High power consumption of cloud data centres is a crucial challenge in modern cloud computing. To comply with the conceptions of green computing, power consumption prediction of the computing cluster has a major role to play in these energy conservation efforts. However, due to complexity and heterogeneity in cloud computing scenarios, it is difficult to accurately predict the power consumption using conventional approaches. To this end, this study presents a power consumption model based on feature selection and deep learning to powerfully cope with low energy efficiency. Different from other methods focusing on only a few performance attributes, the proposed method takes into account up to 12 energy-related features and introduces deep neural network architecture, aiming at making full use of massive data to train model completely. In particular, this approach is composed of three main phases including (i) performance monitoring and energy-related feature acquisition, (ii) essential feature selection, and (iii) model establishment and optimisation. Representative results of comprehensive experiments, in terms of the relative error, reveal that the proposed power consumption model can undoubtedly achieve state-of-the-art predictive capability when compared with other models in most cases.

中文翻译：

---

云计算场景中基于特征选择和深度学习的功耗模型

云数据中心的高功耗是现代云计算中的关键挑战。为了符合绿色计算的概念，计算集群的功耗预测在这些节能工作中起着重要作用。但是，由于云计算场景中的复杂性和异构性，使用传统方法很难准确预测功耗。为此，本研究提出了一种基于特征选择和深度学习的功耗模型，以有效应对低能效。与仅关注少数性能属性的其他方法不同，该方法考虑了多达12个与能量相关的特征，并引入了深度神经网络体系结构，旨在充分利用海量数据来完全训练模型。尤其是，该方法包括三个主要阶段，包括（i）性能监控和与能源相关的特征获取，（ii）基本特征选择，以及（iii）模型建立和优化。综合实验的代表性结果，相对误差而言，表明在大多数情况下，与其他模型相比，所提出的功耗模型无疑可以实现最新的预测能力。