云计算环境下嵌入式CPU负载预测仿真

Embedded CPU Load Prediction Simulation in Cloud Computing Environment

由于嵌入式中央处理器(Central Processing Unit, CPU)负载需要同时考虑CPU利用率、内存利用率等相关因素,导致对其预测时难度较大且无法保证精准度。因此,提出一种新的自适应预测算法。构建嵌入式CPU负载预测框架,对其负载数据预处理,降低非平稳数据对预测结果精度的影响;在整合移动平均自回归模型中加入周期变动因素,构建季节性差分自回归滑动平均模型,分析CPU负载数据时间序列变化特征;并对其迭代计算,得到季节性差分自回归滑动平均模型的参数和CPU负载预测结果。实验结果表明,所提方法的MAPE值低于25%,表明该方法的预测精度高。

At present,it is necessary to simultaneously consider CPU utilization and memory utilization when predicting the load of the embedded Central Processing Unit(CPU),but it is difficult to guarantee accuracy.There⁃fore,a new adaptive prediction algorithm was proposed.Firstly,we built a framework for predicting the embedded CPU load and preprocessed the load data,thus reducing the impact of non-stationary data on the accuracy of predic⁃tion results.Secondly,we added some periodic variation factors to the integrated moving average autoregressive model and then constructed a seasonal differential autoregressive moving average model.Thirdly,we analyzed the character⁃istics of time series changes of CPU load data.Finally,we calculated the parameters of the seasonal difference autore⁃gressive moving average model and the CPU load prediction results iteratively.Experimental results show that the MAPE value of the proposed method is less than 25%,indicating high prediction accuracy of the method.