基于鲁棒学习的最小二乘支持向量机及其应用

Robust least squares support vector machine based on robust learning algorithm and its application

鉴于最小二乘支持向量机比标准支持向量机具有更高的计算效率和拟合精度,但缺少标准支持向量机的鲁棒性,即当采样数据存在奇异点或者误差变量的高斯分布假设不成立时,会导致不稳健的估计结果,提出了一种鲁棒最小二乘支持向量机方法.该方法在最小二乘支持向量机基础上,通过引入鲁棒学习方法来获得鲁棒估计.仿真分析及某湿法冶金厂的应用实例验证了该方法的可行性和有效性.

Least squares support vector machine（LS-SVM） is computationally more efficient than the standard SVM, but unfortunately the robustness of standard SVM is lost. LS-SVM might lead to estimates which are less robust with respect to outliers on the data or when the assumption of a Gaussian distribution for error variables is not realistic. Therefore, an approach based on the robust least squares support vector machine（RLS-SVM） is proposed, in which robust learning algorithm（RLA） is employed to enhance the robust capability of LS-SVM. Finally, simulation analysis and the modeling of a typical plant for hydrometallurgy illustrate the effectiveness and feasibility of the presented method.