摘要： 针对传统在线贯序极限学习机存在的过学习和分类器输出不稳定等问题，将结构风险最小化理论引入到极限学习机中，用小波函数替代原有的隐层激励函数构建正则小波极限学习机，并与在线学习方法结合，提出在线正则小波极限学习机。仿真实验结果表明，在线正则小波极限学习机克服过学习和局部最优等问题，能够实现快速在线学习，具有良好的泛化性和鲁棒性。

关键词: 在线贯序极限学习机, 小波分析, 在线学习, 模式识别, 结构风险, 泛化性能, 鲁棒性

Abstract: The traditional Online Sequential Extreme Learning Machine(OS-ELM) has variations in different trials of simulations and the over-learning problem. Using wavelet substitutes network’s traditional activation function, structural risk minimization is used to modify the problem, so a novel algorithm called regularized wavelet extreme learning machine is proposed. Motivated by online learning method, Online Sequential Regularized Wavelet Extreme Learning Machine(OS-RWELM) is designed. Experimental results show that this algorithm avoids the local minimum and over-learning problem, has a fast online learning speed and a good generalization and robustness.

Key words: Online Sequential Extreme Learning Machine(OS-ELM), wavelet analysis, online learning, pattern recognition, structural risk, generalization performance, robustness

中图分类号: 

• TP18