基于深层LSTM的分布式负载预测模型

Time Series-Oriented Load Prediction Using Deep LSTM

负载不平衡往往会导致分布式系统的性能损失,因此大量的动态负载平衡策略被引入来管理共享资源和分布式负载,它们基于负载和应用特性来实现所需平衡。其中负载预测是一种广泛应用于改善负载分布以避免负载不平衡的技术。由于传统预测模型精度低,所以有很大的局限性。为了提高负载预测精度,针对负载序列的变化特性,提出一种基于LSTM(长短期记忆网络)的负载预测算法。所提出的算法在真实数据集上进行验证,实验结果表明该算法能够精准地预测负载信息,其性能优于其他负载预测算法,如ARIMA、EWMA、RNN等。

Load imbalances are inclined to cause performance losses in distributed systems.Load forecasting is a technique that has been used widely in improving load redistribution to avoid load imbalances.In order to address the limitations of traditional forecasting models these are time consuming,complex and low precision,and in view of the fact that the load sequence has the characteristics of randomness,non-stationari ty,nonlinearity and its continuous oscillation,a model based on Long Short-Term Memory(LSTM)is proposed to enhance load prediction precision in this paper.The proposed approach is Deep Long Short-Term Memory(DLSTM),and it tested on a real dataset,the experiment results show that the DLSTM is capable of predicting load information efficiently and producing very close values to the actual load values,it outperforms the other load forecasting approaches.