核极限学习机的在线状态预测方法综述

Survey of Kernel Extreme Learning Machine Methods for Online Prediction

对非平稳混沌时间序列进行在线预测是当前科学和工程领域中的一个重要研究方向,核极限学习机(kernel extreme learning machine,KELM)为其提供了一种有效的数学模型。由于学习速度快、泛化性能好,在线贯序核极限学习机(online sequential KELM,OSKELM)在状态预测中得到了广泛的研究与扩展。首先,描述了问题并介绍了OSKELM的数学模型;然后,以混沌时间序列为应用背景,对基于OSKELM的各种改进方法进行了分类综述,包括基于数据增量的OSKELM、基于稀疏字典的OSKELM、基于参数寻优和遗忘因子的OSKELM以及其他方法,并对算法性能进行比较和分析;最后总结并讨论了该方法的未来研究方向。

The online prediction of nonstationary chaotic time series is an important research direction in the field of science and engineering,for which kernel extreme learning machine(KELM)provides an effective mathematical model.Due to its fast learning speed and good generalization performance,online sequential KELM(OSKELM)has been extensively studied and extended in state prediction.The problem was described and mathematical model of OSKELM were introduced.Then,with chaotic time series as the application background,various improved OSKELM-based methods were classified and summarized,including OSKELM based on incremental learning method,OSKELM based on sparse dictionary,OSKELM based on parameter optimization,OSKELM based on forgetting factor and other methods.Afterwards,the performance of the algorithm was compared and analyzed.The thesis was summarized and the future research direction of this method was discussed.