适用于小子样时间序列预测的动态回归极端学习机  

Dynamic Regression Extreme Learning Machine and Its Application to Small-sample Time Series Prediction

作　　者：张弦[1] 王宏力[1] 

机构地区：[1]第二炮兵工程学院自动控制工程系,陕西西安710025

出　　处：《信息与控制》2011年第5期704-709,共6页Information and Control

摘　　要：针对设备状态在线监测中的小子样建模问题,提出一种基于动态回归极端学习机(dynamic regressionextreme leaming machine,DR-ELM)的设备状态在线监测方法.该方法利用设备状态数据训练基于DR-ELM的预测模型,通过逐次增加新数据与删减旧数据的方式,对DR-ELM预测模型进行在线训练,从而实现对设备状态的准确预测.混沌时间序列预测仿真与基于时间序列预测的风机状态监测实例表明,相比于极端学习机(extreme learningmachine,ELM)与在线贯序极端学习机(on-line sequential extreme learning machine,OS-ELM),该方法的计算效率与预测精度更高.To deal with the problem of small-sample modeling in equipment condition on-line monitoring,an on-line monitoring method based on dynamic regression extreme learning machine（DR-ELM） is proposed.Condition data of mechanical equipment are used to train a prediction model based on DR-ELM.In an iterative manner,the latest condition data are adopted and the oldest condition data are abandoned,to achieve the DR-ELM prediction model training on-line.Thus, the current condition of mechanical equipment can be effectively predicted by the method.Simulation on chaotic time series prediction and fan condition monitoring based on time series prediction indicate that the method has better performance in training computational cost and prediction accuracy in comparison with conventional condition monitoring methods based on extreme learning machine（ELM） and on-line sequential extreme learning machine（OS-ELM）.

关 键 词：极端学习机 在线训练 小子样 时间序列预测 状态监测 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]