基于Holt-Winters及长短期记忆的云资源组合预测模型

云资源的预测分析对于响应资源请求并及时作出决策非常重要，针对容器云资源的过度调配、供应不足的资源管理问题，云资源预测精度低，数据波动性问题，为使云资源的预测能够为工作负载的需求提前响应并作出合理分配，提出了一种基于Holt-Winters和长短期记忆神经网络( HW-LSTM )的云资源组合预测模型，并以预测残差的变异系数赋权。对亚马逊CPU数据集的预测实验表明，提出的组合模型比Holt-Winters、LSTM及CNN模型预测性能及稳定性更好，各项误差指标优化范围在0.065-1.026、0.023-0.269、0.001-0.007、0.004-0.039和0.079-4.125之间。

Cloud Resource Combination Prediction Model Based on HW-LSTM

Predictive analysis of cloud resources is very important for responding to resource requests and making timely decisions. For resource management issues such as over-allocation and insufficient supply of container cloud resources, low cloud resource prediction accuracy and data volatility issues, in order to make cloud resource predictions work load responding to the demand in advance and making a reasonable allocation, a cloud resource combination forecasting model is proposed that based on Holt-Winters and long short-term memory network ( HW-LSTM ) , and the coefficient of variation of the forecast residuals is used to give weights. Prediction experiments on Amazon''s CPU data set show that the combined model has better prediction performance and stability than Holt-Winters, LSTM and CNN models, and the optimization range of various error indicators is 0.065-1.026, 0.023-0.269, 0.001-0.007, 0.004-0.039 and 0.079-4.125.