动态最小二乘支持向量机学习算法

Learning Algorithm of Dynamic Least Square Support Vector Machine

针对大样本集的训练问题和动态训练样本的模型更新问题,提出了动态最小二乘支持向量机学习算法。该算法充分利用已建好的模型,逐渐加入新样本,并可删除位于任何位置的非支持向量,避免了矩阵求逆运算,保证了算法的高效率。大坝变形及电离层延迟两个时间序列的预报实例表明,该算法具有计算时间短、预报精度高的特点。

Aiming at solving the training problem of large scale sample sets and model modif- ying problem of dynamic training sets, the dynamic least square support vector machine method is presented. The novel method can take full advantage of the model which is built by incremental algorithm. Based on the updated model, the new samples can be added gradu ally, the non support vectors located in any position of the training set can be found and dele ted easily. The matrix inverse algorithm is avoided; then a high calculation efficiency can be obtained theoretically. Two examples, one is dam deformation prediction and the other is ionosphere delay prediction, show the excellent performance of the proposed algorithm in modeling time and prediction precision. All the samples indicate that the method proposed is superior than other methods at present.