基于支持向量机的时间序列预测模型分析与应用  

Analysis and applications of time series forecasting model via support vector machines

作　　者：尉询楷[1] 李应红[1] 张朴[1] 路建明[1] 

机构地区：[1]空军工程大学工程学院飞行器与动力工程系,陕西西安710038

出　　处：《系统工程与电子技术》2005年第3期529-532,共4页Systems Engineering and Electronics

摘　　要：阐述了支持向量机在时间序列预测中应用的理论基础,给出了时间序列预测分析的基本框架。将支持向量机预测模型应用于某型航空发动机的滑油金属含量监测中,并与递归神经网络预测器进行了比较。得出支持向量机由于采用了新型的结构风险最小化准则表现出优秀的推广能力,可预测区间较长且具有较高的准确度,而递归神经网络模型在中、短期预测中与支持向量机相差不大,在较长区间预测中效果较差的结论。Firstly the theoretical basis of support vector machines (SVM) in time series forecasting and the general framework for time series forecasting are described in detail. Then, lubrication metal content time series of a type of aeroengine is used to demonstrate the feasibility of the SVM-based forecasting model. In order to show the superiority of the model, recurrent neural network (RNN) forecaster is also used in numerical simulations. Compared with conventional forecasting method, an SVM-shows an excellent generalization ability because it adopts a new structural risk minimization principle. Finally, a conclusion is made that SVM-based forecasting model possesses high precision in both short-interval forecasting and long-interval forecasting, while RNN-based model does almost the same as SVM-based one in short intervals but has low precision in long intervals.

关 键 词：支持向量回归 递归神经网络 时间序列预测 建模与应用 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]