论文部分内容阅读
为改善直接支持向量回归机(DSVMR)的稀疏性,提出一种适用于DSVMR的剪样训练算法.该算法利用矩阵变换实现剪样前后DSVMR的递推求解,提高了剪样训练过程中DSVMR多次训练的计算效率.混沌时间序列预测仿真表明,该算法有效改善了DSVMR的稀疏性,且计算效率较基于Cholesky分解的剪样训练算法有显著提高.飞机故障率预测实例表明,经剪样训练后的DSVMR的预测精度高于BP(back-propagation)神经网络预测方法与RBF(radial casis function)神经网络预测方法.
In order to improve the sparsity of direct support vector regression machine (DSVMR), a shear training algorithm suitable for DSVMR is proposed, which uses matrix transformation to solve recursive DSVMR before and after cutting and improves DSVMR The simulation results show that this algorithm effectively improves the sparsity of DSVMR and significantly improves the computational efficiency compared to the shear-based training algorithm based on Cholesky decomposition.The examples of aircraft failure rate prediction show that the shear-proof training The DSVMR prediction accuracy is higher than the BP (back-propagation) neural network prediction method and the RBF (radial casis function) neural network prediction method.