非晶态合金晶化电特性时间序列模型及预测  

Modeling and predicting of time series of electric characteristic for amorphous alloy crystallization

作　　者：张万宏[1] 张妍[1] 周维琴 

机构地区：[1]青海大学化工学院,青海西宁810016 [2]长庆油田第三采油厂,宁夏银川750100

出　　处：《青海大学学报（自然科学版）》2012年第5期14-19,共6页Journal of Qinghai University(Natural Science)

基　　金：青海大学中青年基金项目(2010-QG-07)

摘　　要：在非晶态合金晶化过程中,尽管无法知道那些系统变量及它们之间如何相互作用来支配晶化的演化,但可以肯定的是过程变量尤其是晶化电特性蕴含着不同的时频特性。因此,将材料科学与信息科学技术相结合,利用非线性系统辨识方法,提出一种基于小波变换和AR-FILSSVM方法的晶化电特性时间序列模型及预测方法。首先利用小波变换对时间序列进行分解,分离出时间序列中的低频和高频序列,然后利用AR模型预测高频序列,利用FILSSVM方法预测低频序列,最后将各模型的预测结果叠加,从而得到原始序列的模型及预测。仿真试验表明,该方法是非晶态合金晶化电特性时间序列的有效预测方法。Although the system variables and their interaction controls the evolution of crystallization cannot been known,it is sure that these process variables has time-frequency properties,especially the electric characteristic in the process of crystallization.Combining the materials science and information science,using the nonlinear system identification method,an electric time series prediction method is proposed based on the wavelet analysis and AR-FILSSVM in this paper.By wavelet decomposition and reconstruction,the non-stationary time series are decomposed into a low frequency series and several high frequency series.The high frequency series are predicted with auto-regression models,and the low frequency is predicted with fussy incremental least square support vector machines.The prediction result of the original time series is the superimposition of the respective prediction.Experiments results show that the predicting method is effective.

关 键 词：非晶态合金 小波变换 模糊增量最小二乘支持向量机 晶化电特性时间序列 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]