鲁棒最小二乘支持向量机及其在软测量中的应用

Robust Least Squares Support Vector Machines with Applications to Soft-Sensing

针对最小二乘支持向量机在利用产生于工业现场的非理想数据集进行建模预测时,稀疏化模型鲁棒性差的问题,提出了一种基于模糊C均值聚类和密度加权的稀疏化方法.首先通过模糊C均值聚类将训练样本划分为若干个子类;然后计算每个子类中各样本的可能贡献度,依次从每个子类中选取具有最大可能贡献度的样本作为支持向量;最后更新每个样本的可能贡献度,继续从各个子集中增选支持向量,直至稀疏化后的模型性能满足要求.仿真结果和磨机负荷实际应用表明,该方法能够兼顾模型在整体样本集和各工况子集上的性能,在实现模型稀疏化的同时,能够显著改善最小二乘支持向量机模型的鲁棒性.

The sparse model for forecasting established by least squares support vector machines （LSSVM） is lacking in robustness, especially, for case of non-ideal data set produced in the industrial field as training date set. A fuzzy C-means clustering and density weighted based sparsity strategy is proposed. The training data set is divided into several subsets by fuzzy C- means clustering; the potential contribution of each sample is calculated and the sample with the greatest potential contribution in its own subset is selected as the support vector; the potential contribution of each sample is updated, more support vectors in the training data set are iteratively selected, until the user-defined performance is achieved. The simulation and applied examples indicate that the proposed strategy enables to achieve a sparse model with the corresponding character in the whole training data set and each subset, and the model robustness is improved significantly.