论文部分内容阅读
采集不同微水含量的变压器油的近红外光谱,利用集合经验模分解(EEMD)与连续投影算法(SPA),建立变压器油中微量水分的最小二乘支持向量机(LS-SVM)回归模型。结果表明,原始求导光谱经EEMD分解后得到8个本征模态函数(IMF),在去掉第一个IMF后重构数据比原始求导光谱数据直接建模具有较好的效果,而利用去掉第一个IMF后重构数据经SPA筛选出的4个特征光谱(只占全谱的0.78%)来建模则具有更好的预测效果,预测均方根误差为1.04776×10-3,预测相关系数为0.9840,说明EEMD与SPA联用具有比EEMD及SPA单独运用更好的效果,且最优模型应用于实际油品的检测同样具有很好的效果,对实现油中水分的高精度检测以及低维度变量建模具有实际意义。
Near-infrared spectra of transformer oil with different water contents were collected. Least square support vector machine (LS-SVM) regression model was established by using Set EMPD and SPA, and establishing trace moisture in transformer oil. The results show that the Eigenfunctions (IMFs) of the original derivative spectra are obtained after EEMD decomposition and the reconstruction of the data after removing the first IMF is better than the direct derivation of the original derivative data. After the first IMF was removed and the four characteristic spectra screened by SPA (only 0.78% of the full spectrum) were reconstructed, the prediction results were better. The root mean square error of prediction was 1.04776 × 10-3, The correlation coefficient of prediction is 0.9840, which shows that the combination of EEMD and SPA has better effect than EEMD and SPA alone. The optimal model is also applied to the detection of actual oil products with good results. Detection and modeling of low-dimensional variables have practical significance.