster LA14YW)
出处 《小型微型计算机系统》 CSCD 北大核心 2024年第2期359-366,共8页 Journal of Chinese Computer Systems
基金 中央引导地方科技发展资金项目(2022L3004)资助 国家自然科学基金项目(62072108)资助 福建省自然科学基金杰青项目(2020J06014)资助 福建省财政厅科研专项经费项目(83021094)资助。
关键词 边缘计算 负载预测 概率分布 深度自回归 循环神经网络 edge computing load prediction probability distribution deep autoregression recurrent neural networks
  • 相关文献

相关作者

内容加载中请稍等...

相关机构

内容加载中请稍等...

相关主题

内容加载中请稍等...

浏览历史

内容加载中请稍等...
;
使用帮助 返回顶部