基于EMD与LS-SVM的网络控制系统时延预测方法  

Time-Delay Prediction Method of Networked Control System Based on EMD and LS-SVM

作　　者：田中大[1] 高宪文[1] 李琨[1] 

机构地区：[1]东北大学信息科学与工程学院,辽宁沈阳110819

出　　处：《电子学报》2014年第5期868-874,共7页Acta Electronica Sinica

基　　金：国家自然科学基金重点项目(No.61034005)

摘　　要：为了提高基于Internet的网络控制系统中随机时延的预测精度,提出了基于经验模式分解(empirical mode decomposition,EMD)与最小二乘支持向量机(Least Squared Support Vector Machines,LS-SVM)的一步时延预测方法.首先利用EMD将时延序列分解成若干个本征模式函数分量,分解后的分量去除了原始时延序列的长相关性,同时突出时延序列不同的局部特征.然后根据各个分量的变化规律,选择不同的LS-SVM模型分别进行预测.最后将各分量的预测值叠加得到最终的预测值.仿真结果表明本文方法具有较高的预测精度.In order to predict the random time-delay of Intemet-based networked control system effectively, a hybrid one step time-delay forecasting meth~ based on Empirical Mode Decomposition（EMD） and Least Squared Support Vector Machines（ LS- SVM）is presented in the paper. Firstly, EMD algorithm can decompose time-delay sequences into some intrinsic mode functions （IMF） ,the IMF after decomposed remove the long-range dependence of original time-delay sequences and prominent the different local feature of the time-delay sequences. Secondly, according to the change law of each IMF, chose different LS-SVM model to predict time-delay.At last, all the forecasted values corresponded to these partitions are superposed to get the forecasted time-delay. Simulation results show that the proposed method has higher prediction accuracy.

关 键 词：网络控制系统 经验模式分解 最小二乘支持向量机 时延 预测 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]