面向动态数据流的改进OSELM算法研究  

Research on Improved OSELM Algorithm for Dynamic Data Flow

作　　者：乔延松 贺泽华 赵绪营 QIAO Yansong;HE Zehua;ZHAO Xuying(Beijing Electronic Science and Technology Institute,Beijing 100070,P.R.China)

机构地区：[1]北京电子科技学院,北京市100070

出　　处：《北京电子科技学院学报》2020年第3期1-12,共12页Journal of Beijing Electronic Science And Technology Institute

摘　　要：随着信息化时代的发展,动态数据流分析成为一个值得深入研究的课题,而在线学习方法是解决这一问题的关键。在众多的在线学习算法中,在线贯序超限学习机(Online sequential Extreme Learning Machine,OSELM)作为一种优秀的在线学习算法,具有着泛化能力强,学习速度快等明显优点,在动态数据流分析中得到了广泛应用。本文首先介绍了OSELM的理论基础和算法实现,然后以动态数据流分析为应用背景,对引入遗忘因子和正则化技术的OSELM改进算法进行了研究。接着对原始OSELM算法和各种改进的OSELM算法进行了实验比较与分析,得出的实验结论如下:发现遗忘因子能减少模型的在线预测误差,正则化技术可以解决因病态矩阵带来的模型不稳定问题。最后总结出基于正则化与遗忘因子的在线学习模型最适合工程应用的结论。With the development of information,dynamic data flow analysis has become a topic worthy of further research,and online learning method is the key to solve this problem.In numerous online learning algorithms,online sequential Extreme Learning Machine(OSELM)stands out from the rest due to the obvious advantages of strong generalization ability and fast learning speed,and finds a wide application in dynamic data flow analysis.In this paper,theoretical fundamental and algorithm implementation of the OSELM are first introduced.For realizing the dynamic data flow analysis,improved OSELM algorithms with forgetting factor and regularization technology introduced are then studied.Finally,the original OSELM algorithm and various improved OSELM algorithms are compared and analyzed by experiments.Experiment result indicate that forgetting factor reduces the online prediction error of the model,and regularization technology could solve the model instability problem caused by the Ill-Conditioned Matrix.The paper concludes that the online learning model based on regularization and forgetting factor best suits the engineering application.

关 键 词：动态数据流 在线学习 OSELM 正则化 遗忘因子 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]