物理学报Issue(7)：1-8,8.DOI:10.7498/aps.64.070504
多元混沌时间序列的多核极端学习机建模预测
Multivariate chaotic time series prediction using multiple kernel extreme learning machine
摘要
Abstract
Multivariate chaotic time series is widely present in nature, such as in economy, society, industry and other fields. Modeling and predicting multivariate time series will help human to better manage, control, and make decision. A prediction method based on multiple kernel extreme learning machine is proposed in this paper to model the complex dynamics of multivariate chaotic time series. First, the multivariate chaotic time series is reconstructed in phase space, transforming the temporal correlation into spatial correlation. Then, a prediction model-multiple kernel extreme learn-ing machine, which combines the multiple kernel learning and extreme learning machine with kernels, is proposed to approximate the nonlinear function of the input – output data in phase space. The proposed multiple kernel extreme learning machine could effectively combine the simple training of extreme learning machine with kernels and the data fusion capabilities of multiple kernel learning. Simulation results based on Lorenz chaotic time series prediction and San Francisco monthly runoff prediction demonstrate that, compared with other state-of-art chaotic time series prediction methods, the proposed multiple kernel extreme learning machine could get a better prediction accuracy.关键词
混沌时间序列/神经网络/核方法/预测Key words
chaotic time series/neural networks/kernel methods/prediction引用本文复制引用
王新迎,韩敏..多元混沌时间序列的多核极端学习机建模预测[J].物理学报,2015,(7):1-8,8.基金项目
国家自然科学基金(批准号:61374154)和国家重点基础研究发展计划(973计划)(批准号:2013CB430403)资助的课题 （批准号:61374154）
