基于非线性时间序列的预测模型检验与优化的研究

Research of the Optimizing and Testing of Forecasting Model Based on the Non-linear Time Series

模型的适用性检验和参数优化是系统建模的最关键环节,对于预测模型的适用性检验,常采用残差方差图、最小信息准则和AIC准则等方法,存在计算量大、准确性低、模型不唯一等缺点.本文给出采用自相关系数和偏自相关系数的拖尾先对ARIMA模型检验,再对其进行F适用性检验,克服了由于观测样本的长度是有限的,偏相关的估计存在误差,拖尾时不能为ARMA定阶的缺陷,并采用具有超线性收敛性等诸多优点的变尺度法对模型参数进行了优化,得到了较为精确的、单一AIRMA模型,该方法可应用于网络流量模型的适用性检验和模型优化,为网络流量的预测、异常检测和服务器负载预测的应用奠定了坚实的基础.

The adaptability testing of model and the optimization of model parameter are the most critical part in system modeling. Residual variance plot, minimum information criteria and AIC criteria are usually adopted in the adaptability testing of forecasting model, which includes disadvantages of big amount of calculating,low veracity and model is not the only. To adopt the trailing of auto-correlation coefficients and partial correlation coefficients to test the ARMA model and then through the F adaptability testing, the disadvantages that the observation sample length is limited, estimation of the partial correlation has error and ARMA model order can not be determined when trailing have been overcome, and then optimized the model parameters by means of scale transformation which have many of the advantages such as nitro-linear convergence and so on,thus the haploid ARMA model with higher veracity can be gained, which is a convincing method of network traffic model adaptability testing and model optimizing and established a stable foundation of the application of network traffic forecasting, abnorrnity testing and server loading forecast.