具有增加删除机制的正则化极端学习机的混沌时间序列预测

Chaotic time series prediction using add-delete mechanism based regularized extreme learning machine

针对正则化极端学习机的隐层具有随机选择的特性,提出了一种增加删除机制来自适应地确定正则化极端学习机的隐层节点数.这种机制以对优化目标函数影响的大小作为评价隐层节点优劣的标准,从而淘汰那些比较"差"的节点,将那些比较"优"的节点保留下来,起到一个优化正则化极端学习机隐层节点数的目的.与已有的只具有增加隐层节点数的机制相比较,本文提出的增加删除机制在减少正则化极端学习机隐层节点数、增强其泛化性能、提高其实时性等方面具有一定的优势.典型混沌时间序列的实例证明了具有增加删除机制的正则化极端学习机的有效性和可行性.

Considering a regularized extreme learning machine（RELM） with randomly generated hidden nodes, an add-delete mechanism is proposed to determine the number of hidden nodes adaptively, where the extent of contribution to the objective function of RELM is treated as the criterion for judging each hidden node, that is, the large the better, and vice versa. As a result, the better hidden nodes are kept. On the contrary, the so-called worse hidden nodes are deleted. Naturally, the hidden nodes of RELM are selected optimally. In contrast to the other method only with the add mechanism, the proposed one has some advantages in the number of hidden nodes, generalization performance, and the real time. The experimental results on classical chaotic time series demonstrate the effectiveness and feasibility of the proposed add-delete mechanism for RELM.