摘要
最小二乘支持向量机在提高了支持向量机的运算速度的同时,失去了解的稀疏性。构造的多尺度稀疏最小二乘支持向量机,首先通过小波包分解对于数据进行多尺度描述,同时采用最小二乘支持向量机的学习算法获得数据之间的尺度相关性,可以实现解的稀疏性和可解释性,从而实现了系统的多尺度分解、子系统建模与合成的一体化。通过在时间序列预测上的应用可以发现,此模型在获得稀疏解的同时,极大地提高了系统的性能。而且,可以获得输出结果在不同尺度上的贡献度,增加了系统的可解释性。
Least squares support vector machine achieves faster speed at the cost of loosing the sparseness. A new method, called multiscale sparse least squares support vector machine, was proposed to obtain the sparseness and interpretability. It was the very core of this method that the multi-scale decomposition, modeling for the sub-systems and the integration is achieved adaptively. The multi-scale decomposition for the original data was obtained by wavelet packet and the correlations among these scales are obtained by the way of learning using multi-scale sparse least square support vector machine. Experiments in time series prediction demonstrate that multi-scale sparse least squares support vector machine can achieve excellent performance and sparseness at one time. In addition, the effect of different scales for the output can be achieved. It improves the interpretability and gives another way for model evaluation.
出处
《计算机技术与发展》
2011年第3期117-120,124,共5页
Computer Technology and Development
基金
国核院科研业务专项基金项目(#100-KY2010-FZ-E001)
关键词
多尺度稀疏最小二乘支持向量机
小波包分解
最小二乘支持向量机
金融时间序列
时间序列预测
multi-scale least squares support vector machine
wavelet package decomposition
least squares support vector machine
financial time series
time series prediction
